xref: /qemu/hw/ppc/spapr.c (revision b5513584a08db477160ea7f0b700a2367028720c)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qemu/datadir.h"
30 #include "qapi/error.h"
31 #include "qapi/qapi-events-machine.h"
32 #include "qapi/qapi-events-qdev.h"
33 #include "qapi/visitor.h"
34 #include "sysemu/sysemu.h"
35 #include "sysemu/hostmem.h"
36 #include "sysemu/numa.h"
37 #include "sysemu/qtest.h"
38 #include "sysemu/reset.h"
39 #include "sysemu/runstate.h"
40 #include "qemu/log.h"
41 #include "hw/fw-path-provider.h"
42 #include "elf.h"
43 #include "net/net.h"
44 #include "sysemu/device_tree.h"
45 #include "sysemu/cpus.h"
46 #include "sysemu/hw_accel.h"
47 #include "kvm_ppc.h"
48 #include "migration/misc.h"
49 #include "migration/qemu-file-types.h"
50 #include "migration/global_state.h"
51 #include "migration/register.h"
52 #include "migration/blocker.h"
53 #include "mmu-hash64.h"
54 #include "mmu-book3s-v3.h"
55 #include "cpu-models.h"
56 #include "hw/core/cpu.h"
57 
58 #include "hw/ppc/ppc.h"
59 #include "hw/loader.h"
60 
61 #include "hw/ppc/fdt.h"
62 #include "hw/ppc/spapr.h"
63 #include "hw/ppc/spapr_vio.h"
64 #include "hw/qdev-properties.h"
65 #include "hw/pci-host/spapr.h"
66 #include "hw/pci/msi.h"
67 
68 #include "hw/pci/pci.h"
69 #include "hw/scsi/scsi.h"
70 #include "hw/virtio/virtio-scsi.h"
71 #include "hw/virtio/vhost-scsi-common.h"
72 
73 #include "exec/ram_addr.h"
74 #include "hw/usb.h"
75 #include "qemu/config-file.h"
76 #include "qemu/error-report.h"
77 #include "trace.h"
78 #include "hw/nmi.h"
79 #include "hw/intc/intc.h"
80 
81 #include "hw/ppc/spapr_cpu_core.h"
82 #include "hw/mem/memory-device.h"
83 #include "hw/ppc/spapr_tpm_proxy.h"
84 #include "hw/ppc/spapr_nvdimm.h"
85 #include "hw/ppc/spapr_numa.h"
86 #include "hw/ppc/pef.h"
87 
88 #include "monitor/monitor.h"
89 
90 #include <libfdt.h>
91 
92 /* SLOF memory layout:
93  *
94  * SLOF raw image loaded at 0, copies its romfs right below the flat
95  * device-tree, then position SLOF itself 31M below that
96  *
97  * So we set FW_OVERHEAD to 40MB which should account for all of that
98  * and more
99  *
100  * We load our kernel at 4M, leaving space for SLOF initial image
101  */
102 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
103 #define FW_MAX_SIZE             0x400000
104 #define FW_FILE_NAME            "slof.bin"
105 #define FW_FILE_NAME_VOF        "vof.bin"
106 #define FW_OVERHEAD             0x2800000
107 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
108 
109 #define MIN_RMA_SLOF            (128 * MiB)
110 
111 #define PHANDLE_INTC            0x00001111
112 
113 /* These two functions implement the VCPU id numbering: one to compute them
114  * all and one to identify thread 0 of a VCORE. Any change to the first one
115  * is likely to have an impact on the second one, so let's keep them close.
116  */
117 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
118 {
119     MachineState *ms = MACHINE(spapr);
120     unsigned int smp_threads = ms->smp.threads;
121 
122     assert(spapr->vsmt);
123     return
124         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
125 }
126 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
127                                       PowerPCCPU *cpu)
128 {
129     assert(spapr->vsmt);
130     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
131 }
132 
133 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
134 {
135     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
136      * and newer QEMUs don't even have them. In both cases, we don't want
137      * to send anything on the wire.
138      */
139     return false;
140 }
141 
142 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
143     .name = "icp/server",
144     .version_id = 1,
145     .minimum_version_id = 1,
146     .needed = pre_2_10_vmstate_dummy_icp_needed,
147     .fields = (VMStateField[]) {
148         VMSTATE_UNUSED(4), /* uint32_t xirr */
149         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
150         VMSTATE_UNUSED(1), /* uint8_t mfrr */
151         VMSTATE_END_OF_LIST()
152     },
153 };
154 
155 static void pre_2_10_vmstate_register_dummy_icp(int i)
156 {
157     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
158                      (void *)(uintptr_t) i);
159 }
160 
161 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
162 {
163     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
164                        (void *)(uintptr_t) i);
165 }
166 
167 int spapr_max_server_number(SpaprMachineState *spapr)
168 {
169     MachineState *ms = MACHINE(spapr);
170 
171     assert(spapr->vsmt);
172     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
173 }
174 
175 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
176                                   int smt_threads)
177 {
178     int i, ret = 0;
179     uint32_t servers_prop[smt_threads];
180     uint32_t gservers_prop[smt_threads * 2];
181     int index = spapr_get_vcpu_id(cpu);
182 
183     if (cpu->compat_pvr) {
184         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
185         if (ret < 0) {
186             return ret;
187         }
188     }
189 
190     /* Build interrupt servers and gservers properties */
191     for (i = 0; i < smt_threads; i++) {
192         servers_prop[i] = cpu_to_be32(index + i);
193         /* Hack, direct the group queues back to cpu 0 */
194         gservers_prop[i*2] = cpu_to_be32(index + i);
195         gservers_prop[i*2 + 1] = 0;
196     }
197     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
198                       servers_prop, sizeof(servers_prop));
199     if (ret < 0) {
200         return ret;
201     }
202     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
203                       gservers_prop, sizeof(gservers_prop));
204 
205     return ret;
206 }
207 
208 static void spapr_dt_pa_features(SpaprMachineState *spapr,
209                                  PowerPCCPU *cpu,
210                                  void *fdt, int offset)
211 {
212     uint8_t pa_features_206[] = { 6, 0,
213         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214     uint8_t pa_features_207[] = { 24, 0,
215         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219     uint8_t pa_features_300[] = { 66, 0,
220         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
223         /* 6: DS207 */
224         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
225         /* 16: Vector */
226         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235         /* 42: PM, 44: PC RA, 46: SC vec'd */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237         /* 48: SIMD, 50: QP BFP, 52: String */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239         /* 54: DecFP, 56: DecI, 58: SHA */
240         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241         /* 60: NM atomic, 62: RNG */
242         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
243     };
244     uint8_t *pa_features = NULL;
245     size_t pa_size;
246 
247     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
248         pa_features = pa_features_206;
249         pa_size = sizeof(pa_features_206);
250     }
251     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
252         pa_features = pa_features_207;
253         pa_size = sizeof(pa_features_207);
254     }
255     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
256         pa_features = pa_features_300;
257         pa_size = sizeof(pa_features_300);
258     }
259     if (!pa_features) {
260         return;
261     }
262 
263     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
264         /*
265          * Note: we keep CI large pages off by default because a 64K capable
266          * guest provisioned with large pages might otherwise try to map a qemu
267          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268          * even if that qemu runs on a 4k host.
269          * We dd this bit back here if we are confident this is not an issue
270          */
271         pa_features[3] |= 0x20;
272     }
273     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
274         pa_features[24] |= 0x80;    /* Transactional memory support */
275     }
276     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
277         /* Workaround for broken kernels that attempt (guest) radix
278          * mode when they can't handle it, if they see the radix bit set
279          * in pa-features. So hide it from them. */
280         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
281     }
282 
283     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
284 }
285 
286 static hwaddr spapr_node0_size(MachineState *machine)
287 {
288     if (machine->numa_state->num_nodes) {
289         int i;
290         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
291             if (machine->numa_state->nodes[i].node_mem) {
292                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
293                            machine->ram_size);
294             }
295         }
296     }
297     return machine->ram_size;
298 }
299 
300 static void add_str(GString *s, const gchar *s1)
301 {
302     g_string_append_len(s, s1, strlen(s1) + 1);
303 }
304 
305 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
306                                 hwaddr start, hwaddr size)
307 {
308     char mem_name[32];
309     uint64_t mem_reg_property[2];
310     int off;
311 
312     mem_reg_property[0] = cpu_to_be64(start);
313     mem_reg_property[1] = cpu_to_be64(size);
314 
315     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
316     off = fdt_add_subnode(fdt, 0, mem_name);
317     _FDT(off);
318     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
319     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
320                       sizeof(mem_reg_property))));
321     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
322     return off;
323 }
324 
325 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
326 {
327     MemoryDeviceInfoList *info;
328 
329     for (info = list; info; info = info->next) {
330         MemoryDeviceInfo *value = info->value;
331 
332         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
333             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
334 
335             if (addr >= pcdimm_info->addr &&
336                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
337                 return pcdimm_info->node;
338             }
339         }
340     }
341 
342     return -1;
343 }
344 
345 struct sPAPRDrconfCellV2 {
346      uint32_t seq_lmbs;
347      uint64_t base_addr;
348      uint32_t drc_index;
349      uint32_t aa_index;
350      uint32_t flags;
351 } QEMU_PACKED;
352 
353 typedef struct DrconfCellQueue {
354     struct sPAPRDrconfCellV2 cell;
355     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
356 } DrconfCellQueue;
357 
358 static DrconfCellQueue *
359 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
360                       uint32_t drc_index, uint32_t aa_index,
361                       uint32_t flags)
362 {
363     DrconfCellQueue *elem;
364 
365     elem = g_malloc0(sizeof(*elem));
366     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
367     elem->cell.base_addr = cpu_to_be64(base_addr);
368     elem->cell.drc_index = cpu_to_be32(drc_index);
369     elem->cell.aa_index = cpu_to_be32(aa_index);
370     elem->cell.flags = cpu_to_be32(flags);
371 
372     return elem;
373 }
374 
375 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
376                                       int offset, MemoryDeviceInfoList *dimms)
377 {
378     MachineState *machine = MACHINE(spapr);
379     uint8_t *int_buf, *cur_index;
380     int ret;
381     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
382     uint64_t addr, cur_addr, size;
383     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
384     uint64_t mem_end = machine->device_memory->base +
385                        memory_region_size(&machine->device_memory->mr);
386     uint32_t node, buf_len, nr_entries = 0;
387     SpaprDrc *drc;
388     DrconfCellQueue *elem, *next;
389     MemoryDeviceInfoList *info;
390     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
391         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
392 
393     /* Entry to cover RAM and the gap area */
394     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
395                                  SPAPR_LMB_FLAGS_RESERVED |
396                                  SPAPR_LMB_FLAGS_DRC_INVALID);
397     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
398     nr_entries++;
399 
400     cur_addr = machine->device_memory->base;
401     for (info = dimms; info; info = info->next) {
402         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
403 
404         addr = di->addr;
405         size = di->size;
406         node = di->node;
407 
408         /*
409          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
410          * area is marked hotpluggable in the next iteration for the bigger
411          * chunk including the NVDIMM occupied area.
412          */
413         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
414             continue;
415 
416         /* Entry for hot-pluggable area */
417         if (cur_addr < addr) {
418             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
419             g_assert(drc);
420             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
421                                          cur_addr, spapr_drc_index(drc), -1, 0);
422             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
423             nr_entries++;
424         }
425 
426         /* Entry for DIMM */
427         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
428         g_assert(drc);
429         elem = spapr_get_drconf_cell(size / lmb_size, addr,
430                                      spapr_drc_index(drc), node,
431                                      (SPAPR_LMB_FLAGS_ASSIGNED |
432                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
433         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
434         nr_entries++;
435         cur_addr = addr + size;
436     }
437 
438     /* Entry for remaining hotpluggable area */
439     if (cur_addr < mem_end) {
440         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
441         g_assert(drc);
442         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
443                                      cur_addr, spapr_drc_index(drc), -1, 0);
444         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
445         nr_entries++;
446     }
447 
448     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
449     int_buf = cur_index = g_malloc0(buf_len);
450     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
451     cur_index += sizeof(nr_entries);
452 
453     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
454         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
455         cur_index += sizeof(elem->cell);
456         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
457         g_free(elem);
458     }
459 
460     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
461     g_free(int_buf);
462     if (ret < 0) {
463         return -1;
464     }
465     return 0;
466 }
467 
468 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
469                                    int offset, MemoryDeviceInfoList *dimms)
470 {
471     MachineState *machine = MACHINE(spapr);
472     int i, ret;
473     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
474     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
475     uint32_t nr_lmbs = (machine->device_memory->base +
476                        memory_region_size(&machine->device_memory->mr)) /
477                        lmb_size;
478     uint32_t *int_buf, *cur_index, buf_len;
479 
480     /*
481      * Allocate enough buffer size to fit in ibm,dynamic-memory
482      */
483     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
484     cur_index = int_buf = g_malloc0(buf_len);
485     int_buf[0] = cpu_to_be32(nr_lmbs);
486     cur_index++;
487     for (i = 0; i < nr_lmbs; i++) {
488         uint64_t addr = i * lmb_size;
489         uint32_t *dynamic_memory = cur_index;
490 
491         if (i >= device_lmb_start) {
492             SpaprDrc *drc;
493 
494             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
495             g_assert(drc);
496 
497             dynamic_memory[0] = cpu_to_be32(addr >> 32);
498             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
499             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
500             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
501             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
502             if (memory_region_present(get_system_memory(), addr)) {
503                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
504             } else {
505                 dynamic_memory[5] = cpu_to_be32(0);
506             }
507         } else {
508             /*
509              * LMB information for RMA, boot time RAM and gap b/n RAM and
510              * device memory region -- all these are marked as reserved
511              * and as having no valid DRC.
512              */
513             dynamic_memory[0] = cpu_to_be32(addr >> 32);
514             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
515             dynamic_memory[2] = cpu_to_be32(0);
516             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
517             dynamic_memory[4] = cpu_to_be32(-1);
518             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
519                                             SPAPR_LMB_FLAGS_DRC_INVALID);
520         }
521 
522         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
523     }
524     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
525     g_free(int_buf);
526     if (ret < 0) {
527         return -1;
528     }
529     return 0;
530 }
531 
532 /*
533  * Adds ibm,dynamic-reconfiguration-memory node.
534  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
535  * of this device tree node.
536  */
537 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
538                                                    void *fdt)
539 {
540     MachineState *machine = MACHINE(spapr);
541     int ret, offset;
542     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
543     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
544                                 cpu_to_be32(lmb_size & 0xffffffff)};
545     MemoryDeviceInfoList *dimms = NULL;
546 
547     /*
548      * Don't create the node if there is no device memory
549      */
550     if (machine->ram_size == machine->maxram_size) {
551         return 0;
552     }
553 
554     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
555 
556     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
557                     sizeof(prop_lmb_size));
558     if (ret < 0) {
559         return ret;
560     }
561 
562     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
563     if (ret < 0) {
564         return ret;
565     }
566 
567     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
568     if (ret < 0) {
569         return ret;
570     }
571 
572     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
573     dimms = qmp_memory_device_list();
574     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
575         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
576     } else {
577         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
578     }
579     qapi_free_MemoryDeviceInfoList(dimms);
580 
581     if (ret < 0) {
582         return ret;
583     }
584 
585     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
586 
587     return ret;
588 }
589 
590 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
591 {
592     MachineState *machine = MACHINE(spapr);
593     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
594     hwaddr mem_start, node_size;
595     int i, nb_nodes = machine->numa_state->num_nodes;
596     NodeInfo *nodes = machine->numa_state->nodes;
597 
598     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
599         if (!nodes[i].node_mem) {
600             continue;
601         }
602         if (mem_start >= machine->ram_size) {
603             node_size = 0;
604         } else {
605             node_size = nodes[i].node_mem;
606             if (node_size > machine->ram_size - mem_start) {
607                 node_size = machine->ram_size - mem_start;
608             }
609         }
610         if (!mem_start) {
611             /* spapr_machine_init() checks for rma_size <= node0_size
612              * already */
613             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
614             mem_start += spapr->rma_size;
615             node_size -= spapr->rma_size;
616         }
617         for ( ; node_size; ) {
618             hwaddr sizetmp = pow2floor(node_size);
619 
620             /* mem_start != 0 here */
621             if (ctzl(mem_start) < ctzl(sizetmp)) {
622                 sizetmp = 1ULL << ctzl(mem_start);
623             }
624 
625             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
626             node_size -= sizetmp;
627             mem_start += sizetmp;
628         }
629     }
630 
631     /* Generate ibm,dynamic-reconfiguration-memory node if required */
632     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
633         int ret;
634 
635         g_assert(smc->dr_lmb_enabled);
636         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
637         if (ret) {
638             return ret;
639         }
640     }
641 
642     return 0;
643 }
644 
645 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
646                          SpaprMachineState *spapr)
647 {
648     MachineState *ms = MACHINE(spapr);
649     PowerPCCPU *cpu = POWERPC_CPU(cs);
650     CPUPPCState *env = &cpu->env;
651     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
652     int index = spapr_get_vcpu_id(cpu);
653     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
654                        0xffffffff, 0xffffffff};
655     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
656         : SPAPR_TIMEBASE_FREQ;
657     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
658     uint32_t page_sizes_prop[64];
659     size_t page_sizes_prop_size;
660     unsigned int smp_threads = ms->smp.threads;
661     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
662     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
663     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
664     SpaprDrc *drc;
665     int drc_index;
666     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
667     int i;
668 
669     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
670     if (drc) {
671         drc_index = spapr_drc_index(drc);
672         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
673     }
674 
675     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
676     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
677 
678     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
679     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
680                            env->dcache_line_size)));
681     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
682                            env->dcache_line_size)));
683     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
684                            env->icache_line_size)));
685     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
686                            env->icache_line_size)));
687 
688     if (pcc->l1_dcache_size) {
689         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
690                                pcc->l1_dcache_size)));
691     } else {
692         warn_report("Unknown L1 dcache size for cpu");
693     }
694     if (pcc->l1_icache_size) {
695         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
696                                pcc->l1_icache_size)));
697     } else {
698         warn_report("Unknown L1 icache size for cpu");
699     }
700 
701     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
702     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
703     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
704     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
705     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
706     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
707 
708     if (ppc_has_spr(cpu, SPR_PURR)) {
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
710     }
711     if (ppc_has_spr(cpu, SPR_PURR)) {
712         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
713     }
714 
715     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
716         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
717                           segs, sizeof(segs))));
718     }
719 
720     /* Advertise VSX (vector extensions) if available
721      *   1               == VMX / Altivec available
722      *   2               == VSX available
723      *
724      * Only CPUs for which we create core types in spapr_cpu_core.c
725      * are possible, and all of those have VMX */
726     if (env->insns_flags & PPC_ALTIVEC) {
727         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
728             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
729         } else {
730             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
731         }
732     }
733 
734     /* Advertise DFP (Decimal Floating Point) if available
735      *   0 / no property == no DFP
736      *   1               == DFP available */
737     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
738         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
739     }
740 
741     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
742                                                       sizeof(page_sizes_prop));
743     if (page_sizes_prop_size) {
744         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
745                           page_sizes_prop, page_sizes_prop_size)));
746     }
747 
748     spapr_dt_pa_features(spapr, cpu, fdt, offset);
749 
750     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
751                            cs->cpu_index / vcpus_per_socket)));
752 
753     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
754                       pft_size_prop, sizeof(pft_size_prop))));
755 
756     if (ms->numa_state->num_nodes > 1) {
757         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
758     }
759 
760     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
761 
762     if (pcc->radix_page_info) {
763         for (i = 0; i < pcc->radix_page_info->count; i++) {
764             radix_AP_encodings[i] =
765                 cpu_to_be32(pcc->radix_page_info->entries[i]);
766         }
767         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
768                           radix_AP_encodings,
769                           pcc->radix_page_info->count *
770                           sizeof(radix_AP_encodings[0]))));
771     }
772 
773     /*
774      * We set this property to let the guest know that it can use the large
775      * decrementer and its width in bits.
