xref: /openbmc/qemu/hw/ppc/spapr.c (revision 64547a3b)
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 
85 #include "monitor/monitor.h"
86 
87 #include <libfdt.h>
88 
89 /* SLOF memory layout:
90  *
91  * SLOF raw image loaded at 0, copies its romfs right below the flat
92  * device-tree, then position SLOF itself 31M below that
93  *
94  * So we set FW_OVERHEAD to 40MB which should account for all of that
95  * and more
96  *
97  * We load our kernel at 4M, leaving space for SLOF initial image
98  */
99 #define FDT_MAX_SIZE            0x100000
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 int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
206 {
207     int index = spapr_get_vcpu_id(cpu);
208     uint32_t associativity[] = {cpu_to_be32(0x5),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(0x0),
212                                 cpu_to_be32(cpu->node_id),
213                                 cpu_to_be32(index)};
214 
215     /* Advertise NUMA via ibm,associativity */
216     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217                           sizeof(associativity));
218 }
219 
220 static void spapr_dt_pa_features(SpaprMachineState *spapr,
221                                  PowerPCCPU *cpu,
222                                  void *fdt, int offset)
223 {
224     uint8_t pa_features_206[] = { 6, 0,
225         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
226     uint8_t pa_features_207[] = { 24, 0,
227         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
228         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
229         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
230         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
231     uint8_t pa_features_300[] = { 66, 0,
232         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
233         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
234         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
235         /* 6: DS207 */
236         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
237         /* 16: Vector */
238         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
239         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
241         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
242         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
243         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
244         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
245         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
246         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
247         /* 42: PM, 44: PC RA, 46: SC vec'd */
248         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
249         /* 48: SIMD, 50: QP BFP, 52: String */
250         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
251         /* 54: DecFP, 56: DecI, 58: SHA */
252         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
253         /* 60: NM atomic, 62: RNG */
254         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
255     };
256     uint8_t *pa_features = NULL;
257     size_t pa_size;
258 
259     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
260         pa_features = pa_features_206;
261         pa_size = sizeof(pa_features_206);
262     }
263     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
264         pa_features = pa_features_207;
265         pa_size = sizeof(pa_features_207);
266     }
267     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
268         pa_features = pa_features_300;
269         pa_size = sizeof(pa_features_300);
270     }
271     if (!pa_features) {
272         return;
273     }
274 
275     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
276         /*
277          * Note: we keep CI large pages off by default because a 64K capable
278          * guest provisioned with large pages might otherwise try to map a qemu
279          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
280          * even if that qemu runs on a 4k host.
281          * We dd this bit back here if we are confident this is not an issue
282          */
283         pa_features[3] |= 0x20;
284     }
285     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
286         pa_features[24] |= 0x80;    /* Transactional memory support */
287     }
288     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
289         /* Workaround for broken kernels that attempt (guest) radix
290          * mode when they can't handle it, if they see the radix bit set
291          * in pa-features. So hide it from them. */
292         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
293     }
294 
295     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
296 }
297 
298 static hwaddr spapr_node0_size(MachineState *machine)
299 {
300     if (machine->numa_state->num_nodes) {
301         int i;
302         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
303             if (machine->numa_state->nodes[i].node_mem) {
304                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
305                            machine->ram_size);
306             }
307         }
308     }
309     return machine->ram_size;
310 }
311 
312 static void add_str(GString *s, const gchar *s1)
313 {
314     g_string_append_len(s, s1, strlen(s1) + 1);
315 }
316 
317 static int spapr_dt_memory_node(void *fdt, int nodeid, hwaddr start,
318                                 hwaddr size)
319 {
320     uint32_t associativity[] = {
321         cpu_to_be32(0x4), /* length */
322         cpu_to_be32(0x0), cpu_to_be32(0x0),
323         cpu_to_be32(0x0), cpu_to_be32(nodeid)
324     };
325     char mem_name[32];
326     uint64_t mem_reg_property[2];
327     int off;
328 
329     mem_reg_property[0] = cpu_to_be64(start);
330     mem_reg_property[1] = cpu_to_be64(size);
331 
332     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
333     off = fdt_add_subnode(fdt, 0, mem_name);
334     _FDT(off);
335     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
336     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
337                       sizeof(mem_reg_property))));
338     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
339                       sizeof(associativity))));
340     return off;
341 }
342 
343 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
344 {
345     MemoryDeviceInfoList *info;
346 
347     for (info = list; info; info = info->next) {
348         MemoryDeviceInfo *value = info->value;
349 
350         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
351             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
352 
353             if (addr >= pcdimm_info->addr &&
354                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
355                 return pcdimm_info->node;
356             }
357         }
358     }
359 
360     return -1;
361 }
362 
363 struct sPAPRDrconfCellV2 {
364      uint32_t seq_lmbs;
365      uint64_t base_addr;
366      uint32_t drc_index;
367      uint32_t aa_index;
368      uint32_t flags;
369 } QEMU_PACKED;
370 
371 typedef struct DrconfCellQueue {
372     struct sPAPRDrconfCellV2 cell;
373     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
374 } DrconfCellQueue;
375 
376 static DrconfCellQueue *
377 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
378                       uint32_t drc_index, uint32_t aa_index,
379                       uint32_t flags)
380 {
381     DrconfCellQueue *elem;
382 
383     elem = g_malloc0(sizeof(*elem));
384     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
385     elem->cell.base_addr = cpu_to_be64(base_addr);
386     elem->cell.drc_index = cpu_to_be32(drc_index);
387     elem->cell.aa_index = cpu_to_be32(aa_index);
388     elem->cell.flags = cpu_to_be32(flags);
389 
390     return elem;
391 }
392 
393 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
394                                       int offset, MemoryDeviceInfoList *dimms)
395 {
396     MachineState *machine = MACHINE(spapr);
397     uint8_t *int_buf, *cur_index;
398     int ret;
399     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
400     uint64_t addr, cur_addr, size;
401     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
402     uint64_t mem_end = machine->device_memory->base +
403                        memory_region_size(&machine->device_memory->mr);
404     uint32_t node, buf_len, nr_entries = 0;
405     SpaprDrc *drc;
406     DrconfCellQueue *elem, *next;
407     MemoryDeviceInfoList *info;
408     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
409         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
410 
411     /* Entry to cover RAM and the gap area */
412     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
413                                  SPAPR_LMB_FLAGS_RESERVED |
414                                  SPAPR_LMB_FLAGS_DRC_INVALID);
415     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
416     nr_entries++;
417 
418     cur_addr = machine->device_memory->base;
419     for (info = dimms; info; info = info->next) {
420         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
421 
422         addr = di->addr;
423         size = di->size;
424         node = di->node;
425 
426         /*
427          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
428          * area is marked hotpluggable in the next iteration for the bigger
429          * chunk including the NVDIMM occupied area.
430          */
431         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
432             continue;
433 
434         /* Entry for hot-pluggable area */
435         if (cur_addr < addr) {
436             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
437             g_assert(drc);
438             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
439                                          cur_addr, spapr_drc_index(drc), -1, 0);
440             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
441             nr_entries++;
442         }
443 
444         /* Entry for DIMM */
445         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
446         g_assert(drc);
447         elem = spapr_get_drconf_cell(size / lmb_size, addr,
448                                      spapr_drc_index(drc), node,
449                                      SPAPR_LMB_FLAGS_ASSIGNED);
450         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
451         nr_entries++;
452         cur_addr = addr + size;
453     }
454 
455     /* Entry for remaining hotpluggable area */
456     if (cur_addr < mem_end) {
457         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
458         g_assert(drc);
459         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
460                                      cur_addr, spapr_drc_index(drc), -1, 0);
461         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
462         nr_entries++;
463     }
464 
465     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
466     int_buf = cur_index = g_malloc0(buf_len);
467     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
468     cur_index += sizeof(nr_entries);
469 
470     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
471         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
472         cur_index += sizeof(elem->cell);
473         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
474         g_free(elem);
475     }
476 
477     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
478     g_free(int_buf);
479     if (ret < 0) {
480         return -1;
481     }
482     return 0;
483 }
484 
485 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
486                                    int offset, MemoryDeviceInfoList *dimms)
487 {
488     MachineState *machine = MACHINE(spapr);
489     int i, ret;
490     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
491     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
492     uint32_t nr_lmbs = (machine->device_memory->base +
493                        memory_region_size(&machine->device_memory->mr)) /
494                        lmb_size;
495     uint32_t *int_buf, *cur_index, buf_len;
496 
497     /*
498      * Allocate enough buffer size to fit in ibm,dynamic-memory
499      */
500     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
501     cur_index = int_buf = g_malloc0(buf_len);
502     int_buf[0] = cpu_to_be32(nr_lmbs);
503     cur_index++;
504     for (i = 0; i < nr_lmbs; i++) {
505         uint64_t addr = i * lmb_size;
506         uint32_t *dynamic_memory = cur_index;
507 
508         if (i >= device_lmb_start) {
509             SpaprDrc *drc;
510 
511             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
512             g_assert(drc);
513 
514             dynamic_memory[0] = cpu_to_be32(addr >> 32);
515             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
516             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
517             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
518             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
519             if (memory_region_present(get_system_memory(), addr)) {
520                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
521             } else {
522                 dynamic_memory[5] = cpu_to_be32(0);
523             }
524         } else {
525             /*
526              * LMB information for RMA, boot time RAM and gap b/n RAM and
527              * device memory region -- all these are marked as reserved
528              * and as having no valid DRC.
529              */
530             dynamic_memory[0] = cpu_to_be32(addr >> 32);
531             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
532             dynamic_memory[2] = cpu_to_be32(0);
533             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
534             dynamic_memory[4] = cpu_to_be32(-1);
535             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
536                                             SPAPR_LMB_FLAGS_DRC_INVALID);
537         }
538 
539         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
540     }
541     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
542     g_free(int_buf);
543     if (ret < 0) {
544         return -1;
545     }
546     return 0;
547 }
548 
549 /*
550  * Adds ibm,dynamic-reconfiguration-memory node.
551  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
552  * of this device tree node.
