xref: /qemu/hw/ppc/spapr.c (revision 17aa684ff7941f44af57cbb1c7b46bf5dc4f4fe2)
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/datadir.h"
29 #include "qemu/memalign.h"
30 #include "qemu/guest-random.h"
31 #include "qapi/error.h"
32 #include "qapi/qapi-events-machine.h"
33 #include "qapi/qapi-events-qdev.h"
34 #include "qapi/visitor.h"
35 #include "sysemu/sysemu.h"
36 #include "sysemu/hostmem.h"
37 #include "sysemu/numa.h"
38 #include "sysemu/qtest.h"
39 #include "sysemu/reset.h"
40 #include "sysemu/runstate.h"
41 #include "qemu/log.h"
42 #include "hw/fw-path-provider.h"
43 #include "elf.h"
44 #include "net/net.h"
45 #include "sysemu/device_tree.h"
46 #include "sysemu/cpus.h"
47 #include "sysemu/hw_accel.h"
48 #include "kvm_ppc.h"
49 #include "migration/misc.h"
50 #include "migration/qemu-file-types.h"
51 #include "migration/global_state.h"
52 #include "migration/register.h"
53 #include "migration/blocker.h"
54 #include "mmu-hash64.h"
55 #include "mmu-book3s-v3.h"
56 #include "cpu-models.h"
57 #include "hw/core/cpu.h"
58 
59 #include "hw/ppc/ppc.h"
60 #include "hw/loader.h"
61 
62 #include "hw/ppc/fdt.h"
63 #include "hw/ppc/spapr.h"
64 #include "hw/ppc/spapr_nested.h"
65 #include "hw/ppc/spapr_vio.h"
66 #include "hw/ppc/vof.h"
67 #include "hw/qdev-properties.h"
68 #include "hw/pci-host/spapr.h"
69 #include "hw/pci/msi.h"
70 
71 #include "hw/pci/pci.h"
72 #include "hw/scsi/scsi.h"
73 #include "hw/virtio/virtio-scsi.h"
74 #include "hw/virtio/vhost-scsi-common.h"
75 
76 #include "exec/ram_addr.h"
77 #include "hw/usb.h"
78 #include "qemu/config-file.h"
79 #include "qemu/error-report.h"
80 #include "trace.h"
81 #include "hw/nmi.h"
82 #include "hw/intc/intc.h"
83 
84 #include "hw/ppc/spapr_cpu_core.h"
85 #include "hw/mem/memory-device.h"
86 #include "hw/ppc/spapr_tpm_proxy.h"
87 #include "hw/ppc/spapr_nvdimm.h"
88 #include "hw/ppc/spapr_numa.h"
89 #include "hw/ppc/pef.h"
90 
91 #include "monitor/monitor.h"
92 
93 #include <libfdt.h>
94 
95 /* SLOF memory layout:
96  *
97  * SLOF raw image loaded at 0, copies its romfs right below the flat
98  * device-tree, then position SLOF itself 31M below that
99  *
100  * So we set FW_OVERHEAD to 40MB which should account for all of that
101  * and more
102  *
103  * We load our kernel at 4M, leaving space for SLOF initial image
104  */
105 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
106 #define FW_MAX_SIZE             0x400000
107 #define FW_FILE_NAME            "slof.bin"
108 #define FW_FILE_NAME_VOF        "vof.bin"
109 #define FW_OVERHEAD             0x2800000
110 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
111 
112 #define MIN_RMA_SLOF            (128 * MiB)
113 
114 #define PHANDLE_INTC            0x00001111
115 
116 /* These two functions implement the VCPU id numbering: one to compute them
117  * all and one to identify thread 0 of a VCORE. Any change to the first one
118  * is likely to have an impact on the second one, so let's keep them close.
119  */
120 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
121 {
122     MachineState *ms = MACHINE(spapr);
123     unsigned int smp_threads = ms->smp.threads;
124 
125     assert(spapr->vsmt);
126     return
127         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
128 }
129 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
130                                       PowerPCCPU *cpu)
131 {
132     assert(spapr->vsmt);
133     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
134 }
135 
136 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
137 {
138     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
139      * and newer QEMUs don't even have them. In both cases, we don't want
140      * to send anything on the wire.
141      */
142     return false;
143 }
144 
145 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
146     /*
147      * Hack ahead.  We can't have two devices with the same name and
148      * instance id.  So I rename this to pass make check.
149      * Real help from people who knows the hardware is needed.
150      */
151     .name = "icp/server",
152     .version_id = 1,
153     .minimum_version_id = 1,
154     .needed = pre_2_10_vmstate_dummy_icp_needed,
155     .fields = (const VMStateField[]) {
156         VMSTATE_UNUSED(4), /* uint32_t xirr */
157         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
158         VMSTATE_UNUSED(1), /* uint8_t mfrr */
159         VMSTATE_END_OF_LIST()
160     },
161 };
162 
163 /*
164  * See comment in hw/intc/xics.c:icp_realize()
165  *
166  * You have to remove vmstate_replace_hack_for_ppc() when you remove
167  * the machine types that need the following function.
168  */
169 static void pre_2_10_vmstate_register_dummy_icp(int i)
170 {
171     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
172                      (void *)(uintptr_t) i);
173 }
174 
175 /*
176  * See comment in hw/intc/xics.c:icp_realize()
177  *
178  * You have to remove vmstate_replace_hack_for_ppc() when you remove
179  * the machine types that need the following function.
180  */
181 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
182 {
183     /*
184      * This used to be:
185      *
186      *    vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
187      *                      (void *)(uintptr_t) i);
188      */
189 }
190 
191 int spapr_max_server_number(SpaprMachineState *spapr)
192 {
193     MachineState *ms = MACHINE(spapr);
194 
195     assert(spapr->vsmt);
196     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
197 }
198 
199 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
200                                   int smt_threads)
201 {
202     int i, ret = 0;
203     g_autofree uint32_t *servers_prop = g_new(uint32_t, smt_threads);
204     g_autofree uint32_t *gservers_prop = g_new(uint32_t, smt_threads * 2);
205     int index = spapr_get_vcpu_id(cpu);
206 
207     if (cpu->compat_pvr) {
208         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
209         if (ret < 0) {
210             return ret;
211         }
212     }
213 
214     /* Build interrupt servers and gservers properties */
215     for (i = 0; i < smt_threads; i++) {
216         servers_prop[i] = cpu_to_be32(index + i);
217         /* Hack, direct the group queues back to cpu 0 */
218         gservers_prop[i*2] = cpu_to_be32(index + i);
219         gservers_prop[i*2 + 1] = 0;
220     }
221     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
222                       servers_prop, sizeof(*servers_prop) * smt_threads);
223     if (ret < 0) {
224         return ret;
225     }
226     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
227                       gservers_prop, sizeof(*gservers_prop) * smt_threads * 2);
228 
229     return ret;
230 }
231 
232 static void spapr_dt_pa_features(SpaprMachineState *spapr,
233                                  PowerPCCPU *cpu,
234                                  void *fdt, int offset)
235 {
236     /*
237      * SSO (SAO) ordering is supported on KVM and thread=single hosts,
238      * but not MTTCG, so disable it. To advertise it, a cap would have
239      * to be added, or support implemented for MTTCG.
240      *
241      * Copy/paste is not supported by TCG, so it is not advertised. KVM
242      * can execute them but it has no accelerator drivers which are usable,
243      * so there isn't much need for it anyway.
244      */
245 
246     uint8_t pa_features_206[] = { 6, 0,
247         0xf6, 0x1f, 0xc7, 0x00, 0x00, 0xc0 };
248     uint8_t pa_features_207[] = { 24, 0,
249         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0,
250         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
251         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
252         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
253     uint8_t pa_features_300[] = { 66, 0,
254         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
255         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, 5: LE|CFAR|EB|LSQ */
256         0xf6, 0x1f, 0xc7, 0xc0, 0x00, 0xf0, /* 0 - 5 */
257         /* 6: DS207 */
258         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
259         /* 16: Vector */
260         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
261         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
262         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
263         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
264         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
265         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
266         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
267         /* 36: SPR SO, 40: Radix MMU */
268         0x80, 0x00, 0x00, 0x00, 0x80, 0x00, /* 36 - 41 */
269         /* 42: PM, 44: PC RA, 46: SC vec'd */
270         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
271         /* 48: SIMD, 50: QP BFP, 52: String */
272         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
273         /* 54: DecFP, 56: DecI, 58: SHA */
274         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
275         /* 60: NM atomic, 62: RNG */
276         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
277     };
278     uint8_t *pa_features = NULL;
279     size_t pa_size;
280 
281     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
282         pa_features = pa_features_206;
283         pa_size = sizeof(pa_features_206);
284     }
285     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
286         pa_features = pa_features_207;
287         pa_size = sizeof(pa_features_207);
288     }
289     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
290         pa_features = pa_features_300;
291         pa_size = sizeof(pa_features_300);
292     }
293     if (!pa_features) {
294         return;
295     }
296 
297     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
298         /*
299          * Note: we keep CI large pages off by default because a 64K capable
300          * guest provisioned with large pages might otherwise try to map a qemu
301          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
302          * even if that qemu runs on a 4k host.
303          * We dd this bit back here if we are confident this is not an issue
304          */
305         pa_features[3] |= 0x20;
306     }
307     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
308         pa_features[24] |= 0x80;    /* Transactional memory support */
309     }
310     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
311         /* Workaround for broken kernels that attempt (guest) radix
312          * mode when they can't handle it, if they see the radix bit set
313          * in pa-features. So hide it from them. */
314         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
315     }
316 
317     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
318 }
319 
320 static hwaddr spapr_node0_size(MachineState *machine)
321 {
322     if (machine->numa_state->num_nodes) {
323         int i;
324         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
325             if (machine->numa_state->nodes[i].node_mem) {
326                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
327                            machine->ram_size);
328             }
329         }
330     }
331     return machine->ram_size;
332 }
333 
334 static void add_str(GString *s, const gchar *s1)
335 {
336     g_string_append_len(s, s1, strlen(s1) + 1);
337 }
338 
339 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
340                                 hwaddr start, hwaddr size)
341 {
342     char mem_name[32];
343     uint64_t mem_reg_property[2];
344     int off;
345 
346     mem_reg_property[0] = cpu_to_be64(start);
347     mem_reg_property[1] = cpu_to_be64(size);
348 
349     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
350     off = fdt_add_subnode(fdt, 0, mem_name);
351     _FDT(off);
352     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
353     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
354                       sizeof(mem_reg_property))));
355     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
356     return off;
357 }
358 
359 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
360 {
361     MemoryDeviceInfoList *info;
362 
363     for (info = list; info; info = info->next) {
364         MemoryDeviceInfo *value = info->value;
365 
366         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
367             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
368 
369             if (addr >= pcdimm_info->addr &&
370                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
371                 return pcdimm_info->node;
372             }
373         }
374     }
375 
376     return -1;
377 }
378 
379 struct sPAPRDrconfCellV2 {
380      uint32_t seq_lmbs;
381      uint64_t base_addr;
382      uint32_t drc_index;
383      uint32_t aa_index;
384      uint32_t flags;
385 } QEMU_PACKED;
386 
387 typedef struct DrconfCellQueue {
388     struct sPAPRDrconfCellV2 cell;
389     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
390 } DrconfCellQueue;
391 
392 static DrconfCellQueue *
393 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
394                       uint32_t drc_index, uint32_t aa_index,
395                       uint32_t flags)
396 {
397     DrconfCellQueue *elem;
398 
399     elem = g_malloc0(sizeof(*elem));
400     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
401     elem->cell.base_addr = cpu_to_be64(base_addr);
402     elem->cell.drc_index = cpu_to_be32(drc_index);
403     elem->cell.aa_index = cpu_to_be32(aa_index);
404     elem->cell.flags = cpu_to_be32(flags);
405 
406     return elem;
407 }
408 
409 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
410                                       int offset, MemoryDeviceInfoList *dimms)
411 {
412     MachineState *machine = MACHINE(spapr);
413     uint8_t *int_buf, *cur_index;
414     int ret;
415     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
416     uint64_t addr, cur_addr, size;
417     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
418     uint64_t mem_end = machine->device_memory->base +
419                        memory_region_size(&machine->device_memory->mr);
420     uint32_t node, buf_len, nr_entries = 0;
421     SpaprDrc *drc;
422     DrconfCellQueue *elem, *next;
423     MemoryDeviceInfoList *info;
424     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
425         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
426 
427     /* Entry to cover RAM and the gap area */
428     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
429                                  SPAPR_LMB_FLAGS_RESERVED |
430                                  SPAPR_LMB_FLAGS_DRC_INVALID);
431     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
432     nr_entries++;
433 
434     cur_addr = machine->device_memory->base;
435     for (info = dimms; info; info = info->next) {
436         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
437 
438         addr = di->addr;
439         size = di->size;
440         node = di->node;
441 
442         /*
443          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
444          * area is marked hotpluggable in the next iteration for the bigger
445          * chunk including the NVDIMM occupied area.
446          */
447         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
448             continue;
449 
450         /* Entry for hot-pluggable area */
451         if (cur_addr < addr) {
452             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
453             g_assert(drc);
454             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
455                                          cur_addr, spapr_drc_index(drc), -1, 0);
456             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
457             nr_entries++;
458         }
459 
460         /* Entry for DIMM */
461         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
462         g_assert(drc);
463         elem = spapr_get_drconf_cell(size / lmb_size, addr,
464                                      spapr_drc_index(drc), node,
465                                      (SPAPR_LMB_FLAGS_ASSIGNED |
466                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
467         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
468         nr_entries++;
469         cur_addr = addr + size;
470     }
471 
472     /* Entry for remaining hotpluggable area */
473     if (cur_addr < mem_end) {
474         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
475         g_assert(drc);
476         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
477                                      cur_addr, spapr_drc_index(drc), -1, 0);
478         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
479         nr_entries++;
480     }
481 
482     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
483     int_buf = cur_index = g_malloc0(buf_len);
484     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
485     cur_index += sizeof(nr_entries);
486 
487     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
488         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
489         cur_index += sizeof(elem->cell);
490         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
491         g_free(elem);
492     }
493 
494     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
495     g_free(int_buf);
496     if (ret < 0) {
497         return -1;
498     }
499     return 0;
500 }
501 
502 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
503                                    int offset, MemoryDeviceInfoList *dimms)
504 {
505     MachineState *machine = MACHINE(spapr);
506     int i, ret;
507     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
508     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
509     uint32_t nr_lmbs = (machine->device_memory->base +
510                        memory_region_size(&machine->device_memory->mr)) /
511                        lmb_size;
512     uint32_t *int_buf, *cur_index, buf_len;
513 
514     /*
515      * Allocate enough buffer size to fit in ibm,dynamic-memory
516      */
517     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
518     cur_index = int_buf = g_malloc0(buf_len);
519     int_buf[0] = cpu_to_be32(nr_lmbs);
520     cur_index++;
521     for (i = 0; i < nr_lmbs; i++) {
522         uint64_t addr = i * lmb_size;
523         uint32_t *dynamic_memory = cur_index;
524 
525         if (i >= device_lmb_start) {
526             SpaprDrc *drc;
527 
528             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
529             g_assert(drc);
530 
531             dynamic_memory[0] = cpu_to_be32(addr >> 32);
532             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
533             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
534             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
535             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
536             if (memory_region_present(get_system_memory(), addr)) {
537                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
538             } else {
539                 dynamic_memory[5] = cpu_to_be32(0);
540             }
541         } else {
542             /*
543              * LMB information for RMA, boot time RAM and gap b/n RAM and
544              * device memory region -- all these are marked as reserved
545              * and as having no valid DRC.
546              */
547             dynamic_memory[0] = cpu_to_be32(addr >> 32);
548             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
549             dynamic_memory[2] = cpu_to_be32(0);
550             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
551             dynamic_memory[4] = cpu_to_be32(-1);
552             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
553                                             SPAPR_LMB_FLAGS_DRC_INVALID);
554         }
555 
556         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
557     }
558     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
559     g_free(int_buf);
560     if (ret < 0) {
561         return -1;
562     }
563     return 0;
564 }
565 
566 /*
567  * Adds ibm,dynamic-reconfiguration-memory node.
568  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
569  * of this device tree node.