776      */
777     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
778         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
779                               pcc->lrg_decr_bits)));
780 }
781 
782 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
783 {
784     CPUState **rev;
785     CPUState *cs;
786     int n_cpus;
787     int cpus_offset;
788     int i;
789 
790     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
791     _FDT(cpus_offset);
792     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
793     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
794 
795     /*
796      * We walk the CPUs in reverse order to ensure that CPU DT nodes
797      * created by fdt_add_subnode() end up in the right order in FDT
798      * for the guest kernel the enumerate the CPUs correctly.
799      *
800      * The CPU list cannot be traversed in reverse order, so we need
801      * to do extra work.
802      */
803     n_cpus = 0;
804     rev = NULL;
805     CPU_FOREACH(cs) {
806         rev = g_renew(CPUState *, rev, n_cpus + 1);
807         rev[n_cpus++] = cs;
808     }
809 
810     for (i = n_cpus - 1; i >= 0; i--) {
811         CPUState *cs = rev[i];
812         PowerPCCPU *cpu = POWERPC_CPU(cs);
813         int index = spapr_get_vcpu_id(cpu);
814         DeviceClass *dc = DEVICE_GET_CLASS(cs);
815         g_autofree char *nodename = NULL;
816         int offset;
817 
818         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
819             continue;
820         }
821 
822         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
823         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
824         _FDT(offset);
825         spapr_dt_cpu(cs, fdt, offset, spapr);
826     }
827 
828     g_free(rev);
829 }
830 
831 static int spapr_dt_rng(void *fdt)
832 {
833     int node;
834     int ret;
835 
836     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
837     if (node <= 0) {
838         return -1;
839     }
840     ret = fdt_setprop_string(fdt, node, "device_type",
841                              "ibm,platform-facilities");
842     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
843     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
844 
845     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
846     if (node <= 0) {
847         return -1;
848     }
849     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
850 
851     return ret ? -1 : 0;
852 }
853 
854 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
855 {
856     MachineState *ms = MACHINE(spapr);
857     int rtas;
858     GString *hypertas = g_string_sized_new(256);
859     GString *qemu_hypertas = g_string_sized_new(256);
860     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
861         memory_region_size(&MACHINE(spapr)->device_memory->mr);
862     uint32_t lrdr_capacity[] = {
863         cpu_to_be32(max_device_addr >> 32),
864         cpu_to_be32(max_device_addr & 0xffffffff),
865         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
866         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
867         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
868     };
869 
870     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
871 
872     /* hypertas */
873     add_str(hypertas, "hcall-pft");
874     add_str(hypertas, "hcall-term");
875     add_str(hypertas, "hcall-dabr");
876     add_str(hypertas, "hcall-interrupt");
877     add_str(hypertas, "hcall-tce");
878     add_str(hypertas, "hcall-vio");
879     add_str(hypertas, "hcall-splpar");
880     add_str(hypertas, "hcall-join");
881     add_str(hypertas, "hcall-bulk");
882     add_str(hypertas, "hcall-set-mode");
883     add_str(hypertas, "hcall-sprg0");
884     add_str(hypertas, "hcall-copy");
885     add_str(hypertas, "hcall-debug");
886     add_str(hypertas, "hcall-vphn");
887     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
888         add_str(hypertas, "hcall-rpt-invalidate");
889     }
890 
891     add_str(qemu_hypertas, "hcall-memop1");
892 
893     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
894         add_str(hypertas, "hcall-multi-tce");
895     }
896 
897     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
898         add_str(hypertas, "hcall-hpt-resize");
899     }
900 
901     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
902                      hypertas->str, hypertas->len));
903     g_string_free(hypertas, TRUE);
904     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
905                      qemu_hypertas->str, qemu_hypertas->len));
906     g_string_free(qemu_hypertas, TRUE);
907 
908     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
909 
910     /*
911      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
912      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
913      *
914      * The system reset requirements are driven by existing Linux and PowerVM
915      * implementation which (contrary to PAPR) saves r3 in the error log
916      * structure like machine check, so Linux expects to find the saved r3
917      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
918      * does not look at the error value).
919      *
920      * System reset interrupts are not subject to interlock like machine
921      * check, so this memory area could be corrupted if the sreset is
922      * interrupted by a machine check (or vice versa) if it was shared. To
923      * prevent this, system reset uses per-CPU areas for the sreset save
924      * area. A system reset that interrupts a system reset handler could
925      * still overwrite this area, but Linux doesn't try to recover in that
926      * case anyway.
927      *
928      * The extra 8 bytes is required because Linux's FWNMI error log check
929      * is off-by-one.
930      *
931      * RTAS_MIN_SIZE is required for the RTAS blob itself.
932      */
933     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
934                           RTAS_ERROR_LOG_MAX +
935                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
936                           sizeof(uint64_t)));
937     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
938                           RTAS_ERROR_LOG_MAX));
939     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
940                           RTAS_EVENT_SCAN_RATE));
941 
942     g_assert(msi_nonbroken);
943     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
944 
945     /*
946      * According to PAPR, rtas ibm,os-term does not guarantee a return
947      * back to the guest cpu.
948      *
949      * While an additional ibm,extended-os-term property indicates
950      * that rtas call return will always occur. Set this property.
951      */
952     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
953 
954     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
955                      lrdr_capacity, sizeof(lrdr_capacity)));
956 
957     spapr_dt_rtas_tokens(fdt, rtas);
958 }
959 
960 /*
961  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
962  * and the XIVE features that the guest may request and thus the valid
963  * values for bytes 23..26 of option vector 5:
964  */
965 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
966                                           int chosen)
967 {
968     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
969 
970     char val[2 * 4] = {
971         23, 0x00, /* XICS / XIVE mode */
972         24, 0x00, /* Hash/Radix, filled in below. */
973         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
974         26, 0x40, /* Radix options: GTSE == yes. */
975     };
976 
977     if (spapr->irq->xics && spapr->irq->xive) {
978         val[1] = SPAPR_OV5_XIVE_BOTH;
979     } else if (spapr->irq->xive) {
980         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
981     } else {
982         assert(spapr->irq->xics);
983         val[1] = SPAPR_OV5_XIVE_LEGACY;
984     }
985 
986     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
987                           first_ppc_cpu->compat_pvr)) {
988         /*
989          * If we're in a pre POWER9 compat mode then the guest should
990          * do hash and use the legacy interrupt mode
991          */
992         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
993         val[3] = 0x00; /* Hash */
994         spapr_check_mmu_mode(false);
995     } else if (kvm_enabled()) {
996         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
997             val[3] = 0x80; /* OV5_MMU_BOTH */
998         } else if (kvmppc_has_cap_mmu_radix()) {
999             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1000         } else {
1001             val[3] = 0x00; /* Hash */
1002         }
1003     } else {
1004         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1005         val[3] = 0xC0;
1006     }
1007     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1008                      val, sizeof(val)));
1009 }
1010 
1011 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1012 {
1013     MachineState *machine = MACHINE(spapr);
1014     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1015     int chosen;
1016 
1017     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1018 
1019     if (reset) {
1020         const char *boot_device = spapr->boot_device;
1021         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1022         size_t cb = 0;
1023         char *bootlist = get_boot_devices_list(&cb);
1024 
1025         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1026             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1027                                     machine->kernel_cmdline));
1028         }
1029 
1030         if (spapr->initrd_size) {
1031             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1032                                   spapr->initrd_base));
1033             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1034                                   spapr->initrd_base + spapr->initrd_size));
1035         }
1036 
1037         if (spapr->kernel_size) {
1038             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1039                                   cpu_to_be64(spapr->kernel_size) };
1040 
1041             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1042                          &kprop, sizeof(kprop)));
1043             if (spapr->kernel_le) {
1044                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1045             }
1046         }
1047         if (boot_menu) {
1048             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1049         }
1050         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1051         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1052         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1053 
1054         if (cb && bootlist) {
1055             int i;
1056 
1057             for (i = 0; i < cb; i++) {
1058                 if (bootlist[i] == '\n') {
1059                     bootlist[i] = ' ';
1060                 }
1061             }
1062             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1063         }
1064 
1065         if (boot_device && strlen(boot_device)) {
1066             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1067         }
1068 
1069         if (!spapr->has_graphics && stdout_path) {
1070             /*
1071              * "linux,stdout-path" and "stdout" properties are
1072              * deprecated by linux kernel. New platforms should only
1073              * use the "stdout-path" property. Set the new property
1074              * and continue using older property to remain compatible
1075              * with the existing firmware.
1076              */
1077             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1078             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1079         }
1080 
1081         /*
1082          * We can deal with BAR reallocation just fine, advertise it
1083          * to the guest
1084          */
1085         if (smc->linux_pci_probe) {
1086             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1087         }
1088 
1089         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1090 
1091         g_free(stdout_path);
1092         g_free(bootlist);
1093     }
1094 
1095     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1096 }
1097 
1098 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1099 {
1100     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1101      * KVM to work under pHyp with some guest co-operation */
1102     int hypervisor;
1103     uint8_t hypercall[16];
1104 
1105     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1106     /* indicate KVM hypercall interface */
1107     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1108     if (kvmppc_has_cap_fixup_hcalls()) {
1109         /*
1110          * Older KVM versions with older guest kernels were broken
1111          * with the magic page, don't allow the guest to map it.
1112          */
1113         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1114                                   sizeof(hypercall))) {
1115             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1116                              hypercall, sizeof(hypercall)));
1117         }
1118     }
1119 }
1120 
1121 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1122 {
1123     MachineState *machine = MACHINE(spapr);
1124     MachineClass *mc = MACHINE_GET_CLASS(machine);
1125     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1126     uint32_t root_drc_type_mask = 0;
1127     int ret;
1128     void *fdt;
1129     SpaprPhbState *phb;
1130     char *buf;
1131 
1132     fdt = g_malloc0(space);
1133     _FDT((fdt_create_empty_tree(fdt, space)));
1134 
1135     /* Root node */
1136     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1137     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1138     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1139 
1140     /* Guest UUID & Name*/
1141     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1142     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1143     if (qemu_uuid_set) {
1144         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1145     }
1146     g_free(buf);
1147 
1148     if (qemu_get_vm_name()) {
1149         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1150                                 qemu_get_vm_name()));
1151     }
1152 
1153     /* Host Model & Serial Number */
1154     if (spapr->host_model) {
1155         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1156     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1157         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1158         g_free(buf);
1159     }
1160 
1161     if (spapr->host_serial) {
1162         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1163     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1164         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1165         g_free(buf);
1166     }
1167 
1168     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1169     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1170 
1171     /* /interrupt controller */
1172     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1173 
1174     ret = spapr_dt_memory(spapr, fdt);
1175     if (ret < 0) {
1176         error_report("couldn't setup memory nodes in fdt");
1177         exit(1);
1178     }
1179 
1180     /* /vdevice */
1181     spapr_dt_vdevice(spapr->vio_bus, fdt);
1182 
1183     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1184         ret = spapr_dt_rng(fdt);
1185         if (ret < 0) {
1186             error_report("could not set up rng device in the fdt");
1187             exit(1);
1188         }
1189     }
1190 
1191     QLIST_FOREACH(phb, &spapr->phbs, list) {
1192         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1193         if (ret < 0) {
1194             error_report("couldn't setup PCI devices in fdt");
1195             exit(1);
1196         }
1197     }
1198 
1199     spapr_dt_cpus(fdt, spapr);
1200 
1201     /* ibm,drc-indexes and friends */
1202     if (smc->dr_lmb_enabled) {
1203         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1204     }
1205     if (smc->dr_phb_enabled) {
1206         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1207     }
1208     if (mc->nvdimm_supported) {
1209         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1210     }
1211     if (root_drc_type_mask) {
1212         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1213     }
1214 
1215     if (mc->has_hotpluggable_cpus) {
1216         int offset = fdt_path_offset(fdt, "/cpus");
1217         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1218         if (ret < 0) {
1219             error_report("Couldn't set up CPU DR device tree properties");
1220             exit(1);
1221         }
1222     }
1223 
1224     /* /event-sources */
1225     spapr_dt_events(spapr, fdt);
1226 
1227     /* /rtas */
1228     spapr_dt_rtas(spapr, fdt);
1229 
1230     /* /chosen */
1231     spapr_dt_chosen(spapr, fdt, reset);
1232 
1233     /* /hypervisor */
1234     if (kvm_enabled()) {
1235         spapr_dt_hypervisor(spapr, fdt);
1236     }
1237 
1238     /* Build memory reserve map */
1239     if (reset) {
1240         if (spapr->kernel_size) {
1241             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1242                                   spapr->kernel_size)));
1243         }
1244         if (spapr->initrd_size) {
1245             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1246                                   spapr->initrd_size)));
1247         }
1248     }
1249 
1250     /* NVDIMM devices */
1251     if (mc->nvdimm_supported) {
1252         spapr_dt_persistent_memory(spapr, fdt);
1253     }
1254 
1255     return fdt;
1256 }
1257 
1258 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1259 {
1260     SpaprMachineState *spapr = opaque;
1261 
1262     return (addr & 0x0fffffff) + spapr->kernel_addr;
1263 }
1264 
1265 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1266                                     PowerPCCPU *cpu)
1267 {
1268     CPUPPCState *env = &cpu->env;
1269 
1270     /* The TCG path should also be holding the BQL at this point */
1271     g_assert(qemu_mutex_iothread_locked());
1272 
1273     if (msr_pr) {
1274         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1275         env->gpr[3] = H_PRIVILEGE;
1276     } else {
1277         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1278     }
1279 }
1280 
1281 struct LPCRSyncState {
1282     target_ulong value;
1283     target_ulong mask;
1284 };
1285 
1286 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1287 {
1288     struct LPCRSyncState *s = arg.host_ptr;
1289     PowerPCCPU *cpu = POWERPC_CPU(cs);
1290     CPUPPCState *env = &cpu->env;
1291     target_ulong lpcr;
1292 
1293     cpu_synchronize_state(cs);
1294     lpcr = env->spr[SPR_LPCR];
1295     lpcr &= ~s->mask;
1296     lpcr |= s->value;
1297     ppc_store_lpcr(cpu, lpcr);
1298 }
1299 
1300 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1301 {
1302     CPUState *cs;
1303     struct LPCRSyncState s = {
1304         .value = value,
1305         .mask = mask
1306     };
1307     CPU_FOREACH(cs) {
1308         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1309     }
1310 }
1311 
1312 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1313 {
1314     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1315 
1316     /* Copy PATE1:GR into PATE0:HR */
1317     entry->dw0 = spapr->patb_entry & PATE0_HR;
1318     entry->dw1 = spapr->patb_entry;
1319 }
1320 
1321 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1322 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1323 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1324 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1325 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1326 
1327 /*
1328  * Get the fd to access the kernel htab, re-opening it if necessary
1329  */
1330 static int get_htab_fd(SpaprMachineState *spapr)
1331 {
1332     Error *local_err = NULL;
1333 
1334     if (spapr->htab_fd >= 0) {
1335         return spapr->htab_fd;
1336     }
1337 
1338     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1339     if (spapr->htab_fd < 0) {
1340         error_report_err(local_err);
1341     }
1342 
1343     return spapr->htab_fd;
1344 }
1345 
1346 void close_htab_fd(SpaprMachineState *spapr)
1347 {
1348     if (spapr->htab_fd >= 0) {
1349         close(spapr->htab_fd);
1350     }
1351     spapr->htab_fd = -1;
1352 }
1353 
1354 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1355 {
1356     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1357 
1358     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1359 }
1360 
1361 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1362 {
1363     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1364 
1365     assert(kvm_enabled());
1366 
1367     if (!spapr->htab) {
1368         return 0;
1369     }
1370 
1371     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1372 }
1373 
1374 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1375                                                 hwaddr ptex, int n)
1376 {
1377     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1378     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1379 
1380     if (!spapr->htab) {
1381         /*
1382          * HTAB is controlled by KVM. Fetch into temporary buffer
1383          */
1384         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1385         kvmppc_read_hptes(hptes, ptex, n);
1386         return hptes;
1387     }
1388 
1389     /*
1390      * HTAB is controlled by QEMU. Just point to the internally
1391      * accessible PTEG.
