xref: /openbmc/qemu/hw/ppc/spapr.c (revision 35dce34fbc1cfa6a26f95b83f3a8949a4150412f)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 #include "hw/ppc/spapr_numa.h"
85 
86 #include "monitor/monitor.h"
87 
88 #include <libfdt.h>
89 
90 /* SLOF memory layout:
91  *
92  * SLOF raw image loaded at 0, copies its romfs right below the flat
93  * device-tree, then position SLOF itself 31M below that
94  *
95  * So we set FW_OVERHEAD to 40MB which should account for all of that
96  * and more
97  *
98  * We load our kernel at 4M, leaving space for SLOF initial image
99  */
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            (128 * MiB)
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static void spapr_dt_pa_features(SpaprMachineState *spapr,
206                                  PowerPCCPU *cpu,
207                                  void *fdt, int offset)
208 {
209     uint8_t pa_features_206[] = { 6, 0,
210         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
211     uint8_t pa_features_207[] = { 24, 0,
212         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
213         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
214         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
215         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
216     uint8_t pa_features_300[] = { 66, 0,
217         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
218         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
219         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
220         /* 6: DS207 */
221         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
222         /* 16: Vector */
223         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
224         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
225         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
226         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
227         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
228         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
229         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
230         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
231         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
232         /* 42: PM, 44: PC RA, 46: SC vec'd */
233         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
234         /* 48: SIMD, 50: QP BFP, 52: String */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
236         /* 54: DecFP, 56: DecI, 58: SHA */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
238         /* 60: NM atomic, 62: RNG */
239         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
240     };
241     uint8_t *pa_features = NULL;
242     size_t pa_size;
243 
244     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
245         pa_features = pa_features_206;
246         pa_size = sizeof(pa_features_206);
247     }
248     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
249         pa_features = pa_features_207;
250         pa_size = sizeof(pa_features_207);
251     }
252     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
253         pa_features = pa_features_300;
254         pa_size = sizeof(pa_features_300);
255     }
256     if (!pa_features) {
257         return;
258     }
259 
260     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
261         /*
262          * Note: we keep CI large pages off by default because a 64K capable
263          * guest provisioned with large pages might otherwise try to map a qemu
264          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
265          * even if that qemu runs on a 4k host.
266          * We dd this bit back here if we are confident this is not an issue
267          */
268         pa_features[3] |= 0x20;
269     }
270     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
271         pa_features[24] |= 0x80;    /* Transactional memory support */
272     }
273     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
274         /* Workaround for broken kernels that attempt (guest) radix
275          * mode when they can't handle it, if they see the radix bit set
276          * in pa-features. So hide it from them. */
277         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
278     }
279 
280     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
281 }
282 
283 static hwaddr spapr_node0_size(MachineState *machine)
284 {
285     if (machine->numa_state->num_nodes) {
286         int i;
287         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
288             if (machine->numa_state->nodes[i].node_mem) {
289                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
290                            machine->ram_size);
291             }
292         }
293     }
294     return machine->ram_size;
295 }
296 
297 static void add_str(GString *s, const gchar *s1)
298 {
299     g_string_append_len(s, s1, strlen(s1) + 1);
300 }
301 
302 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
303                                 hwaddr start, hwaddr size)
304 {
305     char mem_name[32];
306     uint64_t mem_reg_property[2];
307     int off;
308 
309     mem_reg_property[0] = cpu_to_be64(start);
310     mem_reg_property[1] = cpu_to_be64(size);
311 
312     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
313     off = fdt_add_subnode(fdt, 0, mem_name);
314     _FDT(off);
315     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
316     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
317                       sizeof(mem_reg_property))));
318     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
319     return off;
320 }
321 
322 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
323 {
324     MemoryDeviceInfoList *info;
325 
326     for (info = list; info; info = info->next) {
327         MemoryDeviceInfo *value = info->value;
328 
329         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
330             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
331 
332             if (addr >= pcdimm_info->addr &&
333                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
334                 return pcdimm_info->node;
335             }
336         }
337     }
338 
339     return -1;
340 }
341 
342 struct sPAPRDrconfCellV2 {
343      uint32_t seq_lmbs;
344      uint64_t base_addr;
345      uint32_t drc_index;
346      uint32_t aa_index;
347      uint32_t flags;
348 } QEMU_PACKED;
349 
350 typedef struct DrconfCellQueue {
351     struct sPAPRDrconfCellV2 cell;
352     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
353 } DrconfCellQueue;
354 
355 static DrconfCellQueue *
356 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
357                       uint32_t drc_index, uint32_t aa_index,
358                       uint32_t flags)
359 {
360     DrconfCellQueue *elem;
361 
362     elem = g_malloc0(sizeof(*elem));
363     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
364     elem->cell.base_addr = cpu_to_be64(base_addr);
365     elem->cell.drc_index = cpu_to_be32(drc_index);
366     elem->cell.aa_index = cpu_to_be32(aa_index);
367     elem->cell.flags = cpu_to_be32(flags);
368 
369     return elem;
370 }
371 
372 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
373                                       int offset, MemoryDeviceInfoList *dimms)
374 {
375     MachineState *machine = MACHINE(spapr);
376     uint8_t *int_buf, *cur_index;
377     int ret;
378     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
379     uint64_t addr, cur_addr, size;
380     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
381     uint64_t mem_end = machine->device_memory->base +
382                        memory_region_size(&machine->device_memory->mr);
383     uint32_t node, buf_len, nr_entries = 0;
384     SpaprDrc *drc;
385     DrconfCellQueue *elem, *next;
386     MemoryDeviceInfoList *info;
387     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
388         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
389 
390     /* Entry to cover RAM and the gap area */
391     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
392                                  SPAPR_LMB_FLAGS_RESERVED |
393                                  SPAPR_LMB_FLAGS_DRC_INVALID);
394     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
395     nr_entries++;
396 
397     cur_addr = machine->device_memory->base;
398     for (info = dimms; info; info = info->next) {
399         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
400 
401         addr = di->addr;
402         size = di->size;
403         node = di->node;
404 
405         /*
406          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
407          * area is marked hotpluggable in the next iteration for the bigger
408          * chunk including the NVDIMM occupied area.
409          */
410         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
411             continue;
412 
413         /* Entry for hot-pluggable area */
414         if (cur_addr < addr) {
415             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
416             g_assert(drc);
417             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
418                                          cur_addr, spapr_drc_index(drc), -1, 0);
419             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
420             nr_entries++;
421         }
422 
423         /* Entry for DIMM */
424         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
425         g_assert(drc);
426         elem = spapr_get_drconf_cell(size / lmb_size, addr,
427                                      spapr_drc_index(drc), node,
428                                      (SPAPR_LMB_FLAGS_ASSIGNED |
429                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
430         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
431         nr_entries++;
432         cur_addr = addr + size;
433     }
434 
435     /* Entry for remaining hotpluggable area */
436     if (cur_addr < mem_end) {
437         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
438         g_assert(drc);
439         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
440                                      cur_addr, spapr_drc_index(drc), -1, 0);
441         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
442         nr_entries++;
443     }
444 
445     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
446     int_buf = cur_index = g_malloc0(buf_len);
447     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
448     cur_index += sizeof(nr_entries);
449 
450     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
451         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
452         cur_index += sizeof(elem->cell);
453         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
454         g_free(elem);
455     }
456 
457     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
458     g_free(int_buf);
459     if (ret < 0) {
460         return -1;
461     }
462     return 0;
463 }
464 
465 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
466                                    int offset, MemoryDeviceInfoList *dimms)
467 {
468     MachineState *machine = MACHINE(spapr);
469     int i, ret;
470     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
471     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
472     uint32_t nr_lmbs = (machine->device_memory->base +
473                        memory_region_size(&machine->device_memory->mr)) /
474                        lmb_size;
475     uint32_t *int_buf, *cur_index, buf_len;
476 
477     /*
478      * Allocate enough buffer size to fit in ibm,dynamic-memory
479      */
480     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
481     cur_index = int_buf = g_malloc0(buf_len);
482     int_buf[0] = cpu_to_be32(nr_lmbs);
483     cur_index++;
484     for (i = 0; i < nr_lmbs; i++) {
485         uint64_t addr = i * lmb_size;
486         uint32_t *dynamic_memory = cur_index;
487 
488         if (i >= device_lmb_start) {
489             SpaprDrc *drc;
490 
491             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
492             g_assert(drc);
493 
494             dynamic_memory[0] = cpu_to_be32(addr >> 32);
495             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
496             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
497             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
498             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
499             if (memory_region_present(get_system_memory(), addr)) {
500                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
501             } else {
502                 dynamic_memory[5] = cpu_to_be32(0);
503             }
504         } else {
505             /*
506              * LMB information for RMA, boot time RAM and gap b/n RAM and
507              * device memory region -- all these are marked as reserved
508              * and as having no valid DRC.
509              */
510             dynamic_memory[0] = cpu_to_be32(addr >> 32);
511             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
512             dynamic_memory[2] = cpu_to_be32(0);
513             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
514             dynamic_memory[4] = cpu_to_be32(-1);
515             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
516                                             SPAPR_LMB_FLAGS_DRC_INVALID);
517         }
518 
519         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
520     }
521     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
522     g_free(int_buf);
523     if (ret < 0) {
524         return -1;
525     }
526     return 0;
527 }
528 
529 /*
530  * Adds ibm,dynamic-reconfiguration-memory node.
531  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
532  * of this device tree node.
533  */
534 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
535                                                    void *fdt)
536 {
537     MachineState *machine = MACHINE(spapr);
538     int ret, offset;
539     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
540     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
541                                 cpu_to_be32(lmb_size & 0xffffffff)};
542     MemoryDeviceInfoList *dimms = NULL;
543 
544     /*
545      * Don't create the node if there is no device memory
546      */
547     if (machine->ram_size == machine->maxram_size) {
548         return 0;
549     }
550 
551     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
552 
553     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
554                     sizeof(prop_lmb_size));
555     if (ret < 0) {
556         return ret;
557     }
558 
559     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
560     if (ret < 0) {
561         return ret;
562     }
563 
564     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
565     if (ret < 0) {
566         return ret;
567     }
568 
569     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
570     dimms = qmp_memory_device_list();
571     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
572         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
573     } else {
574         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
575     }
576     qapi_free_MemoryDeviceInfoList(dimms);
577 
578     if (ret < 0) {
579         return ret;
580     }
581 
582     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
583 
584     return ret;
585 }
586 
587 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
588 {
589     MachineState *machine = MACHINE(spapr);
590     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
591     hwaddr mem_start, node_size;
592     int i, nb_nodes = machine->numa_state->num_nodes;
593     NodeInfo *nodes = machine->numa_state->nodes;
594 
595     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
596         if (!nodes[i].node_mem) {
597             continue;
598         }
599         if (mem_start >= machine->ram_size) {
600             node_size = 0;
601         } else {
602             node_size = nodes[i].node_mem;
603             if (node_size > machine->ram_size - mem_start) {
604                 node_size = machine->ram_size - mem_start;
605             }
606         }
607         if (!mem_start) {
608             /* spapr_machine_init() checks for rma_size <= node0_size
609              * already */
610             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
611             mem_start += spapr->rma_size;
612             node_size -= spapr->rma_size;
613         }
614         for ( ; node_size; ) {
615             hwaddr sizetmp = pow2floor(node_size);
616 
617             /* mem_start != 0 here */
618             if (ctzl(mem_start) < ctzl(sizetmp)) {
619                 sizetmp = 1ULL << ctzl(mem_start);
620             }
621 
622             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
623             node_size -= sizetmp;
624             mem_start += sizetmp;
625         }
626     }
627 
628     /* Generate ibm,dynamic-reconfiguration-memory node if required */
629     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
630         int ret;
631 
632         g_assert(smc->dr_lmb_enabled);
633         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
634         if (ret) {
635             return ret;
636         }
637     }
638 
639     return 0;
640 }
641 
642 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
643                          SpaprMachineState *spapr)
644 {
645     MachineState *ms = MACHINE(spapr);
646     PowerPCCPU *cpu = POWERPC_CPU(cs);
647     CPUPPCState *env = &cpu->env;
648     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
649     int index = spapr_get_vcpu_id(cpu);
650     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
651                        0xffffffff, 0xffffffff};
652     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
653         : SPAPR_TIMEBASE_FREQ;
654     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
655     uint32_t page_sizes_prop[64];
656     size_t page_sizes_prop_size;
657     unsigned int smp_threads = ms->smp.threads;
658     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
659     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
660     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
661     SpaprDrc *drc;
662     int drc_index;
663     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
664     int i;
665 
666     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
667     if (drc) {
668         drc_index = spapr_drc_index(drc);
669         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
670     }
671 
672     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
673     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
674 
675     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
676     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
677                            env->dcache_line_size)));
678     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
679                            env->dcache_line_size)));
680     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
681                            env->icache_line_size)));
682     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
683                            env->icache_line_size)));
684 
685     if (pcc->l1_dcache_size) {
686         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
687                                pcc->l1_dcache_size)));
688     } else {
689         warn_report("Unknown L1 dcache size for cpu");
690     }
691     if (pcc->l1_icache_size) {
692         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
693                                pcc->l1_icache_size)));
694     } else {
695         warn_report("Unknown L1 icache size for cpu");
696     }
697 
698     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
699     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
700     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
701     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
702     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
703     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
704 
705     if (env->spr_cb[SPR_PURR].oea_read) {
706         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
707     }
708     if (env->spr_cb[SPR_SPURR].oea_read) {
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
710     }
711 
712     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
713         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
714                           segs, sizeof(segs))));
715     }
716 
717     /* Advertise VSX (vector extensions) if available
718      *   1               == VMX / Altivec available
719      *   2               == VSX available
720      *
721      * Only CPUs for which we create core types in spapr_cpu_core.c
722      * are possible, and all of those have VMX */
723     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
724         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
725     } else {
726         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
727     }
728 
729     /* Advertise DFP (Decimal Floating Point) if available
730      *   0 / no property == no DFP
731      *   1               == DFP available */
732     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
733         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
734     }
735 
736     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
737                                                       sizeof(page_sizes_prop));
738     if (page_sizes_prop_size) {
739         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
740                           page_sizes_prop, page_sizes_prop_size)));
741     }
742 
743     spapr_dt_pa_features(spapr, cpu, fdt, offset);
744 
745     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
746                            cs->cpu_index / vcpus_per_socket)));
747 
748     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
749                       pft_size_prop, sizeof(pft_size_prop))));
750 
751     if (ms->numa_state->num_nodes > 1) {
752         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
753     }
754 
755     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
756 
757     if (pcc->radix_page_info) {
758         for (i = 0; i < pcc->radix_page_info->count; i++) {
759             radix_AP_encodings[i] =
760                 cpu_to_be32(pcc->radix_page_info->entries[i]);
761         }
762         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
763                           radix_AP_encodings,
764                           pcc->radix_page_info->count *
765                           sizeof(radix_AP_encodings[0]))));
766     }
767 
768     /*
769      * We set this property to let the guest know that it can use the large
770      * decrementer and its width in bits.
771      */
772     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
773         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
774                               pcc->lrg_decr_bits)));
775 }
776 
777 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
778 {
779     CPUState **rev;
780     CPUState *cs;
781     int n_cpus;
782     int cpus_offset;
783     char *nodename;
784     int i;
785 
786     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
787     _FDT(cpus_offset);
788     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
789     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
790 
791     /*
792      * We walk the CPUs in reverse order to ensure that CPU DT nodes
793      * created by fdt_add_subnode() end up in the right order in FDT
794      * for the guest kernel the enumerate the CPUs correctly.
795      *
796      * The CPU list cannot be traversed in reverse order, so we need
797      * to do extra work.
