xref: /openbmc/qemu/hw/ppc/spapr.c (revision e7cff9c6)
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 bool spapr_machine_using_legacy_numa(SpaprMachineState *spapr)
298 {
299     MachineState *machine = MACHINE(spapr);
300     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
301 
302     return smc->pre_5_2_numa_associativity ||
303            machine->numa_state->num_nodes <= 1;
304 }
305 
306 static void add_str(GString *s, const gchar *s1)
307 {
308     g_string_append_len(s, s1, strlen(s1) + 1);
309 }
310 
311 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
312                                 hwaddr start, hwaddr size)
313 {
314     char mem_name[32];
315     uint64_t mem_reg_property[2];
316     int off;
317 
318     mem_reg_property[0] = cpu_to_be64(start);
319     mem_reg_property[1] = cpu_to_be64(size);
320 
321     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
322     off = fdt_add_subnode(fdt, 0, mem_name);
323     _FDT(off);
324     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
325     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
326                       sizeof(mem_reg_property))));
327     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
328     return off;
329 }
330 
331 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
332 {
333     MemoryDeviceInfoList *info;
334 
335     for (info = list; info; info = info->next) {
336         MemoryDeviceInfo *value = info->value;
337 
338         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
339             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
340 
341             if (addr >= pcdimm_info->addr &&
342                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
343                 return pcdimm_info->node;
344             }
345         }
346     }
347 
348     return -1;
349 }
350 
351 struct sPAPRDrconfCellV2 {
352      uint32_t seq_lmbs;
353      uint64_t base_addr;
354      uint32_t drc_index;
355      uint32_t aa_index;
356      uint32_t flags;
357 } QEMU_PACKED;
358 
359 typedef struct DrconfCellQueue {
360     struct sPAPRDrconfCellV2 cell;
361     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
362 } DrconfCellQueue;
363 
364 static DrconfCellQueue *
365 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
366                       uint32_t drc_index, uint32_t aa_index,
367                       uint32_t flags)
368 {
369     DrconfCellQueue *elem;
370 
371     elem = g_malloc0(sizeof(*elem));
372     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
373     elem->cell.base_addr = cpu_to_be64(base_addr);
374     elem->cell.drc_index = cpu_to_be32(drc_index);
375     elem->cell.aa_index = cpu_to_be32(aa_index);
376     elem->cell.flags = cpu_to_be32(flags);
377 
378     return elem;
379 }
380 
381 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
382                                       int offset, MemoryDeviceInfoList *dimms)
383 {
384     MachineState *machine = MACHINE(spapr);
385     uint8_t *int_buf, *cur_index;
386     int ret;
387     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
388     uint64_t addr, cur_addr, size;
389     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
390     uint64_t mem_end = machine->device_memory->base +
391                        memory_region_size(&machine->device_memory->mr);
392     uint32_t node, buf_len, nr_entries = 0;
393     SpaprDrc *drc;
394     DrconfCellQueue *elem, *next;
395     MemoryDeviceInfoList *info;
396     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
397         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
398 
399     /* Entry to cover RAM and the gap area */
400     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
401                                  SPAPR_LMB_FLAGS_RESERVED |
402                                  SPAPR_LMB_FLAGS_DRC_INVALID);
403     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
404     nr_entries++;
405 
406     cur_addr = machine->device_memory->base;
407     for (info = dimms; info; info = info->next) {
408         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
409 
410         addr = di->addr;
411         size = di->size;
412         node = di->node;
413 
414         /*
415          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
416          * area is marked hotpluggable in the next iteration for the bigger
417          * chunk including the NVDIMM occupied area.
418          */
419         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
420             continue;
421 
422         /* Entry for hot-pluggable area */
423         if (cur_addr < addr) {
424             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
425             g_assert(drc);
426             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
427                                          cur_addr, spapr_drc_index(drc), -1, 0);
428             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
429             nr_entries++;
430         }
431 
432         /* Entry for DIMM */
433         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
434         g_assert(drc);
435         elem = spapr_get_drconf_cell(size / lmb_size, addr,
436                                      spapr_drc_index(drc), node,
437                                      (SPAPR_LMB_FLAGS_ASSIGNED |
438                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
439         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
440         nr_entries++;
441         cur_addr = addr + size;
442     }
443 
444     /* Entry for remaining hotpluggable area */
445     if (cur_addr < mem_end) {
446         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
447         g_assert(drc);
448         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
449                                      cur_addr, spapr_drc_index(drc), -1, 0);
450         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
451         nr_entries++;
452     }
453 
454     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
455     int_buf = cur_index = g_malloc0(buf_len);
456     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
457     cur_index += sizeof(nr_entries);
458 
459     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
460         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
461         cur_index += sizeof(elem->cell);
462         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
463         g_free(elem);
464     }
465 
466     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
467     g_free(int_buf);
468     if (ret < 0) {
469         return -1;
470     }
471     return 0;
472 }
473 
474 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
475                                    int offset, MemoryDeviceInfoList *dimms)
476 {
477     MachineState *machine = MACHINE(spapr);
478     int i, ret;
479     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
480     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
481     uint32_t nr_lmbs = (machine->device_memory->base +
482                        memory_region_size(&machine->device_memory->mr)) /
483                        lmb_size;
484     uint32_t *int_buf, *cur_index, buf_len;
485 
486     /*
487      * Allocate enough buffer size to fit in ibm,dynamic-memory
488      */
489     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
490     cur_index = int_buf = g_malloc0(buf_len);
491     int_buf[0] = cpu_to_be32(nr_lmbs);
492     cur_index++;
493     for (i = 0; i < nr_lmbs; i++) {
494         uint64_t addr = i * lmb_size;
495         uint32_t *dynamic_memory = cur_index;
496 
497         if (i >= device_lmb_start) {
498             SpaprDrc *drc;
499 
500             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
501             g_assert(drc);
502 
503             dynamic_memory[0] = cpu_to_be32(addr >> 32);
504             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
505             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
506             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
507             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
508             if (memory_region_present(get_system_memory(), addr)) {
509                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
510             } else {
511                 dynamic_memory[5] = cpu_to_be32(0);
512             }
513         } else {
514             /*
515              * LMB information for RMA, boot time RAM and gap b/n RAM and
516              * device memory region -- all these are marked as reserved
517              * and as having no valid DRC.
518              */
519             dynamic_memory[0] = cpu_to_be32(addr >> 32);
520             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
521             dynamic_memory[2] = cpu_to_be32(0);
522             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
523             dynamic_memory[4] = cpu_to_be32(-1);
524             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
525                                             SPAPR_LMB_FLAGS_DRC_INVALID);
526         }
527 
528         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
529     }
530     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
531     g_free(int_buf);
532     if (ret < 0) {
533         return -1;
534     }
535     return 0;
536 }
537 
538 /*
539  * Adds ibm,dynamic-reconfiguration-memory node.
540  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
541  * of this device tree node.
542  */
543 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
544                                                    void *fdt)
545 {
546     MachineState *machine = MACHINE(spapr);
547     int ret, offset;
548     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
549     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
550                                 cpu_to_be32(lmb_size & 0xffffffff)};
551     MemoryDeviceInfoList *dimms = NULL;
552 
553     /*
554      * Don't create the node if there is no device memory
555      */
556     if (machine->ram_size == machine->maxram_size) {
557         return 0;
558     }
559 
560     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
561 
562     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
563                     sizeof(prop_lmb_size));
564     if (ret < 0) {
565         return ret;
566     }
567 
568     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
569     if (ret < 0) {
570         return ret;
571     }
572 
573     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
574     if (ret < 0) {
575         return ret;
576     }
577 
578     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
579     dimms = qmp_memory_device_list();
580     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
581         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
582     } else {
583         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
584     }
585     qapi_free_MemoryDeviceInfoList(dimms);
586 
587     if (ret < 0) {
588         return ret;
589     }
590 
591     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
592 
593     return ret;
594 }
595 
596 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
597 {
598     MachineState *machine = MACHINE(spapr);
599     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
600     hwaddr mem_start, node_size;
601     int i, nb_nodes = machine->numa_state->num_nodes;
602     NodeInfo *nodes = machine->numa_state->nodes;
603 
604     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
605         if (!nodes[i].node_mem) {
606             continue;
607         }
608         if (mem_start >= machine->ram_size) {
609             node_size = 0;
610         } else {
611             node_size = nodes[i].node_mem;
612             if (node_size > machine->ram_size - mem_start) {
613                 node_size = machine->ram_size - mem_start;
614             }
615         }
616         if (!mem_start) {
617             /* spapr_machine_init() checks for rma_size <= node0_size
618              * already */
619             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
620             mem_start += spapr->rma_size;
621             node_size -= spapr->rma_size;
622         }
623         for ( ; node_size; ) {
624             hwaddr sizetmp = pow2floor(node_size);
625 
626             /* mem_start != 0 here */
627             if (ctzl(mem_start) < ctzl(sizetmp)) {
628                 sizetmp = 1ULL << ctzl(mem_start);
629             }
630 
631             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
632             node_size -= sizetmp;
633             mem_start += sizetmp;
634         }
635     }
636 
637     /* Generate ibm,dynamic-reconfiguration-memory node if required */
638     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
639         int ret;
640 
641         g_assert(smc->dr_lmb_enabled);
642         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
643         if (ret) {
644             return ret;
645         }
646     }
647 
648     return 0;
649 }
650 
651 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
652                          SpaprMachineState *spapr)
653 {
654     MachineState *ms = MACHINE(spapr);
655     PowerPCCPU *cpu = POWERPC_CPU(cs);
656     CPUPPCState *env = &cpu->env;
657     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
658     int index = spapr_get_vcpu_id(cpu);
659     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
660                        0xffffffff, 0xffffffff};
661     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
662         : SPAPR_TIMEBASE_FREQ;
663     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
664     uint32_t page_sizes_prop[64];
665     size_t page_sizes_prop_size;
666     unsigned int smp_threads = ms->smp.threads;
667     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
668     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
669     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
670     SpaprDrc *drc;
671     int drc_index;
672     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
673     int i;
674 
675     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
676     if (drc) {
677         drc_index = spapr_drc_index(drc);
678         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
679     }
680 
681     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
682     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
683 
684     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
685     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
686                            env->dcache_line_size)));
687     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
688                            env->dcache_line_size)));
689     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
690                            env->icache_line_size)));
691     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
692                            env->icache_line_size)));
693 
694     if (pcc->l1_dcache_size) {
695         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
696                                pcc->l1_dcache_size)));
697     } else {
698         warn_report("Unknown L1 dcache size for cpu");
699     }
700     if (pcc->l1_icache_size) {
701         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
702                                pcc->l1_icache_size)));
703     } else {
704         warn_report("Unknown L1 icache size for cpu");
705     }
706 
707     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
708     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
709     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
710     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
711     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
712     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
713 
714     if (env->spr_cb[SPR_PURR].oea_read) {
715         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
716     }
717     if (env->spr_cb[SPR_SPURR].oea_read) {
718         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
719     }
720 
721     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
722         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
723                           segs, sizeof(segs))));
724     }
725 
726     /* Advertise VSX (vector extensions) if available
727      *   1               == VMX / Altivec available
728      *   2               == VSX available
729      *
730      * Only CPUs for which we create core types in spapr_cpu_core.c
731      * are possible, and all of those have VMX */
732     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
733         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
734     } else {
735         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
736     }
737 
738     /* Advertise DFP (Decimal Floating Point) if available
739      *   0 / no property == no DFP
740      *   1               == DFP available */
741     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
742         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
743     }
744 
745     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
746                                                       sizeof(page_sizes_prop));
747     if (page_sizes_prop_size) {
748         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
749                           page_sizes_prop, page_sizes_prop_size)));
750     }
751 
752     spapr_dt_pa_features(spapr, cpu, fdt, offset);
753 
754     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
755                            cs->cpu_index / vcpus_per_socket)));
756 
757     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
758                       pft_size_prop, sizeof(pft_size_prop))));
759 
760     if (ms->numa_state->num_nodes > 1) {
761         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
762     }
763 
764     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
765 
766     if (pcc->radix_page_info) {
767         for (i = 0; i < pcc->radix_page_info->count; i++) {
768             radix_AP_encodings[i] =
769                 cpu_to_be32(pcc->radix_page_info->entries[i]);
770         }
771         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
772                           radix_AP_encodings,
773                           pcc->radix_page_info->count *
774                           sizeof(radix_AP_encodings[0]))));
775     }
776 
777     /*
778      * We set this property to let the guest know that it can use the large
779      * decrementer and its width in bits.
780      */
781     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
782         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
783                               pcc->lrg_decr_bits)));
784 }
785 
786 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
787 {
788     CPUState **rev;
789     CPUState *cs;
790     int n_cpus;
791     int cpus_offset;
792     char *nodename;
793     int i;
794 
795     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
796     _FDT(cpus_offset);
797     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
798     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
799 
800     /*
801      * We walk the CPUs in reverse order to ensure that CPU DT nodes
802      * created by fdt_add_subnode() end up in the right order in FDT
803      * for the guest kernel the enumerate the CPUs correctly.
804      *
805      * The CPU list cannot be traversed in reverse order, so we need
806      * to do extra work.
807      */
808     n_cpus = 0;
809     rev = NULL;
810     CPU_FOREACH(cs) {
811         rev = g_renew(CPUState *, rev, n_cpus + 1);
812         rev[n_cpus++] = cs;
813     }
814 
815     for (i = n_cpus - 1; i >= 0; i--) {
816         CPUState *cs = rev[i];
817         PowerPCCPU *cpu = POWERPC_CPU(cs);
818         int index = spapr_get_vcpu_id(cpu);
819         DeviceClass *dc = DEVICE_GET_CLASS(cs);
820         int offset;
821 
822         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
823             continue;
824         }
825 
826         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
827         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
828         g_free(nodename);
829         _FDT(offset);
830         spapr_dt_cpu(cs, fdt, offset, spapr);
831     }
832 
833     g_free(rev);
834 }
835 
836 static int spapr_dt_rng(void *fdt)
837 {
838     int node;
839     int ret;
840 
841     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
842     if (node <= 0) {
843         return -1;
844     }
845     ret = fdt_setprop_string(fdt, node, "device_type",
846                              "ibm,platform-facilities");
847     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
848     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
849 
850     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
851     if (node <= 0) {
852         return -1;
853     }
854     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
855 
856     return ret ? -1 : 0;
857 }
858 
859 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
860 {
861     MachineState *ms = MACHINE(spapr);
862     int rtas;
863     GString *hypertas = g_string_sized_new(256);
864     GString *qemu_hypertas = g_string_sized_new(256);
865     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
866         memory_region_size(&MACHINE(spapr)->device_memory->mr);
867     uint32_t lrdr_capacity[] = {
868         cpu_to_be32(max_device_addr >> 32),
869         cpu_to_be32(max_device_addr & 0xffffffff),
870         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
871         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
872         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
873     };
874 
875     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
876 
877     /* hypertas */
878     add_str(hypertas, "hcall-pft");
879     add_str(hypertas, "hcall-term");
880     add_str(hypertas, "hcall-dabr");
881     add_str(hypertas, "hcall-interrupt");
882     add_str(hypertas, "hcall-tce");
883     add_str(hypertas, "hcall-vio");
884     add_str(hypertas, "hcall-splpar");
885     add_str(hypertas, "hcall-join");
886     add_str(hypertas, "hcall-bulk");
887     add_str(hypertas, "hcall-set-mode");
888     add_str(hypertas, "hcall-sprg0");
889     add_str(hypertas, "hcall-copy");
890     add_str(hypertas, "hcall-debug");
891     add_str(hypertas, "hcall-vphn");
892     add_str(qemu_hypertas, "hcall-memop1");
893 
894     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
895         add_str(hypertas, "hcall-multi-tce");
896     }
897 
898     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
899         add_str(hypertas, "hcall-hpt-resize");
900     }
901 
902     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
903                      hypertas->str, hypertas->len));
904     g_string_free(hypertas, TRUE);
905     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
906                      qemu_hypertas->str, qemu_hypertas->len));
907     g_string_free(qemu_hypertas, TRUE);
908 
909     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
910 
911     /*
912      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
913      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
914      *
915      * The system reset requirements are driven by existing Linux and PowerVM
916      * implementation which (contrary to PAPR) saves r3 in the error log
917      * structure like machine check, so Linux expects to find the saved r3
918      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
919      * does not look at the error value).
