xref: /openbmc/qemu/hw/ppc/spapr.c (revision 5dab5abe623d97929382158b43b3fd545e8edd62)
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 (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
727         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
728     } else {
729         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
730     }
731 
732     /* Advertise DFP (Decimal Floating Point) if available
733      *   0 / no property == no DFP
734      *   1               == DFP available */
735     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
736         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
737     }
738 
739     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
740                                                       sizeof(page_sizes_prop));
741     if (page_sizes_prop_size) {
742         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
743                           page_sizes_prop, page_sizes_prop_size)));
744     }
745 
746     spapr_dt_pa_features(spapr, cpu, fdt, offset);
747 
748     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
749                            cs->cpu_index / vcpus_per_socket)));
750 
751     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
752                       pft_size_prop, sizeof(pft_size_prop))));
753 
754     if (ms->numa_state->num_nodes > 1) {
755         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
756     }
757 
758     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
759 
760     if (pcc->radix_page_info) {
761         for (i = 0; i < pcc->radix_page_info->count; i++) {
762             radix_AP_encodings[i] =
763                 cpu_to_be32(pcc->radix_page_info->entries[i]);
764         }
765         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
766                           radix_AP_encodings,
767                           pcc->radix_page_info->count *
768                           sizeof(radix_AP_encodings[0]))));
769     }
770 
771     /*
772      * We set this property to let the guest know that it can use the large
773      * decrementer and its width in bits.
774      */
775     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
776         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
777                               pcc->lrg_decr_bits)));
778 }
779 
780 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
781 {
782     CPUState **rev;
783     CPUState *cs;
784     int n_cpus;
785     int cpus_offset;
786     int i;
787 
788     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
789     _FDT(cpus_offset);
790     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
791     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
792 
793     /*
794      * We walk the CPUs in reverse order to ensure that CPU DT nodes
795      * created by fdt_add_subnode() end up in the right order in FDT
796      * for the guest kernel the enumerate the CPUs correctly.
797      *
798      * The CPU list cannot be traversed in reverse order, so we need
799      * to do extra work.
800      */
801     n_cpus = 0;
802     rev = NULL;
803     CPU_FOREACH(cs) {
804         rev = g_renew(CPUState *, rev, n_cpus + 1);
805         rev[n_cpus++] = cs;
806     }
807 
808     for (i = n_cpus - 1; i >= 0; i--) {
809         CPUState *cs = rev[i];
810         PowerPCCPU *cpu = POWERPC_CPU(cs);
811         int index = spapr_get_vcpu_id(cpu);
812         DeviceClass *dc = DEVICE_GET_CLASS(cs);
813         g_autofree char *nodename = NULL;
814         int offset;
815 
816         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
817             continue;
818         }
819 
820         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
821         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
822         _FDT(offset);
823         spapr_dt_cpu(cs, fdt, offset, spapr);
824     }
825 
826     g_free(rev);
827 }
828 
829 static int spapr_dt_rng(void *fdt)
830 {
831     int node;
832     int ret;
833 
834     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
835     if (node <= 0) {
836         return -1;
837     }
838     ret = fdt_setprop_string(fdt, node, "device_type",
839                              "ibm,platform-facilities");
840     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
841     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
842 
843     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
844     if (node <= 0) {
845         return -1;
846     }
847     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
848 
849     return ret ? -1 : 0;
850 }
851 
852 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
853 {
854     MachineState *ms = MACHINE(spapr);
855     int rtas;
856     GString *hypertas = g_string_sized_new(256);
857     GString *qemu_hypertas = g_string_sized_new(256);
858     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
859         memory_region_size(&MACHINE(spapr)->device_memory->mr);
860     uint32_t lrdr_capacity[] = {
861         cpu_to_be32(max_device_addr >> 32),
862         cpu_to_be32(max_device_addr & 0xffffffff),
863         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
864         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
865         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
866     };
867 
868     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
869 
870     /* hypertas */
871     add_str(hypertas, "hcall-pft");
872     add_str(hypertas, "hcall-term");
873     add_str(hypertas, "hcall-dabr");
874     add_str(hypertas, "hcall-interrupt");
875     add_str(hypertas, "hcall-tce");
876     add_str(hypertas, "hcall-vio");
877     add_str(hypertas, "hcall-splpar");
878     add_str(hypertas, "hcall-join");
879     add_str(hypertas, "hcall-bulk");
880     add_str(hypertas, "hcall-set-mode");
881     add_str(hypertas, "hcall-sprg0");
882     add_str(hypertas, "hcall-copy");
883     add_str(hypertas, "hcall-debug");
884     add_str(hypertas, "hcall-vphn");
885     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
886         add_str(hypertas, "hcall-rpt-invalidate");
887     }
888 
889     add_str(qemu_hypertas, "hcall-memop1");
890 
891     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
892         add_str(hypertas, "hcall-multi-tce");
893     }
894 
895     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
896         add_str(hypertas, "hcall-hpt-resize");
897     }
898 
899     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
900                      hypertas->str, hypertas->len));
901     g_string_free(hypertas, TRUE);
902     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
903                      qemu_hypertas->str, qemu_hypertas->len));
904     g_string_free(qemu_hypertas, TRUE);
905 
906     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
907 
908     /*
909      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
910      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
911      *
912      * The system reset requirements are driven by existing Linux and PowerVM
913      * implementation which (contrary to PAPR) saves r3 in the error log
914      * structure like machine check, so Linux expects to find the saved r3
915      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
916      * does not look at the error value).
917      *
918      * System reset interrupts are not subject to interlock like machine
919      * check, so this memory area could be corrupted if the sreset is
920      * interrupted by a machine check (or vice versa) if it was shared. To
921      * prevent this, system reset uses per-CPU areas for the sreset save
922      * area. A system reset that interrupts a system reset handler could
923      * still overwrite this area, but Linux doesn't try to recover in that
924      * case anyway.
925      *
926      * The extra 8 bytes is required because Linux's FWNMI error log check
927      * is off-by-one.
928      *
929      * RTAS_MIN_SIZE is required for the RTAS blob itself.
930      */
931     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
932                           RTAS_ERROR_LOG_MAX +
933                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
934                           sizeof(uint64_t)));
935     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
936                           RTAS_ERROR_LOG_MAX));
937     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
938                           RTAS_EVENT_SCAN_RATE));
939 
940     g_assert(msi_nonbroken);
941     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
942 
943     /*
944      * According to PAPR, rtas ibm,os-term does not guarantee a return
945      * back to the guest cpu.
946      *
947      * While an additional ibm,extended-os-term property indicates
948      * that rtas call return will always occur. Set this property.
949      */
950     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
951 
952     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
953                      lrdr_capacity, sizeof(lrdr_capacity)));
954 
955     spapr_dt_rtas_tokens(fdt, rtas);
956 }
957 
958 /*
959  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
960  * and the XIVE features that the guest may request and thus the valid
961  * values for bytes 23..26 of option vector 5:
962  */
963 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
964                                           int chosen)
965 {
966     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
967 
968     char val[2 * 4] = {
969         23, 0x00, /* XICS / XIVE mode */
970         24, 0x00, /* Hash/Radix, filled in below. */
971         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
972         26, 0x40, /* Radix options: GTSE == yes. */
973     };
974 
975     if (spapr->irq->xics && spapr->irq->xive) {
976         val[1] = SPAPR_OV5_XIVE_BOTH;
977     } else if (spapr->irq->xive) {
978         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
979     } else {
980         assert(spapr->irq->xics);
981         val[1] = SPAPR_OV5_XIVE_LEGACY;
982     }
983 
984     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
985                           first_ppc_cpu->compat_pvr)) {
986         /*
987          * If we're in a pre POWER9 compat mode then the guest should
988          * do hash and use the legacy interrupt mode
989          */
990         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
991         val[3] = 0x00; /* Hash */
992         spapr_check_mmu_mode(false);
993     } else if (kvm_enabled()) {
994         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
995             val[3] = 0x80; /* OV5_MMU_BOTH */
996         } else if (kvmppc_has_cap_mmu_radix()) {
997             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
998         } else {
999             val[3] = 0x00; /* Hash */
1000         }
1001     } else {
1002         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1003         val[3] = 0xC0;
1004     }
1005     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1006                      val, sizeof(val)));
1007 }
1008 
1009 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1010 {
1011     MachineState *machine = MACHINE(spapr);
1012     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1013     int chosen;
1014 
1015     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1016 
1017     if (reset) {
1018         const char *boot_device = spapr->boot_device;
1019         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1020         size_t cb = 0;
1021         char *bootlist = get_boot_devices_list(&cb);
1022 
1023         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1024             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1025                                     machine->kernel_cmdline));
1026         }
1027 
1028         if (spapr->initrd_size) {
1029             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1030                                   spapr->initrd_base));
1031             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1032                                   spapr->initrd_base + spapr->initrd_size));
1033         }
1034 
1035         if (spapr->kernel_size) {
1036             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1037                                   cpu_to_be64(spapr->kernel_size) };
1038 
1039             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1040                          &kprop, sizeof(kprop)));
1041             if (spapr->kernel_le) {
1042                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1043             }
1044         }
1045         if (boot_menu) {
1046             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1047         }
1048         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1049         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1050         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1051 
1052         if (cb && bootlist) {
1053             int i;
1054 
1055             for (i = 0; i < cb; i++) {
1056                 if (bootlist[i] == '\n') {
1057                     bootlist[i] = ' ';
1058                 }
1059             }
1060             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1061         }
1062 
1063         if (boot_device && strlen(boot_device)) {
1064             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1065         }
1066 
1067         if (!spapr->has_graphics && stdout_path) {
1068             /*
1069              * "linux,stdout-path" and "stdout" properties are
1070              * deprecated by linux kernel. New platforms should only
1071              * use the "stdout-path" property. Set the new property
1072              * and continue using older property to remain compatible
1073              * with the existing firmware.
1074              */
1075             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1076             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1077         }
1078 
1079         /*
1080          * We can deal with BAR reallocation just fine, advertise it
1081          * to the guest
1082          */
1083         if (smc->linux_pci_probe) {
1084             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1085         }
1086 
1087         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1088 
1089         g_free(stdout_path);
1090         g_free(bootlist);
1091     }
1092 
1093     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1094 }
1095 
1096 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1097 {
1098     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1099      * KVM to work under pHyp with some guest co-operation */
1100     int hypervisor;
1101     uint8_t hypercall[16];
1102 
1103     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1104     /* indicate KVM hypercall interface */
1105     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1106     if (kvmppc_has_cap_fixup_hcalls()) {
1107         /*
1108          * Older KVM versions with older guest kernels were broken
1109          * with the magic page, don't allow the guest to map it.
1110          */
1111         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1112                                   sizeof(hypercall))) {
1113             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1114                              hypercall, sizeof(hypercall)));
1115         }
1116     }
1117 }
1118 
1119 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1120 {
1121     MachineState *machine = MACHINE(spapr);
1122     MachineClass *mc = MACHINE_GET_CLASS(machine);
1123     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1124     uint32_t root_drc_type_mask = 0;
1125     int ret;
1126     void *fdt;
1127     SpaprPhbState *phb;
1128     char *buf;
1129 
1130     fdt = g_malloc0(space);
1131     _FDT((fdt_create_empty_tree(fdt, space)));
1132 
1133     /* Root node */
1134     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1135     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1136     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1137 
1138     /* Guest UUID & Name*/
1139     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1140     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1141     if (qemu_uuid_set) {
1142         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1143     }
1144     g_free(buf);
1145 
1146     if (qemu_get_vm_name()) {
1147         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1148                                 qemu_get_vm_name()));
1149     }
1150 
1151     /* Host Model & Serial Number */
1152     if (spapr->host_model) {
1153         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1154     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1155         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1156         g_free(buf);
1157     }
1158 
1159     if (spapr->host_serial) {
1160         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1161     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1162         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1163         g_free(buf);
1164     }
1165 
1166     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1167     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1168 
1169     /* /interrupt controller */
1170     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1171 
1172     ret = spapr_dt_memory(spapr, fdt);
1173     if (ret < 0) {
1174         error_report("couldn't setup memory nodes in fdt");
1175         exit(1);
1176     }
1177 
1178     /* /vdevice */
1179     spapr_dt_vdevice(spapr->vio_bus, fdt);
1180 
1181     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1182         ret = spapr_dt_rng(fdt);
1183         if (ret < 0) {
1184             error_report("could not set up rng device in the fdt");
1185             exit(1);
1186         }
1187     }
1188 
1189     QLIST_FOREACH(phb, &spapr->phbs, list) {
1190         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1191         if (ret < 0) {
1192             error_report("couldn't setup PCI devices in fdt");
1193             exit(1);
1194         }
1195     }
1196 
1197     spapr_dt_cpus(fdt, spapr);
1198 
1199     /* ibm,drc-indexes and friends */
1200     if (smc->dr_lmb_enabled) {
1201         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1202     }
1203     if (smc->dr_phb_enabled) {
1204         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1205     }
1206     if (mc->nvdimm_supported) {
1207         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1208     }
1209     if (root_drc_type_mask) {
1210         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1211     }
1212 
1213     if (mc->has_hotpluggable_cpus) {
1214         int offset = fdt_path_offset(fdt, "/cpus");
1215         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1216         if (ret < 0) {
1217             error_report("Couldn't set up CPU DR device tree properties");
1218             exit(1);
1219         }
1220     }
1221 
1222     /* /event-sources */
1223     spapr_dt_events(spapr, fdt);
1224 
1225     /* /rtas */
1226     spapr_dt_rtas(spapr, fdt);
1227 
1228     /* /chosen */
1229     spapr_dt_chosen(spapr, fdt, reset);
1230 
1231     /* /hypervisor */
1232     if (kvm_enabled()) {
1233         spapr_dt_hypervisor(spapr, fdt);
1234     }
1235 
1236     /* Build memory reserve map */
1237     if (reset) {
1238         if (spapr->kernel_size) {
1239             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1240                                   spapr->kernel_size)));
1241         }
1242         if (spapr->initrd_size) {
1243             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1244                                   spapr->initrd_size)));
1245         }
1246     }
1247 
1248     /* NVDIMM devices */
1249     if (mc->nvdimm_supported) {
1250         spapr_dt_persistent_memory(spapr, fdt);
1251     }
1252 
1253     return fdt;
1254 }
1255 
1256 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1257 {
1258     SpaprMachineState *spapr = opaque;
1259 
1260     return (addr & 0x0fffffff) + spapr->kernel_addr;
1261 }
1262 
1263 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1264                                     PowerPCCPU *cpu)
1265 {
1266     CPUPPCState *env = &cpu->env;
1267 
1268     /* The TCG path should also be holding the BQL at this point */
1269     g_assert(qemu_mutex_iothread_locked());
1270 
1271     if (msr_pr) {
1272         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1273         env->gpr[3] = H_PRIVILEGE;
1274     } else {
1275         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1276     }
1277 }
1278 
1279 struct LPCRSyncState {
1280     target_ulong value;
1281     target_ulong mask;
1282 };
1283 
1284 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1285 {
1286     struct LPCRSyncState *s = arg.host_ptr;
1287     PowerPCCPU *cpu = POWERPC_CPU(cs);
1288     CPUPPCState *env = &cpu->env;
1289     target_ulong lpcr;
1290 
1291     cpu_synchronize_state(cs);
1292     lpcr = env->spr[SPR_LPCR];
1293     lpcr &= ~s->mask;
1294     lpcr |= s->value;
1295     ppc_store_lpcr(cpu, lpcr);
1296 }
1297 
1298 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1299 {
1300     CPUState *cs;
1301     struct LPCRSyncState s = {
1302         .value = value,
1303         .mask = mask
1304     };
1305     CPU_FOREACH(cs) {
1306         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1307     }
1308 }
1309 
1310 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1311 {
1312     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1313 
1314     /* Copy PATE1:GR into PATE0:HR */
1315     entry->dw0 = spapr->patb_entry & PATE0_HR;
1316     entry->dw1 = spapr->patb_entry;
1317 }
1318 
1319 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1320 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1321 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1322 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1323 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1324 
1325 /*
1326  * Get the fd to access the kernel htab, re-opening it if necessary
1327  */
1328 static int get_htab_fd(SpaprMachineState *spapr)
1329 {
1330     Error *local_err = NULL;
1331 
1332     if (spapr->htab_fd >= 0) {
1333         return spapr->htab_fd;
1334     }
1335 
1336     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1337     if (spapr->htab_fd < 0) {
1338         error_report_err(local_err);
1339     }
1340 
1341     return spapr->htab_fd;
1342 }
1343 
1344 void close_htab_fd(SpaprMachineState *spapr)
1345 {
1346     if (spapr->htab_fd >= 0) {
1347         close(spapr->htab_fd);
1348     }
1349     spapr->htab_fd = -1;
1350 }
1351 
1352 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1353 {
1354     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1355 
1356     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1357 }
1358 
1359 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1360 {
1361     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1362 
1363     assert(kvm_enabled());
1364 
1365     if (!spapr->htab) {
1366         return 0;
1367     }
1368 
1369     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1370 }
1371 
1372 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1373                                                 hwaddr ptex, int n)
1374 {
1375     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1376     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1377 
1378     if (!spapr->htab) {
1379         /*
1380          * HTAB is controlled by KVM. Fetch into temporary buffer
1381          */
1382         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1383         kvmppc_read_hptes(hptes, ptex, n);
1384         return hptes;
1385     }
1386 
1387     /*
1388      * HTAB is controlled by QEMU. Just point to the internally
1389      * accessible PTEG.