798      */
799     n_cpus = 0;
800     rev = NULL;
801     CPU_FOREACH(cs) {
802         rev = g_renew(CPUState *, rev, n_cpus + 1);
803         rev[n_cpus++] = cs;
804     }
805 
806     for (i = n_cpus - 1; i >= 0; i--) {
807         CPUState *cs = rev[i];
808         PowerPCCPU *cpu = POWERPC_CPU(cs);
809         int index = spapr_get_vcpu_id(cpu);
810         DeviceClass *dc = DEVICE_GET_CLASS(cs);
811         int offset;
812 
813         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
814             continue;
815         }
816 
817         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
818         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
819         g_free(nodename);
820         _FDT(offset);
821         spapr_dt_cpu(cs, fdt, offset, spapr);
822     }
823 
824     g_free(rev);
825 }
826 
827 static int spapr_dt_rng(void *fdt)
828 {
829     int node;
830     int ret;
831 
832     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
833     if (node <= 0) {
834         return -1;
835     }
836     ret = fdt_setprop_string(fdt, node, "device_type",
837                              "ibm,platform-facilities");
838     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
839     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
840 
841     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
842     if (node <= 0) {
843         return -1;
844     }
845     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
846 
847     return ret ? -1 : 0;
848 }
849 
850 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
851 {
852     MachineState *ms = MACHINE(spapr);
853     int rtas;
854     GString *hypertas = g_string_sized_new(256);
855     GString *qemu_hypertas = g_string_sized_new(256);
856     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
857         memory_region_size(&MACHINE(spapr)->device_memory->mr);
858     uint32_t lrdr_capacity[] = {
859         cpu_to_be32(max_device_addr >> 32),
860         cpu_to_be32(max_device_addr & 0xffffffff),
861         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
862         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
863         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
864     };
865 
866     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
867 
868     /* hypertas */
869     add_str(hypertas, "hcall-pft");
870     add_str(hypertas, "hcall-term");
871     add_str(hypertas, "hcall-dabr");
872     add_str(hypertas, "hcall-interrupt");
873     add_str(hypertas, "hcall-tce");
874     add_str(hypertas, "hcall-vio");
875     add_str(hypertas, "hcall-splpar");
876     add_str(hypertas, "hcall-join");
877     add_str(hypertas, "hcall-bulk");
878     add_str(hypertas, "hcall-set-mode");
879     add_str(hypertas, "hcall-sprg0");
880     add_str(hypertas, "hcall-copy");
881     add_str(hypertas, "hcall-debug");
882     add_str(hypertas, "hcall-vphn");
883     add_str(qemu_hypertas, "hcall-memop1");
884 
885     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
886         add_str(hypertas, "hcall-multi-tce");
887     }
888 
889     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
890         add_str(hypertas, "hcall-hpt-resize");
891     }
892 
893     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
894                      hypertas->str, hypertas->len));
895     g_string_free(hypertas, TRUE);
896     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
897                      qemu_hypertas->str, qemu_hypertas->len));
898     g_string_free(qemu_hypertas, TRUE);
899 
900     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
901 
902     /*
903      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
904      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
905      *
906      * The system reset requirements are driven by existing Linux and PowerVM
907      * implementation which (contrary to PAPR) saves r3 in the error log
908      * structure like machine check, so Linux expects to find the saved r3
909      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
910      * does not look at the error value).
911      *
912      * System reset interrupts are not subject to interlock like machine
913      * check, so this memory area could be corrupted if the sreset is
914      * interrupted by a machine check (or vice versa) if it was shared. To
915      * prevent this, system reset uses per-CPU areas for the sreset save
916      * area. A system reset that interrupts a system reset handler could
917      * still overwrite this area, but Linux doesn't try to recover in that
918      * case anyway.
919      *
920      * The extra 8 bytes is required because Linux's FWNMI error log check
921      * is off-by-one.
922      */
923     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
924 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
925     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
926                           RTAS_ERROR_LOG_MAX));
927     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
928                           RTAS_EVENT_SCAN_RATE));
929 
930     g_assert(msi_nonbroken);
931     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
932 
933     /*
934      * According to PAPR, rtas ibm,os-term does not guarantee a return
935      * back to the guest cpu.
936      *
937      * While an additional ibm,extended-os-term property indicates
938      * that rtas call return will always occur. Set this property.
939      */
940     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
941 
942     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
943                      lrdr_capacity, sizeof(lrdr_capacity)));
944 
945     spapr_dt_rtas_tokens(fdt, rtas);
946 }
947 
948 /*
949  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
950  * and the XIVE features that the guest may request and thus the valid
951  * values for bytes 23..26 of option vector 5:
952  */
953 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
954                                           int chosen)
955 {
956     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
957 
958     char val[2 * 4] = {
959         23, 0x00, /* XICS / XIVE mode */
960         24, 0x00, /* Hash/Radix, filled in below. */
961         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
962         26, 0x40, /* Radix options: GTSE == yes. */
963     };
964 
965     if (spapr->irq->xics && spapr->irq->xive) {
966         val[1] = SPAPR_OV5_XIVE_BOTH;
967     } else if (spapr->irq->xive) {
968         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
969     } else {
970         assert(spapr->irq->xics);
971         val[1] = SPAPR_OV5_XIVE_LEGACY;
972     }
973 
974     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
975                           first_ppc_cpu->compat_pvr)) {
976         /*
977          * If we're in a pre POWER9 compat mode then the guest should
978          * do hash and use the legacy interrupt mode
979          */
980         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
981         val[3] = 0x00; /* Hash */
982     } else if (kvm_enabled()) {
983         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
984             val[3] = 0x80; /* OV5_MMU_BOTH */
985         } else if (kvmppc_has_cap_mmu_radix()) {
986             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
987         } else {
988             val[3] = 0x00; /* Hash */
989         }
990     } else {
991         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
992         val[3] = 0xC0;
993     }
994     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
995                      val, sizeof(val)));
996 }
997 
998 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
999 {
1000     MachineState *machine = MACHINE(spapr);
1001     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1002     int chosen;
1003 
1004     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1005 
1006     if (reset) {
1007         const char *boot_device = machine->boot_order;
1008         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1009         size_t cb = 0;
1010         char *bootlist = get_boot_devices_list(&cb);
1011 
1012         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1013             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1014                                     machine->kernel_cmdline));
1015         }
1016 
1017         if (spapr->initrd_size) {
1018             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1019                                   spapr->initrd_base));
1020             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1021                                   spapr->initrd_base + spapr->initrd_size));
1022         }
1023 
1024         if (spapr->kernel_size) {
1025             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1026                                   cpu_to_be64(spapr->kernel_size) };
1027 
1028             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1029                          &kprop, sizeof(kprop)));
1030             if (spapr->kernel_le) {
1031                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1032             }
1033         }
1034         if (boot_menu) {
1035             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1036         }
1037         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1038         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1039         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1040 
1041         if (cb && bootlist) {
1042             int i;
1043 
1044             for (i = 0; i < cb; i++) {
1045                 if (bootlist[i] == '\n') {
1046                     bootlist[i] = ' ';
1047                 }
1048             }
1049             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1050         }
1051 
1052         if (boot_device && strlen(boot_device)) {
1053             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1054         }
1055 
1056         if (!spapr->has_graphics && stdout_path) {
1057             /*
1058              * "linux,stdout-path" and "stdout" properties are
1059              * deprecated by linux kernel. New platforms should only
1060              * use the "stdout-path" property. Set the new property
1061              * and continue using older property to remain compatible
1062              * with the existing firmware.
1063              */
1064             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1065             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1066         }
1067 
1068         /*
1069          * We can deal with BAR reallocation just fine, advertise it
1070          * to the guest
1071          */
1072         if (smc->linux_pci_probe) {
1073             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1074         }
1075 
1076         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1077 
1078         g_free(stdout_path);
1079         g_free(bootlist);
1080     }
1081 
1082     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1083 }
1084 
1085 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1086 {
1087     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1088      * KVM to work under pHyp with some guest co-operation */
1089     int hypervisor;
1090     uint8_t hypercall[16];
1091 
1092     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1093     /* indicate KVM hypercall interface */
1094     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1095     if (kvmppc_has_cap_fixup_hcalls()) {
1096         /*
1097          * Older KVM versions with older guest kernels were broken
1098          * with the magic page, don't allow the guest to map it.
1099          */
1100         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1101                                   sizeof(hypercall))) {
1102             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1103                              hypercall, sizeof(hypercall)));
1104         }
1105     }
1106 }
1107 
1108 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1109 {
1110     MachineState *machine = MACHINE(spapr);
1111     MachineClass *mc = MACHINE_GET_CLASS(machine);
1112     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1113     int ret;
1114     void *fdt;
1115     SpaprPhbState *phb;
1116     char *buf;
1117 
1118     fdt = g_malloc0(space);
1119     _FDT((fdt_create_empty_tree(fdt, space)));
1120 
1121     /* Root node */
1122     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1123     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1124     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1125 
1126     /* Guest UUID & Name*/
1127     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1128     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1129     if (qemu_uuid_set) {
1130         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1131     }
1132     g_free(buf);
1133 
1134     if (qemu_get_vm_name()) {
1135         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1136                                 qemu_get_vm_name()));
1137     }
1138 
1139     /* Host Model & Serial Number */
1140     if (spapr->host_model) {
1141         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1142     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1143         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1144         g_free(buf);
1145     }
1146 
1147     if (spapr->host_serial) {
1148         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1149     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1150         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1151         g_free(buf);
1152     }
1153 
1154     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1155     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1156 
1157     /* /interrupt controller */
1158     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1159 
1160     ret = spapr_dt_memory(spapr, fdt);
1161     if (ret < 0) {
1162         error_report("couldn't setup memory nodes in fdt");
1163         exit(1);
1164     }
1165 
1166     /* /vdevice */
1167     spapr_dt_vdevice(spapr->vio_bus, fdt);
1168 
1169     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1170         ret = spapr_dt_rng(fdt);
1171         if (ret < 0) {
1172             error_report("could not set up rng device in the fdt");
1173             exit(1);
1174         }
1175     }
1176 
1177     QLIST_FOREACH(phb, &spapr->phbs, list) {
1178         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1179         if (ret < 0) {
1180             error_report("couldn't setup PCI devices in fdt");
1181             exit(1);
1182         }
1183     }
1184 
1185     spapr_dt_cpus(fdt, spapr);
1186 
1187     if (smc->dr_lmb_enabled) {
1188         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1189     }
1190 
1191     if (mc->has_hotpluggable_cpus) {
1192         int offset = fdt_path_offset(fdt, "/cpus");
1193         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1194         if (ret < 0) {
1195             error_report("Couldn't set up CPU DR device tree properties");
1196             exit(1);
1197         }
1198     }
1199 
1200     /* /event-sources */
1201     spapr_dt_events(spapr, fdt);
1202 
1203     /* /rtas */
1204     spapr_dt_rtas(spapr, fdt);
1205 
1206     /* /chosen */
1207     spapr_dt_chosen(spapr, fdt, reset);
1208 
1209     /* /hypervisor */
1210     if (kvm_enabled()) {
1211         spapr_dt_hypervisor(spapr, fdt);
1212     }
1213 
1214     /* Build memory reserve map */
1215     if (reset) {
1216         if (spapr->kernel_size) {
1217             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1218                                   spapr->kernel_size)));
1219         }
1220         if (spapr->initrd_size) {
1221             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1222                                   spapr->initrd_size)));
1223         }
1224     }
1225 
1226     if (smc->dr_phb_enabled) {
1227         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1228         if (ret < 0) {
1229             error_report("Couldn't set up PHB DR device tree properties");
1230             exit(1);
1231         }
1232     }
1233 
1234     /* NVDIMM devices */
1235     if (mc->nvdimm_supported) {
1236         spapr_dt_persistent_memory(spapr, fdt);
1237     }
1238 
1239     return fdt;
1240 }
1241 
1242 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1243 {
1244     SpaprMachineState *spapr = opaque;
1245 
1246     return (addr & 0x0fffffff) + spapr->kernel_addr;
1247 }
1248 
1249 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1250                                     PowerPCCPU *cpu)
1251 {
1252     CPUPPCState *env = &cpu->env;
1253 
1254     /* The TCG path should also be holding the BQL at this point */
1255     g_assert(qemu_mutex_iothread_locked());
1256 
1257     if (msr_pr) {
1258         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1259         env->gpr[3] = H_PRIVILEGE;
1260     } else {
1261         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1262     }
1263 }
1264 
1265 struct LPCRSyncState {
1266     target_ulong value;
1267     target_ulong mask;
1268 };
1269 
1270 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1271 {
1272     struct LPCRSyncState *s = arg.host_ptr;
1273     PowerPCCPU *cpu = POWERPC_CPU(cs);
1274     CPUPPCState *env = &cpu->env;
1275     target_ulong lpcr;
1276 
1277     cpu_synchronize_state(cs);
1278     lpcr = env->spr[SPR_LPCR];
1279     lpcr &= ~s->mask;
1280     lpcr |= s->value;
1281     ppc_store_lpcr(cpu, lpcr);
1282 }
1283 
1284 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1285 {
1286     CPUState *cs;
1287     struct LPCRSyncState s = {
1288         .value = value,
1289         .mask = mask
1290     };
1291     CPU_FOREACH(cs) {
1292         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1293     }
1294 }
1295 
1296 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1297 {
1298     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1299 
1300     /* Copy PATE1:GR into PATE0:HR */
1301     entry->dw0 = spapr->patb_entry & PATE0_HR;
1302     entry->dw1 = spapr->patb_entry;
1303 }
1304 
1305 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1306 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1307 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1308 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1309 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1310 
1311 /*
1312  * Get the fd to access the kernel htab, re-opening it if necessary
1313  */
1314 static int get_htab_fd(SpaprMachineState *spapr)
1315 {
1316     Error *local_err = NULL;
1317 
1318     if (spapr->htab_fd >= 0) {
1319         return spapr->htab_fd;
1320     }
1321 
1322     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1323     if (spapr->htab_fd < 0) {
1324         error_report_err(local_err);
1325     }
1326 
1327     return spapr->htab_fd;
1328 }
1329 
1330 void close_htab_fd(SpaprMachineState *spapr)
1331 {
1332     if (spapr->htab_fd >= 0) {
1333         close(spapr->htab_fd);
1334     }
1335     spapr->htab_fd = -1;
1336 }
1337 
1338 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1339 {
1340     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1341 
1342     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1343 }
1344 
1345 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1346 {
1347     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1348 
1349     assert(kvm_enabled());
1350 
1351     if (!spapr->htab) {
1352         return 0;
1353     }
1354 
1355     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1356 }
1357 
1358 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1359                                                 hwaddr ptex, int n)
1360 {
1361     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1362     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1363 
1364     if (!spapr->htab) {
1365         /*
1366          * HTAB is controlled by KVM. Fetch into temporary buffer
1367          */
1368         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1369         kvmppc_read_hptes(hptes, ptex, n);
1370         return hptes;
1371     }
1372 
1373     /*
1374      * HTAB is controlled by QEMU. Just point to the internally
1375      * accessible PTEG.