570  */
571 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
572                                                    void *fdt)
573 {
574     MachineState *machine = MACHINE(spapr);
575     int ret, offset;
576     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
577     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
578                                 cpu_to_be32(lmb_size & 0xffffffff)};
579     MemoryDeviceInfoList *dimms = NULL;
580 
581     /* Don't create the node if there is no device memory. */
582     if (!machine->device_memory) {
583         return 0;
584     }
585 
586     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
587 
588     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
589                     sizeof(prop_lmb_size));
590     if (ret < 0) {
591         return ret;
592     }
593 
594     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
595     if (ret < 0) {
596         return ret;
597     }
598 
599     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
600     if (ret < 0) {
601         return ret;
602     }
603 
604     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
605     dimms = qmp_memory_device_list();
606     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
607         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
608     } else {
609         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
610     }
611     qapi_free_MemoryDeviceInfoList(dimms);
612 
613     if (ret < 0) {
614         return ret;
615     }
616 
617     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
618 
619     return ret;
620 }
621 
622 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
623 {
624     MachineState *machine = MACHINE(spapr);
625     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
626     hwaddr mem_start, node_size;
627     int i, nb_nodes = machine->numa_state->num_nodes;
628     NodeInfo *nodes = machine->numa_state->nodes;
629 
630     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
631         if (!nodes[i].node_mem) {
632             continue;
633         }
634         if (mem_start >= machine->ram_size) {
635             node_size = 0;
636         } else {
637             node_size = nodes[i].node_mem;
638             if (node_size > machine->ram_size - mem_start) {
639                 node_size = machine->ram_size - mem_start;
640             }
641         }
642         if (!mem_start) {
643             /* spapr_machine_init() checks for rma_size <= node0_size
644              * already */
645             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
646             mem_start += spapr->rma_size;
647             node_size -= spapr->rma_size;
648         }
649         for ( ; node_size; ) {
650             hwaddr sizetmp = pow2floor(node_size);
651 
652             /* mem_start != 0 here */
653             if (ctzl(mem_start) < ctzl(sizetmp)) {
654                 sizetmp = 1ULL << ctzl(mem_start);
655             }
656 
657             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
658             node_size -= sizetmp;
659             mem_start += sizetmp;
660         }
661     }
662 
663     /* Generate ibm,dynamic-reconfiguration-memory node if required */
664     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
665         int ret;
666 
667         g_assert(smc->dr_lmb_enabled);
668         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
669         if (ret) {
670             return ret;
671         }
672     }
673 
674     return 0;
675 }
676 
677 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
678                          SpaprMachineState *spapr)
679 {
680     MachineState *ms = MACHINE(spapr);
681     PowerPCCPU *cpu = POWERPC_CPU(cs);
682     CPUPPCState *env = &cpu->env;
683     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
684     int index = spapr_get_vcpu_id(cpu);
685     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
686                        0xffffffff, 0xffffffff};
687     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
688         : SPAPR_TIMEBASE_FREQ;
689     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
690     uint32_t page_sizes_prop[64];
691     size_t page_sizes_prop_size;
692     unsigned int smp_threads = ms->smp.threads;
693     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
694     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
695     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
696     SpaprDrc *drc;
697     int drc_index;
698     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
699     int i;
700 
701     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
702     if (drc) {
703         drc_index = spapr_drc_index(drc);
704         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
705     }
706 
707     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
708     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
709 
710     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
711     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
712                            env->dcache_line_size)));
713     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
714                            env->dcache_line_size)));
715     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
716                            env->icache_line_size)));
717     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
718                            env->icache_line_size)));
719 
720     if (pcc->l1_dcache_size) {
721         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
722                                pcc->l1_dcache_size)));
723     } else {
724         warn_report("Unknown L1 dcache size for cpu");
725     }
726     if (pcc->l1_icache_size) {
727         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
728                                pcc->l1_icache_size)));
729     } else {
730         warn_report("Unknown L1 icache size for cpu");
731     }
732 
733     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
734     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
735     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
736     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
737     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
738     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
739 
740     if (ppc_has_spr(cpu, SPR_PURR)) {
741         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
742     }
743     if (ppc_has_spr(cpu, SPR_PURR)) {
744         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
745     }
746 
747     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
748         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
749                           segs, sizeof(segs))));
750     }
751 
752     /* Advertise VSX (vector extensions) if available
753      *   1               == VMX / Altivec available
754      *   2               == VSX available
755      *
756      * Only CPUs for which we create core types in spapr_cpu_core.c
757      * are possible, and all of those have VMX */
758     if (env->insns_flags & PPC_ALTIVEC) {
759         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
760             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
761         } else {
762             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
763         }
764     }
765 
766     /* Advertise DFP (Decimal Floating Point) if available
767      *   0 / no property == no DFP
768      *   1               == DFP available */
769     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
770         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
771     }
772 
773     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
774                                                       sizeof(page_sizes_prop));
775     if (page_sizes_prop_size) {
776         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
777                           page_sizes_prop, page_sizes_prop_size)));
778     }
779 
780     spapr_dt_pa_features(spapr, cpu, fdt, offset);
781 
782     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
783                            cs->cpu_index / vcpus_per_socket)));
784 
785     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
786                       pft_size_prop, sizeof(pft_size_prop))));
787 
788     if (ms->numa_state->num_nodes > 1) {
789         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
790     }
791 
792     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
793 
794     if (pcc->radix_page_info) {
795         for (i = 0; i < pcc->radix_page_info->count; i++) {
796             radix_AP_encodings[i] =
797                 cpu_to_be32(pcc->radix_page_info->entries[i]);
798         }
799         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
800                           radix_AP_encodings,
801                           pcc->radix_page_info->count *
802                           sizeof(radix_AP_encodings[0]))));
803     }
804 
805     /*
806      * We set this property to let the guest know that it can use the large
807      * decrementer and its width in bits.
808      */
809     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
810         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
811                               pcc->lrg_decr_bits)));
812 }
813 
814 static void spapr_dt_one_cpu(void *fdt, SpaprMachineState *spapr, CPUState *cs,
815                              int cpus_offset)
816 {
817     PowerPCCPU *cpu = POWERPC_CPU(cs);
818     int index = spapr_get_vcpu_id(cpu);
819     DeviceClass *dc = DEVICE_GET_CLASS(cs);
820     g_autofree char *nodename = NULL;
821     int offset;
822 
823     if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
824         return;
825     }
826 
827     nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
828     offset = fdt_add_subnode(fdt, cpus_offset, nodename);
829     _FDT(offset);
830     spapr_dt_cpu(cs, fdt, offset, spapr);
831 }
832 
833 
834 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
835 {
836     CPUState **rev;
837     CPUState *cs;
838     int n_cpus;
839     int cpus_offset;
840     int i;
841 
842     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
843     _FDT(cpus_offset);
844     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
845     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
846 
847     /*
848      * We walk the CPUs in reverse order to ensure that CPU DT nodes
849      * created by fdt_add_subnode() end up in the right order in FDT
850      * for the guest kernel the enumerate the CPUs correctly.
851      *
852      * The CPU list cannot be traversed in reverse order, so we need
853      * to do extra work.
854      */
855     n_cpus = 0;
856     rev = NULL;
857     CPU_FOREACH(cs) {
858         rev = g_renew(CPUState *, rev, n_cpus + 1);
859         rev[n_cpus++] = cs;
860     }
861 
862     for (i = n_cpus - 1; i >= 0; i--) {
863         spapr_dt_one_cpu(fdt, spapr, rev[i], cpus_offset);
864     }
865 
866     g_free(rev);
867 }
868 
869 static int spapr_dt_rng(void *fdt)
870 {
871     int node;
872     int ret;
873 
874     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
875     if (node <= 0) {
876         return -1;
877     }
878     ret = fdt_setprop_string(fdt, node, "device_type",
879                              "ibm,platform-facilities");
880     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
881     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
882 
883     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
884     if (node <= 0) {
885         return -1;
886     }
887     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
888 
889     return ret ? -1 : 0;
890 }
891 
892 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
893 {
894     MachineState *ms = MACHINE(spapr);
895     int rtas;
896     GString *hypertas = g_string_sized_new(256);
897     GString *qemu_hypertas = g_string_sized_new(256);
898     uint32_t lrdr_capacity[] = {
899         0,
900         0,
901         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
902         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
903         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
904     };
905 
906     /* Do we have device memory? */
907     if (MACHINE(spapr)->device_memory) {
908         uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
909             memory_region_size(&MACHINE(spapr)->device_memory->mr);
910 
911         lrdr_capacity[0] = cpu_to_be32(max_device_addr >> 32);
912         lrdr_capacity[1] = cpu_to_be32(max_device_addr & 0xffffffff);
913     }
914 
915     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
916 
917     /* hypertas */
918     add_str(hypertas, "hcall-pft");
919     add_str(hypertas, "hcall-term");
920     add_str(hypertas, "hcall-dabr");
921     add_str(hypertas, "hcall-interrupt");
922     add_str(hypertas, "hcall-tce");
923     add_str(hypertas, "hcall-vio");
924     add_str(hypertas, "hcall-splpar");
925     add_str(hypertas, "hcall-join");
926     add_str(hypertas, "hcall-bulk");
927     add_str(hypertas, "hcall-set-mode");
928     add_str(hypertas, "hcall-sprg0");
929     add_str(hypertas, "hcall-copy");
930     add_str(hypertas, "hcall-debug");
931     add_str(hypertas, "hcall-vphn");
932     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
933         add_str(hypertas, "hcall-rpt-invalidate");
934     }
935 
936     add_str(qemu_hypertas, "hcall-memop1");
937 
938     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
939         add_str(hypertas, "hcall-multi-tce");
940     }
941 
942     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
943         add_str(hypertas, "hcall-hpt-resize");
944     }
945 
946     add_str(hypertas, "hcall-watchdog");
947 
948     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
949                      hypertas->str, hypertas->len));
950     g_string_free(hypertas, TRUE);
951     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
952                      qemu_hypertas->str, qemu_hypertas->len));
953     g_string_free(qemu_hypertas, TRUE);
954 
955     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
956 
957     /*
958      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
959      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
960      *
961      * The system reset requirements are driven by existing Linux and PowerVM
962      * implementation which (contrary to PAPR) saves r3 in the error log
963      * structure like machine check, so Linux expects to find the saved r3
964      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
965      * does not look at the error value).
966      *
967      * System reset interrupts are not subject to interlock like machine
968      * check, so this memory area could be corrupted if the sreset is
969      * interrupted by a machine check (or vice versa) if it was shared. To
970      * prevent this, system reset uses per-CPU areas for the sreset save
971      * area. A system reset that interrupts a system reset handler could
972      * still overwrite this area, but Linux doesn't try to recover in that
973      * case anyway.
974      *
975      * The extra 8 bytes is required because Linux's FWNMI error log check
976      * is off-by-one.
977      *
978      * RTAS_MIN_SIZE is required for the RTAS blob itself.
979      */
980     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
981                           RTAS_ERROR_LOG_MAX +
982                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
983                           sizeof(uint64_t)));
984     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
985                           RTAS_ERROR_LOG_MAX));
986     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
987                           RTAS_EVENT_SCAN_RATE));
988 
989     g_assert(msi_nonbroken);
990     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
991 
992     /*
993      * According to PAPR, rtas ibm,os-term does not guarantee a return
994      * back to the guest cpu.
995      *
996      * While an additional ibm,extended-os-term property indicates
997      * that rtas call return will always occur. Set this property.
998      */
999     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1000 
1001     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1002                      lrdr_capacity, sizeof(lrdr_capacity)));
1003 
1004     spapr_dt_rtas_tokens(fdt, rtas);
1005 }
1006 
1007 /*
1008  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1009  * and the XIVE features that the guest may request and thus the valid
1010  * values for bytes 23..26 of option vector 5:
1011  */
1012 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1013                                           int chosen)
1014 {
1015     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1016 
1017     char val[2 * 4] = {
1018         23, 0x00, /* XICS / XIVE mode */
1019         24, 0x00, /* Hash/Radix, filled in below. */
1020         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1021         26, 0x40, /* Radix options: GTSE == yes. */
1022     };
1023 
1024     if (spapr->irq->xics && spapr->irq->xive) {
1025         val[1] = SPAPR_OV5_XIVE_BOTH;
1026     } else if (spapr->irq->xive) {
1027         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1028     } else {
1029         assert(spapr->irq->xics);
1030         val[1] = SPAPR_OV5_XIVE_LEGACY;
1031     }
1032 
1033     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1034                           first_ppc_cpu->compat_pvr)) {
1035         /*
1036          * If we're in a pre POWER9 compat mode then the guest should
1037          * do hash and use the legacy interrupt mode
1038          */
1039         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1040         val[3] = 0x00; /* Hash */
1041         spapr_check_mmu_mode(false);
1042     } else if (kvm_enabled()) {
1043         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1044             val[3] = 0x80; /* OV5_MMU_BOTH */
1045         } else if (kvmppc_has_cap_mmu_radix()) {
1046             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1047         } else {
1048             val[3] = 0x00; /* Hash */
1049         }
1050     } else {
1051         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1052         val[3] = 0xC0;
1053     }
1054     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1055                      val, sizeof(val)));
1056 }
1057 
1058 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1059 {
1060     MachineState *machine = MACHINE(spapr);
1061     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1062     int chosen;
1063 
1064     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1065 
1066     if (reset) {
1067         const char *boot_device = spapr->boot_device;
1068         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1069         size_t cb = 0;
1070         g_autofree char *bootlist = get_boot_devices_list(&cb);
1071 
1072         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1073             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1074                                     machine->kernel_cmdline));
1075         }
1076 
1077         if (spapr->initrd_size) {
1078             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1079                                   spapr->initrd_base));
1080             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1081                                   spapr->initrd_base + spapr->initrd_size));
1082         }
1083 
1084         if (spapr->kernel_size) {
1085             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1086                                   cpu_to_be64(spapr->kernel_size) };
1087 
1088             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1089                          &kprop, sizeof(kprop)));
1090             if (spapr->kernel_le) {
1091                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1092             }
1093         }
1094         if (machine->boot_config.has_menu && machine->boot_config.menu) {
1095             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", true)));
1096         }
1097         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1098         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1099         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1100 
1101         if (cb && bootlist) {
1102             int i;
1103 
1104             for (i = 0; i < cb; i++) {
1105                 if (bootlist[i] == '\n') {
1106                     bootlist[i] = ' ';
1107                 }
1108             }
1109             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1110         }
1111 
1112         if (boot_device && strlen(boot_device)) {
1113             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1114         }
1115 
1116         if (spapr->want_stdout_path && stdout_path) {
1117             /*
1118              * "linux,stdout-path" and "stdout" properties are
1119              * deprecated by linux kernel. New platforms should only
1120              * use the "stdout-path" property. Set the new property
1121              * and continue using older property to remain compatible
1122              * with the existing firmware.
1123              */
1124             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1125             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1126         }
1127 
1128         /*
1129          * We can deal with BAR reallocation just fine, advertise it
1130          * to the guest
1131          */
1132         if (smc->linux_pci_probe) {
1133             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1134         }
1135 
1136         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1137     }
1138 
1139     _FDT(fdt_setprop(fdt, chosen, "rng-seed", spapr->fdt_rng_seed, 32));
1140 
1141     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1142 }
1143 
1144 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1145 {
1146     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1147      * KVM to work under pHyp with some guest co-operation */
1148     int hypervisor;
1149     uint8_t hypercall[16];
1150 
1151     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1152     /* indicate KVM hypercall interface */
1153     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1154     if (kvmppc_has_cap_fixup_hcalls()) {
1155         /*
1156          * Older KVM versions with older guest kernels were broken
1157          * with the magic page, don't allow the guest to map it.
1158          */
1159         if (!kvmppc_get_hypercall(cpu_env(first_cpu), hypercall,
1160                                   sizeof(hypercall))) {
1161             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1162                              hypercall, sizeof(hypercall)));
1163         }
1164     }
1165 }
1166 
1167 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1168 {
1169     MachineState *machine = MACHINE(spapr);
1170     MachineClass *mc = MACHINE_GET_CLASS(machine);
1171     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1172     uint32_t root_drc_type_mask = 0;
1173     int ret;
1174     void *fdt;
1175     SpaprPhbState *phb;
1176     char *buf;
1177 
1178     fdt = g_malloc0(space);
1179     _FDT((fdt_create_empty_tree(fdt, space)));
1180 
1181     /* Root node */
1182     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1183     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1184     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1185 
1186     /* Guest UUID & Name*/
1187     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1188     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1189     if (qemu_uuid_set) {
1190         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1191     }
1192     g_free(buf);
1193 
1194     if (qemu_get_vm_name()) {
1195         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1196                                 qemu_get_vm_name()));
1197     }
1198 
1199     /* Host Model & Serial Number */
1200     if (spapr->host_model) {
1201         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1202     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1203         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1204         g_free(buf);
1205     }
1206 
1207     if (spapr->host_serial) {
1208         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1209     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1210         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1211         g_free(buf);
1212     }
1213 
1214     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1215     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1216 
1217     /* /interrupt controller */
1218     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1219 
1220     ret = spapr_dt_memory(spapr, fdt);
1221     if (ret < 0) {
1222         error_report("couldn't setup memory nodes in fdt");
1223         exit(1);
1224     }
1225 
1226     /* /vdevice */
1227     spapr_dt_vdevice(spapr->vio_bus, fdt);
1228 
1229     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1230         ret = spapr_dt_rng(fdt);
1231         if (ret < 0) {
1232             error_report("could not set up rng device in the fdt");
1233             exit(1);
1234         }
1235     }
1236 
1237     QLIST_FOREACH(phb, &spapr->phbs, list) {
1238         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1239         if (ret < 0) {
1240             error_report("couldn't setup PCI devices in fdt");
1241             exit(1);
1242         }
1243     }
1244 
1245     spapr_dt_cpus(fdt, spapr);
1246 
1247     /* ibm,drc-indexes and friends */
1248     if (smc->dr_lmb_enabled) {
1249         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1250     }
1251     if (smc->dr_phb_enabled) {
1252         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1253     }
1254     if (mc->nvdimm_supported) {
1255         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1256     }
1257     if (root_drc_type_mask) {
1258         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1259     }
1260 
1261     if (mc->has_hotpluggable_cpus) {
1262         int offset = fdt_path_offset(fdt, "/cpus");
1263         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1264         if (ret < 0) {
1265             error_report("Couldn't set up CPU DR device tree properties");
1266             exit(1);
1267         }
1268     }
1269 
1270     /* /event-sources */
1271     spapr_dt_events(spapr, fdt);
1272 
1273     /* /rtas */
1274     spapr_dt_rtas(spapr, fdt);
1275 
1276     /* /chosen */
1277     spapr_dt_chosen(spapr, fdt, reset);
1278 
1279     /* /hypervisor */
1280     if (kvm_enabled()) {
1281         spapr_dt_hypervisor(spapr, fdt);
1282     }
1283 
1284     /* Build memory reserve map */
1285     if (reset) {
1286         if (spapr->kernel_size) {
1287             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1288                                   spapr->kernel_size)));
1289         }
1290         if (spapr->initrd_size) {
1291             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1292                                   spapr->initrd_size)));
1293         }
1294     }
1295 
1296     /* NVDIMM devices */
1297     if (mc->nvdimm_supported) {
1298         spapr_dt_persistent_memory(spapr, fdt);
1299     }
1300 
1301     return fdt;
1302 }
1303 
1304 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1305 {
1306     SpaprMachineState *spapr = opaque;
1307 
1308     return (addr & 0x0fffffff) + spapr->kernel_addr;
1309 }
1310 
1311 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1312                                     PowerPCCPU *cpu)
1313 {
1314     CPUPPCState *env = &cpu->env;
1315 
1316     /* The TCG path should also be holding the BQL at this point */
1317     g_assert(bql_locked());
1318 
1319     g_assert(!vhyp_cpu_in_nested(cpu));
1320 
1321     if (FIELD_EX64(env->msr, MSR, PR)) {
1322         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1323         env->gpr[3] = H_PRIVILEGE;
1324     } else {
1325         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1326     }
1327 }
1328 
1329 struct LPCRSyncState {
1330     target_ulong value;
1331     target_ulong mask;
1332 };
1333 
1334 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1335 {
1336     struct LPCRSyncState *s = arg.host_ptr;
1337     PowerPCCPU *cpu = POWERPC_CPU(cs);
1338     CPUPPCState *env = &cpu->env;
1339     target_ulong lpcr;
1340 
1341     cpu_synchronize_state(cs);
1342     lpcr = env->spr[SPR_LPCR];
1343     lpcr &= ~s->mask;
1344     lpcr |= s->value;
1345     ppc_store_lpcr(cpu, lpcr);
1346 }
1347 
1348 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1349 {
1350     CPUState *cs;
1351     struct LPCRSyncState s = {
1352         .value = value,
1353         .mask = mask
1354     };
1355     CPU_FOREACH(cs) {
1356         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1357     }
1358 }
1359 
1360 /* May be used when the machine is not running */
1361 void spapr_init_all_lpcrs(target_ulong value, target_ulong mask)
1362 {
1363     CPUState *cs;
1364     CPU_FOREACH(cs) {
1365         PowerPCCPU *cpu = POWERPC_CPU(cs);
1366         CPUPPCState *env = &cpu->env;
1367         target_ulong lpcr;
1368 
1369         lpcr = env->spr[SPR_LPCR];
1370         lpcr &= ~(LPCR_HR | LPCR_UPRT);
1371         ppc_store_lpcr(cpu, lpcr);
1372     }
1373 }
1374 
1375 
1376 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1377                            target_ulong lpid, ppc_v3_pate_t *entry)
1378 {
1379     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1380     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1381 
1382     if (!spapr_cpu->in_nested) {
1383         assert(lpid == 0);
1384 
1385         /* Copy PATE1:GR into PATE0:HR */
1386         entry->dw0 = spapr->patb_entry & PATE0_HR;
1387         entry->dw1 = spapr->patb_entry;
1388 
1389     } else {
1390         uint64_t patb, pats;
1391 
1392         assert(lpid != 0);
1393 
1394         patb = spapr->nested_ptcr & PTCR_PATB;
1395         pats = spapr->nested_ptcr & PTCR_PATS;
1396 
1397         /* Check if partition table is properly aligned */
1398         if (patb & MAKE_64BIT_MASK(0, pats + 12)) {
1399             return false;
1400         }
1401 
1402         /* Calculate number of entries */
1403         pats = 1ull << (pats + 12 - 4);
1404         if (pats <= lpid) {
1405             return false;
1406         }
1407 
1408         /* Grab entry */
1409         patb += 16 * lpid;
1410         entry->dw0 = ldq_phys(CPU(cpu)->as, patb);
1411         entry->dw1 = ldq_phys(CPU(cpu)->as, patb + 8);
1412     }
1413 
1414     return true;
1415 }
1416 
1417 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1418 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1419 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1420 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1421 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1422 
1423 /*
1424  * Get the fd to access the kernel htab, re-opening it if necessary
1425  */
1426 static int get_htab_fd(SpaprMachineState *spapr)
1427 {
1428     Error *local_err = NULL;
1429 
1430     if (spapr->htab_fd >= 0) {
1431         return spapr->htab_fd;
1432     }
1433 
1434     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1435     if (spapr->htab_fd < 0) {
1436         error_report_err(local_err);
1437     }
1438 
1439     return spapr->htab_fd;
1440 }
1441 
1442 void close_htab_fd(SpaprMachineState *spapr)
1443 {
1444     if (spapr->htab_fd >= 0) {
1445         close(spapr->htab_fd);
1446     }
1447     spapr->htab_fd = -1;
1448 }
1449 
1450 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1451 {
1452     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1453 
1454     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1455 }
1456 
1457 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1458 {
1459     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1460 
1461     assert(kvm_enabled());
1462 
1463     if (!spapr->htab) {
1464         return 0;
1465     }
1466 
1467     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1468 }
1469 
1470 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1471                                                 hwaddr ptex, int n)
1472 {
1473     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1474     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1475 
1476     if (!spapr->htab) {
1477         /*
1478          * HTAB is controlled by KVM. Fetch into temporary buffer
1479          */
1480         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1481         kvmppc_read_hptes(hptes, ptex, n);
1482         return hptes;
1483     }
1484 
1485     /*
1486      * HTAB is controlled by QEMU. Just point to the internally
1487      * accessible PTEG.