920      *
921      * System reset interrupts are not subject to interlock like machine
922      * check, so this memory area could be corrupted if the sreset is
923      * interrupted by a machine check (or vice versa) if it was shared. To
924      * prevent this, system reset uses per-CPU areas for the sreset save
925      * area. A system reset that interrupts a system reset handler could
926      * still overwrite this area, but Linux doesn't try to recover in that
927      * case anyway.
928      *
929      * The extra 8 bytes is required because Linux's FWNMI error log check
930      * is off-by-one.
931      */
932     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
933 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
934     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
935                           RTAS_ERROR_LOG_MAX));
936     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
937                           RTAS_EVENT_SCAN_RATE));
938 
939     g_assert(msi_nonbroken);
940     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
941 
942     /*
943      * According to PAPR, rtas ibm,os-term does not guarantee a return
944      * back to the guest cpu.
945      *
946      * While an additional ibm,extended-os-term property indicates
947      * that rtas call return will always occur. Set this property.
948      */
949     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
950 
951     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
952                      lrdr_capacity, sizeof(lrdr_capacity)));
953 
954     spapr_dt_rtas_tokens(fdt, rtas);
955 }
956 
957 /*
958  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
959  * and the XIVE features that the guest may request and thus the valid
960  * values for bytes 23..26 of option vector 5:
961  */
962 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
963                                           int chosen)
964 {
965     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
966 
967     char val[2 * 4] = {
968         23, 0x00, /* XICS / XIVE mode */
969         24, 0x00, /* Hash/Radix, filled in below. */
970         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
971         26, 0x40, /* Radix options: GTSE == yes. */
972     };
973 
974     if (spapr->irq->xics && spapr->irq->xive) {
975         val[1] = SPAPR_OV5_XIVE_BOTH;
976     } else if (spapr->irq->xive) {
977         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
978     } else {
979         assert(spapr->irq->xics);
980         val[1] = SPAPR_OV5_XIVE_LEGACY;
981     }
982 
983     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
984                           first_ppc_cpu->compat_pvr)) {
985         /*
986          * If we're in a pre POWER9 compat mode then the guest should
987          * do hash and use the legacy interrupt mode
988          */
989         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
990         val[3] = 0x00; /* Hash */
991     } else if (kvm_enabled()) {
992         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
993             val[3] = 0x80; /* OV5_MMU_BOTH */
994         } else if (kvmppc_has_cap_mmu_radix()) {
995             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
996         } else {
997             val[3] = 0x00; /* Hash */
998         }
999     } else {
1000         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1001         val[3] = 0xC0;
1002     }
1003     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1004                      val, sizeof(val)));
1005 }
1006 
1007 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1008 {
1009     MachineState *machine = MACHINE(spapr);
1010     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1011     int chosen;
1012 
1013     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1014 
1015     if (reset) {
1016         const char *boot_device = machine->boot_order;
1017         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1018         size_t cb = 0;
1019         char *bootlist = get_boot_devices_list(&cb);
1020 
1021         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1022             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1023                                     machine->kernel_cmdline));
1024         }
1025 
1026         if (spapr->initrd_size) {
1027             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1028                                   spapr->initrd_base));
1029             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1030                                   spapr->initrd_base + spapr->initrd_size));
1031         }
1032 
1033         if (spapr->kernel_size) {
1034             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1035                                   cpu_to_be64(spapr->kernel_size) };
1036 
1037             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1038                          &kprop, sizeof(kprop)));
1039             if (spapr->kernel_le) {
1040                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1041             }
1042         }
1043         if (boot_menu) {
1044             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1045         }
1046         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1047         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1048         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1049 
1050         if (cb && bootlist) {
1051             int i;
1052 
1053             for (i = 0; i < cb; i++) {
1054                 if (bootlist[i] == '\n') {
1055                     bootlist[i] = ' ';
1056                 }
1057             }
1058             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1059         }
1060 
1061         if (boot_device && strlen(boot_device)) {
1062             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1063         }
1064 
1065         if (!spapr->has_graphics && stdout_path) {
1066             /*
1067              * "linux,stdout-path" and "stdout" properties are
1068              * deprecated by linux kernel. New platforms should only
1069              * use the "stdout-path" property. Set the new property
1070              * and continue using older property to remain compatible
1071              * with the existing firmware.
1072              */
1073             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1074             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1075         }
1076 
1077         /*
1078          * We can deal with BAR reallocation just fine, advertise it
1079          * to the guest
1080          */
1081         if (smc->linux_pci_probe) {
1082             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1083         }
1084 
1085         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1086 
1087         g_free(stdout_path);
1088         g_free(bootlist);
1089     }
1090 
1091     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1092 }
1093 
1094 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1095 {
1096     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1097      * KVM to work under pHyp with some guest co-operation */
1098     int hypervisor;
1099     uint8_t hypercall[16];
1100 
1101     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1102     /* indicate KVM hypercall interface */
1103     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1104     if (kvmppc_has_cap_fixup_hcalls()) {
1105         /*
1106          * Older KVM versions with older guest kernels were broken
1107          * with the magic page, don't allow the guest to map it.
1108          */
1109         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1110                                   sizeof(hypercall))) {
1111             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1112                              hypercall, sizeof(hypercall)));
1113         }
1114     }
1115 }
1116 
1117 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1118 {
1119     MachineState *machine = MACHINE(spapr);
1120     MachineClass *mc = MACHINE_GET_CLASS(machine);
1121     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1122     int ret;
1123     void *fdt;
1124     SpaprPhbState *phb;
1125     char *buf;
1126 
1127     fdt = g_malloc0(space);
1128     _FDT((fdt_create_empty_tree(fdt, space)));
1129 
1130     /* Root node */
1131     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1132     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1133     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1134 
1135     /* Guest UUID & Name*/
1136     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1137     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1138     if (qemu_uuid_set) {
1139         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1140     }
1141     g_free(buf);
1142 
1143     if (qemu_get_vm_name()) {
1144         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1145                                 qemu_get_vm_name()));
1146     }
1147 
1148     /* Host Model & Serial Number */
1149     if (spapr->host_model) {
1150         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1151     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1152         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1153         g_free(buf);
1154     }
1155 
1156     if (spapr->host_serial) {
1157         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1158     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1159         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1160         g_free(buf);
1161     }
1162 
1163     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1164     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1165 
1166     /* /interrupt controller */
1167     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1168 
1169     ret = spapr_dt_memory(spapr, fdt);
1170     if (ret < 0) {
1171         error_report("couldn't setup memory nodes in fdt");
1172         exit(1);
1173     }
1174 
1175     /* /vdevice */
1176     spapr_dt_vdevice(spapr->vio_bus, fdt);
1177 
1178     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1179         ret = spapr_dt_rng(fdt);
1180         if (ret < 0) {
1181             error_report("could not set up rng device in the fdt");
1182             exit(1);
1183         }
1184     }
1185 
1186     QLIST_FOREACH(phb, &spapr->phbs, list) {
1187         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1188         if (ret < 0) {
1189             error_report("couldn't setup PCI devices in fdt");
1190             exit(1);
1191         }
1192     }
1193 
1194     spapr_dt_cpus(fdt, spapr);
1195 
1196     if (smc->dr_lmb_enabled) {
1197         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1198     }
1199 
1200     if (mc->has_hotpluggable_cpus) {
1201         int offset = fdt_path_offset(fdt, "/cpus");
1202         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1203         if (ret < 0) {
1204             error_report("Couldn't set up CPU DR device tree properties");
1205             exit(1);
1206         }
1207     }
1208 
1209     /* /event-sources */
1210     spapr_dt_events(spapr, fdt);
1211 
1212     /* /rtas */
1213     spapr_dt_rtas(spapr, fdt);
1214 
1215     /* /chosen */
1216     spapr_dt_chosen(spapr, fdt, reset);
1217 
1218     /* /hypervisor */
1219     if (kvm_enabled()) {
1220         spapr_dt_hypervisor(spapr, fdt);
1221     }
1222 
1223     /* Build memory reserve map */
1224     if (reset) {
1225         if (spapr->kernel_size) {
1226             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1227                                   spapr->kernel_size)));
1228         }
1229         if (spapr->initrd_size) {
1230             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1231                                   spapr->initrd_size)));
1232         }
1233     }
1234 
1235     if (smc->dr_phb_enabled) {
1236         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1237         if (ret < 0) {
1238             error_report("Couldn't set up PHB DR device tree properties");
1239             exit(1);
1240         }
1241     }
1242 
1243     /* NVDIMM devices */
1244     if (mc->nvdimm_supported) {
1245         spapr_dt_persistent_memory(spapr, fdt);
1246     }
1247 
1248     return fdt;
1249 }
1250 
1251 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1252 {
1253     SpaprMachineState *spapr = opaque;
1254 
1255     return (addr & 0x0fffffff) + spapr->kernel_addr;
1256 }
1257 
1258 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1259                                     PowerPCCPU *cpu)
1260 {
1261     CPUPPCState *env = &cpu->env;
1262 
1263     /* The TCG path should also be holding the BQL at this point */
1264     g_assert(qemu_mutex_iothread_locked());
1265 
1266     if (msr_pr) {
1267         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1268         env->gpr[3] = H_PRIVILEGE;
1269     } else {
1270         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1271     }
1272 }
1273 
1274 struct LPCRSyncState {
1275     target_ulong value;
1276     target_ulong mask;
1277 };
1278 
1279 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1280 {
1281     struct LPCRSyncState *s = arg.host_ptr;
1282     PowerPCCPU *cpu = POWERPC_CPU(cs);
1283     CPUPPCState *env = &cpu->env;
1284     target_ulong lpcr;
1285 
1286     cpu_synchronize_state(cs);
1287     lpcr = env->spr[SPR_LPCR];
1288     lpcr &= ~s->mask;
1289     lpcr |= s->value;
1290     ppc_store_lpcr(cpu, lpcr);
1291 }
1292 
1293 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1294 {
1295     CPUState *cs;
1296     struct LPCRSyncState s = {
1297         .value = value,
1298         .mask = mask
1299     };
1300     CPU_FOREACH(cs) {
1301         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1302     }
1303 }
1304 
1305 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1306 {
1307     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1308 
1309     /* Copy PATE1:GR into PATE0:HR */
1310     entry->dw0 = spapr->patb_entry & PATE0_HR;
1311     entry->dw1 = spapr->patb_entry;
1312 }
1313 
1314 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1315 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1316 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1317 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1318 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1319 
1320 /*
1321  * Get the fd to access the kernel htab, re-opening it if necessary
1322  */
1323 static int get_htab_fd(SpaprMachineState *spapr)
1324 {
1325     Error *local_err = NULL;
1326 
1327     if (spapr->htab_fd >= 0) {
1328         return spapr->htab_fd;
1329     }
1330 
1331     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1332     if (spapr->htab_fd < 0) {
1333         error_report_err(local_err);
1334     }
1335 
1336     return spapr->htab_fd;
1337 }
1338 
1339 void close_htab_fd(SpaprMachineState *spapr)
1340 {
1341     if (spapr->htab_fd >= 0) {
1342         close(spapr->htab_fd);
1343     }
1344     spapr->htab_fd = -1;
1345 }
1346 
1347 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1348 {
1349     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1350 
1351     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1352 }
1353 
1354 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1355 {
1356     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1357 
1358     assert(kvm_enabled());
1359 
1360     if (!spapr->htab) {
1361         return 0;
1362     }
1363 
1364     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1365 }
1366 
1367 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1368                                                 hwaddr ptex, int n)
1369 {
1370     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1371     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1372 
1373     if (!spapr->htab) {
1374         /*
1375          * HTAB is controlled by KVM. Fetch into temporary buffer
1376          */
1377         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1378         kvmppc_read_hptes(hptes, ptex, n);
1379         return hptes;
1380     }
1381 
1382     /*
1383      * HTAB is controlled by QEMU. Just point to the internally
1384      * accessible PTEG.
1385      */
1386     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1387 }
1388 
1389 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1390                               const ppc_hash_pte64_t *hptes,
1391                               hwaddr ptex, int n)
1392 {
1393     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1394 
1395     if (!spapr->htab) {
1396         g_free((void *)hptes);
1397     }
1398 
1399     /* Nothing to do for qemu managed HPT */
1400 }
1401 
1402 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1403                       uint64_t pte0, uint64_t pte1)
1404 {
1405     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1406     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1407 
1408     if (!spapr->htab) {
1409         kvmppc_write_hpte(ptex, pte0, pte1);
1410     } else {
1411         if (pte0 & HPTE64_V_VALID) {
1412             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1413             /*
1414              * When setting valid, we write PTE1 first. This ensures
1415              * proper synchronization with the reading code in
1416              * ppc_hash64_pteg_search()
1417              */
1418             smp_wmb();
1419             stq_p(spapr->htab + offset, pte0);
1420         } else {
1421             stq_p(spapr->htab + offset, pte0);
1422             /*
1423              * When clearing it we set PTE0 first. This ensures proper
1424              * synchronization with the reading code in
1425              * ppc_hash64_pteg_search()
1426              */
1427             smp_wmb();
1428             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1429         }
1430     }
1431 }
1432 
1433 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1434                              uint64_t pte1)
1435 {
1436     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1437     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1438 
1439     if (!spapr->htab) {
1440         /* There should always be a hash table when this is called */
1441         error_report("spapr_hpte_set_c called with no hash table !");
1442         return;
1443     }
1444 
1445     /* The HW performs a non-atomic byte update */
1446     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1447 }
1448 
1449 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1450                              uint64_t pte1)
1451 {
1452     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1453     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1454 
1455     if (!spapr->htab) {
1456         /* There should always be a hash table when this is called */
1457         error_report("spapr_hpte_set_r called with no hash table !");
1458         return;
1459     }
1460 
1461     /* The HW performs a non-atomic byte update */
1462     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1463 }
1464 
1465 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1466 {
1467     int shift;
1468 
1469     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1470      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1471      * that's much more than is needed for Linux guests */
1472     shift = ctz64(pow2ceil(ramsize)) - 7;
1473     shift = MAX(shift, 18); /* Minimum architected size */
1474     shift = MIN(shift, 46); /* Maximum architected size */
1475     return shift;
1476 }
1477 
1478 void spapr_free_hpt(SpaprMachineState *spapr)
1479 {
1480     g_free(spapr->htab);
1481     spapr->htab = NULL;
1482     spapr->htab_shift = 0;
1483     close_htab_fd(spapr);
1484 }
1485 
1486 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1487 {
1488     ERRP_GUARD();
1489     long rc;
1490 
1491     /* Clean up any HPT info from a previous boot */
1492     spapr_free_hpt(spapr);
1493 
1494     rc = kvmppc_reset_htab(shift);
1495 
1496     if (rc == -EOPNOTSUPP) {
1497         error_setg(errp, "HPT not supported in nested guests");
1498         return -EOPNOTSUPP;
1499     }
1500 
1501     if (rc < 0) {
1502         /* kernel-side HPT needed, but couldn't allocate one */
1503         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1504                          shift);
1505         error_append_hint(errp, "Try smaller maxmem?\n");
1506         return -errno;
1507     } else if (rc > 0) {
1508         /* kernel-side HPT allocated */
1509         if (rc != shift) {
1510             error_setg(errp,
1511                        "Requested order %d HPT, but kernel allocated order %ld",
1512                        shift, rc);
1513             error_append_hint(errp, "Try smaller maxmem?\n");
1514             return -ENOSPC;
1515         }
1516 
1517         spapr->htab_shift = shift;
1518         spapr->htab = NULL;
1519     } else {
1520         /* kernel-side HPT not needed, allocate in userspace instead */
1521         size_t size = 1ULL << shift;
1522         int i;
1523 
1524         spapr->htab = qemu_memalign(size, size);
1525         memset(spapr->htab, 0, size);
1526         spapr->htab_shift = shift;
1527 
1528         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1529             DIRTY_HPTE(HPTE(spapr->htab, i));
1530         }
1531     }
1532     /* We're setting up a hash table, so that means we're not radix */
1533     spapr->patb_entry = 0;
1534     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1535     return 0;
1536 }
1537 
1538 void spapr_setup_hpt(SpaprMachineState *spapr)
1539 {
1540     int hpt_shift;
1541 
1542     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1543         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1544     } else {
1545         uint64_t current_ram_size;
1546 
1547         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1548         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1549     }
1550     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1551 
1552     if (kvm_enabled()) {
1553         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1554 
1555         /* Check our RMA fits in the possible VRMA */
1556         if (vrma_limit < spapr->rma_size) {
1557             error_report("Unable to create %" HWADDR_PRIu
1558                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1559                          spapr->rma_size / MiB, vrma_limit / MiB);
1560             exit(EXIT_FAILURE);
1561         }
1562     }
1563 }
1564 
1565 static int spapr_reset_drcs(Object *child, void *opaque)
1566 {
1567     SpaprDrc *drc =
1568         (SpaprDrc *) object_dynamic_cast(child,
1569                                                  TYPE_SPAPR_DR_CONNECTOR);
1570 
1571     if (drc) {
1572         spapr_drc_reset(drc);
1573     }
1574 
1575     return 0;
1576 }
1577 
1578 static void spapr_machine_reset(MachineState *machine)
1579 {
1580     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1581     PowerPCCPU *first_ppc_cpu;
1582     hwaddr fdt_addr;
1583     void *fdt;
1584     int rc;
1585 
1586     kvmppc_svm_off(&error_fatal);
1587     spapr_caps_apply(spapr);
1588 
1589     first_ppc_cpu = POWERPC_CPU(first_cpu);
1590     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1591         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1592                               spapr->max_compat_pvr)) {
1593         /*
1594          * If using KVM with radix mode available, VCPUs can be started
1595          * without a HPT because KVM will start them in radix mode.
