xref: /qemu/hw/ppc/spapr.c (revision 8af7e1fe6fa4bdd3c4561c13a4bd4ee525e6f02f)
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 FDT_MAX_SIZE            0x100000
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            (128 * MiB)
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
206 {
207     int index = spapr_get_vcpu_id(cpu);
208     uint32_t associativity[] = {cpu_to_be32(0x5),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(0x0),
212                                 cpu_to_be32(cpu->node_id),
213                                 cpu_to_be32(index)};
214 
215     /* Advertise NUMA via ibm,associativity */
216     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217                           sizeof(associativity));
218 }
219 
220 static void spapr_dt_pa_features(SpaprMachineState *spapr,
221                                  PowerPCCPU *cpu,
222                                  void *fdt, int offset)
223 {
224     uint8_t pa_features_206[] = { 6, 0,
225         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
226     uint8_t pa_features_207[] = { 24, 0,
227         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
228         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
229         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
230         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
231     uint8_t pa_features_300[] = { 66, 0,
232         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
233         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
234         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
235         /* 6: DS207 */
236         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
237         /* 16: Vector */
238         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
239         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
241         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
242         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
243         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
244         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
245         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
246         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
247         /* 42: PM, 44: PC RA, 46: SC vec'd */
248         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
249         /* 48: SIMD, 50: QP BFP, 52: String */
250         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
251         /* 54: DecFP, 56: DecI, 58: SHA */
252         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
253         /* 60: NM atomic, 62: RNG */
254         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
255     };
256     uint8_t *pa_features = NULL;
257     size_t pa_size;
258 
259     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
260         pa_features = pa_features_206;
261         pa_size = sizeof(pa_features_206);
262     }
263     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
264         pa_features = pa_features_207;
265         pa_size = sizeof(pa_features_207);
266     }
267     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
268         pa_features = pa_features_300;
269         pa_size = sizeof(pa_features_300);
270     }
271     if (!pa_features) {
272         return;
273     }
274 
275     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
276         /*
277          * Note: we keep CI large pages off by default because a 64K capable
278          * guest provisioned with large pages might otherwise try to map a qemu
279          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
280          * even if that qemu runs on a 4k host.
281          * We dd this bit back here if we are confident this is not an issue
282          */
283         pa_features[3] |= 0x20;
284     }
285     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
286         pa_features[24] |= 0x80;    /* Transactional memory support */
287     }
288     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
289         /* Workaround for broken kernels that attempt (guest) radix
290          * mode when they can't handle it, if they see the radix bit set
291          * in pa-features. So hide it from them. */
292         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
293     }
294 
295     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
296 }
297 
298 static hwaddr spapr_node0_size(MachineState *machine)
299 {
300     if (machine->numa_state->num_nodes) {
301         int i;
302         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
303             if (machine->numa_state->nodes[i].node_mem) {
304                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
305                            machine->ram_size);
306             }
307         }
308     }
309     return machine->ram_size;
310 }
311 
312 static void add_str(GString *s, const gchar *s1)
313 {
314     g_string_append_len(s, s1, strlen(s1) + 1);
315 }
316 
317 static int spapr_dt_memory_node(void *fdt, int nodeid, hwaddr start,
318                                 hwaddr size)
319 {
320     uint32_t associativity[] = {
321         cpu_to_be32(0x4), /* length */
322         cpu_to_be32(0x0), cpu_to_be32(0x0),
323         cpu_to_be32(0x0), cpu_to_be32(nodeid)
324     };
325     char mem_name[32];
326     uint64_t mem_reg_property[2];
327     int off;
328 
329     mem_reg_property[0] = cpu_to_be64(start);
330     mem_reg_property[1] = cpu_to_be64(size);
331 
332     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
333     off = fdt_add_subnode(fdt, 0, mem_name);
334     _FDT(off);
335     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
336     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
337                       sizeof(mem_reg_property))));
338     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
339                       sizeof(associativity))));
340     return off;
341 }
342 
343 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
344 {
345     MemoryDeviceInfoList *info;
346 
347     for (info = list; info; info = info->next) {
348         MemoryDeviceInfo *value = info->value;
349 
350         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
351             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
352 
353             if (addr >= pcdimm_info->addr &&
354                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
355                 return pcdimm_info->node;
356             }
357         }
358     }
359 
360     return -1;
361 }
362 
363 struct sPAPRDrconfCellV2 {
364      uint32_t seq_lmbs;
365      uint64_t base_addr;
366      uint32_t drc_index;
367      uint32_t aa_index;
368      uint32_t flags;
369 } QEMU_PACKED;
370 
371 typedef struct DrconfCellQueue {
372     struct sPAPRDrconfCellV2 cell;
373     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
374 } DrconfCellQueue;
375 
376 static DrconfCellQueue *
377 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
378                       uint32_t drc_index, uint32_t aa_index,
379                       uint32_t flags)
380 {
381     DrconfCellQueue *elem;
382 
383     elem = g_malloc0(sizeof(*elem));
384     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
385     elem->cell.base_addr = cpu_to_be64(base_addr);
386     elem->cell.drc_index = cpu_to_be32(drc_index);
387     elem->cell.aa_index = cpu_to_be32(aa_index);
388     elem->cell.flags = cpu_to_be32(flags);
389 
390     return elem;
391 }
392 
393 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
394                                       int offset, MemoryDeviceInfoList *dimms)
395 {
396     MachineState *machine = MACHINE(spapr);
397     uint8_t *int_buf, *cur_index;
398     int ret;
399     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
400     uint64_t addr, cur_addr, size;
401     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
402     uint64_t mem_end = machine->device_memory->base +
403                        memory_region_size(&machine->device_memory->mr);
404     uint32_t node, buf_len, nr_entries = 0;
405     SpaprDrc *drc;
406     DrconfCellQueue *elem, *next;
407     MemoryDeviceInfoList *info;
408     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
409         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
410 
411     /* Entry to cover RAM and the gap area */
412     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
413                                  SPAPR_LMB_FLAGS_RESERVED |
414                                  SPAPR_LMB_FLAGS_DRC_INVALID);
415     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
416     nr_entries++;
417 
418     cur_addr = machine->device_memory->base;
419     for (info = dimms; info; info = info->next) {
420         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
421 
422         addr = di->addr;
423         size = di->size;
424         node = di->node;
425 
426         /*
427          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
428          * area is marked hotpluggable in the next iteration for the bigger
429          * chunk including the NVDIMM occupied area.
430          */
431         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
432             continue;
433 
434         /* Entry for hot-pluggable area */
435         if (cur_addr < addr) {
436             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
437             g_assert(drc);
438             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
439                                          cur_addr, spapr_drc_index(drc), -1, 0);
440             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
441             nr_entries++;
442         }
443 
444         /* Entry for DIMM */
445         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
446         g_assert(drc);
447         elem = spapr_get_drconf_cell(size / lmb_size, addr,
448                                      spapr_drc_index(drc), node,
449                                      SPAPR_LMB_FLAGS_ASSIGNED);
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     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_SIZE));
955     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
956                           RTAS_ERROR_LOG_MAX));
957     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
958                           RTAS_EVENT_SCAN_RATE));
959 
960     g_assert(msi_nonbroken);
961     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
962 
963     /*
964      * According to PAPR, rtas ibm,os-term does not guarantee a return
965      * back to the guest cpu.
966      *
967      * While an additional ibm,extended-os-term property indicates
968      * that rtas call return will always occur. Set this property.
969      */
970     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
971 
972     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
973                      lrdr_capacity, sizeof(lrdr_capacity)));
974 
975     spapr_dt_rtas_tokens(fdt, rtas);
976 }
977 
978 /*
979  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
980  * and the XIVE features that the guest may request and thus the valid
981  * values for bytes 23..26 of option vector 5:
982  */
983 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
984                                           int chosen)
985 {
986     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
987 
988     char val[2 * 4] = {
989         23, 0x00, /* XICS / XIVE mode */
990         24, 0x00, /* Hash/Radix, filled in below. */
991         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
992         26, 0x40, /* Radix options: GTSE == yes. */
993     };
994 
995     if (spapr->irq->xics && spapr->irq->xive) {
996         val[1] = SPAPR_OV5_XIVE_BOTH;
997     } else if (spapr->irq->xive) {
998         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
999     } else {
1000         assert(spapr->irq->xics);
1001         val[1] = SPAPR_OV5_XIVE_LEGACY;
1002     }
1003 
1004     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1005                           first_ppc_cpu->compat_pvr)) {
1006         /*
1007          * If we're in a pre POWER9 compat mode then the guest should
1008          * do hash and use the legacy interrupt mode
1009          */
1010         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1011         val[3] = 0x00; /* Hash */
1012     } else if (kvm_enabled()) {
1013         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1014             val[3] = 0x80; /* OV5_MMU_BOTH */
1015         } else if (kvmppc_has_cap_mmu_radix()) {
1016             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1017         } else {
1018             val[3] = 0x00; /* Hash */
1019         }
1020     } else {
1021         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1022         val[3] = 0xC0;
1023     }
1024     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1025                      val, sizeof(val)));
1026 }
1027 
1028 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1029 {
1030     MachineState *machine = MACHINE(spapr);
1031     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1032     int chosen;
1033 
1034     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1035 
1036     if (reset) {
1037         const char *boot_device = machine->boot_order;
1038         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1039         size_t cb = 0;
1040         char *bootlist = get_boot_devices_list(&cb);
1041 
1042         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1043             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1044                                     machine->kernel_cmdline));
1045         }
1046 
1047         if (spapr->initrd_size) {
1048             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1049                                   spapr->initrd_base));
1050             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1051                                   spapr->initrd_base + spapr->initrd_size));
1052         }
1053 
1054         if (spapr->kernel_size) {
1055             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1056                                   cpu_to_be64(spapr->kernel_size) };
1057 
1058             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1059                          &kprop, sizeof(kprop)));
1060             if (spapr->kernel_le) {
1061                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1062             }
1063         }
1064         if (boot_menu) {
1065             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1066         }
1067         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1068         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1069         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1070 
1071         if (cb && bootlist) {
1072             int i;
1073 
1074             for (i = 0; i < cb; i++) {
1075                 if (bootlist[i] == '\n') {
1076                     bootlist[i] = ' ';
1077                 }
1078             }
1079             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1080         }
1081 
1082         if (boot_device && strlen(boot_device)) {
1083             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1084         }
1085 
1086         if (!spapr->has_graphics && stdout_path) {
1087             /*
1088              * "linux,stdout-path" and "stdout" properties are
1089              * deprecated by linux kernel. New platforms should only
1090              * use the "stdout-path" property. Set the new property
1091              * and continue using older property to remain compatible
1092              * with the existing firmware.
1093              */
1094             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1095             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1096         }
1097 
1098         /*
1099          * We can deal with BAR reallocation just fine, advertise it
1100          * to the guest
1101          */
1102         if (smc->linux_pci_probe) {
1103             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1104         }
1105 
1106         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1107 
1108         g_free(stdout_path);
1109         g_free(bootlist);
1110     }
1111 
1112     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1113 }
1114 
1115 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1116 {
1117     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1118      * KVM to work under pHyp with some guest co-operation */
1119     int hypervisor;
1120     uint8_t hypercall[16];
1121 
1122     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1123     /* indicate KVM hypercall interface */
1124     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1125     if (kvmppc_has_cap_fixup_hcalls()) {
1126         /*
1127          * Older KVM versions with older guest kernels were broken
1128          * with the magic page, don't allow the guest to map it.