1488      */
1489     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1490 }
1491 
1492 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1493                               const ppc_hash_pte64_t *hptes,
1494                               hwaddr ptex, int n)
1495 {
1496     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1497 
1498     if (!spapr->htab) {
1499         g_free((void *)hptes);
1500     }
1501 
1502     /* Nothing to do for qemu managed HPT */
1503 }
1504 
1505 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1506                       uint64_t pte0, uint64_t pte1)
1507 {
1508     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1509     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1510 
1511     if (!spapr->htab) {
1512         kvmppc_write_hpte(ptex, pte0, pte1);
1513     } else {
1514         if (pte0 & HPTE64_V_VALID) {
1515             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1516             /*
1517              * When setting valid, we write PTE1 first. This ensures
1518              * proper synchronization with the reading code in
1519              * ppc_hash64_pteg_search()
1520              */
1521             smp_wmb();
1522             stq_p(spapr->htab + offset, pte0);
1523         } else {
1524             stq_p(spapr->htab + offset, pte0);
1525             /*
1526              * When clearing it we set PTE0 first. This ensures proper
1527              * synchronization with the reading code in
1528              * ppc_hash64_pteg_search()
1529              */
1530             smp_wmb();
1531             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1532         }
1533     }
1534 }
1535 
1536 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1537                              uint64_t pte1)
1538 {
1539     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1540     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1541 
1542     if (!spapr->htab) {
1543         /* There should always be a hash table when this is called */
1544         error_report("spapr_hpte_set_c called with no hash table !");
1545         return;
1546     }
1547 
1548     /* The HW performs a non-atomic byte update */
1549     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1550 }
1551 
1552 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1553                              uint64_t pte1)
1554 {
1555     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1556     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1557 
1558     if (!spapr->htab) {
1559         /* There should always be a hash table when this is called */
1560         error_report("spapr_hpte_set_r called with no hash table !");
1561         return;
1562     }
1563 
1564     /* The HW performs a non-atomic byte update */
1565     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1566 }
1567 
1568 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1569 {
1570     int shift;
1571 
1572     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1573      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1574      * that's much more than is needed for Linux guests */
1575     shift = ctz64(pow2ceil(ramsize)) - 7;
1576     shift = MAX(shift, 18); /* Minimum architected size */
1577     shift = MIN(shift, 46); /* Maximum architected size */
1578     return shift;
1579 }
1580 
1581 void spapr_free_hpt(SpaprMachineState *spapr)
1582 {
1583     qemu_vfree(spapr->htab);
1584     spapr->htab = NULL;
1585     spapr->htab_shift = 0;
1586     close_htab_fd(spapr);
1587 }
1588 
1589 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1590 {
1591     ERRP_GUARD();
1592     long rc;
1593 
1594     /* Clean up any HPT info from a previous boot */
1595     spapr_free_hpt(spapr);
1596 
1597     rc = kvmppc_reset_htab(shift);
1598 
1599     if (rc == -EOPNOTSUPP) {
1600         error_setg(errp, "HPT not supported in nested guests");
1601         return -EOPNOTSUPP;
1602     }
1603 
1604     if (rc < 0) {
1605         /* kernel-side HPT needed, but couldn't allocate one */
1606         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1607                          shift);
1608         error_append_hint(errp, "Try smaller maxmem?\n");
1609         return -errno;
1610     } else if (rc > 0) {
1611         /* kernel-side HPT allocated */
1612         if (rc != shift) {
1613             error_setg(errp,
1614                        "Requested order %d HPT, but kernel allocated order %ld",
1615                        shift, rc);
1616             error_append_hint(errp, "Try smaller maxmem?\n");
1617             return -ENOSPC;
1618         }
1619 
1620         spapr->htab_shift = shift;
1621         spapr->htab = NULL;
1622     } else {
1623         /* kernel-side HPT not needed, allocate in userspace instead */
1624         size_t size = 1ULL << shift;
1625         int i;
1626 
1627         spapr->htab = qemu_memalign(size, size);
1628         memset(spapr->htab, 0, size);
1629         spapr->htab_shift = shift;
1630 
1631         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1632             DIRTY_HPTE(HPTE(spapr->htab, i));
1633         }
1634     }
1635     /* We're setting up a hash table, so that means we're not radix */
1636     spapr->patb_entry = 0;
1637     spapr_init_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1638     return 0;
1639 }
1640 
1641 void spapr_setup_hpt(SpaprMachineState *spapr)
1642 {
1643     int hpt_shift;
1644 
1645     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1646         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1647     } else {
1648         uint64_t current_ram_size;
1649 
1650         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1651         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1652     }
1653     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1654 
1655     if (kvm_enabled()) {
1656         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1657 
1658         /* Check our RMA fits in the possible VRMA */
1659         if (vrma_limit < spapr->rma_size) {
1660             error_report("Unable to create %" HWADDR_PRIu
1661                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1662                          spapr->rma_size / MiB, vrma_limit / MiB);
1663             exit(EXIT_FAILURE);
1664         }
1665     }
1666 }
1667 
1668 void spapr_check_mmu_mode(bool guest_radix)
1669 {
1670     if (guest_radix) {
1671         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1672             error_report("Guest requested unavailable MMU mode (radix).");
1673             exit(EXIT_FAILURE);
1674         }
1675     } else {
1676         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1677             && !kvmppc_has_cap_mmu_hash_v3()) {
1678             error_report("Guest requested unavailable MMU mode (hash).");
1679             exit(EXIT_FAILURE);
1680         }
1681     }
1682 }
1683 
1684 static void spapr_machine_reset(MachineState *machine, ShutdownCause reason)
1685 {
1686     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1687     PowerPCCPU *first_ppc_cpu;
1688     hwaddr fdt_addr;
1689     void *fdt;
1690     int rc;
1691 
1692     if (reason != SHUTDOWN_CAUSE_SNAPSHOT_LOAD) {
1693         /*
1694          * Record-replay snapshot load must not consume random, this was
1695          * already replayed from initial machine reset.
1696          */
1697         qemu_guest_getrandom_nofail(spapr->fdt_rng_seed, 32);
1698     }
1699 
1700     pef_kvm_reset(machine->cgs, &error_fatal);
1701     spapr_caps_apply(spapr);
1702 
1703     first_ppc_cpu = POWERPC_CPU(first_cpu);
1704     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1705         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1706                               spapr->max_compat_pvr)) {
1707         /*
1708          * If using KVM with radix mode available, VCPUs can be started
1709          * without a HPT because KVM will start them in radix mode.
1710          * Set the GR bit in PATE so that we know there is no HPT.
1711          */
1712         spapr->patb_entry = PATE1_GR;
1713         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1714     } else {
1715         spapr_setup_hpt(spapr);
1716     }
1717 
1718     qemu_devices_reset(reason);
1719 
1720     spapr_ovec_cleanup(spapr->ov5_cas);
1721     spapr->ov5_cas = spapr_ovec_new();
1722 
1723     ppc_init_compat_all(spapr->max_compat_pvr, &error_fatal);
1724 
1725     /*
1726      * This is fixing some of the default configuration of the XIVE
1727      * devices. To be called after the reset of the machine devices.
1728      */
1729     spapr_irq_reset(spapr, &error_fatal);
1730 
1731     /*
1732      * There is no CAS under qtest. Simulate one to please the code that
1733      * depends on spapr->ov5_cas. This is especially needed to test device
1734      * unplug, so we do that before resetting the DRCs.
1735      */
1736     if (qtest_enabled()) {
1737         spapr_ovec_cleanup(spapr->ov5_cas);
1738         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1739     }
1740 
1741     spapr_nvdimm_finish_flushes();
1742 
1743     /* DRC reset may cause a device to be unplugged. This will cause troubles
1744      * if this device is used by another device (eg, a running vhost backend
1745      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1746      * situations, we reset DRCs after all devices have been reset.
1747      */
1748     spapr_drc_reset_all(spapr);
1749 
1750     spapr_clear_pending_events(spapr);
1751 
1752     /*
1753      * We place the device tree just below either the top of the RMA,
1754      * or just below 2GB, whichever is lower, so that it can be
1755      * processed with 32-bit real mode code if necessary
1756      */
1757     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1758 
1759     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1760     if (spapr->vof) {
1761         spapr_vof_reset(spapr, fdt, &error_fatal);
1762         /*
1763          * Do not pack the FDT as the client may change properties.
1764          * VOF client does not expect the FDT so we do not load it to the VM.
1765          */
1766     } else {
1767         rc = fdt_pack(fdt);
1768         /* Should only fail if we've built a corrupted tree */
1769         assert(rc == 0);
1770 
1771         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1772                                   0, fdt_addr, 0);
1773         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1774     }
1775     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1776 
1777     g_free(spapr->fdt_blob);
1778     spapr->fdt_size = fdt_totalsize(fdt);
1779     spapr->fdt_initial_size = spapr->fdt_size;
1780     spapr->fdt_blob = fdt;
1781 
1782     /* Set machine->fdt for 'dumpdtb' QMP/HMP command */
1783     machine->fdt = fdt;
1784 
1785     /* Set up the entry state */
1786     first_ppc_cpu->env.gpr[5] = 0;
1787 
1788     spapr->fwnmi_system_reset_addr = -1;
1789     spapr->fwnmi_machine_check_addr = -1;
1790     spapr->fwnmi_machine_check_interlock = -1;
1791 
1792     /* Signal all vCPUs waiting on this condition */
1793     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1794 
1795     migrate_del_blocker(&spapr->fwnmi_migration_blocker);
1796 }
1797 
1798 static void spapr_create_nvram(SpaprMachineState *spapr)
1799 {
1800     DeviceState *dev = qdev_new("spapr-nvram");
1801     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1802 
1803     if (dinfo) {
1804         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1805                                 &error_fatal);
1806     }
1807 
1808     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1809 
1810     spapr->nvram = (struct SpaprNvram *)dev;
1811 }
1812 
1813 static void spapr_rtc_create(SpaprMachineState *spapr)
1814 {
1815     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1816                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1817                                        &error_fatal, NULL);
1818     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1819     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1820                               "date");
1821 }
1822 
1823 /* Returns whether we want to use VGA or not */
1824 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1825 {
1826     vga_interface_created = true;
1827     switch (vga_interface_type) {
1828     case VGA_NONE:
1829         return false;
1830     case VGA_DEVICE:
1831         return true;
1832     case VGA_STD:
1833     case VGA_VIRTIO:
1834     case VGA_CIRRUS:
1835         return pci_vga_init(pci_bus) != NULL;
1836     default:
1837         error_setg(errp,
1838                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1839         return false;
1840     }
1841 }
1842 
1843 static int spapr_pre_load(void *opaque)
1844 {
1845     int rc;
1846 
1847     rc = spapr_caps_pre_load(opaque);
1848     if (rc) {
1849         return rc;
1850     }
1851 
1852     return 0;
1853 }
1854 
1855 static int spapr_post_load(void *opaque, int version_id)
1856 {
1857     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1858     int err = 0;
1859 
1860     err = spapr_caps_post_migration(spapr);
1861     if (err) {
1862         return err;
1863     }
1864 
1865     /*
1866      * In earlier versions, there was no separate qdev for the PAPR
1867      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1868      * So when migrating from those versions, poke the incoming offset
1869      * value into the RTC device
1870      */
1871     if (version_id < 3) {
1872         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1873         if (err) {
1874             return err;
1875         }
1876     }
1877 
1878     if (kvm_enabled() && spapr->patb_entry) {
1879         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1880         bool radix = !!(spapr->patb_entry & PATE1_GR);
1881         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1882 
1883         /*
1884          * Update LPCR:HR and UPRT as they may not be set properly in
1885          * the stream
1886          */
1887         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1888                             LPCR_HR | LPCR_UPRT);
1889 
1890         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1891         if (err) {
1892             error_report("Process table config unsupported by the host");
1893             return -EINVAL;
1894         }
1895     }
1896 
1897     err = spapr_irq_post_load(spapr, version_id);
1898     if (err) {
1899         return err;
1900     }
1901 
1902     return err;
1903 }
1904 
1905 static int spapr_pre_save(void *opaque)
1906 {
1907     int rc;
1908 
1909     rc = spapr_caps_pre_save(opaque);
1910     if (rc) {
1911         return rc;
1912     }
1913 
1914     return 0;
1915 }
1916 
1917 static bool version_before_3(void *opaque, int version_id)
1918 {
1919     return version_id < 3;
1920 }
1921 
1922 static bool spapr_pending_events_needed(void *opaque)
1923 {
1924     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1925     return !QTAILQ_EMPTY(&spapr->pending_events);
1926 }
1927 
1928 static const VMStateDescription vmstate_spapr_event_entry = {
1929     .name = "spapr_event_log_entry",
1930     .version_id = 1,
1931     .minimum_version_id = 1,
1932     .fields = (const VMStateField[]) {
1933         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1934         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1935         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1936                                      NULL, extended_length),
1937         VMSTATE_END_OF_LIST()
1938     },
1939 };
1940 
1941 static const VMStateDescription vmstate_spapr_pending_events = {
1942     .name = "spapr_pending_events",
1943     .version_id = 1,
1944     .minimum_version_id = 1,
1945     .needed = spapr_pending_events_needed,
1946     .fields = (const VMStateField[]) {
1947         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1948                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1949         VMSTATE_END_OF_LIST()
1950     },
1951 };
1952 
1953 static bool spapr_ov5_cas_needed(void *opaque)
1954 {
1955     SpaprMachineState *spapr = opaque;
1956     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1957     bool cas_needed;
1958 
1959     /* Prior to the introduction of SpaprOptionVector, we had two option
1960      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1961      * Both of these options encode machine topology into the device-tree
1962      * in such a way that the now-booted OS should still be able to interact
1963      * appropriately with QEMU regardless of what options were actually
1964      * negotiatied on the source side.
1965      *
1966      * As such, we can avoid migrating the CAS-negotiated options if these
1967      * are the only options available on the current machine/platform.
1968      * Since these are the only options available for pseries-2.7 and
1969      * earlier, this allows us to maintain old->new/new->old migration
1970      * compatibility.
1971      *
1972      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1973      * via default pseries-2.8 machines and explicit command-line parameters.
1974      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1975      * of the actual CAS-negotiated values to continue working properly. For
1976      * example, availability of memory unplug depends on knowing whether
1977      * OV5_HP_EVT was negotiated via CAS.
1978      *
1979      * Thus, for any cases where the set of available CAS-negotiatable
1980      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1981      * include the CAS-negotiated options in the migration stream, unless
1982      * if they affect boot time behaviour only.