1596          * Set the GR bit in PATE so that we know there is no HPT.
1597          */
1598         spapr->patb_entry = PATE1_GR;
1599         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1600     } else {
1601         spapr_setup_hpt(spapr);
1602     }
1603 
1604     qemu_devices_reset();
1605 
1606     spapr_ovec_cleanup(spapr->ov5_cas);
1607     spapr->ov5_cas = spapr_ovec_new();
1608 
1609     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1610 
1611     /*
1612      * This is fixing some of the default configuration of the XIVE
1613      * devices. To be called after the reset of the machine devices.
1614      */
1615     spapr_irq_reset(spapr, &error_fatal);
1616 
1617     /*
1618      * There is no CAS under qtest. Simulate one to please the code that
1619      * depends on spapr->ov5_cas. This is especially needed to test device
1620      * unplug, so we do that before resetting the DRCs.
1621      */
1622     if (qtest_enabled()) {
1623         spapr_ovec_cleanup(spapr->ov5_cas);
1624         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1625     }
1626 
1627     /* DRC reset may cause a device to be unplugged. This will cause troubles
1628      * if this device is used by another device (eg, a running vhost backend
1629      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1630      * situations, we reset DRCs after all devices have been reset.
1631      */
1632     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1633 
1634     spapr_clear_pending_events(spapr);
1635 
1636     /*
1637      * We place the device tree and RTAS just below either the top of the RMA,
1638      * or just below 2GB, whichever is lower, so that it can be
1639      * processed with 32-bit real mode code if necessary
1640      */
1641     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1642 
1643     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1644 
1645     rc = fdt_pack(fdt);
1646 
1647     /* Should only fail if we've built a corrupted tree */
1648     assert(rc == 0);
1649 
1650     /* Load the fdt */
1651     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1652     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1653     g_free(spapr->fdt_blob);
1654     spapr->fdt_size = fdt_totalsize(fdt);
1655     spapr->fdt_initial_size = spapr->fdt_size;
1656     spapr->fdt_blob = fdt;
1657 
1658     /* Set up the entry state */
1659     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1660     first_ppc_cpu->env.gpr[5] = 0;
1661 
1662     spapr->fwnmi_system_reset_addr = -1;
1663     spapr->fwnmi_machine_check_addr = -1;
1664     spapr->fwnmi_machine_check_interlock = -1;
1665 
1666     /* Signal all vCPUs waiting on this condition */
1667     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1668 
1669     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1670 }
1671 
1672 static void spapr_create_nvram(SpaprMachineState *spapr)
1673 {
1674     DeviceState *dev = qdev_new("spapr-nvram");
1675     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1676 
1677     if (dinfo) {
1678         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1679                                 &error_fatal);
1680     }
1681 
1682     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1683 
1684     spapr->nvram = (struct SpaprNvram *)dev;
1685 }
1686 
1687 static void spapr_rtc_create(SpaprMachineState *spapr)
1688 {
1689     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1690                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1691                                        &error_fatal, NULL);
1692     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1693     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1694                               "date");
1695 }
1696 
1697 /* Returns whether we want to use VGA or not */
1698 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1699 {
1700     switch (vga_interface_type) {
1701     case VGA_NONE:
1702         return false;
1703     case VGA_DEVICE:
1704         return true;
1705     case VGA_STD:
1706     case VGA_VIRTIO:
1707     case VGA_CIRRUS:
1708         return pci_vga_init(pci_bus) != NULL;
1709     default:
1710         error_setg(errp,
1711                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1712         return false;
1713     }
1714 }
1715 
1716 static int spapr_pre_load(void *opaque)
1717 {
1718     int rc;
1719 
1720     rc = spapr_caps_pre_load(opaque);
1721     if (rc) {
1722         return rc;
1723     }
1724 
1725     return 0;
1726 }
1727 
1728 static int spapr_post_load(void *opaque, int version_id)
1729 {
1730     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1731     int err = 0;
1732 
1733     err = spapr_caps_post_migration(spapr);
1734     if (err) {
1735         return err;
1736     }
1737 
1738     /*
1739      * In earlier versions, there was no separate qdev for the PAPR
1740      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1741      * So when migrating from those versions, poke the incoming offset
1742      * value into the RTC device
1743      */
1744     if (version_id < 3) {
1745         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1746         if (err) {
1747             return err;
1748         }
1749     }
1750 
1751     if (kvm_enabled() && spapr->patb_entry) {
1752         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1753         bool radix = !!(spapr->patb_entry & PATE1_GR);
1754         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1755 
1756         /*
1757          * Update LPCR:HR and UPRT as they may not be set properly in
1758          * the stream
1759          */
1760         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1761                             LPCR_HR | LPCR_UPRT);
1762 
1763         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1764         if (err) {
1765             error_report("Process table config unsupported by the host");
1766             return -EINVAL;
1767         }
1768     }
1769 
1770     err = spapr_irq_post_load(spapr, version_id);
1771     if (err) {
1772         return err;
1773     }
1774 
1775     return err;
1776 }
1777 
1778 static int spapr_pre_save(void *opaque)
1779 {
1780     int rc;
1781 
1782     rc = spapr_caps_pre_save(opaque);
1783     if (rc) {
1784         return rc;
1785     }
1786 
1787     return 0;
1788 }
1789 
1790 static bool version_before_3(void *opaque, int version_id)
1791 {
1792     return version_id < 3;
1793 }
1794 
1795 static bool spapr_pending_events_needed(void *opaque)
1796 {
1797     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1798     return !QTAILQ_EMPTY(&spapr->pending_events);
1799 }
1800 
1801 static const VMStateDescription vmstate_spapr_event_entry = {
1802     .name = "spapr_event_log_entry",
1803     .version_id = 1,
1804     .minimum_version_id = 1,
1805     .fields = (VMStateField[]) {
1806         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1807         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1808         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1809                                      NULL, extended_length),
1810         VMSTATE_END_OF_LIST()
1811     },
1812 };
1813 
1814 static const VMStateDescription vmstate_spapr_pending_events = {
1815     .name = "spapr_pending_events",
1816     .version_id = 1,
1817     .minimum_version_id = 1,
1818     .needed = spapr_pending_events_needed,
1819     .fields = (VMStateField[]) {
1820         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1821                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1822         VMSTATE_END_OF_LIST()
1823     },
1824 };
1825 
1826 static bool spapr_ov5_cas_needed(void *opaque)
1827 {
1828     SpaprMachineState *spapr = opaque;
1829     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1830     bool cas_needed;
1831 
1832     /* Prior to the introduction of SpaprOptionVector, we had two option
1833      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1834      * Both of these options encode machine topology into the device-tree
1835      * in such a way that the now-booted OS should still be able to interact
1836      * appropriately with QEMU regardless of what options were actually
1837      * negotiatied on the source side.
1838      *
1839      * As such, we can avoid migrating the CAS-negotiated options if these
1840      * are the only options available on the current machine/platform.
1841      * Since these are the only options available for pseries-2.7 and
1842      * earlier, this allows us to maintain old->new/new->old migration
1843      * compatibility.
1844      *
1845      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1846      * via default pseries-2.8 machines and explicit command-line parameters.
1847      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1848      * of the actual CAS-negotiated values to continue working properly. For
1849      * example, availability of memory unplug depends on knowing whether
1850      * OV5_HP_EVT was negotiated via CAS.
1851      *
1852      * Thus, for any cases where the set of available CAS-negotiatable
1853      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1854      * include the CAS-negotiated options in the migration stream, unless
1855      * if they affect boot time behaviour only.
1856      */
1857     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1858     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1859     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1860 
1861     /* We need extra information if we have any bits outside the mask
1862      * defined above */
1863     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1864 
1865     spapr_ovec_cleanup(ov5_mask);
1866 
1867     return cas_needed;
1868 }
1869 
1870 static const VMStateDescription vmstate_spapr_ov5_cas = {
1871     .name = "spapr_option_vector_ov5_cas",
1872     .version_id = 1,
1873     .minimum_version_id = 1,
1874     .needed = spapr_ov5_cas_needed,
1875     .fields = (VMStateField[]) {
1876         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1877                                  vmstate_spapr_ovec, SpaprOptionVector),
1878         VMSTATE_END_OF_LIST()
1879     },
1880 };
1881 
1882 static bool spapr_patb_entry_needed(void *opaque)
1883 {
1884     SpaprMachineState *spapr = opaque;
1885 
1886     return !!spapr->patb_entry;
1887 }
1888 
1889 static const VMStateDescription vmstate_spapr_patb_entry = {
1890     .name = "spapr_patb_entry",
1891     .version_id = 1,
1892     .minimum_version_id = 1,
1893     .needed = spapr_patb_entry_needed,
1894     .fields = (VMStateField[]) {
1895         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1896         VMSTATE_END_OF_LIST()
1897     },
1898 };
1899 
1900 static bool spapr_irq_map_needed(void *opaque)
1901 {
1902     SpaprMachineState *spapr = opaque;
1903 
1904     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1905 }
1906 
1907 static const VMStateDescription vmstate_spapr_irq_map = {
1908     .name = "spapr_irq_map",
1909     .version_id = 1,
1910     .minimum_version_id = 1,
1911     .needed = spapr_irq_map_needed,
1912     .fields = (VMStateField[]) {
1913         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1914         VMSTATE_END_OF_LIST()
1915     },
1916 };
1917 
1918 static bool spapr_dtb_needed(void *opaque)
1919 {
1920     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1921 
1922     return smc->update_dt_enabled;
1923 }
1924 
1925 static int spapr_dtb_pre_load(void *opaque)
1926 {
1927     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1928 
1929     g_free(spapr->fdt_blob);
1930     spapr->fdt_blob = NULL;
1931     spapr->fdt_size = 0;
1932 
1933     return 0;
1934 }
1935 
1936 static const VMStateDescription vmstate_spapr_dtb = {
1937     .name = "spapr_dtb",
1938     .version_id = 1,
1939     .minimum_version_id = 1,
1940     .needed = spapr_dtb_needed,
1941     .pre_load = spapr_dtb_pre_load,
1942     .fields = (VMStateField[]) {
1943         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1944         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1945         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1946                                      fdt_size),
1947         VMSTATE_END_OF_LIST()
1948     },
1949 };
1950 
1951 static bool spapr_fwnmi_needed(void *opaque)
1952 {
1953     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1954 
1955     return spapr->fwnmi_machine_check_addr != -1;
1956 }
1957 
1958 static int spapr_fwnmi_pre_save(void *opaque)
1959 {
1960     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1961 
1962     /*
1963      * Check if machine check handling is in progress and print a
1964      * warning message.