553  */
554 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
555                                                    void *fdt)
556 {
557     MachineState *machine = MACHINE(spapr);
558     int nb_numa_nodes = machine->numa_state->num_nodes;
559     int ret, i, offset;
560     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
561     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
562     uint32_t *int_buf, *cur_index, buf_len;
563     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
564     MemoryDeviceInfoList *dimms = NULL;
565 
566     /*
567      * Don't create the node if there is no device memory
568      */
569     if (machine->ram_size == machine->maxram_size) {
570         return 0;
571     }
572 
573     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
574 
575     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
576                     sizeof(prop_lmb_size));
577     if (ret < 0) {
578         return ret;
579     }
580 
581     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
582     if (ret < 0) {
583         return ret;
584     }
585 
586     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
587     if (ret < 0) {
588         return ret;
589     }
590 
591     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
592     dimms = qmp_memory_device_list();
593     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
594         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
595     } else {
596         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
597     }
598     qapi_free_MemoryDeviceInfoList(dimms);
599 
600     if (ret < 0) {
601         return ret;
602     }
603 
604     /* ibm,associativity-lookup-arrays */
605     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
606     cur_index = int_buf = g_malloc0(buf_len);
607     int_buf[0] = cpu_to_be32(nr_nodes);
608     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
609     cur_index += 2;
610     for (i = 0; i < nr_nodes; i++) {
611         uint32_t associativity[] = {
612             cpu_to_be32(0x0),
613             cpu_to_be32(0x0),
614             cpu_to_be32(0x0),
615             cpu_to_be32(i)
616         };
617         memcpy(cur_index, associativity, sizeof(associativity));
618         cur_index += 4;
619     }
620     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
621             (cur_index - int_buf) * sizeof(uint32_t));
622     g_free(int_buf);
623 
624     return ret;
625 }
626 
627 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
628 {
629     MachineState *machine = MACHINE(spapr);
630     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
631     hwaddr mem_start, node_size;
632     int i, nb_nodes = machine->numa_state->num_nodes;
633     NodeInfo *nodes = machine->numa_state->nodes;
634 
635     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
636         if (!nodes[i].node_mem) {
637             continue;
638         }
639         if (mem_start >= machine->ram_size) {
640             node_size = 0;
641         } else {
642             node_size = nodes[i].node_mem;
643             if (node_size > machine->ram_size - mem_start) {
644                 node_size = machine->ram_size - mem_start;
645             }
646         }
647         if (!mem_start) {
648             /* spapr_machine_init() checks for rma_size <= node0_size
649              * already */
650             spapr_dt_memory_node(fdt, i, 0, spapr->rma_size);
651             mem_start += spapr->rma_size;
652             node_size -= spapr->rma_size;
653         }
654         for ( ; node_size; ) {
655             hwaddr sizetmp = pow2floor(node_size);
656 
657             /* mem_start != 0 here */
658             if (ctzl(mem_start) < ctzl(sizetmp)) {
659                 sizetmp = 1ULL << ctzl(mem_start);
660             }
661 
662             spapr_dt_memory_node(fdt, i, mem_start, sizetmp);
663             node_size -= sizetmp;
664             mem_start += sizetmp;
665         }
666     }
667 
668     /* Generate ibm,dynamic-reconfiguration-memory node if required */
669     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
670         int ret;
671 
672         g_assert(smc->dr_lmb_enabled);
673         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
674         if (ret) {
675             return ret;
676         }
677     }
678 
679     return 0;
680 }
681 
682 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
683                          SpaprMachineState *spapr)
684 {
685     MachineState *ms = MACHINE(spapr);
686     PowerPCCPU *cpu = POWERPC_CPU(cs);
687     CPUPPCState *env = &cpu->env;
688     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
689     int index = spapr_get_vcpu_id(cpu);
690     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
691                        0xffffffff, 0xffffffff};
692     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
693         : SPAPR_TIMEBASE_FREQ;
694     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
695     uint32_t page_sizes_prop[64];
696     size_t page_sizes_prop_size;
697     unsigned int smp_threads = ms->smp.threads;
698     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
699     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
700     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
701     SpaprDrc *drc;
702     int drc_index;
703     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
704     int i;
705 
706     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
707     if (drc) {
708         drc_index = spapr_drc_index(drc);
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
710     }
711 
712     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
713     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
714 
715     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
716     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
717                            env->dcache_line_size)));
718     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
719                            env->dcache_line_size)));
720     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
721                            env->icache_line_size)));
722     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
723                            env->icache_line_size)));
724 
725     if (pcc->l1_dcache_size) {
726         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
727                                pcc->l1_dcache_size)));
728     } else {
729         warn_report("Unknown L1 dcache size for cpu");
730     }
731     if (pcc->l1_icache_size) {
732         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
733                                pcc->l1_icache_size)));
734     } else {
735         warn_report("Unknown L1 icache size for cpu");
736     }
737 
738     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
739     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
740     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
741     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
742     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
743     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
744 
745     if (env->spr_cb[SPR_PURR].oea_read) {
746         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
747     }
748     if (env->spr_cb[SPR_SPURR].oea_read) {
749         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
750     }
751 
752     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
753         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
754                           segs, sizeof(segs))));
755     }
756 
757     /* Advertise VSX (vector extensions) if available
758      *   1               == VMX / Altivec available
759      *   2               == VSX available
760      *
761      * Only CPUs for which we create core types in spapr_cpu_core.c
762      * are possible, and all of those have VMX */
763     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
764         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
765     } else {
766         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
767     }
768 
769     /* Advertise DFP (Decimal Floating Point) if available
770      *   0 / no property == no DFP
771      *   1               == DFP available */
772     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
773         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
774     }
775 
776     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
777                                                       sizeof(page_sizes_prop));
778     if (page_sizes_prop_size) {
779         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
780                           page_sizes_prop, page_sizes_prop_size)));
781     }
782 
783     spapr_dt_pa_features(spapr, cpu, fdt, offset);
784 
785     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
786                            cs->cpu_index / vcpus_per_socket)));
787 
788     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
789                       pft_size_prop, sizeof(pft_size_prop))));
790 
791     if (ms->numa_state->num_nodes > 1) {
792         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
793     }
794 
795     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
796 
797     if (pcc->radix_page_info) {
798         for (i = 0; i < pcc->radix_page_info->count; i++) {
799             radix_AP_encodings[i] =
800                 cpu_to_be32(pcc->radix_page_info->entries[i]);
801         }
802         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
803                           radix_AP_encodings,
804                           pcc->radix_page_info->count *
805                           sizeof(radix_AP_encodings[0]))));
806     }
807 
808     /*
809      * We set this property to let the guest know that it can use the large
810      * decrementer and its width in bits.
811      */
812     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
813         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
814                               pcc->lrg_decr_bits)));
815 }
816 
817 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
818 {
819     CPUState **rev;
820     CPUState *cs;
821     int n_cpus;
822     int cpus_offset;
823     char *nodename;
824     int i;
825 
826     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
827     _FDT(cpus_offset);
828     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
829     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
830 
831     /*
832      * We walk the CPUs in reverse order to ensure that CPU DT nodes
833      * created by fdt_add_subnode() end up in the right order in FDT
834      * for the guest kernel the enumerate the CPUs correctly.
835      *
836      * The CPU list cannot be traversed in reverse order, so we need
837      * to do extra work.
838      */
839     n_cpus = 0;
840     rev = NULL;
841     CPU_FOREACH(cs) {
842         rev = g_renew(CPUState *, rev, n_cpus + 1);
843         rev[n_cpus++] = cs;
844     }
845 
846     for (i = n_cpus - 1; i >= 0; i--) {
847         CPUState *cs = rev[i];
848         PowerPCCPU *cpu = POWERPC_CPU(cs);
849         int index = spapr_get_vcpu_id(cpu);
850         DeviceClass *dc = DEVICE_GET_CLASS(cs);
851         int offset;
852 
853         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
854             continue;
855         }
856 
857         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
858         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
859         g_free(nodename);
860         _FDT(offset);
861         spapr_dt_cpu(cs, fdt, offset, spapr);
862     }
863 
864     g_free(rev);
865 }
866 
867 static int spapr_dt_rng(void *fdt)
868 {
869     int node;
870     int ret;
871 
872     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
873     if (node <= 0) {
874         return -1;
875     }
876     ret = fdt_setprop_string(fdt, node, "device_type",
877                              "ibm,platform-facilities");
878     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
879     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
880 
881     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
882     if (node <= 0) {
883         return -1;
884     }
885     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
886 
887     return ret ? -1 : 0;
888 }
889 
890 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
891 {
892     MachineState *ms = MACHINE(spapr);
893     int rtas;
894     GString *hypertas = g_string_sized_new(256);
895     GString *qemu_hypertas = g_string_sized_new(256);
896     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
897     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
898         memory_region_size(&MACHINE(spapr)->device_memory->mr);
899     uint32_t lrdr_capacity[] = {
900         cpu_to_be32(max_device_addr >> 32),
901         cpu_to_be32(max_device_addr & 0xffffffff),
902         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
903         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
904     };
905     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
906     uint32_t maxdomains[] = {
907         cpu_to_be32(4),
908         maxdomain,
909         maxdomain,
910         maxdomain,
911         cpu_to_be32(spapr->gpu_numa_id),
912     };
913 
914     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
915 
916     /* hypertas */
917     add_str(hypertas, "hcall-pft");
918     add_str(hypertas, "hcall-term");
919     add_str(hypertas, "hcall-dabr");
920     add_str(hypertas, "hcall-interrupt");
921     add_str(hypertas, "hcall-tce");
922     add_str(hypertas, "hcall-vio");
923     add_str(hypertas, "hcall-splpar");
924     add_str(hypertas, "hcall-join");
925     add_str(hypertas, "hcall-bulk");
926     add_str(hypertas, "hcall-set-mode");
927     add_str(hypertas, "hcall-sprg0");
928     add_str(hypertas, "hcall-copy");
929     add_str(hypertas, "hcall-debug");
930     add_str(hypertas, "hcall-vphn");
931     add_str(qemu_hypertas, "hcall-memop1");
932 
933     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
934         add_str(hypertas, "hcall-multi-tce");
935     }
936 
937     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
938         add_str(hypertas, "hcall-hpt-resize");
939     }
940 
941     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
942                      hypertas->str, hypertas->len));
943     g_string_free(hypertas, TRUE);
944     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
945                      qemu_hypertas->str, qemu_hypertas->len));
946     g_string_free(qemu_hypertas, TRUE);
947 
948     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
949                      refpoints, sizeof(refpoints)));
950 
951     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
952                      maxdomains, sizeof(maxdomains)));
953 
954     /*
955      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
956      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
957      *
958      * The system reset requirements are driven by existing Linux and PowerVM
959      * implementation which (contrary to PAPR) saves r3 in the error log
960      * structure like machine check, so Linux expects to find the saved r3
961      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
962      * does not look at the error value).
963      *
964      * System reset interrupts are not subject to interlock like machine
965      * check, so this memory area could be corrupted if the sreset is
966      * interrupted by a machine check (or vice versa) if it was shared. To
967      * prevent this, system reset uses per-CPU areas for the sreset save
968      * area. A system reset that interrupts a system reset handler could
969      * still overwrite this area, but Linux doesn't try to recover in that
970      * case anyway.
971      *
972      * The extra 8 bytes is required because Linux's FWNMI error log check
973      * is off-by-one.
974      */
975     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
976 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
977     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
978                           RTAS_ERROR_LOG_MAX));
979     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
980                           RTAS_EVENT_SCAN_RATE));
981 
982     g_assert(msi_nonbroken);
983     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
984 
985     /*
986      * According to PAPR, rtas ibm,os-term does not guarantee a return
987      * back to the guest cpu.
988      *
989      * While an additional ibm,extended-os-term property indicates
990      * that rtas call return will always occur. Set this property.