1983      */
1984     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1985     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1986     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1987 
1988     /* We need extra information if we have any bits outside the mask
1989      * defined above */
1990     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1991 
1992     spapr_ovec_cleanup(ov5_mask);
1993 
1994     return cas_needed;
1995 }
1996 
1997 static const VMStateDescription vmstate_spapr_ov5_cas = {
1998     .name = "spapr_option_vector_ov5_cas",
1999     .version_id = 1,
2000     .minimum_version_id = 1,
2001     .needed = spapr_ov5_cas_needed,
2002     .fields = (const VMStateField[]) {
2003         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
2004                                  vmstate_spapr_ovec, SpaprOptionVector),
2005         VMSTATE_END_OF_LIST()
2006     },
2007 };
2008 
2009 static bool spapr_patb_entry_needed(void *opaque)
2010 {
2011     SpaprMachineState *spapr = opaque;
2012 
2013     return !!spapr->patb_entry;
2014 }
2015 
2016 static const VMStateDescription vmstate_spapr_patb_entry = {
2017     .name = "spapr_patb_entry",
2018     .version_id = 1,
2019     .minimum_version_id = 1,
2020     .needed = spapr_patb_entry_needed,
2021     .fields = (const VMStateField[]) {
2022         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2023         VMSTATE_END_OF_LIST()
2024     },
2025 };
2026 
2027 static bool spapr_irq_map_needed(void *opaque)
2028 {
2029     SpaprMachineState *spapr = opaque;
2030 
2031     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2032 }
2033 
2034 static const VMStateDescription vmstate_spapr_irq_map = {
2035     .name = "spapr_irq_map",
2036     .version_id = 1,
2037     .minimum_version_id = 1,
2038     .needed = spapr_irq_map_needed,
2039     .fields = (const VMStateField[]) {
2040         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2041         VMSTATE_END_OF_LIST()
2042     },
2043 };
2044 
2045 static bool spapr_dtb_needed(void *opaque)
2046 {
2047     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2048 
2049     return smc->update_dt_enabled;
2050 }
2051 
2052 static int spapr_dtb_pre_load(void *opaque)
2053 {
2054     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2055 
2056     g_free(spapr->fdt_blob);
2057     spapr->fdt_blob = NULL;
2058     spapr->fdt_size = 0;
2059 
2060     return 0;
2061 }
2062 
2063 static const VMStateDescription vmstate_spapr_dtb = {
2064     .name = "spapr_dtb",
2065     .version_id = 1,
2066     .minimum_version_id = 1,
2067     .needed = spapr_dtb_needed,
2068     .pre_load = spapr_dtb_pre_load,
2069     .fields = (const VMStateField[]) {
2070         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2071         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2072         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2073                                      fdt_size),
2074         VMSTATE_END_OF_LIST()
2075     },
2076 };
2077 
2078 static bool spapr_fwnmi_needed(void *opaque)
2079 {
2080     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2081 
2082     return spapr->fwnmi_machine_check_addr != -1;
2083 }
2084 
2085 static int spapr_fwnmi_pre_save(void *opaque)
2086 {
2087     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2088 
2089     /*
2090      * Check if machine check handling is in progress and print a
2091      * warning message.
2092      */
2093     if (spapr->fwnmi_machine_check_interlock != -1) {
2094         warn_report("A machine check is being handled during migration. The"
2095                 "handler may run and log hardware error on the destination");
2096     }
2097 
2098     return 0;
2099 }
2100 
2101 static const VMStateDescription vmstate_spapr_fwnmi = {
2102     .name = "spapr_fwnmi",
2103     .version_id = 1,
2104     .minimum_version_id = 1,
2105     .needed = spapr_fwnmi_needed,
2106     .pre_save = spapr_fwnmi_pre_save,
2107     .fields = (const VMStateField[]) {
2108         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2109         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2110         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2111         VMSTATE_END_OF_LIST()
2112     },
2113 };
2114 
2115 static const VMStateDescription vmstate_spapr = {
2116     .name = "spapr",
2117     .version_id = 3,
2118     .minimum_version_id = 1,
2119     .pre_load = spapr_pre_load,
2120     .post_load = spapr_post_load,
2121     .pre_save = spapr_pre_save,
2122     .fields = (const VMStateField[]) {
2123         /* used to be @next_irq */
2124         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2125 
2126         /* RTC offset */
2127         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2128 
2129         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2130         VMSTATE_END_OF_LIST()
2131     },
2132     .subsections = (const VMStateDescription * const []) {
2133         &vmstate_spapr_ov5_cas,
2134         &vmstate_spapr_patb_entry,
2135         &vmstate_spapr_pending_events,
2136         &vmstate_spapr_cap_htm,
2137         &vmstate_spapr_cap_vsx,
2138         &vmstate_spapr_cap_dfp,
2139         &vmstate_spapr_cap_cfpc,
2140         &vmstate_spapr_cap_sbbc,
2141         &vmstate_spapr_cap_ibs,
2142         &vmstate_spapr_cap_hpt_maxpagesize,
2143         &vmstate_spapr_irq_map,
2144         &vmstate_spapr_cap_nested_kvm_hv,
2145         &vmstate_spapr_dtb,
2146         &vmstate_spapr_cap_large_decr,
2147         &vmstate_spapr_cap_ccf_assist,
2148         &vmstate_spapr_cap_fwnmi,
2149         &vmstate_spapr_fwnmi,
2150         &vmstate_spapr_cap_rpt_invalidate,
2151         NULL
2152     }
2153 };
2154 
2155 static int htab_save_setup(QEMUFile *f, void *opaque)
2156 {
2157     SpaprMachineState *spapr = opaque;
2158 
2159     /* "Iteration" header */
2160     if (!spapr->htab_shift) {
2161         qemu_put_be32(f, -1);
2162     } else {
2163         qemu_put_be32(f, spapr->htab_shift);
2164     }
2165 
2166     if (spapr->htab) {
2167         spapr->htab_save_index = 0;
2168         spapr->htab_first_pass = true;
2169     } else {
2170         if (spapr->htab_shift) {
2171             assert(kvm_enabled());
2172         }
2173     }
2174 
2175 
2176     return 0;
2177 }
2178 
2179 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2180                             int chunkstart, int n_valid, int n_invalid)
2181 {
2182     qemu_put_be32(f, chunkstart);
2183     qemu_put_be16(f, n_valid);
2184     qemu_put_be16(f, n_invalid);
2185     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2186                     HASH_PTE_SIZE_64 * n_valid);
2187 }
2188 
2189 static void htab_save_end_marker(QEMUFile *f)
2190 {
2191     qemu_put_be32(f, 0);
2192     qemu_put_be16(f, 0);
2193     qemu_put_be16(f, 0);
2194 }
2195 
2196 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2197                                  int64_t max_ns)
2198 {
2199     bool has_timeout = max_ns != -1;
2200     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2201     int index = spapr->htab_save_index;
2202     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2203 
2204     assert(spapr->htab_first_pass);
2205 
2206     do {
2207         int chunkstart;
2208 
2209         /* Consume invalid HPTEs */
2210         while ((index < htabslots)
2211                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2212             CLEAN_HPTE(HPTE(spapr->htab, index));
2213             index++;
2214         }
2215 
2216         /* Consume valid HPTEs */
2217         chunkstart = index;
2218         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2219                && HPTE_VALID(HPTE(spapr->htab, index))) {
2220             CLEAN_HPTE(HPTE(spapr->htab, index));
2221             index++;
2222         }
2223 
2224         if (index > chunkstart) {
2225             int n_valid = index - chunkstart;
2226 
2227             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2228 
2229             if (has_timeout &&
2230                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2231                 break;
2232             }
2233         }
2234     } while ((index < htabslots) && !migration_rate_exceeded(f));
2235 
2236     if (index >= htabslots) {
2237         assert(index == htabslots);
2238         index = 0;
2239         spapr->htab_first_pass = false;
2240     }
2241     spapr->htab_save_index = index;
2242 }
2243 
2244 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2245                                 int64_t max_ns)
2246 {
2247     bool final = max_ns < 0;
2248     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2249     int examined = 0, sent = 0;
2250     int index = spapr->htab_save_index;
2251     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2252 
2253     assert(!spapr->htab_first_pass);
2254 
2255     do {
2256         int chunkstart, invalidstart;
2257 
2258         /* Consume non-dirty HPTEs */
2259         while ((index < htabslots)
2260                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2261             index++;
2262             examined++;
2263         }
2264 
2265         chunkstart = index;
2266         /* Consume valid dirty HPTEs */
2267         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2268                && HPTE_DIRTY(HPTE(spapr->htab, index))
2269                && HPTE_VALID(HPTE(spapr->htab, index))) {
2270             CLEAN_HPTE(HPTE(spapr->htab, index));
2271             index++;
2272             examined++;
2273         }
2274 
2275         invalidstart = index;
2276         /* Consume invalid dirty HPTEs */
2277         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2278                && HPTE_DIRTY(HPTE(spapr->htab, index))
2279                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2280             CLEAN_HPTE(HPTE(spapr->htab, index));
2281             index++;
2282             examined++;
2283         }
2284 
2285         if (index > chunkstart) {
2286             int n_valid = invalidstart - chunkstart;
2287             int n_invalid = index - invalidstart;
2288 
2289             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2290             sent += index - chunkstart;
2291 
2292             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2293                 break;
2294             }
2295         }
2296 
2297         if (examined >= htabslots) {
2298             break;
2299         }
2300 
2301         if (index >= htabslots) {
2302             assert(index == htabslots);
2303             index = 0;
2304         }
2305     } while ((examined < htabslots) && (!migration_rate_exceeded(f) || final));
2306 
2307     if (index >= htabslots) {
2308         assert(index == htabslots);
2309         index = 0;
2310     }
2311 
2312     spapr->htab_save_index = index;
2313 
2314     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2315 }
2316 
2317 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2318 #define MAX_KVM_BUF_SIZE    2048
2319 
2320 static int htab_save_iterate(QEMUFile *f, void *opaque)
2321 {
2322     SpaprMachineState *spapr = opaque;
2323     int fd;
2324     int rc = 0;
2325 
2326     /* Iteration header */
2327     if (!spapr->htab_shift) {
2328         qemu_put_be32(f, -1);
2329         return 1;
2330     } else {
2331         qemu_put_be32(f, 0);
2332     }
2333 
2334     if (!spapr->htab) {
2335         assert(kvm_enabled());
2336 
2337         fd = get_htab_fd(spapr);
2338         if (fd < 0) {
2339             return fd;
2340         }
2341 
2342         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2343         if (rc < 0) {
2344             return rc;
2345         }
2346     } else  if (spapr->htab_first_pass) {
2347         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2348     } else {
2349         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2350     }
2351 
2352     htab_save_end_marker(f);
2353 
2354     return rc;
2355 }
2356 
2357 static int htab_save_complete(QEMUFile *f, void *opaque)
2358 {
2359     SpaprMachineState *spapr = opaque;
2360     int fd;
2361 
2362     /* Iteration header */
2363     if (!spapr->htab_shift) {
2364         qemu_put_be32(f, -1);
2365         return 0;
2366     } else {
2367         qemu_put_be32(f, 0);
2368     }
2369 
2370     if (!spapr->htab) {
2371         int rc;
2372 
2373         assert(kvm_enabled());
2374 
2375         fd = get_htab_fd(spapr);
2376         if (fd < 0) {
2377             return fd;
2378         }
2379 
2380         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2381         if (rc < 0) {
2382             return rc;
2383         }
2384     } else {
2385         if (spapr->htab_first_pass) {
2386             htab_save_first_pass(f, spapr, -1);
2387         }
2388         htab_save_later_pass(f, spapr, -1);
2389     }
2390 
2391     /* End marker */
2392     htab_save_end_marker(f);
2393 
2394     return 0;
2395 }
2396 
2397 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2398 {
2399     SpaprMachineState *spapr = opaque;
2400     uint32_t section_hdr;
2401     int fd = -1;
2402     Error *local_err = NULL;
2403 
2404     if (version_id < 1 || version_id > 1) {
2405         error_report("htab_load() bad version");
2406         return -EINVAL;
2407     }
2408 
2409     section_hdr = qemu_get_be32(f);
2410 
2411     if (section_hdr == -1) {
2412         spapr_free_hpt(spapr);
2413         return 0;
2414     }
2415 
2416     if (section_hdr) {
2417         int ret;
2418 
2419         /* First section gives the htab size */
2420         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2421         if (ret < 0) {
2422             error_report_err(local_err);
2423             return ret;
2424         }
2425         return 0;
2426     }
2427 
2428     if (!spapr->htab) {
2429         assert(kvm_enabled());
2430 
2431         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2432         if (fd < 0) {
2433             error_report_err(local_err);
2434             return fd;
2435         }
2436     }
2437 
2438     while (true) {
2439         uint32_t index;
2440         uint16_t n_valid, n_invalid;
2441 
2442         index = qemu_get_be32(f);
2443         n_valid = qemu_get_be16(f);
2444         n_invalid = qemu_get_be16(f);
2445 
2446         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2447             /* End of Stream */
2448             break;
2449         }
2450 
2451         if ((index + n_valid + n_invalid) >
2452             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2453             /* Bad index in stream */
2454             error_report(
2455                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2456                 index, n_valid, n_invalid, spapr->htab_shift);
2457             return -EINVAL;
2458         }
2459 
2460         if (spapr->htab) {
2461             if (n_valid) {
2462                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2463                                 HASH_PTE_SIZE_64 * n_valid);
2464             }
2465             if (n_invalid) {
2466                 memset(HPTE(spapr->htab, index + n_valid), 0,
2467                        HASH_PTE_SIZE_64 * n_invalid);
2468             }
2469         } else {
2470             int rc;
2471 
2472             assert(fd >= 0);
2473 
2474             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2475                                         &local_err);
2476             if (rc < 0) {
2477                 error_report_err(local_err);
2478                 return rc;
2479             }
2480         }
2481     }
2482 
2483     if (!spapr->htab) {
2484         assert(fd >= 0);
2485         close(fd);
2486     }
2487 
2488     return 0;
2489 }
2490 
2491 static void htab_save_cleanup(void *opaque)
2492 {
2493     SpaprMachineState *spapr = opaque;
2494 
2495     close_htab_fd(spapr);
2496 }
2497 
2498 static SaveVMHandlers savevm_htab_handlers = {
2499     .save_setup = htab_save_setup,
2500     .save_live_iterate = htab_save_iterate,
2501     .save_live_complete_precopy = htab_save_complete,
2502     .save_cleanup = htab_save_cleanup,
2503     .load_state = htab_load,
2504 };
2505 
2506 static void spapr_boot_set(void *opaque, const char *boot_device,
2507                            Error **errp)
2508 {
2509     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2510 
2511     g_free(spapr->boot_device);
2512     spapr->boot_device = g_strdup(boot_device);
2513 }
2514 
2515 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2516 {
2517     MachineState *machine = MACHINE(spapr);
2518     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2519     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2520     int i;
2521 
2522     g_assert(!nr_lmbs || machine->device_memory);
2523     for (i = 0; i < nr_lmbs; i++) {
2524         uint64_t addr;
2525 
2526         addr = i * lmb_size + machine->device_memory->base;
2527         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2528                                addr / lmb_size);
2529     }
2530 }
2531 
2532 /*
2533  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2534  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2535  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2536  */
2537 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2538 {
2539     int i;
2540 
2541     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2542         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2543                    " is not aligned to %" PRIu64 " MiB",
2544                    machine->ram_size,
2545                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2546         return;
2547     }
2548 
2549     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2550         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2551                    " is not aligned to %" PRIu64 " MiB",
2552                    machine->ram_size,
2553                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2554         return;
2555     }
2556 
2557     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2558         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2559             error_setg(errp,
2560                        "Node %d memory size 0x%" PRIx64
2561                        " is not aligned to %" PRIu64 " MiB",
2562                        i, machine->numa_state->nodes[i].node_mem,
2563                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2564             return;
2565         }
2566     }
2567 }
2568 
2569 /* find cpu slot in machine->possible_cpus by core_id */
2570 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2571 {
2572     int index = id / ms->smp.threads;
2573 
2574     if (index >= ms->possible_cpus->len) {
2575         return NULL;
2576     }
2577     if (idx) {
2578         *idx = index;
2579     }
2580     return &ms->possible_cpus->cpus[index];
2581 }
2582 
2583 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2584 {
2585     MachineState *ms = MACHINE(spapr);
2586     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2587     Error *local_err = NULL;
2588     bool vsmt_user = !!spapr->vsmt;
2589     int kvm_smt = kvmppc_smt_threads();
2590     int ret;
2591     unsigned int smp_threads = ms->smp.threads;
2592 
2593     if (tcg_enabled()) {
2594         if (smp_threads > 1 &&
2595             !ppc_type_check_compat(ms->cpu_type, CPU_POWERPC_LOGICAL_2_07, 0,
2596                                    spapr->max_compat_pvr)) {
2597             error_setg(errp, "TCG only supports SMT on POWER8 or newer CPUs");
2598             return;
2599         }
2600 
2601         if (smp_threads > 8) {
2602             error_setg(errp, "TCG cannot support more than 8 threads/core "
2603                        "on a pseries machine");
2604             return;
2605         }
2606     }
2607     if (!is_power_of_2(smp_threads)) {
2608         error_setg(errp, "Cannot support %d threads/core on a pseries "
2609                    "machine because it must be a power of 2", smp_threads);
2610         return;
2611     }
2612 
2613     /* Determine the VSMT mode to use: */
2614     if (vsmt_user) {
2615         if (spapr->vsmt < smp_threads) {
2616             error_setg(errp, "Cannot support VSMT mode %d"
2617                        " because it must be >= threads/core (%d)",
2618                        spapr->vsmt, smp_threads);
2619             return;
2620         }
2621         /* In this case, spapr->vsmt has been set by the command line */
2622     } else if (!smc->smp_threads_vsmt) {
2623         /*
2624          * Default VSMT value is tricky, because we need it to be as
2625          * consistent as possible (for migration), but this requires
2626          * changing it for at least some existing cases.  We pick 8 as
2627          * the value that we'd get with KVM on POWER8, the
2628          * overwhelmingly common case in production systems.