1392      */
1393     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1394 }
1395 
1396 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1397                               const ppc_hash_pte64_t *hptes,
1398                               hwaddr ptex, int n)
1399 {
1400     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1401 
1402     if (!spapr->htab) {
1403         g_free((void *)hptes);
1404     }
1405 
1406     /* Nothing to do for qemu managed HPT */
1407 }
1408 
1409 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1410                       uint64_t pte0, uint64_t pte1)
1411 {
1412     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1413     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1414 
1415     if (!spapr->htab) {
1416         kvmppc_write_hpte(ptex, pte0, pte1);
1417     } else {
1418         if (pte0 & HPTE64_V_VALID) {
1419             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1420             /*
1421              * When setting valid, we write PTE1 first. This ensures
1422              * proper synchronization with the reading code in
1423              * ppc_hash64_pteg_search()
1424              */
1425             smp_wmb();
1426             stq_p(spapr->htab + offset, pte0);
1427         } else {
1428             stq_p(spapr->htab + offset, pte0);
1429             /*
1430              * When clearing it we set PTE0 first. This ensures proper
1431              * synchronization with the reading code in
1432              * ppc_hash64_pteg_search()
1433              */
1434             smp_wmb();
1435             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1436         }
1437     }
1438 }
1439 
1440 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1441                              uint64_t pte1)
1442 {
1443     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1444     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1445 
1446     if (!spapr->htab) {
1447         /* There should always be a hash table when this is called */
1448         error_report("spapr_hpte_set_c called with no hash table !");
1449         return;
1450     }
1451 
1452     /* The HW performs a non-atomic byte update */
1453     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1454 }
1455 
1456 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1457                              uint64_t pte1)
1458 {
1459     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1460     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1461 
1462     if (!spapr->htab) {
1463         /* There should always be a hash table when this is called */
1464         error_report("spapr_hpte_set_r called with no hash table !");
1465         return;
1466     }
1467 
1468     /* The HW performs a non-atomic byte update */
1469     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1470 }
1471 
1472 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1473 {
1474     int shift;
1475 
1476     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1477      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1478      * that's much more than is needed for Linux guests */
1479     shift = ctz64(pow2ceil(ramsize)) - 7;
1480     shift = MAX(shift, 18); /* Minimum architected size */
1481     shift = MIN(shift, 46); /* Maximum architected size */
1482     return shift;
1483 }
1484 
1485 void spapr_free_hpt(SpaprMachineState *spapr)
1486 {
1487     g_free(spapr->htab);
1488     spapr->htab = NULL;
1489     spapr->htab_shift = 0;
1490     close_htab_fd(spapr);
1491 }
1492 
1493 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1494 {
1495     ERRP_GUARD();
1496     long rc;
1497 
1498     /* Clean up any HPT info from a previous boot */
1499     spapr_free_hpt(spapr);
1500 
1501     rc = kvmppc_reset_htab(shift);
1502 
1503     if (rc == -EOPNOTSUPP) {
1504         error_setg(errp, "HPT not supported in nested guests");
1505         return -EOPNOTSUPP;
1506     }
1507 
1508     if (rc < 0) {
1509         /* kernel-side HPT needed, but couldn't allocate one */
1510         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1511                          shift);
1512         error_append_hint(errp, "Try smaller maxmem?\n");
1513         return -errno;
1514     } else if (rc > 0) {
1515         /* kernel-side HPT allocated */
1516         if (rc != shift) {
1517             error_setg(errp,
1518                        "Requested order %d HPT, but kernel allocated order %ld",
1519                        shift, rc);
1520             error_append_hint(errp, "Try smaller maxmem?\n");
1521             return -ENOSPC;
1522         }
1523 
1524         spapr->htab_shift = shift;
1525         spapr->htab = NULL;
1526     } else {
1527         /* kernel-side HPT not needed, allocate in userspace instead */
1528         size_t size = 1ULL << shift;
1529         int i;
1530 
1531         spapr->htab = qemu_memalign(size, size);
1532         memset(spapr->htab, 0, size);
1533         spapr->htab_shift = shift;
1534 
1535         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1536             DIRTY_HPTE(HPTE(spapr->htab, i));
1537         }
1538     }
1539     /* We're setting up a hash table, so that means we're not radix */
1540     spapr->patb_entry = 0;
1541     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1542     return 0;
1543 }
1544 
1545 void spapr_setup_hpt(SpaprMachineState *spapr)
1546 {
1547     int hpt_shift;
1548 
1549     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1550         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1551     } else {
1552         uint64_t current_ram_size;
1553 
1554         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1555         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1556     }
1557     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1558 
1559     if (kvm_enabled()) {
1560         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1561 
1562         /* Check our RMA fits in the possible VRMA */
1563         if (vrma_limit < spapr->rma_size) {
1564             error_report("Unable to create %" HWADDR_PRIu
1565                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1566                          spapr->rma_size / MiB, vrma_limit / MiB);
1567             exit(EXIT_FAILURE);
1568         }
1569     }
1570 }
1571 
1572 void spapr_check_mmu_mode(bool guest_radix)
1573 {
1574     if (guest_radix) {
1575         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1576             error_report("Guest requested unavailable MMU mode (radix).");
1577             exit(EXIT_FAILURE);
1578         }
1579     } else {
1580         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1581             && !kvmppc_has_cap_mmu_hash_v3()) {
1582             error_report("Guest requested unavailable MMU mode (hash).");
1583             exit(EXIT_FAILURE);
1584         }
1585     }
1586 }
1587 
1588 static void spapr_machine_reset(MachineState *machine)
1589 {
1590     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1591     PowerPCCPU *first_ppc_cpu;
1592     hwaddr fdt_addr;
1593     void *fdt;
1594     int rc;
1595 
1596     pef_kvm_reset(machine->cgs, &error_fatal);
1597     spapr_caps_apply(spapr);
1598 
1599     first_ppc_cpu = POWERPC_CPU(first_cpu);
1600     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1601         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1602                               spapr->max_compat_pvr)) {
1603         /*
1604          * If using KVM with radix mode available, VCPUs can be started
1605          * without a HPT because KVM will start them in radix mode.
1606          * Set the GR bit in PATE so that we know there is no HPT.
1607          */
1608         spapr->patb_entry = PATE1_GR;
1609         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1610     } else {
1611         spapr_setup_hpt(spapr);
1612     }
1613 
1614     qemu_devices_reset();
1615 
1616     spapr_ovec_cleanup(spapr->ov5_cas);
1617     spapr->ov5_cas = spapr_ovec_new();
1618 
1619     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1620 
1621     /*
1622      * This is fixing some of the default configuration of the XIVE
1623      * devices. To be called after the reset of the machine devices.
1624      */
1625     spapr_irq_reset(spapr, &error_fatal);
1626 
1627     /*
1628      * There is no CAS under qtest. Simulate one to please the code that
1629      * depends on spapr->ov5_cas. This is especially needed to test device
1630      * unplug, so we do that before resetting the DRCs.
1631      */
1632     if (qtest_enabled()) {
1633         spapr_ovec_cleanup(spapr->ov5_cas);
1634         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1635     }
1636 
1637     spapr_nvdimm_finish_flushes();
1638 
1639     /* DRC reset may cause a device to be unplugged. This will cause troubles
1640      * if this device is used by another device (eg, a running vhost backend
1641      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1642      * situations, we reset DRCs after all devices have been reset.
1643      */
1644     spapr_drc_reset_all(spapr);
1645 
1646     spapr_clear_pending_events(spapr);
1647 
1648     /*
1649      * We place the device tree just below either the top of the RMA,
1650      * or just below 2GB, whichever is lower, so that it can be
1651      * processed with 32-bit real mode code if necessary
1652      */
1653     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1654 
1655     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1656     if (spapr->vof) {
1657         spapr_vof_reset(spapr, fdt, &error_fatal);
1658         /*
1659          * Do not pack the FDT as the client may change properties.
1660          * VOF client does not expect the FDT so we do not load it to the VM.
1661          */
1662     } else {
1663         rc = fdt_pack(fdt);
1664         /* Should only fail if we've built a corrupted tree */
1665         assert(rc == 0);
1666 
1667         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1668                                   0, fdt_addr, 0);
1669         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1670     }
1671     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1672 
1673     g_free(spapr->fdt_blob);
1674     spapr->fdt_size = fdt_totalsize(fdt);
1675     spapr->fdt_initial_size = spapr->fdt_size;
1676     spapr->fdt_blob = fdt;
1677 
1678     /* Set up the entry state */
1679     first_ppc_cpu->env.gpr[5] = 0;
1680 
1681     spapr->fwnmi_system_reset_addr = -1;
1682     spapr->fwnmi_machine_check_addr = -1;
1683     spapr->fwnmi_machine_check_interlock = -1;
1684 
1685     /* Signal all vCPUs waiting on this condition */
1686     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1687 
1688     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1689 }
1690 
1691 static void spapr_create_nvram(SpaprMachineState *spapr)
1692 {
1693     DeviceState *dev = qdev_new("spapr-nvram");
1694     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1695 
1696     if (dinfo) {
1697         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1698                                 &error_fatal);
1699     }
1700 
1701     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1702 
1703     spapr->nvram = (struct SpaprNvram *)dev;
1704 }
1705 
1706 static void spapr_rtc_create(SpaprMachineState *spapr)
1707 {
1708     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1709                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1710                                        &error_fatal, NULL);
1711     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1712     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1713                               "date");
1714 }
1715 
1716 /* Returns whether we want to use VGA or not */
1717 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1718 {
1719     switch (vga_interface_type) {
1720     case VGA_NONE:
1721         return false;
1722     case VGA_DEVICE:
1723         return true;
1724     case VGA_STD:
1725     case VGA_VIRTIO:
1726     case VGA_CIRRUS:
1727         return pci_vga_init(pci_bus) != NULL;
1728     default:
1729         error_setg(errp,
1730                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1731         return false;
1732     }
1733 }
1734 
1735 static int spapr_pre_load(void *opaque)
1736 {
1737     int rc;
1738 
1739     rc = spapr_caps_pre_load(opaque);
1740     if (rc) {
1741         return rc;
1742     }
1743 
1744     return 0;
1745 }
1746 
1747 static int spapr_post_load(void *opaque, int version_id)
1748 {
1749     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1750     int err = 0;
1751 
1752     err = spapr_caps_post_migration(spapr);
1753     if (err) {
1754         return err;
1755     }
1756 
1757     /*
1758      * In earlier versions, there was no separate qdev for the PAPR
1759      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1760      * So when migrating from those versions, poke the incoming offset
1761      * value into the RTC device
1762      */
1763     if (version_id < 3) {
1764         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1765         if (err) {
1766             return err;
1767         }
1768     }
1769 
1770     if (kvm_enabled() && spapr->patb_entry) {
1771         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1772         bool radix = !!(spapr->patb_entry & PATE1_GR);
1773         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1774 
1775         /*
1776          * Update LPCR:HR and UPRT as they may not be set properly in
1777          * the stream
1778          */
1779         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1780                             LPCR_HR | LPCR_UPRT);
1781 
1782         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1783         if (err) {
1784             error_report("Process table config unsupported by the host");
1785             return -EINVAL;
1786         }
1787     }
1788 
1789     err = spapr_irq_post_load(spapr, version_id);
1790     if (err) {
1791         return err;
1792     }
1793 
1794     return err;
1795 }
1796 
1797 static int spapr_pre_save(void *opaque)
1798 {
1799     int rc;
1800 
1801     rc = spapr_caps_pre_save(opaque);
1802     if (rc) {
1803         return rc;
1804     }
1805 
1806     return 0;
1807 }
1808 
1809 static bool version_before_3(void *opaque, int version_id)
1810 {
1811     return version_id < 3;
1812 }
1813 
1814 static bool spapr_pending_events_needed(void *opaque)
1815 {
1816     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1817     return !QTAILQ_EMPTY(&spapr->pending_events);
1818 }
1819 
1820 static const VMStateDescription vmstate_spapr_event_entry = {
1821     .name = "spapr_event_log_entry",
1822     .version_id = 1,
1823     .minimum_version_id = 1,
1824     .fields = (VMStateField[]) {
1825         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1826         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1827         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1828                                      NULL, extended_length),
1829         VMSTATE_END_OF_LIST()
1830     },
1831 };
1832 
1833 static const VMStateDescription vmstate_spapr_pending_events = {
1834     .name = "spapr_pending_events",
1835     .version_id = 1,
1836     .minimum_version_id = 1,
1837     .needed = spapr_pending_events_needed,
1838     .fields = (VMStateField[]) {
1839         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1840                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1841         VMSTATE_END_OF_LIST()
1842     },
1843 };
1844 
1845 static bool spapr_ov5_cas_needed(void *opaque)
1846 {
1847     SpaprMachineState *spapr = opaque;
1848     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1849     bool cas_needed;
1850 
1851     /* Prior to the introduction of SpaprOptionVector, we had two option
1852      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1853      * Both of these options encode machine topology into the device-tree
1854      * in such a way that the now-booted OS should still be able to interact
1855      * appropriately with QEMU regardless of what options were actually
1856      * negotiatied on the source side.
1857      *
1858      * As such, we can avoid migrating the CAS-negotiated options if these
1859      * are the only options available on the current machine/platform.
1860      * Since these are the only options available for pseries-2.7 and
1861      * earlier, this allows us to maintain old->new/new->old migration
1862      * compatibility.
1863      *
1864      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1865      * via default pseries-2.8 machines and explicit command-line parameters.
1866      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1867      * of the actual CAS-negotiated values to continue working properly. For
1868      * example, availability of memory unplug depends on knowing whether
1869      * OV5_HP_EVT was negotiated via CAS.
1870      *
1871      * Thus, for any cases where the set of available CAS-negotiatable
1872      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1873      * include the CAS-negotiated options in the migration stream, unless
1874      * if they affect boot time behaviour only.
1875      */
1876     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1877     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1878     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1879 
1880     /* We need extra information if we have any bits outside the mask
1881      * defined above */
1882     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1883 
1884     spapr_ovec_cleanup(ov5_mask);
1885 
1886     return cas_needed;
1887 }
1888 
1889 static const VMStateDescription vmstate_spapr_ov5_cas = {
1890     .name = "spapr_option_vector_ov5_cas",
1891     .version_id = 1,
1892     .minimum_version_id = 1,
1893     .needed = spapr_ov5_cas_needed,
1894     .fields = (VMStateField[]) {
1895         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1896                                  vmstate_spapr_ovec, SpaprOptionVector),
1897         VMSTATE_END_OF_LIST()
1898     },
1899 };
1900 
1901 static bool spapr_patb_entry_needed(void *opaque)
1902 {
1903     SpaprMachineState *spapr = opaque;
1904 
1905     return !!spapr->patb_entry;
1906 }
1907 
1908 static const VMStateDescription vmstate_spapr_patb_entry = {
1909     .name = "spapr_patb_entry",
1910     .version_id = 1,
1911     .minimum_version_id = 1,
1912     .needed = spapr_patb_entry_needed,
1913     .fields = (VMStateField[]) {
1914         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1915         VMSTATE_END_OF_LIST()
1916     },
1917 };
1918 
1919 static bool spapr_irq_map_needed(void *opaque)
1920 {
1921     SpaprMachineState *spapr = opaque;
1922 
1923     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1924 }
1925 
1926 static const VMStateDescription vmstate_spapr_irq_map = {
1927     .name = "spapr_irq_map",
1928     .version_id = 1,
1929     .minimum_version_id = 1,
1930     .needed = spapr_irq_map_needed,
1931     .fields = (VMStateField[]) {
1932         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1933         VMSTATE_END_OF_LIST()
1934     },
1935 };
1936 
1937 static bool spapr_dtb_needed(void *opaque)
1938 {
1939     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1940 
1941     return smc->update_dt_enabled;
1942 }
1943 
1944 static int spapr_dtb_pre_load(void *opaque)
1945 {
1946     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1947 
1948     g_free(spapr->fdt_blob);
1949     spapr->fdt_blob = NULL;
1950     spapr->fdt_size = 0;
1951 
1952     return 0;
1953 }
1954 
1955 static const VMStateDescription vmstate_spapr_dtb = {
1956     .name = "spapr_dtb",
1957     .version_id = 1,
1958     .minimum_version_id = 1,
1959     .needed = spapr_dtb_needed,
1960     .pre_load = spapr_dtb_pre_load,
1961     .fields = (VMStateField[]) {
1962         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1963         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1964         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1965                                      fdt_size),
1966         VMSTATE_END_OF_LIST()
1967     },
1968 };
1969 
1970 static bool spapr_fwnmi_needed(void *opaque)
1971 {
1972     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1973 
1974     return spapr->fwnmi_machine_check_addr != -1;
1975 }
1976 
1977 static int spapr_fwnmi_pre_save(void *opaque)
1978 {
1979     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1980 
1981     /*
1982      * Check if machine check handling is in progress and print a
1983      * warning message.