991      */
992     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
993 
994     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
995                      lrdr_capacity, sizeof(lrdr_capacity)));
996 
997     spapr_dt_rtas_tokens(fdt, rtas);
998 }
999 
1000 /*
1001  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1002  * and the XIVE features that the guest may request and thus the valid
1003  * values for bytes 23..26 of option vector 5:
1004  */
1005 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1006                                           int chosen)
1007 {
1008     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1009 
1010     char val[2 * 4] = {
1011         23, 0x00, /* XICS / XIVE mode */
1012         24, 0x00, /* Hash/Radix, filled in below. */
1013         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1014         26, 0x40, /* Radix options: GTSE == yes. */
1015     };
1016 
1017     if (spapr->irq->xics && spapr->irq->xive) {
1018         val[1] = SPAPR_OV5_XIVE_BOTH;
1019     } else if (spapr->irq->xive) {
1020         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1021     } else {
1022         assert(spapr->irq->xics);
1023         val[1] = SPAPR_OV5_XIVE_LEGACY;
1024     }
1025 
1026     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1027                           first_ppc_cpu->compat_pvr)) {
1028         /*
1029          * If we're in a pre POWER9 compat mode then the guest should
1030          * do hash and use the legacy interrupt mode
1031          */
1032         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1033         val[3] = 0x00; /* Hash */
1034     } else if (kvm_enabled()) {
1035         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1036             val[3] = 0x80; /* OV5_MMU_BOTH */
1037         } else if (kvmppc_has_cap_mmu_radix()) {
1038             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1039         } else {
1040             val[3] = 0x00; /* Hash */
1041         }
1042     } else {
1043         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1044         val[3] = 0xC0;
1045     }
1046     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1047                      val, sizeof(val)));
1048 }
1049 
1050 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1051 {
1052     MachineState *machine = MACHINE(spapr);
1053     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1054     int chosen;
1055 
1056     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1057 
1058     if (reset) {
1059         const char *boot_device = machine->boot_order;
1060         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1061         size_t cb = 0;
1062         char *bootlist = get_boot_devices_list(&cb);
1063 
1064         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1065             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1066                                     machine->kernel_cmdline));
1067         }
1068 
1069         if (spapr->initrd_size) {
1070             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1071                                   spapr->initrd_base));
1072             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1073                                   spapr->initrd_base + spapr->initrd_size));
1074         }
1075 
1076         if (spapr->kernel_size) {
1077             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1078                                   cpu_to_be64(spapr->kernel_size) };
1079 
1080             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1081                          &kprop, sizeof(kprop)));
1082             if (spapr->kernel_le) {
1083                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1084             }
1085         }
1086         if (boot_menu) {
1087             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1088         }
1089         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1090         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1091         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1092 
1093         if (cb && bootlist) {
1094             int i;
1095 
1096             for (i = 0; i < cb; i++) {
1097                 if (bootlist[i] == '\n') {
1098                     bootlist[i] = ' ';
1099                 }
1100             }
1101             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1102         }
1103 
1104         if (boot_device && strlen(boot_device)) {
1105             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1106         }
1107 
1108         if (!spapr->has_graphics && stdout_path) {
1109             /*
1110              * "linux,stdout-path" and "stdout" properties are
1111              * deprecated by linux kernel. New platforms should only
1112              * use the "stdout-path" property. Set the new property
1113              * and continue using older property to remain compatible
1114              * with the existing firmware.
1115              */
1116             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1117             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1118         }
1119 
1120         /*
1121          * We can deal with BAR reallocation just fine, advertise it
1122          * to the guest
1123          */
1124         if (smc->linux_pci_probe) {
1125             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1126         }
1127 
1128         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1129 
1130         g_free(stdout_path);
1131         g_free(bootlist);
1132     }
1133 
1134     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1135 }
1136 
1137 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1138 {
1139     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1140      * KVM to work under pHyp with some guest co-operation */
1141     int hypervisor;
1142     uint8_t hypercall[16];
1143 
1144     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1145     /* indicate KVM hypercall interface */
1146     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1147     if (kvmppc_has_cap_fixup_hcalls()) {
1148         /*
1149          * Older KVM versions with older guest kernels were broken
1150          * with the magic page, don't allow the guest to map it.
1151          */
1152         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1153                                   sizeof(hypercall))) {
1154             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1155                              hypercall, sizeof(hypercall)));
1156         }
1157     }
1158 }
1159 
1160 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1161 {
1162     MachineState *machine = MACHINE(spapr);
1163     MachineClass *mc = MACHINE_GET_CLASS(machine);
1164     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1165     int ret;
1166     void *fdt;
1167     SpaprPhbState *phb;
1168     char *buf;
1169 
1170     fdt = g_malloc0(space);
1171     _FDT((fdt_create_empty_tree(fdt, space)));
1172 
1173     /* Root node */
1174     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1175     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1176     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1177 
1178     /* Guest UUID & Name*/
1179     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1180     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1181     if (qemu_uuid_set) {
1182         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1183     }
1184     g_free(buf);
1185 
1186     if (qemu_get_vm_name()) {
1187         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1188                                 qemu_get_vm_name()));
1189     }
1190 
1191     /* Host Model & Serial Number */
1192     if (spapr->host_model) {
1193         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1194     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1195         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1196         g_free(buf);
1197     }
1198 
1199     if (spapr->host_serial) {
1200         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1201     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1202         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1203         g_free(buf);
1204     }
1205 
1206     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1207     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1208 
1209     /* /interrupt controller */
1210     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1211 
1212     ret = spapr_dt_memory(spapr, fdt);
1213     if (ret < 0) {
1214         error_report("couldn't setup memory nodes in fdt");
1215         exit(1);
1216     }
1217 
1218     /* /vdevice */
1219     spapr_dt_vdevice(spapr->vio_bus, fdt);
1220 
1221     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1222         ret = spapr_dt_rng(fdt);
1223         if (ret < 0) {
1224             error_report("could not set up rng device in the fdt");
1225             exit(1);
1226         }
1227     }
1228 
1229     QLIST_FOREACH(phb, &spapr->phbs, list) {
1230         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1231         if (ret < 0) {
1232             error_report("couldn't setup PCI devices in fdt");
1233             exit(1);
1234         }
1235     }
1236 
1237     spapr_dt_cpus(fdt, spapr);
1238 
1239     if (smc->dr_lmb_enabled) {
1240         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1241     }
1242 
1243     if (mc->has_hotpluggable_cpus) {
1244         int offset = fdt_path_offset(fdt, "/cpus");
1245         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1246         if (ret < 0) {
1247             error_report("Couldn't set up CPU DR device tree properties");
1248             exit(1);
1249         }
1250     }
1251 
1252     /* /event-sources */
1253     spapr_dt_events(spapr, fdt);
1254 
1255     /* /rtas */
1256     spapr_dt_rtas(spapr, fdt);
1257 
1258     /* /chosen */
1259     spapr_dt_chosen(spapr, fdt, reset);
1260 
1261     /* /hypervisor */
1262     if (kvm_enabled()) {
1263         spapr_dt_hypervisor(spapr, fdt);
1264     }
1265 
1266     /* Build memory reserve map */
1267     if (reset) {
1268         if (spapr->kernel_size) {
1269             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1270                                   spapr->kernel_size)));
1271         }
1272         if (spapr->initrd_size) {
1273             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1274                                   spapr->initrd_size)));
1275         }
1276     }
1277 
1278     if (smc->dr_phb_enabled) {
1279         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1280         if (ret < 0) {
1281             error_report("Couldn't set up PHB DR device tree properties");
1282             exit(1);
1283         }
1284     }
1285 
1286     /* NVDIMM devices */
1287     if (mc->nvdimm_supported) {
1288         spapr_dt_persistent_memory(fdt);
1289     }
1290 
1291     return fdt;
1292 }
1293 
1294 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1295 {
1296     SpaprMachineState *spapr = opaque;
1297 
1298     return (addr & 0x0fffffff) + spapr->kernel_addr;
1299 }
1300 
1301 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1302                                     PowerPCCPU *cpu)
1303 {
1304     CPUPPCState *env = &cpu->env;
1305 
1306     /* The TCG path should also be holding the BQL at this point */
1307     g_assert(qemu_mutex_iothread_locked());
1308 
1309     if (msr_pr) {
1310         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1311         env->gpr[3] = H_PRIVILEGE;
1312     } else {
1313         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1314     }
1315 }
1316 
1317 struct LPCRSyncState {
1318     target_ulong value;
1319     target_ulong mask;
1320 };
1321 
1322 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1323 {
1324     struct LPCRSyncState *s = arg.host_ptr;
1325     PowerPCCPU *cpu = POWERPC_CPU(cs);
1326     CPUPPCState *env = &cpu->env;
1327     target_ulong lpcr;
1328 
1329     cpu_synchronize_state(cs);
1330     lpcr = env->spr[SPR_LPCR];
1331     lpcr &= ~s->mask;
1332     lpcr |= s->value;
1333     ppc_store_lpcr(cpu, lpcr);
1334 }
1335 
1336 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1337 {
1338     CPUState *cs;
1339     struct LPCRSyncState s = {
1340         .value = value,
1341         .mask = mask
1342     };
1343     CPU_FOREACH(cs) {
1344         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1345     }
1346 }
1347 
1348 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1349 {
1350     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1351 
1352     /* Copy PATE1:GR into PATE0:HR */
1353     entry->dw0 = spapr->patb_entry & PATE0_HR;
1354     entry->dw1 = spapr->patb_entry;
1355 }
1356 
1357 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1358 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1359 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1360 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1361 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1362 
1363 /*
1364  * Get the fd to access the kernel htab, re-opening it if necessary
1365  */
1366 static int get_htab_fd(SpaprMachineState *spapr)
1367 {
1368     Error *local_err = NULL;
1369 
1370     if (spapr->htab_fd >= 0) {
1371         return spapr->htab_fd;
1372     }
1373 
1374     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1375     if (spapr->htab_fd < 0) {
1376         error_report_err(local_err);
1377     }
1378 
1379     return spapr->htab_fd;
1380 }
1381 
1382 void close_htab_fd(SpaprMachineState *spapr)
1383 {
1384     if (spapr->htab_fd >= 0) {
1385         close(spapr->htab_fd);
1386     }
1387     spapr->htab_fd = -1;
1388 }
1389 
1390 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1391 {
1392     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1393 
1394     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1395 }
1396 
1397 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1398 {
1399     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1400 
1401     assert(kvm_enabled());
1402 
1403     if (!spapr->htab) {
1404         return 0;
1405     }
1406 
1407     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1408 }
1409 
1410 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1411                                                 hwaddr ptex, int n)
1412 {
1413     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1414     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1415 
1416     if (!spapr->htab) {
1417         /*
1418          * HTAB is controlled by KVM. Fetch into temporary buffer
1419          */
1420         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1421         kvmppc_read_hptes(hptes, ptex, n);
1422         return hptes;
1423     }
1424 
1425     /*
1426      * HTAB is controlled by QEMU. Just point to the internally
1427      * accessible PTEG.