2629          */
2630         spapr->vsmt = MAX(8, smp_threads);
2631     } else {
2632         spapr->vsmt = smp_threads;
2633     }
2634 
2635     /* KVM: If necessary, set the SMT mode: */
2636     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2637         ret = kvmppc_set_smt_threads(spapr->vsmt);
2638         if (ret) {
2639             /* Looks like KVM isn't able to change VSMT mode */
2640             error_setg(&local_err,
2641                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2642                        spapr->vsmt, ret);
2643             /* We can live with that if the default one is big enough
2644              * for the number of threads, and a submultiple of the one
2645              * we want.  In this case we'll waste some vcpu ids, but
2646              * behaviour will be correct */
2647             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2648                 warn_report_err(local_err);
2649             } else {
2650                 if (!vsmt_user) {
2651                     error_append_hint(&local_err,
2652                                       "On PPC, a VM with %d threads/core"
2653                                       " on a host with %d threads/core"
2654                                       " requires the use of VSMT mode %d.\n",
2655                                       smp_threads, kvm_smt, spapr->vsmt);
2656                 }
2657                 kvmppc_error_append_smt_possible_hint(&local_err);
2658                 error_propagate(errp, local_err);
2659             }
2660         }
2661     }
2662     /* else TCG: nothing to do currently */
2663 }
2664 
2665 static void spapr_init_cpus(SpaprMachineState *spapr)
2666 {
2667     MachineState *machine = MACHINE(spapr);
2668     MachineClass *mc = MACHINE_GET_CLASS(machine);
2669     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2670     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2671     const CPUArchIdList *possible_cpus;
2672     unsigned int smp_cpus = machine->smp.cpus;
2673     unsigned int smp_threads = machine->smp.threads;
2674     unsigned int max_cpus = machine->smp.max_cpus;
2675     int boot_cores_nr = smp_cpus / smp_threads;
2676     int i;
2677 
2678     possible_cpus = mc->possible_cpu_arch_ids(machine);
2679     if (mc->has_hotpluggable_cpus) {
2680         if (smp_cpus % smp_threads) {
2681             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2682                          smp_cpus, smp_threads);
2683             exit(1);
2684         }
2685         if (max_cpus % smp_threads) {
2686             error_report("max_cpus (%u) must be multiple of threads (%u)",
2687                          max_cpus, smp_threads);
2688             exit(1);
2689         }
2690     } else {
2691         if (max_cpus != smp_cpus) {
2692             error_report("This machine version does not support CPU hotplug");
2693             exit(1);
2694         }
2695         boot_cores_nr = possible_cpus->len;
2696     }
2697 
2698     if (smc->pre_2_10_has_unused_icps) {
2699         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2700             /* Dummy entries get deregistered when real ICPState objects
2701              * are registered during CPU core hotplug.
2702              */
2703             pre_2_10_vmstate_register_dummy_icp(i);
2704         }
2705     }
2706 
2707     for (i = 0; i < possible_cpus->len; i++) {
2708         int core_id = i * smp_threads;
2709 
2710         if (mc->has_hotpluggable_cpus) {
2711             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2712                                    spapr_vcpu_id(spapr, core_id));
2713         }
2714 
2715         if (i < boot_cores_nr) {
2716             Object *core  = object_new(type);
2717             int nr_threads = smp_threads;
2718 
2719             /* Handle the partially filled core for older machine types */
2720             if ((i + 1) * smp_threads >= smp_cpus) {
2721                 nr_threads = smp_cpus - i * smp_threads;
2722             }
2723 
2724             object_property_set_int(core, "nr-threads", nr_threads,
2725                                     &error_fatal);
2726             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2727                                     &error_fatal);
2728             qdev_realize(DEVICE(core), NULL, &error_fatal);
2729 
2730             object_unref(core);
2731         }
2732     }
2733 }
2734 
2735 static PCIHostState *spapr_create_default_phb(void)
2736 {
2737     DeviceState *dev;
2738 
2739     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2740     qdev_prop_set_uint32(dev, "index", 0);
2741     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2742 
2743     return PCI_HOST_BRIDGE(dev);
2744 }
2745 
2746 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2747 {
2748     MachineState *machine = MACHINE(spapr);
2749     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2750     hwaddr rma_size = machine->ram_size;
2751     hwaddr node0_size = spapr_node0_size(machine);
2752 
2753     /* RMA has to fit in the first NUMA node */
2754     rma_size = MIN(rma_size, node0_size);
2755 
2756     /*
2757      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2758      * never exceed that
2759      */
2760     rma_size = MIN(rma_size, 1 * TiB);
2761 
2762     /*
2763      * Clamp the RMA size based on machine type.  This is for
2764      * migration compatibility with older qemu versions, which limited
2765      * the RMA size for complicated and mostly bad reasons.
2766      */
2767     if (smc->rma_limit) {
2768         rma_size = MIN(rma_size, smc->rma_limit);
2769     }
2770 
2771     if (rma_size < MIN_RMA_SLOF) {
2772         error_setg(errp,
2773                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2774                    "ldMiB guest RMA (Real Mode Area memory)",
2775                    MIN_RMA_SLOF / MiB);
2776         return 0;
2777     }
2778 
2779     return rma_size;
2780 }
2781 
2782 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2783 {
2784     MachineState *machine = MACHINE(spapr);
2785     int i;
2786 
2787     for (i = 0; i < machine->ram_slots; i++) {
2788         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2789     }
2790 }
2791 
2792 /* pSeries LPAR / sPAPR hardware init */
2793 static void spapr_machine_init(MachineState *machine)
2794 {
2795     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2796     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2797     MachineClass *mc = MACHINE_GET_CLASS(machine);
2798     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2799     const char *bios_name = machine->firmware ?: bios_default;
2800     g_autofree char *filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2801     const char *kernel_filename = machine->kernel_filename;
2802     const char *initrd_filename = machine->initrd_filename;
2803     PCIHostState *phb;
2804     bool has_vga;
2805     int i;
2806     MemoryRegion *sysmem = get_system_memory();
2807     long load_limit, fw_size;
2808     Error *resize_hpt_err = NULL;
2809     NICInfo *nd;
2810 
2811     if (!filename) {
2812         error_report("Could not find LPAR firmware '%s'", bios_name);
2813         exit(1);
2814     }
2815     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2816     if (fw_size <= 0) {
2817         error_report("Could not load LPAR firmware '%s'", filename);
2818         exit(1);
2819     }
2820 
2821     /*
2822      * if Secure VM (PEF) support is configured, then initialize it
2823      */
2824     pef_kvm_init(machine->cgs, &error_fatal);
2825 
2826     msi_nonbroken = true;
2827 
2828     QLIST_INIT(&spapr->phbs);
2829     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2830 
2831     /* Determine capabilities to run with */
2832     spapr_caps_init(spapr);
2833 
2834     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2835     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2836         /*
2837          * If the user explicitly requested a mode we should either
2838          * supply it, or fail completely (which we do below).  But if
2839          * it's not set explicitly, we reset our mode to something
2840          * that works
2841          */
2842         if (resize_hpt_err) {
2843             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2844             error_free(resize_hpt_err);
2845             resize_hpt_err = NULL;
2846         } else {
2847             spapr->resize_hpt = smc->resize_hpt_default;
2848         }
2849     }
2850 
2851     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2852 
2853     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2854         /*
2855          * User requested HPT resize, but this host can't supply it.  Bail out
2856          */
2857         error_report_err(resize_hpt_err);
2858         exit(1);
2859     }
2860     error_free(resize_hpt_err);
2861 
2862     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2863 
2864     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2865     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2866 
2867     /*
2868      * VSMT must be set in order to be able to compute VCPU ids, ie to
2869      * call spapr_max_server_number() or spapr_vcpu_id().
2870      */
2871     spapr_set_vsmt_mode(spapr, &error_fatal);
2872 
2873     /* Set up Interrupt Controller before we create the VCPUs */
2874     spapr_irq_init(spapr, &error_fatal);
2875 
2876     /* Set up containers for ibm,client-architecture-support negotiated options
2877      */
2878     spapr->ov5 = spapr_ovec_new();
2879     spapr->ov5_cas = spapr_ovec_new();
2880 
2881     if (smc->dr_lmb_enabled) {
2882         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2883         spapr_validate_node_memory(machine, &error_fatal);
2884     }
2885 
2886     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2887 
2888     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2889     if (!smc->pre_6_2_numa_affinity) {
2890         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2891     }
2892 
2893     /* advertise support for dedicated HP event source to guests */
2894     if (spapr->use_hotplug_event_source) {
2895         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2896     }
2897 
2898     /* advertise support for HPT resizing */
2899     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2900         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2901     }
2902 
2903     /* advertise support for ibm,dyamic-memory-v2 */
2904     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2905 
2906     /* advertise XIVE on POWER9 machines */
2907     if (spapr->irq->xive) {
2908         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2909     }
2910 
2911     /* init CPUs */
2912     spapr_init_cpus(spapr);
2913 
2914     /* Init numa_assoc_array */
2915     spapr_numa_associativity_init(spapr, machine);
2916 
2917     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2918         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2919                               spapr->max_compat_pvr)) {
2920         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2921         /* KVM and TCG always allow GTSE with radix... */
2922         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2923     }
2924     /* ... but not with hash (currently). */
2925 
2926     if (kvm_enabled()) {
2927         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2928         kvmppc_enable_logical_ci_hcalls();
2929         kvmppc_enable_set_mode_hcall();
2930 
2931         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2932         kvmppc_enable_clear_ref_mod_hcalls();
2933 
2934         /* Enable H_PAGE_INIT */
2935         kvmppc_enable_h_page_init();
2936     }
2937 
2938     /* map RAM */
2939     memory_region_add_subregion(sysmem, 0, machine->ram);
2940 
2941     /* initialize hotplug memory address space */
2942     if (machine->ram_size < machine->maxram_size) {
2943         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2944         hwaddr device_mem_base;
2945 
2946         /*
2947          * Limit the number of hotpluggable memory slots to half the number
2948          * slots that KVM supports, leaving the other half for PCI and other
2949          * devices. However ensure that number of slots doesn't drop below 32.
2950          */
2951         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2952                            SPAPR_MAX_RAM_SLOTS;
2953 
2954         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2955             max_memslots = SPAPR_MAX_RAM_SLOTS;
2956         }
2957         if (machine->ram_slots > max_memslots) {
2958             error_report("Specified number of memory slots %"
2959                          PRIu64" exceeds max supported %d",
2960                          machine->ram_slots, max_memslots);
2961             exit(1);
2962         }
2963 
2964         device_mem_base = ROUND_UP(machine->ram_size, SPAPR_DEVICE_MEM_ALIGN);
2965         machine_memory_devices_init(machine, device_mem_base, device_mem_size);
2966     }
2967 
2968     if (smc->dr_lmb_enabled) {
2969         spapr_create_lmb_dr_connectors(spapr);
2970     }
2971 
2972     if (mc->nvdimm_supported) {
2973         spapr_create_nvdimm_dr_connectors(spapr);
2974     }
2975 
2976     /* Set up RTAS event infrastructure */
2977     spapr_events_init(spapr);
2978 
2979     /* Set up the RTC RTAS interfaces */
2980     spapr_rtc_create(spapr);
2981 
2982     /* Set up VIO bus */
2983     spapr->vio_bus = spapr_vio_bus_init();
2984 
2985     for (i = 0; serial_hd(i); i++) {
2986         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2987     }
2988 
2989     /* We always have at least the nvram device on VIO */
2990     spapr_create_nvram(spapr);
2991 
2992     /*
2993      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2994      * connectors (described in root DT node's "ibm,drc-types" property)
2995      * are pre-initialized here. additional child connectors (such as
2996      * connectors for a PHBs PCI slots) are added as needed during their
2997      * parent's realization.
2998      */
2999     if (smc->dr_phb_enabled) {
3000         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
3001             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
3002         }
3003     }
3004 
3005     /* Set up PCI */
3006     spapr_pci_rtas_init();
3007 
3008     phb = spapr_create_default_phb();
3009 
3010     while ((nd = qemu_find_nic_info("spapr-vlan", true, "ibmveth"))) {
3011         spapr_vlan_create(spapr->vio_bus, nd);
3012     }
3013 
3014     pci_init_nic_devices(phb->bus, NULL);
3015 
3016     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
3017         spapr_vscsi_create(spapr->vio_bus);
3018     }
3019 
3020     /* Graphics */
3021     has_vga = spapr_vga_init(phb->bus, &error_fatal);
3022     if (has_vga) {
3023         spapr->want_stdout_path = !machine->enable_graphics;
3024         machine->usb |= defaults_enabled() && !machine->usb_disabled;
3025     } else {
3026         spapr->want_stdout_path = true;
3027     }
3028 
3029     if (machine->usb) {
3030         if (smc->use_ohci_by_default) {
3031             pci_create_simple(phb->bus, -1, "pci-ohci");
3032         } else {
3033             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
3034         }
3035 
3036         if (has_vga) {
3037             USBBus *usb_bus;
3038 
3039             usb_bus = USB_BUS(object_resolve_type_unambiguous(TYPE_USB_BUS,
3040                                                               &error_abort));
3041             usb_create_simple(usb_bus, "usb-kbd");
3042             usb_create_simple(usb_bus, "usb-mouse");
3043         }
3044     }
3045 
3046     if (kernel_filename) {
3047         uint64_t loaded_addr = 0;
3048 
3049         spapr->kernel_size = load_elf(kernel_filename, NULL,
3050                                       translate_kernel_address, spapr,
3051                                       NULL, &loaded_addr, NULL, NULL, 1,
3052                                       PPC_ELF_MACHINE, 0, 0);
3053         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
3054             spapr->kernel_size = load_elf(kernel_filename, NULL,
3055                                           translate_kernel_address, spapr,
3056                                           NULL, &loaded_addr, NULL, NULL, 0,
3057                                           PPC_ELF_MACHINE, 0, 0);
3058             spapr->kernel_le = spapr->kernel_size > 0;
3059         }
3060         if (spapr->kernel_size < 0) {
3061             error_report("error loading %s: %s", kernel_filename,
3062                          load_elf_strerror(spapr->kernel_size));
3063             exit(1);
3064         }
3065 
3066         if (spapr->kernel_addr != loaded_addr) {
3067             warn_report("spapr: kernel_addr changed from 0x%"PRIx64
3068                         " to 0x%"PRIx64,
3069                         spapr->kernel_addr, loaded_addr);
3070             spapr->kernel_addr = loaded_addr;
3071         }
3072 
3073         /* load initrd */
3074         if (initrd_filename) {
3075             /* Try to locate the initrd in the gap between the kernel
3076              * and the firmware. Add a bit of space just in case
3077              */
3078             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3079                                   + 0x1ffff) & ~0xffff;
3080             spapr->initrd_size = load_image_targphys(initrd_filename,
3081                                                      spapr->initrd_base,
3082                                                      load_limit
3083                                                      - spapr->initrd_base);
3084             if (spapr->initrd_size < 0) {
3085                 error_report("could not load initial ram disk '%s'",
3086                              initrd_filename);
3087                 exit(1);
3088             }
3089         }
3090     }
3091 
3092     /* FIXME: Should register things through the MachineState's qdev
3093      * interface, this is a legacy from the sPAPREnvironment structure
3094      * which predated MachineState but had a similar function */
3095     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3096     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3097                          &savevm_htab_handlers, spapr);
3098 
3099     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3100 
3101     qemu_register_boot_set(spapr_boot_set, spapr);
3102 
3103     /*
3104      * Nothing needs to be done to resume a suspended guest because
3105      * suspending does not change the machine state, so no need for
3106      * a ->wakeup method.
3107      */
3108     qemu_register_wakeup_support();
3109 
3110     if (kvm_enabled()) {
3111         /* to stop and start vmclock */
3112         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3113                                          &spapr->tb);
3114 
3115         kvmppc_spapr_enable_inkernel_multitce();
3116     }
3117 
3118     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3119     if (spapr->vof) {
3120         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3121         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3122     }
3123 
3124     spapr_watchdog_init(spapr);
3125 }
3126 
3127 #define DEFAULT_KVM_TYPE "auto"
3128 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3129 {
3130     /*
3131      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3132      * accommodate the 'HV' and 'PV' formats that exists in the
3133      * wild. The 'auto' mode is being introduced already as
3134      * lower-case, thus we don't need to bother checking for
3135      * "AUTO".
3136      */
3137     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3138         return 0;
3139     }
3140 
3141     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3142         return 1;
3143     }
3144 
3145     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3146         return 2;
3147     }
3148 
3149     error_report("Unknown kvm-type specified '%s'", vm_type);
3150     return -1;
3151 }
3152 
3153 /*
3154  * Implementation of an interface to adjust firmware path
3155  * for the bootindex property handling.