1376      */
1377     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1378 }
1379 
1380 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1381                               const ppc_hash_pte64_t *hptes,
1382                               hwaddr ptex, int n)
1383 {
1384     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1385 
1386     if (!spapr->htab) {
1387         g_free((void *)hptes);
1388     }
1389 
1390     /* Nothing to do for qemu managed HPT */
1391 }
1392 
1393 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1394                       uint64_t pte0, uint64_t pte1)
1395 {
1396     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1397     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1398 
1399     if (!spapr->htab) {
1400         kvmppc_write_hpte(ptex, pte0, pte1);
1401     } else {
1402         if (pte0 & HPTE64_V_VALID) {
1403             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1404             /*
1405              * When setting valid, we write PTE1 first. This ensures
1406              * proper synchronization with the reading code in
1407              * ppc_hash64_pteg_search()
1408              */
1409             smp_wmb();
1410             stq_p(spapr->htab + offset, pte0);
1411         } else {
1412             stq_p(spapr->htab + offset, pte0);
1413             /*
1414              * When clearing it we set PTE0 first. This ensures proper
1415              * synchronization with the reading code in
1416              * ppc_hash64_pteg_search()
1417              */
1418             smp_wmb();
1419             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1420         }
1421     }
1422 }
1423 
1424 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1425                              uint64_t pte1)
1426 {
1427     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1428     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1429 
1430     if (!spapr->htab) {
1431         /* There should always be a hash table when this is called */
1432         error_report("spapr_hpte_set_c called with no hash table !");
1433         return;
1434     }
1435 
1436     /* The HW performs a non-atomic byte update */
1437     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1438 }
1439 
1440 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1441                              uint64_t pte1)
1442 {
1443     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1444     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1445 
1446     if (!spapr->htab) {
1447         /* There should always be a hash table when this is called */
1448         error_report("spapr_hpte_set_r called with no hash table !");
1449         return;
1450     }
1451 
1452     /* The HW performs a non-atomic byte update */
1453     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1454 }
1455 
1456 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1457 {
1458     int shift;
1459 
1460     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1461      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1462      * that's much more than is needed for Linux guests */
1463     shift = ctz64(pow2ceil(ramsize)) - 7;
1464     shift = MAX(shift, 18); /* Minimum architected size */
1465     shift = MIN(shift, 46); /* Maximum architected size */
1466     return shift;
1467 }
1468 
1469 void spapr_free_hpt(SpaprMachineState *spapr)
1470 {
1471     g_free(spapr->htab);
1472     spapr->htab = NULL;
1473     spapr->htab_shift = 0;
1474     close_htab_fd(spapr);
1475 }
1476 
1477 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1478                           Error **errp)
1479 {
1480     long rc;
1481 
1482     /* Clean up any HPT info from a previous boot */
1483     spapr_free_hpt(spapr);
1484 
1485     rc = kvmppc_reset_htab(shift);
1486 
1487     if (rc == -EOPNOTSUPP) {
1488         error_setg(errp, "HPT not supported in nested guests");
1489         return;
1490     }
1491 
1492     if (rc < 0) {
1493         /* kernel-side HPT needed, but couldn't allocate one */
1494         error_setg_errno(errp, errno,
1495                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1496                          shift);
1497         /* This is almost certainly fatal, but if the caller really
1498          * wants to carry on with shift == 0, it's welcome to try */
1499     } else if (rc > 0) {
1500         /* kernel-side HPT allocated */
1501         if (rc != shift) {
1502             error_setg(errp,
1503                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1504                        shift, rc);
1505         }
1506 
1507         spapr->htab_shift = shift;
1508         spapr->htab = NULL;
1509     } else {
1510         /* kernel-side HPT not needed, allocate in userspace instead */
1511         size_t size = 1ULL << shift;
1512         int i;
1513 
1514         spapr->htab = qemu_memalign(size, size);
1515         if (!spapr->htab) {
1516             error_setg_errno(errp, errno,
1517                              "Could not allocate HPT of order %d", shift);
1518             return;
1519         }
1520 
1521         memset(spapr->htab, 0, size);
1522         spapr->htab_shift = shift;
1523 
1524         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1525             DIRTY_HPTE(HPTE(spapr->htab, i));
1526         }
1527     }
1528     /* We're setting up a hash table, so that means we're not radix */
1529     spapr->patb_entry = 0;
1530     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1531 }
1532 
1533 void spapr_setup_hpt(SpaprMachineState *spapr)
1534 {
1535     int hpt_shift;
1536 
1537     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1538         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1539     } else {
1540         uint64_t current_ram_size;
1541 
1542         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1543         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1544     }
1545     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1546 
1547     if (kvm_enabled()) {
1548         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1549 
1550         /* Check our RMA fits in the possible VRMA */
1551         if (vrma_limit < spapr->rma_size) {
1552             error_report("Unable to create %" HWADDR_PRIu
1553                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1554                          spapr->rma_size / MiB, vrma_limit / MiB);
1555             exit(EXIT_FAILURE);
1556         }
1557     }
1558 }
1559 
1560 static int spapr_reset_drcs(Object *child, void *opaque)
1561 {
1562     SpaprDrc *drc =
1563         (SpaprDrc *) object_dynamic_cast(child,
1564                                                  TYPE_SPAPR_DR_CONNECTOR);
1565 
1566     if (drc) {
1567         spapr_drc_reset(drc);
1568     }
1569 
1570     return 0;
1571 }
1572 
1573 static void spapr_machine_reset(MachineState *machine)
1574 {
1575     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1576     PowerPCCPU *first_ppc_cpu;
1577     hwaddr fdt_addr;
1578     void *fdt;
1579     int rc;
1580 
1581     kvmppc_svm_off(&error_fatal);
1582     spapr_caps_apply(spapr);
1583 
1584     first_ppc_cpu = POWERPC_CPU(first_cpu);
1585     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1586         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1587                               spapr->max_compat_pvr)) {
1588         /*
1589          * If using KVM with radix mode available, VCPUs can be started
1590          * without a HPT because KVM will start them in radix mode.
1591          * Set the GR bit in PATE so that we know there is no HPT.
1592          */
1593         spapr->patb_entry = PATE1_GR;
1594         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1595     } else {
1596         spapr_setup_hpt(spapr);
1597     }
1598 
1599     qemu_devices_reset();
1600 
1601     spapr_ovec_cleanup(spapr->ov5_cas);
1602     spapr->ov5_cas = spapr_ovec_new();
1603 
1604     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1605 
1606     /*
1607      * This is fixing some of the default configuration of the XIVE
1608      * devices. To be called after the reset of the machine devices.
1609      */
1610     spapr_irq_reset(spapr, &error_fatal);
1611 
1612     /*
1613      * There is no CAS under qtest. Simulate one to please the code that
1614      * depends on spapr->ov5_cas. This is especially needed to test device
1615      * unplug, so we do that before resetting the DRCs.
1616      */
1617     if (qtest_enabled()) {
1618         spapr_ovec_cleanup(spapr->ov5_cas);
1619         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1620     }
1621 
1622     /* DRC reset may cause a device to be unplugged. This will cause troubles
1623      * if this device is used by another device (eg, a running vhost backend
1624      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1625      * situations, we reset DRCs after all devices have been reset.
1626      */
1627     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1628 
1629     spapr_clear_pending_events(spapr);
1630 
1631     /*
1632      * We place the device tree and RTAS just below either the top of the RMA,
1633      * or just below 2GB, whichever is lower, so that it can be
1634      * processed with 32-bit real mode code if necessary
1635      */
1636     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1637 
1638     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1639 
1640     rc = fdt_pack(fdt);
1641 
1642     /* Should only fail if we've built a corrupted tree */
1643     assert(rc == 0);
1644 
1645     /* Load the fdt */
1646     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1647     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1648     g_free(spapr->fdt_blob);
1649     spapr->fdt_size = fdt_totalsize(fdt);
1650     spapr->fdt_initial_size = spapr->fdt_size;
1651     spapr->fdt_blob = fdt;
1652 
1653     /* Set up the entry state */
1654     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1655     first_ppc_cpu->env.gpr[5] = 0;
1656 
1657     spapr->fwnmi_system_reset_addr = -1;
1658     spapr->fwnmi_machine_check_addr = -1;
1659     spapr->fwnmi_machine_check_interlock = -1;
1660 
1661     /* Signal all vCPUs waiting on this condition */
1662     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1663 
1664     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1665 }
1666 
1667 static void spapr_create_nvram(SpaprMachineState *spapr)
1668 {
1669     DeviceState *dev = qdev_new("spapr-nvram");
1670     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1671 
1672     if (dinfo) {
1673         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1674                                 &error_fatal);
1675     }
1676 
1677     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1678 
1679     spapr->nvram = (struct SpaprNvram *)dev;
1680 }
1681 
1682 static void spapr_rtc_create(SpaprMachineState *spapr)
1683 {
1684     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1685                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1686                                        &error_fatal, NULL);
1687     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1688     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1689                               "date");
1690 }
1691 
1692 /* Returns whether we want to use VGA or not */
1693 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1694 {
1695     switch (vga_interface_type) {
1696     case VGA_NONE:
1697         return false;
1698     case VGA_DEVICE:
1699         return true;
1700     case VGA_STD:
1701     case VGA_VIRTIO:
1702     case VGA_CIRRUS:
1703         return pci_vga_init(pci_bus) != NULL;
1704     default:
1705         error_setg(errp,
1706                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1707         return false;
1708     }
1709 }
1710 
1711 static int spapr_pre_load(void *opaque)
1712 {
1713     int rc;
1714 
1715     rc = spapr_caps_pre_load(opaque);
1716     if (rc) {
1717         return rc;
1718     }
1719 
1720     return 0;
1721 }
1722 
1723 static int spapr_post_load(void *opaque, int version_id)
1724 {
1725     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1726     int err = 0;
1727 
1728     err = spapr_caps_post_migration(spapr);
1729     if (err) {
1730         return err;
1731     }
1732 
1733     /*
1734      * In earlier versions, there was no separate qdev for the PAPR
1735      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1736      * So when migrating from those versions, poke the incoming offset
1737      * value into the RTC device
1738      */
1739     if (version_id < 3) {
1740         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1741         if (err) {
1742             return err;
1743         }
1744     }
1745 
1746     if (kvm_enabled() && spapr->patb_entry) {
1747         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1748         bool radix = !!(spapr->patb_entry & PATE1_GR);
1749         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1750 
1751         /*
1752          * Update LPCR:HR and UPRT as they may not be set properly in
1753          * the stream
1754          */
1755         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1756                             LPCR_HR | LPCR_UPRT);
1757 
1758         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1759         if (err) {
1760             error_report("Process table config unsupported by the host");
1761             return -EINVAL;
1762         }
1763     }
1764 
1765     err = spapr_irq_post_load(spapr, version_id);
1766     if (err) {
1767         return err;
1768     }
1769 
1770     return err;
1771 }
1772 
1773 static int spapr_pre_save(void *opaque)
1774 {
1775     int rc;
1776 
1777     rc = spapr_caps_pre_save(opaque);
1778     if (rc) {
1779         return rc;
1780     }
1781 
1782     return 0;
1783 }
1784 
1785 static bool version_before_3(void *opaque, int version_id)
1786 {
1787     return version_id < 3;
1788 }
1789 
1790 static bool spapr_pending_events_needed(void *opaque)
1791 {
1792     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1793     return !QTAILQ_EMPTY(&spapr->pending_events);
1794 }
1795 
1796 static const VMStateDescription vmstate_spapr_event_entry = {
1797     .name = "spapr_event_log_entry",
1798     .version_id = 1,
1799     .minimum_version_id = 1,
1800     .fields = (VMStateField[]) {
1801         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1802         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1803         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1804                                      NULL, extended_length),
1805         VMSTATE_END_OF_LIST()
1806     },
1807 };
1808 
1809 static const VMStateDescription vmstate_spapr_pending_events = {
1810     .name = "spapr_pending_events",
1811     .version_id = 1,
1812     .minimum_version_id = 1,
1813     .needed = spapr_pending_events_needed,
1814     .fields = (VMStateField[]) {
1815         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1816                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1817         VMSTATE_END_OF_LIST()
1818     },
1819 };
1820 
1821 static bool spapr_ov5_cas_needed(void *opaque)
1822 {
1823     SpaprMachineState *spapr = opaque;
1824     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1825     bool cas_needed;
1826 
1827     /* Prior to the introduction of SpaprOptionVector, we had two option
1828      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1829      * Both of these options encode machine topology into the device-tree
1830      * in such a way that the now-booted OS should still be able to interact
1831      * appropriately with QEMU regardless of what options were actually
1832      * negotiatied on the source side.
1833      *
1834      * As such, we can avoid migrating the CAS-negotiated options if these
1835      * are the only options available on the current machine/platform.
1836      * Since these are the only options available for pseries-2.7 and
1837      * earlier, this allows us to maintain old->new/new->old migration
1838      * compatibility.
1839      *
1840      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1841      * via default pseries-2.8 machines and explicit command-line parameters.
1842      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1843      * of the actual CAS-negotiated values to continue working properly. For
1844      * example, availability of memory unplug depends on knowing whether
1845      * OV5_HP_EVT was negotiated via CAS.
1846      *
1847      * Thus, for any cases where the set of available CAS-negotiatable
1848      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1849      * include the CAS-negotiated options in the migration stream, unless
1850      * if they affect boot time behaviour only.
1851      */
1852     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1853     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1854     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1855 
1856     /* We need extra information if we have any bits outside the mask
1857      * defined above */
1858     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1859 
1860     spapr_ovec_cleanup(ov5_mask);
1861 
1862     return cas_needed;
1863 }
1864 
1865 static const VMStateDescription vmstate_spapr_ov5_cas = {
1866     .name = "spapr_option_vector_ov5_cas",
1867     .version_id = 1,
1868     .minimum_version_id = 1,
1869     .needed = spapr_ov5_cas_needed,
1870     .fields = (VMStateField[]) {
1871         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1872                                  vmstate_spapr_ovec, SpaprOptionVector),
1873         VMSTATE_END_OF_LIST()
1874     },
1875 };
1876 
1877 static bool spapr_patb_entry_needed(void *opaque)
1878 {
1879     SpaprMachineState *spapr = opaque;
1880 
1881     return !!spapr->patb_entry;
1882 }
1883 
1884 static const VMStateDescription vmstate_spapr_patb_entry = {
1885     .name = "spapr_patb_entry",
1886     .version_id = 1,
1887     .minimum_version_id = 1,
1888     .needed = spapr_patb_entry_needed,
1889     .fields = (VMStateField[]) {
1890         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1891         VMSTATE_END_OF_LIST()
1892     },
1893 };
1894 
1895 static bool spapr_irq_map_needed(void *opaque)
1896 {
1897     SpaprMachineState *spapr = opaque;
1898 
1899     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1900 }
1901 
1902 static const VMStateDescription vmstate_spapr_irq_map = {
1903     .name = "spapr_irq_map",
1904     .version_id = 1,
1905     .minimum_version_id = 1,
1906     .needed = spapr_irq_map_needed,
1907     .fields = (VMStateField[]) {
1908         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1909         VMSTATE_END_OF_LIST()
1910     },
1911 };
1912 
1913 static bool spapr_dtb_needed(void *opaque)
1914 {
1915     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1916 
1917     return smc->update_dt_enabled;
1918 }
1919 
1920 static int spapr_dtb_pre_load(void *opaque)
1921 {
1922     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1923 
1924     g_free(spapr->fdt_blob);
1925     spapr->fdt_blob = NULL;
1926     spapr->fdt_size = 0;
1927 
1928     return 0;
1929 }
1930 
1931 static const VMStateDescription vmstate_spapr_dtb = {
1932     .name = "spapr_dtb",
1933     .version_id = 1,
1934     .minimum_version_id = 1,
1935     .needed = spapr_dtb_needed,
1936     .pre_load = spapr_dtb_pre_load,
1937     .fields = (VMStateField[]) {
1938         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1939         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1940         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1941                                      fdt_size),
1942         VMSTATE_END_OF_LIST()
1943     },
1944 };
1945 
1946 static bool spapr_fwnmi_needed(void *opaque)
1947 {
1948     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1949 
1950     return spapr->fwnmi_machine_check_addr != -1;
1951 }
1952 
1953 static int spapr_fwnmi_pre_save(void *opaque)
1954 {
1955     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1956 
1957     /*
1958      * Check if machine check handling is in progress and print a
1959      * warning message.