1390      */
1391     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1392 }
1393 
1394 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1395                               const ppc_hash_pte64_t *hptes,
1396                               hwaddr ptex, int n)
1397 {
1398     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1399 
1400     if (!spapr->htab) {
1401         g_free((void *)hptes);
1402     }
1403 
1404     /* Nothing to do for qemu managed HPT */
1405 }
1406 
1407 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1408                       uint64_t pte0, uint64_t pte1)
1409 {
1410     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1411     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1412 
1413     if (!spapr->htab) {
1414         kvmppc_write_hpte(ptex, pte0, pte1);
1415     } else {
1416         if (pte0 & HPTE64_V_VALID) {
1417             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1418             /*
1419              * When setting valid, we write PTE1 first. This ensures
1420              * proper synchronization with the reading code in
1421              * ppc_hash64_pteg_search()
1422              */
1423             smp_wmb();
1424             stq_p(spapr->htab + offset, pte0);
1425         } else {
1426             stq_p(spapr->htab + offset, pte0);
1427             /*
1428              * When clearing it we set PTE0 first. This ensures proper
1429              * synchronization with the reading code in
1430              * ppc_hash64_pteg_search()
1431              */
1432             smp_wmb();
1433             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1434         }
1435     }
1436 }
1437 
1438 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1439                              uint64_t pte1)
1440 {
1441     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1442     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1443 
1444     if (!spapr->htab) {
1445         /* There should always be a hash table when this is called */
1446         error_report("spapr_hpte_set_c called with no hash table !");
1447         return;
1448     }
1449 
1450     /* The HW performs a non-atomic byte update */
1451     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1452 }
1453 
1454 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1455                              uint64_t pte1)
1456 {
1457     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1458     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1459 
1460     if (!spapr->htab) {
1461         /* There should always be a hash table when this is called */
1462         error_report("spapr_hpte_set_r called with no hash table !");
1463         return;
1464     }
1465 
1466     /* The HW performs a non-atomic byte update */
1467     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1468 }
1469 
1470 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1471 {
1472     int shift;
1473 
1474     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1475      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1476      * that's much more than is needed for Linux guests */
1477     shift = ctz64(pow2ceil(ramsize)) - 7;
1478     shift = MAX(shift, 18); /* Minimum architected size */
1479     shift = MIN(shift, 46); /* Maximum architected size */
1480     return shift;
1481 }
1482 
1483 void spapr_free_hpt(SpaprMachineState *spapr)
1484 {
1485     g_free(spapr->htab);
1486     spapr->htab = NULL;
1487     spapr->htab_shift = 0;
1488     close_htab_fd(spapr);
1489 }
1490 
1491 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1492 {
1493     ERRP_GUARD();
1494     long rc;
1495 
1496     /* Clean up any HPT info from a previous boot */
1497     spapr_free_hpt(spapr);
1498 
1499     rc = kvmppc_reset_htab(shift);
1500 
1501     if (rc == -EOPNOTSUPP) {
1502         error_setg(errp, "HPT not supported in nested guests");
1503         return -EOPNOTSUPP;
1504     }
1505 
1506     if (rc < 0) {
1507         /* kernel-side HPT needed, but couldn't allocate one */
1508         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1509                          shift);
1510         error_append_hint(errp, "Try smaller maxmem?\n");
1511         return -errno;
1512     } else if (rc > 0) {
1513         /* kernel-side HPT allocated */
1514         if (rc != shift) {
1515             error_setg(errp,
1516                        "Requested order %d HPT, but kernel allocated order %ld",
1517                        shift, rc);
1518             error_append_hint(errp, "Try smaller maxmem?\n");
1519             return -ENOSPC;
1520         }
1521 
1522         spapr->htab_shift = shift;
1523         spapr->htab = NULL;
1524     } else {
1525         /* kernel-side HPT not needed, allocate in userspace instead */
1526         size_t size = 1ULL << shift;
1527         int i;
1528 
1529         spapr->htab = qemu_memalign(size, size);
1530         memset(spapr->htab, 0, size);
1531         spapr->htab_shift = shift;
1532 
1533         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1534             DIRTY_HPTE(HPTE(spapr->htab, i));
1535         }
1536     }
1537     /* We're setting up a hash table, so that means we're not radix */
1538     spapr->patb_entry = 0;
1539     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1540     return 0;
1541 }
1542 
1543 void spapr_setup_hpt(SpaprMachineState *spapr)
1544 {
1545     int hpt_shift;
1546 
1547     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1548         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1549     } else {
1550         uint64_t current_ram_size;
1551 
1552         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1553         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1554     }
1555     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1556 
1557     if (kvm_enabled()) {
1558         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1559 
1560         /* Check our RMA fits in the possible VRMA */
1561         if (vrma_limit < spapr->rma_size) {
1562             error_report("Unable to create %" HWADDR_PRIu
1563                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1564                          spapr->rma_size / MiB, vrma_limit / MiB);
1565             exit(EXIT_FAILURE);
1566         }
1567     }
1568 }
1569 
1570 void spapr_check_mmu_mode(bool guest_radix)
1571 {
1572     if (guest_radix) {
1573         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1574             error_report("Guest requested unavailable MMU mode (radix).");
1575             exit(EXIT_FAILURE);
1576         }
1577     } else {
1578         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1579             && !kvmppc_has_cap_mmu_hash_v3()) {
1580             error_report("Guest requested unavailable MMU mode (hash).");
1581             exit(EXIT_FAILURE);
1582         }
1583     }
1584 }
1585 
1586 static void spapr_machine_reset(MachineState *machine)
1587 {
1588     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1589     PowerPCCPU *first_ppc_cpu;
1590     hwaddr fdt_addr;
1591     void *fdt;
1592     int rc;
1593 
1594     pef_kvm_reset(machine->cgs, &error_fatal);
1595     spapr_caps_apply(spapr);
1596 
1597     first_ppc_cpu = POWERPC_CPU(first_cpu);
1598     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1599         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1600                               spapr->max_compat_pvr)) {
1601         /*
1602          * If using KVM with radix mode available, VCPUs can be started
1603          * without a HPT because KVM will start them in radix mode.
1604          * Set the GR bit in PATE so that we know there is no HPT.
1605          */
1606         spapr->patb_entry = PATE1_GR;
1607         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1608     } else {
1609         spapr_setup_hpt(spapr);
1610     }
1611 
1612     qemu_devices_reset();
1613 
1614     spapr_ovec_cleanup(spapr->ov5_cas);
1615     spapr->ov5_cas = spapr_ovec_new();
1616 
1617     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1618 
1619     /*
1620      * This is fixing some of the default configuration of the XIVE
1621      * devices. To be called after the reset of the machine devices.
1622      */
1623     spapr_irq_reset(spapr, &error_fatal);
1624 
1625     /*
1626      * There is no CAS under qtest. Simulate one to please the code that
1627      * depends on spapr->ov5_cas. This is especially needed to test device
1628      * unplug, so we do that before resetting the DRCs.
1629      */
1630     if (qtest_enabled()) {
1631         spapr_ovec_cleanup(spapr->ov5_cas);
1632         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1633     }
1634 
1635     /* DRC reset may cause a device to be unplugged. This will cause troubles
1636      * if this device is used by another device (eg, a running vhost backend
1637      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1638      * situations, we reset DRCs after all devices have been reset.
1639      */
1640     spapr_drc_reset_all(spapr);
1641 
1642     spapr_clear_pending_events(spapr);
1643 
1644     /*
1645      * We place the device tree just below either the top of the RMA,
1646      * or just below 2GB, whichever is lower, so that it can be
1647      * processed with 32-bit real mode code if necessary
1648      */
1649     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1650 
1651     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1652     if (spapr->vof) {
1653         spapr_vof_reset(spapr, fdt, &error_fatal);
1654         /*
1655          * Do not pack the FDT as the client may change properties.
1656          * VOF client does not expect the FDT so we do not load it to the VM.
1657          */
1658     } else {
1659         rc = fdt_pack(fdt);
1660         /* Should only fail if we've built a corrupted tree */
1661         assert(rc == 0);
1662 
1663         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1664                                   0, fdt_addr, 0);
1665         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1666     }
1667     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1668 
1669     g_free(spapr->fdt_blob);
1670     spapr->fdt_size = fdt_totalsize(fdt);
1671     spapr->fdt_initial_size = spapr->fdt_size;
1672     spapr->fdt_blob = fdt;
1673 
1674     /* Set up the entry state */
1675     first_ppc_cpu->env.gpr[5] = 0;
1676 
1677     spapr->fwnmi_system_reset_addr = -1;
1678     spapr->fwnmi_machine_check_addr = -1;
1679     spapr->fwnmi_machine_check_interlock = -1;
1680 
1681     /* Signal all vCPUs waiting on this condition */
1682     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1683 
1684     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1685 }
1686 
1687 static void spapr_create_nvram(SpaprMachineState *spapr)
1688 {
1689     DeviceState *dev = qdev_new("spapr-nvram");
1690     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1691 
1692     if (dinfo) {
1693         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1694                                 &error_fatal);
1695     }
1696 
1697     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1698 
1699     spapr->nvram = (struct SpaprNvram *)dev;
1700 }
1701 
1702 static void spapr_rtc_create(SpaprMachineState *spapr)
1703 {
1704     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1705                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1706                                        &error_fatal, NULL);
1707     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1708     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1709                               "date");
1710 }
1711 
1712 /* Returns whether we want to use VGA or not */
1713 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1714 {
1715     switch (vga_interface_type) {
1716     case VGA_NONE:
1717         return false;
1718     case VGA_DEVICE:
1719         return true;
1720     case VGA_STD:
1721     case VGA_VIRTIO:
1722     case VGA_CIRRUS:
1723         return pci_vga_init(pci_bus) != NULL;
1724     default:
1725         error_setg(errp,
1726                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1727         return false;
1728     }
1729 }
1730 
1731 static int spapr_pre_load(void *opaque)
1732 {
1733     int rc;
1734 
1735     rc = spapr_caps_pre_load(opaque);
1736     if (rc) {
1737         return rc;
1738     }
1739 
1740     return 0;
1741 }
1742 
1743 static int spapr_post_load(void *opaque, int version_id)
1744 {
1745     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1746     int err = 0;
1747 
1748     err = spapr_caps_post_migration(spapr);
1749     if (err) {
1750         return err;
1751     }
1752 
1753     /*
1754      * In earlier versions, there was no separate qdev for the PAPR
1755      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1756      * So when migrating from those versions, poke the incoming offset
1757      * value into the RTC device
1758      */
1759     if (version_id < 3) {
1760         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1761         if (err) {
1762             return err;
1763         }
1764     }
1765 
1766     if (kvm_enabled() && spapr->patb_entry) {
1767         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1768         bool radix = !!(spapr->patb_entry & PATE1_GR);
1769         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1770 
1771         /*
1772          * Update LPCR:HR and UPRT as they may not be set properly in
1773          * the stream
1774          */
1775         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1776                             LPCR_HR | LPCR_UPRT);
1777 
1778         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1779         if (err) {
1780             error_report("Process table config unsupported by the host");
1781             return -EINVAL;
1782         }
1783     }
1784 
1785     err = spapr_irq_post_load(spapr, version_id);
1786     if (err) {
1787         return err;
1788     }
1789 
1790     return err;
1791 }
1792 
1793 static int spapr_pre_save(void *opaque)
1794 {
1795     int rc;
1796 
1797     rc = spapr_caps_pre_save(opaque);
1798     if (rc) {
1799         return rc;
1800     }
1801 
1802     return 0;
1803 }
1804 
1805 static bool version_before_3(void *opaque, int version_id)
1806 {
1807     return version_id < 3;
1808 }
1809 
1810 static bool spapr_pending_events_needed(void *opaque)
1811 {
1812     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1813     return !QTAILQ_EMPTY(&spapr->pending_events);
1814 }
1815 
1816 static const VMStateDescription vmstate_spapr_event_entry = {
1817     .name = "spapr_event_log_entry",
1818     .version_id = 1,
1819     .minimum_version_id = 1,
1820     .fields = (VMStateField[]) {
1821         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1822         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1823         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1824                                      NULL, extended_length),
1825         VMSTATE_END_OF_LIST()
1826     },
1827 };
1828 
1829 static const VMStateDescription vmstate_spapr_pending_events = {
1830     .name = "spapr_pending_events",
1831     .version_id = 1,
1832     .minimum_version_id = 1,
1833     .needed = spapr_pending_events_needed,
1834     .fields = (VMStateField[]) {
1835         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1836                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1837         VMSTATE_END_OF_LIST()
1838     },
1839 };
1840 
1841 static bool spapr_ov5_cas_needed(void *opaque)
1842 {
1843     SpaprMachineState *spapr = opaque;
1844     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1845     bool cas_needed;
1846 
1847     /* Prior to the introduction of SpaprOptionVector, we had two option
1848      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1849      * Both of these options encode machine topology into the device-tree
1850      * in such a way that the now-booted OS should still be able to interact
1851      * appropriately with QEMU regardless of what options were actually
1852      * negotiatied on the source side.