3156  */
3157 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3158                                    DeviceState *dev)
3159 {
3160 #define CAST(type, obj, name) \
3161     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3162     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3163     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3164     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3165     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3166 
3167     if (d && bus) {
3168         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3169         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3170         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3171 
3172         if (spapr) {
3173             /*
3174              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3175              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3176              * 0x8000 | (target << 8) | (bus << 5) | lun
3177              * (see the "Logical unit addressing format" table in SAM5)
3178              */
3179             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3180             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3181                                    (uint64_t)id << 48);
3182         } else if (virtio) {
3183             /*
3184              * We use SRP luns of the form 01000000 | (target << 8) | lun
3185              * in the top 32 bits of the 64-bit LUN
3186              * Note: the quote above is from SLOF and it is wrong,
3187              * the actual binding is:
3188              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3189              */
3190             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3191             if (d->lun >= 256) {
3192                 /* Use the LUN "flat space addressing method" */
3193                 id |= 0x4000;
3194             }
3195             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3196                                    (uint64_t)id << 32);
3197         } else if (usb) {
3198             /*
3199              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3200              * in the top 32 bits of the 64-bit LUN
3201              */
3202             unsigned usb_port = atoi(usb->port->path);
3203             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3204             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3205                                    (uint64_t)id << 32);
3206         }
3207     }
3208 
3209     /*
3210      * SLOF probes the USB devices, and if it recognizes that the device is a
3211      * storage device, it changes its name to "storage" instead of "usb-host",
3212      * and additionally adds a child node for the SCSI LUN, so the correct
3213      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3214      */
3215     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3216         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3217         if (usb_device_is_scsi_storage(usbdev)) {
3218             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3219         }
3220     }
3221 
3222     if (phb) {
3223         /* Replace "pci" with "pci@800000020000000" */
3224         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3225     }
3226 
3227     if (vsc) {
3228         /* Same logic as virtio above */
3229         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3230         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3231     }
3232 
3233     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3234         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3235         PCIDevice *pdev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3236         return g_strdup_printf("pci@%x", PCI_SLOT(pdev->devfn));
3237     }
3238 
3239     if (pcidev) {
3240         return spapr_pci_fw_dev_name(pcidev);
3241     }
3242 
3243     return NULL;
3244 }
3245 
3246 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3247 {
3248     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3249 
3250     return g_strdup(spapr->kvm_type);
3251 }
3252 
3253 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3254 {
3255     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3256 
3257     g_free(spapr->kvm_type);
3258     spapr->kvm_type = g_strdup(value);
3259 }
3260 
3261 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3262 {
3263     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3264 
3265     return spapr->use_hotplug_event_source;
3266 }
3267 
3268 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3269                                             Error **errp)
3270 {
3271     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3272 
3273     spapr->use_hotplug_event_source = value;
3274 }
3275 
3276 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3277 {
3278     return true;
3279 }
3280 
3281 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3282 {
3283     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3284 
3285     switch (spapr->resize_hpt) {
3286     case SPAPR_RESIZE_HPT_DEFAULT:
3287         return g_strdup("default");
3288     case SPAPR_RESIZE_HPT_DISABLED:
3289         return g_strdup("disabled");
3290     case SPAPR_RESIZE_HPT_ENABLED:
3291         return g_strdup("enabled");
3292     case SPAPR_RESIZE_HPT_REQUIRED:
3293         return g_strdup("required");
3294     }
3295     g_assert_not_reached();
3296 }
3297 
3298 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3299 {
3300     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3301 
3302     if (strcmp(value, "default") == 0) {
3303         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3304     } else if (strcmp(value, "disabled") == 0) {
3305         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3306     } else if (strcmp(value, "enabled") == 0) {
3307         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3308     } else if (strcmp(value, "required") == 0) {
3309         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3310     } else {
3311         error_setg(errp, "Bad value for \"resize-hpt\" property");
3312     }
3313 }
3314 
3315 static bool spapr_get_vof(Object *obj, Error **errp)
3316 {
3317     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3318 
3319     return spapr->vof != NULL;
3320 }
3321 
3322 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3323 {
3324     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3325 
3326     if (spapr->vof) {
3327         vof_cleanup(spapr->vof);
3328         g_free(spapr->vof);
3329         spapr->vof = NULL;
3330     }
3331     if (!value) {
3332         return;
3333     }
3334     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3335 }
3336 
3337 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3338 {
3339     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3340 
3341     if (spapr->irq == &spapr_irq_xics_legacy) {
3342         return g_strdup("legacy");
3343     } else if (spapr->irq == &spapr_irq_xics) {
3344         return g_strdup("xics");
3345     } else if (spapr->irq == &spapr_irq_xive) {
3346         return g_strdup("xive");
3347     } else if (spapr->irq == &spapr_irq_dual) {
3348         return g_strdup("dual");
3349     }
3350     g_assert_not_reached();
3351 }
3352 
3353 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3354 {
3355     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3356 
3357     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3358         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3359         return;
3360     }
3361 
3362     /* The legacy IRQ backend can not be set */
3363     if (strcmp(value, "xics") == 0) {
3364         spapr->irq = &spapr_irq_xics;
3365     } else if (strcmp(value, "xive") == 0) {
3366         spapr->irq = &spapr_irq_xive;
3367     } else if (strcmp(value, "dual") == 0) {
3368         spapr->irq = &spapr_irq_dual;
3369     } else {
3370         error_setg(errp, "Bad value for \"ic-mode\" property");
3371     }
3372 }
3373 
3374 static char *spapr_get_host_model(Object *obj, Error **errp)
3375 {
3376     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3377 
3378     return g_strdup(spapr->host_model);
3379 }
3380 
3381 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3382 {
3383     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3384 
3385     g_free(spapr->host_model);
3386     spapr->host_model = g_strdup(value);
3387 }
3388 
3389 static char *spapr_get_host_serial(Object *obj, Error **errp)
3390 {
3391     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3392 
3393     return g_strdup(spapr->host_serial);
3394 }
3395 
3396 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3397 {
3398     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3399 
3400     g_free(spapr->host_serial);
3401     spapr->host_serial = g_strdup(value);
3402 }
3403 
3404 static void spapr_instance_init(Object *obj)
3405 {
3406     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3407     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3408     MachineState *ms = MACHINE(spapr);
3409     MachineClass *mc = MACHINE_GET_CLASS(ms);
3410 
3411     /*
3412      * NVDIMM support went live in 5.1 without considering that, in
3413      * other archs, the user needs to enable NVDIMM support with the
3414      * 'nvdimm' machine option and the default behavior is NVDIMM
3415      * support disabled. It is too late to roll back to the standard
3416      * behavior without breaking 5.1 guests.
3417      */
3418     if (mc->nvdimm_supported) {
3419         ms->nvdimms_state->is_enabled = true;
3420     }
3421 
3422     spapr->htab_fd = -1;
3423     spapr->use_hotplug_event_source = true;
3424     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3425     object_property_add_str(obj, "kvm-type",
3426                             spapr_get_kvm_type, spapr_set_kvm_type);
3427     object_property_set_description(obj, "kvm-type",
3428                                     "Specifies the KVM virtualization mode (auto,"
3429                                     " hv, pr). Defaults to 'auto'. This mode will use"
3430                                     " any available KVM module loaded in the host,"
3431                                     " where kvm_hv takes precedence if both kvm_hv and"
3432                                     " kvm_pr are loaded.");
3433     object_property_add_bool(obj, "modern-hotplug-events",
3434                             spapr_get_modern_hotplug_events,
3435                             spapr_set_modern_hotplug_events);
3436     object_property_set_description(obj, "modern-hotplug-events",
3437                                     "Use dedicated hotplug event mechanism in"
3438                                     " place of standard EPOW events when possible"
3439                                     " (required for memory hot-unplug support)");
3440     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3441                             "Maximum permitted CPU compatibility mode");
3442 
3443     object_property_add_str(obj, "resize-hpt",
3444                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3445     object_property_set_description(obj, "resize-hpt",
3446                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3447     object_property_add_uint32_ptr(obj, "vsmt",
3448                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3449     object_property_set_description(obj, "vsmt",
3450                                     "Virtual SMT: KVM behaves as if this were"
3451                                     " the host's SMT mode");
3452 
3453     object_property_add_bool(obj, "vfio-no-msix-emulation",
3454                              spapr_get_msix_emulation, NULL);
3455 
3456     object_property_add_uint64_ptr(obj, "kernel-addr",
3457                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3458     object_property_set_description(obj, "kernel-addr",
3459                                     stringify(KERNEL_LOAD_ADDR)
3460                                     " for -kernel is the default");
3461     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3462 
3463     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3464     object_property_set_description(obj, "x-vof",
3465                                     "Enable Virtual Open Firmware (experimental)");
3466 
3467     /* The machine class defines the default interrupt controller mode */
3468     spapr->irq = smc->irq;
3469     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3470                             spapr_set_ic_mode);
3471     object_property_set_description(obj, "ic-mode",
3472                  "Specifies the interrupt controller mode (xics, xive, dual)");
3473 
3474     object_property_add_str(obj, "host-model",
3475         spapr_get_host_model, spapr_set_host_model);
3476     object_property_set_description(obj, "host-model",
3477         "Host model to advertise in guest device tree");
3478     object_property_add_str(obj, "host-serial",
3479         spapr_get_host_serial, spapr_set_host_serial);
3480     object_property_set_description(obj, "host-serial",
3481         "Host serial number to advertise in guest device tree");
3482 }
3483 
3484 static void spapr_machine_finalizefn(Object *obj)
3485 {
3486     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3487 
3488     g_free(spapr->kvm_type);
3489 }
3490 
3491 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3492 {
3493     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3494     PowerPCCPU *cpu = POWERPC_CPU(cs);
3495     CPUPPCState *env = &cpu->env;
3496 
3497     cpu_synchronize_state(cs);
3498     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3499     if (spapr->fwnmi_system_reset_addr != -1) {
3500         uint64_t rtas_addr, addr;
3501 
3502         /* get rtas addr from fdt */
3503         rtas_addr = spapr_get_rtas_addr();
3504         if (!rtas_addr) {
3505             qemu_system_guest_panicked(NULL);
3506             return;
3507         }
3508 
3509         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3510         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3511         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3512         env->gpr[3] = addr;
3513     }
3514     ppc_cpu_do_system_reset(cs);
3515     if (spapr->fwnmi_system_reset_addr != -1) {
3516         env->nip = spapr->fwnmi_system_reset_addr;
3517     }
3518 }
3519 
3520 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3521 {
3522     CPUState *cs;
3523 
3524     CPU_FOREACH(cs) {
3525         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3526     }
3527 }
3528 
3529 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3530                           void *fdt, int *fdt_start_offset, Error **errp)
3531 {
3532     uint64_t addr;
3533     uint32_t node;
3534 
3535     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3536     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3537                                     &error_abort);
3538     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3539                                              SPAPR_MEMORY_BLOCK_SIZE);
3540     return 0;
3541 }
3542 
3543 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3544                            bool dedicated_hp_event_source)
3545 {
3546     SpaprDrc *drc;
3547     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3548     int i;
3549     uint64_t addr = addr_start;
3550     bool hotplugged = spapr_drc_hotplugged(dev);
3551 
3552     for (i = 0; i < nr_lmbs; i++) {
3553         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3554                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3555         g_assert(drc);
3556 
3557         /*
3558          * memory_device_get_free_addr() provided a range of free addresses
3559          * that doesn't overlap with any existing mapping at pre-plug. The
3560          * corresponding LMB DRCs are thus assumed to be all attachable.
3561          */
3562         spapr_drc_attach(drc, dev);
3563         if (!hotplugged) {
3564             spapr_drc_reset(drc);
3565         }
3566         addr += SPAPR_MEMORY_BLOCK_SIZE;
3567     }
3568     /* send hotplug notification to the
3569      * guest only in case of hotplugged memory
3570      */
3571     if (hotplugged) {
3572         if (dedicated_hp_event_source) {
3573             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3574                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3575             g_assert(drc);
3576             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3577                                                    nr_lmbs,
3578                                                    spapr_drc_index(drc));
3579         } else {
3580             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3581                                            nr_lmbs);
3582         }
3583     }
3584 }
3585 
3586 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3587 {
3588     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3589     PCDIMMDevice *dimm = PC_DIMM(dev);
3590     uint64_t size, addr;
3591     int64_t slot;
3592     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3593 
3594     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3595 
3596     pc_dimm_plug(dimm, MACHINE(ms));
3597 
3598     if (!is_nvdimm) {
3599         addr = object_property_get_uint(OBJECT(dimm),
3600                                         PC_DIMM_ADDR_PROP, &error_abort);
3601         spapr_add_lmbs(dev, addr, size,
3602                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3603     } else {
3604         slot = object_property_get_int(OBJECT(dimm),
3605                                        PC_DIMM_SLOT_PROP, &error_abort);
3606         /* We should have valid slot number at this point */
3607         g_assert(slot >= 0);
3608         spapr_add_nvdimm(dev, slot);
3609     }
3610 }
3611 
3612 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3613                                   Error **errp)
3614 {
3615     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3616     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3617     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3618     PCDIMMDevice *dimm = PC_DIMM(dev);
3619     Error *local_err = NULL;
3620     uint64_t size;
3621     Object *memdev;
3622     hwaddr pagesize;
3623 
3624     if (!smc->dr_lmb_enabled) {
3625         error_setg(errp, "Memory hotplug not supported for this machine");
3626         return;
3627     }
3628 
3629     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3630     if (local_err) {
3631         error_propagate(errp, local_err);
3632         return;
3633     }
3634 
3635     if (is_nvdimm) {
3636         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3637             return;
3638         }
3639     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3640         error_setg(errp, "Hotplugged memory size must be a multiple of "
3641                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3642         return;
3643     }
3644 
3645     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3646                                       &error_abort);
3647     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3648     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3649         return;
3650     }
3651 
3652     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3653 }
3654 
3655 struct SpaprDimmState {
3656     PCDIMMDevice *dimm;
3657     uint32_t nr_lmbs;
3658     QTAILQ_ENTRY(SpaprDimmState) next;
3659 };
3660 
3661 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3662                                                        PCDIMMDevice *dimm)
3663 {
3664     SpaprDimmState *dimm_state = NULL;
3665 
3666     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3667         if (dimm_state->dimm == dimm) {
3668             break;
3669         }
3670     }
3671     return dimm_state;
3672 }
3673 
3674 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3675                                                       uint32_t nr_lmbs,
3676                                                       PCDIMMDevice *dimm)
3677 {
3678     SpaprDimmState *ds = NULL;
3679 
3680     /*
3681      * If this request is for a DIMM whose removal had failed earlier
3682      * (due to guest's refusal to remove the LMBs), we would have this
3683      * dimm already in the pending_dimm_unplugs list. In that
3684      * case don't add again.
3685      */
3686     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3687     if (!ds) {
3688         ds = g_new0(SpaprDimmState, 1);
3689         ds->nr_lmbs = nr_lmbs;
3690         ds->dimm = dimm;
3691         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3692     }
3693     return ds;
3694 }
3695 
3696 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3697                                               SpaprDimmState *dimm_state)
3698 {
3699     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3700     g_free(dimm_state);
3701 }
3702 
3703 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3704                                                         PCDIMMDevice *dimm)
3705 {
3706     SpaprDrc *drc;
3707     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3708                                                   &error_abort);
3709     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3710     uint32_t avail_lmbs = 0;
3711     uint64_t addr_start, addr;
3712     int i;
3713 
3714     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3715                                           &error_abort);
3716 
3717     addr = addr_start;
3718     for (i = 0; i < nr_lmbs; i++) {
3719         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3720                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3721         g_assert(drc);
3722         if (drc->dev) {
3723             avail_lmbs++;
3724         }
3725         addr += SPAPR_MEMORY_BLOCK_SIZE;
3726     }
3727 
3728     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3729 }
3730 
3731 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3732 {
3733     SpaprDimmState *ds;
3734     PCDIMMDevice *dimm;
3735     SpaprDrc *drc;
3736     uint32_t nr_lmbs;
3737     uint64_t size, addr_start, addr;
3738     g_autofree char *qapi_error = NULL;
3739     int i;
3740 
3741     if (!dev) {
3742         return;
3743     }
3744 
3745     dimm = PC_DIMM(dev);
3746     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3747 
3748     /*
3749      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3750      * unplug state, but one of its DRC is marked as unplug_requested.
3751      * This is bad and weird enough to g_assert() out.
3752      */
3753     g_assert(ds);
3754 
3755     spapr_pending_dimm_unplugs_remove(spapr, ds);
3756 
3757     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3758     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3759 
3760     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3761                                           &error_abort);
3762 
3763     addr = addr_start;
3764     for (i = 0; i < nr_lmbs; i++) {
3765         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3766                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3767         g_assert(drc);
3768 
3769         drc->unplug_requested = false;
3770         addr += SPAPR_MEMORY_BLOCK_SIZE;
3771     }
3772 
3773     /*
3774      * Tell QAPI that something happened and the memory
3775      * hotunplug wasn't successful. Keep sending
3776      * MEM_UNPLUG_ERROR even while sending
3777      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3778      * MEM_UNPLUG_ERROR is due.
3779      */
3780     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3781                                  "for device %s", dev->id);
3782 
3783     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3784 
3785     qapi_event_send_device_unplug_guest_error(dev->id,
3786                                               dev->canonical_path);
3787 }
3788 
3789 /* Callback to be called during DRC release. */
3790 void spapr_lmb_release(DeviceState *dev)
3791 {
3792     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3793     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3794     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3795 
3796     /* This information will get lost if a migration occurs
3797      * during the unplug process. In this case recover it. */
3798     if (ds == NULL) {
3799         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3800         g_assert(ds);
3801         /* The DRC being examined by the caller at least must be counted */
3802         g_assert(ds->nr_lmbs);
3803     }
3804 
3805     if (--ds->nr_lmbs) {
3806         return;
3807     }
3808 
3809     /*
3810      * Now that all the LMBs have been removed by the guest, call the
3811      * unplug handler chain. This can never fail.
3812      */
3813     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3814     object_unparent(OBJECT(dev));
3815 }
3816 
3817 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3818 {
3819     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3820     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3821 
3822     /* We really shouldn't get this far without anything to unplug */
3823     g_assert(ds);
3824 
3825     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3826     qdev_unrealize(dev);
3827     spapr_pending_dimm_unplugs_remove(spapr, ds);
3828 }
3829 
3830 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3831                                         DeviceState *dev, Error **errp)
3832 {
3833     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3834     PCDIMMDevice *dimm = PC_DIMM(dev);
3835     uint32_t nr_lmbs;
3836     uint64_t size, addr_start, addr;
3837     int i;
3838     SpaprDrc *drc;
3839 
3840     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3841         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3842         return;
3843     }
3844 
3845     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3846     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3847 
3848     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3849                                           &error_abort);
3850 
3851     /*
3852      * An existing pending dimm state for this DIMM means that there is an
3853      * unplug operation in progress, waiting for the spapr_lmb_release
3854      * callback to complete the job (BQL can't cover that far). In this case,
3855      * bail out to avoid detaching DRCs that were already released.
3856      */
3857     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3858         error_setg(errp, "Memory unplug already in progress for device %s",
3859                    dev->id);
3860         return;
3861     }
3862 
3863     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3864 
3865     addr = addr_start;
3866     for (i = 0; i < nr_lmbs; i++) {
3867         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3868                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3869         g_assert(drc);
3870 
3871         spapr_drc_unplug_request(drc);
3872         addr += SPAPR_MEMORY_BLOCK_SIZE;
3873     }
3874 
3875     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3876                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3877     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3878                                               nr_lmbs, spapr_drc_index(drc));
3879 }
3880 
3881 /* Callback to be called during DRC release. */
3882 void spapr_core_release(DeviceState *dev)
3883 {
3884     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3885 
3886     /* Call the unplug handler chain. This can never fail. */
3887     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3888     object_unparent(OBJECT(dev));
3889 }
3890 
3891 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3892 {
3893     MachineState *ms = MACHINE(hotplug_dev);
3894     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3895     CPUCore *cc = CPU_CORE(dev);
3896     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3897 
3898     if (smc->pre_2_10_has_unused_icps) {
3899         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3900         int i;
3901 
3902         for (i = 0; i < cc->nr_threads; i++) {
3903             CPUState *cs = CPU(sc->threads[i]);
3904 
3905             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3906         }
3907     }
3908 
3909     assert(core_slot);
3910     core_slot->cpu = NULL;
3911     qdev_unrealize(dev);
3912 }
3913 
3914 static
3915 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3916                                Error **errp)
3917 {
3918     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3919     int index;
3920     SpaprDrc *drc;
3921     CPUCore *cc = CPU_CORE(dev);
3922 
3923     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3924         error_setg(errp, "Unable to find CPU core with core-id: %d",
3925                    cc->core_id);
3926         return;
3927     }
3928     if (index == 0) {
3929         error_setg(errp, "Boot CPU core may not be unplugged");
3930         return;
3931     }
3932 
3933     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3934                           spapr_vcpu_id(spapr, cc->core_id));
3935     g_assert(drc);
3936 
3937     if (!spapr_drc_unplug_requested(drc)) {
3938         spapr_drc_unplug_request(drc);
3939     }
3940 
3941     /*
3942      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3943      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3944      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3945      * attempt (e.g. the kernel will refuse to remove the last online
3946      * CPU), we will never attempt it again because unplug_requested
3947      * will still be 'true' in that case.