1129          */
1130         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1131                                   sizeof(hypercall))) {
1132             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1133                              hypercall, sizeof(hypercall)));
1134         }
1135     }
1136 }
1137 
1138 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1139 {
1140     MachineState *machine = MACHINE(spapr);
1141     MachineClass *mc = MACHINE_GET_CLASS(machine);
1142     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1143     int ret;
1144     void *fdt;
1145     SpaprPhbState *phb;
1146     char *buf;
1147 
1148     fdt = g_malloc0(space);
1149     _FDT((fdt_create_empty_tree(fdt, space)));
1150 
1151     /* Root node */
1152     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1153     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1154     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1155 
1156     /* Guest UUID & Name*/
1157     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1158     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1159     if (qemu_uuid_set) {
1160         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1161     }
1162     g_free(buf);
1163 
1164     if (qemu_get_vm_name()) {
1165         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1166                                 qemu_get_vm_name()));
1167     }
1168 
1169     /* Host Model & Serial Number */
1170     if (spapr->host_model) {
1171         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1172     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1173         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1174         g_free(buf);
1175     }
1176 
1177     if (spapr->host_serial) {
1178         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1179     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1180         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1181         g_free(buf);
1182     }
1183 
1184     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1185     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1186 
1187     /* /interrupt controller */
1188     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1189 
1190     ret = spapr_dt_memory(spapr, fdt);
1191     if (ret < 0) {
1192         error_report("couldn't setup memory nodes in fdt");
1193         exit(1);
1194     }
1195 
1196     /* /vdevice */
1197     spapr_dt_vdevice(spapr->vio_bus, fdt);
1198 
1199     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1200         ret = spapr_dt_rng(fdt);
1201         if (ret < 0) {
1202             error_report("could not set up rng device in the fdt");
1203             exit(1);
1204         }
1205     }
1206 
1207     QLIST_FOREACH(phb, &spapr->phbs, list) {
1208         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1209         if (ret < 0) {
1210             error_report("couldn't setup PCI devices in fdt");
1211             exit(1);
1212         }
1213     }
1214 
1215     spapr_dt_cpus(fdt, spapr);
1216 
1217     if (smc->dr_lmb_enabled) {
1218         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1219     }
1220 
1221     if (mc->has_hotpluggable_cpus) {
1222         int offset = fdt_path_offset(fdt, "/cpus");
1223         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1224         if (ret < 0) {
1225             error_report("Couldn't set up CPU DR device tree properties");
1226             exit(1);
1227         }
1228     }
1229 
1230     /* /event-sources */
1231     spapr_dt_events(spapr, fdt);
1232 
1233     /* /rtas */
1234     spapr_dt_rtas(spapr, fdt);
1235 
1236     /* /chosen */
1237     spapr_dt_chosen(spapr, fdt, reset);
1238 
1239     /* /hypervisor */
1240     if (kvm_enabled()) {
1241         spapr_dt_hypervisor(spapr, fdt);
1242     }
1243 
1244     /* Build memory reserve map */
1245     if (reset) {
1246         if (spapr->kernel_size) {
1247             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1248                                   spapr->kernel_size)));
1249         }
1250         if (spapr->initrd_size) {
1251             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1252                                   spapr->initrd_size)));
1253         }
1254     }
1255 
1256     if (smc->dr_phb_enabled) {
1257         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1258         if (ret < 0) {
1259             error_report("Couldn't set up PHB DR device tree properties");
1260             exit(1);
1261         }
1262     }
1263 
1264     /* NVDIMM devices */
1265     if (mc->nvdimm_supported) {
1266         spapr_dt_persistent_memory(fdt);
1267     }
1268 
1269     return fdt;
1270 }
1271 
1272 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1273 {
1274     SpaprMachineState *spapr = opaque;
1275 
1276     return (addr & 0x0fffffff) + spapr->kernel_addr;
1277 }
1278 
1279 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1280                                     PowerPCCPU *cpu)
1281 {
1282     CPUPPCState *env = &cpu->env;
1283 
1284     /* The TCG path should also be holding the BQL at this point */
1285     g_assert(qemu_mutex_iothread_locked());
1286 
1287     if (msr_pr) {
1288         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1289         env->gpr[3] = H_PRIVILEGE;
1290     } else {
1291         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1292     }
1293 }
1294 
1295 struct LPCRSyncState {
1296     target_ulong value;
1297     target_ulong mask;
1298 };
1299 
1300 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1301 {
1302     struct LPCRSyncState *s = arg.host_ptr;
1303     PowerPCCPU *cpu = POWERPC_CPU(cs);
1304     CPUPPCState *env = &cpu->env;
1305     target_ulong lpcr;
1306 
1307     cpu_synchronize_state(cs);
1308     lpcr = env->spr[SPR_LPCR];
1309     lpcr &= ~s->mask;
1310     lpcr |= s->value;
1311     ppc_store_lpcr(cpu, lpcr);
1312 }
1313 
1314 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1315 {
1316     CPUState *cs;
1317     struct LPCRSyncState s = {
1318         .value = value,
1319         .mask = mask
1320     };
1321     CPU_FOREACH(cs) {
1322         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1323     }
1324 }
1325 
1326 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1327 {
1328     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1329 
1330     /* Copy PATE1:GR into PATE0:HR */
1331     entry->dw0 = spapr->patb_entry & PATE0_HR;
1332     entry->dw1 = spapr->patb_entry;
1333 }
1334 
1335 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1336 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1337 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1338 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1339 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1340 
1341 /*
1342  * Get the fd to access the kernel htab, re-opening it if necessary
1343  */
1344 static int get_htab_fd(SpaprMachineState *spapr)
1345 {
1346     Error *local_err = NULL;
1347 
1348     if (spapr->htab_fd >= 0) {
1349         return spapr->htab_fd;
1350     }
1351 
1352     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1353     if (spapr->htab_fd < 0) {
1354         error_report_err(local_err);
1355     }
1356 
1357     return spapr->htab_fd;
1358 }
1359 
1360 void close_htab_fd(SpaprMachineState *spapr)
1361 {
1362     if (spapr->htab_fd >= 0) {
1363         close(spapr->htab_fd);
1364     }
1365     spapr->htab_fd = -1;
1366 }
1367 
1368 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1369 {
1370     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1371 
1372     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1373 }
1374 
1375 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1376 {
1377     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1378 
1379     assert(kvm_enabled());
1380 
1381     if (!spapr->htab) {
1382         return 0;
1383     }
1384 
1385     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1386 }
1387 
1388 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1389                                                 hwaddr ptex, int n)
1390 {
1391     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1392     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1393 
1394     if (!spapr->htab) {
1395         /*
1396          * HTAB is controlled by KVM. Fetch into temporary buffer
1397          */
1398         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1399         kvmppc_read_hptes(hptes, ptex, n);
1400         return hptes;
1401     }
1402 
1403     /*
1404      * HTAB is controlled by QEMU. Just point to the internally
1405      * accessible PTEG.
1406      */
1407     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1408 }
1409 
1410 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1411                               const ppc_hash_pte64_t *hptes,
1412                               hwaddr ptex, int n)
1413 {
1414     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1415 
1416     if (!spapr->htab) {
1417         g_free((void *)hptes);
1418     }
1419 
1420     /* Nothing to do for qemu managed HPT */
1421 }
1422 
1423 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1424                       uint64_t pte0, uint64_t pte1)
1425 {
1426     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1427     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1428 
1429     if (!spapr->htab) {
1430         kvmppc_write_hpte(ptex, pte0, pte1);
1431     } else {
1432         if (pte0 & HPTE64_V_VALID) {
1433             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1434             /*
1435              * When setting valid, we write PTE1 first. This ensures
1436              * proper synchronization with the reading code in
1437              * ppc_hash64_pteg_search()
1438              */
1439             smp_wmb();
1440             stq_p(spapr->htab + offset, pte0);
1441         } else {
1442             stq_p(spapr->htab + offset, pte0);
1443             /*
1444              * When clearing it we set PTE0 first. This ensures proper
1445              * synchronization with the reading code in
1446              * ppc_hash64_pteg_search()
1447              */
1448             smp_wmb();
1449             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1450         }
1451     }
1452 }
1453 
1454 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1455                              uint64_t pte1)
1456 {
1457     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1458     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1459 
1460     if (!spapr->htab) {
1461         /* There should always be a hash table when this is called */
1462         error_report("spapr_hpte_set_c called with no hash table !");
1463         return;
1464     }
1465 
1466     /* The HW performs a non-atomic byte update */
1467     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1468 }
1469 
1470 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1471                              uint64_t pte1)
1472 {
1473     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1474     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1475 
1476     if (!spapr->htab) {
1477         /* There should always be a hash table when this is called */
1478         error_report("spapr_hpte_set_r called with no hash table !");
1479         return;
1480     }
1481 
1482     /* The HW performs a non-atomic byte update */
1483     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1484 }
1485 
1486 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1487 {
1488     int shift;
1489 
1490     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1491      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1492      * that's much more than is needed for Linux guests */
1493     shift = ctz64(pow2ceil(ramsize)) - 7;
1494     shift = MAX(shift, 18); /* Minimum architected size */
1495     shift = MIN(shift, 46); /* Maximum architected size */
1496     return shift;
1497 }
1498 
1499 void spapr_free_hpt(SpaprMachineState *spapr)
1500 {
1501     g_free(spapr->htab);
1502     spapr->htab = NULL;
1503     spapr->htab_shift = 0;
1504     close_htab_fd(spapr);
1505 }
1506 
1507 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1508                           Error **errp)
1509 {
1510     long rc;
1511 
1512     /* Clean up any HPT info from a previous boot */
1513     spapr_free_hpt(spapr);
1514 
1515     rc = kvmppc_reset_htab(shift);
1516     if (rc < 0) {
1517         /* kernel-side HPT needed, but couldn't allocate one */
1518         error_setg_errno(errp, errno,
1519                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1520                          shift);
1521         /* This is almost certainly fatal, but if the caller really
1522          * wants to carry on with shift == 0, it's welcome to try */
1523     } else if (rc > 0) {
1524         /* kernel-side HPT allocated */
1525         if (rc != shift) {
1526             error_setg(errp,
1527                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1528                        shift, rc);
1529         }
1530 
1531         spapr->htab_shift = shift;
1532         spapr->htab = NULL;
1533     } else {
1534         /* kernel-side HPT not needed, allocate in userspace instead */
1535         size_t size = 1ULL << shift;
1536         int i;
1537 
1538         spapr->htab = qemu_memalign(size, size);
1539         if (!spapr->htab) {
1540             error_setg_errno(errp, errno,
1541                              "Could not allocate HPT of order %d", shift);
1542             return;
1543         }
1544 
1545         memset(spapr->htab, 0, size);
1546         spapr->htab_shift = shift;
1547 
1548         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1549             DIRTY_HPTE(HPTE(spapr->htab, i));
1550         }
1551     }
1552     /* We're setting up a hash table, so that means we're not radix */
1553     spapr->patb_entry = 0;
1554     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1555 }
1556 
1557 void spapr_setup_hpt(SpaprMachineState *spapr)
1558 {
1559     int hpt_shift;
1560 
1561     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1562         || (spapr->cas_reboot
1563             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1564         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1565     } else {
1566         uint64_t current_ram_size;
1567 
1568         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1569         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1570     }
1571     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1572 
1573     if (kvm_enabled()) {
1574         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1575 
1576         /* Check our RMA fits in the possible VRMA */
1577         if (vrma_limit < spapr->rma_size) {
1578             error_report("Unable to create %" HWADDR_PRIu
1579                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1580                          spapr->rma_size / MiB, vrma_limit / MiB);
1581             exit(EXIT_FAILURE);
1582         }
1583     }
1584 }
1585 
1586 static int spapr_reset_drcs(Object *child, void *opaque)
1587 {
1588     SpaprDrc *drc =
1589         (SpaprDrc *) object_dynamic_cast(child,
1590                                                  TYPE_SPAPR_DR_CONNECTOR);
1591 
1592     if (drc) {
1593         spapr_drc_reset(drc);
1594     }
1595 
1596     return 0;
1597 }
1598 
1599 static void spapr_machine_reset(MachineState *machine)
1600 {
1601     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1602     PowerPCCPU *first_ppc_cpu;
1603     hwaddr fdt_addr;
1604     void *fdt;
1605     int rc;
1606 
1607     kvmppc_svm_off(&error_fatal);
1608     spapr_caps_apply(spapr);
1609 
1610     first_ppc_cpu = POWERPC_CPU(first_cpu);
1611     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1612         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1613                               spapr->max_compat_pvr)) {
1614         /*
1615          * If using KVM with radix mode available, VCPUs can be started
1616          * without a HPT because KVM will start them in radix mode.
1617          * Set the GR bit in PATE so that we know there is no HPT.
1618          */
1619         spapr->patb_entry = PATE1_GR;
1620         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1621     } else {
1622         spapr_setup_hpt(spapr);
1623     }
1624 
1625     qemu_devices_reset();
1626 
1627     /*
1628      * If this reset wasn't generated by CAS, we should reset our
1629      * negotiated options and start from scratch
1630      */
1631     if (!spapr->cas_reboot) {
1632         spapr_ovec_cleanup(spapr->ov5_cas);
1633         spapr->ov5_cas = spapr_ovec_new();
1634 
1635         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1636     }
1637 
1638     /*
1639      * This is fixing some of the default configuration of the XIVE
1640      * devices. To be called after the reset of the machine devices.
1641      */
1642     spapr_irq_reset(spapr, &error_fatal);
1643 
1644     /*
1645      * There is no CAS under qtest. Simulate one to please the code that
1646      * depends on spapr->ov5_cas. This is especially needed to test device
1647      * unplug, so we do that before resetting the DRCs.
1648      */
1649     if (qtest_enabled()) {
1650         spapr_ovec_cleanup(spapr->ov5_cas);
1651         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1652     }
1653 
1654     /* DRC reset may cause a device to be unplugged. This will cause troubles
1655      * if this device is used by another device (eg, a running vhost backend
1656      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1657      * situations, we reset DRCs after all devices have been reset.
1658      */
1659     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1660 
1661     spapr_clear_pending_events(spapr);
1662 
1663     /*
1664      * We place the device tree and RTAS just below either the top of the RMA,
1665      * or just below 2GB, whichever is lower, so that it can be
1666      * processed with 32-bit real mode code if necessary
1667      */
1668     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1669 
1670     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1671 
1672     rc = fdt_pack(fdt);
1673 
1674     /* Should only fail if we've built a corrupted tree */
1675     assert(rc == 0);
1676 
1677     /* Load the fdt */
1678     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1679     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1680     g_free(spapr->fdt_blob);
1681     spapr->fdt_size = fdt_totalsize(fdt);
1682     spapr->fdt_initial_size = spapr->fdt_size;
1683     spapr->fdt_blob = fdt;
1684 
1685     /* Set up the entry state */
1686     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1687     first_ppc_cpu->env.gpr[5] = 0;
1688 
1689     spapr->cas_reboot = false;
1690 
1691     spapr->fwnmi_machine_check_addr = -1;
1692     spapr->fwnmi_machine_check_interlock = -1;
1693 
1694     /* Signal all vCPUs waiting on this condition */
1695     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1696 
1697     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1698 }
1699 
1700 static void spapr_create_nvram(SpaprMachineState *spapr)
1701 {
1702     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1703     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1704 
1705     if (dinfo) {
1706         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1707                             &error_fatal);
1708     }
1709 
1710     qdev_init_nofail(dev);
1711 
1712     spapr->nvram = (struct SpaprNvram *)dev;
1713 }
1714 
1715 static void spapr_rtc_create(SpaprMachineState *spapr)
1716 {
1717     object_initialize_child(OBJECT(spapr), "rtc",
1718                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1719                             &error_fatal, NULL);
1720     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1721                               &error_fatal);
1722     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1723                               "date", &error_fatal);
1724 }
1725 
1726 /* Returns whether we want to use VGA or not */
1727 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1728 {
1729     switch (vga_interface_type) {
1730     case VGA_NONE:
1731         return false;
1732     case VGA_DEVICE:
1733         return true;
1734     case VGA_STD:
1735     case VGA_VIRTIO:
1736     case VGA_CIRRUS:
1737         return pci_vga_init(pci_bus) != NULL;
1738     default:
1739         error_setg(errp,
1740                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1741         return false;
1742     }
1743 }
1744 
1745 static int spapr_pre_load(void *opaque)
1746 {
1747     int rc;
1748 
1749     rc = spapr_caps_pre_load(opaque);
1750     if (rc) {
1751         return rc;
1752     }
1753 
1754     return 0;
1755 }
1756 
1757 static int spapr_post_load(void *opaque, int version_id)
1758 {
1759     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1760     int err = 0;
1761 
1762     err = spapr_caps_post_migration(spapr);
1763     if (err) {
1764         return err;
1765     }
1766 
1767     /*
1768      * In earlier versions, there was no separate qdev for the PAPR
1769      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1770      * So when migrating from those versions, poke the incoming offset
1771      * value into the RTC device
1772      */
1773     if (version_id < 3) {
1774         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1775         if (err) {
1776             return err;
1777         }
1778     }
1779 
1780     if (kvm_enabled() && spapr->patb_entry) {
1781         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1782         bool radix = !!(spapr->patb_entry & PATE1_GR);
1783         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1784 
1785         /*
1786          * Update LPCR:HR and UPRT as they may not be set properly in
1787          * the stream
1788          */
1789         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1790                             LPCR_HR | LPCR_UPRT);
1791 
1792         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1793         if (err) {
1794             error_report("Process table config unsupported by the host");
1795             return -EINVAL;
1796         }
1797     }
1798 
1799     err = spapr_irq_post_load(spapr, version_id);
1800     if (err) {
1801         return err;
1802     }
1803 
1804     return err;
1805 }
1806 
1807 static int spapr_pre_save(void *opaque)
1808 {
1809     int rc;
1810 
1811     rc = spapr_caps_pre_save(opaque);
1812     if (rc) {
1813         return rc;
1814     }
1815 
1816     return 0;
1817 }
1818 
1819 static bool version_before_3(void *opaque, int version_id)
1820 {
1821     return version_id < 3;
1822 }
1823 
1824 static bool spapr_pending_events_needed(void *opaque)
1825 {
1826     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1827     return !QTAILQ_EMPTY(&spapr->pending_events);
1828 }
1829 
1830 static const VMStateDescription vmstate_spapr_event_entry = {
1831     .name = "spapr_event_log_entry",
1832     .version_id = 1,
1833     .minimum_version_id = 1,
1834     .fields = (VMStateField[]) {
1835         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1836         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1837         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1838                                      NULL, extended_length),
1839         VMSTATE_END_OF_LIST()
1840     },
1841 };
1842 
1843 static const VMStateDescription vmstate_spapr_pending_events = {
1844     .name = "spapr_pending_events",
1845     .version_id = 1,
1846     .minimum_version_id = 1,
1847     .needed = spapr_pending_events_needed,
1848     .fields = (VMStateField[]) {
1849         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1850                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1851         VMSTATE_END_OF_LIST()
1852     },
1853 };
1854 
1855 static bool spapr_ov5_cas_needed(void *opaque)
1856 {
1857     SpaprMachineState *spapr = opaque;
1858     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1859     bool cas_needed;
1860 
1861     /* Prior to the introduction of SpaprOptionVector, we had two option
1862      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1863      * Both of these options encode machine topology into the device-tree
1864      * in such a way that the now-booted OS should still be able to interact
1865      * appropriately with QEMU regardless of what options were actually
1866      * negotiatied on the source side.