1960      */
1961     if (spapr->fwnmi_machine_check_interlock != -1) {
1962         warn_report("A machine check is being handled during migration. The"
1963                 "handler may run and log hardware error on the destination");
1964     }
1965 
1966     return 0;
1967 }
1968 
1969 static const VMStateDescription vmstate_spapr_fwnmi = {
1970     .name = "spapr_fwnmi",
1971     .version_id = 1,
1972     .minimum_version_id = 1,
1973     .needed = spapr_fwnmi_needed,
1974     .pre_save = spapr_fwnmi_pre_save,
1975     .fields = (VMStateField[]) {
1976         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1977         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1978         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1979         VMSTATE_END_OF_LIST()
1980     },
1981 };
1982 
1983 static const VMStateDescription vmstate_spapr = {
1984     .name = "spapr",
1985     .version_id = 3,
1986     .minimum_version_id = 1,
1987     .pre_load = spapr_pre_load,
1988     .post_load = spapr_post_load,
1989     .pre_save = spapr_pre_save,
1990     .fields = (VMStateField[]) {
1991         /* used to be @next_irq */
1992         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1993 
1994         /* RTC offset */
1995         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
1996 
1997         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
1998         VMSTATE_END_OF_LIST()
1999     },
2000     .subsections = (const VMStateDescription*[]) {
2001         &vmstate_spapr_ov5_cas,
2002         &vmstate_spapr_patb_entry,
2003         &vmstate_spapr_pending_events,
2004         &vmstate_spapr_cap_htm,
2005         &vmstate_spapr_cap_vsx,
2006         &vmstate_spapr_cap_dfp,
2007         &vmstate_spapr_cap_cfpc,
2008         &vmstate_spapr_cap_sbbc,
2009         &vmstate_spapr_cap_ibs,
2010         &vmstate_spapr_cap_hpt_maxpagesize,
2011         &vmstate_spapr_irq_map,
2012         &vmstate_spapr_cap_nested_kvm_hv,
2013         &vmstate_spapr_dtb,
2014         &vmstate_spapr_cap_large_decr,
2015         &vmstate_spapr_cap_ccf_assist,
2016         &vmstate_spapr_cap_fwnmi,
2017         &vmstate_spapr_fwnmi,
2018         NULL
2019     }
2020 };
2021 
2022 static int htab_save_setup(QEMUFile *f, void *opaque)
2023 {
2024     SpaprMachineState *spapr = opaque;
2025 
2026     /* "Iteration" header */
2027     if (!spapr->htab_shift) {
2028         qemu_put_be32(f, -1);
2029     } else {
2030         qemu_put_be32(f, spapr->htab_shift);
2031     }
2032 
2033     if (spapr->htab) {
2034         spapr->htab_save_index = 0;
2035         spapr->htab_first_pass = true;
2036     } else {
2037         if (spapr->htab_shift) {
2038             assert(kvm_enabled());
2039         }
2040     }
2041 
2042 
2043     return 0;
2044 }
2045 
2046 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2047                             int chunkstart, int n_valid, int n_invalid)
2048 {
2049     qemu_put_be32(f, chunkstart);
2050     qemu_put_be16(f, n_valid);
2051     qemu_put_be16(f, n_invalid);
2052     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2053                     HASH_PTE_SIZE_64 * n_valid);
2054 }
2055 
2056 static void htab_save_end_marker(QEMUFile *f)
2057 {
2058     qemu_put_be32(f, 0);
2059     qemu_put_be16(f, 0);
2060     qemu_put_be16(f, 0);
2061 }
2062 
2063 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2064                                  int64_t max_ns)
2065 {
2066     bool has_timeout = max_ns != -1;
2067     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2068     int index = spapr->htab_save_index;
2069     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2070 
2071     assert(spapr->htab_first_pass);
2072 
2073     do {
2074         int chunkstart;
2075 
2076         /* Consume invalid HPTEs */
2077         while ((index < htabslots)
2078                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2079             CLEAN_HPTE(HPTE(spapr->htab, index));
2080             index++;
2081         }
2082 
2083         /* Consume valid HPTEs */
2084         chunkstart = index;
2085         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2086                && HPTE_VALID(HPTE(spapr->htab, index))) {
2087             CLEAN_HPTE(HPTE(spapr->htab, index));
2088             index++;
2089         }
2090 
2091         if (index > chunkstart) {
2092             int n_valid = index - chunkstart;
2093 
2094             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2095 
2096             if (has_timeout &&
2097                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2098                 break;
2099             }
2100         }
2101     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2102 
2103     if (index >= htabslots) {
2104         assert(index == htabslots);
2105         index = 0;
2106         spapr->htab_first_pass = false;
2107     }
2108     spapr->htab_save_index = index;
2109 }
2110 
2111 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2112                                 int64_t max_ns)
2113 {
2114     bool final = max_ns < 0;
2115     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2116     int examined = 0, sent = 0;
2117     int index = spapr->htab_save_index;
2118     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2119 
2120     assert(!spapr->htab_first_pass);
2121 
2122     do {
2123         int chunkstart, invalidstart;
2124 
2125         /* Consume non-dirty HPTEs */
2126         while ((index < htabslots)
2127                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2128             index++;
2129             examined++;
2130         }
2131 
2132         chunkstart = index;
2133         /* Consume valid dirty HPTEs */
2134         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2135                && HPTE_DIRTY(HPTE(spapr->htab, index))
2136                && HPTE_VALID(HPTE(spapr->htab, index))) {
2137             CLEAN_HPTE(HPTE(spapr->htab, index));
2138             index++;
2139             examined++;
2140         }
2141 
2142         invalidstart = index;
2143         /* Consume invalid dirty HPTEs */
2144         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2145                && HPTE_DIRTY(HPTE(spapr->htab, index))
2146                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2147             CLEAN_HPTE(HPTE(spapr->htab, index));
2148             index++;
2149             examined++;
2150         }
2151 
2152         if (index > chunkstart) {
2153             int n_valid = invalidstart - chunkstart;
2154             int n_invalid = index - invalidstart;
2155 
2156             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2157             sent += index - chunkstart;
2158 
2159             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2160                 break;
2161             }
2162         }
2163 
2164         if (examined >= htabslots) {
2165             break;
2166         }
2167 
2168         if (index >= htabslots) {
2169             assert(index == htabslots);
2170             index = 0;
2171         }
2172     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2173 
2174     if (index >= htabslots) {
2175         assert(index == htabslots);
2176         index = 0;
2177     }
2178 
2179     spapr->htab_save_index = index;
2180 
2181     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2182 }
2183 
2184 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2185 #define MAX_KVM_BUF_SIZE    2048
2186 
2187 static int htab_save_iterate(QEMUFile *f, void *opaque)
2188 {
2189     SpaprMachineState *spapr = opaque;
2190     int fd;
2191     int rc = 0;
2192 
2193     /* Iteration header */
2194     if (!spapr->htab_shift) {
2195         qemu_put_be32(f, -1);
2196         return 1;
2197     } else {
2198         qemu_put_be32(f, 0);
2199     }
2200 
2201     if (!spapr->htab) {
2202         assert(kvm_enabled());
2203 
2204         fd = get_htab_fd(spapr);
2205         if (fd < 0) {
2206             return fd;
2207         }
2208 
2209         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2210         if (rc < 0) {
2211             return rc;
2212         }
2213     } else  if (spapr->htab_first_pass) {
2214         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2215     } else {
2216         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2217     }
2218 
2219     htab_save_end_marker(f);
2220 
2221     return rc;
2222 }
2223 
2224 static int htab_save_complete(QEMUFile *f, void *opaque)
2225 {
2226     SpaprMachineState *spapr = opaque;
2227     int fd;
2228 
2229     /* Iteration header */
2230     if (!spapr->htab_shift) {
2231         qemu_put_be32(f, -1);
2232         return 0;
2233     } else {
2234         qemu_put_be32(f, 0);
2235     }
2236 
2237     if (!spapr->htab) {
2238         int rc;
2239 
2240         assert(kvm_enabled());
2241 
2242         fd = get_htab_fd(spapr);
2243         if (fd < 0) {
2244             return fd;
2245         }
2246 
2247         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2248         if (rc < 0) {
2249             return rc;
2250         }
2251     } else {
2252         if (spapr->htab_first_pass) {
2253             htab_save_first_pass(f, spapr, -1);
2254         }
2255         htab_save_later_pass(f, spapr, -1);
2256     }
2257 
2258     /* End marker */
2259     htab_save_end_marker(f);
2260 
2261     return 0;
2262 }
2263 
2264 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2265 {
2266     SpaprMachineState *spapr = opaque;
2267     uint32_t section_hdr;
2268     int fd = -1;
2269     Error *local_err = NULL;
2270 
2271     if (version_id < 1 || version_id > 1) {
2272         error_report("htab_load() bad version");
2273         return -EINVAL;
2274     }
2275 
2276     section_hdr = qemu_get_be32(f);
2277 
2278     if (section_hdr == -1) {
2279         spapr_free_hpt(spapr);
2280         return 0;
2281     }
2282 
2283     if (section_hdr) {
2284         /* First section gives the htab size */
2285         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2286         if (local_err) {
2287             error_report_err(local_err);
2288             return -EINVAL;
2289         }
2290         return 0;
2291     }
2292 
2293     if (!spapr->htab) {
2294         assert(kvm_enabled());
2295 
2296         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2297         if (fd < 0) {
2298             error_report_err(local_err);
2299             return fd;
2300         }
2301     }
2302 
2303     while (true) {
2304         uint32_t index;
2305         uint16_t n_valid, n_invalid;
2306 
2307         index = qemu_get_be32(f);
2308         n_valid = qemu_get_be16(f);
2309         n_invalid = qemu_get_be16(f);
2310 
2311         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2312             /* End of Stream */
2313             break;
2314         }
2315 
2316         if ((index + n_valid + n_invalid) >
2317             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2318             /* Bad index in stream */
2319             error_report(
2320                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2321                 index, n_valid, n_invalid, spapr->htab_shift);
2322             return -EINVAL;
2323         }
2324 
2325         if (spapr->htab) {
2326             if (n_valid) {
2327                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2328                                 HASH_PTE_SIZE_64 * n_valid);
2329             }
2330             if (n_invalid) {
2331                 memset(HPTE(spapr->htab, index + n_valid), 0,
2332                        HASH_PTE_SIZE_64 * n_invalid);
2333             }
2334         } else {
2335             int rc;
2336 
2337             assert(fd >= 0);
2338 
2339             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2340             if (rc < 0) {
2341                 return rc;
2342             }
2343         }
2344     }
2345 
2346     if (!spapr->htab) {
2347         assert(fd >= 0);
2348         close(fd);
2349     }
2350 
2351     return 0;
2352 }
2353 
2354 static void htab_save_cleanup(void *opaque)
2355 {
2356     SpaprMachineState *spapr = opaque;
2357 
2358     close_htab_fd(spapr);
2359 }
2360 
2361 static SaveVMHandlers savevm_htab_handlers = {
2362     .save_setup = htab_save_setup,
2363     .save_live_iterate = htab_save_iterate,
2364     .save_live_complete_precopy = htab_save_complete,
2365     .save_cleanup = htab_save_cleanup,
2366     .load_state = htab_load,
2367 };
2368 
2369 static void spapr_boot_set(void *opaque, const char *boot_device,
2370                            Error **errp)
2371 {
2372     MachineState *machine = MACHINE(opaque);
2373     machine->boot_order = g_strdup(boot_device);
2374 }
2375 
2376 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2377 {
2378     MachineState *machine = MACHINE(spapr);
2379     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2380     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2381     int i;
2382 
2383     for (i = 0; i < nr_lmbs; i++) {
2384         uint64_t addr;
2385 
2386         addr = i * lmb_size + machine->device_memory->base;
2387         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2388                                addr / lmb_size);
2389     }
2390 }
2391 
2392 /*
2393  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2394  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2395  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2396  */
2397 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2398 {
2399     int i;
2400 
2401     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2402         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2403                    " is not aligned to %" PRIu64 " MiB",
2404                    machine->ram_size,
2405                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2406         return;
2407     }
2408 
2409     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2410         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2411                    " is not aligned to %" PRIu64 " MiB",
2412                    machine->ram_size,
2413                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2414         return;
2415     }
2416 
2417     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2418         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2419             error_setg(errp,
2420                        "Node %d memory size 0x%" PRIx64
2421                        " is not aligned to %" PRIu64 " MiB",
2422                        i, machine->numa_state->nodes[i].node_mem,
2423                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2424             return;
2425         }
2426     }
2427 }
2428 
2429 /* find cpu slot in machine->possible_cpus by core_id */
2430 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2431 {
2432     int index = id / ms->smp.threads;
2433 
2434     if (index >= ms->possible_cpus->len) {
2435         return NULL;
2436     }
2437     if (idx) {
2438         *idx = index;
2439     }
2440     return &ms->possible_cpus->cpus[index];
2441 }
2442 
2443 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2444 {
2445     MachineState *ms = MACHINE(spapr);
2446     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2447     Error *local_err = NULL;
2448     bool vsmt_user = !!spapr->vsmt;
2449     int kvm_smt = kvmppc_smt_threads();
2450     int ret;
2451     unsigned int smp_threads = ms->smp.threads;
2452 
2453     if (!kvm_enabled() && (smp_threads > 1)) {
2454         error_setg(errp, "TCG cannot support more than 1 thread/core "
2455                    "on a pseries machine");
2456         return;
2457     }
2458     if (!is_power_of_2(smp_threads)) {
2459         error_setg(errp, "Cannot support %d threads/core on a pseries "
2460                    "machine because it must be a power of 2", smp_threads);
2461         return;
2462     }
2463 
2464     /* Detemine the VSMT mode to use: */
2465     if (vsmt_user) {
2466         if (spapr->vsmt < smp_threads) {
2467             error_setg(errp, "Cannot support VSMT mode %d"
2468                        " because it must be >= threads/core (%d)",
2469                        spapr->vsmt, smp_threads);
2470             return;
2471         }
2472         /* In this case, spapr->vsmt has been set by the command line */
2473     } else if (!smc->smp_threads_vsmt) {
2474         /*
2475          * Default VSMT value is tricky, because we need it to be as
2476          * consistent as possible (for migration), but this requires
2477          * changing it for at least some existing cases.  We pick 8 as
2478          * the value that we'd get with KVM on POWER8, the
2479          * overwhelmingly common case in production systems.
2480          */
2481         spapr->vsmt = MAX(8, smp_threads);
2482     } else {
2483         spapr->vsmt = smp_threads;
2484     }
2485 
2486     /* KVM: If necessary, set the SMT mode: */
2487     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2488         ret = kvmppc_set_smt_threads(spapr->vsmt);
2489         if (ret) {
2490             /* Looks like KVM isn't able to change VSMT mode */
2491             error_setg(&local_err,
2492                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2493                        spapr->vsmt, ret);
2494             /* We can live with that if the default one is big enough
2495              * for the number of threads, and a submultiple of the one
2496              * we want.  In this case we'll waste some vcpu ids, but
2497              * behaviour will be correct */
2498             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2499                 warn_report_err(local_err);
2500             } else {
2501                 if (!vsmt_user) {
2502                     error_append_hint(&local_err,
2503                                       "On PPC, a VM with %d threads/core"
2504                                       " on a host with %d threads/core"
2505                                       " requires the use of VSMT mode %d.\n",
2506                                       smp_threads, kvm_smt, spapr->vsmt);
2507                 }
2508                 kvmppc_error_append_smt_possible_hint(&local_err);
2509                 error_propagate(errp, local_err);
2510             }
2511         }
2512     }
2513     /* else TCG: nothing to do currently */
2514 }
2515 
2516 static void spapr_init_cpus(SpaprMachineState *spapr)
2517 {
2518     MachineState *machine = MACHINE(spapr);
2519     MachineClass *mc = MACHINE_GET_CLASS(machine);
2520     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2521     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2522     const CPUArchIdList *possible_cpus;
2523     unsigned int smp_cpus = machine->smp.cpus;
2524     unsigned int smp_threads = machine->smp.threads;
2525     unsigned int max_cpus = machine->smp.max_cpus;
2526     int boot_cores_nr = smp_cpus / smp_threads;
2527     int i;
2528 
2529     possible_cpus = mc->possible_cpu_arch_ids(machine);
2530     if (mc->has_hotpluggable_cpus) {
2531         if (smp_cpus % smp_threads) {
2532             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2533                          smp_cpus, smp_threads);
2534             exit(1);
2535         }
2536         if (max_cpus % smp_threads) {
2537             error_report("max_cpus (%u) must be multiple of threads (%u)",
2538                          max_cpus, smp_threads);
2539             exit(1);
2540         }
2541     } else {
2542         if (max_cpus != smp_cpus) {
2543             error_report("This machine version does not support CPU hotplug");
2544             exit(1);
2545         }
2546         boot_cores_nr = possible_cpus->len;
2547     }
2548 
2549     if (smc->pre_2_10_has_unused_icps) {
2550         int i;
2551 
2552         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2553             /* Dummy entries get deregistered when real ICPState objects
2554              * are registered during CPU core hotplug.