3948      */
3949     spapr_hotplug_req_remove_by_index(drc);
3950 }
3951 
3952 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3953                            void *fdt, int *fdt_start_offset, Error **errp)
3954 {
3955     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3956     CPUState *cs = CPU(core->threads[0]);
3957     PowerPCCPU *cpu = POWERPC_CPU(cs);
3958     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3959     int id = spapr_get_vcpu_id(cpu);
3960     g_autofree char *nodename = NULL;
3961     int offset;
3962 
3963     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3964     offset = fdt_add_subnode(fdt, 0, nodename);
3965 
3966     spapr_dt_cpu(cs, fdt, offset, spapr);
3967 
3968     /*
3969      * spapr_dt_cpu() does not fill the 'name' property in the
3970      * CPU node. The function is called during boot process, before
3971      * and after CAS, and overwriting the 'name' property written
3972      * by SLOF is not allowed.
3973      *
3974      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3975      * CPUs more compatible with the coldplugged ones, which have
3976      * the 'name' property. Linux Kernel also relies on this
3977      * property to identify CPU nodes.
3978      */
3979     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3980 
3981     *fdt_start_offset = offset;
3982     return 0;
3983 }
3984 
3985 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3986 {
3987     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3988     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3989     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3990     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3991     CPUCore *cc = CPU_CORE(dev);
3992     CPUState *cs;
3993     SpaprDrc *drc;
3994     CPUArchId *core_slot;
3995     int index;
3996     bool hotplugged = spapr_drc_hotplugged(dev);
3997     int i;
3998 
3999     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4000     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
4001 
4002     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
4003                           spapr_vcpu_id(spapr, cc->core_id));
4004 
4005     g_assert(drc || !mc->has_hotpluggable_cpus);
4006 
4007     if (drc) {
4008         /*
4009          * spapr_core_pre_plug() already buys us this is a brand new
4010          * core being plugged into a free slot. Nothing should already
4011          * be attached to the corresponding DRC.
4012          */
4013         spapr_drc_attach(drc, dev);
4014 
4015         if (hotplugged) {
4016             /*
4017              * Send hotplug notification interrupt to the guest only
4018              * in case of hotplugged CPUs.
4019              */
4020             spapr_hotplug_req_add_by_index(drc);
4021         } else {
4022             spapr_drc_reset(drc);
4023         }
4024     }
4025 
4026     core_slot->cpu = OBJECT(dev);
4027 
4028     /*
4029      * Set compatibility mode to match the boot CPU, which was either set
4030      * by the machine reset code or by CAS. This really shouldn't fail at
4031      * this point.
4032      */
4033     if (hotplugged) {
4034         for (i = 0; i < cc->nr_threads; i++) {
4035             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
4036                            &error_abort);
4037         }
4038     }
4039 
4040     if (smc->pre_2_10_has_unused_icps) {
4041         for (i = 0; i < cc->nr_threads; i++) {
4042             cs = CPU(core->threads[i]);
4043             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
4044         }
4045     }
4046 }
4047 
4048 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4049                                 Error **errp)
4050 {
4051     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
4052     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
4053     CPUCore *cc = CPU_CORE(dev);
4054     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
4055     const char *type = object_get_typename(OBJECT(dev));
4056     CPUArchId *core_slot;
4057     int index;
4058     unsigned int smp_threads = machine->smp.threads;
4059 
4060     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
4061         error_setg(errp, "CPU hotplug not supported for this machine");
4062         return;
4063     }
4064 
4065     if (strcmp(base_core_type, type)) {
4066         error_setg(errp, "CPU core type should be %s", base_core_type);
4067         return;
4068     }
4069 
4070     if (cc->core_id % smp_threads) {
4071         error_setg(errp, "invalid core id %d", cc->core_id);
4072         return;
4073     }
4074 
4075     /*
4076      * In general we should have homogeneous threads-per-core, but old
4077      * (pre hotplug support) machine types allow the last core to have
4078      * reduced threads as a compatibility hack for when we allowed
4079      * total vcpus not a multiple of threads-per-core.
4080      */
4081     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
4082         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4083                    smp_threads);
4084         return;
4085     }
4086 
4087     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4088     if (!core_slot) {
4089         error_setg(errp, "core id %d out of range", cc->core_id);
4090         return;
4091     }
4092 
4093     if (core_slot->cpu) {
4094         error_setg(errp, "core %d already populated", cc->core_id);
4095         return;
4096     }
4097 
4098     numa_cpu_pre_plug(core_slot, dev, errp);
4099 }
4100 
4101 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4102                           void *fdt, int *fdt_start_offset, Error **errp)
4103 {
4104     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4105     int intc_phandle;
4106 
4107     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4108     if (intc_phandle <= 0) {
4109         return -1;
4110     }
4111 
4112     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4113         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4114         return -1;
4115     }
4116 
4117     /* generally SLOF creates these, for hotplug it's up to QEMU */
4118     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4119 
4120     return 0;
4121 }
4122 
4123 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4124                                Error **errp)
4125 {
4126     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4127     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4128     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4129     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4130     SpaprDrc *drc;
4131 
4132     if (dev->hotplugged && !smc->dr_phb_enabled) {
4133         error_setg(errp, "PHB hotplug not supported for this machine");
4134         return false;
4135     }
4136 
4137     if (sphb->index == (uint32_t)-1) {
4138         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4139         return false;
4140     }
4141 
4142     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4143     if (drc && drc->dev) {
4144         error_setg(errp, "PHB %d already attached", sphb->index);
4145         return false;
4146     }
4147 
4148     /*
4149      * This will check that sphb->index doesn't exceed the maximum number of
4150      * PHBs for the current machine type.
4151      */
4152     return
4153         smc->phb_placement(spapr, sphb->index,
4154                            &sphb->buid, &sphb->io_win_addr,
4155                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4156                            windows_supported, sphb->dma_liobn,
4157                            errp);
4158 }
4159 
4160 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4161 {
4162     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4163     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4164     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4165     SpaprDrc *drc;
4166     bool hotplugged = spapr_drc_hotplugged(dev);
4167 
4168     if (!smc->dr_phb_enabled) {
4169         return;
4170     }
4171 
4172     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4173     /* hotplug hooks should check it's enabled before getting this far */
4174     assert(drc);
4175 
4176     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4177     spapr_drc_attach(drc, dev);
4178 
4179     if (hotplugged) {
4180         spapr_hotplug_req_add_by_index(drc);
4181     } else {
4182         spapr_drc_reset(drc);
4183     }
4184 }
4185 
4186 void spapr_phb_release(DeviceState *dev)
4187 {
4188     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4189 
4190     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4191     object_unparent(OBJECT(dev));
4192 }
4193 
4194 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4195 {
4196     qdev_unrealize(dev);
4197 }
4198 
4199 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4200                                      DeviceState *dev, Error **errp)
4201 {
4202     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4203     SpaprDrc *drc;
4204 
4205     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4206     assert(drc);
4207 
4208     if (!spapr_drc_unplug_requested(drc)) {
4209         spapr_drc_unplug_request(drc);
4210         spapr_hotplug_req_remove_by_index(drc);
4211     } else {
4212         error_setg(errp,
4213                    "PCI Host Bridge unplug already in progress for device %s",
4214                    dev->id);
4215     }
4216 }
4217 
4218 static
4219 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4220                               Error **errp)
4221 {
4222     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4223 
4224     if (spapr->tpm_proxy != NULL) {
4225         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4226         return false;
4227     }
4228 
4229     return true;
4230 }
4231 
4232 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4233 {
4234     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4235     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4236 
4237     /* Already checked in spapr_tpm_proxy_pre_plug() */
4238     g_assert(spapr->tpm_proxy == NULL);
4239 
4240     spapr->tpm_proxy = tpm_proxy;
4241 }
4242 
4243 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4244 {
4245     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4246 
4247     qdev_unrealize(dev);
4248     object_unparent(OBJECT(dev));
4249     spapr->tpm_proxy = NULL;
4250 }
4251 
4252 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4253                                       DeviceState *dev, Error **errp)
4254 {
4255     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4256         spapr_memory_plug(hotplug_dev, dev);
4257     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4258         spapr_core_plug(hotplug_dev, dev);
4259     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4260         spapr_phb_plug(hotplug_dev, dev);
4261     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4262         spapr_tpm_proxy_plug(hotplug_dev, dev);
4263     }
4264 }
4265 
4266 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4267                                         DeviceState *dev, Error **errp)
4268 {
4269     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4270         spapr_memory_unplug(hotplug_dev, dev);
4271     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4272         spapr_core_unplug(hotplug_dev, dev);
4273     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4274         spapr_phb_unplug(hotplug_dev, dev);
4275     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4276         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4277     }
4278 }
4279 
4280 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4281 {
4282     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4283         /*
4284          * CAS will process all pending unplug requests.
4285          *
4286          * HACK: a guest could theoretically have cleared all bits in OV5,
4287          * but none of the guests we care for do.
4288          */
4289         spapr_ovec_empty(spapr->ov5_cas);
4290 }
4291 
4292 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4293                                                 DeviceState *dev, Error **errp)
4294 {
4295     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4296     MachineClass *mc = MACHINE_GET_CLASS(sms);
4297     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4298 
4299     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4300         if (spapr_memory_hot_unplug_supported(sms)) {
4301             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4302         } else {
4303             error_setg(errp, "Memory hot unplug not supported for this guest");
4304         }
4305     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4306         if (!mc->has_hotpluggable_cpus) {
4307             error_setg(errp, "CPU hot unplug not supported on this machine");
4308             return;
4309         }
4310         spapr_core_unplug_request(hotplug_dev, dev, errp);
4311     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4312         if (!smc->dr_phb_enabled) {
4313             error_setg(errp, "PHB hot unplug not supported on this machine");
4314             return;
4315         }
4316         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4317     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4318         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4319     }
4320 }
4321 
4322 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4323                                           DeviceState *dev, Error **errp)
4324 {
4325     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4326         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4327     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4328         spapr_core_pre_plug(hotplug_dev, dev, errp);
4329     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4330         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4331     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4332         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4333     }
4334 }
4335 
4336 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4337                                                  DeviceState *dev)
4338 {
4339     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4340         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4341         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4342         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4343         return HOTPLUG_HANDLER(machine);
4344     }
4345     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4346         PCIDevice *pcidev = PCI_DEVICE(dev);
4347         PCIBus *root = pci_device_root_bus(pcidev);
4348         SpaprPhbState *phb =
4349             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4350                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4351 
4352         if (phb) {
4353             return HOTPLUG_HANDLER(phb);
4354         }
4355     }
4356     return NULL;
4357 }
4358 
4359 static CpuInstanceProperties
4360 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4361 {
4362     CPUArchId *core_slot;
4363     MachineClass *mc = MACHINE_GET_CLASS(machine);
4364 
4365     /* make sure possible_cpu are initialized */
4366     mc->possible_cpu_arch_ids(machine);
4367     /* get CPU core slot containing thread that matches cpu_index */
4368     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4369     assert(core_slot);
4370     return core_slot->props;
4371 }
4372 
4373 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4374 {
4375     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4376 }
4377 
4378 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4379 {
4380     int i;
4381     unsigned int smp_threads = machine->smp.threads;
4382     unsigned int smp_cpus = machine->smp.cpus;
4383     const char *core_type;
4384     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4385     MachineClass *mc = MACHINE_GET_CLASS(machine);
4386 
4387     if (!mc->has_hotpluggable_cpus) {
4388         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4389     }
4390     if (machine->possible_cpus) {
4391         assert(machine->possible_cpus->len == spapr_max_cores);
4392         return machine->possible_cpus;
4393     }
4394 
4395     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4396     if (!core_type) {
4397         error_report("Unable to find sPAPR CPU Core definition");
4398         exit(1);
4399     }
4400 
4401     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4402                              sizeof(CPUArchId) * spapr_max_cores);
4403     machine->possible_cpus->len = spapr_max_cores;
4404     for (i = 0; i < machine->possible_cpus->len; i++) {
4405         int core_id = i * smp_threads;
4406 
4407         machine->possible_cpus->cpus[i].type = core_type;
4408         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4409         machine->possible_cpus->cpus[i].arch_id = core_id;
4410         machine->possible_cpus->cpus[i].props.has_core_id = true;
4411         machine->possible_cpus->cpus[i].props.core_id = core_id;
4412     }
4413     return machine->possible_cpus;
4414 }
4415 
4416 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4417                                 uint64_t *buid, hwaddr *pio,
4418                                 hwaddr *mmio32, hwaddr *mmio64,
4419                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4420 {
4421     /*
4422      * New-style PHB window placement.
4423      *
4424      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4425      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4426      * windows.
4427      *
4428      * Some guest kernels can't work with MMIO windows above 1<<46
4429      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4430      *
4431      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4432      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4433      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4434      * 1TiB 64-bit MMIO windows for each PHB.
4435      */
4436     const uint64_t base_buid = 0x800000020000000ULL;
4437     int i;
4438 
4439     /* Sanity check natural alignments */
4440     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4441     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4442     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4443     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4444     /* Sanity check bounds */
4445     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4446                       SPAPR_PCI_MEM32_WIN_SIZE);
4447     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4448                       SPAPR_PCI_MEM64_WIN_SIZE);
4449 
4450     if (index >= SPAPR_MAX_PHBS) {
4451         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4452                    SPAPR_MAX_PHBS - 1);
4453         return false;
4454     }
4455 
4456     *buid = base_buid + index;
4457     for (i = 0; i < n_dma; ++i) {
4458         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4459     }
4460 
4461     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4462     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4463     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4464     return true;
4465 }
4466 
4467 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4468 {
4469     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4470 
4471     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4472 }
4473 
4474 static void spapr_ics_resend(XICSFabric *dev)
4475 {
4476     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4477 
4478     ics_resend(spapr->ics);
4479 }
4480 
4481 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4482 {
4483     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4484 
4485     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4486 }
4487 
4488 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4489                                  Monitor *mon)
4490 {
4491     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4492 
4493     spapr_irq_print_info(spapr, mon);
4494     monitor_printf(mon, "irqchip: %s\n",
4495                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4496 }
4497 
4498 /*
4499  * This is a XIVE only operation
4500  */
4501 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4502                            uint8_t nvt_blk, uint32_t nvt_idx,
4503                            bool cam_ignore, uint8_t priority,
4504                            uint32_t logic_serv, XiveTCTXMatch *match)
4505 {
4506     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4507     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4508     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4509     int count;
4510 
4511     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4512                            priority, logic_serv, match);
4513     if (count < 0) {
4514         return count;
4515     }
4516 
4517     /*
4518      * When we implement the save and restore of the thread interrupt
4519      * contexts in the enter/exit CPU handlers of the machine and the
4520      * escalations in QEMU, we should be able to handle non dispatched
4521      * vCPUs.
4522      *
4523      * Until this is done, the sPAPR machine should find at least one
4524      * matching context always.
4525      */
4526     if (count == 0) {
4527         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4528                       nvt_blk, nvt_idx);
4529     }
4530 
4531     return count;
4532 }
4533 
4534 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4535 {
4536     return cpu->vcpu_id;
4537 }
4538 
4539 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4540 {
4541     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4542     MachineState *ms = MACHINE(spapr);
4543     int vcpu_id;
4544 
4545     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4546 
4547     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4548         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4549         error_append_hint(errp, "Adjust the number of cpus to %d "
4550                           "or try to raise the number of threads per core\n",
4551                           vcpu_id * ms->smp.threads / spapr->vsmt);
4552         return false;
4553     }
4554 
4555     cpu->vcpu_id = vcpu_id;
4556     return true;
4557 }
4558 
4559 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4560 {
4561     CPUState *cs;
4562 
4563     CPU_FOREACH(cs) {
4564         PowerPCCPU *cpu = POWERPC_CPU(cs);
4565 
4566         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4567             return cpu;
4568         }
4569     }
4570 
4571     return NULL;
4572 }
4573 
4574 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4575 {
4576     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4577 
4578     return spapr_cpu->in_nested;
4579 }
4580 
4581 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4582 {
4583     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4584 
4585     /* These are only called by TCG, KVM maintains dispatch state */
4586 
4587     spapr_cpu->prod = false;
4588     if (spapr_cpu->vpa_addr) {
4589         CPUState *cs = CPU(cpu);
4590         uint32_t dispatch;
4591 
4592         dispatch = ldl_be_phys(cs->as,
4593                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4594         dispatch++;
4595         if ((dispatch & 1) != 0) {
4596             qemu_log_mask(LOG_GUEST_ERROR,
4597                           "VPA: incorrect dispatch counter value for "
4598                           "dispatched partition %u, correcting.\n", dispatch);
4599             dispatch++;
4600         }
4601         stl_be_phys(cs->as,
4602                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4603     }
4604 }
4605 
4606 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4607 {
4608     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4609 
4610     if (spapr_cpu->vpa_addr) {
4611         CPUState *cs = CPU(cpu);
4612         uint32_t dispatch;
4613 
4614         dispatch = ldl_be_phys(cs->as,
4615                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4616         dispatch++;
4617         if ((dispatch & 1) != 1) {
4618             qemu_log_mask(LOG_GUEST_ERROR,
4619                           "VPA: incorrect dispatch counter value for "
4620                           "preempted partition %u, correcting.\n", dispatch);
4621             dispatch++;
4622         }
4623         stl_be_phys(cs->as,
4624                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4625     }
4626 }
4627 
4628 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4629 {
4630     MachineClass *mc = MACHINE_CLASS(oc);
4631     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4632     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4633     NMIClass *nc = NMI_CLASS(oc);
4634     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4635     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4636     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4637     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4638     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4639     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4640 
4641     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4642     mc->ignore_boot_device_suffixes = true;
4643 
4644     /*
4645      * We set up the default / latest behaviour here.  The class_init
4646      * functions for the specific versioned machine types can override
4647      * these details for backwards compatibility
4648      */
4649     mc->init = spapr_machine_init;
4650     mc->reset = spapr_machine_reset;
4651     mc->block_default_type = IF_SCSI;
4652 
4653     /*
4654      * While KVM determines max cpus in kvm_init() using kvm_max_vcpus(),
4655      * In TCG the limit is restricted by the range of CPU IPIs available.