1965      */
1966     if (spapr->fwnmi_machine_check_interlock != -1) {
1967         warn_report("A machine check is being handled during migration. The"
1968                 "handler may run and log hardware error on the destination");
1969     }
1970 
1971     return 0;
1972 }
1973 
1974 static const VMStateDescription vmstate_spapr_fwnmi = {
1975     .name = "spapr_fwnmi",
1976     .version_id = 1,
1977     .minimum_version_id = 1,
1978     .needed = spapr_fwnmi_needed,
1979     .pre_save = spapr_fwnmi_pre_save,
1980     .fields = (VMStateField[]) {
1981         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1982         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1983         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1984         VMSTATE_END_OF_LIST()
1985     },
1986 };
1987 
1988 static const VMStateDescription vmstate_spapr = {
1989     .name = "spapr",
1990     .version_id = 3,
1991     .minimum_version_id = 1,
1992     .pre_load = spapr_pre_load,
1993     .post_load = spapr_post_load,
1994     .pre_save = spapr_pre_save,
1995     .fields = (VMStateField[]) {
1996         /* used to be @next_irq */
1997         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1998 
1999         /* RTC offset */
2000         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2001 
2002         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2003         VMSTATE_END_OF_LIST()
2004     },
2005     .subsections = (const VMStateDescription*[]) {
2006         &vmstate_spapr_ov5_cas,
2007         &vmstate_spapr_patb_entry,
2008         &vmstate_spapr_pending_events,
2009         &vmstate_spapr_cap_htm,
2010         &vmstate_spapr_cap_vsx,
2011         &vmstate_spapr_cap_dfp,
2012         &vmstate_spapr_cap_cfpc,
2013         &vmstate_spapr_cap_sbbc,
2014         &vmstate_spapr_cap_ibs,
2015         &vmstate_spapr_cap_hpt_maxpagesize,
2016         &vmstate_spapr_irq_map,
2017         &vmstate_spapr_cap_nested_kvm_hv,
2018         &vmstate_spapr_dtb,
2019         &vmstate_spapr_cap_large_decr,
2020         &vmstate_spapr_cap_ccf_assist,
2021         &vmstate_spapr_cap_fwnmi,
2022         &vmstate_spapr_fwnmi,
2023         NULL
2024     }
2025 };
2026 
2027 static int htab_save_setup(QEMUFile *f, void *opaque)
2028 {
2029     SpaprMachineState *spapr = opaque;
2030 
2031     /* "Iteration" header */
2032     if (!spapr->htab_shift) {
2033         qemu_put_be32(f, -1);
2034     } else {
2035         qemu_put_be32(f, spapr->htab_shift);
2036     }
2037 
2038     if (spapr->htab) {
2039         spapr->htab_save_index = 0;
2040         spapr->htab_first_pass = true;
2041     } else {
2042         if (spapr->htab_shift) {
2043             assert(kvm_enabled());
2044         }
2045     }
2046 
2047 
2048     return 0;
2049 }
2050 
2051 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2052                             int chunkstart, int n_valid, int n_invalid)
2053 {
2054     qemu_put_be32(f, chunkstart);
2055     qemu_put_be16(f, n_valid);
2056     qemu_put_be16(f, n_invalid);
2057     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2058                     HASH_PTE_SIZE_64 * n_valid);
2059 }
2060 
2061 static void htab_save_end_marker(QEMUFile *f)
2062 {
2063     qemu_put_be32(f, 0);
2064     qemu_put_be16(f, 0);
2065     qemu_put_be16(f, 0);
2066 }
2067 
2068 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2069                                  int64_t max_ns)
2070 {
2071     bool has_timeout = max_ns != -1;
2072     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2073     int index = spapr->htab_save_index;
2074     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2075 
2076     assert(spapr->htab_first_pass);
2077 
2078     do {
2079         int chunkstart;
2080 
2081         /* Consume invalid HPTEs */
2082         while ((index < htabslots)
2083                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2084             CLEAN_HPTE(HPTE(spapr->htab, index));
2085             index++;
2086         }
2087 
2088         /* Consume valid HPTEs */
2089         chunkstart = index;
2090         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2091                && HPTE_VALID(HPTE(spapr->htab, index))) {
2092             CLEAN_HPTE(HPTE(spapr->htab, index));
2093             index++;
2094         }
2095 
2096         if (index > chunkstart) {
2097             int n_valid = index - chunkstart;
2098 
2099             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2100 
2101             if (has_timeout &&
2102                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2103                 break;
2104             }
2105         }
2106     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2107 
2108     if (index >= htabslots) {
2109         assert(index == htabslots);
2110         index = 0;
2111         spapr->htab_first_pass = false;
2112     }
2113     spapr->htab_save_index = index;
2114 }
2115 
2116 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2117                                 int64_t max_ns)
2118 {
2119     bool final = max_ns < 0;
2120     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2121     int examined = 0, sent = 0;
2122     int index = spapr->htab_save_index;
2123     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2124 
2125     assert(!spapr->htab_first_pass);
2126 
2127     do {
2128         int chunkstart, invalidstart;
2129 
2130         /* Consume non-dirty HPTEs */
2131         while ((index < htabslots)
2132                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2133             index++;
2134             examined++;
2135         }
2136 
2137         chunkstart = index;
2138         /* Consume valid dirty HPTEs */
2139         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2140                && HPTE_DIRTY(HPTE(spapr->htab, index))
2141                && HPTE_VALID(HPTE(spapr->htab, index))) {
2142             CLEAN_HPTE(HPTE(spapr->htab, index));
2143             index++;
2144             examined++;
2145         }
2146 
2147         invalidstart = index;
2148         /* Consume invalid dirty HPTEs */
2149         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2150                && HPTE_DIRTY(HPTE(spapr->htab, index))
2151                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2152             CLEAN_HPTE(HPTE(spapr->htab, index));
2153             index++;
2154             examined++;
2155         }
2156 
2157         if (index > chunkstart) {
2158             int n_valid = invalidstart - chunkstart;
2159             int n_invalid = index - invalidstart;
2160 
2161             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2162             sent += index - chunkstart;
2163 
2164             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2165                 break;
2166             }
2167         }
2168 
2169         if (examined >= htabslots) {
2170             break;
2171         }
2172 
2173         if (index >= htabslots) {
2174             assert(index == htabslots);
2175             index = 0;
2176         }
2177     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2178 
2179     if (index >= htabslots) {
2180         assert(index == htabslots);
2181         index = 0;
2182     }
2183 
2184     spapr->htab_save_index = index;
2185 
2186     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2187 }
2188 
2189 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2190 #define MAX_KVM_BUF_SIZE    2048
2191 
2192 static int htab_save_iterate(QEMUFile *f, void *opaque)
2193 {
2194     SpaprMachineState *spapr = opaque;
2195     int fd;
2196     int rc = 0;
2197 
2198     /* Iteration header */
2199     if (!spapr->htab_shift) {
2200         qemu_put_be32(f, -1);
2201         return 1;
2202     } else {
2203         qemu_put_be32(f, 0);
2204     }
2205 
2206     if (!spapr->htab) {
2207         assert(kvm_enabled());
2208 
2209         fd = get_htab_fd(spapr);
2210         if (fd < 0) {
2211             return fd;
2212         }
2213 
2214         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2215         if (rc < 0) {
2216             return rc;
2217         }
2218     } else  if (spapr->htab_first_pass) {
2219         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2220     } else {
2221         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2222     }
2223 
2224     htab_save_end_marker(f);
2225 
2226     return rc;
2227 }
2228 
2229 static int htab_save_complete(QEMUFile *f, void *opaque)
2230 {
2231     SpaprMachineState *spapr = opaque;
2232     int fd;
2233 
2234     /* Iteration header */
2235     if (!spapr->htab_shift) {
2236         qemu_put_be32(f, -1);
2237         return 0;
2238     } else {
2239         qemu_put_be32(f, 0);
2240     }
2241 
2242     if (!spapr->htab) {
2243         int rc;
2244 
2245         assert(kvm_enabled());
2246 
2247         fd = get_htab_fd(spapr);
2248         if (fd < 0) {
2249             return fd;
2250         }
2251 
2252         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2253         if (rc < 0) {
2254             return rc;
2255         }
2256     } else {
2257         if (spapr->htab_first_pass) {
2258             htab_save_first_pass(f, spapr, -1);
2259         }
2260         htab_save_later_pass(f, spapr, -1);
2261     }
2262 
2263     /* End marker */
2264     htab_save_end_marker(f);
2265 
2266     return 0;
2267 }
2268 
2269 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2270 {
2271     SpaprMachineState *spapr = opaque;
2272     uint32_t section_hdr;
2273     int fd = -1;
2274     Error *local_err = NULL;
2275 
2276     if (version_id < 1 || version_id > 1) {
2277         error_report("htab_load() bad version");
2278         return -EINVAL;
2279     }
2280 
2281     section_hdr = qemu_get_be32(f);
2282 
2283     if (section_hdr == -1) {
2284         spapr_free_hpt(spapr);
2285         return 0;
2286     }
2287 
2288     if (section_hdr) {
2289         int ret;
2290 
2291         /* First section gives the htab size */
2292         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2293         if (ret < 0) {
2294             error_report_err(local_err);
2295             return ret;
2296         }
2297         return 0;
2298     }
2299 
2300     if (!spapr->htab) {
2301         assert(kvm_enabled());
2302 
2303         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2304         if (fd < 0) {
2305             error_report_err(local_err);
2306             return fd;
2307         }
2308     }
2309 
2310     while (true) {
2311         uint32_t index;
2312         uint16_t n_valid, n_invalid;
2313 
2314         index = qemu_get_be32(f);
2315         n_valid = qemu_get_be16(f);
2316         n_invalid = qemu_get_be16(f);
2317 
2318         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2319             /* End of Stream */
2320             break;
2321         }
2322 
2323         if ((index + n_valid + n_invalid) >
2324             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2325             /* Bad index in stream */
2326             error_report(
2327                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2328                 index, n_valid, n_invalid, spapr->htab_shift);
2329             return -EINVAL;
2330         }
2331 
2332         if (spapr->htab) {
2333             if (n_valid) {
2334                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2335                                 HASH_PTE_SIZE_64 * n_valid);
2336             }
2337             if (n_invalid) {
2338                 memset(HPTE(spapr->htab, index + n_valid), 0,
2339                        HASH_PTE_SIZE_64 * n_invalid);
2340             }
2341         } else {
2342             int rc;
2343 
2344             assert(fd >= 0);
2345 
2346             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2347                                         &local_err);
2348             if (rc < 0) {
2349                 error_report_err(local_err);
2350                 return rc;
2351             }
2352         }
2353     }
2354 
2355     if (!spapr->htab) {
2356         assert(fd >= 0);
2357         close(fd);
2358     }
2359 
2360     return 0;
2361 }
2362 
2363 static void htab_save_cleanup(void *opaque)
2364 {
2365     SpaprMachineState *spapr = opaque;
2366 
2367     close_htab_fd(spapr);
2368 }
2369 
2370 static SaveVMHandlers savevm_htab_handlers = {
2371     .save_setup = htab_save_setup,
2372     .save_live_iterate = htab_save_iterate,
2373     .save_live_complete_precopy = htab_save_complete,
2374     .save_cleanup = htab_save_cleanup,
2375     .load_state = htab_load,
2376 };
2377 
2378 static void spapr_boot_set(void *opaque, const char *boot_device,
2379                            Error **errp)
2380 {
2381     MachineState *machine = MACHINE(opaque);
2382     machine->boot_order = g_strdup(boot_device);
2383 }
2384 
2385 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2386 {
2387     MachineState *machine = MACHINE(spapr);
2388     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2389     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2390     int i;
2391 
2392     for (i = 0; i < nr_lmbs; i++) {
2393         uint64_t addr;
2394 
2395         addr = i * lmb_size + machine->device_memory->base;
2396         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2397                                addr / lmb_size);
2398     }
2399 }
2400 
2401 /*
2402  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2403  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2404  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2405  */
2406 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2407 {
2408     int i;
2409 
2410     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2411         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2412                    " is not aligned to %" PRIu64 " MiB",
2413                    machine->ram_size,
2414                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2415         return;
2416     }
2417 
2418     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2419         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2420                    " is not aligned to %" PRIu64 " MiB",
2421                    machine->ram_size,
2422                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2423         return;
2424     }
2425 
2426     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2427         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2428             error_setg(errp,
2429                        "Node %d memory size 0x%" PRIx64
2430                        " is not aligned to %" PRIu64 " MiB",
2431                        i, machine->numa_state->nodes[i].node_mem,
2432                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2433             return;
2434         }
2435     }
2436 }
2437 
2438 /* find cpu slot in machine->possible_cpus by core_id */
2439 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2440 {
2441     int index = id / ms->smp.threads;
2442 
2443     if (index >= ms->possible_cpus->len) {
2444         return NULL;
2445     }
2446     if (idx) {
2447         *idx = index;
2448     }
2449     return &ms->possible_cpus->cpus[index];
2450 }
2451 
2452 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2453 {
2454     MachineState *ms = MACHINE(spapr);
2455     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2456     Error *local_err = NULL;
2457     bool vsmt_user = !!spapr->vsmt;
2458     int kvm_smt = kvmppc_smt_threads();
2459     int ret;
2460     unsigned int smp_threads = ms->smp.threads;
2461 
2462     if (!kvm_enabled() && (smp_threads > 1)) {
2463         error_setg(errp, "TCG cannot support more than 1 thread/core "
2464                    "on a pseries machine");
2465         return;
2466     }
2467     if (!is_power_of_2(smp_threads)) {
2468         error_setg(errp, "Cannot support %d threads/core on a pseries "
2469                    "machine because it must be a power of 2", smp_threads);
2470         return;
2471     }
2472 
2473     /* Detemine the VSMT mode to use: */
2474     if (vsmt_user) {
2475         if (spapr->vsmt < smp_threads) {
2476             error_setg(errp, "Cannot support VSMT mode %d"
2477                        " because it must be >= threads/core (%d)",
2478                        spapr->vsmt, smp_threads);
2479             return;
2480         }
2481         /* In this case, spapr->vsmt has been set by the command line */
2482     } else if (!smc->smp_threads_vsmt) {
2483         /*
2484          * Default VSMT value is tricky, because we need it to be as
2485          * consistent as possible (for migration), but this requires
2486          * changing it for at least some existing cases.  We pick 8 as
2487          * the value that we'd get with KVM on POWER8, the
2488          * overwhelmingly common case in production systems.
2489          */
2490         spapr->vsmt = MAX(8, smp_threads);
2491     } else {
2492         spapr->vsmt = smp_threads;
2493     }
2494 
2495     /* KVM: If necessary, set the SMT mode: */
2496     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2497         ret = kvmppc_set_smt_threads(spapr->vsmt);
2498         if (ret) {
2499             /* Looks like KVM isn't able to change VSMT mode */
2500             error_setg(&local_err,
2501                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2502                        spapr->vsmt, ret);
2503             /* We can live with that if the default one is big enough
2504              * for the number of threads, and a submultiple of the one
2505              * we want.  In this case we'll waste some vcpu ids, but
2506              * behaviour will be correct */
2507             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2508                 warn_report_err(local_err);
2509             } else {
2510                 if (!vsmt_user) {
2511                     error_append_hint(&local_err,
2512                                       "On PPC, a VM with %d threads/core"
2513                                       " on a host with %d threads/core"
2514                                       " requires the use of VSMT mode %d.\n",
2515                                       smp_threads, kvm_smt, spapr->vsmt);
2516                 }
2517                 kvmppc_error_append_smt_possible_hint(&local_err);
2518                 error_propagate(errp, local_err);
2519             }
2520         }
2521     }
2522     /* else TCG: nothing to do currently */
2523 }
2524 
2525 static void spapr_init_cpus(SpaprMachineState *spapr)
2526 {
2527     MachineState *machine = MACHINE(spapr);
2528     MachineClass *mc = MACHINE_GET_CLASS(machine);
2529     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2530     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2531     const CPUArchIdList *possible_cpus;
2532     unsigned int smp_cpus = machine->smp.cpus;
2533     unsigned int smp_threads = machine->smp.threads;
2534     unsigned int max_cpus = machine->smp.max_cpus;
2535     int boot_cores_nr = smp_cpus / smp_threads;
2536     int i;
2537 
2538     possible_cpus = mc->possible_cpu_arch_ids(machine);
2539     if (mc->has_hotpluggable_cpus) {
2540         if (smp_cpus % smp_threads) {
2541             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2542                          smp_cpus, smp_threads);
2543             exit(1);
2544         }
2545         if (max_cpus % smp_threads) {
2546             error_report("max_cpus (%u) must be multiple of threads (%u)",
2547                          max_cpus, smp_threads);
2548             exit(1);
2549         }
2550     } else {
2551         if (max_cpus != smp_cpus) {
2552             error_report("This machine version does not support CPU hotplug");
2553             exit(1);
2554         }
2555         boot_cores_nr = possible_cpus->len;
2556     }
2557 
2558     if (smc->pre_2_10_has_unused_icps) {
2559         int i;
2560 
2561         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2562             /* Dummy entries get deregistered when real ICPState objects
2563              * are registered during CPU core hotplug.
2564              */
2565             pre_2_10_vmstate_register_dummy_icp(i);
2566         }
2567     }
2568 
2569     for (i = 0; i < possible_cpus->len; i++) {
2570         int core_id = i * smp_threads;
2571 
2572         if (mc->has_hotpluggable_cpus) {
2573             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2574                                    spapr_vcpu_id(spapr, core_id));
2575         }
2576 
2577         if (i < boot_cores_nr) {
2578             Object *core  = object_new(type);
2579             int nr_threads = smp_threads;
2580 
2581             /* Handle the partially filled core for older machine types */
2582             if ((i + 1) * smp_threads >= smp_cpus) {
2583                 nr_threads = smp_cpus - i * smp_threads;
2584             }
2585 
2586             object_property_set_int(core, "nr-threads", nr_threads,
2587                                     &error_fatal);
2588             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2589                                     &error_fatal);
2590             qdev_realize(DEVICE(core), NULL, &error_fatal);
2591 
2592             object_unref(core);
2593         }
2594     }
2595 }
2596 
2597 static PCIHostState *spapr_create_default_phb(void)
2598 {
2599     DeviceState *dev;
2600 
2601     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2602     qdev_prop_set_uint32(dev, "index", 0);
2603     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2604 
2605     return PCI_HOST_BRIDGE(dev);
2606 }
2607 
2608 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2609 {
2610     MachineState *machine = MACHINE(spapr);
2611     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2612     hwaddr rma_size = machine->ram_size;
2613     hwaddr node0_size = spapr_node0_size(machine);
2614 
2615     /* RMA has to fit in the first NUMA node */
2616     rma_size = MIN(rma_size, node0_size);
2617 
2618     /*
2619      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2620      * never exceed that
2621      */
2622     rma_size = MIN(rma_size, 1 * TiB);
2623 
2624     /*
2625      * Clamp the RMA size based on machine type.  This is for
2626      * migration compatibility with older qemu versions, which limited
2627      * the RMA size for complicated and mostly bad reasons.