1867      *
1868      * As such, we can avoid migrating the CAS-negotiated options if these
1869      * are the only options available on the current machine/platform.
1870      * Since these are the only options available for pseries-2.7 and
1871      * earlier, this allows us to maintain old->new/new->old migration
1872      * compatibility.
1873      *
1874      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1875      * via default pseries-2.8 machines and explicit command-line parameters.
1876      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1877      * of the actual CAS-negotiated values to continue working properly. For
1878      * example, availability of memory unplug depends on knowing whether
1879      * OV5_HP_EVT was negotiated via CAS.
1880      *
1881      * Thus, for any cases where the set of available CAS-negotiatable
1882      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1883      * include the CAS-negotiated options in the migration stream, unless
1884      * if they affect boot time behaviour only.
1885      */
1886     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1887     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1888     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1889 
1890     /* We need extra information if we have any bits outside the mask
1891      * defined above */
1892     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1893 
1894     spapr_ovec_cleanup(ov5_mask);
1895 
1896     return cas_needed;
1897 }
1898 
1899 static const VMStateDescription vmstate_spapr_ov5_cas = {
1900     .name = "spapr_option_vector_ov5_cas",
1901     .version_id = 1,
1902     .minimum_version_id = 1,
1903     .needed = spapr_ov5_cas_needed,
1904     .fields = (VMStateField[]) {
1905         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1906                                  vmstate_spapr_ovec, SpaprOptionVector),
1907         VMSTATE_END_OF_LIST()
1908     },
1909 };
1910 
1911 static bool spapr_patb_entry_needed(void *opaque)
1912 {
1913     SpaprMachineState *spapr = opaque;
1914 
1915     return !!spapr->patb_entry;
1916 }
1917 
1918 static const VMStateDescription vmstate_spapr_patb_entry = {
1919     .name = "spapr_patb_entry",
1920     .version_id = 1,
1921     .minimum_version_id = 1,
1922     .needed = spapr_patb_entry_needed,
1923     .fields = (VMStateField[]) {
1924         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1925         VMSTATE_END_OF_LIST()
1926     },
1927 };
1928 
1929 static bool spapr_irq_map_needed(void *opaque)
1930 {
1931     SpaprMachineState *spapr = opaque;
1932 
1933     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1934 }
1935 
1936 static const VMStateDescription vmstate_spapr_irq_map = {
1937     .name = "spapr_irq_map",
1938     .version_id = 1,
1939     .minimum_version_id = 1,
1940     .needed = spapr_irq_map_needed,
1941     .fields = (VMStateField[]) {
1942         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1943         VMSTATE_END_OF_LIST()
1944     },
1945 };
1946 
1947 static bool spapr_dtb_needed(void *opaque)
1948 {
1949     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1950 
1951     return smc->update_dt_enabled;
1952 }
1953 
1954 static int spapr_dtb_pre_load(void *opaque)
1955 {
1956     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1957 
1958     g_free(spapr->fdt_blob);
1959     spapr->fdt_blob = NULL;
1960     spapr->fdt_size = 0;
1961 
1962     return 0;
1963 }
1964 
1965 static const VMStateDescription vmstate_spapr_dtb = {
1966     .name = "spapr_dtb",
1967     .version_id = 1,
1968     .minimum_version_id = 1,
1969     .needed = spapr_dtb_needed,
1970     .pre_load = spapr_dtb_pre_load,
1971     .fields = (VMStateField[]) {
1972         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1973         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1974         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1975                                      fdt_size),
1976         VMSTATE_END_OF_LIST()
1977     },
1978 };
1979 
1980 static bool spapr_fwnmi_needed(void *opaque)
1981 {
1982     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1983 
1984     return spapr->fwnmi_machine_check_addr != -1;
1985 }
1986 
1987 static int spapr_fwnmi_pre_save(void *opaque)
1988 {
1989     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1990 
1991     /*
1992      * Check if machine check handling is in progress and print a
1993      * warning message.
1994      */
1995     if (spapr->fwnmi_machine_check_interlock != -1) {
1996         warn_report("A machine check is being handled during migration. The"
1997                 "handler may run and log hardware error on the destination");
1998     }
1999 
2000     return 0;
2001 }
2002 
2003 static const VMStateDescription vmstate_spapr_fwnmi = {
2004     .name = "spapr_fwnmi",
2005     .version_id = 1,
2006     .minimum_version_id = 1,
2007     .needed = spapr_fwnmi_needed,
2008     .pre_save = spapr_fwnmi_pre_save,
2009     .fields = (VMStateField[]) {
2010         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2011         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2012         VMSTATE_END_OF_LIST()
2013     },
2014 };
2015 
2016 static const VMStateDescription vmstate_spapr = {
2017     .name = "spapr",
2018     .version_id = 3,
2019     .minimum_version_id = 1,
2020     .pre_load = spapr_pre_load,
2021     .post_load = spapr_post_load,
2022     .pre_save = spapr_pre_save,
2023     .fields = (VMStateField[]) {
2024         /* used to be @next_irq */
2025         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2026 
2027         /* RTC offset */
2028         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2029 
2030         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2031         VMSTATE_END_OF_LIST()
2032     },
2033     .subsections = (const VMStateDescription*[]) {
2034         &vmstate_spapr_ov5_cas,
2035         &vmstate_spapr_patb_entry,
2036         &vmstate_spapr_pending_events,
2037         &vmstate_spapr_cap_htm,
2038         &vmstate_spapr_cap_vsx,
2039         &vmstate_spapr_cap_dfp,
2040         &vmstate_spapr_cap_cfpc,
2041         &vmstate_spapr_cap_sbbc,
2042         &vmstate_spapr_cap_ibs,
2043         &vmstate_spapr_cap_hpt_maxpagesize,
2044         &vmstate_spapr_irq_map,
2045         &vmstate_spapr_cap_nested_kvm_hv,
2046         &vmstate_spapr_dtb,
2047         &vmstate_spapr_cap_large_decr,
2048         &vmstate_spapr_cap_ccf_assist,
2049         &vmstate_spapr_cap_fwnmi,
2050         &vmstate_spapr_fwnmi,
2051         NULL
2052     }
2053 };
2054 
2055 static int htab_save_setup(QEMUFile *f, void *opaque)
2056 {
2057     SpaprMachineState *spapr = opaque;
2058 
2059     /* "Iteration" header */
2060     if (!spapr->htab_shift) {
2061         qemu_put_be32(f, -1);
2062     } else {
2063         qemu_put_be32(f, spapr->htab_shift);
2064     }
2065 
2066     if (spapr->htab) {
2067         spapr->htab_save_index = 0;
2068         spapr->htab_first_pass = true;
2069     } else {
2070         if (spapr->htab_shift) {
2071             assert(kvm_enabled());
2072         }
2073     }
2074 
2075 
2076     return 0;
2077 }
2078 
2079 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2080                             int chunkstart, int n_valid, int n_invalid)
2081 {
2082     qemu_put_be32(f, chunkstart);
2083     qemu_put_be16(f, n_valid);
2084     qemu_put_be16(f, n_invalid);
2085     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2086                     HASH_PTE_SIZE_64 * n_valid);
2087 }
2088 
2089 static void htab_save_end_marker(QEMUFile *f)
2090 {
2091     qemu_put_be32(f, 0);
2092     qemu_put_be16(f, 0);
2093     qemu_put_be16(f, 0);
2094 }
2095 
2096 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2097                                  int64_t max_ns)
2098 {
2099     bool has_timeout = max_ns != -1;
2100     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2101     int index = spapr->htab_save_index;
2102     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2103 
2104     assert(spapr->htab_first_pass);
2105 
2106     do {
2107         int chunkstart;
2108 
2109         /* Consume invalid HPTEs */
2110         while ((index < htabslots)
2111                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2112             CLEAN_HPTE(HPTE(spapr->htab, index));
2113             index++;
2114         }
2115 
2116         /* Consume valid HPTEs */
2117         chunkstart = index;
2118         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2119                && HPTE_VALID(HPTE(spapr->htab, index))) {
2120             CLEAN_HPTE(HPTE(spapr->htab, index));
2121             index++;
2122         }
2123 
2124         if (index > chunkstart) {
2125             int n_valid = index - chunkstart;
2126 
2127             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2128 
2129             if (has_timeout &&
2130                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2131                 break;
2132             }
2133         }
2134     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2135 
2136     if (index >= htabslots) {
2137         assert(index == htabslots);
2138         index = 0;
2139         spapr->htab_first_pass = false;
2140     }
2141     spapr->htab_save_index = index;
2142 }
2143 
2144 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2145                                 int64_t max_ns)
2146 {
2147     bool final = max_ns < 0;
2148     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2149     int examined = 0, sent = 0;
2150     int index = spapr->htab_save_index;
2151     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2152 
2153     assert(!spapr->htab_first_pass);
2154 
2155     do {
2156         int chunkstart, invalidstart;
2157 
2158         /* Consume non-dirty HPTEs */
2159         while ((index < htabslots)
2160                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2161             index++;
2162             examined++;
2163         }
2164 
2165         chunkstart = index;
2166         /* Consume valid dirty HPTEs */
2167         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2168                && HPTE_DIRTY(HPTE(spapr->htab, index))
2169                && HPTE_VALID(HPTE(spapr->htab, index))) {
2170             CLEAN_HPTE(HPTE(spapr->htab, index));
2171             index++;
2172             examined++;
2173         }
2174 
2175         invalidstart = index;
2176         /* Consume invalid dirty HPTEs */
2177         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2178                && HPTE_DIRTY(HPTE(spapr->htab, index))
2179                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2180             CLEAN_HPTE(HPTE(spapr->htab, index));
2181             index++;
2182             examined++;
2183         }
2184 
2185         if (index > chunkstart) {
2186             int n_valid = invalidstart - chunkstart;
2187             int n_invalid = index - invalidstart;
2188 
2189             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2190             sent += index - chunkstart;
2191 
2192             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2193                 break;
2194             }
2195         }
2196 
2197         if (examined >= htabslots) {
2198             break;
2199         }
2200 
2201         if (index >= htabslots) {
2202             assert(index == htabslots);
2203             index = 0;
2204         }
2205     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2206 
2207     if (index >= htabslots) {
2208         assert(index == htabslots);
2209         index = 0;
2210     }
2211 
2212     spapr->htab_save_index = index;
2213 
2214     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2215 }
2216 
2217 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2218 #define MAX_KVM_BUF_SIZE    2048
2219 
2220 static int htab_save_iterate(QEMUFile *f, void *opaque)
2221 {
2222     SpaprMachineState *spapr = opaque;
2223     int fd;
2224     int rc = 0;
2225 
2226     /* Iteration header */
2227     if (!spapr->htab_shift) {
2228         qemu_put_be32(f, -1);
2229         return 1;
2230     } else {
2231         qemu_put_be32(f, 0);
2232     }
2233 
2234     if (!spapr->htab) {
2235         assert(kvm_enabled());
2236 
2237         fd = get_htab_fd(spapr);
2238         if (fd < 0) {
2239             return fd;
2240         }
2241 
2242         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2243         if (rc < 0) {
2244             return rc;
2245         }
2246     } else  if (spapr->htab_first_pass) {
2247         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2248     } else {
2249         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2250     }
2251 
2252     htab_save_end_marker(f);
2253 
2254     return rc;
2255 }
2256 
2257 static int htab_save_complete(QEMUFile *f, void *opaque)
2258 {
2259     SpaprMachineState *spapr = opaque;
2260     int fd;
2261 
2262     /* Iteration header */
2263     if (!spapr->htab_shift) {
2264         qemu_put_be32(f, -1);
2265         return 0;
2266     } else {
2267         qemu_put_be32(f, 0);
2268     }
2269 
2270     if (!spapr->htab) {
2271         int rc;
2272 
2273         assert(kvm_enabled());
2274 
2275         fd = get_htab_fd(spapr);
2276         if (fd < 0) {
2277             return fd;
2278         }
2279 
2280         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2281         if (rc < 0) {
2282             return rc;
2283         }
2284     } else {
2285         if (spapr->htab_first_pass) {
2286             htab_save_first_pass(f, spapr, -1);
2287         }
2288         htab_save_later_pass(f, spapr, -1);
2289     }
2290 
2291     /* End marker */
2292     htab_save_end_marker(f);
2293 
2294     return 0;
2295 }
2296 
2297 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2298 {
2299     SpaprMachineState *spapr = opaque;
2300     uint32_t section_hdr;
2301     int fd = -1;
2302     Error *local_err = NULL;
2303 
2304     if (version_id < 1 || version_id > 1) {
2305         error_report("htab_load() bad version");
2306         return -EINVAL;
2307     }
2308 
2309     section_hdr = qemu_get_be32(f);
2310 
2311     if (section_hdr == -1) {
2312         spapr_free_hpt(spapr);
2313         return 0;
2314     }
2315 
2316     if (section_hdr) {
2317         /* First section gives the htab size */
2318         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2319         if (local_err) {
2320             error_report_err(local_err);
2321             return -EINVAL;
2322         }
2323         return 0;
2324     }
2325 
2326     if (!spapr->htab) {
2327         assert(kvm_enabled());
2328 
2329         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2330         if (fd < 0) {
2331             error_report_err(local_err);
2332             return fd;
2333         }
2334     }
2335 
2336     while (true) {
2337         uint32_t index;
2338         uint16_t n_valid, n_invalid;
2339 
2340         index = qemu_get_be32(f);
2341         n_valid = qemu_get_be16(f);
2342         n_invalid = qemu_get_be16(f);
2343 
2344         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2345             /* End of Stream */
2346             break;
2347         }
2348 
2349         if ((index + n_valid + n_invalid) >
2350             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2351             /* Bad index in stream */
2352             error_report(
2353                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2354                 index, n_valid, n_invalid, spapr->htab_shift);
2355             return -EINVAL;
2356         }
2357 
2358         if (spapr->htab) {
2359             if (n_valid) {
2360                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2361                                 HASH_PTE_SIZE_64 * n_valid);
2362             }
2363             if (n_invalid) {
2364                 memset(HPTE(spapr->htab, index + n_valid), 0,
2365                        HASH_PTE_SIZE_64 * n_invalid);
2366             }
2367         } else {
2368             int rc;
2369 
2370             assert(fd >= 0);
2371 
2372             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2373             if (rc < 0) {