1984      */
1985     if (spapr->fwnmi_machine_check_interlock != -1) {
1986         warn_report("A machine check is being handled during migration. The"
1987                 "handler may run and log hardware error on the destination");
1988     }
1989 
1990     return 0;
1991 }
1992 
1993 static const VMStateDescription vmstate_spapr_fwnmi = {
1994     .name = "spapr_fwnmi",
1995     .version_id = 1,
1996     .minimum_version_id = 1,
1997     .needed = spapr_fwnmi_needed,
1998     .pre_save = spapr_fwnmi_pre_save,
1999     .fields = (VMStateField[]) {
2000         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2001         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2002         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2003         VMSTATE_END_OF_LIST()
2004     },
2005 };
2006 
2007 static const VMStateDescription vmstate_spapr = {
2008     .name = "spapr",
2009     .version_id = 3,
2010     .minimum_version_id = 1,
2011     .pre_load = spapr_pre_load,
2012     .post_load = spapr_post_load,
2013     .pre_save = spapr_pre_save,
2014     .fields = (VMStateField[]) {
2015         /* used to be @next_irq */
2016         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2017 
2018         /* RTC offset */
2019         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2020 
2021         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2022         VMSTATE_END_OF_LIST()
2023     },
2024     .subsections = (const VMStateDescription*[]) {
2025         &vmstate_spapr_ov5_cas,
2026         &vmstate_spapr_patb_entry,
2027         &vmstate_spapr_pending_events,
2028         &vmstate_spapr_cap_htm,
2029         &vmstate_spapr_cap_vsx,
2030         &vmstate_spapr_cap_dfp,
2031         &vmstate_spapr_cap_cfpc,
2032         &vmstate_spapr_cap_sbbc,
2033         &vmstate_spapr_cap_ibs,
2034         &vmstate_spapr_cap_hpt_maxpagesize,
2035         &vmstate_spapr_irq_map,
2036         &vmstate_spapr_cap_nested_kvm_hv,
2037         &vmstate_spapr_dtb,
2038         &vmstate_spapr_cap_large_decr,
2039         &vmstate_spapr_cap_ccf_assist,
2040         &vmstate_spapr_cap_fwnmi,
2041         &vmstate_spapr_fwnmi,
2042         &vmstate_spapr_cap_rpt_invalidate,
2043         NULL
2044     }
2045 };
2046 
2047 static int htab_save_setup(QEMUFile *f, void *opaque)
2048 {
2049     SpaprMachineState *spapr = opaque;
2050 
2051     /* "Iteration" header */
2052     if (!spapr->htab_shift) {
2053         qemu_put_be32(f, -1);
2054     } else {
2055         qemu_put_be32(f, spapr->htab_shift);
2056     }
2057 
2058     if (spapr->htab) {
2059         spapr->htab_save_index = 0;
2060         spapr->htab_first_pass = true;
2061     } else {
2062         if (spapr->htab_shift) {
2063             assert(kvm_enabled());
2064         }
2065     }
2066 
2067 
2068     return 0;
2069 }
2070 
2071 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2072                             int chunkstart, int n_valid, int n_invalid)
2073 {
2074     qemu_put_be32(f, chunkstart);
2075     qemu_put_be16(f, n_valid);
2076     qemu_put_be16(f, n_invalid);
2077     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2078                     HASH_PTE_SIZE_64 * n_valid);
2079 }
2080 
2081 static void htab_save_end_marker(QEMUFile *f)
2082 {
2083     qemu_put_be32(f, 0);
2084     qemu_put_be16(f, 0);
2085     qemu_put_be16(f, 0);
2086 }
2087 
2088 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2089                                  int64_t max_ns)
2090 {
2091     bool has_timeout = max_ns != -1;
2092     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2093     int index = spapr->htab_save_index;
2094     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2095 
2096     assert(spapr->htab_first_pass);
2097 
2098     do {
2099         int chunkstart;
2100 
2101         /* Consume invalid HPTEs */
2102         while ((index < htabslots)
2103                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2104             CLEAN_HPTE(HPTE(spapr->htab, index));
2105             index++;
2106         }
2107 
2108         /* Consume valid HPTEs */
2109         chunkstart = index;
2110         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2111                && HPTE_VALID(HPTE(spapr->htab, index))) {
2112             CLEAN_HPTE(HPTE(spapr->htab, index));
2113             index++;
2114         }
2115 
2116         if (index > chunkstart) {
2117             int n_valid = index - chunkstart;
2118 
2119             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2120 
2121             if (has_timeout &&
2122                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2123                 break;
2124             }
2125         }
2126     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2127 
2128     if (index >= htabslots) {
2129         assert(index == htabslots);
2130         index = 0;
2131         spapr->htab_first_pass = false;
2132     }
2133     spapr->htab_save_index = index;
2134 }
2135 
2136 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2137                                 int64_t max_ns)
2138 {
2139     bool final = max_ns < 0;
2140     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2141     int examined = 0, sent = 0;
2142     int index = spapr->htab_save_index;
2143     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2144 
2145     assert(!spapr->htab_first_pass);
2146 
2147     do {
2148         int chunkstart, invalidstart;
2149 
2150         /* Consume non-dirty HPTEs */
2151         while ((index < htabslots)
2152                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2153             index++;
2154             examined++;
2155         }
2156 
2157         chunkstart = index;
2158         /* Consume valid dirty HPTEs */
2159         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2160                && HPTE_DIRTY(HPTE(spapr->htab, index))
2161                && HPTE_VALID(HPTE(spapr->htab, index))) {
2162             CLEAN_HPTE(HPTE(spapr->htab, index));
2163             index++;
2164             examined++;
2165         }
2166 
2167         invalidstart = index;
2168         /* Consume invalid dirty HPTEs */
2169         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2170                && HPTE_DIRTY(HPTE(spapr->htab, index))
2171                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2172             CLEAN_HPTE(HPTE(spapr->htab, index));
2173             index++;
2174             examined++;
2175         }
2176 
2177         if (index > chunkstart) {
2178             int n_valid = invalidstart - chunkstart;
2179             int n_invalid = index - invalidstart;
2180 
2181             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2182             sent += index - chunkstart;
2183 
2184             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2185                 break;
2186             }
2187         }
2188 
2189         if (examined >= htabslots) {
2190             break;
2191         }
2192 
2193         if (index >= htabslots) {
2194             assert(index == htabslots);
2195             index = 0;
2196         }
2197     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2198 
2199     if (index >= htabslots) {
2200         assert(index == htabslots);
2201         index = 0;
2202     }
2203 
2204     spapr->htab_save_index = index;
2205 
2206     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2207 }
2208 
2209 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2210 #define MAX_KVM_BUF_SIZE    2048
2211 
2212 static int htab_save_iterate(QEMUFile *f, void *opaque)
2213 {
2214     SpaprMachineState *spapr = opaque;
2215     int fd;
2216     int rc = 0;
2217 
2218     /* Iteration header */
2219     if (!spapr->htab_shift) {
2220         qemu_put_be32(f, -1);
2221         return 1;
2222     } else {
2223         qemu_put_be32(f, 0);
2224     }
2225 
2226     if (!spapr->htab) {
2227         assert(kvm_enabled());
2228 
2229         fd = get_htab_fd(spapr);
2230         if (fd < 0) {
2231             return fd;
2232         }
2233 
2234         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2235         if (rc < 0) {
2236             return rc;
2237         }
2238     } else  if (spapr->htab_first_pass) {
2239         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2240     } else {
2241         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2242     }
2243 
2244     htab_save_end_marker(f);
2245 
2246     return rc;
2247 }
2248 
2249 static int htab_save_complete(QEMUFile *f, void *opaque)
2250 {
2251     SpaprMachineState *spapr = opaque;
2252     int fd;
2253 
2254     /* Iteration header */
2255     if (!spapr->htab_shift) {
2256         qemu_put_be32(f, -1);
2257         return 0;
2258     } else {
2259         qemu_put_be32(f, 0);
2260     }
2261 
2262     if (!spapr->htab) {
2263         int rc;
2264 
2265         assert(kvm_enabled());
2266 
2267         fd = get_htab_fd(spapr);
2268         if (fd < 0) {
2269             return fd;
2270         }
2271 
2272         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2273         if (rc < 0) {
2274             return rc;
2275         }
2276     } else {
2277         if (spapr->htab_first_pass) {
2278             htab_save_first_pass(f, spapr, -1);
2279         }
2280         htab_save_later_pass(f, spapr, -1);
2281     }
2282 
2283     /* End marker */
2284     htab_save_end_marker(f);
2285 
2286     return 0;
2287 }
2288 
2289 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2290 {
2291     SpaprMachineState *spapr = opaque;
2292     uint32_t section_hdr;
2293     int fd = -1;
2294     Error *local_err = NULL;
2295 
2296     if (version_id < 1 || version_id > 1) {
2297         error_report("htab_load() bad version");
2298         return -EINVAL;
2299     }
2300 
2301     section_hdr = qemu_get_be32(f);
2302 
2303     if (section_hdr == -1) {
2304         spapr_free_hpt(spapr);
2305         return 0;
2306     }
2307 
2308     if (section_hdr) {
2309         int ret;
2310 
2311         /* First section gives the htab size */
2312         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2313         if (ret < 0) {
2314             error_report_err(local_err);
2315             return ret;
2316         }
2317         return 0;
2318     }
2319 
2320     if (!spapr->htab) {
2321         assert(kvm_enabled());
2322 
2323         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2324         if (fd < 0) {
2325             error_report_err(local_err);
2326             return fd;
2327         }
2328     }
2329 
2330     while (true) {
2331         uint32_t index;
2332         uint16_t n_valid, n_invalid;
2333 
2334         index = qemu_get_be32(f);
2335         n_valid = qemu_get_be16(f);
2336         n_invalid = qemu_get_be16(f);
2337 
2338         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2339             /* End of Stream */
2340             break;
2341         }
2342 
2343         if ((index + n_valid + n_invalid) >
2344             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2345             /* Bad index in stream */
2346             error_report(
2347                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2348                 index, n_valid, n_invalid, spapr->htab_shift);
2349             return -EINVAL;
2350         }
2351 
2352         if (spapr->htab) {
2353             if (n_valid) {
2354                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2355                                 HASH_PTE_SIZE_64 * n_valid);
2356             }
2357             if (n_invalid) {
2358                 memset(HPTE(spapr->htab, index + n_valid), 0,
2359                        HASH_PTE_SIZE_64 * n_invalid);
2360             }
2361         } else {
2362             int rc;
2363 
2364             assert(fd >= 0);
2365 
2366             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2367                                         &local_err);
2368             if (rc < 0) {
2369                 error_report_err(local_err);
2370                 return rc;
2371             }
2372         }
2373     }
2374 
2375     if (!spapr->htab) {
2376         assert(fd >= 0);
2377         close(fd);
2378     }
2379 
2380     return 0;
2381 }
2382 
2383 static void htab_save_cleanup(void *opaque)
2384 {
2385     SpaprMachineState *spapr = opaque;
2386 
2387     close_htab_fd(spapr);
2388 }
2389 
2390 static SaveVMHandlers savevm_htab_handlers = {
2391     .save_setup = htab_save_setup,
2392     .save_live_iterate = htab_save_iterate,
2393     .save_live_complete_precopy = htab_save_complete,
2394     .save_cleanup = htab_save_cleanup,
2395     .load_state = htab_load,
2396 };
2397 
2398 static void spapr_boot_set(void *opaque, const char *boot_device,
2399                            Error **errp)
2400 {
2401     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2402 
2403     g_free(spapr->boot_device);
2404     spapr->boot_device = g_strdup(boot_device);
2405 }
2406 
2407 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2408 {
2409     MachineState *machine = MACHINE(spapr);
2410     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2411     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2412     int i;
2413 
2414     for (i = 0; i < nr_lmbs; i++) {
2415         uint64_t addr;
2416 
2417         addr = i * lmb_size + machine->device_memory->base;
2418         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2419                                addr / lmb_size);
2420     }
2421 }
2422 
2423 /*
2424  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2425  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2426  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2427  */
2428 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2429 {
2430     int i;
2431 
2432     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2433         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2434                    " is not aligned to %" PRIu64 " MiB",
2435                    machine->ram_size,
2436                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2437         return;
2438     }
2439 
2440     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2441         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2442                    " is not aligned to %" PRIu64 " MiB",
2443                    machine->ram_size,
2444                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2445         return;
2446     }
2447 
2448     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2449         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2450             error_setg(errp,
2451                        "Node %d memory size 0x%" PRIx64
2452                        " is not aligned to %" PRIu64 " MiB",
2453                        i, machine->numa_state->nodes[i].node_mem,
2454                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2455             return;
2456         }
2457     }
2458 }
2459 
2460 /* find cpu slot in machine->possible_cpus by core_id */
2461 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2462 {
2463     int index = id / ms->smp.threads;
2464 
2465     if (index >= ms->possible_cpus->len) {
2466         return NULL;
2467     }
2468     if (idx) {
2469         *idx = index;
2470     }
2471     return &ms->possible_cpus->cpus[index];
2472 }
2473 
2474 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2475 {
2476     MachineState *ms = MACHINE(spapr);
2477     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2478     Error *local_err = NULL;
2479     bool vsmt_user = !!spapr->vsmt;
2480     int kvm_smt = kvmppc_smt_threads();
2481     int ret;
2482     unsigned int smp_threads = ms->smp.threads;
2483 
2484     if (!kvm_enabled() && (smp_threads > 1)) {
2485         error_setg(errp, "TCG cannot support more than 1 thread/core "
2486                    "on a pseries machine");
2487         return;
2488     }
2489     if (!is_power_of_2(smp_threads)) {
2490         error_setg(errp, "Cannot support %d threads/core on a pseries "
2491                    "machine because it must be a power of 2", smp_threads);
2492         return;
2493     }
2494 
2495     /* Detemine the VSMT mode to use: */
2496     if (vsmt_user) {
2497         if (spapr->vsmt < smp_threads) {
2498             error_setg(errp, "Cannot support VSMT mode %d"
2499                        " because it must be >= threads/core (%d)",
2500                        spapr->vsmt, smp_threads);
2501             return;
2502         }
2503         /* In this case, spapr->vsmt has been set by the command line */
2504     } else if (!smc->smp_threads_vsmt) {
2505         /*
2506          * Default VSMT value is tricky, because we need it to be as
2507          * consistent as possible (for migration), but this requires
2508          * changing it for at least some existing cases.  We pick 8 as
2509          * the value that we'd get with KVM on POWER8, the
2510          * overwhelmingly common case in production systems.
2511          */
2512         spapr->vsmt = MAX(8, smp_threads);
2513     } else {
2514         spapr->vsmt = smp_threads;
2515     }
2516 
2517     /* KVM: If necessary, set the SMT mode: */
2518     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2519         ret = kvmppc_set_smt_threads(spapr->vsmt);
2520         if (ret) {
2521             /* Looks like KVM isn't able to change VSMT mode */
2522             error_setg(&local_err,
2523                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2524                        spapr->vsmt, ret);
2525             /* We can live with that if the default one is big enough
2526              * for the number of threads, and a submultiple of the one
2527              * we want.  In this case we'll waste some vcpu ids, but
2528              * behaviour will be correct */
2529             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2530                 warn_report_err(local_err);
2531             } else {
2532                 if (!vsmt_user) {
2533                     error_append_hint(&local_err,
2534                                       "On PPC, a VM with %d threads/core"
2535                                       " on a host with %d threads/core"
2536                                       " requires the use of VSMT mode %d.\n",
2537                                       smp_threads, kvm_smt, spapr->vsmt);
2538                 }
2539                 kvmppc_error_append_smt_possible_hint(&local_err);
2540                 error_propagate(errp, local_err);
2541             }
2542         }
2543     }
2544     /* else TCG: nothing to do currently */
2545 }
2546 
2547 static void spapr_init_cpus(SpaprMachineState *spapr)
2548 {
2549     MachineState *machine = MACHINE(spapr);
2550     MachineClass *mc = MACHINE_GET_CLASS(machine);
2551     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2552     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2553     const CPUArchIdList *possible_cpus;
2554     unsigned int smp_cpus = machine->smp.cpus;
2555     unsigned int smp_threads = machine->smp.threads;
2556     unsigned int max_cpus = machine->smp.max_cpus;
2557     int boot_cores_nr = smp_cpus / smp_threads;
2558     int i;
2559 
2560     possible_cpus = mc->possible_cpu_arch_ids(machine);
2561     if (mc->has_hotpluggable_cpus) {
2562         if (smp_cpus % smp_threads) {
2563             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2564                          smp_cpus, smp_threads);
2565             exit(1);
2566         }
2567         if (max_cpus % smp_threads) {
2568             error_report("max_cpus (%u) must be multiple of threads (%u)",
2569                          max_cpus, smp_threads);
2570             exit(1);
2571         }
2572     } else {
2573         if (max_cpus != smp_cpus) {
2574             error_report("This machine version does not support CPU hotplug");
2575             exit(1);
2576         }
2577         boot_cores_nr = possible_cpus->len;
2578     }
2579 
2580     if (smc->pre_2_10_has_unused_icps) {
2581         int i;
2582 
2583         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2584             /* Dummy entries get deregistered when real ICPState objects
2585              * are registered during CPU core hotplug.