2628      */
2629     if (smc->rma_limit) {
2630         rma_size = MIN(rma_size, smc->rma_limit);
2631     }
2632 
2633     if (rma_size < MIN_RMA_SLOF) {
2634         error_setg(errp,
2635                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2636                    "ldMiB guest RMA (Real Mode Area memory)",
2637                    MIN_RMA_SLOF / MiB);
2638         return 0;
2639     }
2640 
2641     return rma_size;
2642 }
2643 
2644 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2645 {
2646     MachineState *machine = MACHINE(spapr);
2647     int i;
2648 
2649     for (i = 0; i < machine->ram_slots; i++) {
2650         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2651     }
2652 }
2653 
2654 /* pSeries LPAR / sPAPR hardware init */
2655 static void spapr_machine_init(MachineState *machine)
2656 {
2657     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2658     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2659     MachineClass *mc = MACHINE_GET_CLASS(machine);
2660     const char *kernel_filename = machine->kernel_filename;
2661     const char *initrd_filename = machine->initrd_filename;
2662     PCIHostState *phb;
2663     int i;
2664     MemoryRegion *sysmem = get_system_memory();
2665     long load_limit, fw_size;
2666     char *filename;
2667     Error *resize_hpt_err = NULL;
2668 
2669     msi_nonbroken = true;
2670 
2671     QLIST_INIT(&spapr->phbs);
2672     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2673 
2674     /* Determine capabilities to run with */
2675     spapr_caps_init(spapr);
2676 
2677     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2678     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2679         /*
2680          * If the user explicitly requested a mode we should either
2681          * supply it, or fail completely (which we do below).  But if
2682          * it's not set explicitly, we reset our mode to something
2683          * that works
2684          */
2685         if (resize_hpt_err) {
2686             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2687             error_free(resize_hpt_err);
2688             resize_hpt_err = NULL;
2689         } else {
2690             spapr->resize_hpt = smc->resize_hpt_default;
2691         }
2692     }
2693 
2694     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2695 
2696     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2697         /*
2698          * User requested HPT resize, but this host can't supply it.  Bail out
2699          */
2700         error_report_err(resize_hpt_err);
2701         exit(1);
2702     }
2703     error_free(resize_hpt_err);
2704 
2705     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2706 
2707     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2708     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2709 
2710     /*
2711      * VSMT must be set in order to be able to compute VCPU ids, ie to
2712      * call spapr_max_server_number() or spapr_vcpu_id().
2713      */
2714     spapr_set_vsmt_mode(spapr, &error_fatal);
2715 
2716     /* Set up Interrupt Controller before we create the VCPUs */
2717     spapr_irq_init(spapr, &error_fatal);
2718 
2719     /* Set up containers for ibm,client-architecture-support negotiated options
2720      */
2721     spapr->ov5 = spapr_ovec_new();
2722     spapr->ov5_cas = spapr_ovec_new();
2723 
2724     if (smc->dr_lmb_enabled) {
2725         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2726         spapr_validate_node_memory(machine, &error_fatal);
2727     }
2728 
2729     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2730 
2731     /* advertise support for dedicated HP event source to guests */
2732     if (spapr->use_hotplug_event_source) {
2733         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2734     }
2735 
2736     /* advertise support for HPT resizing */
2737     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2738         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2739     }
2740 
2741     /* advertise support for ibm,dyamic-memory-v2 */
2742     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2743 
2744     /* advertise XIVE on POWER9 machines */
2745     if (spapr->irq->xive) {
2746         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2747     }
2748 
2749     /* init CPUs */
2750     spapr_init_cpus(spapr);
2751 
2752     /*
2753      * check we don't have a memory-less/cpu-less NUMA node
2754      * Firmware relies on the existing memory/cpu topology to provide the
2755      * NUMA topology to the kernel.
2756      * And the linux kernel needs to know the NUMA topology at start
2757      * to be able to hotplug CPUs later.
2758      */
2759     if (machine->numa_state->num_nodes) {
2760         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2761             /* check for memory-less node */
2762             if (machine->numa_state->nodes[i].node_mem == 0) {
2763                 CPUState *cs;
2764                 int found = 0;
2765                 /* check for cpu-less node */
2766                 CPU_FOREACH(cs) {
2767                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2768                     if (cpu->node_id == i) {
2769                         found = 1;
2770                         break;
2771                     }
2772                 }
2773                 /* memory-less and cpu-less node */
2774                 if (!found) {
2775                     error_report(
2776                        "Memory-less/cpu-less nodes are not supported (node %d)",
2777                                  i);
2778                     exit(1);
2779                 }
2780             }
2781         }
2782 
2783     }
2784 
2785     /*
2786      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2787      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2788      * called from vPHB reset handler so we initialize the counter here.
2789      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2790      * must be equally distant from any other node.
2791      * The final value of spapr->gpu_numa_id is going to be written to
2792      * max-associativity-domains in spapr_build_fdt().
2793      */
2794     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2795 
2796     /* Init numa_assoc_array */
2797     spapr_numa_associativity_init(spapr, machine);
2798 
2799     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2800         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2801                               spapr->max_compat_pvr)) {
2802         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2803         /* KVM and TCG always allow GTSE with radix... */
2804         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2805     }
2806     /* ... but not with hash (currently). */
2807 
2808     if (kvm_enabled()) {
2809         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2810         kvmppc_enable_logical_ci_hcalls();
2811         kvmppc_enable_set_mode_hcall();
2812 
2813         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2814         kvmppc_enable_clear_ref_mod_hcalls();
2815 
2816         /* Enable H_PAGE_INIT */
2817         kvmppc_enable_h_page_init();
2818     }
2819 
2820     /* map RAM */
2821     memory_region_add_subregion(sysmem, 0, machine->ram);
2822 
2823     /* always allocate the device memory information */
2824     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2825 
2826     /* initialize hotplug memory address space */
2827     if (machine->ram_size < machine->maxram_size) {
2828         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2829         /*
2830          * Limit the number of hotpluggable memory slots to half the number
2831          * slots that KVM supports, leaving the other half for PCI and other
2832          * devices. However ensure that number of slots doesn't drop below 32.
2833          */
2834         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2835                            SPAPR_MAX_RAM_SLOTS;
2836 
2837         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2838             max_memslots = SPAPR_MAX_RAM_SLOTS;
2839         }
2840         if (machine->ram_slots > max_memslots) {
2841             error_report("Specified number of memory slots %"
2842                          PRIu64" exceeds max supported %d",
2843                          machine->ram_slots, max_memslots);
2844             exit(1);
2845         }
2846 
2847         machine->device_memory->base = ROUND_UP(machine->ram_size,
2848                                                 SPAPR_DEVICE_MEM_ALIGN);
2849         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2850                            "device-memory", device_mem_size);
2851         memory_region_add_subregion(sysmem, machine->device_memory->base,
2852                                     &machine->device_memory->mr);
2853     }
2854 
2855     if (smc->dr_lmb_enabled) {
2856         spapr_create_lmb_dr_connectors(spapr);
2857     }
2858 
2859     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2860         /* Create the error string for live migration blocker */
2861         error_setg(&spapr->fwnmi_migration_blocker,
2862             "A machine check is being handled during migration. The handler"
2863             "may run and log hardware error on the destination");
2864     }
2865 
2866     if (mc->nvdimm_supported) {
2867         spapr_create_nvdimm_dr_connectors(spapr);
2868     }
2869 
2870     /* Set up RTAS event infrastructure */
2871     spapr_events_init(spapr);
2872 
2873     /* Set up the RTC RTAS interfaces */
2874     spapr_rtc_create(spapr);
2875 
2876     /* Set up VIO bus */
2877     spapr->vio_bus = spapr_vio_bus_init();
2878 
2879     for (i = 0; i < serial_max_hds(); i++) {
2880         if (serial_hd(i)) {
2881             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2882         }
2883     }
2884 
2885     /* We always have at least the nvram device on VIO */
2886     spapr_create_nvram(spapr);
2887 
2888     /*
2889      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2890      * connectors (described in root DT node's "ibm,drc-types" property)
2891      * are pre-initialized here. additional child connectors (such as
2892      * connectors for a PHBs PCI slots) are added as needed during their
2893      * parent's realization.
2894      */
2895     if (smc->dr_phb_enabled) {
2896         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2897             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2898         }
2899     }
2900 
2901     /* Set up PCI */
2902     spapr_pci_rtas_init();
2903 
2904     phb = spapr_create_default_phb();
2905 
2906     for (i = 0; i < nb_nics; i++) {
2907         NICInfo *nd = &nd_table[i];
2908 
2909         if (!nd->model) {
2910             nd->model = g_strdup("spapr-vlan");
2911         }
2912 
2913         if (g_str_equal(nd->model, "spapr-vlan") ||
2914             g_str_equal(nd->model, "ibmveth")) {
2915             spapr_vlan_create(spapr->vio_bus, nd);
2916         } else {
2917             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2918         }
2919     }
2920 
2921     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2922         spapr_vscsi_create(spapr->vio_bus);
2923     }
2924 
2925     /* Graphics */
2926     if (spapr_vga_init(phb->bus, &error_fatal)) {
2927         spapr->has_graphics = true;
2928         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2929     }
2930 
2931     if (machine->usb) {
2932         if (smc->use_ohci_by_default) {
2933             pci_create_simple(phb->bus, -1, "pci-ohci");
2934         } else {
2935             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2936         }
2937 
2938         if (spapr->has_graphics) {
2939             USBBus *usb_bus = usb_bus_find(-1);
2940 
2941             usb_create_simple(usb_bus, "usb-kbd");
2942             usb_create_simple(usb_bus, "usb-mouse");
2943         }
2944     }
2945 
2946     if (kernel_filename) {
2947         spapr->kernel_size = load_elf(kernel_filename, NULL,
2948                                       translate_kernel_address, spapr,
2949                                       NULL, NULL, NULL, NULL, 1,
2950                                       PPC_ELF_MACHINE, 0, 0);
2951         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2952             spapr->kernel_size = load_elf(kernel_filename, NULL,
2953                                           translate_kernel_address, spapr,
2954                                           NULL, NULL, NULL, NULL, 0,
2955                                           PPC_ELF_MACHINE, 0, 0);
2956             spapr->kernel_le = spapr->kernel_size > 0;
2957         }
2958         if (spapr->kernel_size < 0) {
2959             error_report("error loading %s: %s", kernel_filename,
2960                          load_elf_strerror(spapr->kernel_size));
2961             exit(1);
2962         }
2963 
2964         /* load initrd */
2965         if (initrd_filename) {
2966             /* Try to locate the initrd in the gap between the kernel
2967              * and the firmware. Add a bit of space just in case
2968              */
2969             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2970                                   + 0x1ffff) & ~0xffff;
2971             spapr->initrd_size = load_image_targphys(initrd_filename,
2972                                                      spapr->initrd_base,
2973                                                      load_limit
2974                                                      - spapr->initrd_base);
2975             if (spapr->initrd_size < 0) {
2976                 error_report("could not load initial ram disk '%s'",
2977                              initrd_filename);
2978                 exit(1);
2979             }
2980         }
2981     }
2982 
2983     if (bios_name == NULL) {
2984         bios_name = FW_FILE_NAME;
2985     }
2986     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2987     if (!filename) {
2988         error_report("Could not find LPAR firmware '%s'", bios_name);
2989         exit(1);
2990     }
2991     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2992     if (fw_size <= 0) {
2993         error_report("Could not load LPAR firmware '%s'", filename);
2994         exit(1);
2995     }
2996     g_free(filename);
2997 
2998     /* FIXME: Should register things through the MachineState's qdev
2999      * interface, this is a legacy from the sPAPREnvironment structure
3000      * which predated MachineState but had a similar function */
3001     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3002     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3003                          &savevm_htab_handlers, spapr);
3004 
3005     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3006 
3007     qemu_register_boot_set(spapr_boot_set, spapr);
3008 
3009     /*
3010      * Nothing needs to be done to resume a suspended guest because
3011      * suspending does not change the machine state, so no need for
3012      * a ->wakeup method.
3013      */
3014     qemu_register_wakeup_support();
3015 
3016     if (kvm_enabled()) {
3017         /* to stop and start vmclock */
3018         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3019                                          &spapr->tb);
3020 
3021         kvmppc_spapr_enable_inkernel_multitce();
3022     }
3023 
3024     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3025 }
3026 
3027 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3028 {
3029     if (!vm_type) {
3030         return 0;
3031     }
3032 
3033     if (!strcmp(vm_type, "HV")) {
3034         return 1;
3035     }
3036 
3037     if (!strcmp(vm_type, "PR")) {
3038         return 2;
3039     }
3040 
3041     error_report("Unknown kvm-type specified '%s'", vm_type);
3042     exit(1);
3043 }
3044 
3045 /*
3046  * Implementation of an interface to adjust firmware path
3047  * for the bootindex property handling.