2374                 return rc;
2375             }
2376         }
2377     }
2378 
2379     if (!spapr->htab) {
2380         assert(fd >= 0);
2381         close(fd);
2382     }
2383 
2384     return 0;
2385 }
2386 
2387 static void htab_save_cleanup(void *opaque)
2388 {
2389     SpaprMachineState *spapr = opaque;
2390 
2391     close_htab_fd(spapr);
2392 }
2393 
2394 static SaveVMHandlers savevm_htab_handlers = {
2395     .save_setup = htab_save_setup,
2396     .save_live_iterate = htab_save_iterate,
2397     .save_live_complete_precopy = htab_save_complete,
2398     .save_cleanup = htab_save_cleanup,
2399     .load_state = htab_load,
2400 };
2401 
2402 static void spapr_boot_set(void *opaque, const char *boot_device,
2403                            Error **errp)
2404 {
2405     MachineState *machine = MACHINE(opaque);
2406     machine->boot_order = g_strdup(boot_device);
2407 }
2408 
2409 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2410 {
2411     MachineState *machine = MACHINE(spapr);
2412     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2413     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2414     int i;
2415 
2416     for (i = 0; i < nr_lmbs; i++) {
2417         uint64_t addr;
2418 
2419         addr = i * lmb_size + machine->device_memory->base;
2420         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2421                                addr / lmb_size);
2422     }
2423 }
2424 
2425 /*
2426  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2427  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2428  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2429  */
2430 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2431 {
2432     int i;
2433 
2434     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2435         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2436                    " is not aligned to %" PRIu64 " MiB",
2437                    machine->ram_size,
2438                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2439         return;
2440     }
2441 
2442     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2443         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2444                    " is not aligned to %" PRIu64 " MiB",
2445                    machine->ram_size,
2446                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2447         return;
2448     }
2449 
2450     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2451         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2452             error_setg(errp,
2453                        "Node %d memory size 0x%" PRIx64
2454                        " is not aligned to %" PRIu64 " MiB",
2455                        i, machine->numa_state->nodes[i].node_mem,
2456                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2457             return;
2458         }
2459     }
2460 }
2461 
2462 /* find cpu slot in machine->possible_cpus by core_id */
2463 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2464 {
2465     int index = id / ms->smp.threads;
2466 
2467     if (index >= ms->possible_cpus->len) {
2468         return NULL;
2469     }
2470     if (idx) {
2471         *idx = index;
2472     }
2473     return &ms->possible_cpus->cpus[index];
2474 }
2475 
2476 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2477 {
2478     MachineState *ms = MACHINE(spapr);
2479     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2480     Error *local_err = NULL;
2481     bool vsmt_user = !!spapr->vsmt;
2482     int kvm_smt = kvmppc_smt_threads();
2483     int ret;
2484     unsigned int smp_threads = ms->smp.threads;
2485 
2486     if (!kvm_enabled() && (smp_threads > 1)) {
2487         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2488                      "on a pseries machine");
2489         goto out;
2490     }
2491     if (!is_power_of_2(smp_threads)) {
2492         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2493                      "machine because it must be a power of 2", smp_threads);
2494         goto out;
2495     }
2496 
2497     /* Detemine the VSMT mode to use: */
2498     if (vsmt_user) {
2499         if (spapr->vsmt < smp_threads) {
2500             error_setg(&local_err, "Cannot support VSMT mode %d"
2501                          " because it must be >= threads/core (%d)",
2502                          spapr->vsmt, smp_threads);
2503             goto out;
2504         }
2505         /* In this case, spapr->vsmt has been set by the command line */
2506     } else if (!smc->smp_threads_vsmt) {
2507         /*
2508          * Default VSMT value is tricky, because we need it to be as
2509          * consistent as possible (for migration), but this requires
2510          * changing it for at least some existing cases.  We pick 8 as
2511          * the value that we'd get with KVM on POWER8, the
2512          * overwhelmingly common case in production systems.
2513          */
2514         spapr->vsmt = MAX(8, smp_threads);
2515     } else {
2516         spapr->vsmt = smp_threads;
2517     }
2518 
2519     /* KVM: If necessary, set the SMT mode: */
2520     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2521         ret = kvmppc_set_smt_threads(spapr->vsmt);
2522         if (ret) {
2523             /* Looks like KVM isn't able to change VSMT mode */
2524             error_setg(&local_err,
2525                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2526                        spapr->vsmt, ret);
2527             /* We can live with that if the default one is big enough
2528              * for the number of threads, and a submultiple of the one
2529              * we want.  In this case we'll waste some vcpu ids, but
2530              * behaviour will be correct */
2531             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2532                 warn_report_err(local_err);
2533                 local_err = NULL;
2534                 goto out;
2535             } else {
2536                 if (!vsmt_user) {
2537                     error_append_hint(&local_err,
2538                                       "On PPC, a VM with %d threads/core"
2539                                       " on a host with %d threads/core"
2540                                       " requires the use of VSMT mode %d.\n",
2541                                       smp_threads, kvm_smt, spapr->vsmt);
2542                 }
2543                 kvmppc_error_append_smt_possible_hint(&local_err);
2544                 goto out;
2545             }
2546         }
2547     }
2548     /* else TCG: nothing to do currently */
2549 out:
2550     error_propagate(errp, local_err);
2551 }
2552 
2553 static void spapr_init_cpus(SpaprMachineState *spapr)
2554 {
2555     MachineState *machine = MACHINE(spapr);
2556     MachineClass *mc = MACHINE_GET_CLASS(machine);
2557     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2558     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2559     const CPUArchIdList *possible_cpus;
2560     unsigned int smp_cpus = machine->smp.cpus;
2561     unsigned int smp_threads = machine->smp.threads;
2562     unsigned int max_cpus = machine->smp.max_cpus;
2563     int boot_cores_nr = smp_cpus / smp_threads;
2564     int i;
2565 
2566     possible_cpus = mc->possible_cpu_arch_ids(machine);
2567     if (mc->has_hotpluggable_cpus) {
2568         if (smp_cpus % smp_threads) {
2569             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2570                          smp_cpus, smp_threads);
2571             exit(1);
2572         }
2573         if (max_cpus % smp_threads) {
2574             error_report("max_cpus (%u) must be multiple of threads (%u)",
2575                          max_cpus, smp_threads);
2576             exit(1);
2577         }
2578     } else {
2579         if (max_cpus != smp_cpus) {
2580             error_report("This machine version does not support CPU hotplug");
2581             exit(1);
2582         }
2583         boot_cores_nr = possible_cpus->len;
2584     }
2585 
2586     if (smc->pre_2_10_has_unused_icps) {
2587         int i;
2588 
2589         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2590             /* Dummy entries get deregistered when real ICPState objects
2591              * are registered during CPU core hotplug.
2592              */
2593             pre_2_10_vmstate_register_dummy_icp(i);
2594         }
2595     }
2596 
2597     for (i = 0; i < possible_cpus->len; i++) {
2598         int core_id = i * smp_threads;
2599 
2600         if (mc->has_hotpluggable_cpus) {
2601             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2602                                    spapr_vcpu_id(spapr, core_id));
2603         }
2604 
2605         if (i < boot_cores_nr) {
2606             Object *core  = object_new(type);
2607             int nr_threads = smp_threads;
2608 
2609             /* Handle the partially filled core for older machine types */
2610             if ((i + 1) * smp_threads >= smp_cpus) {
2611                 nr_threads = smp_cpus - i * smp_threads;
2612             }
2613 
2614             object_property_set_int(core, nr_threads, "nr-threads",
2615                                     &error_fatal);
2616             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2617                                     &error_fatal);
2618             object_property_set_bool(core, true, "realized", &error_fatal);
2619 
2620             object_unref(core);
2621         }
2622     }
2623 }
2624 
2625 static PCIHostState *spapr_create_default_phb(void)
2626 {
2627     DeviceState *dev;
2628 
2629     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2630     qdev_prop_set_uint32(dev, "index", 0);
2631     qdev_init_nofail(dev);
2632 
2633     return PCI_HOST_BRIDGE(dev);
2634 }
2635 
2636 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2637 {
2638     MachineState *machine = MACHINE(spapr);
2639     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2640     hwaddr rma_size = machine->ram_size;
2641     hwaddr node0_size = spapr_node0_size(machine);
2642 
2643     /* RMA has to fit in the first NUMA node */
2644     rma_size = MIN(rma_size, node0_size);
2645 
2646     /*
2647      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2648      * never exceed that
2649      */
2650     rma_size = MIN(rma_size, 1 * TiB);
2651 
2652     /*
2653      * Clamp the RMA size based on machine type.  This is for
2654      * migration compatibility with older qemu versions, which limited
2655      * the RMA size for complicated and mostly bad reasons.
2656      */
2657     if (smc->rma_limit) {
2658         rma_size = MIN(rma_size, smc->rma_limit);
2659     }
2660 
2661     if (rma_size < MIN_RMA_SLOF) {
2662         error_setg(errp,
2663                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2664                    "ldMiB guest RMA (Real Mode Area memory)",
2665                    MIN_RMA_SLOF / MiB);
2666         return 0;
2667     }
2668 
2669     return rma_size;
2670 }
2671 
2672 /* pSeries LPAR / sPAPR hardware init */
2673 static void spapr_machine_init(MachineState *machine)
2674 {
2675     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2676     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2677     MachineClass *mc = MACHINE_GET_CLASS(machine);
2678     const char *kernel_filename = machine->kernel_filename;
2679     const char *initrd_filename = machine->initrd_filename;
2680     PCIHostState *phb;
2681     int i;
2682     MemoryRegion *sysmem = get_system_memory();
2683     long load_limit, fw_size;
2684     char *filename;
2685     Error *resize_hpt_err = NULL;
2686 
2687     msi_nonbroken = true;
2688 
2689     QLIST_INIT(&spapr->phbs);
2690     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2691 
2692     /* Determine capabilities to run with */
2693     spapr_caps_init(spapr);
2694 
2695     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2696     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2697         /*
2698          * If the user explicitly requested a mode we should either
2699          * supply it, or fail completely (which we do below).  But if
2700          * it's not set explicitly, we reset our mode to something
2701          * that works
2702          */
2703         if (resize_hpt_err) {
2704             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2705             error_free(resize_hpt_err);
2706             resize_hpt_err = NULL;
2707         } else {
2708             spapr->resize_hpt = smc->resize_hpt_default;
2709         }
2710     }
2711 
2712     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2713 
2714     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2715         /*
2716          * User requested HPT resize, but this host can't supply it.  Bail out
2717          */
2718         error_report_err(resize_hpt_err);
2719         exit(1);
2720     }
2721 
2722     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2723 
2724     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2725     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2726 
2727     /*
2728      * VSMT must be set in order to be able to compute VCPU ids, ie to
2729      * call spapr_max_server_number() or spapr_vcpu_id().
2730      */
2731     spapr_set_vsmt_mode(spapr, &error_fatal);
2732 
2733     /* Set up Interrupt Controller before we create the VCPUs */
2734     spapr_irq_init(spapr, &error_fatal);
2735 
2736     /* Set up containers for ibm,client-architecture-support negotiated options
2737      */
2738     spapr->ov5 = spapr_ovec_new();
2739     spapr->ov5_cas = spapr_ovec_new();
2740 
2741     if (smc->dr_lmb_enabled) {
2742         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2743         spapr_validate_node_memory(machine, &error_fatal);
2744     }
2745 
2746     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2747 
2748     /* advertise support for dedicated HP event source to guests */
2749     if (spapr->use_hotplug_event_source) {
2750         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2751     }
2752 
2753     /* advertise support for HPT resizing */
2754     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2755         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2756     }
2757 
2758     /* advertise support for ibm,dyamic-memory-v2 */
2759     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2760 
2761     /* advertise XIVE on POWER9 machines */
2762     if (spapr->irq->xive) {
2763         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2764     }
2765 
2766     /* init CPUs */
2767     spapr_init_cpus(spapr);
2768 
2769     /*
2770      * check we don't have a memory-less/cpu-less NUMA node
2771      * Firmware relies on the existing memory/cpu topology to provide the
2772      * NUMA topology to the kernel.
2773      * And the linux kernel needs to know the NUMA topology at start
2774      * to be able to hotplug CPUs later.
2775      */
2776     if (machine->numa_state->num_nodes) {
2777         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2778             /* check for memory-less node */
2779             if (machine->numa_state->nodes[i].node_mem == 0) {
2780                 CPUState *cs;
2781                 int found = 0;
2782                 /* check for cpu-less node */
2783                 CPU_FOREACH(cs) {
2784                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2785                     if (cpu->node_id == i) {
2786                         found = 1;
2787                         break;
2788                     }
2789                 }
2790                 /* memory-less and cpu-less node */
2791                 if (!found) {
2792                     error_report(
2793                        "Memory-less/cpu-less nodes are not supported (node %d)",
2794                                  i);
2795                     exit(1);
2796                 }
2797             }
2798         }
2799 
2800     }
2801 
2802     /*
2803      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2804      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2805      * called from vPHB reset handler so we initialize the counter here.