2555              */
2556             pre_2_10_vmstate_register_dummy_icp(i);
2557         }
2558     }
2559 
2560     for (i = 0; i < possible_cpus->len; i++) {
2561         int core_id = i * smp_threads;
2562 
2563         if (mc->has_hotpluggable_cpus) {
2564             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2565                                    spapr_vcpu_id(spapr, core_id));
2566         }
2567 
2568         if (i < boot_cores_nr) {
2569             Object *core  = object_new(type);
2570             int nr_threads = smp_threads;
2571 
2572             /* Handle the partially filled core for older machine types */
2573             if ((i + 1) * smp_threads >= smp_cpus) {
2574                 nr_threads = smp_cpus - i * smp_threads;
2575             }
2576 
2577             object_property_set_int(core, "nr-threads", nr_threads,
2578                                     &error_fatal);
2579             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2580                                     &error_fatal);
2581             qdev_realize(DEVICE(core), NULL, &error_fatal);
2582 
2583             object_unref(core);
2584         }
2585     }
2586 }
2587 
2588 static PCIHostState *spapr_create_default_phb(void)
2589 {
2590     DeviceState *dev;
2591 
2592     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2593     qdev_prop_set_uint32(dev, "index", 0);
2594     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2595 
2596     return PCI_HOST_BRIDGE(dev);
2597 }
2598 
2599 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2600 {
2601     MachineState *machine = MACHINE(spapr);
2602     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2603     hwaddr rma_size = machine->ram_size;
2604     hwaddr node0_size = spapr_node0_size(machine);
2605 
2606     /* RMA has to fit in the first NUMA node */
2607     rma_size = MIN(rma_size, node0_size);
2608 
2609     /*
2610      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2611      * never exceed that
2612      */
2613     rma_size = MIN(rma_size, 1 * TiB);
2614 
2615     /*
2616      * Clamp the RMA size based on machine type.  This is for
2617      * migration compatibility with older qemu versions, which limited
2618      * the RMA size for complicated and mostly bad reasons.
2619      */
2620     if (smc->rma_limit) {
2621         rma_size = MIN(rma_size, smc->rma_limit);
2622     }
2623 
2624     if (rma_size < MIN_RMA_SLOF) {
2625         error_setg(errp,
2626                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2627                    "ldMiB guest RMA (Real Mode Area memory)",
2628                    MIN_RMA_SLOF / MiB);
2629         return 0;
2630     }
2631 
2632     return rma_size;
2633 }
2634 
2635 /* pSeries LPAR / sPAPR hardware init */
2636 static void spapr_machine_init(MachineState *machine)
2637 {
2638     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2639     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2640     MachineClass *mc = MACHINE_GET_CLASS(machine);
2641     const char *kernel_filename = machine->kernel_filename;
2642     const char *initrd_filename = machine->initrd_filename;
2643     PCIHostState *phb;
2644     int i;
2645     MemoryRegion *sysmem = get_system_memory();
2646     long load_limit, fw_size;
2647     char *filename;
2648     Error *resize_hpt_err = NULL;
2649 
2650     msi_nonbroken = true;
2651 
2652     QLIST_INIT(&spapr->phbs);
2653     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2654 
2655     /* Determine capabilities to run with */
2656     spapr_caps_init(spapr);
2657 
2658     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2659     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2660         /*
2661          * If the user explicitly requested a mode we should either
2662          * supply it, or fail completely (which we do below).  But if
2663          * it's not set explicitly, we reset our mode to something
2664          * that works
2665          */
2666         if (resize_hpt_err) {
2667             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2668             error_free(resize_hpt_err);
2669             resize_hpt_err = NULL;
2670         } else {
2671             spapr->resize_hpt = smc->resize_hpt_default;
2672         }
2673     }
2674 
2675     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2676 
2677     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2678         /*
2679          * User requested HPT resize, but this host can't supply it.  Bail out
2680          */
2681         error_report_err(resize_hpt_err);
2682         exit(1);
2683     }
2684     error_free(resize_hpt_err);
2685 
2686     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2687 
2688     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2689     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2690 
2691     /*
2692      * VSMT must be set in order to be able to compute VCPU ids, ie to
2693      * call spapr_max_server_number() or spapr_vcpu_id().
2694      */
2695     spapr_set_vsmt_mode(spapr, &error_fatal);
2696 
2697     /* Set up Interrupt Controller before we create the VCPUs */
2698     spapr_irq_init(spapr, &error_fatal);
2699 
2700     /* Set up containers for ibm,client-architecture-support negotiated options
2701      */
2702     spapr->ov5 = spapr_ovec_new();
2703     spapr->ov5_cas = spapr_ovec_new();
2704 
2705     if (smc->dr_lmb_enabled) {
2706         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2707         spapr_validate_node_memory(machine, &error_fatal);
2708     }
2709 
2710     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2711 
2712     /* advertise support for dedicated HP event source to guests */
2713     if (spapr->use_hotplug_event_source) {
2714         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2715     }
2716 
2717     /* advertise support for HPT resizing */
2718     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2719         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2720     }
2721 
2722     /* advertise support for ibm,dyamic-memory-v2 */
2723     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2724 
2725     /* advertise XIVE on POWER9 machines */
2726     if (spapr->irq->xive) {
2727         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2728     }
2729 
2730     /* init CPUs */
2731     spapr_init_cpus(spapr);
2732 
2733     /*
2734      * check we don't have a memory-less/cpu-less NUMA node
2735      * Firmware relies on the existing memory/cpu topology to provide the
2736      * NUMA topology to the kernel.
2737      * And the linux kernel needs to know the NUMA topology at start
2738      * to be able to hotplug CPUs later.
2739      */
2740     if (machine->numa_state->num_nodes) {
2741         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2742             /* check for memory-less node */
2743             if (machine->numa_state->nodes[i].node_mem == 0) {
2744                 CPUState *cs;
2745                 int found = 0;
2746                 /* check for cpu-less node */
2747                 CPU_FOREACH(cs) {
2748                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2749                     if (cpu->node_id == i) {
2750                         found = 1;
2751                         break;
2752                     }
2753                 }
2754                 /* memory-less and cpu-less node */
2755                 if (!found) {
2756                     error_report(
2757                        "Memory-less/cpu-less nodes are not supported (node %d)",
2758                                  i);
2759                     exit(1);
2760                 }
2761             }
2762         }
2763 
2764     }
2765 
2766     /*
2767      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2768      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2769      * called from vPHB reset handler so we initialize the counter here.
2770      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2771      * must be equally distant from any other node.
2772      * The final value of spapr->gpu_numa_id is going to be written to
2773      * max-associativity-domains in spapr_build_fdt().
2774      */
2775     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2776 
2777     /* Init numa_assoc_array */
2778     spapr_numa_associativity_init(spapr, machine);
2779 
2780     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2781         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2782                               spapr->max_compat_pvr)) {
2783         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2784         /* KVM and TCG always allow GTSE with radix... */
2785         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2786     }
2787     /* ... but not with hash (currently). */
2788 
2789     if (kvm_enabled()) {
2790         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2791         kvmppc_enable_logical_ci_hcalls();
2792         kvmppc_enable_set_mode_hcall();
2793 
2794         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2795         kvmppc_enable_clear_ref_mod_hcalls();
2796 
2797         /* Enable H_PAGE_INIT */
2798         kvmppc_enable_h_page_init();
2799     }
2800 
2801     /* map RAM */
2802     memory_region_add_subregion(sysmem, 0, machine->ram);
2803 
2804     /* always allocate the device memory information */
2805     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2806 
2807     /* initialize hotplug memory address space */
2808     if (machine->ram_size < machine->maxram_size) {
2809         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2810         /*
2811          * Limit the number of hotpluggable memory slots to half the number
2812          * slots that KVM supports, leaving the other half for PCI and other
2813          * devices. However ensure that number of slots doesn't drop below 32.
2814          */
2815         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2816                            SPAPR_MAX_RAM_SLOTS;
2817 
2818         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2819             max_memslots = SPAPR_MAX_RAM_SLOTS;
2820         }
2821         if (machine->ram_slots > max_memslots) {
2822             error_report("Specified number of memory slots %"
2823                          PRIu64" exceeds max supported %d",
2824                          machine->ram_slots, max_memslots);
2825             exit(1);
2826         }
2827 
2828         machine->device_memory->base = ROUND_UP(machine->ram_size,
2829                                                 SPAPR_DEVICE_MEM_ALIGN);
2830         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2831                            "device-memory", device_mem_size);
2832         memory_region_add_subregion(sysmem, machine->device_memory->base,
2833                                     &machine->device_memory->mr);
2834     }
2835 
2836     if (smc->dr_lmb_enabled) {
2837         spapr_create_lmb_dr_connectors(spapr);
2838     }
2839 
2840     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2841         /* Create the error string for live migration blocker */
2842         error_setg(&spapr->fwnmi_migration_blocker,
2843             "A machine check is being handled during migration. The handler"
2844             "may run and log hardware error on the destination");
2845     }
2846 
2847     if (mc->nvdimm_supported) {
2848         spapr_create_nvdimm_dr_connectors(spapr);
2849     }
2850 
2851     /* Set up RTAS event infrastructure */
2852     spapr_events_init(spapr);
2853 
2854     /* Set up the RTC RTAS interfaces */
2855     spapr_rtc_create(spapr);
2856 
2857     /* Set up VIO bus */
2858     spapr->vio_bus = spapr_vio_bus_init();
2859 
2860     for (i = 0; i < serial_max_hds(); i++) {
2861         if (serial_hd(i)) {
2862             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2863         }
2864     }
2865 
2866     /* We always have at least the nvram device on VIO */
2867     spapr_create_nvram(spapr);
2868 
2869     /*
2870      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2871      * connectors (described in root DT node's "ibm,drc-types" property)
2872      * are pre-initialized here. additional child connectors (such as
2873      * connectors for a PHBs PCI slots) are added as needed during their
2874      * parent's realization.
2875      */
2876     if (smc->dr_phb_enabled) {
2877         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2878             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2879         }
2880     }
2881 
2882     /* Set up PCI */
2883     spapr_pci_rtas_init();
2884 
2885     phb = spapr_create_default_phb();
2886 
2887     for (i = 0; i < nb_nics; i++) {
2888         NICInfo *nd = &nd_table[i];
2889 
2890         if (!nd->model) {
2891             nd->model = g_strdup("spapr-vlan");
2892         }
2893 
2894         if (g_str_equal(nd->model, "spapr-vlan") ||
2895             g_str_equal(nd->model, "ibmveth")) {
2896             spapr_vlan_create(spapr->vio_bus, nd);
2897         } else {
2898             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2899         }
2900     }
2901 
2902     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2903         spapr_vscsi_create(spapr->vio_bus);
2904     }
2905 
2906     /* Graphics */
2907     if (spapr_vga_init(phb->bus, &error_fatal)) {
2908         spapr->has_graphics = true;
2909         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2910     }
2911 
2912     if (machine->usb) {
2913         if (smc->use_ohci_by_default) {
2914             pci_create_simple(phb->bus, -1, "pci-ohci");
2915         } else {
2916             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2917         }
2918 
2919         if (spapr->has_graphics) {
2920             USBBus *usb_bus = usb_bus_find(-1);
2921 
2922             usb_create_simple(usb_bus, "usb-kbd");
2923             usb_create_simple(usb_bus, "usb-mouse");
2924         }
2925     }
2926 
2927     if (kernel_filename) {
2928         spapr->kernel_size = load_elf(kernel_filename, NULL,
2929                                       translate_kernel_address, spapr,
2930                                       NULL, NULL, NULL, NULL, 1,
2931                                       PPC_ELF_MACHINE, 0, 0);
2932         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2933             spapr->kernel_size = load_elf(kernel_filename, NULL,
2934                                           translate_kernel_address, spapr,
2935                                           NULL, NULL, NULL, NULL, 0,
2936                                           PPC_ELF_MACHINE, 0, 0);
2937             spapr->kernel_le = spapr->kernel_size > 0;
2938         }
2939         if (spapr->kernel_size < 0) {
2940             error_report("error loading %s: %s", kernel_filename,
2941                          load_elf_strerror(spapr->kernel_size));
2942             exit(1);
2943         }
2944 
2945         /* load initrd */
2946         if (initrd_filename) {
2947             /* Try to locate the initrd in the gap between the kernel
2948              * and the firmware. Add a bit of space just in case
2949              */
2950             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2951                                   + 0x1ffff) & ~0xffff;
2952             spapr->initrd_size = load_image_targphys(initrd_filename,
2953                                                      spapr->initrd_base,
2954                                                      load_limit
2955                                                      - spapr->initrd_base);
2956             if (spapr->initrd_size < 0) {
2957                 error_report("could not load initial ram disk '%s'",
2958                              initrd_filename);
2959                 exit(1);
2960             }
2961         }
2962     }
2963 
2964     if (bios_name == NULL) {
2965         bios_name = FW_FILE_NAME;
2966     }
2967     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2968     if (!filename) {
2969         error_report("Could not find LPAR firmware '%s'", bios_name);
2970         exit(1);
2971     }
2972     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2973     if (fw_size <= 0) {
2974         error_report("Could not load LPAR firmware '%s'", filename);
2975         exit(1);
2976     }
2977     g_free(filename);
2978 
2979     /* FIXME: Should register things through the MachineState's qdev
2980      * interface, this is a legacy from the sPAPREnvironment structure
2981      * which predated MachineState but had a similar function */
2982     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2983     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2984                          &savevm_htab_handlers, spapr);
2985 
2986     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2987 
2988     qemu_register_boot_set(spapr_boot_set, spapr);
2989 
2990     /*
2991      * Nothing needs to be done to resume a suspended guest because
2992      * suspending does not change the machine state, so no need for
2993      * a ->wakeup method.
2994      */
2995     qemu_register_wakeup_support();
2996 
2997     if (kvm_enabled()) {
2998         /* to stop and start vmclock */
2999         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3000                                          &spapr->tb);
3001 
3002         kvmppc_spapr_enable_inkernel_multitce();
3003     }
3004 
3005     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3006 }
3007 
3008 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3009 {
3010     if (!vm_type) {
3011         return 0;
3012     }
3013 
3014     if (!strcmp(vm_type, "HV")) {
3015         return 1;
3016     }
3017 
3018     if (!strcmp(vm_type, "PR")) {
3019         return 2;
3020     }
3021 
3022     error_report("Unknown kvm-type specified '%s'", vm_type);
3023     exit(1);
3024 }
3025 
3026 /*
3027  * Implementation of an interface to adjust firmware path
3028  * for the bootindex property handling.