1428      */
1429     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1430 }
1431 
1432 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1433                               const ppc_hash_pte64_t *hptes,
1434                               hwaddr ptex, int n)
1435 {
1436     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1437 
1438     if (!spapr->htab) {
1439         g_free((void *)hptes);
1440     }
1441 
1442     /* Nothing to do for qemu managed HPT */
1443 }
1444 
1445 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1446                       uint64_t pte0, uint64_t pte1)
1447 {
1448     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1449     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1450 
1451     if (!spapr->htab) {
1452         kvmppc_write_hpte(ptex, pte0, pte1);
1453     } else {
1454         if (pte0 & HPTE64_V_VALID) {
1455             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1456             /*
1457              * When setting valid, we write PTE1 first. This ensures
1458              * proper synchronization with the reading code in
1459              * ppc_hash64_pteg_search()
1460              */
1461             smp_wmb();
1462             stq_p(spapr->htab + offset, pte0);
1463         } else {
1464             stq_p(spapr->htab + offset, pte0);
1465             /*
1466              * When clearing it we set PTE0 first. This ensures proper
1467              * synchronization with the reading code in
1468              * ppc_hash64_pteg_search()
1469              */
1470             smp_wmb();
1471             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1472         }
1473     }
1474 }
1475 
1476 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1477                              uint64_t pte1)
1478 {
1479     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1480     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1481 
1482     if (!spapr->htab) {
1483         /* There should always be a hash table when this is called */
1484         error_report("spapr_hpte_set_c called with no hash table !");
1485         return;
1486     }
1487 
1488     /* The HW performs a non-atomic byte update */
1489     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1490 }
1491 
1492 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1493                              uint64_t pte1)
1494 {
1495     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1496     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1497 
1498     if (!spapr->htab) {
1499         /* There should always be a hash table when this is called */
1500         error_report("spapr_hpte_set_r called with no hash table !");
1501         return;
1502     }
1503 
1504     /* The HW performs a non-atomic byte update */
1505     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1506 }
1507 
1508 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1509 {
1510     int shift;
1511 
1512     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1513      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1514      * that's much more than is needed for Linux guests */
1515     shift = ctz64(pow2ceil(ramsize)) - 7;
1516     shift = MAX(shift, 18); /* Minimum architected size */
1517     shift = MIN(shift, 46); /* Maximum architected size */
1518     return shift;
1519 }
1520 
1521 void spapr_free_hpt(SpaprMachineState *spapr)
1522 {
1523     g_free(spapr->htab);
1524     spapr->htab = NULL;
1525     spapr->htab_shift = 0;
1526     close_htab_fd(spapr);
1527 }
1528 
1529 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1530                           Error **errp)
1531 {
1532     long rc;
1533 
1534     /* Clean up any HPT info from a previous boot */
1535     spapr_free_hpt(spapr);
1536 
1537     rc = kvmppc_reset_htab(shift);
1538     if (rc < 0) {
1539         /* kernel-side HPT needed, but couldn't allocate one */
1540         error_setg_errno(errp, errno,
1541                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1542                          shift);
1543         /* This is almost certainly fatal, but if the caller really
1544          * wants to carry on with shift == 0, it's welcome to try */
1545     } else if (rc > 0) {
1546         /* kernel-side HPT allocated */
1547         if (rc != shift) {
1548             error_setg(errp,
1549                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1550                        shift, rc);
1551         }
1552 
1553         spapr->htab_shift = shift;
1554         spapr->htab = NULL;
1555     } else {
1556         /* kernel-side HPT not needed, allocate in userspace instead */
1557         size_t size = 1ULL << shift;
1558         int i;
1559 
1560         spapr->htab = qemu_memalign(size, size);
1561         if (!spapr->htab) {
1562             error_setg_errno(errp, errno,
1563                              "Could not allocate HPT of order %d", shift);
1564             return;
1565         }
1566 
1567         memset(spapr->htab, 0, size);
1568         spapr->htab_shift = shift;
1569 
1570         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1571             DIRTY_HPTE(HPTE(spapr->htab, i));
1572         }
1573     }
1574     /* We're setting up a hash table, so that means we're not radix */
1575     spapr->patb_entry = 0;
1576     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1577 }
1578 
1579 void spapr_setup_hpt(SpaprMachineState *spapr)
1580 {
1581     int hpt_shift;
1582 
1583     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1584         || (spapr->cas_reboot
1585             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1586         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1587     } else {
1588         uint64_t current_ram_size;
1589 
1590         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1591         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1592     }
1593     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1594 
1595     if (kvm_enabled()) {
1596         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1597 
1598         /* Check our RMA fits in the possible VRMA */
1599         if (vrma_limit < spapr->rma_size) {
1600             error_report("Unable to create %" HWADDR_PRIu
1601                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1602                          spapr->rma_size / MiB, vrma_limit / MiB);
1603             exit(EXIT_FAILURE);
1604         }
1605     }
1606 }
1607 
1608 static int spapr_reset_drcs(Object *child, void *opaque)
1609 {
1610     SpaprDrc *drc =
1611         (SpaprDrc *) object_dynamic_cast(child,
1612                                                  TYPE_SPAPR_DR_CONNECTOR);
1613 
1614     if (drc) {
1615         spapr_drc_reset(drc);
1616     }
1617 
1618     return 0;
1619 }
1620 
1621 static void spapr_machine_reset(MachineState *machine)
1622 {
1623     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1624     PowerPCCPU *first_ppc_cpu;
1625     hwaddr fdt_addr;
1626     void *fdt;
1627     int rc;
1628 
1629     kvmppc_svm_off(&error_fatal);
1630     spapr_caps_apply(spapr);
1631 
1632     first_ppc_cpu = POWERPC_CPU(first_cpu);
1633     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1634         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1635                               spapr->max_compat_pvr)) {
1636         /*
1637          * If using KVM with radix mode available, VCPUs can be started
1638          * without a HPT because KVM will start them in radix mode.
1639          * Set the GR bit in PATE so that we know there is no HPT.
1640          */
1641         spapr->patb_entry = PATE1_GR;
1642         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1643     } else {
1644         spapr_setup_hpt(spapr);
1645     }
1646 
1647     qemu_devices_reset();
1648 
1649     /*
1650      * If this reset wasn't generated by CAS, we should reset our
1651      * negotiated options and start from scratch
1652      */
1653     if (!spapr->cas_reboot) {
1654         spapr_ovec_cleanup(spapr->ov5_cas);
1655         spapr->ov5_cas = spapr_ovec_new();
1656 
1657         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1658     }
1659 
1660     /*
1661      * This is fixing some of the default configuration of the XIVE
1662      * devices. To be called after the reset of the machine devices.
1663      */
1664     spapr_irq_reset(spapr, &error_fatal);
1665 
1666     /*
1667      * There is no CAS under qtest. Simulate one to please the code that
1668      * depends on spapr->ov5_cas. This is especially needed to test device
1669      * unplug, so we do that before resetting the DRCs.
1670      */
1671     if (qtest_enabled()) {
1672         spapr_ovec_cleanup(spapr->ov5_cas);
1673         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1674     }
1675 
1676     /* DRC reset may cause a device to be unplugged. This will cause troubles
1677      * if this device is used by another device (eg, a running vhost backend
1678      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1679      * situations, we reset DRCs after all devices have been reset.
1680      */
1681     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1682 
1683     spapr_clear_pending_events(spapr);
1684 
1685     /*
1686      * We place the device tree and RTAS just below either the top of the RMA,
1687      * or just below 2GB, whichever is lower, so that it can be
1688      * processed with 32-bit real mode code if necessary
1689      */
1690     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1691 
1692     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1693 
1694     rc = fdt_pack(fdt);
1695 
1696     /* Should only fail if we've built a corrupted tree */
1697     assert(rc == 0);
1698 
1699     /* Load the fdt */
1700     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1701     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1702     g_free(spapr->fdt_blob);
1703     spapr->fdt_size = fdt_totalsize(fdt);
1704     spapr->fdt_initial_size = spapr->fdt_size;
1705     spapr->fdt_blob = fdt;
1706 
1707     /* Set up the entry state */
1708     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1709     first_ppc_cpu->env.gpr[5] = 0;
1710 
1711     spapr->cas_reboot = false;
1712 
1713     spapr->fwnmi_system_reset_addr = -1;
1714     spapr->fwnmi_machine_check_addr = -1;
1715     spapr->fwnmi_machine_check_interlock = -1;
1716 
1717     /* Signal all vCPUs waiting on this condition */
1718     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1719 
1720     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1721 }
1722 
1723 static void spapr_create_nvram(SpaprMachineState *spapr)
1724 {
1725     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1726     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1727 
1728     if (dinfo) {
1729         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1730                             &error_fatal);
1731     }
1732 
1733     qdev_init_nofail(dev);
1734 
1735     spapr->nvram = (struct SpaprNvram *)dev;
1736 }
1737 
1738 static void spapr_rtc_create(SpaprMachineState *spapr)
1739 {
1740     object_initialize_child(OBJECT(spapr), "rtc",
1741                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1742                             &error_fatal, NULL);
1743     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1744                               &error_fatal);
1745     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1746                               "date", &error_fatal);
1747 }
1748 
1749 /* Returns whether we want to use VGA or not */
1750 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1751 {
1752     switch (vga_interface_type) {
1753     case VGA_NONE:
1754         return false;
1755     case VGA_DEVICE:
1756         return true;
1757     case VGA_STD:
1758     case VGA_VIRTIO:
1759     case VGA_CIRRUS:
1760         return pci_vga_init(pci_bus) != NULL;
1761     default:
1762         error_setg(errp,
1763                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1764         return false;
1765     }
1766 }
1767 
1768 static int spapr_pre_load(void *opaque)
1769 {
1770     int rc;
1771 
1772     rc = spapr_caps_pre_load(opaque);
1773     if (rc) {
1774         return rc;
1775     }
1776 
1777     return 0;
1778 }
1779 
1780 static int spapr_post_load(void *opaque, int version_id)
1781 {
1782     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1783     int err = 0;
1784 
1785     err = spapr_caps_post_migration(spapr);
1786     if (err) {
1787         return err;
1788     }
1789 
1790     /*
1791      * In earlier versions, there was no separate qdev for the PAPR
1792      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1793      * So when migrating from those versions, poke the incoming offset
1794      * value into the RTC device
1795      */
1796     if (version_id < 3) {
1797         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1798         if (err) {
1799             return err;
1800         }
1801     }
1802 
1803     if (kvm_enabled() && spapr->patb_entry) {
1804         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1805         bool radix = !!(spapr->patb_entry & PATE1_GR);
1806         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1807 
1808         /*
1809          * Update LPCR:HR and UPRT as they may not be set properly in
1810          * the stream
1811          */
1812         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1813                             LPCR_HR | LPCR_UPRT);
1814 
1815         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1816         if (err) {
1817             error_report("Process table config unsupported by the host");
1818             return -EINVAL;
1819         }
1820     }
1821 
1822     err = spapr_irq_post_load(spapr, version_id);
1823     if (err) {
1824         return err;
1825     }
1826 
1827     return err;
1828 }
1829 
1830 static int spapr_pre_save(void *opaque)
1831 {
1832     int rc;
1833 
1834     rc = spapr_caps_pre_save(opaque);
1835     if (rc) {
1836         return rc;
1837     }
1838 
1839     return 0;
1840 }
1841 
1842 static bool version_before_3(void *opaque, int version_id)
1843 {
1844     return version_id < 3;
1845 }
1846 
1847 static bool spapr_pending_events_needed(void *opaque)
1848 {
1849     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1850     return !QTAILQ_EMPTY(&spapr->pending_events);
1851 }
1852 
1853 static const VMStateDescription vmstate_spapr_event_entry = {
1854     .name = "spapr_event_log_entry",
1855     .version_id = 1,
1856     .minimum_version_id = 1,
1857     .fields = (VMStateField[]) {
1858         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1859         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1860         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1861                                      NULL, extended_length),
1862         VMSTATE_END_OF_LIST()
1863     },
1864 };
1865 
1866 static const VMStateDescription vmstate_spapr_pending_events = {
1867     .name = "spapr_pending_events",
1868     .version_id = 1,
1869     .minimum_version_id = 1,
1870     .needed = spapr_pending_events_needed,
1871     .fields = (VMStateField[]) {
1872         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1873                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1874         VMSTATE_END_OF_LIST()
1875     },
1876 };
1877 
1878 static bool spapr_ov5_cas_needed(void *opaque)
1879 {
1880     SpaprMachineState *spapr = opaque;
1881     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1882     bool cas_needed;
1883 
1884     /* Prior to the introduction of SpaprOptionVector, we had two option
1885      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1886      * Both of these options encode machine topology into the device-tree
1887      * in such a way that the now-booted OS should still be able to interact
1888      * appropriately with QEMU regardless of what options were actually
1889      * negotiatied on the source side.
1890      *
1891      * As such, we can avoid migrating the CAS-negotiated options if these
1892      * are the only options available on the current machine/platform.
1893      * Since these are the only options available for pseries-2.7 and
1894      * earlier, this allows us to maintain old->new/new->old migration
1895      * compatibility.
1896      *
1897      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1898      * via default pseries-2.8 machines and explicit command-line parameters.
1899      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1900      * of the actual CAS-negotiated values to continue working properly. For
1901      * example, availability of memory unplug depends on knowing whether
1902      * OV5_HP_EVT was negotiated via CAS.
1903      *
1904      * Thus, for any cases where the set of available CAS-negotiatable
1905      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1906      * include the CAS-negotiated options in the migration stream, unless
1907      * if they affect boot time behaviour only.