2806      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2807      * must be equally distant from any other node.
2808      * The final value of spapr->gpu_numa_id is going to be written to
2809      * max-associativity-domains in spapr_build_fdt().
2810      */
2811     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2812 
2813     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2814         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2815                               spapr->max_compat_pvr)) {
2816         /* KVM and TCG always allow GTSE with radix... */
2817         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2818     }
2819     /* ... but not with hash (currently). */
2820 
2821     if (kvm_enabled()) {
2822         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2823         kvmppc_enable_logical_ci_hcalls();
2824         kvmppc_enable_set_mode_hcall();
2825 
2826         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2827         kvmppc_enable_clear_ref_mod_hcalls();
2828 
2829         /* Enable H_PAGE_INIT */
2830         kvmppc_enable_h_page_init();
2831     }
2832 
2833     /* map RAM */
2834     memory_region_add_subregion(sysmem, 0, machine->ram);
2835 
2836     /* always allocate the device memory information */
2837     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2838 
2839     /* initialize hotplug memory address space */
2840     if (machine->ram_size < machine->maxram_size) {
2841         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2842         /*
2843          * Limit the number of hotpluggable memory slots to half the number
2844          * slots that KVM supports, leaving the other half for PCI and other
2845          * devices. However ensure that number of slots doesn't drop below 32.
2846          */
2847         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2848                            SPAPR_MAX_RAM_SLOTS;
2849 
2850         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2851             max_memslots = SPAPR_MAX_RAM_SLOTS;
2852         }
2853         if (machine->ram_slots > max_memslots) {
2854             error_report("Specified number of memory slots %"
2855                          PRIu64" exceeds max supported %d",
2856                          machine->ram_slots, max_memslots);
2857             exit(1);
2858         }
2859 
2860         machine->device_memory->base = ROUND_UP(machine->ram_size,
2861                                                 SPAPR_DEVICE_MEM_ALIGN);
2862         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2863                            "device-memory", device_mem_size);
2864         memory_region_add_subregion(sysmem, machine->device_memory->base,
2865                                     &machine->device_memory->mr);
2866     }
2867 
2868     if (smc->dr_lmb_enabled) {
2869         spapr_create_lmb_dr_connectors(spapr);
2870     }
2871 
2872     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2873         /* Create the error string for live migration blocker */
2874         error_setg(&spapr->fwnmi_migration_blocker,
2875             "A machine check is being handled during migration. The handler"
2876             "may run and log hardware error on the destination");
2877     }
2878 
2879     if (mc->nvdimm_supported) {
2880         spapr_create_nvdimm_dr_connectors(spapr);
2881     }
2882 
2883     /* Set up RTAS event infrastructure */
2884     spapr_events_init(spapr);
2885 
2886     /* Set up the RTC RTAS interfaces */
2887     spapr_rtc_create(spapr);
2888 
2889     /* Set up VIO bus */
2890     spapr->vio_bus = spapr_vio_bus_init();
2891 
2892     for (i = 0; i < serial_max_hds(); i++) {
2893         if (serial_hd(i)) {
2894             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2895         }
2896     }
2897 
2898     /* We always have at least the nvram device on VIO */
2899     spapr_create_nvram(spapr);
2900 
2901     /*
2902      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2903      * connectors (described in root DT node's "ibm,drc-types" property)
2904      * are pre-initialized here. additional child connectors (such as
2905      * connectors for a PHBs PCI slots) are added as needed during their
2906      * parent's realization.
2907      */
2908     if (smc->dr_phb_enabled) {
2909         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2910             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2911         }
2912     }
2913 
2914     /* Set up PCI */
2915     spapr_pci_rtas_init();
2916 
2917     phb = spapr_create_default_phb();
2918 
2919     for (i = 0; i < nb_nics; i++) {
2920         NICInfo *nd = &nd_table[i];
2921 
2922         if (!nd->model) {
2923             nd->model = g_strdup("spapr-vlan");
2924         }
2925 
2926         if (g_str_equal(nd->model, "spapr-vlan") ||
2927             g_str_equal(nd->model, "ibmveth")) {
2928             spapr_vlan_create(spapr->vio_bus, nd);
2929         } else {
2930             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2931         }
2932     }
2933 
2934     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2935         spapr_vscsi_create(spapr->vio_bus);
2936     }
2937 
2938     /* Graphics */
2939     if (spapr_vga_init(phb->bus, &error_fatal)) {
2940         spapr->has_graphics = true;
2941         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2942     }
2943 
2944     if (machine->usb) {
2945         if (smc->use_ohci_by_default) {
2946             pci_create_simple(phb->bus, -1, "pci-ohci");
2947         } else {
2948             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2949         }
2950 
2951         if (spapr->has_graphics) {
2952             USBBus *usb_bus = usb_bus_find(-1);
2953 
2954             usb_create_simple(usb_bus, "usb-kbd");
2955             usb_create_simple(usb_bus, "usb-mouse");
2956         }
2957     }
2958 
2959     if (kernel_filename) {
2960         uint64_t lowaddr = 0;
2961 
2962         spapr->kernel_size = load_elf(kernel_filename, NULL,
2963                                       translate_kernel_address, spapr,
2964                                       NULL, &lowaddr, NULL, NULL, 1,
2965                                       PPC_ELF_MACHINE, 0, 0);
2966         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2967             spapr->kernel_size = load_elf(kernel_filename, NULL,
2968                                           translate_kernel_address, spapr, NULL,
2969                                           &lowaddr, NULL, NULL, 0,
2970                                           PPC_ELF_MACHINE,
2971                                           0, 0);
2972             spapr->kernel_le = spapr->kernel_size > 0;
2973         }
2974         if (spapr->kernel_size < 0) {
2975             error_report("error loading %s: %s", kernel_filename,
2976                          load_elf_strerror(spapr->kernel_size));
2977             exit(1);
2978         }
2979 
2980         /* load initrd */
2981         if (initrd_filename) {
2982             /* Try to locate the initrd in the gap between the kernel
2983              * and the firmware. Add a bit of space just in case
2984              */
2985             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2986                                   + 0x1ffff) & ~0xffff;
2987             spapr->initrd_size = load_image_targphys(initrd_filename,
2988                                                      spapr->initrd_base,
2989                                                      load_limit
2990                                                      - spapr->initrd_base);
2991             if (spapr->initrd_size < 0) {
2992                 error_report("could not load initial ram disk '%s'",
2993                              initrd_filename);
2994                 exit(1);
2995             }
2996         }
2997     }
2998 
2999     if (bios_name == NULL) {
3000         bios_name = FW_FILE_NAME;
3001     }
3002     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3003     if (!filename) {
3004         error_report("Could not find LPAR firmware '%s'", bios_name);
3005         exit(1);
3006     }
3007     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3008     if (fw_size <= 0) {
3009         error_report("Could not load LPAR firmware '%s'", filename);
3010         exit(1);
3011     }
3012     g_free(filename);
3013 
3014     /* FIXME: Should register things through the MachineState's qdev
3015      * interface, this is a legacy from the sPAPREnvironment structure
3016      * which predated MachineState but had a similar function */
3017     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3018     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3019                          &savevm_htab_handlers, spapr);
3020 
3021     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3022                              &error_fatal);
3023 
3024     qemu_register_boot_set(spapr_boot_set, spapr);
3025 
3026     /*
3027      * Nothing needs to be done to resume a suspended guest because
3028      * suspending does not change the machine state, so no need for
3029      * a ->wakeup method.
3030      */
3031     qemu_register_wakeup_support();
3032 
3033     if (kvm_enabled()) {
3034         /* to stop and start vmclock */
3035         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3036                                          &spapr->tb);
3037 
3038         kvmppc_spapr_enable_inkernel_multitce();
3039     }
3040 
3041     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3042 }
3043 
3044 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3045 {
3046     if (!vm_type) {
3047         return 0;
3048     }
3049 
3050     if (!strcmp(vm_type, "HV")) {
3051         return 1;
3052     }
3053 
3054     if (!strcmp(vm_type, "PR")) {
3055         return 2;
3056     }
3057 
3058     error_report("Unknown kvm-type specified '%s'", vm_type);
3059     exit(1);
3060 }
3061 
3062 /*
3063  * Implementation of an interface to adjust firmware path
3064  * for the bootindex property handling.
3065  */
3066 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3067                                    DeviceState *dev)
3068 {
3069 #define CAST(type, obj, name) \
3070     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3071     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3072     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3073     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3074 
3075     if (d) {
3076         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3077         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3078         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3079 
3080         if (spapr) {
3081             /*
3082              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3083              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3084              * 0x8000 | (target << 8) | (bus << 5) | lun
3085              * (see the "Logical unit addressing format" table in SAM5)
3086              */
3087             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3088             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3089                                    (uint64_t)id << 48);
3090         } else if (virtio) {
3091             /*
3092              * We use SRP luns of the form 01000000 | (target << 8) | lun
3093              * in the top 32 bits of the 64-bit LUN
3094              * Note: the quote above is from SLOF and it is wrong,
3095              * the actual binding is:
3096              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3097              */
3098             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3099             if (d->lun >= 256) {
3100                 /* Use the LUN "flat space addressing method" */
3101                 id |= 0x4000;
3102             }
3103             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3104                                    (uint64_t)id << 32);
3105         } else if (usb) {
3106             /*
3107              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3108              * in the top 32 bits of the 64-bit LUN
3109              */
3110             unsigned usb_port = atoi(usb->port->path);
3111             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3112             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3113                                    (uint64_t)id << 32);
3114         }
3115     }
3116 
3117     /*
3118      * SLOF probes the USB devices, and if it recognizes that the device is a
3119      * storage device, it changes its name to "storage" instead of "usb-host",
3120      * and additionally adds a child node for the SCSI LUN, so the correct
3121      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3122      */
3123     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3124         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3125         if (usb_host_dev_is_scsi_storage(usbdev)) {
3126             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3127         }
3128     }
3129 
3130     if (phb) {
3131         /* Replace "pci" with "pci@800000020000000" */
3132         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3133     }
3134 
3135     if (vsc) {
3136         /* Same logic as virtio above */
3137         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3138         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3139     }
3140 
3141     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3142         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3143         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3144         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3145     }
3146 
3147     return NULL;
3148 }
3149 
3150 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3151 {
3152     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3153 
3154     return g_strdup(spapr->kvm_type);
3155 }
3156 
3157 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     g_free(spapr->kvm_type);
3162     spapr->kvm_type = g_strdup(value);
3163 }
3164 
3165 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3166 {
3167     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3168 
3169     return spapr->use_hotplug_event_source;
3170 }
3171 
3172 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3173                                             Error **errp)
3174 {
3175     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3176 
3177     spapr->use_hotplug_event_source = value;
3178 }
3179 
3180 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3181 {
3182     return true;
3183 }
3184 
3185 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3186 {
3187     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3188 
3189     switch (spapr->resize_hpt) {
3190     case SPAPR_RESIZE_HPT_DEFAULT:
3191         return g_strdup("default");
3192     case SPAPR_RESIZE_HPT_DISABLED:
3193         return g_strdup("disabled");
3194     case SPAPR_RESIZE_HPT_ENABLED:
3195         return g_strdup("enabled");
3196     case SPAPR_RESIZE_HPT_REQUIRED:
3197         return g_strdup("required");
3198     }
3199     g_assert_not_reached();
3200 }
3201 
3202 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3203 {
3204     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3205 
3206     if (strcmp(value, "default") == 0) {
3207         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3208     } else if (strcmp(value, "disabled") == 0) {
3209         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3210     } else if (strcmp(value, "enabled") == 0) {
3211         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3212     } else if (strcmp(value, "required") == 0) {
3213         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3214     } else {
3215         error_setg(errp, "Bad value for \"resize-hpt\" property");
3216     }
3217 }
3218 
3219 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3220                                    void *opaque, Error **errp)
3221 {
3222     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3223 }
3224 
3225 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3226                                    void *opaque, Error **errp)
3227 {
3228     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3229 }
3230 
3231 static void spapr_get_kernel_addr(Object *obj, Visitor *v, const char *name,
3232                                   void *opaque, Error **errp)
3233 {
3234     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3235 }
3236 
3237 static void spapr_set_kernel_addr(Object *obj, Visitor *v, const char *name,
3238                                   void *opaque, Error **errp)
3239 {
3240     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3241 }
3242 
3243 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3244 {
3245     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3246 
3247     if (spapr->irq == &spapr_irq_xics_legacy) {
3248         return g_strdup("legacy");
3249     } else if (spapr->irq == &spapr_irq_xics) {
3250         return g_strdup("xics");
3251     } else if (spapr->irq == &spapr_irq_xive) {
3252         return g_strdup("xive");
3253     } else if (spapr->irq == &spapr_irq_dual) {
3254         return g_strdup("dual");
3255     }
3256     g_assert_not_reached();
3257 }
3258 
3259 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3260 {
3261     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3262 
3263     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3264         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3265         return;
3266     }
3267 
3268     /* The legacy IRQ backend can not be set */
3269     if (strcmp(value, "xics") == 0) {
3270         spapr->irq = &spapr_irq_xics;
3271     } else if (strcmp(value, "xive") == 0) {
3272         spapr->irq = &spapr_irq_xive;
3273     } else if (strcmp(value, "dual") == 0) {
3274         spapr->irq = &spapr_irq_dual;
3275     } else {
3276         error_setg(errp, "Bad value for \"ic-mode\" property");
3277     }
3278 }
3279 
3280 static char *spapr_get_host_model(Object *obj, Error **errp)
3281 {
3282     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3283 
3284     return g_strdup(spapr->host_model);
3285 }
3286 
3287 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3288 {
3289     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3290 
3291     g_free(spapr->host_model);
3292     spapr->host_model = g_strdup(value);
3293 }
3294 
3295 static char *spapr_get_host_serial(Object *obj, Error **errp)
3296 {
3297     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3298 
3299     return g_strdup(spapr->host_serial);
3300 }
3301 
3302 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3303 {
3304     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3305 
3306     g_free(spapr->host_serial);
3307     spapr->host_serial = g_strdup(value);
3308 }
3309 
3310 static void spapr_instance_init(Object *obj)
3311 {
3312     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3313     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3314 
3315     spapr->htab_fd = -1;
3316     spapr->use_hotplug_event_source = true;
3317     object_property_add_str(obj, "kvm-type",
3318                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3319     object_property_set_description(obj, "kvm-type",
3320                                     "Specifies the KVM virtualization mode (HV, PR)",
3321                                     NULL);
3322     object_property_add_bool(obj, "modern-hotplug-events",
3323                             spapr_get_modern_hotplug_events,
3324                             spapr_set_modern_hotplug_events,