1853      *
1854      * As such, we can avoid migrating the CAS-negotiated options if these
1855      * are the only options available on the current machine/platform.
1856      * Since these are the only options available for pseries-2.7 and
1857      * earlier, this allows us to maintain old->new/new->old migration
1858      * compatibility.
1859      *
1860      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1861      * via default pseries-2.8 machines and explicit command-line parameters.
1862      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1863      * of the actual CAS-negotiated values to continue working properly. For
1864      * example, availability of memory unplug depends on knowing whether
1865      * OV5_HP_EVT was negotiated via CAS.
1866      *
1867      * Thus, for any cases where the set of available CAS-negotiatable
1868      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1869      * include the CAS-negotiated options in the migration stream, unless
1870      * if they affect boot time behaviour only.
1871      */
1872     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1873     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1874     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1875 
1876     /* We need extra information if we have any bits outside the mask
1877      * defined above */
1878     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1879 
1880     spapr_ovec_cleanup(ov5_mask);
1881 
1882     return cas_needed;
1883 }
1884 
1885 static const VMStateDescription vmstate_spapr_ov5_cas = {
1886     .name = "spapr_option_vector_ov5_cas",
1887     .version_id = 1,
1888     .minimum_version_id = 1,
1889     .needed = spapr_ov5_cas_needed,
1890     .fields = (VMStateField[]) {
1891         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1892                                  vmstate_spapr_ovec, SpaprOptionVector),
1893         VMSTATE_END_OF_LIST()
1894     },
1895 };
1896 
1897 static bool spapr_patb_entry_needed(void *opaque)
1898 {
1899     SpaprMachineState *spapr = opaque;
1900 
1901     return !!spapr->patb_entry;
1902 }
1903 
1904 static const VMStateDescription vmstate_spapr_patb_entry = {
1905     .name = "spapr_patb_entry",
1906     .version_id = 1,
1907     .minimum_version_id = 1,
1908     .needed = spapr_patb_entry_needed,
1909     .fields = (VMStateField[]) {
1910         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1911         VMSTATE_END_OF_LIST()
1912     },
1913 };
1914 
1915 static bool spapr_irq_map_needed(void *opaque)
1916 {
1917     SpaprMachineState *spapr = opaque;
1918 
1919     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1920 }
1921 
1922 static const VMStateDescription vmstate_spapr_irq_map = {
1923     .name = "spapr_irq_map",
1924     .version_id = 1,
1925     .minimum_version_id = 1,
1926     .needed = spapr_irq_map_needed,
1927     .fields = (VMStateField[]) {
1928         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1929         VMSTATE_END_OF_LIST()
1930     },
1931 };
1932 
1933 static bool spapr_dtb_needed(void *opaque)
1934 {
1935     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1936 
1937     return smc->update_dt_enabled;
1938 }
1939 
1940 static int spapr_dtb_pre_load(void *opaque)
1941 {
1942     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1943 
1944     g_free(spapr->fdt_blob);
1945     spapr->fdt_blob = NULL;
1946     spapr->fdt_size = 0;
1947 
1948     return 0;
1949 }
1950 
1951 static const VMStateDescription vmstate_spapr_dtb = {
1952     .name = "spapr_dtb",
1953     .version_id = 1,
1954     .minimum_version_id = 1,
1955     .needed = spapr_dtb_needed,
1956     .pre_load = spapr_dtb_pre_load,
1957     .fields = (VMStateField[]) {
1958         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1959         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1960         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1961                                      fdt_size),
1962         VMSTATE_END_OF_LIST()
1963     },
1964 };
1965 
1966 static bool spapr_fwnmi_needed(void *opaque)
1967 {
1968     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1969 
1970     return spapr->fwnmi_machine_check_addr != -1;
1971 }
1972 
1973 static int spapr_fwnmi_pre_save(void *opaque)
1974 {
1975     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1976 
1977     /*
1978      * Check if machine check handling is in progress and print a
1979      * warning message.
1980      */
1981     if (spapr->fwnmi_machine_check_interlock != -1) {
1982         warn_report("A machine check is being handled during migration. The"
1983                 "handler may run and log hardware error on the destination");
1984     }
1985 
1986     return 0;
1987 }
1988 
1989 static const VMStateDescription vmstate_spapr_fwnmi = {
1990     .name = "spapr_fwnmi",
1991     .version_id = 1,
1992     .minimum_version_id = 1,
1993     .needed = spapr_fwnmi_needed,
1994     .pre_save = spapr_fwnmi_pre_save,
1995     .fields = (VMStateField[]) {
1996         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1997         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1998         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1999         VMSTATE_END_OF_LIST()
2000     },
2001 };
2002 
2003 static const VMStateDescription vmstate_spapr = {
2004     .name = "spapr",
2005     .version_id = 3,
2006     .minimum_version_id = 1,
2007     .pre_load = spapr_pre_load,
2008     .post_load = spapr_post_load,
2009     .pre_save = spapr_pre_save,
2010     .fields = (VMStateField[]) {
2011         /* used to be @next_irq */
2012         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2013 
2014         /* RTC offset */
2015         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2016 
2017         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2018         VMSTATE_END_OF_LIST()
2019     },
2020     .subsections = (const VMStateDescription*[]) {
2021         &vmstate_spapr_ov5_cas,
2022         &vmstate_spapr_patb_entry,
2023         &vmstate_spapr_pending_events,
2024         &vmstate_spapr_cap_htm,
2025         &vmstate_spapr_cap_vsx,
2026         &vmstate_spapr_cap_dfp,
2027         &vmstate_spapr_cap_cfpc,
2028         &vmstate_spapr_cap_sbbc,
2029         &vmstate_spapr_cap_ibs,
2030         &vmstate_spapr_cap_hpt_maxpagesize,
2031         &vmstate_spapr_irq_map,
2032         &vmstate_spapr_cap_nested_kvm_hv,
2033         &vmstate_spapr_dtb,
2034         &vmstate_spapr_cap_large_decr,
2035         &vmstate_spapr_cap_ccf_assist,
2036         &vmstate_spapr_cap_fwnmi,
2037         &vmstate_spapr_fwnmi,
2038         &vmstate_spapr_cap_rpt_invalidate,
2039         NULL
2040     }
2041 };
2042 
2043 static int htab_save_setup(QEMUFile *f, void *opaque)
2044 {
2045     SpaprMachineState *spapr = opaque;
2046 
2047     /* "Iteration" header */
2048     if (!spapr->htab_shift) {
2049         qemu_put_be32(f, -1);
2050     } else {
2051         qemu_put_be32(f, spapr->htab_shift);
2052     }
2053 
2054     if (spapr->htab) {
2055         spapr->htab_save_index = 0;
2056         spapr->htab_first_pass = true;
2057     } else {
2058         if (spapr->htab_shift) {
2059             assert(kvm_enabled());
2060         }
2061     }
2062 
2063 
2064     return 0;
2065 }
2066 
2067 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2068                             int chunkstart, int n_valid, int n_invalid)
2069 {
2070     qemu_put_be32(f, chunkstart);
2071     qemu_put_be16(f, n_valid);
2072     qemu_put_be16(f, n_invalid);
2073     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2074                     HASH_PTE_SIZE_64 * n_valid);
2075 }
2076 
2077 static void htab_save_end_marker(QEMUFile *f)
2078 {
2079     qemu_put_be32(f, 0);
2080     qemu_put_be16(f, 0);
2081     qemu_put_be16(f, 0);
2082 }
2083 
2084 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2085                                  int64_t max_ns)
2086 {
2087     bool has_timeout = max_ns != -1;
2088     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2089     int index = spapr->htab_save_index;
2090     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2091 
2092     assert(spapr->htab_first_pass);
2093 
2094     do {
2095         int chunkstart;
2096 
2097         /* Consume invalid HPTEs */
2098         while ((index < htabslots)
2099                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2100             CLEAN_HPTE(HPTE(spapr->htab, index));
2101             index++;
2102         }
2103 
2104         /* Consume valid HPTEs */
2105         chunkstart = index;
2106         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2107                && HPTE_VALID(HPTE(spapr->htab, index))) {
2108             CLEAN_HPTE(HPTE(spapr->htab, index));
2109             index++;
2110         }
2111 
2112         if (index > chunkstart) {
2113             int n_valid = index - chunkstart;
2114 
2115             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2116 
2117             if (has_timeout &&
2118                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2119                 break;
2120             }
2121         }
2122     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2123 
2124     if (index >= htabslots) {
2125         assert(index == htabslots);
2126         index = 0;
2127         spapr->htab_first_pass = false;
2128     }
2129     spapr->htab_save_index = index;
2130 }
2131 
2132 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2133                                 int64_t max_ns)
2134 {
2135     bool final = max_ns < 0;
2136     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2137     int examined = 0, sent = 0;
2138     int index = spapr->htab_save_index;
2139     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2140 
2141     assert(!spapr->htab_first_pass);
2142 
2143     do {
2144         int chunkstart, invalidstart;
2145 
2146         /* Consume non-dirty HPTEs */
2147         while ((index < htabslots)
2148                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2149             index++;
2150             examined++;
2151         }
2152 
2153         chunkstart = index;
2154         /* Consume valid dirty HPTEs */
2155         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2156                && HPTE_DIRTY(HPTE(spapr->htab, index))
2157                && HPTE_VALID(HPTE(spapr->htab, index))) {
2158             CLEAN_HPTE(HPTE(spapr->htab, index));
2159             index++;
2160             examined++;
2161         }
2162 
2163         invalidstart = index;
2164         /* Consume invalid dirty HPTEs */
2165         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2166                && HPTE_DIRTY(HPTE(spapr->htab, index))
2167                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2168             CLEAN_HPTE(HPTE(spapr->htab, index));
2169             index++;
2170             examined++;
2171         }
2172 
2173         if (index > chunkstart) {
2174             int n_valid = invalidstart - chunkstart;
2175             int n_invalid = index - invalidstart;
2176 
2177             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2178             sent += index - chunkstart;
2179 
2180             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2181                 break;
2182             }
2183         }
2184 
2185         if (examined >= htabslots) {
2186             break;
2187         }
2188 
2189         if (index >= htabslots) {
2190             assert(index == htabslots);
2191             index = 0;
2192         }
2193     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2194 
2195     if (index >= htabslots) {
2196         assert(index == htabslots);
2197         index = 0;
2198     }
2199 
2200     spapr->htab_save_index = index;
2201 
2202     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2203 }
2204 
2205 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2206 #define MAX_KVM_BUF_SIZE    2048
2207 
2208 static int htab_save_iterate(QEMUFile *f, void *opaque)
2209 {
2210     SpaprMachineState *spapr = opaque;
2211     int fd;
2212     int rc = 0;
2213 
2214     /* Iteration header */
2215     if (!spapr->htab_shift) {
2216         qemu_put_be32(f, -1);
2217         return 1;
2218     } else {
2219         qemu_put_be32(f, 0);
2220     }
2221 
2222     if (!spapr->htab) {
2223         assert(kvm_enabled());
2224 
2225         fd = get_htab_fd(spapr);
2226         if (fd < 0) {
2227             return fd;
2228         }
2229 
2230         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2231         if (rc < 0) {
2232             return rc;
2233         }
2234     } else  if (spapr->htab_first_pass) {
2235         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2236     } else {
2237         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2238     }
2239 
2240     htab_save_end_marker(f);
2241 
2242     return rc;
2243 }
2244 
2245 static int htab_save_complete(QEMUFile *f, void *opaque)
2246 {
2247     SpaprMachineState *spapr = opaque;
2248     int fd;
2249 
2250     /* Iteration header */
2251     if (!spapr->htab_shift) {
2252         qemu_put_be32(f, -1);
2253         return 0;
2254     } else {
2255         qemu_put_be32(f, 0);
2256     }
2257 
2258     if (!spapr->htab) {
2259         int rc;
2260 
2261         assert(kvm_enabled());
2262 
2263         fd = get_htab_fd(spapr);
2264         if (fd < 0) {
2265             return fd;
2266         }
2267 
2268         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2269         if (rc < 0) {
2270             return rc;
2271         }
2272     } else {
2273         if (spapr->htab_first_pass) {
2274             htab_save_first_pass(f, spapr, -1);
2275         }
2276         htab_save_later_pass(f, spapr, -1);
2277     }
2278 
2279     /* End marker */
2280     htab_save_end_marker(f);
2281 
2282     return 0;
2283 }
2284 
2285 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2286 {
2287     SpaprMachineState *spapr = opaque;
2288     uint32_t section_hdr;
2289     int fd = -1;
2290     Error *local_err = NULL;
2291 
2292     if (version_id < 1 || version_id > 1) {
2293         error_report("htab_load() bad version");
2294         return -EINVAL;
2295     }
2296 
2297     section_hdr = qemu_get_be32(f);
2298 
2299     if (section_hdr == -1) {
2300         spapr_free_hpt(spapr);
2301         return 0;
2302     }
2303 
2304     if (section_hdr) {
2305         int ret;
2306 
2307         /* First section gives the htab size */
2308         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2309         if (ret < 0) {
2310             error_report_err(local_err);
2311             return ret;
2312         }
2313         return 0;
2314     }
2315 
2316     if (!spapr->htab) {
2317         assert(kvm_enabled());
2318 
2319         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2320         if (fd < 0) {
2321             error_report_err(local_err);
2322             return fd;
2323         }
2324     }
2325 
2326     while (true) {
2327         uint32_t index;
2328         uint16_t n_valid, n_invalid;
2329 
2330         index = qemu_get_be32(f);
2331         n_valid = qemu_get_be16(f);
2332         n_invalid = qemu_get_be16(f);
2333 
2334         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2335             /* End of Stream */
2336             break;
2337         }
2338 
2339         if ((index + n_valid + n_invalid) >
2340             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2341             /* Bad index in stream */
2342             error_report(
2343                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2344                 index, n_valid, n_invalid, spapr->htab_shift);
2345             return -EINVAL;
2346         }
2347 
2348         if (spapr->htab) {
2349             if (n_valid) {
2350                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2351                                 HASH_PTE_SIZE_64 * n_valid);
2352             }
2353             if (n_invalid) {
2354                 memset(HPTE(spapr->htab, index + n_valid), 0,
2355                        HASH_PTE_SIZE_64 * n_invalid);
2356             }
2357         } else {
2358             int rc;
2359 
2360             assert(fd >= 0);
2361 
2362             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2363                                         &local_err);
2364             if (rc < 0) {
2365                 error_report_err(local_err);
2366                 return rc;
2367             }
2368         }
2369     }
2370 
2371     if (!spapr->htab) {
2372         assert(fd >= 0);
2373         close(fd);
2374     }
2375 
2376     return 0;
2377 }
2378 
2379 static void htab_save_cleanup(void *opaque)
2380 {
2381     SpaprMachineState *spapr = opaque;
2382 
2383     close_htab_fd(spapr);
2384 }
2385 
2386 static SaveVMHandlers savevm_htab_handlers = {
2387     .save_setup = htab_save_setup,
2388     .save_live_iterate = htab_save_iterate,
2389     .save_live_complete_precopy = htab_save_complete,
2390     .save_cleanup = htab_save_cleanup,
2391     .load_state = htab_load,
2392 };
2393 
2394 static void spapr_boot_set(void *opaque, const char *boot_device,
2395                            Error **errp)
2396 {
2397     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2398 
2399     g_free(spapr->boot_device);
2400     spapr->boot_device = g_strdup(boot_device);
2401 }
2402 
2403 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2404 {
2405     MachineState *machine = MACHINE(spapr);
2406     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2407     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2408     int i;
2409 
2410     for (i = 0; i < nr_lmbs; i++) {
2411         uint64_t addr;
2412 
2413         addr = i * lmb_size + machine->device_memory->base;
2414         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2415                                addr / lmb_size);
2416     }
2417 }
2418 
2419 /*
2420  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2421  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2422  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2423  */
2424 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2425 {
2426     int i;
2427 
2428     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2429         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2430                    " is not aligned to %" PRIu64 " MiB",
2431                    machine->ram_size,
2432                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2433         return;
2434     }
2435 
2436     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2437         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2438                    " is not aligned to %" PRIu64 " MiB",
2439                    machine->ram_size,
2440                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2441         return;
2442     }
2443 
2444     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2445         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2446             error_setg(errp,
2447                        "Node %d memory size 0x%" PRIx64
2448                        " is not aligned to %" PRIu64 " MiB",
2449                        i, machine->numa_state->nodes[i].node_mem,
2450                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2451             return;
2452         }
2453     }
2454 }
2455 
2456 /* find cpu slot in machine->possible_cpus by core_id */
2457 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2458 {
2459     int index = id / ms->smp.threads;
2460 
2461     if (index >= ms->possible_cpus->len) {
2462         return NULL;
2463     }
2464     if (idx) {
2465         *idx = index;
2466     }
2467     return &ms->possible_cpus->cpus[index];
2468 }
2469 
2470 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2471 {
2472     MachineState *ms = MACHINE(spapr);
2473     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2474     Error *local_err = NULL;
2475     bool vsmt_user = !!spapr->vsmt;
2476     int kvm_smt = kvmppc_smt_threads();
2477     int ret;
2478     unsigned int smp_threads = ms->smp.threads;
2479 
2480     if (!kvm_enabled() && (smp_threads > 1)) {
2481         error_setg(errp, "TCG cannot support more than 1 thread/core "
2482                    "on a pseries machine");
2483         return;
2484     }
2485     if (!is_power_of_2(smp_threads)) {
2486         error_setg(errp, "Cannot support %d threads/core on a pseries "
2487                    "machine because it must be a power of 2", smp_threads);
2488         return;
2489     }
2490 
2491     /* Detemine the VSMT mode to use: */
2492     if (vsmt_user) {
2493         if (spapr->vsmt < smp_threads) {
2494             error_setg(errp, "Cannot support VSMT mode %d"
2495                        " because it must be >= threads/core (%d)",
2496                        spapr->vsmt, smp_threads);
2497             return;
2498         }
2499         /* In this case, spapr->vsmt has been set by the command line */
2500     } else if (!smc->smp_threads_vsmt) {
2501         /*
2502          * Default VSMT value is tricky, because we need it to be as
2503          * consistent as possible (for migration), but this requires
2504          * changing it for at least some existing cases.  We pick 8 as
2505          * the value that we'd get with KVM on POWER8, the
2506          * overwhelmingly common case in production systems.