1908      */
1909     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1910     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1911     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1912 
1913     /* We need extra information if we have any bits outside the mask
1914      * defined above */
1915     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1916 
1917     spapr_ovec_cleanup(ov5_mask);
1918 
1919     return cas_needed;
1920 }
1921 
1922 static const VMStateDescription vmstate_spapr_ov5_cas = {
1923     .name = "spapr_option_vector_ov5_cas",
1924     .version_id = 1,
1925     .minimum_version_id = 1,
1926     .needed = spapr_ov5_cas_needed,
1927     .fields = (VMStateField[]) {
1928         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1929                                  vmstate_spapr_ovec, SpaprOptionVector),
1930         VMSTATE_END_OF_LIST()
1931     },
1932 };
1933 
1934 static bool spapr_patb_entry_needed(void *opaque)
1935 {
1936     SpaprMachineState *spapr = opaque;
1937 
1938     return !!spapr->patb_entry;
1939 }
1940 
1941 static const VMStateDescription vmstate_spapr_patb_entry = {
1942     .name = "spapr_patb_entry",
1943     .version_id = 1,
1944     .minimum_version_id = 1,
1945     .needed = spapr_patb_entry_needed,
1946     .fields = (VMStateField[]) {
1947         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1948         VMSTATE_END_OF_LIST()
1949     },
1950 };
1951 
1952 static bool spapr_irq_map_needed(void *opaque)
1953 {
1954     SpaprMachineState *spapr = opaque;
1955 
1956     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1957 }
1958 
1959 static const VMStateDescription vmstate_spapr_irq_map = {
1960     .name = "spapr_irq_map",
1961     .version_id = 1,
1962     .minimum_version_id = 1,
1963     .needed = spapr_irq_map_needed,
1964     .fields = (VMStateField[]) {
1965         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1966         VMSTATE_END_OF_LIST()
1967     },
1968 };
1969 
1970 static bool spapr_dtb_needed(void *opaque)
1971 {
1972     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1973 
1974     return smc->update_dt_enabled;
1975 }
1976 
1977 static int spapr_dtb_pre_load(void *opaque)
1978 {
1979     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1980 
1981     g_free(spapr->fdt_blob);
1982     spapr->fdt_blob = NULL;
1983     spapr->fdt_size = 0;
1984 
1985     return 0;
1986 }
1987 
1988 static const VMStateDescription vmstate_spapr_dtb = {
1989     .name = "spapr_dtb",
1990     .version_id = 1,
1991     .minimum_version_id = 1,
1992     .needed = spapr_dtb_needed,
1993     .pre_load = spapr_dtb_pre_load,
1994     .fields = (VMStateField[]) {
1995         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1996         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1997         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1998                                      fdt_size),
1999         VMSTATE_END_OF_LIST()
2000     },
2001 };
2002 
2003 static bool spapr_fwnmi_needed(void *opaque)
2004 {
2005     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2006 
2007     return spapr->fwnmi_machine_check_addr != -1;
2008 }
2009 
2010 static int spapr_fwnmi_pre_save(void *opaque)
2011 {
2012     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2013 
2014     /*
2015      * Check if machine check handling is in progress and print a
2016      * warning message.
2017      */
2018     if (spapr->fwnmi_machine_check_interlock != -1) {
2019         warn_report("A machine check is being handled during migration. The"
2020                 "handler may run and log hardware error on the destination");
2021     }
2022 
2023     return 0;
2024 }
2025 
2026 static const VMStateDescription vmstate_spapr_fwnmi = {
2027     .name = "spapr_fwnmi",
2028     .version_id = 1,
2029     .minimum_version_id = 1,
2030     .needed = spapr_fwnmi_needed,
2031     .pre_save = spapr_fwnmi_pre_save,
2032     .fields = (VMStateField[]) {
2033         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2034         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2035         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2036         VMSTATE_END_OF_LIST()
2037     },
2038 };
2039 
2040 static const VMStateDescription vmstate_spapr = {
2041     .name = "spapr",
2042     .version_id = 3,
2043     .minimum_version_id = 1,
2044     .pre_load = spapr_pre_load,
2045     .post_load = spapr_post_load,
2046     .pre_save = spapr_pre_save,
2047     .fields = (VMStateField[]) {
2048         /* used to be @next_irq */
2049         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2050 
2051         /* RTC offset */
2052         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2053 
2054         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2055         VMSTATE_END_OF_LIST()
2056     },
2057     .subsections = (const VMStateDescription*[]) {
2058         &vmstate_spapr_ov5_cas,
2059         &vmstate_spapr_patb_entry,
2060         &vmstate_spapr_pending_events,
2061         &vmstate_spapr_cap_htm,
2062         &vmstate_spapr_cap_vsx,
2063         &vmstate_spapr_cap_dfp,
2064         &vmstate_spapr_cap_cfpc,
2065         &vmstate_spapr_cap_sbbc,
2066         &vmstate_spapr_cap_ibs,
2067         &vmstate_spapr_cap_hpt_maxpagesize,
2068         &vmstate_spapr_irq_map,
2069         &vmstate_spapr_cap_nested_kvm_hv,
2070         &vmstate_spapr_dtb,
2071         &vmstate_spapr_cap_large_decr,
2072         &vmstate_spapr_cap_ccf_assist,
2073         &vmstate_spapr_cap_fwnmi,
2074         &vmstate_spapr_fwnmi,
2075         NULL
2076     }
2077 };
2078 
2079 static int htab_save_setup(QEMUFile *f, void *opaque)
2080 {
2081     SpaprMachineState *spapr = opaque;
2082 
2083     /* "Iteration" header */
2084     if (!spapr->htab_shift) {
2085         qemu_put_be32(f, -1);
2086     } else {
2087         qemu_put_be32(f, spapr->htab_shift);
2088     }
2089 
2090     if (spapr->htab) {
2091         spapr->htab_save_index = 0;
2092         spapr->htab_first_pass = true;
2093     } else {
2094         if (spapr->htab_shift) {
2095             assert(kvm_enabled());
2096         }
2097     }
2098 
2099 
2100     return 0;
2101 }
2102 
2103 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2104                             int chunkstart, int n_valid, int n_invalid)
2105 {
2106     qemu_put_be32(f, chunkstart);
2107     qemu_put_be16(f, n_valid);
2108     qemu_put_be16(f, n_invalid);
2109     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2110                     HASH_PTE_SIZE_64 * n_valid);
2111 }
2112 
2113 static void htab_save_end_marker(QEMUFile *f)
2114 {
2115     qemu_put_be32(f, 0);
2116     qemu_put_be16(f, 0);
2117     qemu_put_be16(f, 0);
2118 }
2119 
2120 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2121                                  int64_t max_ns)
2122 {
2123     bool has_timeout = max_ns != -1;
2124     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2125     int index = spapr->htab_save_index;
2126     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2127 
2128     assert(spapr->htab_first_pass);
2129 
2130     do {
2131         int chunkstart;
2132 
2133         /* Consume invalid HPTEs */
2134         while ((index < htabslots)
2135                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2136             CLEAN_HPTE(HPTE(spapr->htab, index));
2137             index++;
2138         }
2139 
2140         /* Consume valid HPTEs */
2141         chunkstart = index;
2142         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2143                && HPTE_VALID(HPTE(spapr->htab, index))) {
2144             CLEAN_HPTE(HPTE(spapr->htab, index));
2145             index++;
2146         }
2147 
2148         if (index > chunkstart) {
2149             int n_valid = index - chunkstart;
2150 
2151             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2152 
2153             if (has_timeout &&
2154                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2155                 break;
2156             }
2157         }
2158     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2159 
2160     if (index >= htabslots) {
2161         assert(index == htabslots);
2162         index = 0;
2163         spapr->htab_first_pass = false;
2164     }
2165     spapr->htab_save_index = index;
2166 }
2167 
2168 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2169                                 int64_t max_ns)
2170 {
2171     bool final = max_ns < 0;
2172     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2173     int examined = 0, sent = 0;
2174     int index = spapr->htab_save_index;
2175     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2176 
2177     assert(!spapr->htab_first_pass);
2178 
2179     do {
2180         int chunkstart, invalidstart;
2181 
2182         /* Consume non-dirty HPTEs */
2183         while ((index < htabslots)
2184                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2185             index++;
2186             examined++;
2187         }
2188 
2189         chunkstart = index;
2190         /* Consume valid dirty HPTEs */
2191         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2192                && HPTE_DIRTY(HPTE(spapr->htab, index))
2193                && HPTE_VALID(HPTE(spapr->htab, index))) {
2194             CLEAN_HPTE(HPTE(spapr->htab, index));
2195             index++;
2196             examined++;
2197         }
2198 
2199         invalidstart = index;
2200         /* Consume invalid dirty HPTEs */
2201         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2202                && HPTE_DIRTY(HPTE(spapr->htab, index))
2203                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2204             CLEAN_HPTE(HPTE(spapr->htab, index));
2205             index++;
2206             examined++;
2207         }
2208 
2209         if (index > chunkstart) {
2210             int n_valid = invalidstart - chunkstart;
2211             int n_invalid = index - invalidstart;
2212 
2213             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2214             sent += index - chunkstart;
2215 
2216             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2217                 break;
2218             }
2219         }
2220 
2221         if (examined >= htabslots) {
2222             break;
2223         }
2224 
2225         if (index >= htabslots) {
2226             assert(index == htabslots);
2227             index = 0;
2228         }
2229     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2230 
2231     if (index >= htabslots) {
2232         assert(index == htabslots);
2233         index = 0;
2234     }
2235 
2236     spapr->htab_save_index = index;
2237 
2238     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2239 }
2240 
2241 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2242 #define MAX_KVM_BUF_SIZE    2048
2243 
2244 static int htab_save_iterate(QEMUFile *f, void *opaque)
2245 {
2246     SpaprMachineState *spapr = opaque;
2247     int fd;
2248     int rc = 0;
2249 
2250     /* Iteration header */
2251     if (!spapr->htab_shift) {
2252         qemu_put_be32(f, -1);
2253         return 1;
2254     } else {
2255         qemu_put_be32(f, 0);
2256     }
2257 
2258     if (!spapr->htab) {
2259         assert(kvm_enabled());
2260 
2261         fd = get_htab_fd(spapr);
2262         if (fd < 0) {
2263             return fd;
2264         }
2265 
2266         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2267         if (rc < 0) {
2268             return rc;
2269         }
2270     } else  if (spapr->htab_first_pass) {
2271         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2272     } else {
2273         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2274     }
2275 
2276     htab_save_end_marker(f);
2277 
2278     return rc;
2279 }
2280 
2281 static int htab_save_complete(QEMUFile *f, void *opaque)
2282 {
2283     SpaprMachineState *spapr = opaque;
2284     int fd;
2285 
2286     /* Iteration header */
2287     if (!spapr->htab_shift) {
2288         qemu_put_be32(f, -1);
2289         return 0;
2290     } else {
2291         qemu_put_be32(f, 0);
2292     }
2293 
2294     if (!spapr->htab) {
2295         int rc;
2296 
2297         assert(kvm_enabled());
2298 
2299         fd = get_htab_fd(spapr);
2300         if (fd < 0) {
2301             return fd;
2302         }
2303 
2304         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2305         if (rc < 0) {
2306             return rc;
2307         }
2308     } else {
2309         if (spapr->htab_first_pass) {
2310             htab_save_first_pass(f, spapr, -1);
2311         }
2312         htab_save_later_pass(f, spapr, -1);
2313     }
2314 
2315     /* End marker */
2316     htab_save_end_marker(f);
2317 
2318     return 0;
2319 }
2320 
2321 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2322 {
2323     SpaprMachineState *spapr = opaque;
2324     uint32_t section_hdr;
2325     int fd = -1;
2326     Error *local_err = NULL;
2327 
2328     if (version_id < 1 || version_id > 1) {
2329         error_report("htab_load() bad version");
2330         return -EINVAL;
2331     }
2332 
2333     section_hdr = qemu_get_be32(f);
2334 
2335     if (section_hdr == -1) {
2336         spapr_free_hpt(spapr);
2337         return 0;
2338     }
2339 
2340     if (section_hdr) {
2341         /* First section gives the htab size */
2342         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2343         if (local_err) {
2344             error_report_err(local_err);
2345             return -EINVAL;
2346         }
2347         return 0;
2348     }
2349 
2350     if (!spapr->htab) {
2351         assert(kvm_enabled());
2352 
2353         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2354         if (fd < 0) {
2355             error_report_err(local_err);
2356             return fd;
2357         }
2358     }
2359 
2360     while (true) {
2361         uint32_t index;
2362         uint16_t n_valid, n_invalid;
2363 
2364         index = qemu_get_be32(f);
2365         n_valid = qemu_get_be16(f);
2366         n_invalid = qemu_get_be16(f);
2367 
2368         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2369             /* End of Stream */
2370             break;