2586              */
2587             pre_2_10_vmstate_register_dummy_icp(i);
2588         }
2589     }
2590 
2591     for (i = 0; i < possible_cpus->len; i++) {
2592         int core_id = i * smp_threads;
2593 
2594         if (mc->has_hotpluggable_cpus) {
2595             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2596                                    spapr_vcpu_id(spapr, core_id));
2597         }
2598 
2599         if (i < boot_cores_nr) {
2600             Object *core  = object_new(type);
2601             int nr_threads = smp_threads;
2602 
2603             /* Handle the partially filled core for older machine types */
2604             if ((i + 1) * smp_threads >= smp_cpus) {
2605                 nr_threads = smp_cpus - i * smp_threads;
2606             }
2607 
2608             object_property_set_int(core, "nr-threads", nr_threads,
2609                                     &error_fatal);
2610             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2611                                     &error_fatal);
2612             qdev_realize(DEVICE(core), NULL, &error_fatal);
2613 
2614             object_unref(core);
2615         }
2616     }
2617 }
2618 
2619 static PCIHostState *spapr_create_default_phb(void)
2620 {
2621     DeviceState *dev;
2622 
2623     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2624     qdev_prop_set_uint32(dev, "index", 0);
2625     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2626 
2627     return PCI_HOST_BRIDGE(dev);
2628 }
2629 
2630 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2631 {
2632     MachineState *machine = MACHINE(spapr);
2633     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2634     hwaddr rma_size = machine->ram_size;
2635     hwaddr node0_size = spapr_node0_size(machine);
2636 
2637     /* RMA has to fit in the first NUMA node */
2638     rma_size = MIN(rma_size, node0_size);
2639 
2640     /*
2641      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2642      * never exceed that
2643      */
2644     rma_size = MIN(rma_size, 1 * TiB);
2645 
2646     /*
2647      * Clamp the RMA size based on machine type.  This is for
2648      * migration compatibility with older qemu versions, which limited
2649      * the RMA size for complicated and mostly bad reasons.
2650      */
2651     if (smc->rma_limit) {
2652         rma_size = MIN(rma_size, smc->rma_limit);
2653     }
2654 
2655     if (rma_size < MIN_RMA_SLOF) {
2656         error_setg(errp,
2657                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2658                    "ldMiB guest RMA (Real Mode Area memory)",
2659                    MIN_RMA_SLOF / MiB);
2660         return 0;
2661     }
2662 
2663     return rma_size;
2664 }
2665 
2666 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2667 {
2668     MachineState *machine = MACHINE(spapr);
2669     int i;
2670 
2671     for (i = 0; i < machine->ram_slots; i++) {
2672         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2673     }
2674 }
2675 
2676 /* pSeries LPAR / sPAPR hardware init */
2677 static void spapr_machine_init(MachineState *machine)
2678 {
2679     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2680     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2681     MachineClass *mc = MACHINE_GET_CLASS(machine);
2682     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2683     const char *bios_name = machine->firmware ?: bios_default;
2684     const char *kernel_filename = machine->kernel_filename;
2685     const char *initrd_filename = machine->initrd_filename;
2686     PCIHostState *phb;
2687     int i;
2688     MemoryRegion *sysmem = get_system_memory();
2689     long load_limit, fw_size;
2690     char *filename;
2691     Error *resize_hpt_err = NULL;
2692 
2693     /*
2694      * if Secure VM (PEF) support is configured, then initialize it
2695      */
2696     pef_kvm_init(machine->cgs, &error_fatal);
2697 
2698     msi_nonbroken = true;
2699 
2700     QLIST_INIT(&spapr->phbs);
2701     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2702 
2703     /* Determine capabilities to run with */
2704     spapr_caps_init(spapr);
2705 
2706     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2707     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2708         /*
2709          * If the user explicitly requested a mode we should either
2710          * supply it, or fail completely (which we do below).  But if
2711          * it's not set explicitly, we reset our mode to something
2712          * that works
2713          */
2714         if (resize_hpt_err) {
2715             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2716             error_free(resize_hpt_err);
2717             resize_hpt_err = NULL;
2718         } else {
2719             spapr->resize_hpt = smc->resize_hpt_default;
2720         }
2721     }
2722 
2723     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2724 
2725     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2726         /*
2727          * User requested HPT resize, but this host can't supply it.  Bail out
2728          */
2729         error_report_err(resize_hpt_err);
2730         exit(1);
2731     }
2732     error_free(resize_hpt_err);
2733 
2734     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2735 
2736     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2737     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2738 
2739     /*
2740      * VSMT must be set in order to be able to compute VCPU ids, ie to
2741      * call spapr_max_server_number() or spapr_vcpu_id().
2742      */
2743     spapr_set_vsmt_mode(spapr, &error_fatal);
2744 
2745     /* Set up Interrupt Controller before we create the VCPUs */
2746     spapr_irq_init(spapr, &error_fatal);
2747 
2748     /* Set up containers for ibm,client-architecture-support negotiated options
2749      */
2750     spapr->ov5 = spapr_ovec_new();
2751     spapr->ov5_cas = spapr_ovec_new();
2752 
2753     if (smc->dr_lmb_enabled) {
2754         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2755         spapr_validate_node_memory(machine, &error_fatal);
2756     }
2757 
2758     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2759 
2760     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2761     if (!smc->pre_6_2_numa_affinity) {
2762         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2763     }
2764 
2765     /* advertise support for dedicated HP event source to guests */
2766     if (spapr->use_hotplug_event_source) {
2767         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2768     }
2769 
2770     /* advertise support for HPT resizing */
2771     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2772         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2773     }
2774 
2775     /* advertise support for ibm,dyamic-memory-v2 */
2776     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2777 
2778     /* advertise XIVE on POWER9 machines */
2779     if (spapr->irq->xive) {
2780         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2781     }
2782 
2783     /* init CPUs */
2784     spapr_init_cpus(spapr);
2785 
2786     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2787 
2788     /* Init numa_assoc_array */
2789     spapr_numa_associativity_init(spapr, machine);
2790 
2791     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2792         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2793                               spapr->max_compat_pvr)) {
2794         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2795         /* KVM and TCG always allow GTSE with radix... */
2796         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2797     }
2798     /* ... but not with hash (currently). */
2799 
2800     if (kvm_enabled()) {
2801         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2802         kvmppc_enable_logical_ci_hcalls();
2803         kvmppc_enable_set_mode_hcall();
2804 
2805         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2806         kvmppc_enable_clear_ref_mod_hcalls();
2807 
2808         /* Enable H_PAGE_INIT */
2809         kvmppc_enable_h_page_init();
2810     }
2811 
2812     /* map RAM */
2813     memory_region_add_subregion(sysmem, 0, machine->ram);
2814 
2815     /* always allocate the device memory information */
2816     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2817 
2818     /* initialize hotplug memory address space */
2819     if (machine->ram_size < machine->maxram_size) {
2820         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2821         /*
2822          * Limit the number of hotpluggable memory slots to half the number
2823          * slots that KVM supports, leaving the other half for PCI and other
2824          * devices. However ensure that number of slots doesn't drop below 32.
2825          */
2826         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2827                            SPAPR_MAX_RAM_SLOTS;
2828 
2829         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2830             max_memslots = SPAPR_MAX_RAM_SLOTS;
2831         }
2832         if (machine->ram_slots > max_memslots) {
2833             error_report("Specified number of memory slots %"
2834                          PRIu64" exceeds max supported %d",
2835                          machine->ram_slots, max_memslots);
2836             exit(1);
2837         }
2838 
2839         machine->device_memory->base = ROUND_UP(machine->ram_size,
2840                                                 SPAPR_DEVICE_MEM_ALIGN);
2841         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2842                            "device-memory", device_mem_size);
2843         memory_region_add_subregion(sysmem, machine->device_memory->base,
2844                                     &machine->device_memory->mr);
2845     }
2846 
2847     if (smc->dr_lmb_enabled) {
2848         spapr_create_lmb_dr_connectors(spapr);
2849     }
2850 
2851     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2852         /* Create the error string for live migration blocker */
2853         error_setg(&spapr->fwnmi_migration_blocker,
2854             "A machine check is being handled during migration. The handler"
2855             "may run and log hardware error on the destination");
2856     }
2857 
2858     if (mc->nvdimm_supported) {
2859         spapr_create_nvdimm_dr_connectors(spapr);
2860     }
2861 
2862     /* Set up RTAS event infrastructure */
2863     spapr_events_init(spapr);
2864 
2865     /* Set up the RTC RTAS interfaces */
2866     spapr_rtc_create(spapr);
2867 
2868     /* Set up VIO bus */
2869     spapr->vio_bus = spapr_vio_bus_init();
2870 
2871     for (i = 0; serial_hd(i); i++) {
2872         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2873     }
2874 
2875     /* We always have at least the nvram device on VIO */
2876     spapr_create_nvram(spapr);
2877 
2878     /*
2879      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2880      * connectors (described in root DT node's "ibm,drc-types" property)
2881      * are pre-initialized here. additional child connectors (such as
2882      * connectors for a PHBs PCI slots) are added as needed during their
2883      * parent's realization.
2884      */
2885     if (smc->dr_phb_enabled) {
2886         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2887             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2888         }
2889     }
2890 
2891     /* Set up PCI */
2892     spapr_pci_rtas_init();
2893 
2894     phb = spapr_create_default_phb();
2895 
2896     for (i = 0; i < nb_nics; i++) {
2897         NICInfo *nd = &nd_table[i];
2898 
2899         if (!nd->model) {
2900             nd->model = g_strdup("spapr-vlan");
2901         }
2902 
2903         if (g_str_equal(nd->model, "spapr-vlan") ||
2904             g_str_equal(nd->model, "ibmveth")) {
2905             spapr_vlan_create(spapr->vio_bus, nd);
2906         } else {
2907             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2908         }
2909     }
2910 
2911     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2912         spapr_vscsi_create(spapr->vio_bus);
2913     }
2914 
2915     /* Graphics */
2916     if (spapr_vga_init(phb->bus, &error_fatal)) {
2917         spapr->has_graphics = true;
2918         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2919     }
2920 
2921     if (machine->usb) {
2922         if (smc->use_ohci_by_default) {
2923             pci_create_simple(phb->bus, -1, "pci-ohci");
2924         } else {
2925             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2926         }
2927 
2928         if (spapr->has_graphics) {
2929             USBBus *usb_bus = usb_bus_find(-1);
2930 
2931             usb_create_simple(usb_bus, "usb-kbd");
2932             usb_create_simple(usb_bus, "usb-mouse");
2933         }
2934     }
2935 
2936     if (kernel_filename) {
2937         spapr->kernel_size = load_elf(kernel_filename, NULL,
2938                                       translate_kernel_address, spapr,
2939                                       NULL, NULL, NULL, NULL, 1,
2940                                       PPC_ELF_MACHINE, 0, 0);
2941         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2942             spapr->kernel_size = load_elf(kernel_filename, NULL,
2943                                           translate_kernel_address, spapr,
2944                                           NULL, NULL, NULL, NULL, 0,
2945                                           PPC_ELF_MACHINE, 0, 0);
2946             spapr->kernel_le = spapr->kernel_size > 0;
2947         }
2948         if (spapr->kernel_size < 0) {
2949             error_report("error loading %s: %s", kernel_filename,
2950                          load_elf_strerror(spapr->kernel_size));
2951             exit(1);
2952         }
2953 
2954         /* load initrd */
2955         if (initrd_filename) {
2956             /* Try to locate the initrd in the gap between the kernel
2957              * and the firmware. Add a bit of space just in case
2958              */
2959             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2960                                   + 0x1ffff) & ~0xffff;
2961             spapr->initrd_size = load_image_targphys(initrd_filename,
2962                                                      spapr->initrd_base,
2963                                                      load_limit
2964                                                      - spapr->initrd_base);
2965             if (spapr->initrd_size < 0) {
2966                 error_report("could not load initial ram disk '%s'",
2967                              initrd_filename);
2968                 exit(1);
2969             }
2970         }
2971     }
2972 
2973     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2974     if (!filename) {
2975         error_report("Could not find LPAR firmware '%s'", bios_name);
2976         exit(1);
2977     }
2978     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2979     if (fw_size <= 0) {
2980         error_report("Could not load LPAR firmware '%s'", filename);
2981         exit(1);
2982     }
2983     g_free(filename);
2984 
2985     /* FIXME: Should register things through the MachineState's qdev
2986      * interface, this is a legacy from the sPAPREnvironment structure
2987      * which predated MachineState but had a similar function */
2988     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2989     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2990                          &savevm_htab_handlers, spapr);
2991 
2992     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2993 
2994     qemu_register_boot_set(spapr_boot_set, spapr);
2995 
2996     /*
2997      * Nothing needs to be done to resume a suspended guest because
2998      * suspending does not change the machine state, so no need for
2999      * a ->wakeup method.
3000      */
3001     qemu_register_wakeup_support();
3002 
3003     if (kvm_enabled()) {
3004         /* to stop and start vmclock */
3005         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3006                                          &spapr->tb);
3007 
3008         kvmppc_spapr_enable_inkernel_multitce();
3009     }
3010 
3011     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3012     if (spapr->vof) {
3013         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3014         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3015     }
3016 }
3017 
3018 #define DEFAULT_KVM_TYPE "auto"
3019 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3020 {
3021     /*
3022      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3023      * accomodate the 'HV' and 'PV' formats that exists in the
3024      * wild. The 'auto' mode is being introduced already as
3025      * lower-case, thus we don't need to bother checking for
3026      * "AUTO".
3027      */
3028     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3029         return 0;
3030     }
3031 
3032     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3033         return 1;
3034     }
3035 
3036     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3037         return 2;
3038     }
3039 
3040     error_report("Unknown kvm-type specified '%s'", vm_type);
3041     exit(1);
3042 }
3043 
3044 /*
3045  * Implementation of an interface to adjust firmware path
3046  * for the bootindex property handling.