2507          */
2508         spapr->vsmt = MAX(8, smp_threads);
2509     } else {
2510         spapr->vsmt = smp_threads;
2511     }
2512 
2513     /* KVM: If necessary, set the SMT mode: */
2514     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2515         ret = kvmppc_set_smt_threads(spapr->vsmt);
2516         if (ret) {
2517             /* Looks like KVM isn't able to change VSMT mode */
2518             error_setg(&local_err,
2519                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2520                        spapr->vsmt, ret);
2521             /* We can live with that if the default one is big enough
2522              * for the number of threads, and a submultiple of the one
2523              * we want.  In this case we'll waste some vcpu ids, but
2524              * behaviour will be correct */
2525             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2526                 warn_report_err(local_err);
2527             } else {
2528                 if (!vsmt_user) {
2529                     error_append_hint(&local_err,
2530                                       "On PPC, a VM with %d threads/core"
2531                                       " on a host with %d threads/core"
2532                                       " requires the use of VSMT mode %d.\n",
2533                                       smp_threads, kvm_smt, spapr->vsmt);
2534                 }
2535                 kvmppc_error_append_smt_possible_hint(&local_err);
2536                 error_propagate(errp, local_err);
2537             }
2538         }
2539     }
2540     /* else TCG: nothing to do currently */
2541 }
2542 
2543 static void spapr_init_cpus(SpaprMachineState *spapr)
2544 {
2545     MachineState *machine = MACHINE(spapr);
2546     MachineClass *mc = MACHINE_GET_CLASS(machine);
2547     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2548     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2549     const CPUArchIdList *possible_cpus;
2550     unsigned int smp_cpus = machine->smp.cpus;
2551     unsigned int smp_threads = machine->smp.threads;
2552     unsigned int max_cpus = machine->smp.max_cpus;
2553     int boot_cores_nr = smp_cpus / smp_threads;
2554     int i;
2555 
2556     possible_cpus = mc->possible_cpu_arch_ids(machine);
2557     if (mc->has_hotpluggable_cpus) {
2558         if (smp_cpus % smp_threads) {
2559             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2560                          smp_cpus, smp_threads);
2561             exit(1);
2562         }
2563         if (max_cpus % smp_threads) {
2564             error_report("max_cpus (%u) must be multiple of threads (%u)",
2565                          max_cpus, smp_threads);
2566             exit(1);
2567         }
2568     } else {
2569         if (max_cpus != smp_cpus) {
2570             error_report("This machine version does not support CPU hotplug");
2571             exit(1);
2572         }
2573         boot_cores_nr = possible_cpus->len;
2574     }
2575 
2576     if (smc->pre_2_10_has_unused_icps) {
2577         int i;
2578 
2579         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2580             /* Dummy entries get deregistered when real ICPState objects
2581              * are registered during CPU core hotplug.
2582              */
2583             pre_2_10_vmstate_register_dummy_icp(i);
2584         }
2585     }
2586 
2587     for (i = 0; i < possible_cpus->len; i++) {
2588         int core_id = i * smp_threads;
2589 
2590         if (mc->has_hotpluggable_cpus) {
2591             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2592                                    spapr_vcpu_id(spapr, core_id));
2593         }
2594 
2595         if (i < boot_cores_nr) {
2596             Object *core  = object_new(type);
2597             int nr_threads = smp_threads;
2598 
2599             /* Handle the partially filled core for older machine types */
2600             if ((i + 1) * smp_threads >= smp_cpus) {
2601                 nr_threads = smp_cpus - i * smp_threads;
2602             }
2603 
2604             object_property_set_int(core, "nr-threads", nr_threads,
2605                                     &error_fatal);
2606             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2607                                     &error_fatal);
2608             qdev_realize(DEVICE(core), NULL, &error_fatal);
2609 
2610             object_unref(core);
2611         }
2612     }
2613 }
2614 
2615 static PCIHostState *spapr_create_default_phb(void)
2616 {
2617     DeviceState *dev;
2618 
2619     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2620     qdev_prop_set_uint32(dev, "index", 0);
2621     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2622 
2623     return PCI_HOST_BRIDGE(dev);
2624 }
2625 
2626 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2627 {
2628     MachineState *machine = MACHINE(spapr);
2629     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2630     hwaddr rma_size = machine->ram_size;
2631     hwaddr node0_size = spapr_node0_size(machine);
2632 
2633     /* RMA has to fit in the first NUMA node */
2634     rma_size = MIN(rma_size, node0_size);
2635 
2636     /*
2637      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2638      * never exceed that
2639      */
2640     rma_size = MIN(rma_size, 1 * TiB);
2641 
2642     /*
2643      * Clamp the RMA size based on machine type.  This is for
2644      * migration compatibility with older qemu versions, which limited
2645      * the RMA size for complicated and mostly bad reasons.
2646      */
2647     if (smc->rma_limit) {
2648         rma_size = MIN(rma_size, smc->rma_limit);
2649     }
2650 
2651     if (rma_size < MIN_RMA_SLOF) {
2652         error_setg(errp,
2653                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2654                    "ldMiB guest RMA (Real Mode Area memory)",
2655                    MIN_RMA_SLOF / MiB);
2656         return 0;
2657     }
2658 
2659     return rma_size;
2660 }
2661 
2662 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2663 {
2664     MachineState *machine = MACHINE(spapr);
2665     int i;
2666 
2667     for (i = 0; i < machine->ram_slots; i++) {
2668         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2669     }
2670 }
2671 
2672 /* pSeries LPAR / sPAPR hardware init */
2673 static void spapr_machine_init(MachineState *machine)
2674 {
2675     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2676     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2677     MachineClass *mc = MACHINE_GET_CLASS(machine);
2678     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2679     const char *bios_name = machine->firmware ?: bios_default;
2680     const char *kernel_filename = machine->kernel_filename;
2681     const char *initrd_filename = machine->initrd_filename;
2682     PCIHostState *phb;
2683     int i;
2684     MemoryRegion *sysmem = get_system_memory();
2685     long load_limit, fw_size;
2686     char *filename;
2687     Error *resize_hpt_err = NULL;
2688 
2689     /*
2690      * if Secure VM (PEF) support is configured, then initialize it
2691      */
2692     pef_kvm_init(machine->cgs, &error_fatal);
2693 
2694     msi_nonbroken = true;
2695 
2696     QLIST_INIT(&spapr->phbs);
2697     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2698 
2699     /* Determine capabilities to run with */
2700     spapr_caps_init(spapr);
2701 
2702     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2703     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2704         /*
2705          * If the user explicitly requested a mode we should either
2706          * supply it, or fail completely (which we do below).  But if
2707          * it's not set explicitly, we reset our mode to something
2708          * that works
2709          */
2710         if (resize_hpt_err) {
2711             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2712             error_free(resize_hpt_err);
2713             resize_hpt_err = NULL;
2714         } else {
2715             spapr->resize_hpt = smc->resize_hpt_default;
2716         }
2717     }
2718 
2719     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2720 
2721     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2722         /*
2723          * User requested HPT resize, but this host can't supply it.  Bail out
2724          */
2725         error_report_err(resize_hpt_err);
2726         exit(1);
2727     }
2728     error_free(resize_hpt_err);
2729 
2730     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2731 
2732     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2733     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2734 
2735     /*
2736      * VSMT must be set in order to be able to compute VCPU ids, ie to
2737      * call spapr_max_server_number() or spapr_vcpu_id().
2738      */
2739     spapr_set_vsmt_mode(spapr, &error_fatal);
2740 
2741     /* Set up Interrupt Controller before we create the VCPUs */
2742     spapr_irq_init(spapr, &error_fatal);
2743 
2744     /* Set up containers for ibm,client-architecture-support negotiated options
2745      */
2746     spapr->ov5 = spapr_ovec_new();
2747     spapr->ov5_cas = spapr_ovec_new();
2748 
2749     if (smc->dr_lmb_enabled) {
2750         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2751         spapr_validate_node_memory(machine, &error_fatal);
2752     }
2753 
2754     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2755 
2756     /* advertise support for dedicated HP event source to guests */
2757     if (spapr->use_hotplug_event_source) {
2758         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2759     }
2760 
2761     /* advertise support for HPT resizing */
2762     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2763         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2764     }
2765 
2766     /* advertise support for ibm,dyamic-memory-v2 */
2767     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2768 
2769     /* advertise XIVE on POWER9 machines */
2770     if (spapr->irq->xive) {
2771         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2772     }
2773 
2774     /* init CPUs */
2775     spapr_init_cpus(spapr);
2776 
2777     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2778 
2779     /* Init numa_assoc_array */
2780     spapr_numa_associativity_init(spapr, machine);
2781 
2782     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2783         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2784                               spapr->max_compat_pvr)) {
2785         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2786         /* KVM and TCG always allow GTSE with radix... */
2787         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2788     }
2789     /* ... but not with hash (currently). */
2790 
2791     if (kvm_enabled()) {
2792         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2793         kvmppc_enable_logical_ci_hcalls();
2794         kvmppc_enable_set_mode_hcall();
2795 
2796         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2797         kvmppc_enable_clear_ref_mod_hcalls();
2798 
2799         /* Enable H_PAGE_INIT */
2800         kvmppc_enable_h_page_init();
2801     }
2802 
2803     /* map RAM */
2804     memory_region_add_subregion(sysmem, 0, machine->ram);
2805 
2806     /* always allocate the device memory information */
2807     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2808 
2809     /* initialize hotplug memory address space */
2810     if (machine->ram_size < machine->maxram_size) {
2811         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2812         /*
2813          * Limit the number of hotpluggable memory slots to half the number
2814          * slots that KVM supports, leaving the other half for PCI and other
2815          * devices. However ensure that number of slots doesn't drop below 32.
2816          */
2817         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2818                            SPAPR_MAX_RAM_SLOTS;
2819 
2820         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2821             max_memslots = SPAPR_MAX_RAM_SLOTS;
2822         }
2823         if (machine->ram_slots > max_memslots) {
2824             error_report("Specified number of memory slots %"
2825                          PRIu64" exceeds max supported %d",
2826                          machine->ram_slots, max_memslots);
2827             exit(1);
2828         }
2829 
2830         machine->device_memory->base = ROUND_UP(machine->ram_size,
2831                                                 SPAPR_DEVICE_MEM_ALIGN);
2832         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2833                            "device-memory", device_mem_size);
2834         memory_region_add_subregion(sysmem, machine->device_memory->base,
2835                                     &machine->device_memory->mr);
2836     }
2837 
2838     if (smc->dr_lmb_enabled) {
2839         spapr_create_lmb_dr_connectors(spapr);
2840     }
2841 
2842     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2843         /* Create the error string for live migration blocker */
2844         error_setg(&spapr->fwnmi_migration_blocker,
2845             "A machine check is being handled during migration. The handler"
2846             "may run and log hardware error on the destination");
2847     }
2848 
2849     if (mc->nvdimm_supported) {
2850         spapr_create_nvdimm_dr_connectors(spapr);
2851     }
2852 
2853     /* Set up RTAS event infrastructure */
2854     spapr_events_init(spapr);
2855 
2856     /* Set up the RTC RTAS interfaces */
2857     spapr_rtc_create(spapr);
2858 
2859     /* Set up VIO bus */
2860     spapr->vio_bus = spapr_vio_bus_init();
2861 
2862     for (i = 0; serial_hd(i); i++) {
2863         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2864     }
2865 
2866     /* We always have at least the nvram device on VIO */
2867     spapr_create_nvram(spapr);
2868 
2869     /*
2870      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2871      * connectors (described in root DT node's "ibm,drc-types" property)
2872      * are pre-initialized here. additional child connectors (such as
2873      * connectors for a PHBs PCI slots) are added as needed during their
2874      * parent's realization.