2371         }
2372 
2373         if ((index + n_valid + n_invalid) >
2374             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2375             /* Bad index in stream */
2376             error_report(
2377                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2378                 index, n_valid, n_invalid, spapr->htab_shift);
2379             return -EINVAL;
2380         }
2381 
2382         if (spapr->htab) {
2383             if (n_valid) {
2384                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2385                                 HASH_PTE_SIZE_64 * n_valid);
2386             }
2387             if (n_invalid) {
2388                 memset(HPTE(spapr->htab, index + n_valid), 0,
2389                        HASH_PTE_SIZE_64 * n_invalid);
2390             }
2391         } else {
2392             int rc;
2393 
2394             assert(fd >= 0);
2395 
2396             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2397             if (rc < 0) {
2398                 return rc;
2399             }
2400         }
2401     }
2402 
2403     if (!spapr->htab) {
2404         assert(fd >= 0);
2405         close(fd);
2406     }
2407 
2408     return 0;
2409 }
2410 
2411 static void htab_save_cleanup(void *opaque)
2412 {
2413     SpaprMachineState *spapr = opaque;
2414 
2415     close_htab_fd(spapr);
2416 }
2417 
2418 static SaveVMHandlers savevm_htab_handlers = {
2419     .save_setup = htab_save_setup,
2420     .save_live_iterate = htab_save_iterate,
2421     .save_live_complete_precopy = htab_save_complete,
2422     .save_cleanup = htab_save_cleanup,
2423     .load_state = htab_load,
2424 };
2425 
2426 static void spapr_boot_set(void *opaque, const char *boot_device,
2427                            Error **errp)
2428 {
2429     MachineState *machine = MACHINE(opaque);
2430     machine->boot_order = g_strdup(boot_device);
2431 }
2432 
2433 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2434 {
2435     MachineState *machine = MACHINE(spapr);
2436     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2437     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2438     int i;
2439 
2440     for (i = 0; i < nr_lmbs; i++) {
2441         uint64_t addr;
2442 
2443         addr = i * lmb_size + machine->device_memory->base;
2444         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2445                                addr / lmb_size);
2446     }
2447 }
2448 
2449 /*
2450  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2451  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2452  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2453  */
2454 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2455 {
2456     int i;
2457 
2458     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2459         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2460                    " is not aligned to %" PRIu64 " MiB",
2461                    machine->ram_size,
2462                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2463         return;
2464     }
2465 
2466     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2467         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2468                    " is not aligned to %" PRIu64 " MiB",
2469                    machine->ram_size,
2470                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2471         return;
2472     }
2473 
2474     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2475         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2476             error_setg(errp,
2477                        "Node %d memory size 0x%" PRIx64
2478                        " is not aligned to %" PRIu64 " MiB",
2479                        i, machine->numa_state->nodes[i].node_mem,
2480                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2481             return;
2482         }
2483     }
2484 }
2485 
2486 /* find cpu slot in machine->possible_cpus by core_id */
2487 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2488 {
2489     int index = id / ms->smp.threads;
2490 
2491     if (index >= ms->possible_cpus->len) {
2492         return NULL;
2493     }
2494     if (idx) {
2495         *idx = index;
2496     }
2497     return &ms->possible_cpus->cpus[index];
2498 }
2499 
2500 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2501 {
2502     MachineState *ms = MACHINE(spapr);
2503     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2504     Error *local_err = NULL;
2505     bool vsmt_user = !!spapr->vsmt;
2506     int kvm_smt = kvmppc_smt_threads();
2507     int ret;
2508     unsigned int smp_threads = ms->smp.threads;
2509 
2510     if (!kvm_enabled() && (smp_threads > 1)) {
2511         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2512                      "on a pseries machine");
2513         goto out;
2514     }
2515     if (!is_power_of_2(smp_threads)) {
2516         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2517                      "machine because it must be a power of 2", smp_threads);
2518         goto out;
2519     }
2520 
2521     /* Detemine the VSMT mode to use: */
2522     if (vsmt_user) {
2523         if (spapr->vsmt < smp_threads) {
2524             error_setg(&local_err, "Cannot support VSMT mode %d"
2525                          " because it must be >= threads/core (%d)",
2526                          spapr->vsmt, smp_threads);
2527             goto out;
2528         }
2529         /* In this case, spapr->vsmt has been set by the command line */
2530     } else if (!smc->smp_threads_vsmt) {
2531         /*
2532          * Default VSMT value is tricky, because we need it to be as
2533          * consistent as possible (for migration), but this requires
2534          * changing it for at least some existing cases.  We pick 8 as
2535          * the value that we'd get with KVM on POWER8, the
2536          * overwhelmingly common case in production systems.
2537          */
2538         spapr->vsmt = MAX(8, smp_threads);
2539     } else {
2540         spapr->vsmt = smp_threads;
2541     }
2542 
2543     /* KVM: If necessary, set the SMT mode: */
2544     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2545         ret = kvmppc_set_smt_threads(spapr->vsmt);
2546         if (ret) {
2547             /* Looks like KVM isn't able to change VSMT mode */
2548             error_setg(&local_err,
2549                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2550                        spapr->vsmt, ret);
2551             /* We can live with that if the default one is big enough
2552              * for the number of threads, and a submultiple of the one
2553              * we want.  In this case we'll waste some vcpu ids, but
2554              * behaviour will be correct */
2555             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2556                 warn_report_err(local_err);
2557                 local_err = NULL;
2558                 goto out;
2559             } else {
2560                 if (!vsmt_user) {
2561                     error_append_hint(&local_err,
2562                                       "On PPC, a VM with %d threads/core"
2563                                       " on a host with %d threads/core"
2564                                       " requires the use of VSMT mode %d.\n",
2565                                       smp_threads, kvm_smt, spapr->vsmt);
2566                 }
2567                 kvmppc_error_append_smt_possible_hint(&local_err);
2568                 goto out;
2569             }
2570         }
2571     }
2572     /* else TCG: nothing to do currently */
2573 out:
2574     error_propagate(errp, local_err);
2575 }
2576 
2577 static void spapr_init_cpus(SpaprMachineState *spapr)
2578 {
2579     MachineState *machine = MACHINE(spapr);
2580     MachineClass *mc = MACHINE_GET_CLASS(machine);
2581     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2582     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2583     const CPUArchIdList *possible_cpus;
2584     unsigned int smp_cpus = machine->smp.cpus;
2585     unsigned int smp_threads = machine->smp.threads;
2586     unsigned int max_cpus = machine->smp.max_cpus;
2587     int boot_cores_nr = smp_cpus / smp_threads;
2588     int i;
2589 
2590     possible_cpus = mc->possible_cpu_arch_ids(machine);
2591     if (mc->has_hotpluggable_cpus) {
2592         if (smp_cpus % smp_threads) {
2593             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2594                          smp_cpus, smp_threads);
2595             exit(1);
2596         }
2597         if (max_cpus % smp_threads) {
2598             error_report("max_cpus (%u) must be multiple of threads (%u)",
2599                          max_cpus, smp_threads);
2600             exit(1);
2601         }
2602     } else {
2603         if (max_cpus != smp_cpus) {
2604             error_report("This machine version does not support CPU hotplug");
2605             exit(1);
2606         }
2607         boot_cores_nr = possible_cpus->len;
2608     }
2609 
2610     if (smc->pre_2_10_has_unused_icps) {
2611         int i;
2612 
2613         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2614             /* Dummy entries get deregistered when real ICPState objects
2615              * are registered during CPU core hotplug.
2616              */
2617             pre_2_10_vmstate_register_dummy_icp(i);
2618         }
2619     }
2620 
2621     for (i = 0; i < possible_cpus->len; i++) {
2622         int core_id = i * smp_threads;
2623 
2624         if (mc->has_hotpluggable_cpus) {
2625             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2626                                    spapr_vcpu_id(spapr, core_id));
2627         }
2628 
2629         if (i < boot_cores_nr) {
2630             Object *core  = object_new(type);
2631             int nr_threads = smp_threads;
2632 
2633             /* Handle the partially filled core for older machine types */
2634             if ((i + 1) * smp_threads >= smp_cpus) {
2635                 nr_threads = smp_cpus - i * smp_threads;
2636             }
2637 
2638             object_property_set_int(core, nr_threads, "nr-threads",
2639                                     &error_fatal);
2640             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2641                                     &error_fatal);
2642             object_property_set_bool(core, true, "realized", &error_fatal);
2643 
2644             object_unref(core);
2645         }
2646     }
2647 }
2648 
2649 static PCIHostState *spapr_create_default_phb(void)
2650 {
2651     DeviceState *dev;
2652 
2653     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2654     qdev_prop_set_uint32(dev, "index", 0);
2655     qdev_init_nofail(dev);
2656 
2657     return PCI_HOST_BRIDGE(dev);
2658 }
2659 
2660 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2661 {
2662     MachineState *machine = MACHINE(spapr);
2663     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2664     hwaddr rma_size = machine->ram_size;
2665     hwaddr node0_size = spapr_node0_size(machine);
2666 
2667     /* RMA has to fit in the first NUMA node */
2668     rma_size = MIN(rma_size, node0_size);
2669 
2670     /*
2671      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2672      * never exceed that
2673      */
2674     rma_size = MIN(rma_size, 1 * TiB);
2675 
2676     /*
2677      * Clamp the RMA size based on machine type.  This is for
2678      * migration compatibility with older qemu versions, which limited
2679      * the RMA size for complicated and mostly bad reasons.
2680      */
2681     if (smc->rma_limit) {
2682         rma_size = MIN(rma_size, smc->rma_limit);
2683     }
2684 
2685     if (rma_size < MIN_RMA_SLOF) {
2686         error_setg(errp,
2687                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2688                    "ldMiB guest RMA (Real Mode Area memory)",
2689                    MIN_RMA_SLOF / MiB);
2690         return 0;
2691     }
2692 
2693     return rma_size;
2694 }
2695 
2696 /* pSeries LPAR / sPAPR hardware init */
2697 static void spapr_machine_init(MachineState *machine)
2698 {
2699     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2700     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2701     MachineClass *mc = MACHINE_GET_CLASS(machine);
2702     const char *kernel_filename = machine->kernel_filename;
2703     const char *initrd_filename = machine->initrd_filename;
2704     PCIHostState *phb;
2705     int i;
2706     MemoryRegion *sysmem = get_system_memory();
2707     long load_limit, fw_size;
2708     char *filename;
2709     Error *resize_hpt_err = NULL;
2710 
2711     msi_nonbroken = true;
2712 
2713     QLIST_INIT(&spapr->phbs);
2714     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2715 
2716     /* Determine capabilities to run with */
2717     spapr_caps_init(spapr);
2718 
2719     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2720     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2721         /*
2722          * If the user explicitly requested a mode we should either
2723          * supply it, or fail completely (which we do below).  But if
2724          * it's not set explicitly, we reset our mode to something
2725          * that works
2726          */
2727         if (resize_hpt_err) {
2728             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2729             error_free(resize_hpt_err);
2730             resize_hpt_err = NULL;
2731         } else {
2732             spapr->resize_hpt = smc->resize_hpt_default;
2733         }
2734     }
2735 
2736     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2737 
2738     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2739         /*
2740          * User requested HPT resize, but this host can't supply it.  Bail out
2741          */
2742         error_report_err(resize_hpt_err);
2743         exit(1);
2744     }
2745 
2746     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2747 
2748     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2749     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2750 
2751     /*
2752      * VSMT must be set in order to be able to compute VCPU ids, ie to
2753      * call spapr_max_server_number() or spapr_vcpu_id().