3325                             NULL);
3326     object_property_set_description(obj, "modern-hotplug-events",
3327                                     "Use dedicated hotplug event mechanism in"
3328                                     " place of standard EPOW events when possible"
3329                                     " (required for memory hot-unplug support)",
3330                                     NULL);
3331     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3332                             "Maximum permitted CPU compatibility mode",
3333                             &error_fatal);
3334 
3335     object_property_add_str(obj, "resize-hpt",
3336                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3337     object_property_set_description(obj, "resize-hpt",
3338                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3339                                     NULL);
3340     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3341                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3342     object_property_set_description(obj, "vsmt",
3343                                     "Virtual SMT: KVM behaves as if this were"
3344                                     " the host's SMT mode", &error_abort);
3345     object_property_add_bool(obj, "vfio-no-msix-emulation",
3346                              spapr_get_msix_emulation, NULL, NULL);
3347 
3348     object_property_add(obj, "kernel-addr", "uint64", spapr_get_kernel_addr,
3349                         spapr_set_kernel_addr, NULL, &spapr->kernel_addr,
3350                         &error_abort);
3351     object_property_set_description(obj, "kernel-addr",
3352                                     stringify(KERNEL_LOAD_ADDR)
3353                                     " for -kernel is the default",
3354                                     NULL);
3355     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3356     /* The machine class defines the default interrupt controller mode */
3357     spapr->irq = smc->irq;
3358     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3359                             spapr_set_ic_mode, NULL);
3360     object_property_set_description(obj, "ic-mode",
3361                  "Specifies the interrupt controller mode (xics, xive, dual)",
3362                  NULL);
3363 
3364     object_property_add_str(obj, "host-model",
3365         spapr_get_host_model, spapr_set_host_model,
3366         &error_abort);
3367     object_property_set_description(obj, "host-model",
3368         "Host model to advertise in guest device tree", &error_abort);
3369     object_property_add_str(obj, "host-serial",
3370         spapr_get_host_serial, spapr_set_host_serial,
3371         &error_abort);
3372     object_property_set_description(obj, "host-serial",
3373         "Host serial number to advertise in guest device tree", &error_abort);
3374 }
3375 
3376 static void spapr_machine_finalizefn(Object *obj)
3377 {
3378     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3379 
3380     g_free(spapr->kvm_type);
3381 }
3382 
3383 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3384 {
3385     cpu_synchronize_state(cs);
3386     ppc_cpu_do_system_reset(cs);
3387 }
3388 
3389 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3390 {
3391     CPUState *cs;
3392 
3393     CPU_FOREACH(cs) {
3394         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3395     }
3396 }
3397 
3398 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3399                           void *fdt, int *fdt_start_offset, Error **errp)
3400 {
3401     uint64_t addr;
3402     uint32_t node;
3403 
3404     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3405     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3406                                     &error_abort);
3407     *fdt_start_offset = spapr_dt_memory_node(fdt, node, addr,
3408                                              SPAPR_MEMORY_BLOCK_SIZE);
3409     return 0;
3410 }
3411 
3412 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3413                            bool dedicated_hp_event_source, Error **errp)
3414 {
3415     SpaprDrc *drc;
3416     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3417     int i;
3418     uint64_t addr = addr_start;
3419     bool hotplugged = spapr_drc_hotplugged(dev);
3420     Error *local_err = NULL;
3421 
3422     for (i = 0; i < nr_lmbs; i++) {
3423         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3424                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3425         g_assert(drc);
3426 
3427         spapr_drc_attach(drc, dev, &local_err);
3428         if (local_err) {
3429             while (addr > addr_start) {
3430                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3431                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3432                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3433                 spapr_drc_detach(drc);
3434             }
3435             error_propagate(errp, local_err);
3436             return;
3437         }
3438         if (!hotplugged) {
3439             spapr_drc_reset(drc);
3440         }
3441         addr += SPAPR_MEMORY_BLOCK_SIZE;
3442     }
3443     /* send hotplug notification to the
3444      * guest only in case of hotplugged memory
3445      */
3446     if (hotplugged) {
3447         if (dedicated_hp_event_source) {
3448             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3449                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3450             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3451                                                    nr_lmbs,
3452                                                    spapr_drc_index(drc));
3453         } else {
3454             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3455                                            nr_lmbs);
3456         }
3457     }
3458 }
3459 
3460 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3461                               Error **errp)
3462 {
3463     Error *local_err = NULL;
3464     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3465     PCDIMMDevice *dimm = PC_DIMM(dev);
3466     uint64_t size, addr, slot;
3467     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3468 
3469     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3470 
3471     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3472     if (local_err) {
3473         goto out;
3474     }
3475 
3476     if (!is_nvdimm) {
3477         addr = object_property_get_uint(OBJECT(dimm),
3478                                         PC_DIMM_ADDR_PROP, &local_err);
3479         if (local_err) {
3480             goto out_unplug;
3481         }
3482         spapr_add_lmbs(dev, addr, size,
3483                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3484                        &local_err);
3485     } else {
3486         slot = object_property_get_uint(OBJECT(dimm),
3487                                         PC_DIMM_SLOT_PROP, &local_err);
3488         if (local_err) {
3489             goto out_unplug;
3490         }
3491         spapr_add_nvdimm(dev, slot, &local_err);
3492     }
3493 
3494     if (local_err) {
3495         goto out_unplug;
3496     }
3497 
3498     return;
3499 
3500 out_unplug:
3501     pc_dimm_unplug(dimm, MACHINE(ms));
3502 out:
3503     error_propagate(errp, local_err);
3504 }
3505 
3506 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3507                                   Error **errp)
3508 {
3509     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3510     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3511     const MachineClass *mc = MACHINE_CLASS(smc);
3512     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3513     PCDIMMDevice *dimm = PC_DIMM(dev);
3514     Error *local_err = NULL;
3515     uint64_t size;
3516     Object *memdev;
3517     hwaddr pagesize;
3518 
3519     if (!smc->dr_lmb_enabled) {
3520         error_setg(errp, "Memory hotplug not supported for this machine");
3521         return;
3522     }
3523 
3524     if (is_nvdimm && !mc->nvdimm_supported) {
3525         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3526         return;
3527     }
3528 
3529     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3530     if (local_err) {
3531         error_propagate(errp, local_err);
3532         return;
3533     }
3534 
3535     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3536         error_setg(errp, "Hotplugged memory size must be a multiple of "
3537                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3538         return;
3539     } else if (is_nvdimm) {
3540         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3541         if (local_err) {
3542             error_propagate(errp, local_err);
3543             return;
3544         }
3545     }
3546 
3547     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3548                                       &error_abort);
3549     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3550     spapr_check_pagesize(spapr, pagesize, &local_err);
3551     if (local_err) {
3552         error_propagate(errp, local_err);
3553         return;
3554     }
3555 
3556     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3557 }
3558 
3559 struct SpaprDimmState {
3560     PCDIMMDevice *dimm;
3561     uint32_t nr_lmbs;
3562     QTAILQ_ENTRY(SpaprDimmState) next;
3563 };
3564 
3565 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3566                                                        PCDIMMDevice *dimm)
3567 {
3568     SpaprDimmState *dimm_state = NULL;
3569 
3570     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3571         if (dimm_state->dimm == dimm) {
3572             break;
3573         }
3574     }
3575     return dimm_state;
3576 }
3577 
3578 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3579                                                       uint32_t nr_lmbs,
3580                                                       PCDIMMDevice *dimm)
3581 {
3582     SpaprDimmState *ds = NULL;
3583 
3584     /*
3585      * If this request is for a DIMM whose removal had failed earlier
3586      * (due to guest's refusal to remove the LMBs), we would have this
3587      * dimm already in the pending_dimm_unplugs list. In that
3588      * case don't add again.
3589      */
3590     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3591     if (!ds) {
3592         ds = g_malloc0(sizeof(SpaprDimmState));
3593         ds->nr_lmbs = nr_lmbs;
3594         ds->dimm = dimm;
3595         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3596     }
3597     return ds;
3598 }
3599 
3600 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3601                                               SpaprDimmState *dimm_state)
3602 {
3603     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3604     g_free(dimm_state);
3605 }
3606 
3607 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3608                                                         PCDIMMDevice *dimm)
3609 {
3610     SpaprDrc *drc;
3611     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3612                                                   &error_abort);
3613     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3614     uint32_t avail_lmbs = 0;
3615     uint64_t addr_start, addr;
3616     int i;
3617 
3618     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3619                                          &error_abort);
3620 
3621     addr = addr_start;
3622     for (i = 0; i < nr_lmbs; i++) {
3623         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3624                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3625         g_assert(drc);
3626         if (drc->dev) {
3627             avail_lmbs++;
3628         }
3629         addr += SPAPR_MEMORY_BLOCK_SIZE;
3630     }
3631 
3632     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3633 }
3634 
3635 /* Callback to be called during DRC release. */
3636 void spapr_lmb_release(DeviceState *dev)
3637 {
3638     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3639     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3640     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3641 
3642     /* This information will get lost if a migration occurs
3643      * during the unplug process. In this case recover it. */
3644     if (ds == NULL) {
3645         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3646         g_assert(ds);
3647         /* The DRC being examined by the caller at least must be counted */
3648         g_assert(ds->nr_lmbs);
3649     }
3650 
3651     if (--ds->nr_lmbs) {
3652         return;
3653     }
3654 
3655     /*
3656      * Now that all the LMBs have been removed by the guest, call the
3657      * unplug handler chain. This can never fail.
3658      */
3659     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3660     object_unparent(OBJECT(dev));
3661 }
3662 
3663 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3664 {
3665     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3666     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3667 
3668     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3669     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3670     spapr_pending_dimm_unplugs_remove(spapr, ds);
3671 }
3672 
3673 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3674                                         DeviceState *dev, Error **errp)
3675 {
3676     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3677     Error *local_err = NULL;
3678     PCDIMMDevice *dimm = PC_DIMM(dev);
3679     uint32_t nr_lmbs;
3680     uint64_t size, addr_start, addr;
3681     int i;
3682     SpaprDrc *drc;
3683 
3684     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3685         error_setg(&local_err,
3686                    "nvdimm device hot unplug is not supported yet.");
3687         goto out;
3688     }
3689 
3690     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3691     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3692 
3693     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3694                                          &local_err);
3695     if (local_err) {
3696         goto out;
3697     }
3698 
3699     /*
3700      * An existing pending dimm state for this DIMM means that there is an
3701      * unplug operation in progress, waiting for the spapr_lmb_release
3702      * callback to complete the job (BQL can't cover that far). In this case,
3703      * bail out to avoid detaching DRCs that were already released.
3704      */
3705     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3706         error_setg(&local_err,
3707                    "Memory unplug already in progress for device %s",
3708                    dev->id);
3709         goto out;
3710     }
3711 
3712     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3713 
3714     addr = addr_start;
3715     for (i = 0; i < nr_lmbs; i++) {
3716         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3717                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3718         g_assert(drc);
3719 
3720         spapr_drc_detach(drc);
3721         addr += SPAPR_MEMORY_BLOCK_SIZE;
3722     }
3723 
3724     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3725                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3726     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3727                                               nr_lmbs, spapr_drc_index(drc));
3728 out:
3729     error_propagate(errp, local_err);
3730 }
3731 
3732 /* Callback to be called during DRC release. */
3733 void spapr_core_release(DeviceState *dev)
3734 {
3735     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3736 
3737     /* Call the unplug handler chain. This can never fail. */
3738     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3739     object_unparent(OBJECT(dev));
3740 }
3741 
3742 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3743 {
3744     MachineState *ms = MACHINE(hotplug_dev);
3745     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3746     CPUCore *cc = CPU_CORE(dev);
3747     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3748 
3749     if (smc->pre_2_10_has_unused_icps) {
3750         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3751         int i;
3752 
3753         for (i = 0; i < cc->nr_threads; i++) {
3754             CPUState *cs = CPU(sc->threads[i]);
3755 
3756             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3757         }
3758     }
3759 
3760     assert(core_slot);
3761     core_slot->cpu = NULL;
3762     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3763 }
3764 
3765 static
3766 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3767                                Error **errp)
3768 {
3769     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3770     int index;
3771     SpaprDrc *drc;
3772     CPUCore *cc = CPU_CORE(dev);
3773 
3774     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3775         error_setg(errp, "Unable to find CPU core with core-id: %d",
3776                    cc->core_id);
3777         return;
3778     }
3779     if (index == 0) {
3780         error_setg(errp, "Boot CPU core may not be unplugged");
3781         return;
3782     }
3783 
3784     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3785                           spapr_vcpu_id(spapr, cc->core_id));
3786     g_assert(drc);
3787 
3788     if (!spapr_drc_unplug_requested(drc)) {
3789         spapr_drc_detach(drc);
3790         spapr_hotplug_req_remove_by_index(drc);
3791     }
3792 }
3793 
3794 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3795                            void *fdt, int *fdt_start_offset, Error **errp)
3796 {
3797     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3798     CPUState *cs = CPU(core->threads[0]);
3799     PowerPCCPU *cpu = POWERPC_CPU(cs);
3800     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3801     int id = spapr_get_vcpu_id(cpu);
3802     char *nodename;
3803     int offset;
3804 
3805     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3806     offset = fdt_add_subnode(fdt, 0, nodename);
3807     g_free(nodename);
3808 
3809     spapr_dt_cpu(cs, fdt, offset, spapr);
3810 
3811     *fdt_start_offset = offset;
3812     return 0;
3813 }
3814 
3815 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3816                             Error **errp)
3817 {
3818     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3819     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3820     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3821     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3822     CPUCore *cc = CPU_CORE(dev);
3823     CPUState *cs;
3824     SpaprDrc *drc;
3825     Error *local_err = NULL;
3826     CPUArchId *core_slot;
3827     int index;
3828     bool hotplugged = spapr_drc_hotplugged(dev);
3829     int i;
3830 
3831     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3832     if (!core_slot) {
3833         error_setg(errp, "Unable to find CPU core with core-id: %d",
3834                    cc->core_id);
3835         return;
3836     }
3837     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3838                           spapr_vcpu_id(spapr, cc->core_id));
3839 
3840     g_assert(drc || !mc->has_hotpluggable_cpus);
3841 
3842     if (drc) {
3843         spapr_drc_attach(drc, dev, &local_err);
3844         if (local_err) {
3845             error_propagate(errp, local_err);
3846             return;
3847         }
3848 
3849         if (hotplugged) {
3850             /*
3851              * Send hotplug notification interrupt to the guest only
3852              * in case of hotplugged CPUs.