2875      */
2876     if (smc->dr_phb_enabled) {
2877         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2878             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2879         }
2880     }
2881 
2882     /* Set up PCI */
2883     spapr_pci_rtas_init();
2884 
2885     phb = spapr_create_default_phb();
2886 
2887     for (i = 0; i < nb_nics; i++) {
2888         NICInfo *nd = &nd_table[i];
2889 
2890         if (!nd->model) {
2891             nd->model = g_strdup("spapr-vlan");
2892         }
2893 
2894         if (g_str_equal(nd->model, "spapr-vlan") ||
2895             g_str_equal(nd->model, "ibmveth")) {
2896             spapr_vlan_create(spapr->vio_bus, nd);
2897         } else {
2898             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2899         }
2900     }
2901 
2902     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2903         spapr_vscsi_create(spapr->vio_bus);
2904     }
2905 
2906     /* Graphics */
2907     if (spapr_vga_init(phb->bus, &error_fatal)) {
2908         spapr->has_graphics = true;
2909         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2910     }
2911 
2912     if (machine->usb) {
2913         if (smc->use_ohci_by_default) {
2914             pci_create_simple(phb->bus, -1, "pci-ohci");
2915         } else {
2916             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2917         }
2918 
2919         if (spapr->has_graphics) {
2920             USBBus *usb_bus = usb_bus_find(-1);
2921 
2922             usb_create_simple(usb_bus, "usb-kbd");
2923             usb_create_simple(usb_bus, "usb-mouse");
2924         }
2925     }
2926 
2927     if (kernel_filename) {
2928         spapr->kernel_size = load_elf(kernel_filename, NULL,
2929                                       translate_kernel_address, spapr,
2930                                       NULL, NULL, NULL, NULL, 1,
2931                                       PPC_ELF_MACHINE, 0, 0);
2932         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2933             spapr->kernel_size = load_elf(kernel_filename, NULL,
2934                                           translate_kernel_address, spapr,
2935                                           NULL, NULL, NULL, NULL, 0,
2936                                           PPC_ELF_MACHINE, 0, 0);
2937             spapr->kernel_le = spapr->kernel_size > 0;
2938         }
2939         if (spapr->kernel_size < 0) {
2940             error_report("error loading %s: %s", kernel_filename,
2941                          load_elf_strerror(spapr->kernel_size));
2942             exit(1);
2943         }
2944 
2945         /* load initrd */
2946         if (initrd_filename) {
2947             /* Try to locate the initrd in the gap between the kernel
2948              * and the firmware. Add a bit of space just in case
2949              */
2950             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2951                                   + 0x1ffff) & ~0xffff;
2952             spapr->initrd_size = load_image_targphys(initrd_filename,
2953                                                      spapr->initrd_base,
2954                                                      load_limit
2955                                                      - spapr->initrd_base);
2956             if (spapr->initrd_size < 0) {
2957                 error_report("could not load initial ram disk '%s'",
2958                              initrd_filename);
2959                 exit(1);
2960             }
2961         }
2962     }
2963 
2964     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2965     if (!filename) {
2966         error_report("Could not find LPAR firmware '%s'", bios_name);
2967         exit(1);
2968     }
2969     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2970     if (fw_size <= 0) {
2971         error_report("Could not load LPAR firmware '%s'", filename);
2972         exit(1);
2973     }
2974     g_free(filename);
2975 
2976     /* FIXME: Should register things through the MachineState's qdev
2977      * interface, this is a legacy from the sPAPREnvironment structure
2978      * which predated MachineState but had a similar function */
2979     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2980     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2981                          &savevm_htab_handlers, spapr);
2982 
2983     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2984 
2985     qemu_register_boot_set(spapr_boot_set, spapr);
2986 
2987     /*
2988      * Nothing needs to be done to resume a suspended guest because
2989      * suspending does not change the machine state, so no need for
2990      * a ->wakeup method.
2991      */
2992     qemu_register_wakeup_support();
2993 
2994     if (kvm_enabled()) {
2995         /* to stop and start vmclock */
2996         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2997                                          &spapr->tb);
2998 
2999         kvmppc_spapr_enable_inkernel_multitce();
3000     }
3001 
3002     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3003     if (spapr->vof) {
3004         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3005         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3006     }
3007 }
3008 
3009 #define DEFAULT_KVM_TYPE "auto"
3010 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3011 {
3012     /*
3013      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3014      * accomodate the 'HV' and 'PV' formats that exists in the
3015      * wild. The 'auto' mode is being introduced already as
3016      * lower-case, thus we don't need to bother checking for
3017      * "AUTO".
3018      */
3019     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3020         return 0;
3021     }
3022 
3023     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3024         return 1;
3025     }
3026 
3027     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3028         return 2;
3029     }
3030 
3031     error_report("Unknown kvm-type specified '%s'", vm_type);
3032     exit(1);
3033 }
3034 
3035 /*
3036  * Implementation of an interface to adjust firmware path
3037  * for the bootindex property handling.
3038  */
3039 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3040                                    DeviceState *dev)
3041 {
3042 #define CAST(type, obj, name) \
3043     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3044     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3045     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3046     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3047     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3048 
3049     if (d) {
3050         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3051         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3052         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3053 
3054         if (spapr) {
3055             /*
3056              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3057              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3058              * 0x8000 | (target << 8) | (bus << 5) | lun
3059              * (see the "Logical unit addressing format" table in SAM5)
3060              */
3061             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3062             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3063                                    (uint64_t)id << 48);
3064         } else if (virtio) {
3065             /*
3066              * We use SRP luns of the form 01000000 | (target << 8) | lun
3067              * in the top 32 bits of the 64-bit LUN
3068              * Note: the quote above is from SLOF and it is wrong,
3069              * the actual binding is:
3070              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3071              */
3072             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3073             if (d->lun >= 256) {
3074                 /* Use the LUN "flat space addressing method" */
3075                 id |= 0x4000;
3076             }
3077             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3078                                    (uint64_t)id << 32);
3079         } else if (usb) {
3080             /*
3081              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3082              * in the top 32 bits of the 64-bit LUN
3083              */
3084             unsigned usb_port = atoi(usb->port->path);
3085             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3086             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3087                                    (uint64_t)id << 32);
3088         }
3089     }
3090 
3091     /*
3092      * SLOF probes the USB devices, and if it recognizes that the device is a
3093      * storage device, it changes its name to "storage" instead of "usb-host",
3094      * and additionally adds a child node for the SCSI LUN, so the correct
3095      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3096      */
3097     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3098         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3099         if (usb_device_is_scsi_storage(usbdev)) {
3100             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3101         }
3102     }
3103 
3104     if (phb) {
3105         /* Replace "pci" with "pci@800000020000000" */
3106         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3107     }
3108 
3109     if (vsc) {
3110         /* Same logic as virtio above */
3111         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3112         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3113     }
3114 
3115     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3116         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3117         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3118         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3119     }
3120 
3121     if (pcidev) {
3122         return spapr_pci_fw_dev_name(pcidev);
3123     }
3124 
3125     return NULL;
3126 }
3127 
3128 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3129 {
3130     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3131 
3132     return g_strdup(spapr->kvm_type);
3133 }
3134 
3135 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3136 {
3137     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3138 
3139     g_free(spapr->kvm_type);
3140     spapr->kvm_type = g_strdup(value);
3141 }
3142 
3143 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3144 {
3145     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3146 
3147     return spapr->use_hotplug_event_source;
3148 }
3149 
3150 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3151                                             Error **errp)
3152 {
3153     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3154 
3155     spapr->use_hotplug_event_source = value;
3156 }
3157 
3158 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3159 {
3160     return true;
3161 }
3162 
3163 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3164 {
3165     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3166 
3167     switch (spapr->resize_hpt) {
3168     case SPAPR_RESIZE_HPT_DEFAULT:
3169         return g_strdup("default");
3170     case SPAPR_RESIZE_HPT_DISABLED:
3171         return g_strdup("disabled");
3172     case SPAPR_RESIZE_HPT_ENABLED:
3173         return g_strdup("enabled");
3174     case SPAPR_RESIZE_HPT_REQUIRED:
3175         return g_strdup("required");
3176     }
3177     g_assert_not_reached();
3178 }
3179 
3180 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3181 {
3182     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3183 
3184     if (strcmp(value, "default") == 0) {
3185         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3186     } else if (strcmp(value, "disabled") == 0) {
3187         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3188     } else if (strcmp(value, "enabled") == 0) {
3189         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3190     } else if (strcmp(value, "required") == 0) {
3191         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3192     } else {
3193         error_setg(errp, "Bad value for \"resize-hpt\" property");
3194     }
3195 }
3196 
3197 static bool spapr_get_vof(Object *obj, Error **errp)
3198 {
3199     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3200 
3201     return spapr->vof != NULL;
3202 }
3203 
3204 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3205 {
3206     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3207 
3208     if (spapr->vof) {
3209         vof_cleanup(spapr->vof);
3210         g_free(spapr->vof);
3211         spapr->vof = NULL;
3212     }
3213     if (!value) {
3214         return;
3215     }
3216     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3217 }
3218 
3219 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3220 {
3221     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3222 
3223     if (spapr->irq == &spapr_irq_xics_legacy) {
3224         return g_strdup("legacy");
3225     } else if (spapr->irq == &spapr_irq_xics) {
3226         return g_strdup("xics");
3227     } else if (spapr->irq == &spapr_irq_xive) {
3228         return g_strdup("xive");
3229     } else if (spapr->irq == &spapr_irq_dual) {
3230         return g_strdup("dual");
3231     }
3232     g_assert_not_reached();
3233 }
3234 
3235 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3236 {
3237     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3238 
3239     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3240         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3241         return;
3242     }
3243 
3244     /* The legacy IRQ backend can not be set */
3245     if (strcmp(value, "xics") == 0) {
3246         spapr->irq = &spapr_irq_xics;
3247     } else if (strcmp(value, "xive") == 0) {
3248         spapr->irq = &spapr_irq_xive;
3249     } else if (strcmp(value, "dual") == 0) {
3250         spapr->irq = &spapr_irq_dual;
3251     } else {
3252         error_setg(errp, "Bad value for \"ic-mode\" property");
3253     }
3254 }
3255 
3256 static char *spapr_get_host_model(Object *obj, Error **errp)
3257 {
3258     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3259 
3260     return g_strdup(spapr->host_model);
3261 }
3262 
3263 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3264 {
3265     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3266 
3267     g_free(spapr->host_model);
3268     spapr->host_model = g_strdup(value);
3269 }
3270 
3271 static char *spapr_get_host_serial(Object *obj, Error **errp)
3272 {
3273     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3274 
3275     return g_strdup(spapr->host_serial);
3276 }
3277 
3278 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3279 {
3280     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3281 
3282     g_free(spapr->host_serial);
3283     spapr->host_serial = g_strdup(value);
3284 }
3285 
3286 static void spapr_instance_init(Object *obj)
3287 {
3288     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3289     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3290     MachineState *ms = MACHINE(spapr);
3291     MachineClass *mc = MACHINE_GET_CLASS(ms);
3292 
3293     /*
3294      * NVDIMM support went live in 5.1 without considering that, in
3295      * other archs, the user needs to enable NVDIMM support with the
3296      * 'nvdimm' machine option and the default behavior is NVDIMM
3297      * support disabled. It is too late to roll back to the standard
3298      * behavior without breaking 5.1 guests.
3299      */
3300     if (mc->nvdimm_supported) {
3301         ms->nvdimms_state->is_enabled = true;
3302     }
3303 
3304     spapr->htab_fd = -1;
3305     spapr->use_hotplug_event_source = true;
3306     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3307     object_property_add_str(obj, "kvm-type",
3308                             spapr_get_kvm_type, spapr_set_kvm_type);
3309     object_property_set_description(obj, "kvm-type",
3310                                     "Specifies the KVM virtualization mode (auto,"
3311                                     " hv, pr). Defaults to 'auto'. This mode will use"
3312                                     " any available KVM module loaded in the host,"
3313                                     " where kvm_hv takes precedence if both kvm_hv and"
3314                                     " kvm_pr are loaded.");
3315     object_property_add_bool(obj, "modern-hotplug-events",
3316                             spapr_get_modern_hotplug_events,
3317                             spapr_set_modern_hotplug_events);
3318     object_property_set_description(obj, "modern-hotplug-events",
3319                                     "Use dedicated hotplug event mechanism in"
3320                                     " place of standard EPOW events when possible"
3321                                     " (required for memory hot-unplug support)");
3322     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3323                             "Maximum permitted CPU compatibility mode");
3324 
3325     object_property_add_str(obj, "resize-hpt",
3326                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3327     object_property_set_description(obj, "resize-hpt",
3328                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3329     object_property_add_uint32_ptr(obj, "vsmt",
3330                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3331     object_property_set_description(obj, "vsmt",
3332                                     "Virtual SMT: KVM behaves as if this were"
3333                                     " the host's SMT mode");
3334 
3335     object_property_add_bool(obj, "vfio-no-msix-emulation",
3336                              spapr_get_msix_emulation, NULL);
3337 
3338     object_property_add_uint64_ptr(obj, "kernel-addr",
3339                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3340     object_property_set_description(obj, "kernel-addr",
3341                                     stringify(KERNEL_LOAD_ADDR)
3342                                     " for -kernel is the default");
3343     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3344 
3345     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3346     object_property_set_description(obj, "x-vof",
3347                                     "Enable Virtual Open Firmware (experimental)");
3348 
3349     /* The machine class defines the default interrupt controller mode */
3350     spapr->irq = smc->irq;
3351     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3352                             spapr_set_ic_mode);
3353     object_property_set_description(obj, "ic-mode",
3354                  "Specifies the interrupt controller mode (xics, xive, dual)");
3355 
3356     object_property_add_str(obj, "host-model",
3357         spapr_get_host_model, spapr_set_host_model);
3358     object_property_set_description(obj, "host-model",
3359         "Host model to advertise in guest device tree");
3360     object_property_add_str(obj, "host-serial",
3361         spapr_get_host_serial, spapr_set_host_serial);
3362     object_property_set_description(obj, "host-serial",
3363         "Host serial number to advertise in guest device tree");
3364 }
3365 
3366 static void spapr_machine_finalizefn(Object *obj)
3367 {
3368     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3369 
3370     g_free(spapr->kvm_type);
3371 }
3372 
3373 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3374 {
3375     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3376     PowerPCCPU *cpu = POWERPC_CPU(cs);
3377     CPUPPCState *env = &cpu->env;
3378 
3379     cpu_synchronize_state(cs);
3380     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3381     if (spapr->fwnmi_system_reset_addr != -1) {
3382         uint64_t rtas_addr, addr;
3383 
3384         /* get rtas addr from fdt */
3385         rtas_addr = spapr_get_rtas_addr();
3386         if (!rtas_addr) {
3387             qemu_system_guest_panicked(NULL);
3388             return;
3389         }
3390 
3391         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3392         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3393         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3394         env->gpr[3] = addr;
3395     }
3396     ppc_cpu_do_system_reset(cs);
3397     if (spapr->fwnmi_system_reset_addr != -1) {
3398         env->nip = spapr->fwnmi_system_reset_addr;
3399     }
3400 }
3401 
3402 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3403 {
3404     CPUState *cs;
3405 
3406     CPU_FOREACH(cs) {
3407         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3408     }
3409 }
3410 
3411 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3412                           void *fdt, int *fdt_start_offset, Error **errp)
3413 {
3414     uint64_t addr;
3415     uint32_t node;
3416 
3417     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3418     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3419                                     &error_abort);
3420     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3421                                              SPAPR_MEMORY_BLOCK_SIZE);
3422     return 0;
3423 }
3424 
3425 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3426                            bool dedicated_hp_event_source)
3427 {
3428     SpaprDrc *drc;
3429     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3430     int i;
3431     uint64_t addr = addr_start;
3432     bool hotplugged = spapr_drc_hotplugged(dev);
3433 
3434     for (i = 0; i < nr_lmbs; i++) {
3435         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3436                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3437         g_assert(drc);
3438 
3439         /*
3440          * memory_device_get_free_addr() provided a range of free addresses
3441          * that doesn't overlap with any existing mapping at pre-plug. The
3442          * corresponding LMB DRCs are thus assumed to be all attachable.