2754      */
2755     spapr_set_vsmt_mode(spapr, &error_fatal);
2756 
2757     /* Set up Interrupt Controller before we create the VCPUs */
2758     spapr_irq_init(spapr, &error_fatal);
2759 
2760     /* Set up containers for ibm,client-architecture-support negotiated options
2761      */
2762     spapr->ov5 = spapr_ovec_new();
2763     spapr->ov5_cas = spapr_ovec_new();
2764 
2765     if (smc->dr_lmb_enabled) {
2766         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2767         spapr_validate_node_memory(machine, &error_fatal);
2768     }
2769 
2770     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2771 
2772     /* advertise support for dedicated HP event source to guests */
2773     if (spapr->use_hotplug_event_source) {
2774         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2775     }
2776 
2777     /* advertise support for HPT resizing */
2778     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2779         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2780     }
2781 
2782     /* advertise support for ibm,dyamic-memory-v2 */
2783     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2784 
2785     /* advertise XIVE on POWER9 machines */
2786     if (spapr->irq->xive) {
2787         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2788     }
2789 
2790     /* init CPUs */
2791     spapr_init_cpus(spapr);
2792 
2793     /*
2794      * check we don't have a memory-less/cpu-less NUMA node
2795      * Firmware relies on the existing memory/cpu topology to provide the
2796      * NUMA topology to the kernel.
2797      * And the linux kernel needs to know the NUMA topology at start
2798      * to be able to hotplug CPUs later.
2799      */
2800     if (machine->numa_state->num_nodes) {
2801         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2802             /* check for memory-less node */
2803             if (machine->numa_state->nodes[i].node_mem == 0) {
2804                 CPUState *cs;
2805                 int found = 0;
2806                 /* check for cpu-less node */
2807                 CPU_FOREACH(cs) {
2808                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2809                     if (cpu->node_id == i) {
2810                         found = 1;
2811                         break;
2812                     }
2813                 }
2814                 /* memory-less and cpu-less node */
2815                 if (!found) {
2816                     error_report(
2817                        "Memory-less/cpu-less nodes are not supported (node %d)",
2818                                  i);
2819                     exit(1);
2820                 }
2821             }
2822         }
2823 
2824     }
2825 
2826     /*
2827      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2828      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2829      * called from vPHB reset handler so we initialize the counter here.
2830      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2831      * must be equally distant from any other node.
2832      * The final value of spapr->gpu_numa_id is going to be written to
2833      * max-associativity-domains in spapr_build_fdt().
2834      */
2835     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2836 
2837     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2838         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2839                               spapr->max_compat_pvr)) {
2840         /* KVM and TCG always allow GTSE with radix... */
2841         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2842     }
2843     /* ... but not with hash (currently). */
2844 
2845     if (kvm_enabled()) {
2846         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2847         kvmppc_enable_logical_ci_hcalls();
2848         kvmppc_enable_set_mode_hcall();
2849 
2850         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2851         kvmppc_enable_clear_ref_mod_hcalls();
2852 
2853         /* Enable H_PAGE_INIT */
2854         kvmppc_enable_h_page_init();
2855     }
2856 
2857     /* map RAM */
2858     memory_region_add_subregion(sysmem, 0, machine->ram);
2859 
2860     /* always allocate the device memory information */
2861     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2862 
2863     /* initialize hotplug memory address space */
2864     if (machine->ram_size < machine->maxram_size) {
2865         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2866         /*
2867          * Limit the number of hotpluggable memory slots to half the number
2868          * slots that KVM supports, leaving the other half for PCI and other
2869          * devices. However ensure that number of slots doesn't drop below 32.
2870          */
2871         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2872                            SPAPR_MAX_RAM_SLOTS;
2873 
2874         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2875             max_memslots = SPAPR_MAX_RAM_SLOTS;
2876         }
2877         if (machine->ram_slots > max_memslots) {
2878             error_report("Specified number of memory slots %"
2879                          PRIu64" exceeds max supported %d",
2880                          machine->ram_slots, max_memslots);
2881             exit(1);
2882         }
2883 
2884         machine->device_memory->base = ROUND_UP(machine->ram_size,
2885                                                 SPAPR_DEVICE_MEM_ALIGN);
2886         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2887                            "device-memory", device_mem_size);
2888         memory_region_add_subregion(sysmem, machine->device_memory->base,
2889                                     &machine->device_memory->mr);
2890     }
2891 
2892     if (smc->dr_lmb_enabled) {
2893         spapr_create_lmb_dr_connectors(spapr);
2894     }
2895 
2896     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2897         /* Create the error string for live migration blocker */
2898         error_setg(&spapr->fwnmi_migration_blocker,
2899             "A machine check is being handled during migration. The handler"
2900             "may run and log hardware error on the destination");
2901     }
2902 
2903     if (mc->nvdimm_supported) {
2904         spapr_create_nvdimm_dr_connectors(spapr);
2905     }
2906 
2907     /* Set up RTAS event infrastructure */
2908     spapr_events_init(spapr);
2909 
2910     /* Set up the RTC RTAS interfaces */
2911     spapr_rtc_create(spapr);
2912 
2913     /* Set up VIO bus */
2914     spapr->vio_bus = spapr_vio_bus_init();
2915 
2916     for (i = 0; i < serial_max_hds(); i++) {
2917         if (serial_hd(i)) {
2918             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2919         }
2920     }
2921 
2922     /* We always have at least the nvram device on VIO */
2923     spapr_create_nvram(spapr);
2924 
2925     /*
2926      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2927      * connectors (described in root DT node's "ibm,drc-types" property)
2928      * are pre-initialized here. additional child connectors (such as
2929      * connectors for a PHBs PCI slots) are added as needed during their
2930      * parent's realization.
2931      */
2932     if (smc->dr_phb_enabled) {
2933         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2934             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2935         }
2936     }
2937 
2938     /* Set up PCI */
2939     spapr_pci_rtas_init();
2940 
2941     phb = spapr_create_default_phb();
2942 
2943     for (i = 0; i < nb_nics; i++) {
2944         NICInfo *nd = &nd_table[i];
2945 
2946         if (!nd->model) {
2947             nd->model = g_strdup("spapr-vlan");
2948         }
2949 
2950         if (g_str_equal(nd->model, "spapr-vlan") ||
2951             g_str_equal(nd->model, "ibmveth")) {
2952             spapr_vlan_create(spapr->vio_bus, nd);
2953         } else {
2954             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2955         }
2956     }
2957 
2958     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2959         spapr_vscsi_create(spapr->vio_bus);
2960     }
2961 
2962     /* Graphics */
2963     if (spapr_vga_init(phb->bus, &error_fatal)) {
2964         spapr->has_graphics = true;
2965         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2966     }
2967 
2968     if (machine->usb) {
2969         if (smc->use_ohci_by_default) {
2970             pci_create_simple(phb->bus, -1, "pci-ohci");
2971         } else {
2972             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2973         }
2974 
2975         if (spapr->has_graphics) {
2976             USBBus *usb_bus = usb_bus_find(-1);
2977 
2978             usb_create_simple(usb_bus, "usb-kbd");
2979             usb_create_simple(usb_bus, "usb-mouse");
2980         }
2981     }
2982 
2983     if (kernel_filename) {
2984         uint64_t lowaddr = 0;
2985 
2986         spapr->kernel_size = load_elf(kernel_filename, NULL,
2987                                       translate_kernel_address, spapr,
2988                                       NULL, &lowaddr, NULL, NULL, 1,
2989                                       PPC_ELF_MACHINE, 0, 0);
2990         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2991             spapr->kernel_size = load_elf(kernel_filename, NULL,
2992                                           translate_kernel_address, spapr, NULL,
2993                                           &lowaddr, NULL, NULL, 0,
2994                                           PPC_ELF_MACHINE,
2995                                           0, 0);
2996             spapr->kernel_le = spapr->kernel_size > 0;
2997         }
2998         if (spapr->kernel_size < 0) {
2999             error_report("error loading %s: %s", kernel_filename,
3000                          load_elf_strerror(spapr->kernel_size));
3001             exit(1);
3002         }
3003 
3004         /* load initrd */
3005         if (initrd_filename) {
3006             /* Try to locate the initrd in the gap between the kernel
3007              * and the firmware. Add a bit of space just in case
3008              */
3009             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3010                                   + 0x1ffff) & ~0xffff;
3011             spapr->initrd_size = load_image_targphys(initrd_filename,
3012                                                      spapr->initrd_base,
3013                                                      load_limit
3014                                                      - spapr->initrd_base);
3015             if (spapr->initrd_size < 0) {
3016                 error_report("could not load initial ram disk '%s'",
3017                              initrd_filename);
3018                 exit(1);
3019             }
3020         }
3021     }
3022 
3023     if (bios_name == NULL) {
3024         bios_name = FW_FILE_NAME;
3025     }
3026     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3027     if (!filename) {
3028         error_report("Could not find LPAR firmware '%s'", bios_name);
3029         exit(1);
3030     }
3031     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3032     if (fw_size <= 0) {
3033         error_report("Could not load LPAR firmware '%s'", filename);
3034         exit(1);
3035     }
3036     g_free(filename);
3037 
3038     /* FIXME: Should register things through the MachineState's qdev
3039      * interface, this is a legacy from the sPAPREnvironment structure
3040      * which predated MachineState but had a similar function */
3041     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3042     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3043                          &savevm_htab_handlers, spapr);
3044 
3045     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3046                              &error_fatal);
3047 
3048     qemu_register_boot_set(spapr_boot_set, spapr);
3049 
3050     /*
3051      * Nothing needs to be done to resume a suspended guest because
3052      * suspending does not change the machine state, so no need for
3053      * a ->wakeup method.
3054      */
3055     qemu_register_wakeup_support();
3056 
3057     if (kvm_enabled()) {
3058         /* to stop and start vmclock */
3059         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3060                                          &spapr->tb);
3061 
3062         kvmppc_spapr_enable_inkernel_multitce();
3063     }
3064 
3065     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3066 }
3067 
3068 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3069 {
3070     if (!vm_type) {
3071         return 0;
3072     }
3073 
3074     if (!strcmp(vm_type, "HV")) {
3075         return 1;
3076     }
3077 
3078     if (!strcmp(vm_type, "PR")) {
3079         return 2;
3080     }
3081 
3082     error_report("Unknown kvm-type specified '%s'", vm_type);
3083     exit(1);
3084 }
3085 
3086 /*
3087  * Implementation of an interface to adjust firmware path
3088  * for the bootindex property handling.