3047  */
3048 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3049                                    DeviceState *dev)
3050 {
3051 #define CAST(type, obj, name) \
3052     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3053     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3054     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3055     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3056     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3057 
3058     if (d && bus) {
3059         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3060         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3061         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3062 
3063         if (spapr) {
3064             /*
3065              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3066              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3067              * 0x8000 | (target << 8) | (bus << 5) | lun
3068              * (see the "Logical unit addressing format" table in SAM5)
3069              */
3070             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3071             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3072                                    (uint64_t)id << 48);
3073         } else if (virtio) {
3074             /*
3075              * We use SRP luns of the form 01000000 | (target << 8) | lun
3076              * in the top 32 bits of the 64-bit LUN
3077              * Note: the quote above is from SLOF and it is wrong,
3078              * the actual binding is:
3079              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3080              */
3081             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3082             if (d->lun >= 256) {
3083                 /* Use the LUN "flat space addressing method" */
3084                 id |= 0x4000;
3085             }
3086             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3087                                    (uint64_t)id << 32);
3088         } else if (usb) {
3089             /*
3090              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3091              * in the top 32 bits of the 64-bit LUN
3092              */
3093             unsigned usb_port = atoi(usb->port->path);
3094             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3095             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3096                                    (uint64_t)id << 32);
3097         }
3098     }
3099 
3100     /*
3101      * SLOF probes the USB devices, and if it recognizes that the device is a
3102      * storage device, it changes its name to "storage" instead of "usb-host",
3103      * and additionally adds a child node for the SCSI LUN, so the correct
3104      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3105      */
3106     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3107         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3108         if (usb_device_is_scsi_storage(usbdev)) {
3109             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3110         }
3111     }
3112 
3113     if (phb) {
3114         /* Replace "pci" with "pci@800000020000000" */
3115         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3116     }
3117 
3118     if (vsc) {
3119         /* Same logic as virtio above */
3120         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3121         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3122     }
3123 
3124     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3125         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3126         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3127         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3128     }
3129 
3130     if (pcidev) {
3131         return spapr_pci_fw_dev_name(pcidev);
3132     }
3133 
3134     return NULL;
3135 }
3136 
3137 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3138 {
3139     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3140 
3141     return g_strdup(spapr->kvm_type);
3142 }
3143 
3144 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3145 {
3146     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3147 
3148     g_free(spapr->kvm_type);
3149     spapr->kvm_type = g_strdup(value);
3150 }
3151 
3152 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3153 {
3154     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3155 
3156     return spapr->use_hotplug_event_source;
3157 }
3158 
3159 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3160                                             Error **errp)
3161 {
3162     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3163 
3164     spapr->use_hotplug_event_source = value;
3165 }
3166 
3167 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3168 {
3169     return true;
3170 }
3171 
3172 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3173 {
3174     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3175 
3176     switch (spapr->resize_hpt) {
3177     case SPAPR_RESIZE_HPT_DEFAULT:
3178         return g_strdup("default");
3179     case SPAPR_RESIZE_HPT_DISABLED:
3180         return g_strdup("disabled");
3181     case SPAPR_RESIZE_HPT_ENABLED:
3182         return g_strdup("enabled");
3183     case SPAPR_RESIZE_HPT_REQUIRED:
3184         return g_strdup("required");
3185     }
3186     g_assert_not_reached();
3187 }
3188 
3189 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3190 {
3191     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3192 
3193     if (strcmp(value, "default") == 0) {
3194         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3195     } else if (strcmp(value, "disabled") == 0) {
3196         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3197     } else if (strcmp(value, "enabled") == 0) {
3198         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3199     } else if (strcmp(value, "required") == 0) {
3200         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3201     } else {
3202         error_setg(errp, "Bad value for \"resize-hpt\" property");
3203     }
3204 }
3205 
3206 static bool spapr_get_vof(Object *obj, Error **errp)
3207 {
3208     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3209 
3210     return spapr->vof != NULL;
3211 }
3212 
3213 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3214 {
3215     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3216 
3217     if (spapr->vof) {
3218         vof_cleanup(spapr->vof);
3219         g_free(spapr->vof);
3220         spapr->vof = NULL;
3221     }
3222     if (!value) {
3223         return;
3224     }
3225     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3226 }
3227 
3228 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3229 {
3230     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3231 
3232     if (spapr->irq == &spapr_irq_xics_legacy) {
3233         return g_strdup("legacy");
3234     } else if (spapr->irq == &spapr_irq_xics) {
3235         return g_strdup("xics");
3236     } else if (spapr->irq == &spapr_irq_xive) {
3237         return g_strdup("xive");
3238     } else if (spapr->irq == &spapr_irq_dual) {
3239         return g_strdup("dual");
3240     }
3241     g_assert_not_reached();
3242 }
3243 
3244 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3245 {
3246     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3247 
3248     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3249         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3250         return;
3251     }
3252 
3253     /* The legacy IRQ backend can not be set */
3254     if (strcmp(value, "xics") == 0) {
3255         spapr->irq = &spapr_irq_xics;
3256     } else if (strcmp(value, "xive") == 0) {
3257         spapr->irq = &spapr_irq_xive;
3258     } else if (strcmp(value, "dual") == 0) {
3259         spapr->irq = &spapr_irq_dual;
3260     } else {
3261         error_setg(errp, "Bad value for \"ic-mode\" property");
3262     }
3263 }
3264 
3265 static char *spapr_get_host_model(Object *obj, Error **errp)
3266 {
3267     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3268 
3269     return g_strdup(spapr->host_model);
3270 }
3271 
3272 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3273 {
3274     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3275 
3276     g_free(spapr->host_model);
3277     spapr->host_model = g_strdup(value);
3278 }
3279 
3280 static char *spapr_get_host_serial(Object *obj, Error **errp)
3281 {
3282     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3283 
3284     return g_strdup(spapr->host_serial);
3285 }
3286 
3287 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3288 {
3289     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3290 
3291     g_free(spapr->host_serial);
3292     spapr->host_serial = g_strdup(value);
3293 }
3294 
3295 static void spapr_instance_init(Object *obj)
3296 {
3297     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3298     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3299     MachineState *ms = MACHINE(spapr);
3300     MachineClass *mc = MACHINE_GET_CLASS(ms);
3301 
3302     /*
3303      * NVDIMM support went live in 5.1 without considering that, in
3304      * other archs, the user needs to enable NVDIMM support with the
3305      * 'nvdimm' machine option and the default behavior is NVDIMM
3306      * support disabled. It is too late to roll back to the standard
3307      * behavior without breaking 5.1 guests.
3308      */
3309     if (mc->nvdimm_supported) {
3310         ms->nvdimms_state->is_enabled = true;
3311     }
3312 
3313     spapr->htab_fd = -1;
3314     spapr->use_hotplug_event_source = true;
3315     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3316     object_property_add_str(obj, "kvm-type",
3317                             spapr_get_kvm_type, spapr_set_kvm_type);
3318     object_property_set_description(obj, "kvm-type",
3319                                     "Specifies the KVM virtualization mode (auto,"
3320                                     " hv, pr). Defaults to 'auto'. This mode will use"
3321                                     " any available KVM module loaded in the host,"
3322                                     " where kvm_hv takes precedence if both kvm_hv and"
3323                                     " kvm_pr are loaded.");
3324     object_property_add_bool(obj, "modern-hotplug-events",
3325                             spapr_get_modern_hotplug_events,
3326                             spapr_set_modern_hotplug_events);
3327     object_property_set_description(obj, "modern-hotplug-events",
3328                                     "Use dedicated hotplug event mechanism in"
3329                                     " place of standard EPOW events when possible"
3330                                     " (required for memory hot-unplug support)");
3331     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3332                             "Maximum permitted CPU compatibility mode");
3333 
3334     object_property_add_str(obj, "resize-hpt",
3335                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3336     object_property_set_description(obj, "resize-hpt",
3337                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3338     object_property_add_uint32_ptr(obj, "vsmt",
3339                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3340     object_property_set_description(obj, "vsmt",
3341                                     "Virtual SMT: KVM behaves as if this were"
3342                                     " the host's SMT mode");
3343 
3344     object_property_add_bool(obj, "vfio-no-msix-emulation",
3345                              spapr_get_msix_emulation, NULL);
3346 
3347     object_property_add_uint64_ptr(obj, "kernel-addr",
3348                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3349     object_property_set_description(obj, "kernel-addr",
3350                                     stringify(KERNEL_LOAD_ADDR)
3351                                     " for -kernel is the default");
3352     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3353 
3354     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3355     object_property_set_description(obj, "x-vof",
3356                                     "Enable Virtual Open Firmware (experimental)");
3357 
3358     /* The machine class defines the default interrupt controller mode */
3359     spapr->irq = smc->irq;
3360     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3361                             spapr_set_ic_mode);
3362     object_property_set_description(obj, "ic-mode",
3363                  "Specifies the interrupt controller mode (xics, xive, dual)");
3364 
3365     object_property_add_str(obj, "host-model",
3366         spapr_get_host_model, spapr_set_host_model);
3367     object_property_set_description(obj, "host-model",
3368         "Host model to advertise in guest device tree");
3369     object_property_add_str(obj, "host-serial",
3370         spapr_get_host_serial, spapr_set_host_serial);
3371     object_property_set_description(obj, "host-serial",
3372         "Host serial number to advertise in guest device tree");
3373 }
3374 
3375 static void spapr_machine_finalizefn(Object *obj)
3376 {
3377     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3378 
3379     g_free(spapr->kvm_type);
3380 }
3381 
3382 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3383 {
3384     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3385     PowerPCCPU *cpu = POWERPC_CPU(cs);
3386     CPUPPCState *env = &cpu->env;
3387 
3388     cpu_synchronize_state(cs);
3389     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3390     if (spapr->fwnmi_system_reset_addr != -1) {
3391         uint64_t rtas_addr, addr;
3392 
3393         /* get rtas addr from fdt */
3394         rtas_addr = spapr_get_rtas_addr();
3395         if (!rtas_addr) {
3396             qemu_system_guest_panicked(NULL);
3397             return;
3398         }
3399 
3400         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3401         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3402         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3403         env->gpr[3] = addr;
3404     }
3405     ppc_cpu_do_system_reset(cs);
3406     if (spapr->fwnmi_system_reset_addr != -1) {
3407         env->nip = spapr->fwnmi_system_reset_addr;
3408     }
3409 }
3410 
3411 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3412 {
3413     CPUState *cs;
3414 
3415     CPU_FOREACH(cs) {
3416         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3417     }
3418 }
3419 
3420 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3421                           void *fdt, int *fdt_start_offset, Error **errp)
3422 {
3423     uint64_t addr;
3424     uint32_t node;
3425 
3426     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3427     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3428                                     &error_abort);
3429     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3430                                              SPAPR_MEMORY_BLOCK_SIZE);
3431     return 0;
3432 }
3433 
3434 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3435                            bool dedicated_hp_event_source)
3436 {
3437     SpaprDrc *drc;
3438     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3439     int i;
3440     uint64_t addr = addr_start;
3441     bool hotplugged = spapr_drc_hotplugged(dev);
3442 
3443     for (i = 0; i < nr_lmbs; i++) {
3444         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3445                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3446         g_assert(drc);
3447 
3448         /*
3449          * memory_device_get_free_addr() provided a range of free addresses
3450          * that doesn't overlap with any existing mapping at pre-plug. The
3451          * corresponding LMB DRCs are thus assumed to be all attachable.
3452          */
3453         spapr_drc_attach(drc, dev);
3454         if (!hotplugged) {
3455             spapr_drc_reset(drc);
3456         }
3457         addr += SPAPR_MEMORY_BLOCK_SIZE;
3458     }
3459     /* send hotplug notification to the
3460      * guest only in case of hotplugged memory
3461      */
3462     if (hotplugged) {
3463         if (dedicated_hp_event_source) {
3464             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3465                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3466             g_assert(drc);
3467             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3468                                                    nr_lmbs,
3469                                                    spapr_drc_index(drc));
3470         } else {
3471             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3472                                            nr_lmbs);
3473         }
3474     }
3475 }
3476 
3477 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3478 {
3479     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3480     PCDIMMDevice *dimm = PC_DIMM(dev);
3481     uint64_t size, addr;
3482     int64_t slot;
3483     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3484 
3485     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3486 
3487     pc_dimm_plug(dimm, MACHINE(ms));
3488 
3489     if (!is_nvdimm) {
3490         addr = object_property_get_uint(OBJECT(dimm),
3491                                         PC_DIMM_ADDR_PROP, &error_abort);
3492         spapr_add_lmbs(dev, addr, size,
3493                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3494     } else {
3495         slot = object_property_get_int(OBJECT(dimm),
3496                                        PC_DIMM_SLOT_PROP, &error_abort);
3497         /* We should have valid slot number at this point */
3498         g_assert(slot >= 0);
3499         spapr_add_nvdimm(dev, slot);
3500     }
3501 }
3502 
3503 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3504                                   Error **errp)
3505 {
3506     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3507     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3508     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3509     PCDIMMDevice *dimm = PC_DIMM(dev);
3510     Error *local_err = NULL;
3511     uint64_t size;
3512     Object *memdev;
3513     hwaddr pagesize;
3514 
3515     if (!smc->dr_lmb_enabled) {
3516         error_setg(errp, "Memory hotplug not supported for this machine");
3517         return;
3518     }
3519 
3520     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3521     if (local_err) {
3522         error_propagate(errp, local_err);
3523         return;
3524     }
3525 
3526     if (is_nvdimm) {
3527         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3528             return;
3529         }
3530     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3531         error_setg(errp, "Hotplugged memory size must be a multiple of "
3532                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3533         return;
3534     }
3535 
3536     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3537                                       &error_abort);
3538     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3539     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3540         return;
3541     }
3542 
3543     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3544 }
3545 
3546 struct SpaprDimmState {
3547     PCDIMMDevice *dimm;
3548     uint32_t nr_lmbs;
3549     QTAILQ_ENTRY(SpaprDimmState) next;
3550 };
3551 
3552 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3553                                                        PCDIMMDevice *dimm)
3554 {
3555     SpaprDimmState *dimm_state = NULL;
3556 
3557     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3558         if (dimm_state->dimm == dimm) {
3559             break;
3560         }
3561     }
3562     return dimm_state;
3563 }
3564 
3565 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3566                                                       uint32_t nr_lmbs,
3567                                                       PCDIMMDevice *dimm)
3568 {
3569     SpaprDimmState *ds = NULL;
3570 
3571     /*
3572      * If this request is for a DIMM whose removal had failed earlier
3573      * (due to guest's refusal to remove the LMBs), we would have this
3574      * dimm already in the pending_dimm_unplugs list. In that
3575      * case don't add again.
3576      */
3577     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3578     if (!ds) {
3579         ds = g_malloc0(sizeof(SpaprDimmState));
3580         ds->nr_lmbs = nr_lmbs;
3581         ds->dimm = dimm;
3582         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3583     }
3584     return ds;
3585 }
3586 
3587 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3588                                               SpaprDimmState *dimm_state)
3589 {
3590     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3591     g_free(dimm_state);
3592 }
3593 
3594 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3595                                                         PCDIMMDevice *dimm)
3596 {
3597     SpaprDrc *drc;
3598     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3599                                                   &error_abort);
3600     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3601     uint32_t avail_lmbs = 0;
3602     uint64_t addr_start, addr;
3603     int i;
3604 
3605     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3606                                           &error_abort);
3607 
3608     addr = addr_start;
3609     for (i = 0; i < nr_lmbs; i++) {
3610         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3611                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3612         g_assert(drc);
3613         if (drc->dev) {
3614             avail_lmbs++;
3615         }
3616         addr += SPAPR_MEMORY_BLOCK_SIZE;
3617     }
3618 
3619     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3620 }
3621 
3622 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3623 {
3624     SpaprDimmState *ds;
3625     PCDIMMDevice *dimm;
3626     SpaprDrc *drc;
3627     uint32_t nr_lmbs;
3628     uint64_t size, addr_start, addr;
3629     g_autofree char *qapi_error = NULL;
3630     int i;
3631 
3632     if (!dev) {
3633         return;
3634     }
3635 
3636     dimm = PC_DIMM(dev);
3637     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3638 
3639     /*
3640      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3641      * unplug state, but one of its DRC is marked as unplug_requested.
3642      * This is bad and weird enough to g_assert() out.
3643      */
3644     g_assert(ds);
3645 
3646     spapr_pending_dimm_unplugs_remove(spapr, ds);
3647 
3648     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3649     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3650 
3651     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3652                                           &error_abort);
3653 
3654     addr = addr_start;
3655     for (i = 0; i < nr_lmbs; i++) {
3656         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3657                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3658         g_assert(drc);
3659 
3660         drc->unplug_requested = false;
3661         addr += SPAPR_MEMORY_BLOCK_SIZE;
3662     }
3663 
3664     /*
3665      * Tell QAPI that something happened and the memory
3666      * hotunplug wasn't successful. Keep sending
3667      * MEM_UNPLUG_ERROR even while sending
3668      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3669      * MEM_UNPLUG_ERROR is due.
3670      */
3671     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3672                                  "for device %s", dev->id);
3673 
3674     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3675 
3676     qapi_event_send_device_unplug_guest_error(!!dev->id, dev->id,
3677                                               dev->canonical_path);
3678 }
3679 
3680 /* Callback to be called during DRC release. */
3681 void spapr_lmb_release(DeviceState *dev)
3682 {
3683     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3684     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3685     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3686 
3687     /* This information will get lost if a migration occurs
3688      * during the unplug process. In this case recover it. */
3689     if (ds == NULL) {
3690         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3691         g_assert(ds);
3692         /* The DRC being examined by the caller at least must be counted */
3693         g_assert(ds->nr_lmbs);
3694     }
3695 
3696     if (--ds->nr_lmbs) {
3697         return;
3698     }
3699 
3700     /*
3701      * Now that all the LMBs have been removed by the guest, call the
3702      * unplug handler chain. This can never fail.
3703      */
3704     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3705     object_unparent(OBJECT(dev));
3706 }
3707 
3708 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3709 {
3710     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3711     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3712 
3713     /* We really shouldn't get this far without anything to unplug */
3714     g_assert(ds);
3715 
3716     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3717     qdev_unrealize(dev);
3718     spapr_pending_dimm_unplugs_remove(spapr, ds);
3719 }
3720 
3721 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3722                                         DeviceState *dev, Error **errp)
3723 {
3724     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3725     PCDIMMDevice *dimm = PC_DIMM(dev);
3726     uint32_t nr_lmbs;
3727     uint64_t size, addr_start, addr;
3728     int i;
3729     SpaprDrc *drc;
3730 
3731     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3732         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3733         return;
3734     }
3735 
3736     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3737     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3738 
3739     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3740                                           &error_abort);
3741 
3742     /*
3743      * An existing pending dimm state for this DIMM means that there is an
3744      * unplug operation in progress, waiting for the spapr_lmb_release
3745      * callback to complete the job (BQL can't cover that far). In this case,
3746      * bail out to avoid detaching DRCs that were already released.
3747      */
3748     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3749         error_setg(errp, "Memory unplug already in progress for device %s",
3750                    dev->id);
3751         return;
3752     }
3753 
3754     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3755 
3756     addr = addr_start;
3757     for (i = 0; i < nr_lmbs; i++) {
3758         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3759                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3760         g_assert(drc);
3761 
3762         spapr_drc_unplug_request(drc);
3763         addr += SPAPR_MEMORY_BLOCK_SIZE;
3764     }
3765 
3766     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3767                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3768     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3769                                               nr_lmbs, spapr_drc_index(drc));
3770 }
3771 
3772 /* Callback to be called during DRC release. */
3773 void spapr_core_release(DeviceState *dev)
3774 {
3775     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3776 
3777     /* Call the unplug handler chain. This can never fail. */
3778     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3779     object_unparent(OBJECT(dev));
3780 }
3781 
3782 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3783 {
3784     MachineState *ms = MACHINE(hotplug_dev);
3785     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3786     CPUCore *cc = CPU_CORE(dev);
3787     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3788 
3789     if (smc->pre_2_10_has_unused_icps) {
3790         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3791         int i;
3792 
3793         for (i = 0; i < cc->nr_threads; i++) {
3794             CPUState *cs = CPU(sc->threads[i]);
3795 
3796             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3797         }
3798     }
3799 
3800     assert(core_slot);
3801     core_slot->cpu = NULL;
3802     qdev_unrealize(dev);
3803 }
3804 
3805 static
3806 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3807                                Error **errp)
3808 {
3809     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3810     int index;
3811     SpaprDrc *drc;
3812     CPUCore *cc = CPU_CORE(dev);
3813 
3814     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3815         error_setg(errp, "Unable to find CPU core with core-id: %d",
3816                    cc->core_id);
3817         return;
3818     }
3819     if (index == 0) {
3820         error_setg(errp, "Boot CPU core may not be unplugged");
3821         return;
3822     }
3823 
3824     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3825                           spapr_vcpu_id(spapr, cc->core_id));
3826     g_assert(drc);
3827 
3828     if (!spapr_drc_unplug_requested(drc)) {
3829         spapr_drc_unplug_request(drc);
3830     }
3831 
3832     /*
3833      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3834      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3835      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3836      * attempt (e.g. the kernel will refuse to remove the last online
3837      * CPU), we will never attempt it again because unplug_requested
3838      * will still be 'true' in that case.