3443          */
3444         spapr_drc_attach(drc, dev);
3445         if (!hotplugged) {
3446             spapr_drc_reset(drc);
3447         }
3448         addr += SPAPR_MEMORY_BLOCK_SIZE;
3449     }
3450     /* send hotplug notification to the
3451      * guest only in case of hotplugged memory
3452      */
3453     if (hotplugged) {
3454         if (dedicated_hp_event_source) {
3455             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3456                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3457             g_assert(drc);
3458             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3459                                                    nr_lmbs,
3460                                                    spapr_drc_index(drc));
3461         } else {
3462             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3463                                            nr_lmbs);
3464         }
3465     }
3466 }
3467 
3468 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3469 {
3470     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3471     PCDIMMDevice *dimm = PC_DIMM(dev);
3472     uint64_t size, addr;
3473     int64_t slot;
3474     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3475 
3476     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3477 
3478     pc_dimm_plug(dimm, MACHINE(ms));
3479 
3480     if (!is_nvdimm) {
3481         addr = object_property_get_uint(OBJECT(dimm),
3482                                         PC_DIMM_ADDR_PROP, &error_abort);
3483         spapr_add_lmbs(dev, addr, size,
3484                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3485     } else {
3486         slot = object_property_get_int(OBJECT(dimm),
3487                                        PC_DIMM_SLOT_PROP, &error_abort);
3488         /* We should have valid slot number at this point */
3489         g_assert(slot >= 0);
3490         spapr_add_nvdimm(dev, slot);
3491     }
3492 }
3493 
3494 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3495                                   Error **errp)
3496 {
3497     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3498     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3499     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3500     PCDIMMDevice *dimm = PC_DIMM(dev);
3501     Error *local_err = NULL;
3502     uint64_t size;
3503     Object *memdev;
3504     hwaddr pagesize;
3505 
3506     if (!smc->dr_lmb_enabled) {
3507         error_setg(errp, "Memory hotplug not supported for this machine");
3508         return;
3509     }
3510 
3511     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3512     if (local_err) {
3513         error_propagate(errp, local_err);
3514         return;
3515     }
3516 
3517     if (is_nvdimm) {
3518         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3519             return;
3520         }
3521     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3522         error_setg(errp, "Hotplugged memory size must be a multiple of "
3523                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3524         return;
3525     }
3526 
3527     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3528                                       &error_abort);
3529     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3530     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3531         return;
3532     }
3533 
3534     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3535 }
3536 
3537 struct SpaprDimmState {
3538     PCDIMMDevice *dimm;
3539     uint32_t nr_lmbs;
3540     QTAILQ_ENTRY(SpaprDimmState) next;
3541 };
3542 
3543 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3544                                                        PCDIMMDevice *dimm)
3545 {
3546     SpaprDimmState *dimm_state = NULL;
3547 
3548     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3549         if (dimm_state->dimm == dimm) {
3550             break;
3551         }
3552     }
3553     return dimm_state;
3554 }
3555 
3556 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3557                                                       uint32_t nr_lmbs,
3558                                                       PCDIMMDevice *dimm)
3559 {
3560     SpaprDimmState *ds = NULL;
3561 
3562     /*
3563      * If this request is for a DIMM whose removal had failed earlier
3564      * (due to guest's refusal to remove the LMBs), we would have this
3565      * dimm already in the pending_dimm_unplugs list. In that
3566      * case don't add again.
3567      */
3568     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3569     if (!ds) {
3570         ds = g_malloc0(sizeof(SpaprDimmState));
3571         ds->nr_lmbs = nr_lmbs;
3572         ds->dimm = dimm;
3573         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3574     }
3575     return ds;
3576 }
3577 
3578 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3579                                               SpaprDimmState *dimm_state)
3580 {
3581     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3582     g_free(dimm_state);
3583 }
3584 
3585 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3586                                                         PCDIMMDevice *dimm)
3587 {
3588     SpaprDrc *drc;
3589     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3590                                                   &error_abort);
3591     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3592     uint32_t avail_lmbs = 0;
3593     uint64_t addr_start, addr;
3594     int i;
3595 
3596     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3597                                           &error_abort);
3598 
3599     addr = addr_start;
3600     for (i = 0; i < nr_lmbs; i++) {
3601         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3602                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3603         g_assert(drc);
3604         if (drc->dev) {
3605             avail_lmbs++;
3606         }
3607         addr += SPAPR_MEMORY_BLOCK_SIZE;
3608     }
3609 
3610     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3611 }
3612 
3613 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3614 {
3615     SpaprDimmState *ds;
3616     PCDIMMDevice *dimm;
3617     SpaprDrc *drc;
3618     uint32_t nr_lmbs;
3619     uint64_t size, addr_start, addr;
3620     g_autofree char *qapi_error = NULL;
3621     int i;
3622 
3623     if (!dev) {
3624         return;
3625     }
3626 
3627     dimm = PC_DIMM(dev);
3628     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3629 
3630     /*
3631      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3632      * unplug state, but one of its DRC is marked as unplug_requested.
3633      * This is bad and weird enough to g_assert() out.
3634      */
3635     g_assert(ds);
3636 
3637     spapr_pending_dimm_unplugs_remove(spapr, ds);
3638 
3639     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3640     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3641 
3642     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3643                                           &error_abort);
3644 
3645     addr = addr_start;
3646     for (i = 0; i < nr_lmbs; i++) {
3647         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3648                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3649         g_assert(drc);
3650 
3651         drc->unplug_requested = false;
3652         addr += SPAPR_MEMORY_BLOCK_SIZE;
3653     }
3654 
3655     /*
3656      * Tell QAPI that something happened and the memory
3657      * hotunplug wasn't successful. Keep sending
3658      * MEM_UNPLUG_ERROR even while sending
3659      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3660      * MEM_UNPLUG_ERROR is due.
3661      */
3662     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3663                                  "for device %s", dev->id);
3664 
3665     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3666 
3667     qapi_event_send_device_unplug_guest_error(!!dev->id, dev->id,
3668                                               dev->canonical_path);
3669 }
3670 
3671 /* Callback to be called during DRC release. */
3672 void spapr_lmb_release(DeviceState *dev)
3673 {
3674     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3675     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3676     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3677 
3678     /* This information will get lost if a migration occurs
3679      * during the unplug process. In this case recover it. */
3680     if (ds == NULL) {
3681         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3682         g_assert(ds);
3683         /* The DRC being examined by the caller at least must be counted */
3684         g_assert(ds->nr_lmbs);
3685     }
3686 
3687     if (--ds->nr_lmbs) {
3688         return;
3689     }
3690 
3691     /*
3692      * Now that all the LMBs have been removed by the guest, call the
3693      * unplug handler chain. This can never fail.
3694      */
3695     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3696     object_unparent(OBJECT(dev));
3697 }
3698 
3699 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3700 {
3701     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3702     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3703 
3704     /* We really shouldn't get this far without anything to unplug */
3705     g_assert(ds);
3706 
3707     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3708     qdev_unrealize(dev);
3709     spapr_pending_dimm_unplugs_remove(spapr, ds);
3710 }
3711 
3712 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3713                                         DeviceState *dev, Error **errp)
3714 {
3715     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3716     PCDIMMDevice *dimm = PC_DIMM(dev);
3717     uint32_t nr_lmbs;
3718     uint64_t size, addr_start, addr;
3719     int i;
3720     SpaprDrc *drc;
3721 
3722     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3723         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3724         return;
3725     }
3726 
3727     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3728     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3729 
3730     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3731                                           &error_abort);
3732 
3733     /*
3734      * An existing pending dimm state for this DIMM means that there is an
3735      * unplug operation in progress, waiting for the spapr_lmb_release
3736      * callback to complete the job (BQL can't cover that far). In this case,
3737      * bail out to avoid detaching DRCs that were already released.
3738      */
3739     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3740         error_setg(errp, "Memory unplug already in progress for device %s",
3741                    dev->id);
3742         return;
3743     }
3744 
3745     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3746 
3747     addr = addr_start;
3748     for (i = 0; i < nr_lmbs; i++) {
3749         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3750                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3751         g_assert(drc);
3752 
3753         spapr_drc_unplug_request(drc);
3754         addr += SPAPR_MEMORY_BLOCK_SIZE;
3755     }
3756 
3757     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3758                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3759     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3760                                               nr_lmbs, spapr_drc_index(drc));
3761 }
3762 
3763 /* Callback to be called during DRC release. */
3764 void spapr_core_release(DeviceState *dev)
3765 {
3766     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3767 
3768     /* Call the unplug handler chain. This can never fail. */
3769     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3770     object_unparent(OBJECT(dev));
3771 }
3772 
3773 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3774 {
3775     MachineState *ms = MACHINE(hotplug_dev);
3776     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3777     CPUCore *cc = CPU_CORE(dev);
3778     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3779 
3780     if (smc->pre_2_10_has_unused_icps) {
3781         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3782         int i;
3783 
3784         for (i = 0; i < cc->nr_threads; i++) {
3785             CPUState *cs = CPU(sc->threads[i]);
3786 
3787             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3788         }
3789     }
3790 
3791     assert(core_slot);
3792     core_slot->cpu = NULL;
3793     qdev_unrealize(dev);
3794 }
3795 
3796 static
3797 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3798                                Error **errp)
3799 {
3800     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3801     int index;
3802     SpaprDrc *drc;
3803     CPUCore *cc = CPU_CORE(dev);
3804 
3805     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3806         error_setg(errp, "Unable to find CPU core with core-id: %d",
3807                    cc->core_id);
3808         return;
3809     }
3810     if (index == 0) {
3811         error_setg(errp, "Boot CPU core may not be unplugged");
3812         return;
3813     }
3814 
3815     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3816                           spapr_vcpu_id(spapr, cc->core_id));
3817     g_assert(drc);
3818 
3819     if (!spapr_drc_unplug_requested(drc)) {
3820         spapr_drc_unplug_request(drc);
3821     }
3822 
3823     /*
3824      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3825      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3826      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3827      * attempt (e.g. the kernel will refuse to remove the last online
3828      * CPU), we will never attempt it again because unplug_requested
3829      * will still be 'true' in that case.
3830      */
3831     spapr_hotplug_req_remove_by_index(drc);
3832 }
3833 
3834 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3835                            void *fdt, int *fdt_start_offset, Error **errp)
3836 {
3837     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3838     CPUState *cs = CPU(core->threads[0]);
3839     PowerPCCPU *cpu = POWERPC_CPU(cs);
3840     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3841     int id = spapr_get_vcpu_id(cpu);
3842     g_autofree char *nodename = NULL;
3843     int offset;
3844 
3845     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3846     offset = fdt_add_subnode(fdt, 0, nodename);
3847 
3848     spapr_dt_cpu(cs, fdt, offset, spapr);
3849 
3850     /*
3851      * spapr_dt_cpu() does not fill the 'name' property in the
3852      * CPU node. The function is called during boot process, before
3853      * and after CAS, and overwriting the 'name' property written
3854      * by SLOF is not allowed.
3855      *
3856      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3857      * CPUs more compatible with the coldplugged ones, which have
3858      * the 'name' property. Linux Kernel also relies on this
3859      * property to identify CPU nodes.
3860      */
3861     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3862 
3863     *fdt_start_offset = offset;
3864     return 0;
3865 }
3866 
3867 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3868 {
3869     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3870     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3871     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3872     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3873     CPUCore *cc = CPU_CORE(dev);
3874     CPUState *cs;
3875     SpaprDrc *drc;
3876     CPUArchId *core_slot;
3877     int index;
3878     bool hotplugged = spapr_drc_hotplugged(dev);
3879     int i;
3880 
3881     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3882     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3883 
3884     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3885                           spapr_vcpu_id(spapr, cc->core_id));
3886 
3887     g_assert(drc || !mc->has_hotpluggable_cpus);
3888 
3889     if (drc) {
3890         /*
3891          * spapr_core_pre_plug() already buys us this is a brand new
3892          * core being plugged into a free slot. Nothing should already
3893          * be attached to the corresponding DRC.
3894          */
3895         spapr_drc_attach(drc, dev);
3896 
3897         if (hotplugged) {
3898             /*
3899              * Send hotplug notification interrupt to the guest only
3900              * in case of hotplugged CPUs.
3901              */
3902             spapr_hotplug_req_add_by_index(drc);
3903         } else {
3904             spapr_drc_reset(drc);
3905         }
3906     }
3907 
3908     core_slot->cpu = OBJECT(dev);
3909 
3910     /*
3911      * Set compatibility mode to match the boot CPU, which was either set
3912      * by the machine reset code or by CAS. This really shouldn't fail at
3913      * this point.
3914      */
3915     if (hotplugged) {
3916         for (i = 0; i < cc->nr_threads; i++) {
3917             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3918                            &error_abort);
3919         }
3920     }
3921 
3922     if (smc->pre_2_10_has_unused_icps) {
3923         for (i = 0; i < cc->nr_threads; i++) {
3924             cs = CPU(core->threads[i]);
3925             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3926         }
3927     }
3928 }
3929 
3930 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3931                                 Error **errp)
3932 {
3933     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3934     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3935     CPUCore *cc = CPU_CORE(dev);
3936     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3937     const char *type = object_get_typename(OBJECT(dev));
3938     CPUArchId *core_slot;
3939     int index;
3940     unsigned int smp_threads = machine->smp.threads;
3941 
3942     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3943         error_setg(errp, "CPU hotplug not supported for this machine");
3944         return;
3945     }
3946 
3947     if (strcmp(base_core_type, type)) {
3948         error_setg(errp, "CPU core type should be %s", base_core_type);
3949         return;
3950     }
3951 
3952     if (cc->core_id % smp_threads) {
3953         error_setg(errp, "invalid core id %d", cc->core_id);
3954         return;
3955     }
3956 
3957     /*
3958      * In general we should have homogeneous threads-per-core, but old
3959      * (pre hotplug support) machine types allow the last core to have
3960      * reduced threads as a compatibility hack for when we allowed
3961      * total vcpus not a multiple of threads-per-core.