3029  */
3030 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3031                                    DeviceState *dev)
3032 {
3033 #define CAST(type, obj, name) \
3034     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3035     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3036     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3037     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3038 
3039     if (d) {
3040         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3041         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3042         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3043 
3044         if (spapr) {
3045             /*
3046              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3047              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3048              * 0x8000 | (target << 8) | (bus << 5) | lun
3049              * (see the "Logical unit addressing format" table in SAM5)
3050              */
3051             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3052             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3053                                    (uint64_t)id << 48);
3054         } else if (virtio) {
3055             /*
3056              * We use SRP luns of the form 01000000 | (target << 8) | lun
3057              * in the top 32 bits of the 64-bit LUN
3058              * Note: the quote above is from SLOF and it is wrong,
3059              * the actual binding is:
3060              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3061              */
3062             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3063             if (d->lun >= 256) {
3064                 /* Use the LUN "flat space addressing method" */
3065                 id |= 0x4000;
3066             }
3067             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3068                                    (uint64_t)id << 32);
3069         } else if (usb) {
3070             /*
3071              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3072              * in the top 32 bits of the 64-bit LUN
3073              */
3074             unsigned usb_port = atoi(usb->port->path);
3075             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3076             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3077                                    (uint64_t)id << 32);
3078         }
3079     }
3080 
3081     /*
3082      * SLOF probes the USB devices, and if it recognizes that the device is a
3083      * storage device, it changes its name to "storage" instead of "usb-host",
3084      * and additionally adds a child node for the SCSI LUN, so the correct
3085      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3086      */
3087     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3088         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3089         if (usb_host_dev_is_scsi_storage(usbdev)) {
3090             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3091         }
3092     }
3093 
3094     if (phb) {
3095         /* Replace "pci" with "pci@800000020000000" */
3096         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3097     }
3098 
3099     if (vsc) {
3100         /* Same logic as virtio above */
3101         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3102         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3103     }
3104 
3105     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3106         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3107         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3108         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3109     }
3110 
3111     return NULL;
3112 }
3113 
3114 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3115 {
3116     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3117 
3118     return g_strdup(spapr->kvm_type);
3119 }
3120 
3121 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3122 {
3123     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3124 
3125     g_free(spapr->kvm_type);
3126     spapr->kvm_type = g_strdup(value);
3127 }
3128 
3129 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3130 {
3131     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3132 
3133     return spapr->use_hotplug_event_source;
3134 }
3135 
3136 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3137                                             Error **errp)
3138 {
3139     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3140 
3141     spapr->use_hotplug_event_source = value;
3142 }
3143 
3144 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3145 {
3146     return true;
3147 }
3148 
3149 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3150 {
3151     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3152 
3153     switch (spapr->resize_hpt) {
3154     case SPAPR_RESIZE_HPT_DEFAULT:
3155         return g_strdup("default");
3156     case SPAPR_RESIZE_HPT_DISABLED:
3157         return g_strdup("disabled");
3158     case SPAPR_RESIZE_HPT_ENABLED:
3159         return g_strdup("enabled");
3160     case SPAPR_RESIZE_HPT_REQUIRED:
3161         return g_strdup("required");
3162     }
3163     g_assert_not_reached();
3164 }
3165 
3166 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3167 {
3168     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3169 
3170     if (strcmp(value, "default") == 0) {
3171         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3172     } else if (strcmp(value, "disabled") == 0) {
3173         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3174     } else if (strcmp(value, "enabled") == 0) {
3175         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3176     } else if (strcmp(value, "required") == 0) {
3177         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3178     } else {
3179         error_setg(errp, "Bad value for \"resize-hpt\" property");
3180     }
3181 }
3182 
3183 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3184 {
3185     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3186 
3187     if (spapr->irq == &spapr_irq_xics_legacy) {
3188         return g_strdup("legacy");
3189     } else if (spapr->irq == &spapr_irq_xics) {
3190         return g_strdup("xics");
3191     } else if (spapr->irq == &spapr_irq_xive) {
3192         return g_strdup("xive");
3193     } else if (spapr->irq == &spapr_irq_dual) {
3194         return g_strdup("dual");
3195     }
3196     g_assert_not_reached();
3197 }
3198 
3199 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3200 {
3201     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3202 
3203     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3204         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3205         return;
3206     }
3207 
3208     /* The legacy IRQ backend can not be set */
3209     if (strcmp(value, "xics") == 0) {
3210         spapr->irq = &spapr_irq_xics;
3211     } else if (strcmp(value, "xive") == 0) {
3212         spapr->irq = &spapr_irq_xive;
3213     } else if (strcmp(value, "dual") == 0) {
3214         spapr->irq = &spapr_irq_dual;
3215     } else {
3216         error_setg(errp, "Bad value for \"ic-mode\" property");
3217     }
3218 }
3219 
3220 static char *spapr_get_host_model(Object *obj, Error **errp)
3221 {
3222     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3223 
3224     return g_strdup(spapr->host_model);
3225 }
3226 
3227 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3228 {
3229     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3230 
3231     g_free(spapr->host_model);
3232     spapr->host_model = g_strdup(value);
3233 }
3234 
3235 static char *spapr_get_host_serial(Object *obj, Error **errp)
3236 {
3237     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3238 
3239     return g_strdup(spapr->host_serial);
3240 }
3241 
3242 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3243 {
3244     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3245 
3246     g_free(spapr->host_serial);
3247     spapr->host_serial = g_strdup(value);
3248 }
3249 
3250 static void spapr_instance_init(Object *obj)
3251 {
3252     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3253     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3254 
3255     spapr->htab_fd = -1;
3256     spapr->use_hotplug_event_source = true;
3257     object_property_add_str(obj, "kvm-type",
3258                             spapr_get_kvm_type, spapr_set_kvm_type);
3259     object_property_set_description(obj, "kvm-type",
3260                                     "Specifies the KVM virtualization mode (HV, PR)");
3261     object_property_add_bool(obj, "modern-hotplug-events",
3262                             spapr_get_modern_hotplug_events,
3263                             spapr_set_modern_hotplug_events);
3264     object_property_set_description(obj, "modern-hotplug-events",
3265                                     "Use dedicated hotplug event mechanism in"
3266                                     " place of standard EPOW events when possible"
3267                                     " (required for memory hot-unplug support)");
3268     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3269                             "Maximum permitted CPU compatibility mode");
3270 
3271     object_property_add_str(obj, "resize-hpt",
3272                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3273     object_property_set_description(obj, "resize-hpt",
3274                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3275     object_property_add_uint32_ptr(obj, "vsmt",
3276                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3277     object_property_set_description(obj, "vsmt",
3278                                     "Virtual SMT: KVM behaves as if this were"
3279                                     " the host's SMT mode");
3280 
3281     object_property_add_bool(obj, "vfio-no-msix-emulation",
3282                              spapr_get_msix_emulation, NULL);
3283 
3284     object_property_add_uint64_ptr(obj, "kernel-addr",
3285                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3286     object_property_set_description(obj, "kernel-addr",
3287                                     stringify(KERNEL_LOAD_ADDR)
3288                                     " for -kernel is the default");
3289     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3290     /* The machine class defines the default interrupt controller mode */
3291     spapr->irq = smc->irq;
3292     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3293                             spapr_set_ic_mode);
3294     object_property_set_description(obj, "ic-mode",
3295                  "Specifies the interrupt controller mode (xics, xive, dual)");
3296 
3297     object_property_add_str(obj, "host-model",
3298         spapr_get_host_model, spapr_set_host_model);
3299     object_property_set_description(obj, "host-model",
3300         "Host model to advertise in guest device tree");
3301     object_property_add_str(obj, "host-serial",
3302         spapr_get_host_serial, spapr_set_host_serial);
3303     object_property_set_description(obj, "host-serial",
3304         "Host serial number to advertise in guest device tree");
3305 }
3306 
3307 static void spapr_machine_finalizefn(Object *obj)
3308 {
3309     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3310 
3311     g_free(spapr->kvm_type);
3312 }
3313 
3314 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3315 {
3316     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3317     PowerPCCPU *cpu = POWERPC_CPU(cs);
3318     CPUPPCState *env = &cpu->env;
3319 
3320     cpu_synchronize_state(cs);
3321     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3322     if (spapr->fwnmi_system_reset_addr != -1) {
3323         uint64_t rtas_addr, addr;
3324 
3325         /* get rtas addr from fdt */
3326         rtas_addr = spapr_get_rtas_addr();
3327         if (!rtas_addr) {
3328             qemu_system_guest_panicked(NULL);
3329             return;
3330         }
3331 
3332         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3333         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3334         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3335         env->gpr[3] = addr;
3336     }
3337     ppc_cpu_do_system_reset(cs);
3338     if (spapr->fwnmi_system_reset_addr != -1) {
3339         env->nip = spapr->fwnmi_system_reset_addr;
3340     }
3341 }
3342 
3343 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3344 {
3345     CPUState *cs;
3346 
3347     CPU_FOREACH(cs) {
3348         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3349     }
3350 }
3351 
3352 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3353                           void *fdt, int *fdt_start_offset, Error **errp)
3354 {
3355     uint64_t addr;
3356     uint32_t node;
3357 
3358     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3359     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3360                                     &error_abort);
3361     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3362                                              SPAPR_MEMORY_BLOCK_SIZE);
3363     return 0;
3364 }
3365 
3366 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3367                            bool dedicated_hp_event_source, Error **errp)
3368 {
3369     SpaprDrc *drc;
3370     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3371     int i;
3372     uint64_t addr = addr_start;
3373     bool hotplugged = spapr_drc_hotplugged(dev);
3374 
3375     for (i = 0; i < nr_lmbs; i++) {
3376         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3377                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3378         g_assert(drc);
3379 
3380         if (!spapr_drc_attach(drc, dev, errp)) {
3381             while (addr > addr_start) {
3382                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3383                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3384                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3385                 spapr_drc_detach(drc);
3386             }
3387             return;
3388         }
3389         if (!hotplugged) {
3390             spapr_drc_reset(drc);
3391         }
3392         addr += SPAPR_MEMORY_BLOCK_SIZE;
3393     }
3394     /* send hotplug notification to the
3395      * guest only in case of hotplugged memory
3396      */
3397     if (hotplugged) {
3398         if (dedicated_hp_event_source) {
3399             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3400                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3401             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3402                                                    nr_lmbs,
3403                                                    spapr_drc_index(drc));
3404         } else {
3405             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3406                                            nr_lmbs);
3407         }
3408     }
3409 }
3410 
3411 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3412                               Error **errp)
3413 {
3414     Error *local_err = NULL;
3415     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3416     PCDIMMDevice *dimm = PC_DIMM(dev);
3417     uint64_t size, addr, slot;
3418     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3419 
3420     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3421 
3422     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3423     if (local_err) {
3424         goto out;
3425     }
3426 
3427     if (!is_nvdimm) {
3428         addr = object_property_get_uint(OBJECT(dimm),
3429                                         PC_DIMM_ADDR_PROP, &local_err);
3430         if (local_err) {
3431             goto out_unplug;
3432         }
3433         spapr_add_lmbs(dev, addr, size,
3434                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3435                        &local_err);
3436     } else {
3437         slot = object_property_get_uint(OBJECT(dimm),
3438                                         PC_DIMM_SLOT_PROP, &local_err);
3439         if (local_err) {
3440             goto out_unplug;
3441         }
3442         spapr_add_nvdimm(dev, slot, &local_err);
3443     }
3444 
3445     if (local_err) {
3446         goto out_unplug;
3447     }
3448 
3449     return;
3450 
3451 out_unplug:
3452     pc_dimm_unplug(dimm, MACHINE(ms));
3453 out:
3454     error_propagate(errp, local_err);
3455 }
3456 
3457 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3458                                   Error **errp)
3459 {
3460     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3461     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3462     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3463     PCDIMMDevice *dimm = PC_DIMM(dev);
3464     Error *local_err = NULL;
3465     uint64_t size;
3466     Object *memdev;
3467     hwaddr pagesize;
3468 
3469     if (!smc->dr_lmb_enabled) {
3470         error_setg(errp, "Memory hotplug not supported for this machine");
3471         return;
3472     }
3473 
3474     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3475     if (local_err) {
3476         error_propagate(errp, local_err);
3477         return;
3478     }
3479 
3480     if (is_nvdimm) {
3481         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3482             return;
3483         }
3484     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3485         error_setg(errp, "Hotplugged memory size must be a multiple of "
3486                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3487         return;
3488     }
3489 
3490     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3491                                       &error_abort);
3492     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3493     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3494         return;
3495     }
3496 
3497     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3498 }
3499 
3500 struct SpaprDimmState {
3501     PCDIMMDevice *dimm;
3502     uint32_t nr_lmbs;
3503     QTAILQ_ENTRY(SpaprDimmState) next;
3504 };
3505 
3506 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3507                                                        PCDIMMDevice *dimm)
3508 {
3509     SpaprDimmState *dimm_state = NULL;
3510 
3511     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3512         if (dimm_state->dimm == dimm) {
3513             break;
3514         }
3515     }
3516     return dimm_state;
3517 }
3518 
3519 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3520                                                       uint32_t nr_lmbs,
3521                                                       PCDIMMDevice *dimm)
3522 {
3523     SpaprDimmState *ds = NULL;
3524 
3525     /*
3526      * If this request is for a DIMM whose removal had failed earlier
3527      * (due to guest's refusal to remove the LMBs), we would have this
3528      * dimm already in the pending_dimm_unplugs list. In that
3529      * case don't add again.
3530      */
3531     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3532     if (!ds) {
3533         ds = g_malloc0(sizeof(SpaprDimmState));
3534         ds->nr_lmbs = nr_lmbs;
3535         ds->dimm = dimm;
3536         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3537     }
3538     return ds;
3539 }
3540 
3541 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3542                                               SpaprDimmState *dimm_state)
3543 {
3544     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3545     g_free(dimm_state);
3546 }
3547 
3548 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3549                                                         PCDIMMDevice *dimm)
3550 {
3551     SpaprDrc *drc;
3552     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3553                                                   &error_abort);
3554     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3555     uint32_t avail_lmbs = 0;
3556     uint64_t addr_start, addr;
3557     int i;
3558 
3559     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3560                                          &error_abort);
3561 
3562     addr = addr_start;
3563     for (i = 0; i < nr_lmbs; i++) {
3564         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3565                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3566         g_assert(drc);
3567         if (drc->dev) {
3568             avail_lmbs++;
3569         }
3570         addr += SPAPR_MEMORY_BLOCK_SIZE;
3571     }
3572 
3573     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3574 }
3575 
3576 /* Callback to be called during DRC release. */
3577 void spapr_lmb_release(DeviceState *dev)
3578 {
3579     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3580     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3581     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3582 
3583     /* This information will get lost if a migration occurs
3584      * during the unplug process. In this case recover it. */
3585     if (ds == NULL) {
3586         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3587         g_assert(ds);
3588         /* The DRC being examined by the caller at least must be counted */
3589         g_assert(ds->nr_lmbs);
3590     }
3591 
3592     if (--ds->nr_lmbs) {
3593         return;
3594     }
3595 
3596     /*
3597      * Now that all the LMBs have been removed by the guest, call the
3598      * unplug handler chain. This can never fail.
3599      */
3600     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3601     object_unparent(OBJECT(dev));
3602 }
3603 
3604 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3605 {
3606     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3607     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3608 
3609     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3610     qdev_unrealize(dev);
3611     spapr_pending_dimm_unplugs_remove(spapr, ds);
3612 }
3613 
3614 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3615                                         DeviceState *dev, Error **errp)
3616 {
3617     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3618     Error *local_err = NULL;
3619     PCDIMMDevice *dimm = PC_DIMM(dev);
3620     uint32_t nr_lmbs;
3621     uint64_t size, addr_start, addr;
3622     int i;
3623     SpaprDrc *drc;
3624 
3625     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3626         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3627         return;
3628     }
3629 
3630     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3631     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3632 
3633     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3634                                          &local_err);
3635     if (local_err) {
3636         error_propagate(errp, local_err);
3637         return;
3638     }
3639 
3640     /*
3641      * An existing pending dimm state for this DIMM means that there is an
3642      * unplug operation in progress, waiting for the spapr_lmb_release
3643      * callback to complete the job (BQL can't cover that far). In this case,
3644      * bail out to avoid detaching DRCs that were already released.