3048  */
3049 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3050                                    DeviceState *dev)
3051 {
3052 #define CAST(type, obj, name) \
3053     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3054     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3055     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3056     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3057 
3058     if (d) {
3059         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3060         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3061         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3062 
3063         if (spapr) {
3064             /*
3065              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3066              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3067              * 0x8000 | (target << 8) | (bus << 5) | lun
3068              * (see the "Logical unit addressing format" table in SAM5)
3069              */
3070             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3071             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3072                                    (uint64_t)id << 48);
3073         } else if (virtio) {
3074             /*
3075              * We use SRP luns of the form 01000000 | (target << 8) | lun
3076              * in the top 32 bits of the 64-bit LUN
3077              * Note: the quote above is from SLOF and it is wrong,
3078              * the actual binding is:
3079              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3080              */
3081             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3082             if (d->lun >= 256) {
3083                 /* Use the LUN "flat space addressing method" */
3084                 id |= 0x4000;
3085             }
3086             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3087                                    (uint64_t)id << 32);
3088         } else if (usb) {
3089             /*
3090              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3091              * in the top 32 bits of the 64-bit LUN
3092              */
3093             unsigned usb_port = atoi(usb->port->path);
3094             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3095             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3096                                    (uint64_t)id << 32);
3097         }
3098     }
3099 
3100     /*
3101      * SLOF probes the USB devices, and if it recognizes that the device is a
3102      * storage device, it changes its name to "storage" instead of "usb-host",
3103      * and additionally adds a child node for the SCSI LUN, so the correct
3104      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3105      */
3106     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3107         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3108         if (usb_host_dev_is_scsi_storage(usbdev)) {
3109             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3110         }
3111     }
3112 
3113     if (phb) {
3114         /* Replace "pci" with "pci@800000020000000" */
3115         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3116     }
3117 
3118     if (vsc) {
3119         /* Same logic as virtio above */
3120         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3121         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3122     }
3123 
3124     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3125         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3126         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3127         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3128     }
3129 
3130     return NULL;
3131 }
3132 
3133 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3134 {
3135     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3136 
3137     return g_strdup(spapr->kvm_type);
3138 }
3139 
3140 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3141 {
3142     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3143 
3144     g_free(spapr->kvm_type);
3145     spapr->kvm_type = g_strdup(value);
3146 }
3147 
3148 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3149 {
3150     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3151 
3152     return spapr->use_hotplug_event_source;
3153 }
3154 
3155 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3156                                             Error **errp)
3157 {
3158     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3159 
3160     spapr->use_hotplug_event_source = value;
3161 }
3162 
3163 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3164 {
3165     return true;
3166 }
3167 
3168 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3169 {
3170     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3171 
3172     switch (spapr->resize_hpt) {
3173     case SPAPR_RESIZE_HPT_DEFAULT:
3174         return g_strdup("default");
3175     case SPAPR_RESIZE_HPT_DISABLED:
3176         return g_strdup("disabled");
3177     case SPAPR_RESIZE_HPT_ENABLED:
3178         return g_strdup("enabled");
3179     case SPAPR_RESIZE_HPT_REQUIRED:
3180         return g_strdup("required");
3181     }
3182     g_assert_not_reached();
3183 }
3184 
3185 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3186 {
3187     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3188 
3189     if (strcmp(value, "default") == 0) {
3190         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3191     } else if (strcmp(value, "disabled") == 0) {
3192         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3193     } else if (strcmp(value, "enabled") == 0) {
3194         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3195     } else if (strcmp(value, "required") == 0) {
3196         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3197     } else {
3198         error_setg(errp, "Bad value for \"resize-hpt\" property");
3199     }
3200 }
3201 
3202 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3203 {
3204     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3205 
3206     if (spapr->irq == &spapr_irq_xics_legacy) {
3207         return g_strdup("legacy");
3208     } else if (spapr->irq == &spapr_irq_xics) {
3209         return g_strdup("xics");
3210     } else if (spapr->irq == &spapr_irq_xive) {
3211         return g_strdup("xive");
3212     } else if (spapr->irq == &spapr_irq_dual) {
3213         return g_strdup("dual");
3214     }
3215     g_assert_not_reached();
3216 }
3217 
3218 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3219 {
3220     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3221 
3222     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3223         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3224         return;
3225     }
3226 
3227     /* The legacy IRQ backend can not be set */
3228     if (strcmp(value, "xics") == 0) {
3229         spapr->irq = &spapr_irq_xics;
3230     } else if (strcmp(value, "xive") == 0) {
3231         spapr->irq = &spapr_irq_xive;
3232     } else if (strcmp(value, "dual") == 0) {
3233         spapr->irq = &spapr_irq_dual;
3234     } else {
3235         error_setg(errp, "Bad value for \"ic-mode\" property");
3236     }
3237 }
3238 
3239 static char *spapr_get_host_model(Object *obj, Error **errp)
3240 {
3241     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3242 
3243     return g_strdup(spapr->host_model);
3244 }
3245 
3246 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3247 {
3248     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3249 
3250     g_free(spapr->host_model);
3251     spapr->host_model = g_strdup(value);
3252 }
3253 
3254 static char *spapr_get_host_serial(Object *obj, Error **errp)
3255 {
3256     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3257 
3258     return g_strdup(spapr->host_serial);
3259 }
3260 
3261 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3262 {
3263     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3264 
3265     g_free(spapr->host_serial);
3266     spapr->host_serial = g_strdup(value);
3267 }
3268 
3269 static void spapr_instance_init(Object *obj)
3270 {
3271     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3272     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3273 
3274     spapr->htab_fd = -1;
3275     spapr->use_hotplug_event_source = true;
3276     object_property_add_str(obj, "kvm-type",
3277                             spapr_get_kvm_type, spapr_set_kvm_type);
3278     object_property_set_description(obj, "kvm-type",
3279                                     "Specifies the KVM virtualization mode (HV, PR)");
3280     object_property_add_bool(obj, "modern-hotplug-events",
3281                             spapr_get_modern_hotplug_events,
3282                             spapr_set_modern_hotplug_events);
3283     object_property_set_description(obj, "modern-hotplug-events",
3284                                     "Use dedicated hotplug event mechanism in"
3285                                     " place of standard EPOW events when possible"
3286                                     " (required for memory hot-unplug support)");
3287     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3288                             "Maximum permitted CPU compatibility mode");
3289 
3290     object_property_add_str(obj, "resize-hpt",
3291                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3292     object_property_set_description(obj, "resize-hpt",
3293                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3294     object_property_add_uint32_ptr(obj, "vsmt",
3295                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3296     object_property_set_description(obj, "vsmt",
3297                                     "Virtual SMT: KVM behaves as if this were"
3298                                     " the host's SMT mode");
3299 
3300     object_property_add_bool(obj, "vfio-no-msix-emulation",
3301                              spapr_get_msix_emulation, NULL);
3302 
3303     object_property_add_uint64_ptr(obj, "kernel-addr",
3304                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3305     object_property_set_description(obj, "kernel-addr",
3306                                     stringify(KERNEL_LOAD_ADDR)
3307                                     " for -kernel is the default");
3308     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3309     /* The machine class defines the default interrupt controller mode */
3310     spapr->irq = smc->irq;
3311     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3312                             spapr_set_ic_mode);
3313     object_property_set_description(obj, "ic-mode",
3314                  "Specifies the interrupt controller mode (xics, xive, dual)");
3315 
3316     object_property_add_str(obj, "host-model",
3317         spapr_get_host_model, spapr_set_host_model);
3318     object_property_set_description(obj, "host-model",
3319         "Host model to advertise in guest device tree");
3320     object_property_add_str(obj, "host-serial",
3321         spapr_get_host_serial, spapr_set_host_serial);
3322     object_property_set_description(obj, "host-serial",
3323         "Host serial number to advertise in guest device tree");
3324 }
3325 
3326 static void spapr_machine_finalizefn(Object *obj)
3327 {
3328     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3329 
3330     g_free(spapr->kvm_type);
3331 }
3332 
3333 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3334 {
3335     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3336     PowerPCCPU *cpu = POWERPC_CPU(cs);
3337     CPUPPCState *env = &cpu->env;
3338 
3339     cpu_synchronize_state(cs);
3340     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3341     if (spapr->fwnmi_system_reset_addr != -1) {
3342         uint64_t rtas_addr, addr;
3343 
3344         /* get rtas addr from fdt */
3345         rtas_addr = spapr_get_rtas_addr();
3346         if (!rtas_addr) {
3347             qemu_system_guest_panicked(NULL);
3348             return;
3349         }
3350 
3351         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3352         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3353         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3354         env->gpr[3] = addr;
3355     }
3356     ppc_cpu_do_system_reset(cs);
3357     if (spapr->fwnmi_system_reset_addr != -1) {
3358         env->nip = spapr->fwnmi_system_reset_addr;
3359     }
3360 }
3361 
3362 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3363 {
3364     CPUState *cs;
3365 
3366     CPU_FOREACH(cs) {
3367         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3368     }
3369 }
3370 
3371 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3372                           void *fdt, int *fdt_start_offset, Error **errp)
3373 {
3374     uint64_t addr;
3375     uint32_t node;
3376 
3377     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3378     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3379                                     &error_abort);
3380     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3381                                              SPAPR_MEMORY_BLOCK_SIZE);
3382     return 0;
3383 }
3384 
3385 static bool spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3386                            bool dedicated_hp_event_source, Error **errp)
3387 {
3388     SpaprDrc *drc;
3389     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3390     int i;
3391     uint64_t addr = addr_start;
3392     bool hotplugged = spapr_drc_hotplugged(dev);
3393 
3394     for (i = 0; i < nr_lmbs; i++) {
3395         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3396                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3397         g_assert(drc);
3398 
3399         if (!spapr_drc_attach(drc, dev, errp)) {
3400             while (addr > addr_start) {
3401                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3402                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3403                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3404                 spapr_drc_detach(drc);
3405             }
3406             return false;
3407         }
3408         if (!hotplugged) {
3409             spapr_drc_reset(drc);
3410         }
3411         addr += SPAPR_MEMORY_BLOCK_SIZE;
3412     }
3413     /* send hotplug notification to the
3414      * guest only in case of hotplugged memory
3415      */
3416     if (hotplugged) {
3417         if (dedicated_hp_event_source) {
3418             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3419                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3420             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3421                                                    nr_lmbs,
3422                                                    spapr_drc_index(drc));
3423         } else {
3424             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3425                                            nr_lmbs);
3426         }
3427     }
3428     return true;
3429 }
3430 
3431 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3432                               Error **errp)
3433 {
3434     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3435     PCDIMMDevice *dimm = PC_DIMM(dev);
3436     uint64_t size, addr;
3437     int64_t slot;
3438     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3439 
3440     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3441 
3442     pc_dimm_plug(dimm, MACHINE(ms));
3443 
3444     if (!is_nvdimm) {
3445         addr = object_property_get_uint(OBJECT(dimm),
3446                                         PC_DIMM_ADDR_PROP, &error_abort);
3447         if (!spapr_add_lmbs(dev, addr, size,
3448                             spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT), errp)) {
3449             goto out_unplug;
3450         }
3451     } else {
3452         slot = object_property_get_int(OBJECT(dimm),
3453                                        PC_DIMM_SLOT_PROP, &error_abort);
3454         /* We should have valid slot number at this point */
3455         g_assert(slot >= 0);
3456         if (!spapr_add_nvdimm(dev, slot, errp)) {
3457             goto out_unplug;
3458         }
3459     }
3460 
3461     return;
3462 
3463 out_unplug:
3464     pc_dimm_unplug(dimm, MACHINE(ms));
3465 }
3466 
3467 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3468                                   Error **errp)
3469 {
3470     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3471     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3472     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3473     PCDIMMDevice *dimm = PC_DIMM(dev);
3474     Error *local_err = NULL;
3475     uint64_t size;
3476     Object *memdev;
3477     hwaddr pagesize;
3478 
3479     if (!smc->dr_lmb_enabled) {
3480         error_setg(errp, "Memory hotplug not supported for this machine");
3481         return;
3482     }
3483 
3484     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3485     if (local_err) {
3486         error_propagate(errp, local_err);
3487         return;
3488     }
3489 
3490     if (is_nvdimm) {
3491         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3492             return;
3493         }
3494     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3495         error_setg(errp, "Hotplugged memory size must be a multiple of "
3496                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3497         return;
3498     }
3499 
3500     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3501                                       &error_abort);
3502     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3503     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3504         return;
3505     }
3506 
3507     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3508 }
3509 
3510 struct SpaprDimmState {
3511     PCDIMMDevice *dimm;
3512     uint32_t nr_lmbs;
3513     QTAILQ_ENTRY(SpaprDimmState) next;
3514 };
3515 
3516 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3517                                                        PCDIMMDevice *dimm)
3518 {
3519     SpaprDimmState *dimm_state = NULL;
3520 
3521     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3522         if (dimm_state->dimm == dimm) {
3523             break;
3524         }
3525     }
3526     return dimm_state;
3527 }
3528 
3529 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3530                                                       uint32_t nr_lmbs,
3531                                                       PCDIMMDevice *dimm)
3532 {
3533     SpaprDimmState *ds = NULL;
3534 
3535     /*
3536      * If this request is for a DIMM whose removal had failed earlier
3537      * (due to guest's refusal to remove the LMBs), we would have this
3538      * dimm already in the pending_dimm_unplugs list. In that
3539      * case don't add again.
3540      */
3541     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3542     if (!ds) {
3543         ds = g_malloc0(sizeof(SpaprDimmState));
3544         ds->nr_lmbs = nr_lmbs;
3545         ds->dimm = dimm;
3546         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3547     }
3548     return ds;
3549 }
3550 
3551 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3552                                               SpaprDimmState *dimm_state)
3553 {
3554     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3555     g_free(dimm_state);
3556 }
3557 
3558 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3559                                                         PCDIMMDevice *dimm)
3560 {
3561     SpaprDrc *drc;
3562     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3563                                                   &error_abort);
3564     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3565     uint32_t avail_lmbs = 0;
3566     uint64_t addr_start, addr;
3567     int i;
3568 
3569     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3570                                           &error_abort);
3571 
3572     addr = addr_start;
3573     for (i = 0; i < nr_lmbs; i++) {
3574         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3575                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3576         g_assert(drc);
3577         if (drc->dev) {
3578             avail_lmbs++;
3579         }
3580         addr += SPAPR_MEMORY_BLOCK_SIZE;
3581     }
3582 
3583     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3584 }
3585 
3586 /* Callback to be called during DRC release. */
3587 void spapr_lmb_release(DeviceState *dev)
3588 {
3589     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3590     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3591     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3592 
3593     /* This information will get lost if a migration occurs
3594      * during the unplug process. In this case recover it. */
3595     if (ds == NULL) {
3596         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3597         g_assert(ds);
3598         /* The DRC being examined by the caller at least must be counted */
3599         g_assert(ds->nr_lmbs);
3600     }
3601 
3602     if (--ds->nr_lmbs) {
3603         return;
3604     }
3605 
3606     /*
3607      * Now that all the LMBs have been removed by the guest, call the
3608      * unplug handler chain. This can never fail.
3609      */
3610     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3611     object_unparent(OBJECT(dev));
3612 }
3613 
3614 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3615 {
3616     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3617     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3618 
3619     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3620     qdev_unrealize(dev);
3621     spapr_pending_dimm_unplugs_remove(spapr, ds);
3622 }
3623 
3624 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3625                                         DeviceState *dev, Error **errp)
3626 {
3627     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3628     PCDIMMDevice *dimm = PC_DIMM(dev);
3629     uint32_t nr_lmbs;
3630     uint64_t size, addr_start, addr;
3631     int i;
3632     SpaprDrc *drc;
3633 
3634     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3635         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3636         return;
3637     }
3638 
3639     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3640     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3641 
3642     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3643                                           &error_abort);
3644 
3645     /*
3646      * An existing pending dimm state for this DIMM means that there is an
3647      * unplug operation in progress, waiting for the spapr_lmb_release
3648      * callback to complete the job (BQL can't cover that far). In this case,
3649      * bail out to avoid detaching DRCs that were already released.
3650      */
3651     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3652         error_setg(errp, "Memory unplug already in progress for device %s",
3653                    dev->id);
3654         return;
3655     }
3656 
3657     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3658 
3659     addr = addr_start;
3660     for (i = 0; i < nr_lmbs; i++) {
3661         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3662                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3663         g_assert(drc);
3664 
3665         spapr_drc_detach(drc);
3666         addr += SPAPR_MEMORY_BLOCK_SIZE;
3667     }
3668 
3669     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3670                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3671     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3672                                               nr_lmbs, spapr_drc_index(drc));
3673 }
3674 
3675 /* Callback to be called during DRC release. */
3676 void spapr_core_release(DeviceState *dev)
3677 {
3678     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3679 
3680     /* Call the unplug handler chain. This can never fail. */
3681     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3682     object_unparent(OBJECT(dev));
3683 }
3684 
3685 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3686 {
3687     MachineState *ms = MACHINE(hotplug_dev);
3688     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3689     CPUCore *cc = CPU_CORE(dev);
3690     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3691 
3692     if (smc->pre_2_10_has_unused_icps) {
3693         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3694         int i;
3695 
3696         for (i = 0; i < cc->nr_threads; i++) {
3697             CPUState *cs = CPU(sc->threads[i]);
3698 
3699             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3700         }
3701     }
3702 
3703     assert(core_slot);
3704     core_slot->cpu = NULL;
3705     qdev_unrealize(dev);
3706 }
3707 
3708 static
3709 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3710                                Error **errp)
3711 {
3712     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3713     int index;
3714     SpaprDrc *drc;
3715     CPUCore *cc = CPU_CORE(dev);
3716 
3717     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3718         error_setg(errp, "Unable to find CPU core with core-id: %d",
3719                    cc->core_id);
3720         return;
3721     }
3722     if (index == 0) {
3723         error_setg(errp, "Boot CPU core may not be unplugged");
3724         return;
3725     }
3726 
3727     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3728                           spapr_vcpu_id(spapr, cc->core_id));
3729     g_assert(drc);
3730 
3731     if (!spapr_drc_unplug_requested(drc)) {
3732         spapr_drc_detach(drc);
3733         spapr_hotplug_req_remove_by_index(drc);
3734     }
3735 }
3736 
3737 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3738                            void *fdt, int *fdt_start_offset, Error **errp)
3739 {
3740     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3741     CPUState *cs = CPU(core->threads[0]);
3742     PowerPCCPU *cpu = POWERPC_CPU(cs);
3743     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3744     int id = spapr_get_vcpu_id(cpu);
3745     char *nodename;
3746     int offset;
3747 
3748     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3749     offset = fdt_add_subnode(fdt, 0, nodename);
3750     g_free(nodename);
3751 
3752     spapr_dt_cpu(cs, fdt, offset, spapr);
3753 
3754     *fdt_start_offset = offset;
3755     return 0;
3756 }
3757 
3758 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3759                             Error **errp)
3760 {
3761     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3762     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3763     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3764     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3765     CPUCore *cc = CPU_CORE(dev);
3766     CPUState *cs;
3767     SpaprDrc *drc;
3768     CPUArchId *core_slot;
3769     int index;
3770     bool hotplugged = spapr_drc_hotplugged(dev);
3771     int i;
3772 
3773     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3774     if (!core_slot) {
3775         error_setg(errp, "Unable to find CPU core with core-id: %d",
3776                    cc->core_id);
3777         return;
3778     }
3779     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3780                           spapr_vcpu_id(spapr, cc->core_id));
3781 
3782     g_assert(drc || !mc->has_hotpluggable_cpus);
3783 
3784     if (drc) {
3785         if (!spapr_drc_attach(drc, dev, errp)) {
3786             return;
3787         }
3788 
3789         if (hotplugged) {
3790             /*
3791              * Send hotplug notification interrupt to the guest only
3792              * in case of hotplugged CPUs.