3962      */
3963     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3964         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3965                    smp_threads);
3966         return;
3967     }
3968 
3969     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3970     if (!core_slot) {
3971         error_setg(errp, "core id %d out of range", cc->core_id);
3972         return;
3973     }
3974 
3975     if (core_slot->cpu) {
3976         error_setg(errp, "core %d already populated", cc->core_id);
3977         return;
3978     }
3979 
3980     numa_cpu_pre_plug(core_slot, dev, errp);
3981 }
3982 
3983 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3984                           void *fdt, int *fdt_start_offset, Error **errp)
3985 {
3986     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3987     int intc_phandle;
3988 
3989     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3990     if (intc_phandle <= 0) {
3991         return -1;
3992     }
3993 
3994     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3995         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3996         return -1;
3997     }
3998 
3999     /* generally SLOF creates these, for hotplug it's up to QEMU */
4000     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4001 
4002     return 0;
4003 }
4004 
4005 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4006                                Error **errp)
4007 {
4008     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4009     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4010     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4011     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4012     SpaprDrc *drc;
4013 
4014     if (dev->hotplugged && !smc->dr_phb_enabled) {
4015         error_setg(errp, "PHB hotplug not supported for this machine");
4016         return false;
4017     }
4018 
4019     if (sphb->index == (uint32_t)-1) {
4020         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4021         return false;
4022     }
4023 
4024     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4025     if (drc && drc->dev) {
4026         error_setg(errp, "PHB %d already attached", sphb->index);
4027         return false;
4028     }
4029 
4030     /*
4031      * This will check that sphb->index doesn't exceed the maximum number of
4032      * PHBs for the current machine type.
4033      */
4034     return
4035         smc->phb_placement(spapr, sphb->index,
4036                            &sphb->buid, &sphb->io_win_addr,
4037                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4038                            windows_supported, sphb->dma_liobn,
4039                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4040                            errp);
4041 }
4042 
4043 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4044 {
4045     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4046     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4047     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4048     SpaprDrc *drc;
4049     bool hotplugged = spapr_drc_hotplugged(dev);
4050 
4051     if (!smc->dr_phb_enabled) {
4052         return;
4053     }
4054 
4055     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4056     /* hotplug hooks should check it's enabled before getting this far */
4057     assert(drc);
4058 
4059     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4060     spapr_drc_attach(drc, dev);
4061 
4062     if (hotplugged) {
4063         spapr_hotplug_req_add_by_index(drc);
4064     } else {
4065         spapr_drc_reset(drc);
4066     }
4067 }
4068 
4069 void spapr_phb_release(DeviceState *dev)
4070 {
4071     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4072 
4073     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4074     object_unparent(OBJECT(dev));
4075 }
4076 
4077 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4078 {
4079     qdev_unrealize(dev);
4080 }
4081 
4082 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4083                                      DeviceState *dev, Error **errp)
4084 {
4085     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4086     SpaprDrc *drc;
4087 
4088     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4089     assert(drc);
4090 
4091     if (!spapr_drc_unplug_requested(drc)) {
4092         spapr_drc_unplug_request(drc);
4093         spapr_hotplug_req_remove_by_index(drc);
4094     } else {
4095         error_setg(errp,
4096                    "PCI Host Bridge unplug already in progress for device %s",
4097                    dev->id);
4098     }
4099 }
4100 
4101 static
4102 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4103                               Error **errp)
4104 {
4105     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4106 
4107     if (spapr->tpm_proxy != NULL) {
4108         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4109         return false;
4110     }
4111 
4112     return true;
4113 }
4114 
4115 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4116 {
4117     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4118     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4119 
4120     /* Already checked in spapr_tpm_proxy_pre_plug() */
4121     g_assert(spapr->tpm_proxy == NULL);
4122 
4123     spapr->tpm_proxy = tpm_proxy;
4124 }
4125 
4126 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4127 {
4128     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4129 
4130     qdev_unrealize(dev);
4131     object_unparent(OBJECT(dev));
4132     spapr->tpm_proxy = NULL;
4133 }
4134 
4135 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4136                                       DeviceState *dev, Error **errp)
4137 {
4138     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4139         spapr_memory_plug(hotplug_dev, dev);
4140     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4141         spapr_core_plug(hotplug_dev, dev);
4142     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4143         spapr_phb_plug(hotplug_dev, dev);
4144     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4145         spapr_tpm_proxy_plug(hotplug_dev, dev);
4146     }
4147 }
4148 
4149 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4150                                         DeviceState *dev, Error **errp)
4151 {
4152     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4153         spapr_memory_unplug(hotplug_dev, dev);
4154     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4155         spapr_core_unplug(hotplug_dev, dev);
4156     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4157         spapr_phb_unplug(hotplug_dev, dev);
4158     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4159         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4160     }
4161 }
4162 
4163 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4164 {
4165     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4166         /*
4167          * CAS will process all pending unplug requests.
4168          *
4169          * HACK: a guest could theoretically have cleared all bits in OV5,
4170          * but none of the guests we care for do.
4171          */
4172         spapr_ovec_empty(spapr->ov5_cas);
4173 }
4174 
4175 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4176                                                 DeviceState *dev, Error **errp)
4177 {
4178     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4179     MachineClass *mc = MACHINE_GET_CLASS(sms);
4180     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4181 
4182     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4183         if (spapr_memory_hot_unplug_supported(sms)) {
4184             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4185         } else {
4186             error_setg(errp, "Memory hot unplug not supported for this guest");
4187         }
4188     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4189         if (!mc->has_hotpluggable_cpus) {
4190             error_setg(errp, "CPU hot unplug not supported on this machine");
4191             return;
4192         }
4193         spapr_core_unplug_request(hotplug_dev, dev, errp);
4194     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4195         if (!smc->dr_phb_enabled) {
4196             error_setg(errp, "PHB hot unplug not supported on this machine");
4197             return;
4198         }
4199         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4200     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4201         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4202     }
4203 }
4204 
4205 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4206                                           DeviceState *dev, Error **errp)
4207 {
4208     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4209         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4210     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4211         spapr_core_pre_plug(hotplug_dev, dev, errp);
4212     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4213         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4214     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4215         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4216     }
4217 }
4218 
4219 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4220                                                  DeviceState *dev)
4221 {
4222     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4223         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4224         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4225         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4226         return HOTPLUG_HANDLER(machine);
4227     }
4228     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4229         PCIDevice *pcidev = PCI_DEVICE(dev);
4230         PCIBus *root = pci_device_root_bus(pcidev);
4231         SpaprPhbState *phb =
4232             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4233                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4234 
4235         if (phb) {
4236             return HOTPLUG_HANDLER(phb);
4237         }
4238     }
4239     return NULL;
4240 }
4241 
4242 static CpuInstanceProperties
4243 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4244 {
4245     CPUArchId *core_slot;
4246     MachineClass *mc = MACHINE_GET_CLASS(machine);
4247 
4248     /* make sure possible_cpu are intialized */
4249     mc->possible_cpu_arch_ids(machine);
4250     /* get CPU core slot containing thread that matches cpu_index */
4251     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4252     assert(core_slot);
4253     return core_slot->props;
4254 }
4255 
4256 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4257 {
4258     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4259 }
4260 
4261 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4262 {
4263     int i;
4264     unsigned int smp_threads = machine->smp.threads;
4265     unsigned int smp_cpus = machine->smp.cpus;
4266     const char *core_type;
4267     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4268     MachineClass *mc = MACHINE_GET_CLASS(machine);
4269 
4270     if (!mc->has_hotpluggable_cpus) {
4271         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4272     }
4273     if (machine->possible_cpus) {
4274         assert(machine->possible_cpus->len == spapr_max_cores);
4275         return machine->possible_cpus;
4276     }
4277 
4278     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4279     if (!core_type) {
4280         error_report("Unable to find sPAPR CPU Core definition");
4281         exit(1);
4282     }
4283 
4284     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4285                              sizeof(CPUArchId) * spapr_max_cores);
4286     machine->possible_cpus->len = spapr_max_cores;
4287     for (i = 0; i < machine->possible_cpus->len; i++) {
4288         int core_id = i * smp_threads;
4289 
4290         machine->possible_cpus->cpus[i].type = core_type;
4291         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4292         machine->possible_cpus->cpus[i].arch_id = core_id;
4293         machine->possible_cpus->cpus[i].props.has_core_id = true;
4294         machine->possible_cpus->cpus[i].props.core_id = core_id;
4295     }
4296     return machine->possible_cpus;
4297 }
4298 
4299 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4300                                 uint64_t *buid, hwaddr *pio,
4301                                 hwaddr *mmio32, hwaddr *mmio64,
4302                                 unsigned n_dma, uint32_t *liobns,
4303                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4304 {
4305     /*
4306      * New-style PHB window placement.
4307      *
4308      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4309      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4310      * windows.
4311      *
4312      * Some guest kernels can't work with MMIO windows above 1<<46
4313      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4314      *
4315      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4316      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4317      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4318      * 1TiB 64-bit MMIO windows for each PHB.
4319      */
4320     const uint64_t base_buid = 0x800000020000000ULL;
4321     int i;
4322 
4323     /* Sanity check natural alignments */
4324     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4325     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4326     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4327     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4328     /* Sanity check bounds */
4329     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4330                       SPAPR_PCI_MEM32_WIN_SIZE);
4331     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4332                       SPAPR_PCI_MEM64_WIN_SIZE);
4333 
4334     if (index >= SPAPR_MAX_PHBS) {
4335         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4336                    SPAPR_MAX_PHBS - 1);
4337         return false;
4338     }
4339 
4340     *buid = base_buid + index;
4341     for (i = 0; i < n_dma; ++i) {
4342         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4343     }
4344 
4345     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4346     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4347     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4348 
4349     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4350     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4351     return true;
4352 }
4353 
4354 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4355 {
4356     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4357 
4358     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4359 }
4360 
4361 static void spapr_ics_resend(XICSFabric *dev)
4362 {
4363     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4364 
4365     ics_resend(spapr->ics);
4366 }
4367 
4368 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4369 {
4370     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4371 
4372     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4373 }
4374 
4375 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4376                                  Monitor *mon)
4377 {
4378     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4379 
4380     spapr_irq_print_info(spapr, mon);
4381     monitor_printf(mon, "irqchip: %s\n",
4382                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4383 }
4384 
4385 /*
4386  * This is a XIVE only operation
4387  */
4388 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4389                            uint8_t nvt_blk, uint32_t nvt_idx,
4390                            bool cam_ignore, uint8_t priority,
4391                            uint32_t logic_serv, XiveTCTXMatch *match)
4392 {
4393     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4394     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4395     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4396     int count;
4397 
4398     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4399                            priority, logic_serv, match);
4400     if (count < 0) {
4401         return count;
4402     }
4403 
4404     /*
4405      * When we implement the save and restore of the thread interrupt
4406      * contexts in the enter/exit CPU handlers of the machine and the
4407      * escalations in QEMU, we should be able to handle non dispatched
4408      * vCPUs.
4409      *
4410      * Until this is done, the sPAPR machine should find at least one
4411      * matching context always.
4412      */
4413     if (count == 0) {
4414         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4415                       nvt_blk, nvt_idx);
4416     }
4417 
4418     return count;
4419 }
4420 
4421 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4422 {
4423     return cpu->vcpu_id;
4424 }
4425 
4426 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4427 {
4428     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4429     MachineState *ms = MACHINE(spapr);
4430     int vcpu_id;
4431 
4432     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4433 
4434     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4435         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4436         error_append_hint(errp, "Adjust the number of cpus to %d "
4437                           "or try to raise the number of threads per core\n",
4438                           vcpu_id * ms->smp.threads / spapr->vsmt);
4439         return false;
4440     }
4441 
4442     cpu->vcpu_id = vcpu_id;
4443     return true;
4444 }
4445 
4446 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4447 {
4448     CPUState *cs;
4449 
4450     CPU_FOREACH(cs) {
4451         PowerPCCPU *cpu = POWERPC_CPU(cs);
4452 
4453         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4454             return cpu;
4455         }
4456     }
4457 
4458     return NULL;
4459 }
4460 
4461 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4462 {
4463     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4464 
4465     /* These are only called by TCG, KVM maintains dispatch state */
4466 
4467     spapr_cpu->prod = false;
4468     if (spapr_cpu->vpa_addr) {
4469         CPUState *cs = CPU(cpu);
4470         uint32_t dispatch;
4471 
4472         dispatch = ldl_be_phys(cs->as,
4473                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4474         dispatch++;
4475         if ((dispatch & 1) != 0) {
4476             qemu_log_mask(LOG_GUEST_ERROR,
4477                           "VPA: incorrect dispatch counter value for "
4478                           "dispatched partition %u, correcting.\n", dispatch);
4479             dispatch++;
4480         }
4481         stl_be_phys(cs->as,
4482                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4483     }
4484 }
4485 
4486 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4487 {
4488     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4489 
4490     if (spapr_cpu->vpa_addr) {
4491         CPUState *cs = CPU(cpu);
4492         uint32_t dispatch;
4493 
4494         dispatch = ldl_be_phys(cs->as,
4495                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4496         dispatch++;
4497         if ((dispatch & 1) != 1) {
4498             qemu_log_mask(LOG_GUEST_ERROR,
4499                           "VPA: incorrect dispatch counter value for "
4500                           "preempted partition %u, correcting.\n", dispatch);
4501             dispatch++;
4502         }
4503         stl_be_phys(cs->as,
4504                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4505     }
4506 }
4507 
4508 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4509 {
4510     MachineClass *mc = MACHINE_CLASS(oc);
4511     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4512     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4513     NMIClass *nc = NMI_CLASS(oc);
4514     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4515     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4516     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4517     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4518     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4519     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4520 
4521     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4522     mc->ignore_boot_device_suffixes = true;
4523 
4524     /*
4525      * We set up the default / latest behaviour here.  The class_init
4526      * functions for the specific versioned machine types can override
4527      * these details for backwards compatibility
4528      */
4529     mc->init = spapr_machine_init;
4530     mc->reset = spapr_machine_reset;
4531     mc->block_default_type = IF_SCSI;
4532 
4533     /*
4534      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4535      * should be limited by the host capability instead of hardcoded.
4536      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4537      * guests are welcome to have as many CPUs as the host are capable
4538      * of emulate.