3853              */
3854             spapr_hotplug_req_add_by_index(drc);
3855         } else {
3856             spapr_drc_reset(drc);
3857         }
3858     }
3859 
3860     core_slot->cpu = OBJECT(dev);
3861 
3862     if (smc->pre_2_10_has_unused_icps) {
3863         for (i = 0; i < cc->nr_threads; i++) {
3864             cs = CPU(core->threads[i]);
3865             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3866         }
3867     }
3868 
3869     /*
3870      * Set compatibility mode to match the boot CPU, which was either set
3871      * by the machine reset code or by CAS.
3872      */
3873     if (hotplugged) {
3874         for (i = 0; i < cc->nr_threads; i++) {
3875             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3876                            &local_err);
3877             if (local_err) {
3878                 error_propagate(errp, local_err);
3879                 return;
3880             }
3881         }
3882     }
3883 }
3884 
3885 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3886                                 Error **errp)
3887 {
3888     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3889     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3890     Error *local_err = NULL;
3891     CPUCore *cc = CPU_CORE(dev);
3892     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3893     const char *type = object_get_typename(OBJECT(dev));
3894     CPUArchId *core_slot;
3895     int index;
3896     unsigned int smp_threads = machine->smp.threads;
3897 
3898     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3899         error_setg(&local_err, "CPU hotplug not supported for this machine");
3900         goto out;
3901     }
3902 
3903     if (strcmp(base_core_type, type)) {
3904         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3905         goto out;
3906     }
3907 
3908     if (cc->core_id % smp_threads) {
3909         error_setg(&local_err, "invalid core id %d", cc->core_id);
3910         goto out;
3911     }
3912 
3913     /*
3914      * In general we should have homogeneous threads-per-core, but old
3915      * (pre hotplug support) machine types allow the last core to have
3916      * reduced threads as a compatibility hack for when we allowed
3917      * total vcpus not a multiple of threads-per-core.
3918      */
3919     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3920         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3921                    cc->nr_threads, smp_threads);
3922         goto out;
3923     }
3924 
3925     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3926     if (!core_slot) {
3927         error_setg(&local_err, "core id %d out of range", cc->core_id);
3928         goto out;
3929     }
3930 
3931     if (core_slot->cpu) {
3932         error_setg(&local_err, "core %d already populated", cc->core_id);
3933         goto out;
3934     }
3935 
3936     numa_cpu_pre_plug(core_slot, dev, &local_err);
3937 
3938 out:
3939     error_propagate(errp, local_err);
3940 }
3941 
3942 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3943                           void *fdt, int *fdt_start_offset, Error **errp)
3944 {
3945     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3946     int intc_phandle;
3947 
3948     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3949     if (intc_phandle <= 0) {
3950         return -1;
3951     }
3952 
3953     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3954         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3955         return -1;
3956     }
3957 
3958     /* generally SLOF creates these, for hotplug it's up to QEMU */
3959     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3960 
3961     return 0;
3962 }
3963 
3964 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3965                                Error **errp)
3966 {
3967     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3968     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3969     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3970     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3971 
3972     if (dev->hotplugged && !smc->dr_phb_enabled) {
3973         error_setg(errp, "PHB hotplug not supported for this machine");
3974         return;
3975     }
3976 
3977     if (sphb->index == (uint32_t)-1) {
3978         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3979         return;
3980     }
3981 
3982     /*
3983      * This will check that sphb->index doesn't exceed the maximum number of
3984      * PHBs for the current machine type.
3985      */
3986     smc->phb_placement(spapr, sphb->index,
3987                        &sphb->buid, &sphb->io_win_addr,
3988                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3989                        windows_supported, sphb->dma_liobn,
3990                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3991                        errp);
3992 }
3993 
3994 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3995                            Error **errp)
3996 {
3997     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3998     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3999     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4000     SpaprDrc *drc;
4001     bool hotplugged = spapr_drc_hotplugged(dev);
4002     Error *local_err = NULL;
4003 
4004     if (!smc->dr_phb_enabled) {
4005         return;
4006     }
4007 
4008     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4009     /* hotplug hooks should check it's enabled before getting this far */
4010     assert(drc);
4011 
4012     spapr_drc_attach(drc, DEVICE(dev), &local_err);
4013     if (local_err) {
4014         error_propagate(errp, local_err);
4015         return;
4016     }
4017 
4018     if (hotplugged) {
4019         spapr_hotplug_req_add_by_index(drc);
4020     } else {
4021         spapr_drc_reset(drc);
4022     }
4023 }
4024 
4025 void spapr_phb_release(DeviceState *dev)
4026 {
4027     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4028 
4029     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4030     object_unparent(OBJECT(dev));
4031 }
4032 
4033 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4034 {
4035     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4036 }
4037 
4038 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4039                                      DeviceState *dev, Error **errp)
4040 {
4041     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4042     SpaprDrc *drc;
4043 
4044     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4045     assert(drc);
4046 
4047     if (!spapr_drc_unplug_requested(drc)) {
4048         spapr_drc_detach(drc);
4049         spapr_hotplug_req_remove_by_index(drc);
4050     }
4051 }
4052 
4053 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4054                                  Error **errp)
4055 {
4056     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4057     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4058 
4059     if (spapr->tpm_proxy != NULL) {
4060         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4061         return;
4062     }
4063 
4064     spapr->tpm_proxy = tpm_proxy;
4065 }
4066 
4067 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4068 {
4069     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4070 
4071     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4072     object_unparent(OBJECT(dev));
4073     spapr->tpm_proxy = NULL;
4074 }
4075 
4076 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4077                                       DeviceState *dev, Error **errp)
4078 {
4079     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4080         spapr_memory_plug(hotplug_dev, dev, errp);
4081     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4082         spapr_core_plug(hotplug_dev, dev, errp);
4083     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4084         spapr_phb_plug(hotplug_dev, dev, errp);
4085     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4086         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4087     }
4088 }
4089 
4090 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4091                                         DeviceState *dev, Error **errp)
4092 {
4093     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4094         spapr_memory_unplug(hotplug_dev, dev);
4095     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4096         spapr_core_unplug(hotplug_dev, dev);
4097     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4098         spapr_phb_unplug(hotplug_dev, dev);
4099     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4100         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4101     }
4102 }
4103 
4104 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4105                                                 DeviceState *dev, Error **errp)
4106 {
4107     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4108     MachineClass *mc = MACHINE_GET_CLASS(sms);
4109     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4110 
4111     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4112         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4113             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4114         } else {
4115             /* NOTE: this means there is a window after guest reset, prior to
4116              * CAS negotiation, where unplug requests will fail due to the
4117              * capability not being detected yet. This is a bit different than
4118              * the case with PCI unplug, where the events will be queued and
4119              * eventually handled by the guest after boot
4120              */
4121             error_setg(errp, "Memory hot unplug not supported for this guest");
4122         }
4123     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4124         if (!mc->has_hotpluggable_cpus) {
4125             error_setg(errp, "CPU hot unplug not supported on this machine");
4126             return;
4127         }
4128         spapr_core_unplug_request(hotplug_dev, dev, errp);
4129     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4130         if (!smc->dr_phb_enabled) {
4131             error_setg(errp, "PHB hot unplug not supported on this machine");
4132             return;
4133         }
4134         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4135     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4136         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4137     }
4138 }
4139 
4140 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4141                                           DeviceState *dev, Error **errp)
4142 {
4143     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4144         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4145     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4146         spapr_core_pre_plug(hotplug_dev, dev, errp);
4147     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4148         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4149     }
4150 }
4151 
4152 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4153                                                  DeviceState *dev)
4154 {
4155     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4156         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4157         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4158         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4159         return HOTPLUG_HANDLER(machine);
4160     }
4161     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4162         PCIDevice *pcidev = PCI_DEVICE(dev);
4163         PCIBus *root = pci_device_root_bus(pcidev);
4164         SpaprPhbState *phb =
4165             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4166                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4167 
4168         if (phb) {
4169             return HOTPLUG_HANDLER(phb);
4170         }
4171     }
4172     return NULL;
4173 }
4174 
4175 static CpuInstanceProperties
4176 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4177 {
4178     CPUArchId *core_slot;
4179     MachineClass *mc = MACHINE_GET_CLASS(machine);
4180 
4181     /* make sure possible_cpu are intialized */
4182     mc->possible_cpu_arch_ids(machine);
4183     /* get CPU core slot containing thread that matches cpu_index */
4184     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4185     assert(core_slot);
4186     return core_slot->props;
4187 }
4188 
4189 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4190 {
4191     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4192 }
4193 
4194 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4195 {
4196     int i;
4197     unsigned int smp_threads = machine->smp.threads;
4198     unsigned int smp_cpus = machine->smp.cpus;
4199     const char *core_type;
4200     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4201     MachineClass *mc = MACHINE_GET_CLASS(machine);
4202 
4203     if (!mc->has_hotpluggable_cpus) {
4204         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4205     }
4206     if (machine->possible_cpus) {
4207         assert(machine->possible_cpus->len == spapr_max_cores);
4208         return machine->possible_cpus;
4209     }
4210 
4211     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4212     if (!core_type) {
4213         error_report("Unable to find sPAPR CPU Core definition");
4214         exit(1);
4215     }
4216 
4217     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4218                              sizeof(CPUArchId) * spapr_max_cores);
4219     machine->possible_cpus->len = spapr_max_cores;
4220     for (i = 0; i < machine->possible_cpus->len; i++) {
4221         int core_id = i * smp_threads;
4222 
4223         machine->possible_cpus->cpus[i].type = core_type;
4224         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4225         machine->possible_cpus->cpus[i].arch_id = core_id;
4226         machine->possible_cpus->cpus[i].props.has_core_id = true;
4227         machine->possible_cpus->cpus[i].props.core_id = core_id;
4228     }
4229     return machine->possible_cpus;
4230 }
4231 
4232 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4233                                 uint64_t *buid, hwaddr *pio,
4234                                 hwaddr *mmio32, hwaddr *mmio64,
4235                                 unsigned n_dma, uint32_t *liobns,
4236                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4237 {
4238     /*
4239      * New-style PHB window placement.
4240      *
4241      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4242      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4243      * windows.
4244      *
4245      * Some guest kernels can't work with MMIO windows above 1<<46
4246      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4247      *
4248      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4249      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4250      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4251      * 1TiB 64-bit MMIO windows for each PHB.
4252      */
4253     const uint64_t base_buid = 0x800000020000000ULL;
4254     int i;
4255 
4256     /* Sanity check natural alignments */
4257     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4258     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4259     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4260     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4261     /* Sanity check bounds */
4262     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4263                       SPAPR_PCI_MEM32_WIN_SIZE);
4264     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4265                       SPAPR_PCI_MEM64_WIN_SIZE);
4266 
4267     if (index >= SPAPR_MAX_PHBS) {
4268         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4269                    SPAPR_MAX_PHBS - 1);
4270         return;
4271     }
4272 
4273     *buid = base_buid + index;
4274     for (i = 0; i < n_dma; ++i) {
4275         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4276     }
4277 
4278     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4279     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4280     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4281 
4282     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4283     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4284 }
4285 
4286 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4287 {
4288     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4289 
4290     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4291 }
4292 
4293 static void spapr_ics_resend(XICSFabric *dev)
4294 {
4295     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4296 
4297     ics_resend(spapr->ics);
4298 }
4299 
4300 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4301 {
4302     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4303 
4304     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4305 }
4306 
4307 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4308                                  Monitor *mon)
4309 {
4310     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4311 
4312     spapr_irq_print_info(spapr, mon);
4313     monitor_printf(mon, "irqchip: %s\n",
4314                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4315 }
4316 
4317 /*
4318  * This is a XIVE only operation
4319  */
4320 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4321                            uint8_t nvt_blk, uint32_t nvt_idx,
4322                            bool cam_ignore, uint8_t priority,
4323                            uint32_t logic_serv, XiveTCTXMatch *match)
4324 {
4325     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4326     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4327     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4328     int count;
4329 
4330     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4331                            priority, logic_serv, match);
4332     if (count < 0) {
4333         return count;
4334     }
4335 
4336     /*
4337      * When we implement the save and restore of the thread interrupt
4338      * contexts in the enter/exit CPU handlers of the machine and the
4339      * escalations in QEMU, we should be able to handle non dispatched
4340      * vCPUs.
4341      *
4342      * Until this is done, the sPAPR machine should find at least one
4343      * matching context always.