3793              */
3794             spapr_hotplug_req_add_by_index(drc);
3795         } else {
3796             spapr_drc_reset(drc);
3797         }
3798     }
3799 
3800     core_slot->cpu = OBJECT(dev);
3801 
3802     if (smc->pre_2_10_has_unused_icps) {
3803         for (i = 0; i < cc->nr_threads; i++) {
3804             cs = CPU(core->threads[i]);
3805             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3806         }
3807     }
3808 
3809     /*
3810      * Set compatibility mode to match the boot CPU, which was either set
3811      * by the machine reset code or by CAS.
3812      */
3813     if (hotplugged) {
3814         for (i = 0; i < cc->nr_threads; i++) {
3815             if (ppc_set_compat(core->threads[i],
3816                                POWERPC_CPU(first_cpu)->compat_pvr,
3817                                errp) < 0) {
3818                 return;
3819             }
3820         }
3821     }
3822 }
3823 
3824 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3825                                 Error **errp)
3826 {
3827     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3828     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3829     CPUCore *cc = CPU_CORE(dev);
3830     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3831     const char *type = object_get_typename(OBJECT(dev));
3832     CPUArchId *core_slot;
3833     int index;
3834     unsigned int smp_threads = machine->smp.threads;
3835 
3836     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3837         error_setg(errp, "CPU hotplug not supported for this machine");
3838         return;
3839     }
3840 
3841     if (strcmp(base_core_type, type)) {
3842         error_setg(errp, "CPU core type should be %s", base_core_type);
3843         return;
3844     }
3845 
3846     if (cc->core_id % smp_threads) {
3847         error_setg(errp, "invalid core id %d", cc->core_id);
3848         return;
3849     }
3850 
3851     /*
3852      * In general we should have homogeneous threads-per-core, but old
3853      * (pre hotplug support) machine types allow the last core to have
3854      * reduced threads as a compatibility hack for when we allowed
3855      * total vcpus not a multiple of threads-per-core.
3856      */
3857     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3858         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3859                    smp_threads);
3860         return;
3861     }
3862 
3863     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3864     if (!core_slot) {
3865         error_setg(errp, "core id %d out of range", cc->core_id);
3866         return;
3867     }
3868 
3869     if (core_slot->cpu) {
3870         error_setg(errp, "core %d already populated", cc->core_id);
3871         return;
3872     }
3873 
3874     numa_cpu_pre_plug(core_slot, dev, errp);
3875 }
3876 
3877 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3878                           void *fdt, int *fdt_start_offset, Error **errp)
3879 {
3880     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3881     int intc_phandle;
3882 
3883     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3884     if (intc_phandle <= 0) {
3885         return -1;
3886     }
3887 
3888     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3889         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3890         return -1;
3891     }
3892 
3893     /* generally SLOF creates these, for hotplug it's up to QEMU */
3894     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3895 
3896     return 0;
3897 }
3898 
3899 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3900                                Error **errp)
3901 {
3902     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3903     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3904     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3905     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3906 
3907     if (dev->hotplugged && !smc->dr_phb_enabled) {
3908         error_setg(errp, "PHB hotplug not supported for this machine");
3909         return;
3910     }
3911 
3912     if (sphb->index == (uint32_t)-1) {
3913         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3914         return;
3915     }
3916 
3917     /*
3918      * This will check that sphb->index doesn't exceed the maximum number of
3919      * PHBs for the current machine type.
3920      */
3921     smc->phb_placement(spapr, sphb->index,
3922                        &sphb->buid, &sphb->io_win_addr,
3923                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3924                        windows_supported, sphb->dma_liobn,
3925                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3926                        errp);
3927 }
3928 
3929 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3930                            Error **errp)
3931 {
3932     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3933     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3934     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3935     SpaprDrc *drc;
3936     bool hotplugged = spapr_drc_hotplugged(dev);
3937 
3938     if (!smc->dr_phb_enabled) {
3939         return;
3940     }
3941 
3942     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3943     /* hotplug hooks should check it's enabled before getting this far */
3944     assert(drc);
3945 
3946     if (!spapr_drc_attach(drc, dev, errp)) {
3947         return;
3948     }
3949 
3950     if (hotplugged) {
3951         spapr_hotplug_req_add_by_index(drc);
3952     } else {
3953         spapr_drc_reset(drc);
3954     }
3955 }
3956 
3957 void spapr_phb_release(DeviceState *dev)
3958 {
3959     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3960 
3961     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3962     object_unparent(OBJECT(dev));
3963 }
3964 
3965 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3966 {
3967     qdev_unrealize(dev);
3968 }
3969 
3970 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3971                                      DeviceState *dev, Error **errp)
3972 {
3973     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3974     SpaprDrc *drc;
3975 
3976     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3977     assert(drc);
3978 
3979     if (!spapr_drc_unplug_requested(drc)) {
3980         spapr_drc_detach(drc);
3981         spapr_hotplug_req_remove_by_index(drc);
3982     }
3983 }
3984 
3985 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3986                                  Error **errp)
3987 {
3988     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3989     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
3990 
3991     if (spapr->tpm_proxy != NULL) {
3992         error_setg(errp, "Only one TPM proxy can be specified for this machine");
3993         return;
3994     }
3995 
3996     spapr->tpm_proxy = tpm_proxy;
3997 }
3998 
3999 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4000 {
4001     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4002 
4003     qdev_unrealize(dev);
4004     object_unparent(OBJECT(dev));
4005     spapr->tpm_proxy = NULL;
4006 }
4007 
4008 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4009                                       DeviceState *dev, Error **errp)
4010 {
4011     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4012         spapr_memory_plug(hotplug_dev, dev, errp);
4013     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4014         spapr_core_plug(hotplug_dev, dev, errp);
4015     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4016         spapr_phb_plug(hotplug_dev, dev, errp);
4017     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4018         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4019     }
4020 }
4021 
4022 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4023                                         DeviceState *dev, Error **errp)
4024 {
4025     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4026         spapr_memory_unplug(hotplug_dev, dev);
4027     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4028         spapr_core_unplug(hotplug_dev, dev);
4029     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4030         spapr_phb_unplug(hotplug_dev, dev);
4031     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4032         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4033     }
4034 }
4035 
4036 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4037                                                 DeviceState *dev, Error **errp)
4038 {
4039     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4040     MachineClass *mc = MACHINE_GET_CLASS(sms);
4041     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4042 
4043     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4044         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4045             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4046         } else {
4047             /* NOTE: this means there is a window after guest reset, prior to
4048              * CAS negotiation, where unplug requests will fail due to the
4049              * capability not being detected yet. This is a bit different than
4050              * the case with PCI unplug, where the events will be queued and
4051              * eventually handled by the guest after boot
4052              */
4053             error_setg(errp, "Memory hot unplug not supported for this guest");
4054         }
4055     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4056         if (!mc->has_hotpluggable_cpus) {
4057             error_setg(errp, "CPU hot unplug not supported on this machine");
4058             return;
4059         }
4060         spapr_core_unplug_request(hotplug_dev, dev, errp);
4061     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4062         if (!smc->dr_phb_enabled) {
4063             error_setg(errp, "PHB hot unplug not supported on this machine");
4064             return;
4065         }
4066         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4067     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4068         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4069     }
4070 }
4071 
4072 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4073                                           DeviceState *dev, Error **errp)
4074 {
4075     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4076         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4077     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4078         spapr_core_pre_plug(hotplug_dev, dev, errp);
4079     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4080         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4081     }
4082 }
4083 
4084 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4085                                                  DeviceState *dev)
4086 {
4087     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4088         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4089         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4090         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4091         return HOTPLUG_HANDLER(machine);
4092     }
4093     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4094         PCIDevice *pcidev = PCI_DEVICE(dev);
4095         PCIBus *root = pci_device_root_bus(pcidev);
4096         SpaprPhbState *phb =
4097             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4098                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4099 
4100         if (phb) {
4101             return HOTPLUG_HANDLER(phb);
4102         }
4103     }
4104     return NULL;
4105 }
4106 
4107 static CpuInstanceProperties
4108 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4109 {
4110     CPUArchId *core_slot;
4111     MachineClass *mc = MACHINE_GET_CLASS(machine);
4112 
4113     /* make sure possible_cpu are intialized */
4114     mc->possible_cpu_arch_ids(machine);
4115     /* get CPU core slot containing thread that matches cpu_index */
4116     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4117     assert(core_slot);
4118     return core_slot->props;
4119 }
4120 
4121 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4122 {
4123     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4124 }
4125 
4126 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4127 {
4128     int i;
4129     unsigned int smp_threads = machine->smp.threads;
4130     unsigned int smp_cpus = machine->smp.cpus;
4131     const char *core_type;
4132     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4133     MachineClass *mc = MACHINE_GET_CLASS(machine);
4134 
4135     if (!mc->has_hotpluggable_cpus) {
4136         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4137     }
4138     if (machine->possible_cpus) {
4139         assert(machine->possible_cpus->len == spapr_max_cores);
4140         return machine->possible_cpus;
4141     }
4142 
4143     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4144     if (!core_type) {
4145         error_report("Unable to find sPAPR CPU Core definition");
4146         exit(1);
4147     }
4148 
4149     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4150                              sizeof(CPUArchId) * spapr_max_cores);
4151     machine->possible_cpus->len = spapr_max_cores;
4152     for (i = 0; i < machine->possible_cpus->len; i++) {
4153         int core_id = i * smp_threads;
4154 
4155         machine->possible_cpus->cpus[i].type = core_type;
4156         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4157         machine->possible_cpus->cpus[i].arch_id = core_id;
4158         machine->possible_cpus->cpus[i].props.has_core_id = true;
4159         machine->possible_cpus->cpus[i].props.core_id = core_id;
4160     }
4161     return machine->possible_cpus;
4162 }
4163 
4164 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4165                                 uint64_t *buid, hwaddr *pio,
4166                                 hwaddr *mmio32, hwaddr *mmio64,
4167                                 unsigned n_dma, uint32_t *liobns,
4168                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4169 {
4170     /*
4171      * New-style PHB window placement.
4172      *
4173      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4174      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4175      * windows.
4176      *
4177      * Some guest kernels can't work with MMIO windows above 1<<46
4178      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4179      *
4180      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4181      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4182      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4183      * 1TiB 64-bit MMIO windows for each PHB.
4184      */
4185     const uint64_t base_buid = 0x800000020000000ULL;
4186     int i;
4187 
4188     /* Sanity check natural alignments */
4189     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4190     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4191     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4192     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4193     /* Sanity check bounds */
4194     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4195                       SPAPR_PCI_MEM32_WIN_SIZE);
4196     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4197                       SPAPR_PCI_MEM64_WIN_SIZE);
4198 
4199     if (index >= SPAPR_MAX_PHBS) {
4200         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4201                    SPAPR_MAX_PHBS - 1);
4202         return;
4203     }
4204 
4205     *buid = base_buid + index;
4206     for (i = 0; i < n_dma; ++i) {
4207         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4208     }
4209 
4210     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4211     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4212     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4213 
4214     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4215     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4216 }
4217 
4218 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4219 {
4220     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4221 
4222     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4223 }
4224 
4225 static void spapr_ics_resend(XICSFabric *dev)
4226 {
4227     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4228 
4229     ics_resend(spapr->ics);
4230 }
4231 
4232 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4233 {
4234     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4235 
4236     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4237 }
4238 
4239 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4240                                  Monitor *mon)
4241 {
4242     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4243 
4244     spapr_irq_print_info(spapr, mon);
4245     monitor_printf(mon, "irqchip: %s\n",
4246                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4247 }
4248 
4249 /*
4250  * This is a XIVE only operation
4251  */
4252 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4253                            uint8_t nvt_blk, uint32_t nvt_idx,
4254                            bool cam_ignore, uint8_t priority,
4255                            uint32_t logic_serv, XiveTCTXMatch *match)
4256 {
4257     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4258     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4259     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4260     int count;
4261 
4262     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4263                            priority, logic_serv, match);
4264     if (count < 0) {
4265         return count;
4266     }
4267 
4268     /*
4269      * When we implement the save and restore of the thread interrupt
4270      * contexts in the enter/exit CPU handlers of the machine and the
4271      * escalations in QEMU, we should be able to handle non dispatched
4272      * vCPUs.
4273      *
4274      * Until this is done, the sPAPR machine should find at least one
4275      * matching context always.