3645      */
3646     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3647         error_setg(errp, "Memory unplug already in progress for device %s",
3648                    dev->id);
3649         return;
3650     }
3651 
3652     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3653 
3654     addr = addr_start;
3655     for (i = 0; i < nr_lmbs; i++) {
3656         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3657                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3658         g_assert(drc);
3659 
3660         spapr_drc_detach(drc);
3661         addr += SPAPR_MEMORY_BLOCK_SIZE;
3662     }
3663 
3664     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3665                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3666     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3667                                               nr_lmbs, spapr_drc_index(drc));
3668 }
3669 
3670 /* Callback to be called during DRC release. */
3671 void spapr_core_release(DeviceState *dev)
3672 {
3673     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3674 
3675     /* Call the unplug handler chain. This can never fail. */
3676     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3677     object_unparent(OBJECT(dev));
3678 }
3679 
3680 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3681 {
3682     MachineState *ms = MACHINE(hotplug_dev);
3683     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3684     CPUCore *cc = CPU_CORE(dev);
3685     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3686 
3687     if (smc->pre_2_10_has_unused_icps) {
3688         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3689         int i;
3690 
3691         for (i = 0; i < cc->nr_threads; i++) {
3692             CPUState *cs = CPU(sc->threads[i]);
3693 
3694             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3695         }
3696     }
3697 
3698     assert(core_slot);
3699     core_slot->cpu = NULL;
3700     qdev_unrealize(dev);
3701 }
3702 
3703 static
3704 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3705                                Error **errp)
3706 {
3707     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3708     int index;
3709     SpaprDrc *drc;
3710     CPUCore *cc = CPU_CORE(dev);
3711 
3712     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3713         error_setg(errp, "Unable to find CPU core with core-id: %d",
3714                    cc->core_id);
3715         return;
3716     }
3717     if (index == 0) {
3718         error_setg(errp, "Boot CPU core may not be unplugged");
3719         return;
3720     }
3721 
3722     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3723                           spapr_vcpu_id(spapr, cc->core_id));
3724     g_assert(drc);
3725 
3726     if (!spapr_drc_unplug_requested(drc)) {
3727         spapr_drc_detach(drc);
3728         spapr_hotplug_req_remove_by_index(drc);
3729     }
3730 }
3731 
3732 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3733                            void *fdt, int *fdt_start_offset, Error **errp)
3734 {
3735     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3736     CPUState *cs = CPU(core->threads[0]);
3737     PowerPCCPU *cpu = POWERPC_CPU(cs);
3738     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3739     int id = spapr_get_vcpu_id(cpu);
3740     char *nodename;
3741     int offset;
3742 
3743     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3744     offset = fdt_add_subnode(fdt, 0, nodename);
3745     g_free(nodename);
3746 
3747     spapr_dt_cpu(cs, fdt, offset, spapr);
3748 
3749     *fdt_start_offset = offset;
3750     return 0;
3751 }
3752 
3753 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3754                             Error **errp)
3755 {
3756     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3757     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3758     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3759     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3760     CPUCore *cc = CPU_CORE(dev);
3761     CPUState *cs;
3762     SpaprDrc *drc;
3763     CPUArchId *core_slot;
3764     int index;
3765     bool hotplugged = spapr_drc_hotplugged(dev);
3766     int i;
3767 
3768     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3769     if (!core_slot) {
3770         error_setg(errp, "Unable to find CPU core with core-id: %d",
3771                    cc->core_id);
3772         return;
3773     }
3774     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3775                           spapr_vcpu_id(spapr, cc->core_id));
3776 
3777     g_assert(drc || !mc->has_hotpluggable_cpus);
3778 
3779     if (drc) {
3780         if (!spapr_drc_attach(drc, dev, errp)) {
3781             return;
3782         }
3783 
3784         if (hotplugged) {
3785             /*
3786              * Send hotplug notification interrupt to the guest only
3787              * in case of hotplugged CPUs.
3788              */
3789             spapr_hotplug_req_add_by_index(drc);
3790         } else {
3791             spapr_drc_reset(drc);
3792         }
3793     }
3794 
3795     core_slot->cpu = OBJECT(dev);
3796 
3797     if (smc->pre_2_10_has_unused_icps) {
3798         for (i = 0; i < cc->nr_threads; i++) {
3799             cs = CPU(core->threads[i]);
3800             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3801         }
3802     }
3803 
3804     /*
3805      * Set compatibility mode to match the boot CPU, which was either set
3806      * by the machine reset code or by CAS.
3807      */
3808     if (hotplugged) {
3809         for (i = 0; i < cc->nr_threads; i++) {
3810             if (ppc_set_compat(core->threads[i],
3811                                POWERPC_CPU(first_cpu)->compat_pvr,
3812                                errp) < 0) {
3813                 return;
3814             }
3815         }
3816     }
3817 }
3818 
3819 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3820                                 Error **errp)
3821 {
3822     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3823     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3824     CPUCore *cc = CPU_CORE(dev);
3825     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3826     const char *type = object_get_typename(OBJECT(dev));
3827     CPUArchId *core_slot;
3828     int index;
3829     unsigned int smp_threads = machine->smp.threads;
3830 
3831     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3832         error_setg(errp, "CPU hotplug not supported for this machine");
3833         return;
3834     }
3835 
3836     if (strcmp(base_core_type, type)) {
3837         error_setg(errp, "CPU core type should be %s", base_core_type);
3838         return;
3839     }
3840 
3841     if (cc->core_id % smp_threads) {
3842         error_setg(errp, "invalid core id %d", cc->core_id);
3843         return;
3844     }
3845 
3846     /*
3847      * In general we should have homogeneous threads-per-core, but old
3848      * (pre hotplug support) machine types allow the last core to have
3849      * reduced threads as a compatibility hack for when we allowed
3850      * total vcpus not a multiple of threads-per-core.
3851      */
3852     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3853         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3854                    smp_threads);
3855         return;
3856     }
3857 
3858     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3859     if (!core_slot) {
3860         error_setg(errp, "core id %d out of range", cc->core_id);
3861         return;
3862     }
3863 
3864     if (core_slot->cpu) {
3865         error_setg(errp, "core %d already populated", cc->core_id);
3866         return;
3867     }
3868 
3869     numa_cpu_pre_plug(core_slot, dev, errp);
3870 }
3871 
3872 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3873                           void *fdt, int *fdt_start_offset, Error **errp)
3874 {
3875     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3876     int intc_phandle;
3877 
3878     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3879     if (intc_phandle <= 0) {
3880         return -1;
3881     }
3882 
3883     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3884         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3885         return -1;
3886     }
3887 
3888     /* generally SLOF creates these, for hotplug it's up to QEMU */
3889     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3890 
3891     return 0;
3892 }
3893 
3894 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3895                                Error **errp)
3896 {
3897     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3898     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3899     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3900     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3901 
3902     if (dev->hotplugged && !smc->dr_phb_enabled) {
3903         error_setg(errp, "PHB hotplug not supported for this machine");
3904         return;
3905     }
3906 
3907     if (sphb->index == (uint32_t)-1) {
3908         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3909         return;
3910     }
3911 
3912     /*
3913      * This will check that sphb->index doesn't exceed the maximum number of
3914      * PHBs for the current machine type.
3915      */
3916     smc->phb_placement(spapr, sphb->index,
3917                        &sphb->buid, &sphb->io_win_addr,
3918                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3919                        windows_supported, sphb->dma_liobn,
3920                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3921                        errp);
3922 }
3923 
3924 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3925                            Error **errp)
3926 {
3927     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3928     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3929     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3930     SpaprDrc *drc;
3931     bool hotplugged = spapr_drc_hotplugged(dev);
3932 
3933     if (!smc->dr_phb_enabled) {
3934         return;
3935     }
3936 
3937     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3938     /* hotplug hooks should check it's enabled before getting this far */
3939     assert(drc);
3940 
3941     if (!spapr_drc_attach(drc, dev, errp)) {
3942         return;
3943     }
3944 
3945     if (hotplugged) {
3946         spapr_hotplug_req_add_by_index(drc);
3947     } else {
3948         spapr_drc_reset(drc);
3949     }
3950 }
3951 
3952 void spapr_phb_release(DeviceState *dev)
3953 {
3954     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3955 
3956     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3957     object_unparent(OBJECT(dev));
3958 }
3959 
3960 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3961 {
3962     qdev_unrealize(dev);
3963 }
3964 
3965 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3966                                      DeviceState *dev, Error **errp)
3967 {
3968     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3969     SpaprDrc *drc;
3970 
3971     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3972     assert(drc);
3973 
3974     if (!spapr_drc_unplug_requested(drc)) {
3975         spapr_drc_detach(drc);
3976         spapr_hotplug_req_remove_by_index(drc);
3977     }
3978 }
3979 
3980 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3981                                  Error **errp)
3982 {
3983     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3984     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
3985 
3986     if (spapr->tpm_proxy != NULL) {
3987         error_setg(errp, "Only one TPM proxy can be specified for this machine");
3988         return;
3989     }
3990 
3991     spapr->tpm_proxy = tpm_proxy;
3992 }
3993 
3994 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3995 {
3996     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3997 
3998     qdev_unrealize(dev);
3999     object_unparent(OBJECT(dev));
4000     spapr->tpm_proxy = NULL;
4001 }
4002 
4003 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4004                                       DeviceState *dev, Error **errp)
4005 {
4006     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4007         spapr_memory_plug(hotplug_dev, dev, errp);
4008     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4009         spapr_core_plug(hotplug_dev, dev, errp);
4010     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4011         spapr_phb_plug(hotplug_dev, dev, errp);
4012     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4013         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4014     }
4015 }
4016 
4017 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4018                                         DeviceState *dev, Error **errp)
4019 {
4020     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4021         spapr_memory_unplug(hotplug_dev, dev);
4022     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4023         spapr_core_unplug(hotplug_dev, dev);
4024     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4025         spapr_phb_unplug(hotplug_dev, dev);
4026     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4027         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4028     }
4029 }
4030 
4031 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4032                                                 DeviceState *dev, Error **errp)
4033 {
4034     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4035     MachineClass *mc = MACHINE_GET_CLASS(sms);
4036     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4037 
4038     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4039         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4040             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4041         } else {
4042             /* NOTE: this means there is a window after guest reset, prior to
4043              * CAS negotiation, where unplug requests will fail due to the
4044              * capability not being detected yet. This is a bit different than
4045              * the case with PCI unplug, where the events will be queued and
4046              * eventually handled by the guest after boot
4047              */
4048             error_setg(errp, "Memory hot unplug not supported for this guest");
4049         }
4050     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4051         if (!mc->has_hotpluggable_cpus) {
4052             error_setg(errp, "CPU hot unplug not supported on this machine");
4053             return;
4054         }
4055         spapr_core_unplug_request(hotplug_dev, dev, errp);
4056     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4057         if (!smc->dr_phb_enabled) {
4058             error_setg(errp, "PHB hot unplug not supported on this machine");
4059             return;
4060         }
4061         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4062     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4063         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4064     }
4065 }
4066 
4067 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4068                                           DeviceState *dev, Error **errp)
4069 {
4070     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4071         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4072     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4073         spapr_core_pre_plug(hotplug_dev, dev, errp);
4074     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4075         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4076     }
4077 }
4078 
4079 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4080                                                  DeviceState *dev)
4081 {
4082     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4083         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4084         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4085         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4086         return HOTPLUG_HANDLER(machine);
4087     }
4088     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4089         PCIDevice *pcidev = PCI_DEVICE(dev);
4090         PCIBus *root = pci_device_root_bus(pcidev);
4091         SpaprPhbState *phb =
4092             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4093                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4094 
4095         if (phb) {
4096             return HOTPLUG_HANDLER(phb);
4097         }
4098     }
4099     return NULL;
4100 }
4101 
4102 static CpuInstanceProperties
4103 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4104 {
4105     CPUArchId *core_slot;
4106     MachineClass *mc = MACHINE_GET_CLASS(machine);
4107 
4108     /* make sure possible_cpu are intialized */
4109     mc->possible_cpu_arch_ids(machine);
4110     /* get CPU core slot containing thread that matches cpu_index */
4111     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4112     assert(core_slot);
4113     return core_slot->props;
4114 }
4115 
4116 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4117 {
4118     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4119 }
4120 
4121 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4122 {
4123     int i;
4124     unsigned int smp_threads = machine->smp.threads;
4125     unsigned int smp_cpus = machine->smp.cpus;
4126     const char *core_type;
4127     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4128     MachineClass *mc = MACHINE_GET_CLASS(machine);
4129 
4130     if (!mc->has_hotpluggable_cpus) {
4131         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4132     }
4133     if (machine->possible_cpus) {
4134         assert(machine->possible_cpus->len == spapr_max_cores);
4135         return machine->possible_cpus;
4136     }
4137 
4138     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4139     if (!core_type) {
4140         error_report("Unable to find sPAPR CPU Core definition");
4141         exit(1);
4142     }
4143 
4144     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4145                              sizeof(CPUArchId) * spapr_max_cores);
4146     machine->possible_cpus->len = spapr_max_cores;
4147     for (i = 0; i < machine->possible_cpus->len; i++) {
4148         int core_id = i * smp_threads;
4149 
4150         machine->possible_cpus->cpus[i].type = core_type;
4151         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4152         machine->possible_cpus->cpus[i].arch_id = core_id;
4153         machine->possible_cpus->cpus[i].props.has_core_id = true;
4154         machine->possible_cpus->cpus[i].props.core_id = core_id;
4155     }
4156     return machine->possible_cpus;
4157 }
4158 
4159 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4160                                 uint64_t *buid, hwaddr *pio,
4161                                 hwaddr *mmio32, hwaddr *mmio64,
4162                                 unsigned n_dma, uint32_t *liobns,
4163                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4164 {
4165     /*
4166      * New-style PHB window placement.
4167      *
4168      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4169      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4170      * windows.
4171      *
4172      * Some guest kernels can't work with MMIO windows above 1<<46
4173      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4174      *
4175      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4176      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4177      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4178      * 1TiB 64-bit MMIO windows for each PHB.
4179      */
4180     const uint64_t base_buid = 0x800000020000000ULL;
4181     int i;
4182 
4183     /* Sanity check natural alignments */
4184     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4185     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4186     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4187     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4188     /* Sanity check bounds */
4189     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4190                       SPAPR_PCI_MEM32_WIN_SIZE);
4191     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4192                       SPAPR_PCI_MEM64_WIN_SIZE);
4193 
4194     if (index >= SPAPR_MAX_PHBS) {
4195         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4196                    SPAPR_MAX_PHBS - 1);
4197         return;
4198     }
4199 
4200     *buid = base_buid + index;
4201     for (i = 0; i < n_dma; ++i) {
4202         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4203     }
4204 
4205     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4206     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4207     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4208 
4209     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4210     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4211 }
4212 
4213 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4214 {
4215     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4216 
4217     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4218 }
4219 
4220 static void spapr_ics_resend(XICSFabric *dev)
4221 {
4222     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4223 
4224     ics_resend(spapr->ics);
4225 }
4226 
4227 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4228 {
4229     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4230 
4231     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4232 }
4233 
4234 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4235                                  Monitor *mon)
4236 {
4237     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4238 
4239     spapr_irq_print_info(spapr, mon);
4240     monitor_printf(mon, "irqchip: %s\n",
4241                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4242 }
4243 
4244 /*
4245  * This is a XIVE only operation
4246  */
4247 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4248                            uint8_t nvt_blk, uint32_t nvt_idx,
4249                            bool cam_ignore, uint8_t priority,
4250                            uint32_t logic_serv, XiveTCTXMatch *match)
4251 {
4252     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4253     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4254     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4255     int count;
4256 
4257     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4258                            priority, logic_serv, match);
4259     if (count < 0) {
4260         return count;
4261     }
4262 
4263     /*
4264      * When we implement the save and restore of the thread interrupt
4265      * contexts in the enter/exit CPU handlers of the machine and the
4266      * escalations in QEMU, we should be able to handle non dispatched
4267      * vCPUs.
4268      *
4269      * Until this is done, the sPAPR machine should find at least one
4270      * matching context always.