3839      */
3840     spapr_hotplug_req_remove_by_index(drc);
3841 }
3842 
3843 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3844                            void *fdt, int *fdt_start_offset, Error **errp)
3845 {
3846     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3847     CPUState *cs = CPU(core->threads[0]);
3848     PowerPCCPU *cpu = POWERPC_CPU(cs);
3849     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3850     int id = spapr_get_vcpu_id(cpu);
3851     g_autofree char *nodename = NULL;
3852     int offset;
3853 
3854     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3855     offset = fdt_add_subnode(fdt, 0, nodename);
3856 
3857     spapr_dt_cpu(cs, fdt, offset, spapr);
3858 
3859     /*
3860      * spapr_dt_cpu() does not fill the 'name' property in the
3861      * CPU node. The function is called during boot process, before
3862      * and after CAS, and overwriting the 'name' property written
3863      * by SLOF is not allowed.
3864      *
3865      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3866      * CPUs more compatible with the coldplugged ones, which have
3867      * the 'name' property. Linux Kernel also relies on this
3868      * property to identify CPU nodes.
3869      */
3870     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3871 
3872     *fdt_start_offset = offset;
3873     return 0;
3874 }
3875 
3876 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3877 {
3878     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3879     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3880     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3881     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3882     CPUCore *cc = CPU_CORE(dev);
3883     CPUState *cs;
3884     SpaprDrc *drc;
3885     CPUArchId *core_slot;
3886     int index;
3887     bool hotplugged = spapr_drc_hotplugged(dev);
3888     int i;
3889 
3890     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3891     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3892 
3893     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3894                           spapr_vcpu_id(spapr, cc->core_id));
3895 
3896     g_assert(drc || !mc->has_hotpluggable_cpus);
3897 
3898     if (drc) {
3899         /*
3900          * spapr_core_pre_plug() already buys us this is a brand new
3901          * core being plugged into a free slot. Nothing should already
3902          * be attached to the corresponding DRC.
3903          */
3904         spapr_drc_attach(drc, dev);
3905 
3906         if (hotplugged) {
3907             /*
3908              * Send hotplug notification interrupt to the guest only
3909              * in case of hotplugged CPUs.
3910              */
3911             spapr_hotplug_req_add_by_index(drc);
3912         } else {
3913             spapr_drc_reset(drc);
3914         }
3915     }
3916 
3917     core_slot->cpu = OBJECT(dev);
3918 
3919     /*
3920      * Set compatibility mode to match the boot CPU, which was either set
3921      * by the machine reset code or by CAS. This really shouldn't fail at
3922      * this point.
3923      */
3924     if (hotplugged) {
3925         for (i = 0; i < cc->nr_threads; i++) {
3926             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3927                            &error_abort);
3928         }
3929     }
3930 
3931     if (smc->pre_2_10_has_unused_icps) {
3932         for (i = 0; i < cc->nr_threads; i++) {
3933             cs = CPU(core->threads[i]);
3934             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3935         }
3936     }
3937 }
3938 
3939 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3940                                 Error **errp)
3941 {
3942     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3943     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3944     CPUCore *cc = CPU_CORE(dev);
3945     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3946     const char *type = object_get_typename(OBJECT(dev));
3947     CPUArchId *core_slot;
3948     int index;
3949     unsigned int smp_threads = machine->smp.threads;
3950 
3951     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3952         error_setg(errp, "CPU hotplug not supported for this machine");
3953         return;
3954     }
3955 
3956     if (strcmp(base_core_type, type)) {
3957         error_setg(errp, "CPU core type should be %s", base_core_type);
3958         return;
3959     }
3960 
3961     if (cc->core_id % smp_threads) {
3962         error_setg(errp, "invalid core id %d", cc->core_id);
3963         return;
3964     }
3965 
3966     /*
3967      * In general we should have homogeneous threads-per-core, but old
3968      * (pre hotplug support) machine types allow the last core to have
3969      * reduced threads as a compatibility hack for when we allowed
3970      * total vcpus not a multiple of threads-per-core.
3971      */
3972     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3973         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3974                    smp_threads);
3975         return;
3976     }
3977 
3978     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3979     if (!core_slot) {
3980         error_setg(errp, "core id %d out of range", cc->core_id);
3981         return;
3982     }
3983 
3984     if (core_slot->cpu) {
3985         error_setg(errp, "core %d already populated", cc->core_id);
3986         return;
3987     }
3988 
3989     numa_cpu_pre_plug(core_slot, dev, errp);
3990 }
3991 
3992 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3993                           void *fdt, int *fdt_start_offset, Error **errp)
3994 {
3995     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3996     int intc_phandle;
3997 
3998     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3999     if (intc_phandle <= 0) {
4000         return -1;
4001     }
4002 
4003     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4004         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4005         return -1;
4006     }
4007 
4008     /* generally SLOF creates these, for hotplug it's up to QEMU */
4009     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4010 
4011     return 0;
4012 }
4013 
4014 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4015                                Error **errp)
4016 {
4017     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4018     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4019     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4020     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4021     SpaprDrc *drc;
4022 
4023     if (dev->hotplugged && !smc->dr_phb_enabled) {
4024         error_setg(errp, "PHB hotplug not supported for this machine");
4025         return false;
4026     }
4027 
4028     if (sphb->index == (uint32_t)-1) {
4029         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4030         return false;
4031     }
4032 
4033     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4034     if (drc && drc->dev) {
4035         error_setg(errp, "PHB %d already attached", sphb->index);
4036         return false;
4037     }
4038 
4039     /*
4040      * This will check that sphb->index doesn't exceed the maximum number of
4041      * PHBs for the current machine type.
4042      */
4043     return
4044         smc->phb_placement(spapr, sphb->index,
4045                            &sphb->buid, &sphb->io_win_addr,
4046                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4047                            windows_supported, sphb->dma_liobn,
4048                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4049                            errp);
4050 }
4051 
4052 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4053 {
4054     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4055     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4056     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4057     SpaprDrc *drc;
4058     bool hotplugged = spapr_drc_hotplugged(dev);
4059 
4060     if (!smc->dr_phb_enabled) {
4061         return;
4062     }
4063 
4064     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4065     /* hotplug hooks should check it's enabled before getting this far */
4066     assert(drc);
4067 
4068     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4069     spapr_drc_attach(drc, dev);
4070 
4071     if (hotplugged) {
4072         spapr_hotplug_req_add_by_index(drc);
4073     } else {
4074         spapr_drc_reset(drc);
4075     }
4076 }
4077 
4078 void spapr_phb_release(DeviceState *dev)
4079 {
4080     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4081 
4082     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4083     object_unparent(OBJECT(dev));
4084 }
4085 
4086 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4087 {
4088     qdev_unrealize(dev);
4089 }
4090 
4091 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4092                                      DeviceState *dev, Error **errp)
4093 {
4094     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4095     SpaprDrc *drc;
4096 
4097     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4098     assert(drc);
4099 
4100     if (!spapr_drc_unplug_requested(drc)) {
4101         spapr_drc_unplug_request(drc);
4102         spapr_hotplug_req_remove_by_index(drc);
4103     } else {
4104         error_setg(errp,
4105                    "PCI Host Bridge unplug already in progress for device %s",
4106                    dev->id);
4107     }
4108 }
4109 
4110 static
4111 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4112                               Error **errp)
4113 {
4114     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4115 
4116     if (spapr->tpm_proxy != NULL) {
4117         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4118         return false;
4119     }
4120 
4121     return true;
4122 }
4123 
4124 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4125 {
4126     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4127     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4128 
4129     /* Already checked in spapr_tpm_proxy_pre_plug() */
4130     g_assert(spapr->tpm_proxy == NULL);
4131 
4132     spapr->tpm_proxy = tpm_proxy;
4133 }
4134 
4135 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4136 {
4137     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4138 
4139     qdev_unrealize(dev);
4140     object_unparent(OBJECT(dev));
4141     spapr->tpm_proxy = NULL;
4142 }
4143 
4144 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4145                                       DeviceState *dev, Error **errp)
4146 {
4147     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4148         spapr_memory_plug(hotplug_dev, dev);
4149     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4150         spapr_core_plug(hotplug_dev, dev);
4151     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4152         spapr_phb_plug(hotplug_dev, dev);
4153     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4154         spapr_tpm_proxy_plug(hotplug_dev, dev);
4155     }
4156 }
4157 
4158 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4159                                         DeviceState *dev, Error **errp)
4160 {
4161     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4162         spapr_memory_unplug(hotplug_dev, dev);
4163     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4164         spapr_core_unplug(hotplug_dev, dev);
4165     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4166         spapr_phb_unplug(hotplug_dev, dev);
4167     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4168         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4169     }
4170 }
4171 
4172 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4173 {
4174     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4175         /*
4176          * CAS will process all pending unplug requests.
4177          *
4178          * HACK: a guest could theoretically have cleared all bits in OV5,
4179          * but none of the guests we care for do.
4180          */
4181         spapr_ovec_empty(spapr->ov5_cas);
4182 }
4183 
4184 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4185                                                 DeviceState *dev, Error **errp)
4186 {
4187     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4188     MachineClass *mc = MACHINE_GET_CLASS(sms);
4189     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4190 
4191     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4192         if (spapr_memory_hot_unplug_supported(sms)) {
4193             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4194         } else {
4195             error_setg(errp, "Memory hot unplug not supported for this guest");
4196         }
4197     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4198         if (!mc->has_hotpluggable_cpus) {
4199             error_setg(errp, "CPU hot unplug not supported on this machine");
4200             return;
4201         }
4202         spapr_core_unplug_request(hotplug_dev, dev, errp);
4203     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4204         if (!smc->dr_phb_enabled) {
4205             error_setg(errp, "PHB hot unplug not supported on this machine");
4206             return;
4207         }
4208         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4209     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4210         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4211     }
4212 }
4213 
4214 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4215                                           DeviceState *dev, Error **errp)
4216 {
4217     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4218         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4219     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4220         spapr_core_pre_plug(hotplug_dev, dev, errp);
4221     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4222         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4223     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4224         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4225     }
4226 }
4227 
4228 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4229                                                  DeviceState *dev)
4230 {
4231     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4232         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4233         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4234         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4235         return HOTPLUG_HANDLER(machine);
4236     }
4237     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4238         PCIDevice *pcidev = PCI_DEVICE(dev);
4239         PCIBus *root = pci_device_root_bus(pcidev);
4240         SpaprPhbState *phb =
4241             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4242                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4243 
4244         if (phb) {
4245             return HOTPLUG_HANDLER(phb);
4246         }
4247     }
4248     return NULL;
4249 }
4250 
4251 static CpuInstanceProperties
4252 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4253 {
4254     CPUArchId *core_slot;
4255     MachineClass *mc = MACHINE_GET_CLASS(machine);
4256 
4257     /* make sure possible_cpu are intialized */
4258     mc->possible_cpu_arch_ids(machine);
4259     /* get CPU core slot containing thread that matches cpu_index */
4260     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4261     assert(core_slot);
4262     return core_slot->props;
4263 }
4264 
4265 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4266 {
4267     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4268 }
4269 
4270 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4271 {
4272     int i;
4273     unsigned int smp_threads = machine->smp.threads;
4274     unsigned int smp_cpus = machine->smp.cpus;
4275     const char *core_type;
4276     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4277     MachineClass *mc = MACHINE_GET_CLASS(machine);
4278 
4279     if (!mc->has_hotpluggable_cpus) {
4280         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4281     }
4282     if (machine->possible_cpus) {
4283         assert(machine->possible_cpus->len == spapr_max_cores);
4284         return machine->possible_cpus;
4285     }
4286 
4287     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4288     if (!core_type) {
4289         error_report("Unable to find sPAPR CPU Core definition");
4290         exit(1);
4291     }
4292 
4293     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4294                              sizeof(CPUArchId) * spapr_max_cores);
4295     machine->possible_cpus->len = spapr_max_cores;
4296     for (i = 0; i < machine->possible_cpus->len; i++) {
4297         int core_id = i * smp_threads;
4298 
4299         machine->possible_cpus->cpus[i].type = core_type;
4300         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4301         machine->possible_cpus->cpus[i].arch_id = core_id;
4302         machine->possible_cpus->cpus[i].props.has_core_id = true;
4303         machine->possible_cpus->cpus[i].props.core_id = core_id;
4304     }
4305     return machine->possible_cpus;
4306 }
4307 
4308 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4309                                 uint64_t *buid, hwaddr *pio,
4310                                 hwaddr *mmio32, hwaddr *mmio64,
4311                                 unsigned n_dma, uint32_t *liobns,
4312                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4313 {
4314     /*
4315      * New-style PHB window placement.
4316      *
4317      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4318      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4319      * windows.
4320      *
4321      * Some guest kernels can't work with MMIO windows above 1<<46
4322      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4323      *
4324      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4325      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4326      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4327      * 1TiB 64-bit MMIO windows for each PHB.
4328      */
4329     const uint64_t base_buid = 0x800000020000000ULL;
4330     int i;
4331 
4332     /* Sanity check natural alignments */
4333     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4334     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4335     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4336     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4337     /* Sanity check bounds */
4338     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4339                       SPAPR_PCI_MEM32_WIN_SIZE);
4340     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4341                       SPAPR_PCI_MEM64_WIN_SIZE);
4342 
4343     if (index >= SPAPR_MAX_PHBS) {
4344         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4345                    SPAPR_MAX_PHBS - 1);
4346         return false;
4347     }
4348 
4349     *buid = base_buid + index;
4350     for (i = 0; i < n_dma; ++i) {
4351         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4352     }
4353 
4354     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4355     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4356     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4357 
4358     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4359     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4360     return true;
4361 }
4362 
4363 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4364 {
4365     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4366 
4367     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4368 }
4369 
4370 static void spapr_ics_resend(XICSFabric *dev)
4371 {
4372     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4373 
4374     ics_resend(spapr->ics);
4375 }
4376 
4377 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4378 {
4379     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4380 
4381     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4382 }
4383 
4384 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4385                                  Monitor *mon)
4386 {
4387     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4388 
4389     spapr_irq_print_info(spapr, mon);
4390     monitor_printf(mon, "irqchip: %s\n",
4391                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4392 }
4393 
4394 /*
4395  * This is a XIVE only operation
4396  */
4397 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4398                            uint8_t nvt_blk, uint32_t nvt_idx,
4399                            bool cam_ignore, uint8_t priority,
4400                            uint32_t logic_serv, XiveTCTXMatch *match)
4401 {
4402     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4403     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4404     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4405     int count;
4406 
4407     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4408                            priority, logic_serv, match);
4409     if (count < 0) {
4410         return count;
4411     }
4412 
4413     /*
4414      * When we implement the save and restore of the thread interrupt
4415      * contexts in the enter/exit CPU handlers of the machine and the
4416      * escalations in QEMU, we should be able to handle non dispatched
4417      * vCPUs.
4418      *
4419      * Until this is done, the sPAPR machine should find at least one
4420      * matching context always.
4421      */
4422     if (count == 0) {
4423         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4424                       nvt_blk, nvt_idx);
4425     }
4426 
4427     return count;
4428 }
4429 
4430 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4431 {
4432     return cpu->vcpu_id;
4433 }
4434 
4435 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4436 {
4437     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4438     MachineState *ms = MACHINE(spapr);
4439     int vcpu_id;
4440 
4441     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4442 
4443     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4444         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4445         error_append_hint(errp, "Adjust the number of cpus to %d "
4446                           "or try to raise the number of threads per core\n",
4447                           vcpu_id * ms->smp.threads / spapr->vsmt);
4448         return false;
4449     }
4450 
4451     cpu->vcpu_id = vcpu_id;
4452     return true;
4453 }
4454 
4455 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4456 {
4457     CPUState *cs;
4458 
4459     CPU_FOREACH(cs) {
4460         PowerPCCPU *cpu = POWERPC_CPU(cs);
4461 
4462         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4463             return cpu;
4464         }
4465     }
4466 
4467     return NULL;
4468 }
4469 
4470 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4471 {
4472     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4473 
4474     /* These are only called by TCG, KVM maintains dispatch state */
4475 
4476     spapr_cpu->prod = false;
4477     if (spapr_cpu->vpa_addr) {
4478         CPUState *cs = CPU(cpu);
4479         uint32_t dispatch;
4480 
4481         dispatch = ldl_be_phys(cs->as,
4482                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4483         dispatch++;
4484         if ((dispatch & 1) != 0) {
4485             qemu_log_mask(LOG_GUEST_ERROR,
4486                           "VPA: incorrect dispatch counter value for "
4487                           "dispatched partition %u, correcting.\n", dispatch);
4488             dispatch++;
4489         }
4490         stl_be_phys(cs->as,
4491                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4492     }
4493 }
4494 
4495 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4496 {
4497     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4498 
4499     if (spapr_cpu->vpa_addr) {
4500         CPUState *cs = CPU(cpu);
4501         uint32_t dispatch;
4502 
4503         dispatch = ldl_be_phys(cs->as,
4504                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4505         dispatch++;
4506         if ((dispatch & 1) != 1) {
4507             qemu_log_mask(LOG_GUEST_ERROR,
4508                           "VPA: incorrect dispatch counter value for "
4509                           "preempted partition %u, correcting.\n", dispatch);
4510             dispatch++;
4511         }
4512         stl_be_phys(cs->as,
4513                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4514     }
4515 }
4516 
4517 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4518 {
4519     MachineClass *mc = MACHINE_CLASS(oc);
4520     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4521     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4522     NMIClass *nc = NMI_CLASS(oc);
4523     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4524     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4525     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4526     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4527     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4528     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4529 
4530     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4531     mc->ignore_boot_device_suffixes = true;
4532 
4533     /*
4534      * We set up the default / latest behaviour here.  The class_init
4535      * functions for the specific versioned machine types can override
4536      * these details for backwards compatibility
4537      */
4538     mc->init = spapr_machine_init;
4539     mc->reset = spapr_machine_reset;
4540     mc->block_default_type = IF_SCSI;
4541 
4542     /*
4543      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4544      * should be limited by the host capability instead of hardcoded.