4539      */
4540     mc->max_cpus = INT32_MAX;
4541 
4542     mc->no_parallel = 1;
4543     mc->default_boot_order = "";
4544     mc->default_ram_size = 512 * MiB;
4545     mc->default_ram_id = "ppc_spapr.ram";
4546     mc->default_display = "std";
4547     mc->kvm_type = spapr_kvm_type;
4548     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4549     mc->pci_allow_0_address = true;
4550     assert(!mc->get_hotplug_handler);
4551     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4552     hc->pre_plug = spapr_machine_device_pre_plug;
4553     hc->plug = spapr_machine_device_plug;
4554     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4555     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4556     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4557     hc->unplug_request = spapr_machine_device_unplug_request;
4558     hc->unplug = spapr_machine_device_unplug;
4559 
4560     smc->dr_lmb_enabled = true;
4561     smc->update_dt_enabled = true;
4562     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4563     mc->has_hotpluggable_cpus = true;
4564     mc->nvdimm_supported = true;
4565     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4566     fwc->get_dev_path = spapr_get_fw_dev_path;
4567     nc->nmi_monitor_handler = spapr_nmi;
4568     smc->phb_placement = spapr_phb_placement;
4569     vhc->hypercall = emulate_spapr_hypercall;
4570     vhc->hpt_mask = spapr_hpt_mask;
4571     vhc->map_hptes = spapr_map_hptes;
4572     vhc->unmap_hptes = spapr_unmap_hptes;
4573     vhc->hpte_set_c = spapr_hpte_set_c;
4574     vhc->hpte_set_r = spapr_hpte_set_r;
4575     vhc->get_pate = spapr_get_pate;
4576     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4577     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4578     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4579     xic->ics_get = spapr_ics_get;
4580     xic->ics_resend = spapr_ics_resend;
4581     xic->icp_get = spapr_icp_get;
4582     ispc->print_info = spapr_pic_print_info;
4583     /* Force NUMA node memory size to be a multiple of
4584      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4585      * in which LMBs are represented and hot-added
4586      */
4587     mc->numa_mem_align_shift = 28;
4588     mc->auto_enable_numa = true;
4589 
4590     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4591     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4592     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4593     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4594     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4595     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4596     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4597     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4598     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4599     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4600     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4601     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4602     spapr_caps_add_properties(smc);
4603     smc->irq = &spapr_irq_dual;
4604     smc->dr_phb_enabled = true;
4605     smc->linux_pci_probe = true;
4606     smc->smp_threads_vsmt = true;
4607     smc->nr_xirqs = SPAPR_NR_XIRQS;
4608     xfc->match_nvt = spapr_match_nvt;
4609     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4610     vmc->quiesce = spapr_vof_quiesce;
4611     vmc->setprop = spapr_vof_setprop;
4612 }
4613 
4614 static const TypeInfo spapr_machine_info = {
4615     .name          = TYPE_SPAPR_MACHINE,
4616     .parent        = TYPE_MACHINE,
4617     .abstract      = true,
4618     .instance_size = sizeof(SpaprMachineState),
4619     .instance_init = spapr_instance_init,
4620     .instance_finalize = spapr_machine_finalizefn,
4621     .class_size    = sizeof(SpaprMachineClass),
4622     .class_init    = spapr_machine_class_init,
4623     .interfaces = (InterfaceInfo[]) {
4624         { TYPE_FW_PATH_PROVIDER },
4625         { TYPE_NMI },
4626         { TYPE_HOTPLUG_HANDLER },
4627         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4628         { TYPE_XICS_FABRIC },
4629         { TYPE_INTERRUPT_STATS_PROVIDER },
4630         { TYPE_XIVE_FABRIC },
4631         { TYPE_VOF_MACHINE_IF },
4632         { }
4633     },
4634 };
4635 
4636 static void spapr_machine_latest_class_options(MachineClass *mc)
4637 {
4638     mc->alias = "pseries";
4639     mc->is_default = true;
4640 }
4641 
4642 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4643     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4644                                                     void *data)      \
4645     {                                                                \
4646         MachineClass *mc = MACHINE_CLASS(oc);                        \
4647         spapr_machine_##suffix##_class_options(mc);                  \
4648         if (latest) {                                                \
4649             spapr_machine_latest_class_options(mc);                  \
4650         }                                                            \
4651     }                                                                \
4652     static const TypeInfo spapr_machine_##suffix##_info = {          \
4653         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4654         .parent = TYPE_SPAPR_MACHINE,                                \
4655         .class_init = spapr_machine_##suffix##_class_init,           \
4656     };                                                               \
4657     static void spapr_machine_register_##suffix(void)                \
4658     {                                                                \
4659         type_register(&spapr_machine_##suffix##_info);               \
4660     }                                                                \
4661     type_init(spapr_machine_register_##suffix)
4662 
4663 /*
4664  * pseries-6.2
4665  */
4666 static void spapr_machine_6_2_class_options(MachineClass *mc)
4667 {
4668     /* Defaults for the latest behaviour inherited from the base class */
4669 }
4670 
4671 DEFINE_SPAPR_MACHINE(6_2, "6.2", true);
4672 
4673 /*
4674  * pseries-6.1
4675  */
4676 static void spapr_machine_6_1_class_options(MachineClass *mc)
4677 {
4678     spapr_machine_6_2_class_options(mc);
4679     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4680 }
4681 
4682 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4683 
4684 /*
4685  * pseries-6.0
4686  */
4687 static void spapr_machine_6_0_class_options(MachineClass *mc)
4688 {
4689     spapr_machine_6_1_class_options(mc);
4690     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4691 }
4692 
4693 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4694 
4695 /*
4696  * pseries-5.2
4697  */
4698 static void spapr_machine_5_2_class_options(MachineClass *mc)
4699 {
4700     spapr_machine_6_0_class_options(mc);
4701     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4702 }
4703 
4704 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4705 
4706 /*
4707  * pseries-5.1
4708  */
4709 static void spapr_machine_5_1_class_options(MachineClass *mc)
4710 {
4711     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4712 
4713     spapr_machine_5_2_class_options(mc);
4714     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4715     smc->pre_5_2_numa_associativity = true;
4716 }
4717 
4718 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4719 
4720 /*
4721  * pseries-5.0
4722  */
4723 static void spapr_machine_5_0_class_options(MachineClass *mc)
4724 {
4725     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4726     static GlobalProperty compat[] = {
4727         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4728     };
4729 
4730     spapr_machine_5_1_class_options(mc);
4731     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4732     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4733     mc->numa_mem_supported = true;
4734     smc->pre_5_1_assoc_refpoints = true;
4735 }
4736 
4737 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4738 
4739 /*
4740  * pseries-4.2
4741  */
4742 static void spapr_machine_4_2_class_options(MachineClass *mc)
4743 {
4744     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4745 
4746     spapr_machine_5_0_class_options(mc);
4747     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4748     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4749     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4750     smc->rma_limit = 16 * GiB;
4751     mc->nvdimm_supported = false;
4752 }
4753 
4754 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4755 
4756 /*
4757  * pseries-4.1
4758  */
4759 static void spapr_machine_4_1_class_options(MachineClass *mc)
4760 {
4761     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4762     static GlobalProperty compat[] = {
4763         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4764         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4765     };
4766 
4767     spapr_machine_4_2_class_options(mc);
4768     smc->linux_pci_probe = false;
4769     smc->smp_threads_vsmt = false;
4770     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4771     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4772 }
4773 
4774 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4775 
4776 /*
4777  * pseries-4.0
4778  */
4779 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4780                               uint64_t *buid, hwaddr *pio,
4781                               hwaddr *mmio32, hwaddr *mmio64,
4782                               unsigned n_dma, uint32_t *liobns,
4783                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4784 {
4785     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4786                              liobns, nv2gpa, nv2atsd, errp)) {
4787         return false;
4788     }
4789 
4790     *nv2gpa = 0;
4791     *nv2atsd = 0;
4792     return true;
4793 }
4794 static void spapr_machine_4_0_class_options(MachineClass *mc)
4795 {
4796     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4797 
4798     spapr_machine_4_1_class_options(mc);
4799     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4800     smc->phb_placement = phb_placement_4_0;
4801     smc->irq = &spapr_irq_xics;
4802     smc->pre_4_1_migration = true;
4803 }
4804 
4805 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4806 
4807 /*
4808  * pseries-3.1
4809  */
4810 static void spapr_machine_3_1_class_options(MachineClass *mc)
4811 {
4812     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4813 
4814     spapr_machine_4_0_class_options(mc);
4815     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4816 
4817     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4818     smc->update_dt_enabled = false;
4819     smc->dr_phb_enabled = false;
4820     smc->broken_host_serial_model = true;
4821     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4822     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4823     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4824     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4825 }
4826 
4827 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4828 
4829 /*
4830  * pseries-3.0
4831  */
4832 
4833 static void spapr_machine_3_0_class_options(MachineClass *mc)
4834 {
4835     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4836 
4837     spapr_machine_3_1_class_options(mc);
4838     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4839 
4840     smc->legacy_irq_allocation = true;
4841     smc->nr_xirqs = 0x400;
4842     smc->irq = &spapr_irq_xics_legacy;
4843 }
4844 
4845 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4846 
4847 /*
4848  * pseries-2.12
4849  */
4850 static void spapr_machine_2_12_class_options(MachineClass *mc)
4851 {
4852     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4853     static GlobalProperty compat[] = {
4854         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4855         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4856     };
4857 
4858     spapr_machine_3_0_class_options(mc);
4859     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4860     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4861 
4862     /* We depend on kvm_enabled() to choose a default value for the
4863      * hpt-max-page-size capability. Of course we can't do it here
4864      * because this is too early and the HW accelerator isn't initialzed
4865      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4866      */
4867     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4868 }
4869 
4870 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4871 
4872 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4873 {
4874     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4875 
4876     spapr_machine_2_12_class_options(mc);
4877     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4878     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4879     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4880 }
4881 
4882 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4883 
4884 /*
4885  * pseries-2.11
4886  */
4887 
4888 static void spapr_machine_2_11_class_options(MachineClass *mc)
4889 {
4890     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4891 
4892     spapr_machine_2_12_class_options(mc);
4893     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4894     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4895 }
4896 
4897 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4898 
4899 /*
4900  * pseries-2.10
4901  */
4902 
4903 static void spapr_machine_2_10_class_options(MachineClass *mc)
4904 {
4905     spapr_machine_2_11_class_options(mc);
4906     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4907 }
4908 
4909 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4910 
4911 /*
4912  * pseries-2.9
4913  */
4914 
4915 static void spapr_machine_2_9_class_options(MachineClass *mc)
4916 {
4917     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4918     static GlobalProperty compat[] = {
4919         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4920     };
4921 
4922     spapr_machine_2_10_class_options(mc);
4923     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4924     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4925     smc->pre_2_10_has_unused_icps = true;
4926     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4927 }
4928 
4929 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4930 
4931 /*
4932  * pseries-2.8
4933  */
4934 
4935 static void spapr_machine_2_8_class_options(MachineClass *mc)
4936 {
4937     static GlobalProperty compat[] = {
4938         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4939     };
4940 
4941     spapr_machine_2_9_class_options(mc);
4942     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4943     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4944     mc->numa_mem_align_shift = 23;
4945 }
4946 
4947 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4948 
4949 /*
4950  * pseries-2.7
4951  */
4952 
4953 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4954                               uint64_t *buid, hwaddr *pio,
4955                               hwaddr *mmio32, hwaddr *mmio64,
4956                               unsigned n_dma, uint32_t *liobns,
4957                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4958 {
4959     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4960     const uint64_t base_buid = 0x800000020000000ULL;
4961     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4962     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4963     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4964     const uint32_t max_index = 255;
4965     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4966 
4967     uint64_t ram_top = MACHINE(spapr)->ram_size;
4968     hwaddr phb0_base, phb_base;
4969     int i;
4970 
4971     /* Do we have device memory? */
4972     if (MACHINE(spapr)->maxram_size > ram_top) {
4973         /* Can't just use maxram_size, because there may be an
4974          * alignment gap between normal and device memory regions
4975          */
4976         ram_top = MACHINE(spapr)->device_memory->base +
4977             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4978     }
4979 
4980     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4981 
4982     if (index > max_index) {
4983         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4984                    max_index);
4985         return false;
4986     }
4987 
4988     *buid = base_buid + index;
4989     for (i = 0; i < n_dma; ++i) {
4990         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4991     }
4992 
4993     phb_base = phb0_base + index * phb_spacing;
4994     *pio = phb_base + pio_offset;
4995     *mmio32 = phb_base + mmio_offset;
4996     /*
4997      * We don't set the 64-bit MMIO window, relying on the PHB's
4998      * fallback behaviour of automatically splitting a large "32-bit"
4999      * window into contiguous 32-bit and 64-bit windows
5000      */
5001 
5002     *nv2gpa = 0;
5003     *nv2atsd = 0;
5004     return true;
5005 }
5006 
5007 static void spapr_machine_2_7_class_options(MachineClass *mc)
5008 {
5009     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5010     static GlobalProperty compat[] = {
5011         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5012         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5013         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5014         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5015     };
5016 
5017     spapr_machine_2_8_class_options(mc);
5018     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5019     mc->default_machine_opts = "modern-hotplug-events=off";
5020     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5021     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5022     smc->phb_placement = phb_placement_2_7;
5023 }
5024 
5025 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5026 
5027 /*
5028  * pseries-2.6
5029  */
5030 
5031 static void spapr_machine_2_6_class_options(MachineClass *mc)
5032 {
5033     static GlobalProperty compat[] = {
5034         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5035     };
5036 
5037     spapr_machine_2_7_class_options(mc);
5038     mc->has_hotpluggable_cpus = false;
5039     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5040     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5041 }
5042 
5043 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5044 
5045 /*
5046  * pseries-2.5
5047  */
5048 
5049 static void spapr_machine_2_5_class_options(MachineClass *mc)
5050 {
5051     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5052     static GlobalProperty compat[] = {
5053         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5054     };
5055 
5056     spapr_machine_2_6_class_options(mc);
5057     smc->use_ohci_by_default = true;
5058     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5059     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5060 }
5061 
5062 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5063 
5064 /*
5065  * pseries-2.4
5066  */
5067 
5068 static void spapr_machine_2_4_class_options(MachineClass *mc)
5069 {
5070     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5071 
5072     spapr_machine_2_5_class_options(mc);
5073     smc->dr_lmb_enabled = false;
5074     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5075 }
5076 
5077 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5078 
5079 /*
5080  * pseries-2.3
5081  */
5082 
5083 static void spapr_machine_2_3_class_options(MachineClass *mc)
5084 {
5085     static GlobalProperty compat[] = {
5086         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5087     };
5088     spapr_machine_2_4_class_options(mc);
5089     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5090     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5091 }
5092 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5093 
5094 /*
5095  * pseries-2.2
5096  */
5097 
5098 static void spapr_machine_2_2_class_options(MachineClass *mc)
5099 {
5100     static GlobalProperty compat[] = {
5101         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5102     };
5103 
5104     spapr_machine_2_3_class_options(mc);
5105     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5106     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5107     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5108 }
5109 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5110 
5111 /*
5112  * pseries-2.1
5113  */
5114 
5115 static void spapr_machine_2_1_class_options(MachineClass *mc)
5116 {
5117     spapr_machine_2_2_class_options(mc);
5118     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5119 }
5120 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5121 
5122 static void spapr_machine_register_types(void)
5123 {
5124     type_register_static(&spapr_machine_info);
5125 }
5126 
5127 type_init(spapr_machine_register_types)
5128