4344      */
4345     if (count == 0) {
4346         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4347                       nvt_blk, nvt_idx);
4348     }
4349 
4350     return count;
4351 }
4352 
4353 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4354 {
4355     return cpu->vcpu_id;
4356 }
4357 
4358 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4359 {
4360     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4361     MachineState *ms = MACHINE(spapr);
4362     int vcpu_id;
4363 
4364     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4365 
4366     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4367         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4368         error_append_hint(errp, "Adjust the number of cpus to %d "
4369                           "or try to raise the number of threads per core\n",
4370                           vcpu_id * ms->smp.threads / spapr->vsmt);
4371         return;
4372     }
4373 
4374     cpu->vcpu_id = vcpu_id;
4375 }
4376 
4377 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4378 {
4379     CPUState *cs;
4380 
4381     CPU_FOREACH(cs) {
4382         PowerPCCPU *cpu = POWERPC_CPU(cs);
4383 
4384         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4385             return cpu;
4386         }
4387     }
4388 
4389     return NULL;
4390 }
4391 
4392 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4393 {
4394     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4395 
4396     /* These are only called by TCG, KVM maintains dispatch state */
4397 
4398     spapr_cpu->prod = false;
4399     if (spapr_cpu->vpa_addr) {
4400         CPUState *cs = CPU(cpu);
4401         uint32_t dispatch;
4402 
4403         dispatch = ldl_be_phys(cs->as,
4404                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4405         dispatch++;
4406         if ((dispatch & 1) != 0) {
4407             qemu_log_mask(LOG_GUEST_ERROR,
4408                           "VPA: incorrect dispatch counter value for "
4409                           "dispatched partition %u, correcting.\n", dispatch);
4410             dispatch++;
4411         }
4412         stl_be_phys(cs->as,
4413                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4414     }
4415 }
4416 
4417 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4418 {
4419     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4420 
4421     if (spapr_cpu->vpa_addr) {
4422         CPUState *cs = CPU(cpu);
4423         uint32_t dispatch;
4424 
4425         dispatch = ldl_be_phys(cs->as,
4426                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4427         dispatch++;
4428         if ((dispatch & 1) != 1) {
4429             qemu_log_mask(LOG_GUEST_ERROR,
4430                           "VPA: incorrect dispatch counter value for "
4431                           "preempted partition %u, correcting.\n", dispatch);
4432             dispatch++;
4433         }
4434         stl_be_phys(cs->as,
4435                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4436     }
4437 }
4438 
4439 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4440 {
4441     MachineClass *mc = MACHINE_CLASS(oc);
4442     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4443     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4444     NMIClass *nc = NMI_CLASS(oc);
4445     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4446     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4447     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4448     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4449     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4450 
4451     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4452     mc->ignore_boot_device_suffixes = true;
4453 
4454     /*
4455      * We set up the default / latest behaviour here.  The class_init
4456      * functions for the specific versioned machine types can override
4457      * these details for backwards compatibility
4458      */
4459     mc->init = spapr_machine_init;
4460     mc->reset = spapr_machine_reset;
4461     mc->block_default_type = IF_SCSI;
4462     mc->max_cpus = 1024;
4463     mc->no_parallel = 1;
4464     mc->default_boot_order = "";
4465     mc->default_ram_size = 512 * MiB;
4466     mc->default_ram_id = "ppc_spapr.ram";
4467     mc->default_display = "std";
4468     mc->kvm_type = spapr_kvm_type;
4469     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4470     mc->pci_allow_0_address = true;
4471     assert(!mc->get_hotplug_handler);
4472     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4473     hc->pre_plug = spapr_machine_device_pre_plug;
4474     hc->plug = spapr_machine_device_plug;
4475     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4476     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4477     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4478     hc->unplug_request = spapr_machine_device_unplug_request;
4479     hc->unplug = spapr_machine_device_unplug;
4480 
4481     smc->dr_lmb_enabled = true;
4482     smc->update_dt_enabled = true;
4483     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4484     mc->has_hotpluggable_cpus = true;
4485     mc->nvdimm_supported = true;
4486     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4487     fwc->get_dev_path = spapr_get_fw_dev_path;
4488     nc->nmi_monitor_handler = spapr_nmi;
4489     smc->phb_placement = spapr_phb_placement;
4490     vhc->hypercall = emulate_spapr_hypercall;
4491     vhc->hpt_mask = spapr_hpt_mask;
4492     vhc->map_hptes = spapr_map_hptes;
4493     vhc->unmap_hptes = spapr_unmap_hptes;
4494     vhc->hpte_set_c = spapr_hpte_set_c;
4495     vhc->hpte_set_r = spapr_hpte_set_r;
4496     vhc->get_pate = spapr_get_pate;
4497     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4498     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4499     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4500     xic->ics_get = spapr_ics_get;
4501     xic->ics_resend = spapr_ics_resend;
4502     xic->icp_get = spapr_icp_get;
4503     ispc->print_info = spapr_pic_print_info;
4504     /* Force NUMA node memory size to be a multiple of
4505      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4506      * in which LMBs are represented and hot-added
4507      */
4508     mc->numa_mem_align_shift = 28;
4509     mc->numa_mem_supported = true;
4510     mc->auto_enable_numa = true;
4511 
4512     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4513     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4514     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4515     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4516     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4517     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4518     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4519     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4520     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4521     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4522     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4523     spapr_caps_add_properties(smc, &error_abort);
4524     smc->irq = &spapr_irq_dual;
4525     smc->dr_phb_enabled = true;
4526     smc->linux_pci_probe = true;
4527     smc->smp_threads_vsmt = true;
4528     smc->nr_xirqs = SPAPR_NR_XIRQS;
4529     xfc->match_nvt = spapr_match_nvt;
4530 }
4531 
4532 static const TypeInfo spapr_machine_info = {
4533     .name          = TYPE_SPAPR_MACHINE,
4534     .parent        = TYPE_MACHINE,
4535     .abstract      = true,
4536     .instance_size = sizeof(SpaprMachineState),
4537     .instance_init = spapr_instance_init,
4538     .instance_finalize = spapr_machine_finalizefn,
4539     .class_size    = sizeof(SpaprMachineClass),
4540     .class_init    = spapr_machine_class_init,
4541     .interfaces = (InterfaceInfo[]) {
4542         { TYPE_FW_PATH_PROVIDER },
4543         { TYPE_NMI },
4544         { TYPE_HOTPLUG_HANDLER },
4545         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4546         { TYPE_XICS_FABRIC },
4547         { TYPE_INTERRUPT_STATS_PROVIDER },
4548         { TYPE_XIVE_FABRIC },
4549         { }
4550     },
4551 };
4552 
4553 static void spapr_machine_latest_class_options(MachineClass *mc)
4554 {
4555     mc->alias = "pseries";
4556     mc->is_default = true;
4557 }
4558 
4559 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4560     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4561                                                     void *data)      \
4562     {                                                                \
4563         MachineClass *mc = MACHINE_CLASS(oc);                        \
4564         spapr_machine_##suffix##_class_options(mc);                  \
4565         if (latest) {                                                \
4566             spapr_machine_latest_class_options(mc);                  \
4567         }                                                            \
4568     }                                                                \
4569     static const TypeInfo spapr_machine_##suffix##_info = {          \
4570         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4571         .parent = TYPE_SPAPR_MACHINE,                                \
4572         .class_init = spapr_machine_##suffix##_class_init,           \
4573     };                                                               \
4574     static void spapr_machine_register_##suffix(void)                \
4575     {                                                                \
4576         type_register(&spapr_machine_##suffix##_info);               \
4577     }                                                                \
4578     type_init(spapr_machine_register_##suffix)
4579 
4580 /*
4581  * pseries-5.0
4582  */
4583 static void spapr_machine_5_0_class_options(MachineClass *mc)
4584 {
4585     /* Defaults for the latest behaviour inherited from the base class */
4586 }
4587 
4588 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4589 
4590 /*
4591  * pseries-4.2
4592  */
4593 static void spapr_machine_4_2_class_options(MachineClass *mc)
4594 {
4595     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4596 
4597     spapr_machine_5_0_class_options(mc);
4598     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4599     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4600     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4601     smc->rma_limit = 16 * GiB;
4602     mc->nvdimm_supported = false;
4603 }
4604 
4605 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4606 
4607 /*
4608  * pseries-4.1
4609  */
4610 static void spapr_machine_4_1_class_options(MachineClass *mc)
4611 {
4612     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4613     static GlobalProperty compat[] = {
4614         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4615         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4616     };
4617 
4618     spapr_machine_4_2_class_options(mc);
4619     smc->linux_pci_probe = false;
4620     smc->smp_threads_vsmt = false;
4621     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4622     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4623 }
4624 
4625 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4626 
4627 /*
4628  * pseries-4.0
4629  */
4630 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4631                               uint64_t *buid, hwaddr *pio,
4632                               hwaddr *mmio32, hwaddr *mmio64,
4633                               unsigned n_dma, uint32_t *liobns,
4634                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4635 {
4636     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4637                         nv2gpa, nv2atsd, errp);
4638     *nv2gpa = 0;
4639     *nv2atsd = 0;
4640 }
4641 
4642 static void spapr_machine_4_0_class_options(MachineClass *mc)
4643 {
4644     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4645 
4646     spapr_machine_4_1_class_options(mc);
4647     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4648     smc->phb_placement = phb_placement_4_0;
4649     smc->irq = &spapr_irq_xics;
4650     smc->pre_4_1_migration = true;
4651 }
4652 
4653 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4654 
4655 /*
4656  * pseries-3.1
4657  */
4658 static void spapr_machine_3_1_class_options(MachineClass *mc)
4659 {
4660     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4661 
4662     spapr_machine_4_0_class_options(mc);
4663     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4664 
4665     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4666     smc->update_dt_enabled = false;
4667     smc->dr_phb_enabled = false;
4668     smc->broken_host_serial_model = true;
4669     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4670     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4671     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4672     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4673 }
4674 
4675 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4676 
4677 /*
4678  * pseries-3.0
4679  */
4680 
4681 static void spapr_machine_3_0_class_options(MachineClass *mc)
4682 {
4683     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4684 
4685     spapr_machine_3_1_class_options(mc);
4686     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4687 
4688     smc->legacy_irq_allocation = true;
4689     smc->nr_xirqs = 0x400;
4690     smc->irq = &spapr_irq_xics_legacy;
4691 }
4692 
4693 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4694 
4695 /*
4696  * pseries-2.12
4697  */
4698 static void spapr_machine_2_12_class_options(MachineClass *mc)
4699 {
4700     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4701     static GlobalProperty compat[] = {
4702         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4703         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4704     };
4705 
4706     spapr_machine_3_0_class_options(mc);
4707     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4708     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4709 
4710     /* We depend on kvm_enabled() to choose a default value for the
4711      * hpt-max-page-size capability. Of course we can't do it here
4712      * because this is too early and the HW accelerator isn't initialzed
4713      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4714      */
4715     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4716 }
4717 
4718 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4719 
4720 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4721 {
4722     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4723 
4724     spapr_machine_2_12_class_options(mc);
4725     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4726     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4727     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4728 }
4729 
4730 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4731 
4732 /*
4733  * pseries-2.11
4734  */
4735 
4736 static void spapr_machine_2_11_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_HTM] = SPAPR_CAP_ON;
4742     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4743 }
4744 
4745 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4746 
4747 /*
4748  * pseries-2.10
4749  */
4750 
4751 static void spapr_machine_2_10_class_options(MachineClass *mc)
4752 {
4753     spapr_machine_2_11_class_options(mc);
4754     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4755 }
4756 
4757 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4758 
4759 /*
4760  * pseries-2.9
4761  */
4762 
4763 static void spapr_machine_2_9_class_options(MachineClass *mc)
4764 {
4765     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4766     static GlobalProperty compat[] = {
4767         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4768     };
4769 
4770     spapr_machine_2_10_class_options(mc);
4771     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4772     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4773     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4774     smc->pre_2_10_has_unused_icps = true;
4775     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4776 }
4777 
4778 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4779 
4780 /*
4781  * pseries-2.8
4782  */
4783 
4784 static void spapr_machine_2_8_class_options(MachineClass *mc)
4785 {
4786     static GlobalProperty compat[] = {
4787         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4788     };
4789 
4790     spapr_machine_2_9_class_options(mc);
4791     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4792     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4793     mc->numa_mem_align_shift = 23;
4794 }
4795 
4796 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4797 
4798 /*
4799  * pseries-2.7
4800  */
4801 
4802 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4803                               uint64_t *buid, hwaddr *pio,
4804                               hwaddr *mmio32, hwaddr *mmio64,
4805                               unsigned n_dma, uint32_t *liobns,
4806                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4807 {
4808     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4809     const uint64_t base_buid = 0x800000020000000ULL;
4810     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4811     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4812     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4813     const uint32_t max_index = 255;
4814     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4815 
4816     uint64_t ram_top = MACHINE(spapr)->ram_size;
4817     hwaddr phb0_base, phb_base;
4818     int i;
4819 
4820     /* Do we have device memory? */
4821     if (MACHINE(spapr)->maxram_size > ram_top) {
4822         /* Can't just use maxram_size, because there may be an
4823          * alignment gap between normal and device memory regions
4824          */
4825         ram_top = MACHINE(spapr)->device_memory->base +
4826             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4827     }
4828 
4829     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4830 
4831     if (index > max_index) {
4832         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4833                    max_index);
4834         return;
4835     }
4836 
4837     *buid = base_buid + index;
4838     for (i = 0; i < n_dma; ++i) {
4839         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4840     }
4841 
4842     phb_base = phb0_base + index * phb_spacing;
4843     *pio = phb_base + pio_offset;
4844     *mmio32 = phb_base + mmio_offset;
4845     /*
4846      * We don't set the 64-bit MMIO window, relying on the PHB's
4847      * fallback behaviour of automatically splitting a large "32-bit"
4848      * window into contiguous 32-bit and 64-bit windows
4849      */
4850 
4851     *nv2gpa = 0;
4852     *nv2atsd = 0;
4853 }
4854 
4855 static void spapr_machine_2_7_class_options(MachineClass *mc)
4856 {
4857     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4858     static GlobalProperty compat[] = {
4859         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4860         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4861         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4862         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4863     };
4864 
4865     spapr_machine_2_8_class_options(mc);
4866     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4867     mc->default_machine_opts = "modern-hotplug-events=off";
4868     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4869     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4870     smc->phb_placement = phb_placement_2_7;
4871 }
4872 
4873 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4874 
4875 /*
4876  * pseries-2.6
4877  */
4878 
4879 static void spapr_machine_2_6_class_options(MachineClass *mc)
4880 {
4881     static GlobalProperty compat[] = {
4882         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4883     };
4884 
4885     spapr_machine_2_7_class_options(mc);
4886     mc->has_hotpluggable_cpus = false;
4887     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4888     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4889 }
4890 
4891 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4892 
4893 /*
4894  * pseries-2.5
4895  */
4896 
4897 static void spapr_machine_2_5_class_options(MachineClass *mc)
4898 {
4899     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4900     static GlobalProperty compat[] = {
4901         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4902     };
4903 
4904     spapr_machine_2_6_class_options(mc);
4905     smc->use_ohci_by_default = true;
4906     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4907     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4908 }
4909 
4910 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4911 
4912 /*
4913  * pseries-2.4
4914  */
4915 
4916 static void spapr_machine_2_4_class_options(MachineClass *mc)
4917 {
4918     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4919 
4920     spapr_machine_2_5_class_options(mc);
4921     smc->dr_lmb_enabled = false;
4922     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4923 }
4924 
4925 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4926 
4927 /*
4928  * pseries-2.3
4929  */
4930 
4931 static void spapr_machine_2_3_class_options(MachineClass *mc)
4932 {
4933     static GlobalProperty compat[] = {
4934         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4935     };
4936     spapr_machine_2_4_class_options(mc);
4937     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4938     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4939 }
4940 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4941 
4942 /*
4943  * pseries-2.2
4944  */
4945 
4946 static void spapr_machine_2_2_class_options(MachineClass *mc)
4947 {
4948     static GlobalProperty compat[] = {
4949         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4950     };
4951 
4952     spapr_machine_2_3_class_options(mc);
4953     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4954     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4955     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4956 }
4957 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4958 
4959 /*
4960  * pseries-2.1
4961  */
4962 
4963 static void spapr_machine_2_1_class_options(MachineClass *mc)
4964 {
4965     spapr_machine_2_2_class_options(mc);
4966     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4967 }
4968 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4969 
4970 static void spapr_machine_register_types(void)
4971 {
4972     type_register_static(&spapr_machine_info);
4973 }
4974 
4975 type_init(spapr_machine_register_types)
4976