4271      */
4272     if (count == 0) {
4273         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4274                       nvt_blk, nvt_idx);
4275     }
4276 
4277     return count;
4278 }
4279 
4280 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4281 {
4282     return cpu->vcpu_id;
4283 }
4284 
4285 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4286 {
4287     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4288     MachineState *ms = MACHINE(spapr);
4289     int vcpu_id;
4290 
4291     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4292 
4293     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4294         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4295         error_append_hint(errp, "Adjust the number of cpus to %d "
4296                           "or try to raise the number of threads per core\n",
4297                           vcpu_id * ms->smp.threads / spapr->vsmt);
4298         return false;
4299     }
4300 
4301     cpu->vcpu_id = vcpu_id;
4302     return true;
4303 }
4304 
4305 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4306 {
4307     CPUState *cs;
4308 
4309     CPU_FOREACH(cs) {
4310         PowerPCCPU *cpu = POWERPC_CPU(cs);
4311 
4312         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4313             return cpu;
4314         }
4315     }
4316 
4317     return NULL;
4318 }
4319 
4320 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4321 {
4322     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4323 
4324     /* These are only called by TCG, KVM maintains dispatch state */
4325 
4326     spapr_cpu->prod = false;
4327     if (spapr_cpu->vpa_addr) {
4328         CPUState *cs = CPU(cpu);
4329         uint32_t dispatch;
4330 
4331         dispatch = ldl_be_phys(cs->as,
4332                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4333         dispatch++;
4334         if ((dispatch & 1) != 0) {
4335             qemu_log_mask(LOG_GUEST_ERROR,
4336                           "VPA: incorrect dispatch counter value for "
4337                           "dispatched partition %u, correcting.\n", dispatch);
4338             dispatch++;
4339         }
4340         stl_be_phys(cs->as,
4341                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4342     }
4343 }
4344 
4345 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4346 {
4347     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4348 
4349     if (spapr_cpu->vpa_addr) {
4350         CPUState *cs = CPU(cpu);
4351         uint32_t dispatch;
4352 
4353         dispatch = ldl_be_phys(cs->as,
4354                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4355         dispatch++;
4356         if ((dispatch & 1) != 1) {
4357             qemu_log_mask(LOG_GUEST_ERROR,
4358                           "VPA: incorrect dispatch counter value for "
4359                           "preempted partition %u, correcting.\n", dispatch);
4360             dispatch++;
4361         }
4362         stl_be_phys(cs->as,
4363                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4364     }
4365 }
4366 
4367 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4368 {
4369     MachineClass *mc = MACHINE_CLASS(oc);
4370     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4371     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4372     NMIClass *nc = NMI_CLASS(oc);
4373     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4374     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4375     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4376     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4377     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4378 
4379     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4380     mc->ignore_boot_device_suffixes = true;
4381 
4382     /*
4383      * We set up the default / latest behaviour here.  The class_init
4384      * functions for the specific versioned machine types can override
4385      * these details for backwards compatibility
4386      */
4387     mc->init = spapr_machine_init;
4388     mc->reset = spapr_machine_reset;
4389     mc->block_default_type = IF_SCSI;
4390     mc->max_cpus = 1024;
4391     mc->no_parallel = 1;
4392     mc->default_boot_order = "";
4393     mc->default_ram_size = 512 * MiB;
4394     mc->default_ram_id = "ppc_spapr.ram";
4395     mc->default_display = "std";
4396     mc->kvm_type = spapr_kvm_type;
4397     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4398     mc->pci_allow_0_address = true;
4399     assert(!mc->get_hotplug_handler);
4400     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4401     hc->pre_plug = spapr_machine_device_pre_plug;
4402     hc->plug = spapr_machine_device_plug;
4403     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4404     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4405     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4406     hc->unplug_request = spapr_machine_device_unplug_request;
4407     hc->unplug = spapr_machine_device_unplug;
4408 
4409     smc->dr_lmb_enabled = true;
4410     smc->update_dt_enabled = true;
4411     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4412     mc->has_hotpluggable_cpus = true;
4413     mc->nvdimm_supported = true;
4414     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4415     fwc->get_dev_path = spapr_get_fw_dev_path;
4416     nc->nmi_monitor_handler = spapr_nmi;
4417     smc->phb_placement = spapr_phb_placement;
4418     vhc->hypercall = emulate_spapr_hypercall;
4419     vhc->hpt_mask = spapr_hpt_mask;
4420     vhc->map_hptes = spapr_map_hptes;
4421     vhc->unmap_hptes = spapr_unmap_hptes;
4422     vhc->hpte_set_c = spapr_hpte_set_c;
4423     vhc->hpte_set_r = spapr_hpte_set_r;
4424     vhc->get_pate = spapr_get_pate;
4425     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4426     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4427     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4428     xic->ics_get = spapr_ics_get;
4429     xic->ics_resend = spapr_ics_resend;
4430     xic->icp_get = spapr_icp_get;
4431     ispc->print_info = spapr_pic_print_info;
4432     /* Force NUMA node memory size to be a multiple of
4433      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4434      * in which LMBs are represented and hot-added
4435      */
4436     mc->numa_mem_align_shift = 28;
4437     mc->auto_enable_numa = true;
4438 
4439     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4440     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4441     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4442     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4443     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4444     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4445     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4446     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4447     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4448     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4449     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4450     spapr_caps_add_properties(smc);
4451     smc->irq = &spapr_irq_dual;
4452     smc->dr_phb_enabled = true;
4453     smc->linux_pci_probe = true;
4454     smc->smp_threads_vsmt = true;
4455     smc->nr_xirqs = SPAPR_NR_XIRQS;
4456     xfc->match_nvt = spapr_match_nvt;
4457 }
4458 
4459 static const TypeInfo spapr_machine_info = {
4460     .name          = TYPE_SPAPR_MACHINE,
4461     .parent        = TYPE_MACHINE,
4462     .abstract      = true,
4463     .instance_size = sizeof(SpaprMachineState),
4464     .instance_init = spapr_instance_init,
4465     .instance_finalize = spapr_machine_finalizefn,
4466     .class_size    = sizeof(SpaprMachineClass),
4467     .class_init    = spapr_machine_class_init,
4468     .interfaces = (InterfaceInfo[]) {
4469         { TYPE_FW_PATH_PROVIDER },
4470         { TYPE_NMI },
4471         { TYPE_HOTPLUG_HANDLER },
4472         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4473         { TYPE_XICS_FABRIC },
4474         { TYPE_INTERRUPT_STATS_PROVIDER },
4475         { TYPE_XIVE_FABRIC },
4476         { }
4477     },
4478 };
4479 
4480 static void spapr_machine_latest_class_options(MachineClass *mc)
4481 {
4482     mc->alias = "pseries";
4483     mc->is_default = true;
4484 }
4485 
4486 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4487     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4488                                                     void *data)      \
4489     {                                                                \
4490         MachineClass *mc = MACHINE_CLASS(oc);                        \
4491         spapr_machine_##suffix##_class_options(mc);                  \
4492         if (latest) {                                                \
4493             spapr_machine_latest_class_options(mc);                  \
4494         }                                                            \
4495     }                                                                \
4496     static const TypeInfo spapr_machine_##suffix##_info = {          \
4497         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4498         .parent = TYPE_SPAPR_MACHINE,                                \
4499         .class_init = spapr_machine_##suffix##_class_init,           \
4500     };                                                               \
4501     static void spapr_machine_register_##suffix(void)                \
4502     {                                                                \
4503         type_register(&spapr_machine_##suffix##_info);               \
4504     }                                                                \
4505     type_init(spapr_machine_register_##suffix)
4506 
4507 /*
4508  * pseries-5.2
4509  */
4510 static void spapr_machine_5_2_class_options(MachineClass *mc)
4511 {
4512     /* Defaults for the latest behaviour inherited from the base class */
4513 }
4514 
4515 DEFINE_SPAPR_MACHINE(5_2, "5.2", true);
4516 
4517 /*
4518  * pseries-5.1
4519  */
4520 static void spapr_machine_5_1_class_options(MachineClass *mc)
4521 {
4522     spapr_machine_5_2_class_options(mc);
4523     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4524 }
4525 
4526 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4527 
4528 /*
4529  * pseries-5.0
4530  */
4531 static void spapr_machine_5_0_class_options(MachineClass *mc)
4532 {
4533     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4534     static GlobalProperty compat[] = {
4535         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4536     };
4537 
4538     spapr_machine_5_1_class_options(mc);
4539     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4540     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4541     mc->numa_mem_supported = true;
4542     smc->pre_5_1_assoc_refpoints = true;
4543 }
4544 
4545 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4546 
4547 /*
4548  * pseries-4.2
4549  */
4550 static void spapr_machine_4_2_class_options(MachineClass *mc)
4551 {
4552     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4553 
4554     spapr_machine_5_0_class_options(mc);
4555     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4556     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4557     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4558     smc->rma_limit = 16 * GiB;
4559     mc->nvdimm_supported = false;
4560 }
4561 
4562 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4563 
4564 /*
4565  * pseries-4.1
4566  */
4567 static void spapr_machine_4_1_class_options(MachineClass *mc)
4568 {
4569     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4570     static GlobalProperty compat[] = {
4571         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4572         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4573     };
4574 
4575     spapr_machine_4_2_class_options(mc);
4576     smc->linux_pci_probe = false;
4577     smc->smp_threads_vsmt = false;
4578     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4579     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4580 }
4581 
4582 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4583 
4584 /*
4585  * pseries-4.0
4586  */
4587 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4588                               uint64_t *buid, hwaddr *pio,
4589                               hwaddr *mmio32, hwaddr *mmio64,
4590                               unsigned n_dma, uint32_t *liobns,
4591                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4592 {
4593     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4594                         nv2gpa, nv2atsd, errp);
4595     *nv2gpa = 0;
4596     *nv2atsd = 0;
4597 }
4598 
4599 static void spapr_machine_4_0_class_options(MachineClass *mc)
4600 {
4601     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4602 
4603     spapr_machine_4_1_class_options(mc);
4604     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4605     smc->phb_placement = phb_placement_4_0;
4606     smc->irq = &spapr_irq_xics;
4607     smc->pre_4_1_migration = true;
4608 }
4609 
4610 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4611 
4612 /*
4613  * pseries-3.1
4614  */
4615 static void spapr_machine_3_1_class_options(MachineClass *mc)
4616 {
4617     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4618 
4619     spapr_machine_4_0_class_options(mc);
4620     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4621 
4622     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4623     smc->update_dt_enabled = false;
4624     smc->dr_phb_enabled = false;
4625     smc->broken_host_serial_model = true;
4626     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4627     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4628     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4629     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4630 }
4631 
4632 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4633 
4634 /*
4635  * pseries-3.0
4636  */
4637 
4638 static void spapr_machine_3_0_class_options(MachineClass *mc)
4639 {
4640     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4641 
4642     spapr_machine_3_1_class_options(mc);
4643     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4644 
4645     smc->legacy_irq_allocation = true;
4646     smc->nr_xirqs = 0x400;
4647     smc->irq = &spapr_irq_xics_legacy;
4648 }
4649 
4650 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4651 
4652 /*
4653  * pseries-2.12
4654  */
4655 static void spapr_machine_2_12_class_options(MachineClass *mc)
4656 {
4657     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4658     static GlobalProperty compat[] = {
4659         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4660         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4661     };
4662 
4663     spapr_machine_3_0_class_options(mc);
4664     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4665     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4666 
4667     /* We depend on kvm_enabled() to choose a default value for the
4668      * hpt-max-page-size capability. Of course we can't do it here
4669      * because this is too early and the HW accelerator isn't initialzed
4670      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4671      */
4672     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4673 }
4674 
4675 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4676 
4677 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4678 {
4679     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4680 
4681     spapr_machine_2_12_class_options(mc);
4682     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4683     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4684     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4685 }
4686 
4687 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4688 
4689 /*
4690  * pseries-2.11
4691  */
4692 
4693 static void spapr_machine_2_11_class_options(MachineClass *mc)
4694 {
4695     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4696 
4697     spapr_machine_2_12_class_options(mc);
4698     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4699     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4700 }
4701 
4702 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4703 
4704 /*
4705  * pseries-2.10
4706  */
4707 
4708 static void spapr_machine_2_10_class_options(MachineClass *mc)
4709 {
4710     spapr_machine_2_11_class_options(mc);
4711     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4712 }
4713 
4714 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4715 
4716 /*
4717  * pseries-2.9
4718  */
4719 
4720 static void spapr_machine_2_9_class_options(MachineClass *mc)
4721 {
4722     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4723     static GlobalProperty compat[] = {
4724         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4725     };
4726 
4727     spapr_machine_2_10_class_options(mc);
4728     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4729     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4730     smc->pre_2_10_has_unused_icps = true;
4731     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4732 }
4733 
4734 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4735 
4736 /*
4737  * pseries-2.8
4738  */
4739 
4740 static void spapr_machine_2_8_class_options(MachineClass *mc)
4741 {
4742     static GlobalProperty compat[] = {
4743         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4744     };
4745 
4746     spapr_machine_2_9_class_options(mc);
4747     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4748     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4749     mc->numa_mem_align_shift = 23;
4750 }
4751 
4752 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4753 
4754 /*
4755  * pseries-2.7
4756  */
4757 
4758 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4759                               uint64_t *buid, hwaddr *pio,
4760                               hwaddr *mmio32, hwaddr *mmio64,
4761                               unsigned n_dma, uint32_t *liobns,
4762                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4763 {
4764     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4765     const uint64_t base_buid = 0x800000020000000ULL;
4766     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4767     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4768     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4769     const uint32_t max_index = 255;
4770     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4771 
4772     uint64_t ram_top = MACHINE(spapr)->ram_size;
4773     hwaddr phb0_base, phb_base;
4774     int i;
4775 
4776     /* Do we have device memory? */
4777     if (MACHINE(spapr)->maxram_size > ram_top) {
4778         /* Can't just use maxram_size, because there may be an
4779          * alignment gap between normal and device memory regions
4780          */
4781         ram_top = MACHINE(spapr)->device_memory->base +
4782             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4783     }
4784 
4785     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4786 
4787     if (index > max_index) {
4788         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4789                    max_index);
4790         return;
4791     }
4792 
4793     *buid = base_buid + index;
4794     for (i = 0; i < n_dma; ++i) {
4795         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4796     }
4797 
4798     phb_base = phb0_base + index * phb_spacing;
4799     *pio = phb_base + pio_offset;
4800     *mmio32 = phb_base + mmio_offset;
4801     /*
4802      * We don't set the 64-bit MMIO window, relying on the PHB's
4803      * fallback behaviour of automatically splitting a large "32-bit"
4804      * window into contiguous 32-bit and 64-bit windows
4805      */
4806 
4807     *nv2gpa = 0;
4808     *nv2atsd = 0;
4809 }
4810 
4811 static void spapr_machine_2_7_class_options(MachineClass *mc)
4812 {
4813     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4814     static GlobalProperty compat[] = {
4815         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4816         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4817         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4818         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4819     };
4820 
4821     spapr_machine_2_8_class_options(mc);
4822     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4823     mc->default_machine_opts = "modern-hotplug-events=off";
4824     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4825     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4826     smc->phb_placement = phb_placement_2_7;
4827 }
4828 
4829 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4830 
4831 /*
4832  * pseries-2.6
4833  */
4834 
4835 static void spapr_machine_2_6_class_options(MachineClass *mc)
4836 {
4837     static GlobalProperty compat[] = {
4838         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4839     };
4840 
4841     spapr_machine_2_7_class_options(mc);
4842     mc->has_hotpluggable_cpus = false;
4843     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4844     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4845 }
4846 
4847 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4848 
4849 /*
4850  * pseries-2.5
4851  */
4852 
4853 static void spapr_machine_2_5_class_options(MachineClass *mc)
4854 {
4855     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4856     static GlobalProperty compat[] = {
4857         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4858     };
4859 
4860     spapr_machine_2_6_class_options(mc);
4861     smc->use_ohci_by_default = true;
4862     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4863     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4864 }
4865 
4866 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4867 
4868 /*
4869  * pseries-2.4
4870  */
4871 
4872 static void spapr_machine_2_4_class_options(MachineClass *mc)
4873 {
4874     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4875 
4876     spapr_machine_2_5_class_options(mc);
4877     smc->dr_lmb_enabled = false;
4878     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4879 }
4880 
4881 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4882 
4883 /*
4884  * pseries-2.3
4885  */
4886 
4887 static void spapr_machine_2_3_class_options(MachineClass *mc)
4888 {
4889     static GlobalProperty compat[] = {
4890         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4891     };
4892     spapr_machine_2_4_class_options(mc);
4893     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4894     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4895 }
4896 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4897 
4898 /*
4899  * pseries-2.2
4900  */
4901 
4902 static void spapr_machine_2_2_class_options(MachineClass *mc)
4903 {
4904     static GlobalProperty compat[] = {
4905         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4906     };
4907 
4908     spapr_machine_2_3_class_options(mc);
4909     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4910     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4911     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4912 }
4913 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4914 
4915 /*
4916  * pseries-2.1
4917  */
4918 
4919 static void spapr_machine_2_1_class_options(MachineClass *mc)
4920 {
4921     spapr_machine_2_2_class_options(mc);
4922     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4923 }
4924 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4925 
4926 static void spapr_machine_register_types(void)
4927 {
4928     type_register_static(&spapr_machine_info);
4929 }
4930 
4931 type_init(spapr_machine_register_types)
4932