4276      */
4277     if (count == 0) {
4278         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4279                       nvt_blk, nvt_idx);
4280     }
4281 
4282     return count;
4283 }
4284 
4285 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4286 {
4287     return cpu->vcpu_id;
4288 }
4289 
4290 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4291 {
4292     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4293     MachineState *ms = MACHINE(spapr);
4294     int vcpu_id;
4295 
4296     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4297 
4298     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4299         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4300         error_append_hint(errp, "Adjust the number of cpus to %d "
4301                           "or try to raise the number of threads per core\n",
4302                           vcpu_id * ms->smp.threads / spapr->vsmt);
4303         return false;
4304     }
4305 
4306     cpu->vcpu_id = vcpu_id;
4307     return true;
4308 }
4309 
4310 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4311 {
4312     CPUState *cs;
4313 
4314     CPU_FOREACH(cs) {
4315         PowerPCCPU *cpu = POWERPC_CPU(cs);
4316 
4317         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4318             return cpu;
4319         }
4320     }
4321 
4322     return NULL;
4323 }
4324 
4325 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4326 {
4327     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4328 
4329     /* These are only called by TCG, KVM maintains dispatch state */
4330 
4331     spapr_cpu->prod = false;
4332     if (spapr_cpu->vpa_addr) {
4333         CPUState *cs = CPU(cpu);
4334         uint32_t dispatch;
4335 
4336         dispatch = ldl_be_phys(cs->as,
4337                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4338         dispatch++;
4339         if ((dispatch & 1) != 0) {
4340             qemu_log_mask(LOG_GUEST_ERROR,
4341                           "VPA: incorrect dispatch counter value for "
4342                           "dispatched partition %u, correcting.\n", dispatch);
4343             dispatch++;
4344         }
4345         stl_be_phys(cs->as,
4346                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4347     }
4348 }
4349 
4350 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4351 {
4352     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4353 
4354     if (spapr_cpu->vpa_addr) {
4355         CPUState *cs = CPU(cpu);
4356         uint32_t dispatch;
4357 
4358         dispatch = ldl_be_phys(cs->as,
4359                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4360         dispatch++;
4361         if ((dispatch & 1) != 1) {
4362             qemu_log_mask(LOG_GUEST_ERROR,
4363                           "VPA: incorrect dispatch counter value for "
4364                           "preempted partition %u, correcting.\n", dispatch);
4365             dispatch++;
4366         }
4367         stl_be_phys(cs->as,
4368                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4369     }
4370 }
4371 
4372 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4373 {
4374     MachineClass *mc = MACHINE_CLASS(oc);
4375     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4376     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4377     NMIClass *nc = NMI_CLASS(oc);
4378     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4379     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4380     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4381     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4382     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4383 
4384     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4385     mc->ignore_boot_device_suffixes = true;
4386 
4387     /*
4388      * We set up the default / latest behaviour here.  The class_init
4389      * functions for the specific versioned machine types can override
4390      * these details for backwards compatibility
4391      */
4392     mc->init = spapr_machine_init;
4393     mc->reset = spapr_machine_reset;
4394     mc->block_default_type = IF_SCSI;
4395     mc->max_cpus = 1024;
4396     mc->no_parallel = 1;
4397     mc->default_boot_order = "";
4398     mc->default_ram_size = 512 * MiB;
4399     mc->default_ram_id = "ppc_spapr.ram";
4400     mc->default_display = "std";
4401     mc->kvm_type = spapr_kvm_type;
4402     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4403     mc->pci_allow_0_address = true;
4404     assert(!mc->get_hotplug_handler);
4405     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4406     hc->pre_plug = spapr_machine_device_pre_plug;
4407     hc->plug = spapr_machine_device_plug;
4408     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4409     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4410     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4411     hc->unplug_request = spapr_machine_device_unplug_request;
4412     hc->unplug = spapr_machine_device_unplug;
4413 
4414     smc->dr_lmb_enabled = true;
4415     smc->update_dt_enabled = true;
4416     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4417     mc->has_hotpluggable_cpus = true;
4418     mc->nvdimm_supported = true;
4419     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4420     fwc->get_dev_path = spapr_get_fw_dev_path;
4421     nc->nmi_monitor_handler = spapr_nmi;
4422     smc->phb_placement = spapr_phb_placement;
4423     vhc->hypercall = emulate_spapr_hypercall;
4424     vhc->hpt_mask = spapr_hpt_mask;
4425     vhc->map_hptes = spapr_map_hptes;
4426     vhc->unmap_hptes = spapr_unmap_hptes;
4427     vhc->hpte_set_c = spapr_hpte_set_c;
4428     vhc->hpte_set_r = spapr_hpte_set_r;
4429     vhc->get_pate = spapr_get_pate;
4430     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4431     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4432     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4433     xic->ics_get = spapr_ics_get;
4434     xic->ics_resend = spapr_ics_resend;
4435     xic->icp_get = spapr_icp_get;
4436     ispc->print_info = spapr_pic_print_info;
4437     /* Force NUMA node memory size to be a multiple of
4438      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4439      * in which LMBs are represented and hot-added
4440      */
4441     mc->numa_mem_align_shift = 28;
4442     mc->auto_enable_numa = true;
4443 
4444     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4445     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4446     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4447     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4448     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4449     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4450     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4451     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4452     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4453     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4454     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4455     spapr_caps_add_properties(smc);
4456     smc->irq = &spapr_irq_dual;
4457     smc->dr_phb_enabled = true;
4458     smc->linux_pci_probe = true;
4459     smc->smp_threads_vsmt = true;
4460     smc->nr_xirqs = SPAPR_NR_XIRQS;
4461     xfc->match_nvt = spapr_match_nvt;
4462 }
4463 
4464 static const TypeInfo spapr_machine_info = {
4465     .name          = TYPE_SPAPR_MACHINE,
4466     .parent        = TYPE_MACHINE,
4467     .abstract      = true,
4468     .instance_size = sizeof(SpaprMachineState),
4469     .instance_init = spapr_instance_init,
4470     .instance_finalize = spapr_machine_finalizefn,
4471     .class_size    = sizeof(SpaprMachineClass),
4472     .class_init    = spapr_machine_class_init,
4473     .interfaces = (InterfaceInfo[]) {
4474         { TYPE_FW_PATH_PROVIDER },
4475         { TYPE_NMI },
4476         { TYPE_HOTPLUG_HANDLER },
4477         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4478         { TYPE_XICS_FABRIC },
4479         { TYPE_INTERRUPT_STATS_PROVIDER },
4480         { TYPE_XIVE_FABRIC },
4481         { }
4482     },
4483 };
4484 
4485 static void spapr_machine_latest_class_options(MachineClass *mc)
4486 {
4487     mc->alias = "pseries";
4488     mc->is_default = true;
4489 }
4490 
4491 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4492     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4493                                                     void *data)      \
4494     {                                                                \
4495         MachineClass *mc = MACHINE_CLASS(oc);                        \
4496         spapr_machine_##suffix##_class_options(mc);                  \
4497         if (latest) {                                                \
4498             spapr_machine_latest_class_options(mc);                  \
4499         }                                                            \
4500     }                                                                \
4501     static const TypeInfo spapr_machine_##suffix##_info = {          \
4502         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4503         .parent = TYPE_SPAPR_MACHINE,                                \
4504         .class_init = spapr_machine_##suffix##_class_init,           \
4505     };                                                               \
4506     static void spapr_machine_register_##suffix(void)                \
4507     {                                                                \
4508         type_register(&spapr_machine_##suffix##_info);               \
4509     }                                                                \
4510     type_init(spapr_machine_register_##suffix)
4511 
4512 /*
4513  * pseries-5.2
4514  */
4515 static void spapr_machine_5_2_class_options(MachineClass *mc)
4516 {
4517     /* Defaults for the latest behaviour inherited from the base class */
4518 }
4519 
4520 DEFINE_SPAPR_MACHINE(5_2, "5.2", true);
4521 
4522 /*
4523  * pseries-5.1
4524  */
4525 static void spapr_machine_5_1_class_options(MachineClass *mc)
4526 {
4527     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4528 
4529     spapr_machine_5_2_class_options(mc);
4530     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4531     smc->pre_5_2_numa_associativity = true;
4532 }
4533 
4534 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4535 
4536 /*
4537  * pseries-5.0
4538  */
4539 static void spapr_machine_5_0_class_options(MachineClass *mc)
4540 {
4541     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4542     static GlobalProperty compat[] = {
4543         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4544     };
4545 
4546     spapr_machine_5_1_class_options(mc);
4547     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4548     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4549     mc->numa_mem_supported = true;
4550     smc->pre_5_1_assoc_refpoints = true;
4551 }
4552 
4553 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4554 
4555 /*
4556  * pseries-4.2
4557  */
4558 static void spapr_machine_4_2_class_options(MachineClass *mc)
4559 {
4560     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4561 
4562     spapr_machine_5_0_class_options(mc);
4563     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4564     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4565     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4566     smc->rma_limit = 16 * GiB;
4567     mc->nvdimm_supported = false;
4568 }
4569 
4570 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4571 
4572 /*
4573  * pseries-4.1
4574  */
4575 static void spapr_machine_4_1_class_options(MachineClass *mc)
4576 {
4577     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4578     static GlobalProperty compat[] = {
4579         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4580         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4581     };
4582 
4583     spapr_machine_4_2_class_options(mc);
4584     smc->linux_pci_probe = false;
4585     smc->smp_threads_vsmt = false;
4586     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4587     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4588 }
4589 
4590 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4591 
4592 /*
4593  * pseries-4.0
4594  */
4595 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4596                               uint64_t *buid, hwaddr *pio,
4597                               hwaddr *mmio32, hwaddr *mmio64,
4598                               unsigned n_dma, uint32_t *liobns,
4599                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4600 {
4601     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4602                         nv2gpa, nv2atsd, errp);
4603     *nv2gpa = 0;
4604     *nv2atsd = 0;
4605 }
4606 
4607 static void spapr_machine_4_0_class_options(MachineClass *mc)
4608 {
4609     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4610 
4611     spapr_machine_4_1_class_options(mc);
4612     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4613     smc->phb_placement = phb_placement_4_0;
4614     smc->irq = &spapr_irq_xics;
4615     smc->pre_4_1_migration = true;
4616 }
4617 
4618 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4619 
4620 /*
4621  * pseries-3.1
4622  */
4623 static void spapr_machine_3_1_class_options(MachineClass *mc)
4624 {
4625     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4626 
4627     spapr_machine_4_0_class_options(mc);
4628     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4629 
4630     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4631     smc->update_dt_enabled = false;
4632     smc->dr_phb_enabled = false;
4633     smc->broken_host_serial_model = true;
4634     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4635     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4636     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4637     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4638 }
4639 
4640 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4641 
4642 /*
4643  * pseries-3.0
4644  */
4645 
4646 static void spapr_machine_3_0_class_options(MachineClass *mc)
4647 {
4648     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4649 
4650     spapr_machine_3_1_class_options(mc);
4651     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4652 
4653     smc->legacy_irq_allocation = true;
4654     smc->nr_xirqs = 0x400;
4655     smc->irq = &spapr_irq_xics_legacy;
4656 }
4657 
4658 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4659 
4660 /*
4661  * pseries-2.12
4662  */
4663 static void spapr_machine_2_12_class_options(MachineClass *mc)
4664 {
4665     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4666     static GlobalProperty compat[] = {
4667         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4668         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4669     };
4670 
4671     spapr_machine_3_0_class_options(mc);
4672     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4673     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4674 
4675     /* We depend on kvm_enabled() to choose a default value for the
4676      * hpt-max-page-size capability. Of course we can't do it here
4677      * because this is too early and the HW accelerator isn't initialzed
4678      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4679      */
4680     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4681 }
4682 
4683 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4684 
4685 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4686 {
4687     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4688 
4689     spapr_machine_2_12_class_options(mc);
4690     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4691     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4692     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4693 }
4694 
4695 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4696 
4697 /*
4698  * pseries-2.11
4699  */
4700 
4701 static void spapr_machine_2_11_class_options(MachineClass *mc)
4702 {
4703     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4704 
4705     spapr_machine_2_12_class_options(mc);
4706     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4707     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4708 }
4709 
4710 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4711 
4712 /*
4713  * pseries-2.10
4714  */
4715 
4716 static void spapr_machine_2_10_class_options(MachineClass *mc)
4717 {
4718     spapr_machine_2_11_class_options(mc);
4719     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4720 }
4721 
4722 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4723 
4724 /*
4725  * pseries-2.9
4726  */
4727 
4728 static void spapr_machine_2_9_class_options(MachineClass *mc)
4729 {
4730     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4731     static GlobalProperty compat[] = {
4732         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4733     };
4734 
4735     spapr_machine_2_10_class_options(mc);
4736     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4737     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4738     smc->pre_2_10_has_unused_icps = true;
4739     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4740 }
4741 
4742 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4743 
4744 /*
4745  * pseries-2.8
4746  */
4747 
4748 static void spapr_machine_2_8_class_options(MachineClass *mc)
4749 {
4750     static GlobalProperty compat[] = {
4751         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4752     };
4753 
4754     spapr_machine_2_9_class_options(mc);
4755     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4756     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4757     mc->numa_mem_align_shift = 23;
4758 }
4759 
4760 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4761 
4762 /*
4763  * pseries-2.7
4764  */
4765 
4766 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4767                               uint64_t *buid, hwaddr *pio,
4768                               hwaddr *mmio32, hwaddr *mmio64,
4769                               unsigned n_dma, uint32_t *liobns,
4770                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4771 {
4772     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4773     const uint64_t base_buid = 0x800000020000000ULL;
4774     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4775     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4776     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4777     const uint32_t max_index = 255;
4778     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4779 
4780     uint64_t ram_top = MACHINE(spapr)->ram_size;
4781     hwaddr phb0_base, phb_base;
4782     int i;
4783 
4784     /* Do we have device memory? */
4785     if (MACHINE(spapr)->maxram_size > ram_top) {
4786         /* Can't just use maxram_size, because there may be an
4787          * alignment gap between normal and device memory regions
4788          */
4789         ram_top = MACHINE(spapr)->device_memory->base +
4790             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4791     }
4792 
4793     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4794 
4795     if (index > max_index) {
4796         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4797                    max_index);
4798         return;
4799     }
4800 
4801     *buid = base_buid + index;
4802     for (i = 0; i < n_dma; ++i) {
4803         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4804     }
4805 
4806     phb_base = phb0_base + index * phb_spacing;
4807     *pio = phb_base + pio_offset;
4808     *mmio32 = phb_base + mmio_offset;
4809     /*
4810      * We don't set the 64-bit MMIO window, relying on the PHB's
4811      * fallback behaviour of automatically splitting a large "32-bit"
4812      * window into contiguous 32-bit and 64-bit windows
4813      */
4814 
4815     *nv2gpa = 0;
4816     *nv2atsd = 0;
4817 }
4818 
4819 static void spapr_machine_2_7_class_options(MachineClass *mc)
4820 {
4821     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4822     static GlobalProperty compat[] = {
4823         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4824         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4825         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4826         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4827     };
4828 
4829     spapr_machine_2_8_class_options(mc);
4830     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4831     mc->default_machine_opts = "modern-hotplug-events=off";
4832     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4833     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4834     smc->phb_placement = phb_placement_2_7;
4835 }
4836 
4837 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4838 
4839 /*
4840  * pseries-2.6
4841  */
4842 
4843 static void spapr_machine_2_6_class_options(MachineClass *mc)
4844 {
4845     static GlobalProperty compat[] = {
4846         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4847     };
4848 
4849     spapr_machine_2_7_class_options(mc);
4850     mc->has_hotpluggable_cpus = false;
4851     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4852     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4853 }
4854 
4855 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4856 
4857 /*
4858  * pseries-2.5
4859  */
4860 
4861 static void spapr_machine_2_5_class_options(MachineClass *mc)
4862 {
4863     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4864     static GlobalProperty compat[] = {
4865         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4866     };
4867 
4868     spapr_machine_2_6_class_options(mc);
4869     smc->use_ohci_by_default = true;
4870     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4871     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4872 }
4873 
4874 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4875 
4876 /*
4877  * pseries-2.4
4878  */
4879 
4880 static void spapr_machine_2_4_class_options(MachineClass *mc)
4881 {
4882     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4883 
4884     spapr_machine_2_5_class_options(mc);
4885     smc->dr_lmb_enabled = false;
4886     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4887 }
4888 
4889 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4890 
4891 /*
4892  * pseries-2.3
4893  */
4894 
4895 static void spapr_machine_2_3_class_options(MachineClass *mc)
4896 {
4897     static GlobalProperty compat[] = {
4898         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4899     };
4900     spapr_machine_2_4_class_options(mc);
4901     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4902     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4903 }
4904 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4905 
4906 /*
4907  * pseries-2.2
4908  */
4909 
4910 static void spapr_machine_2_2_class_options(MachineClass *mc)
4911 {
4912     static GlobalProperty compat[] = {
4913         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4914     };
4915 
4916     spapr_machine_2_3_class_options(mc);
4917     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4918     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4919     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4920 }
4921 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4922 
4923 /*
4924  * pseries-2.1
4925  */
4926 
4927 static void spapr_machine_2_1_class_options(MachineClass *mc)
4928 {
4929     spapr_machine_2_2_class_options(mc);
4930     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4931 }
4932 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4933 
4934 static void spapr_machine_register_types(void)
4935 {
4936     type_register_static(&spapr_machine_info);
4937 }
4938 
4939 type_init(spapr_machine_register_types)
4940