4656      */
4657     mc->max_cpus = SPAPR_IRQ_NR_IPIS;
4658 
4659     mc->no_parallel = 1;
4660     mc->default_boot_order = "";
4661     mc->default_ram_size = 512 * MiB;
4662     mc->default_ram_id = "ppc_spapr.ram";
4663     mc->default_display = "std";
4664     mc->kvm_type = spapr_kvm_type;
4665     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4666     mc->pci_allow_0_address = true;
4667     assert(!mc->get_hotplug_handler);
4668     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4669     hc->pre_plug = spapr_machine_device_pre_plug;
4670     hc->plug = spapr_machine_device_plug;
4671     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4672     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4673     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4674     hc->unplug_request = spapr_machine_device_unplug_request;
4675     hc->unplug = spapr_machine_device_unplug;
4676 
4677     smc->dr_lmb_enabled = true;
4678     smc->update_dt_enabled = true;
4679     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power10_v2.0");
4680     mc->has_hotpluggable_cpus = true;
4681     mc->nvdimm_supported = true;
4682     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4683     fwc->get_dev_path = spapr_get_fw_dev_path;
4684     nc->nmi_monitor_handler = spapr_nmi;
4685     smc->phb_placement = spapr_phb_placement;
4686     vhc->cpu_in_nested = spapr_cpu_in_nested;
4687     vhc->deliver_hv_excp = spapr_exit_nested;
4688     vhc->hypercall = emulate_spapr_hypercall;
4689     vhc->hpt_mask = spapr_hpt_mask;
4690     vhc->map_hptes = spapr_map_hptes;
4691     vhc->unmap_hptes = spapr_unmap_hptes;
4692     vhc->hpte_set_c = spapr_hpte_set_c;
4693     vhc->hpte_set_r = spapr_hpte_set_r;
4694     vhc->get_pate = spapr_get_pate;
4695     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4696     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4697     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4698     xic->ics_get = spapr_ics_get;
4699     xic->ics_resend = spapr_ics_resend;
4700     xic->icp_get = spapr_icp_get;
4701     ispc->print_info = spapr_pic_print_info;
4702     /* Force NUMA node memory size to be a multiple of
4703      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4704      * in which LMBs are represented and hot-added
4705      */
4706     mc->numa_mem_align_shift = 28;
4707     mc->auto_enable_numa = true;
4708 
4709     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4710     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4711     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4712     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4713     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4714     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4715     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4716     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4717     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4718     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4719     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4720     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4721 
4722     /*
4723      * This cap specifies whether the AIL 3 mode for
4724      * H_SET_RESOURCE is supported. The default is modified
4725      * by default_caps_with_cpu().
4726      */
4727     smc->default_caps.caps[SPAPR_CAP_AIL_MODE_3] = SPAPR_CAP_ON;
4728     spapr_caps_add_properties(smc);
4729     smc->irq = &spapr_irq_dual;
4730     smc->dr_phb_enabled = true;
4731     smc->linux_pci_probe = true;
4732     smc->smp_threads_vsmt = true;
4733     smc->nr_xirqs = SPAPR_NR_XIRQS;
4734     xfc->match_nvt = spapr_match_nvt;
4735     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4736     vmc->quiesce = spapr_vof_quiesce;
4737     vmc->setprop = spapr_vof_setprop;
4738 }
4739 
4740 static const TypeInfo spapr_machine_info = {
4741     .name          = TYPE_SPAPR_MACHINE,
4742     .parent        = TYPE_MACHINE,
4743     .abstract      = true,
4744     .instance_size = sizeof(SpaprMachineState),
4745     .instance_init = spapr_instance_init,
4746     .instance_finalize = spapr_machine_finalizefn,
4747     .class_size    = sizeof(SpaprMachineClass),
4748     .class_init    = spapr_machine_class_init,
4749     .interfaces = (InterfaceInfo[]) {
4750         { TYPE_FW_PATH_PROVIDER },
4751         { TYPE_NMI },
4752         { TYPE_HOTPLUG_HANDLER },
4753         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4754         { TYPE_XICS_FABRIC },
4755         { TYPE_INTERRUPT_STATS_PROVIDER },
4756         { TYPE_XIVE_FABRIC },
4757         { TYPE_VOF_MACHINE_IF },
4758         { }
4759     },
4760 };
4761 
4762 static void spapr_machine_latest_class_options(MachineClass *mc)
4763 {
4764     mc->alias = "pseries";
4765     mc->is_default = true;
4766 }
4767 
4768 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4769     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4770                                                     void *data)      \
4771     {                                                                \
4772         MachineClass *mc = MACHINE_CLASS(oc);                        \
4773         spapr_machine_##suffix##_class_options(mc);                  \
4774         if (latest) {                                                \
4775             spapr_machine_latest_class_options(mc);                  \
4776         }                                                            \
4777     }                                                                \
4778     static const TypeInfo spapr_machine_##suffix##_info = {          \
4779         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4780         .parent = TYPE_SPAPR_MACHINE,                                \
4781         .class_init = spapr_machine_##suffix##_class_init,           \
4782     };                                                               \
4783     static void spapr_machine_register_##suffix(void)                \
4784     {                                                                \
4785         type_register(&spapr_machine_##suffix##_info);               \
4786     }                                                                \
4787     type_init(spapr_machine_register_##suffix)
4788 
4789 /*
4790  * pseries-9.0
4791  */
4792 static void spapr_machine_9_0_class_options(MachineClass *mc)
4793 {
4794     /* Defaults for the latest behaviour inherited from the base class */
4795 }
4796 
4797 DEFINE_SPAPR_MACHINE(9_0, "9.0", true);
4798 
4799 /*
4800  * pseries-8.2
4801  */
4802 static void spapr_machine_8_2_class_options(MachineClass *mc)
4803 {
4804     spapr_machine_9_0_class_options(mc);
4805     compat_props_add(mc->compat_props, hw_compat_8_2, hw_compat_8_2_len);
4806 }
4807 
4808 DEFINE_SPAPR_MACHINE(8_2, "8.2", false);
4809 
4810 /*
4811  * pseries-8.1
4812  */
4813 static void spapr_machine_8_1_class_options(MachineClass *mc)
4814 {
4815     spapr_machine_8_2_class_options(mc);
4816     compat_props_add(mc->compat_props, hw_compat_8_1, hw_compat_8_1_len);
4817 }
4818 
4819 DEFINE_SPAPR_MACHINE(8_1, "8.1", false);
4820 
4821 /*
4822  * pseries-8.0
4823  */
4824 static void spapr_machine_8_0_class_options(MachineClass *mc)
4825 {
4826     spapr_machine_8_1_class_options(mc);
4827     compat_props_add(mc->compat_props, hw_compat_8_0, hw_compat_8_0_len);
4828 }
4829 
4830 DEFINE_SPAPR_MACHINE(8_0, "8.0", false);
4831 
4832 /*
4833  * pseries-7.2
4834  */
4835 static void spapr_machine_7_2_class_options(MachineClass *mc)
4836 {
4837     spapr_machine_8_0_class_options(mc);
4838     compat_props_add(mc->compat_props, hw_compat_7_2, hw_compat_7_2_len);
4839 }
4840 
4841 DEFINE_SPAPR_MACHINE(7_2, "7.2", false);
4842 
4843 /*
4844  * pseries-7.1
4845  */
4846 static void spapr_machine_7_1_class_options(MachineClass *mc)
4847 {
4848     spapr_machine_7_2_class_options(mc);
4849     compat_props_add(mc->compat_props, hw_compat_7_1, hw_compat_7_1_len);
4850 }
4851 
4852 DEFINE_SPAPR_MACHINE(7_1, "7.1", false);
4853 
4854 /*
4855  * pseries-7.0
4856  */
4857 static void spapr_machine_7_0_class_options(MachineClass *mc)
4858 {
4859     spapr_machine_7_1_class_options(mc);
4860     compat_props_add(mc->compat_props, hw_compat_7_0, hw_compat_7_0_len);
4861 }
4862 
4863 DEFINE_SPAPR_MACHINE(7_0, "7.0", false);
4864 
4865 /*
4866  * pseries-6.2
4867  */
4868 static void spapr_machine_6_2_class_options(MachineClass *mc)
4869 {
4870     spapr_machine_7_0_class_options(mc);
4871     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4872 }
4873 
4874 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4875 
4876 /*
4877  * pseries-6.1
4878  */
4879 static void spapr_machine_6_1_class_options(MachineClass *mc)
4880 {
4881     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4882 
4883     spapr_machine_6_2_class_options(mc);
4884     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4885     smc->pre_6_2_numa_affinity = true;
4886     mc->smp_props.prefer_sockets = true;
4887 }
4888 
4889 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4890 
4891 /*
4892  * pseries-6.0
4893  */
4894 static void spapr_machine_6_0_class_options(MachineClass *mc)
4895 {
4896     spapr_machine_6_1_class_options(mc);
4897     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4898 }
4899 
4900 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4901 
4902 /*
4903  * pseries-5.2
4904  */
4905 static void spapr_machine_5_2_class_options(MachineClass *mc)
4906 {
4907     spapr_machine_6_0_class_options(mc);
4908     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4909 }
4910 
4911 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4912 
4913 /*
4914  * pseries-5.1
4915  */
4916 static void spapr_machine_5_1_class_options(MachineClass *mc)
4917 {
4918     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4919 
4920     spapr_machine_5_2_class_options(mc);
4921     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4922     smc->pre_5_2_numa_associativity = true;
4923 }
4924 
4925 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4926 
4927 /*
4928  * pseries-5.0
4929  */
4930 static void spapr_machine_5_0_class_options(MachineClass *mc)
4931 {
4932     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4933     static GlobalProperty compat[] = {
4934         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4935     };
4936 
4937     spapr_machine_5_1_class_options(mc);
4938     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4939     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4940     mc->numa_mem_supported = true;
4941     smc->pre_5_1_assoc_refpoints = true;
4942 }
4943 
4944 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4945 
4946 /*
4947  * pseries-4.2
4948  */
4949 static void spapr_machine_4_2_class_options(MachineClass *mc)
4950 {
4951     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4952 
4953     spapr_machine_5_0_class_options(mc);
4954     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4955     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4956     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4957     smc->rma_limit = 16 * GiB;
4958     mc->nvdimm_supported = false;
4959 }
4960 
4961 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4962 
4963 /*
4964  * pseries-4.1
4965  */
4966 static void spapr_machine_4_1_class_options(MachineClass *mc)
4967 {
4968     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4969     static GlobalProperty compat[] = {
4970         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4971         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4972     };
4973 
4974     spapr_machine_4_2_class_options(mc);
4975     smc->linux_pci_probe = false;
4976     smc->smp_threads_vsmt = false;
4977     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4978     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4979 }
4980 
4981 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4982 
4983 /*
4984  * pseries-4.0
4985  */
4986 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4987                               uint64_t *buid, hwaddr *pio,
4988                               hwaddr *mmio32, hwaddr *mmio64,
4989                               unsigned n_dma, uint32_t *liobns, Error **errp)
4990 {
4991     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4992                              liobns, errp)) {
4993         return false;
4994     }
4995     return true;
4996 }
4997 static void spapr_machine_4_0_class_options(MachineClass *mc)
4998 {
4999     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5000 
5001     spapr_machine_4_1_class_options(mc);
5002     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
5003     smc->phb_placement = phb_placement_4_0;
5004     smc->irq = &spapr_irq_xics;
5005     smc->pre_4_1_migration = true;
5006 }
5007 
5008 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
5009 
5010 /*
5011  * pseries-3.1
5012  */
5013 static void spapr_machine_3_1_class_options(MachineClass *mc)
5014 {
5015     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5016 
5017     spapr_machine_4_0_class_options(mc);
5018     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
5019 
5020     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
5021     smc->update_dt_enabled = false;
5022     smc->dr_phb_enabled = false;
5023     smc->broken_host_serial_model = true;
5024     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
5025     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
5026     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
5027     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
5028 }
5029 
5030 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
5031 
5032 /*
5033  * pseries-3.0
5034  */
5035 
5036 static void spapr_machine_3_0_class_options(MachineClass *mc)
5037 {
5038     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5039 
5040     spapr_machine_3_1_class_options(mc);
5041     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
5042 
5043     smc->legacy_irq_allocation = true;
5044     smc->nr_xirqs = 0x400;
5045     smc->irq = &spapr_irq_xics_legacy;
5046 }
5047 
5048 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
5049 
5050 /*
5051  * pseries-2.12
5052  */
5053 static void spapr_machine_2_12_class_options(MachineClass *mc)
5054 {
5055     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5056     static GlobalProperty compat[] = {
5057         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
5058         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
5059     };
5060 
5061     spapr_machine_3_0_class_options(mc);
5062     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
5063     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5064 
5065     /* We depend on kvm_enabled() to choose a default value for the
5066      * hpt-max-page-size capability. Of course we can't do it here
5067      * because this is too early and the HW accelerator isn't initialized
5068      * yet. Postpone this to machine init (see default_caps_with_cpu()).
5069      */
5070     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
5071 }
5072 
5073 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
5074 
5075 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
5076 {
5077     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5078 
5079     spapr_machine_2_12_class_options(mc);
5080     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
5081     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
5082     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
5083 }
5084 
5085 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
5086 
5087 /*
5088  * pseries-2.11
5089  */
5090 
5091 static void spapr_machine_2_11_class_options(MachineClass *mc)
5092 {
5093     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5094 
5095     spapr_machine_2_12_class_options(mc);
5096     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
5097     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
5098     mc->deprecation_reason = "old and not maintained - use a 2.12+ version";
5099 }
5100 
5101 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
5102 
5103 /*
5104  * pseries-2.10
5105  */
5106 
5107 static void spapr_machine_2_10_class_options(MachineClass *mc)
5108 {
5109     spapr_machine_2_11_class_options(mc);
5110     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
5111 }
5112 
5113 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
5114 
5115 /*
5116  * pseries-2.9
5117  */
5118 
5119 static void spapr_machine_2_9_class_options(MachineClass *mc)
5120 {
5121     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5122     static GlobalProperty compat[] = {
5123         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
5124     };
5125 
5126     spapr_machine_2_10_class_options(mc);
5127     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
5128     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5129     smc->pre_2_10_has_unused_icps = true;
5130     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
5131 }
5132 
5133 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
5134 
5135 /*
5136  * pseries-2.8
5137  */
5138 
5139 static void spapr_machine_2_8_class_options(MachineClass *mc)
5140 {
5141     static GlobalProperty compat[] = {
5142         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
5143     };
5144 
5145     spapr_machine_2_9_class_options(mc);
5146     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5147     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5148     mc->numa_mem_align_shift = 23;
5149 }
5150 
5151 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5152 
5153 /*
5154  * pseries-2.7
5155  */
5156 
5157 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5158                               uint64_t *buid, hwaddr *pio,
5159                               hwaddr *mmio32, hwaddr *mmio64,
5160                               unsigned n_dma, uint32_t *liobns, Error **errp)
5161 {
5162     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5163     const uint64_t base_buid = 0x800000020000000ULL;
5164     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5165     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5166     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5167     const uint32_t max_index = 255;
5168     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5169 
5170     uint64_t ram_top = MACHINE(spapr)->ram_size;
5171     hwaddr phb0_base, phb_base;
5172     int i;
5173 
5174     /* Do we have device memory? */
5175     if (MACHINE(spapr)->device_memory) {
5176         /* Can't just use maxram_size, because there may be an
5177          * alignment gap between normal and device memory regions
5178          */
5179         ram_top = MACHINE(spapr)->device_memory->base +
5180             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5181     }
5182 
5183     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5184 
5185     if (index > max_index) {
5186         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5187                    max_index);
5188         return false;
5189     }
5190 
5191     *buid = base_buid + index;
5192     for (i = 0; i < n_dma; ++i) {
5193         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5194     }
5195 
5196     phb_base = phb0_base + index * phb_spacing;
5197     *pio = phb_base + pio_offset;
5198     *mmio32 = phb_base + mmio_offset;
5199     /*
5200      * We don't set the 64-bit MMIO window, relying on the PHB's
5201      * fallback behaviour of automatically splitting a large "32-bit"
5202      * window into contiguous 32-bit and 64-bit windows
5203      */
5204 
5205     return true;
5206 }
5207 
5208 static void spapr_machine_2_7_class_options(MachineClass *mc)
5209 {
5210     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5211     static GlobalProperty compat[] = {
5212         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5213         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5214         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5215         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5216     };
5217 
5218     spapr_machine_2_8_class_options(mc);
5219     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5220     mc->default_machine_opts = "modern-hotplug-events=off";
5221     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5222     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5223     smc->phb_placement = phb_placement_2_7;
5224 }
5225 
5226 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5227 
5228 /*
5229  * pseries-2.6
5230  */
5231 
5232 static void spapr_machine_2_6_class_options(MachineClass *mc)
5233 {
5234     static GlobalProperty compat[] = {
5235         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5236     };
5237 
5238     spapr_machine_2_7_class_options(mc);
5239     mc->has_hotpluggable_cpus = false;
5240     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5241     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5242 }
5243 
5244 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5245 
5246 /*
5247  * pseries-2.5
5248  */
5249 
5250 static void spapr_machine_2_5_class_options(MachineClass *mc)
5251 {
5252     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5253     static GlobalProperty compat[] = {
5254         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5255     };
5256 
5257     spapr_machine_2_6_class_options(mc);
5258     smc->use_ohci_by_default = true;
5259     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5260     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5261 }
5262 
5263 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5264 
5265 /*
5266  * pseries-2.4
5267  */
5268 
5269 static void spapr_machine_2_4_class_options(MachineClass *mc)
5270 {
5271     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5272 
5273     spapr_machine_2_5_class_options(mc);
5274     smc->dr_lmb_enabled = false;
5275     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5276 }
5277 
5278 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5279 
5280 /*
5281  * pseries-2.3
5282  */
5283 
5284 static void spapr_machine_2_3_class_options(MachineClass *mc)
5285 {
5286     static GlobalProperty compat[] = {
5287         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5288     };
5289     spapr_machine_2_4_class_options(mc);
5290     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5291     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5292 }
5293 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5294 
5295 /*
5296  * pseries-2.2
5297  */
5298 
5299 static void spapr_machine_2_2_class_options(MachineClass *mc)
5300 {
5301     static GlobalProperty compat[] = {
5302         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5303     };
5304 
5305     spapr_machine_2_3_class_options(mc);
5306     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5307     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5308     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5309 }
5310 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5311 
5312 /*
5313  * pseries-2.1
5314  */
5315 
5316 static void spapr_machine_2_1_class_options(MachineClass *mc)
5317 {
5318     spapr_machine_2_2_class_options(mc);
5319     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5320 }
5321 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5322 
5323 static void spapr_machine_register_types(void)
5324 {
5325     type_register_static(&spapr_machine_info);
5326 }
5327 
5328 type_init(spapr_machine_register_types)
5329