4545      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4546      * guests are welcome to have as many CPUs as the host are capable
4547      * of emulate.
4548      */
4549     mc->max_cpus = INT32_MAX;
4550 
4551     mc->no_parallel = 1;
4552     mc->default_boot_order = "";
4553     mc->default_ram_size = 512 * MiB;
4554     mc->default_ram_id = "ppc_spapr.ram";
4555     mc->default_display = "std";
4556     mc->kvm_type = spapr_kvm_type;
4557     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4558     mc->pci_allow_0_address = true;
4559     assert(!mc->get_hotplug_handler);
4560     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4561     hc->pre_plug = spapr_machine_device_pre_plug;
4562     hc->plug = spapr_machine_device_plug;
4563     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4564     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4565     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4566     hc->unplug_request = spapr_machine_device_unplug_request;
4567     hc->unplug = spapr_machine_device_unplug;
4568 
4569     smc->dr_lmb_enabled = true;
4570     smc->update_dt_enabled = true;
4571     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4572     mc->has_hotpluggable_cpus = true;
4573     mc->nvdimm_supported = true;
4574     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4575     fwc->get_dev_path = spapr_get_fw_dev_path;
4576     nc->nmi_monitor_handler = spapr_nmi;
4577     smc->phb_placement = spapr_phb_placement;
4578     vhc->hypercall = emulate_spapr_hypercall;
4579     vhc->hpt_mask = spapr_hpt_mask;
4580     vhc->map_hptes = spapr_map_hptes;
4581     vhc->unmap_hptes = spapr_unmap_hptes;
4582     vhc->hpte_set_c = spapr_hpte_set_c;
4583     vhc->hpte_set_r = spapr_hpte_set_r;
4584     vhc->get_pate = spapr_get_pate;
4585     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4586     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4587     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4588     xic->ics_get = spapr_ics_get;
4589     xic->ics_resend = spapr_ics_resend;
4590     xic->icp_get = spapr_icp_get;
4591     ispc->print_info = spapr_pic_print_info;
4592     /* Force NUMA node memory size to be a multiple of
4593      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4594      * in which LMBs are represented and hot-added
4595      */
4596     mc->numa_mem_align_shift = 28;
4597     mc->auto_enable_numa = true;
4598 
4599     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4600     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4601     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4602     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4603     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4604     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4605     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4606     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4607     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4608     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4609     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4610     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4611     spapr_caps_add_properties(smc);
4612     smc->irq = &spapr_irq_dual;
4613     smc->dr_phb_enabled = true;
4614     smc->linux_pci_probe = true;
4615     smc->smp_threads_vsmt = true;
4616     smc->nr_xirqs = SPAPR_NR_XIRQS;
4617     xfc->match_nvt = spapr_match_nvt;
4618     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4619     vmc->quiesce = spapr_vof_quiesce;
4620     vmc->setprop = spapr_vof_setprop;
4621 }
4622 
4623 static const TypeInfo spapr_machine_info = {
4624     .name          = TYPE_SPAPR_MACHINE,
4625     .parent        = TYPE_MACHINE,
4626     .abstract      = true,
4627     .instance_size = sizeof(SpaprMachineState),
4628     .instance_init = spapr_instance_init,
4629     .instance_finalize = spapr_machine_finalizefn,
4630     .class_size    = sizeof(SpaprMachineClass),
4631     .class_init    = spapr_machine_class_init,
4632     .interfaces = (InterfaceInfo[]) {
4633         { TYPE_FW_PATH_PROVIDER },
4634         { TYPE_NMI },
4635         { TYPE_HOTPLUG_HANDLER },
4636         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4637         { TYPE_XICS_FABRIC },
4638         { TYPE_INTERRUPT_STATS_PROVIDER },
4639         { TYPE_XIVE_FABRIC },
4640         { TYPE_VOF_MACHINE_IF },
4641         { }
4642     },
4643 };
4644 
4645 static void spapr_machine_latest_class_options(MachineClass *mc)
4646 {
4647     mc->alias = "pseries";
4648     mc->is_default = true;
4649 }
4650 
4651 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4652     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4653                                                     void *data)      \
4654     {                                                                \
4655         MachineClass *mc = MACHINE_CLASS(oc);                        \
4656         spapr_machine_##suffix##_class_options(mc);                  \
4657         if (latest) {                                                \
4658             spapr_machine_latest_class_options(mc);                  \
4659         }                                                            \
4660     }                                                                \
4661     static const TypeInfo spapr_machine_##suffix##_info = {          \
4662         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4663         .parent = TYPE_SPAPR_MACHINE,                                \
4664         .class_init = spapr_machine_##suffix##_class_init,           \
4665     };                                                               \
4666     static void spapr_machine_register_##suffix(void)                \
4667     {                                                                \
4668         type_register(&spapr_machine_##suffix##_info);               \
4669     }                                                                \
4670     type_init(spapr_machine_register_##suffix)
4671 
4672 /*
4673  * pseries-7.0
4674  */
4675 static void spapr_machine_7_0_class_options(MachineClass *mc)
4676 {
4677     /* Defaults for the latest behaviour inherited from the base class */
4678 }
4679 
4680 DEFINE_SPAPR_MACHINE(7_0, "7.0", true);
4681 
4682 /*
4683  * pseries-6.2
4684  */
4685 static void spapr_machine_6_2_class_options(MachineClass *mc)
4686 {
4687     spapr_machine_7_0_class_options(mc);
4688     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4689 }
4690 
4691 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4692 
4693 /*
4694  * pseries-6.1
4695  */
4696 static void spapr_machine_6_1_class_options(MachineClass *mc)
4697 {
4698     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4699 
4700     spapr_machine_6_2_class_options(mc);
4701     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4702     smc->pre_6_2_numa_affinity = true;
4703     mc->smp_props.prefer_sockets = true;
4704 }
4705 
4706 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4707 
4708 /*
4709  * pseries-6.0
4710  */
4711 static void spapr_machine_6_0_class_options(MachineClass *mc)
4712 {
4713     spapr_machine_6_1_class_options(mc);
4714     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4715 }
4716 
4717 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4718 
4719 /*
4720  * pseries-5.2
4721  */
4722 static void spapr_machine_5_2_class_options(MachineClass *mc)
4723 {
4724     spapr_machine_6_0_class_options(mc);
4725     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4726 }
4727 
4728 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4729 
4730 /*
4731  * pseries-5.1
4732  */
4733 static void spapr_machine_5_1_class_options(MachineClass *mc)
4734 {
4735     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4736 
4737     spapr_machine_5_2_class_options(mc);
4738     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4739     smc->pre_5_2_numa_associativity = true;
4740 }
4741 
4742 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4743 
4744 /*
4745  * pseries-5.0
4746  */
4747 static void spapr_machine_5_0_class_options(MachineClass *mc)
4748 {
4749     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4750     static GlobalProperty compat[] = {
4751         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4752     };
4753 
4754     spapr_machine_5_1_class_options(mc);
4755     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4756     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4757     mc->numa_mem_supported = true;
4758     smc->pre_5_1_assoc_refpoints = true;
4759 }
4760 
4761 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4762 
4763 /*
4764  * pseries-4.2
4765  */
4766 static void spapr_machine_4_2_class_options(MachineClass *mc)
4767 {
4768     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4769 
4770     spapr_machine_5_0_class_options(mc);
4771     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4772     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4773     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4774     smc->rma_limit = 16 * GiB;
4775     mc->nvdimm_supported = false;
4776 }
4777 
4778 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4779 
4780 /*
4781  * pseries-4.1
4782  */
4783 static void spapr_machine_4_1_class_options(MachineClass *mc)
4784 {
4785     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4786     static GlobalProperty compat[] = {
4787         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4788         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4789     };
4790 
4791     spapr_machine_4_2_class_options(mc);
4792     smc->linux_pci_probe = false;
4793     smc->smp_threads_vsmt = false;
4794     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4795     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4796 }
4797 
4798 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4799 
4800 /*
4801  * pseries-4.0
4802  */
4803 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4804                               uint64_t *buid, hwaddr *pio,
4805                               hwaddr *mmio32, hwaddr *mmio64,
4806                               unsigned n_dma, uint32_t *liobns,
4807                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4808 {
4809     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4810                              liobns, nv2gpa, nv2atsd, errp)) {
4811         return false;
4812     }
4813 
4814     *nv2gpa = 0;
4815     *nv2atsd = 0;
4816     return true;
4817 }
4818 static void spapr_machine_4_0_class_options(MachineClass *mc)
4819 {
4820     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4821 
4822     spapr_machine_4_1_class_options(mc);
4823     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4824     smc->phb_placement = phb_placement_4_0;
4825     smc->irq = &spapr_irq_xics;
4826     smc->pre_4_1_migration = true;
4827 }
4828 
4829 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4830 
4831 /*
4832  * pseries-3.1
4833  */
4834 static void spapr_machine_3_1_class_options(MachineClass *mc)
4835 {
4836     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4837 
4838     spapr_machine_4_0_class_options(mc);
4839     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4840 
4841     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4842     smc->update_dt_enabled = false;
4843     smc->dr_phb_enabled = false;
4844     smc->broken_host_serial_model = true;
4845     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4846     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4847     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4848     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4849 }
4850 
4851 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4852 
4853 /*
4854  * pseries-3.0
4855  */
4856 
4857 static void spapr_machine_3_0_class_options(MachineClass *mc)
4858 {
4859     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4860 
4861     spapr_machine_3_1_class_options(mc);
4862     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4863 
4864     smc->legacy_irq_allocation = true;
4865     smc->nr_xirqs = 0x400;
4866     smc->irq = &spapr_irq_xics_legacy;
4867 }
4868 
4869 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4870 
4871 /*
4872  * pseries-2.12
4873  */
4874 static void spapr_machine_2_12_class_options(MachineClass *mc)
4875 {
4876     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4877     static GlobalProperty compat[] = {
4878         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4879         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4880     };
4881 
4882     spapr_machine_3_0_class_options(mc);
4883     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4884     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4885 
4886     /* We depend on kvm_enabled() to choose a default value for the
4887      * hpt-max-page-size capability. Of course we can't do it here
4888      * because this is too early and the HW accelerator isn't initialzed
4889      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4890      */
4891     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4892 }
4893 
4894 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4895 
4896 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4897 {
4898     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4899 
4900     spapr_machine_2_12_class_options(mc);
4901     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4902     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4903     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4904 }
4905 
4906 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4907 
4908 /*
4909  * pseries-2.11
4910  */
4911 
4912 static void spapr_machine_2_11_class_options(MachineClass *mc)
4913 {
4914     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4915 
4916     spapr_machine_2_12_class_options(mc);
4917     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4918     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4919 }
4920 
4921 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4922 
4923 /*
4924  * pseries-2.10
4925  */
4926 
4927 static void spapr_machine_2_10_class_options(MachineClass *mc)
4928 {
4929     spapr_machine_2_11_class_options(mc);
4930     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4931 }
4932 
4933 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4934 
4935 /*
4936  * pseries-2.9
4937  */
4938 
4939 static void spapr_machine_2_9_class_options(MachineClass *mc)
4940 {
4941     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4942     static GlobalProperty compat[] = {
4943         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4944     };
4945 
4946     spapr_machine_2_10_class_options(mc);
4947     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4948     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4949     smc->pre_2_10_has_unused_icps = true;
4950     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4951 }
4952 
4953 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4954 
4955 /*
4956  * pseries-2.8
4957  */
4958 
4959 static void spapr_machine_2_8_class_options(MachineClass *mc)
4960 {
4961     static GlobalProperty compat[] = {
4962         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4963     };
4964 
4965     spapr_machine_2_9_class_options(mc);
4966     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4967     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4968     mc->numa_mem_align_shift = 23;
4969 }
4970 
4971 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4972 
4973 /*
4974  * pseries-2.7
4975  */
4976 
4977 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4978                               uint64_t *buid, hwaddr *pio,
4979                               hwaddr *mmio32, hwaddr *mmio64,
4980                               unsigned n_dma, uint32_t *liobns,
4981                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4982 {
4983     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4984     const uint64_t base_buid = 0x800000020000000ULL;
4985     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4986     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4987     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4988     const uint32_t max_index = 255;
4989     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4990 
4991     uint64_t ram_top = MACHINE(spapr)->ram_size;
4992     hwaddr phb0_base, phb_base;
4993     int i;
4994 
4995     /* Do we have device memory? */
4996     if (MACHINE(spapr)->maxram_size > ram_top) {
4997         /* Can't just use maxram_size, because there may be an
4998          * alignment gap between normal and device memory regions
4999          */
5000         ram_top = MACHINE(spapr)->device_memory->base +
5001             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5002     }
5003 
5004     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5005 
5006     if (index > max_index) {
5007         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5008                    max_index);
5009         return false;
5010     }
5011 
5012     *buid = base_buid + index;
5013     for (i = 0; i < n_dma; ++i) {
5014         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5015     }
5016 
5017     phb_base = phb0_base + index * phb_spacing;
5018     *pio = phb_base + pio_offset;
5019     *mmio32 = phb_base + mmio_offset;
5020     /*
5021      * We don't set the 64-bit MMIO window, relying on the PHB's
5022      * fallback behaviour of automatically splitting a large "32-bit"
5023      * window into contiguous 32-bit and 64-bit windows
5024      */
5025 
5026     *nv2gpa = 0;
5027     *nv2atsd = 0;
5028     return true;
5029 }
5030 
5031 static void spapr_machine_2_7_class_options(MachineClass *mc)
5032 {
5033     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5034     static GlobalProperty compat[] = {
5035         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5036         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5037         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5038         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5039     };
5040 
5041     spapr_machine_2_8_class_options(mc);
5042     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5043     mc->default_machine_opts = "modern-hotplug-events=off";
5044     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5045     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5046     smc->phb_placement = phb_placement_2_7;
5047 }
5048 
5049 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5050 
5051 /*
5052  * pseries-2.6
5053  */
5054 
5055 static void spapr_machine_2_6_class_options(MachineClass *mc)
5056 {
5057     static GlobalProperty compat[] = {
5058         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5059     };
5060 
5061     spapr_machine_2_7_class_options(mc);
5062     mc->has_hotpluggable_cpus = false;
5063     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5064     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5065 }
5066 
5067 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5068 
5069 /*
5070  * pseries-2.5
5071  */
5072 
5073 static void spapr_machine_2_5_class_options(MachineClass *mc)
5074 {
5075     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5076     static GlobalProperty compat[] = {
5077         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5078     };
5079 
5080     spapr_machine_2_6_class_options(mc);
5081     smc->use_ohci_by_default = true;
5082     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5083     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5084 }
5085 
5086 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5087 
5088 /*
5089  * pseries-2.4
5090  */
5091 
5092 static void spapr_machine_2_4_class_options(MachineClass *mc)
5093 {
5094     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5095 
5096     spapr_machine_2_5_class_options(mc);
5097     smc->dr_lmb_enabled = false;
5098     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5099 }
5100 
5101 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5102 
5103 /*
5104  * pseries-2.3
5105  */
5106 
5107 static void spapr_machine_2_3_class_options(MachineClass *mc)
5108 {
5109     static GlobalProperty compat[] = {
5110         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5111     };
5112     spapr_machine_2_4_class_options(mc);
5113     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5114     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5115 }
5116 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5117 
5118 /*
5119  * pseries-2.2
5120  */
5121 
5122 static void spapr_machine_2_2_class_options(MachineClass *mc)
5123 {
5124     static GlobalProperty compat[] = {
5125         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5126     };
5127 
5128     spapr_machine_2_3_class_options(mc);
5129     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5130     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5131     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5132 }
5133 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5134 
5135 /*
5136  * pseries-2.1
5137  */
5138 
5139 static void spapr_machine_2_1_class_options(MachineClass *mc)
5140 {
5141     spapr_machine_2_2_class_options(mc);
5142     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5143 }
5144 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5145 
5146 static void spapr_machine_register_types(void)
5147 {
5148     type_register_static(&spapr_machine_info);
5149 }
5150 
5151 type_init(spapr_machine_register_types)
5152