3089  */
3090 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3091                                    DeviceState *dev)
3092 {
3093 #define CAST(type, obj, name) \
3094     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3095     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3096     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3097     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3098 
3099     if (d) {
3100         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3101         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3102         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3103 
3104         if (spapr) {
3105             /*
3106              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3107              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3108              * 0x8000 | (target << 8) | (bus << 5) | lun
3109              * (see the "Logical unit addressing format" table in SAM5)
3110              */
3111             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3112             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3113                                    (uint64_t)id << 48);
3114         } else if (virtio) {
3115             /*
3116              * We use SRP luns of the form 01000000 | (target << 8) | lun
3117              * in the top 32 bits of the 64-bit LUN
3118              * Note: the quote above is from SLOF and it is wrong,
3119              * the actual binding is:
3120              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3121              */
3122             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3123             if (d->lun >= 256) {
3124                 /* Use the LUN "flat space addressing method" */
3125                 id |= 0x4000;
3126             }
3127             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3128                                    (uint64_t)id << 32);
3129         } else if (usb) {
3130             /*
3131              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3132              * in the top 32 bits of the 64-bit LUN
3133              */
3134             unsigned usb_port = atoi(usb->port->path);
3135             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3136             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3137                                    (uint64_t)id << 32);
3138         }
3139     }
3140 
3141     /*
3142      * SLOF probes the USB devices, and if it recognizes that the device is a
3143      * storage device, it changes its name to "storage" instead of "usb-host",
3144      * and additionally adds a child node for the SCSI LUN, so the correct
3145      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3146      */
3147     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3148         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3149         if (usb_host_dev_is_scsi_storage(usbdev)) {
3150             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3151         }
3152     }
3153 
3154     if (phb) {
3155         /* Replace "pci" with "pci@800000020000000" */
3156         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3157     }
3158 
3159     if (vsc) {
3160         /* Same logic as virtio above */
3161         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3162         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3163     }
3164 
3165     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3166         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3167         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3168         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3169     }
3170 
3171     return NULL;
3172 }
3173 
3174 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3175 {
3176     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3177 
3178     return g_strdup(spapr->kvm_type);
3179 }
3180 
3181 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3182 {
3183     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3184 
3185     g_free(spapr->kvm_type);
3186     spapr->kvm_type = g_strdup(value);
3187 }
3188 
3189 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3190 {
3191     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3192 
3193     return spapr->use_hotplug_event_source;
3194 }
3195 
3196 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3197                                             Error **errp)
3198 {
3199     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3200 
3201     spapr->use_hotplug_event_source = value;
3202 }
3203 
3204 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3205 {
3206     return true;
3207 }
3208 
3209 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3210 {
3211     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3212 
3213     switch (spapr->resize_hpt) {
3214     case SPAPR_RESIZE_HPT_DEFAULT:
3215         return g_strdup("default");
3216     case SPAPR_RESIZE_HPT_DISABLED:
3217         return g_strdup("disabled");
3218     case SPAPR_RESIZE_HPT_ENABLED:
3219         return g_strdup("enabled");
3220     case SPAPR_RESIZE_HPT_REQUIRED:
3221         return g_strdup("required");
3222     }
3223     g_assert_not_reached();
3224 }
3225 
3226 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3227 {
3228     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3229 
3230     if (strcmp(value, "default") == 0) {
3231         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3232     } else if (strcmp(value, "disabled") == 0) {
3233         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3234     } else if (strcmp(value, "enabled") == 0) {
3235         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3236     } else if (strcmp(value, "required") == 0) {
3237         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3238     } else {
3239         error_setg(errp, "Bad value for \"resize-hpt\" property");
3240     }
3241 }
3242 
3243 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3244 {
3245     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3246 
3247     if (spapr->irq == &spapr_irq_xics_legacy) {
3248         return g_strdup("legacy");
3249     } else if (spapr->irq == &spapr_irq_xics) {
3250         return g_strdup("xics");
3251     } else if (spapr->irq == &spapr_irq_xive) {
3252         return g_strdup("xive");
3253     } else if (spapr->irq == &spapr_irq_dual) {
3254         return g_strdup("dual");
3255     }
3256     g_assert_not_reached();
3257 }
3258 
3259 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3260 {
3261     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3262 
3263     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3264         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3265         return;
3266     }
3267 
3268     /* The legacy IRQ backend can not be set */
3269     if (strcmp(value, "xics") == 0) {
3270         spapr->irq = &spapr_irq_xics;
3271     } else if (strcmp(value, "xive") == 0) {
3272         spapr->irq = &spapr_irq_xive;
3273     } else if (strcmp(value, "dual") == 0) {
3274         spapr->irq = &spapr_irq_dual;
3275     } else {
3276         error_setg(errp, "Bad value for \"ic-mode\" property");
3277     }
3278 }
3279 
3280 static char *spapr_get_host_model(Object *obj, Error **errp)
3281 {
3282     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3283 
3284     return g_strdup(spapr->host_model);
3285 }
3286 
3287 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3288 {
3289     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3290 
3291     g_free(spapr->host_model);
3292     spapr->host_model = g_strdup(value);
3293 }
3294 
3295 static char *spapr_get_host_serial(Object *obj, Error **errp)
3296 {
3297     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3298 
3299     return g_strdup(spapr->host_serial);
3300 }
3301 
3302 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3303 {
3304     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3305 
3306     g_free(spapr->host_serial);
3307     spapr->host_serial = g_strdup(value);
3308 }
3309 
3310 static void spapr_instance_init(Object *obj)
3311 {
3312     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3313     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3314 
3315     spapr->htab_fd = -1;
3316     spapr->use_hotplug_event_source = true;
3317     object_property_add_str(obj, "kvm-type",
3318                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3319     object_property_set_description(obj, "kvm-type",
3320                                     "Specifies the KVM virtualization mode (HV, PR)",
3321                                     NULL);
3322     object_property_add_bool(obj, "modern-hotplug-events",
3323                             spapr_get_modern_hotplug_events,
3324                             spapr_set_modern_hotplug_events,
3325                             NULL);
3326     object_property_set_description(obj, "modern-hotplug-events",
3327                                     "Use dedicated hotplug event mechanism in"
3328                                     " place of standard EPOW events when possible"
3329                                     " (required for memory hot-unplug support)",
3330                                     NULL);
3331     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3332                             "Maximum permitted CPU compatibility mode",
3333                             &error_fatal);
3334 
3335     object_property_add_str(obj, "resize-hpt",
3336                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3337     object_property_set_description(obj, "resize-hpt",
3338                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3339                                     NULL);
3340     object_property_add_uint32_ptr(obj, "vsmt",
3341                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE,
3342                                    &error_abort);
3343     object_property_set_description(obj, "vsmt",
3344                                     "Virtual SMT: KVM behaves as if this were"
3345                                     " the host's SMT mode", &error_abort);
3346 
3347     object_property_add_bool(obj, "vfio-no-msix-emulation",
3348                              spapr_get_msix_emulation, NULL, NULL);
3349 
3350     object_property_add_uint64_ptr(obj, "kernel-addr",
3351                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE,
3352                                    &error_abort);
3353     object_property_set_description(obj, "kernel-addr",
3354                                     stringify(KERNEL_LOAD_ADDR)
3355                                     " for -kernel is the default",
3356                                     NULL);
3357     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3358     /* The machine class defines the default interrupt controller mode */
3359     spapr->irq = smc->irq;
3360     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3361                             spapr_set_ic_mode, NULL);
3362     object_property_set_description(obj, "ic-mode",
3363                  "Specifies the interrupt controller mode (xics, xive, dual)",
3364                  NULL);
3365 
3366     object_property_add_str(obj, "host-model",
3367         spapr_get_host_model, spapr_set_host_model,
3368         &error_abort);
3369     object_property_set_description(obj, "host-model",
3370         "Host model to advertise in guest device tree", &error_abort);
3371     object_property_add_str(obj, "host-serial",
3372         spapr_get_host_serial, spapr_set_host_serial,
3373         &error_abort);
3374     object_property_set_description(obj, "host-serial",
3375         "Host serial number to advertise in guest device tree", &error_abort);
3376 }
3377 
3378 static void spapr_machine_finalizefn(Object *obj)
3379 {
3380     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3381 
3382     g_free(spapr->kvm_type);
3383 }
3384 
3385 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3386 {
3387     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3388 
3389     cpu_synchronize_state(cs);
3390     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3391     if (spapr->fwnmi_system_reset_addr != -1) {
3392         uint64_t rtas_addr, addr;
3393         PowerPCCPU *cpu = POWERPC_CPU(cs);
3394         CPUPPCState *env = &cpu->env;
3395 
3396         /* get rtas addr from fdt */
3397         rtas_addr = spapr_get_rtas_addr();
3398         if (!rtas_addr) {
3399             qemu_system_guest_panicked(NULL);
3400             return;
3401         }
3402 
3403         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3404         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3405         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3406         env->gpr[3] = addr;
3407     }
3408     ppc_cpu_do_system_reset(cs, spapr->fwnmi_system_reset_addr);
3409 }
3410 
3411 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3412 {
3413     CPUState *cs;
3414 
3415     CPU_FOREACH(cs) {
3416         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3417     }
3418 }
3419 
3420 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3421                           void *fdt, int *fdt_start_offset, Error **errp)
3422 {
3423     uint64_t addr;
3424     uint32_t node;
3425 
3426     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3427     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3428                                     &error_abort);
3429     *fdt_start_offset = spapr_dt_memory_node(fdt, node, addr,
3430                                              SPAPR_MEMORY_BLOCK_SIZE);
3431     return 0;
3432 }
3433 
3434 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3435                            bool dedicated_hp_event_source, Error **errp)
3436 {
3437     SpaprDrc *drc;
3438     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3439     int i;
3440     uint64_t addr = addr_start;
3441     bool hotplugged = spapr_drc_hotplugged(dev);
3442     Error *local_err = NULL;
3443 
3444     for (i = 0; i < nr_lmbs; i++) {
3445         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3446                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3447         g_assert(drc);
3448 
3449         spapr_drc_attach(drc, dev, &local_err);
3450         if (local_err) {
3451             while (addr > addr_start) {
3452                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3453                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3454                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3455                 spapr_drc_detach(drc);
3456             }
3457             error_propagate(errp, local_err);
3458             return;
3459         }
3460         if (!hotplugged) {
3461             spapr_drc_reset(drc);
3462         }
3463         addr += SPAPR_MEMORY_BLOCK_SIZE;
3464     }
3465     /* send hotplug notification to the
3466      * guest only in case of hotplugged memory
3467      */
3468     if (hotplugged) {
3469         if (dedicated_hp_event_source) {
3470             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3471                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3472             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3473                                                    nr_lmbs,
3474                                                    spapr_drc_index(drc));
3475         } else {
3476             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3477                                            nr_lmbs);
3478         }
3479     }
3480 }
3481 
3482 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3483                               Error **errp)
3484 {
3485     Error *local_err = NULL;
3486     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3487     PCDIMMDevice *dimm = PC_DIMM(dev);
3488     uint64_t size, addr, slot;
3489     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3490 
3491     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3492 
3493     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3494     if (local_err) {
3495         goto out;
3496     }
3497 
3498     if (!is_nvdimm) {
3499         addr = object_property_get_uint(OBJECT(dimm),
3500                                         PC_DIMM_ADDR_PROP, &local_err);
3501         if (local_err) {
3502             goto out_unplug;
3503         }
3504         spapr_add_lmbs(dev, addr, size,
3505                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3506                        &local_err);
3507     } else {
3508         slot = object_property_get_uint(OBJECT(dimm),
3509                                         PC_DIMM_SLOT_PROP, &local_err);
3510         if (local_err) {
3511             goto out_unplug;
3512         }
3513         spapr_add_nvdimm(dev, slot, &local_err);
3514     }
3515 
3516     if (local_err) {
3517         goto out_unplug;
3518     }
3519 
3520     return;
3521 
3522 out_unplug:
3523     pc_dimm_unplug(dimm, MACHINE(ms));
3524 out:
3525     error_propagate(errp, local_err);
3526 }
3527 
3528 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3529                                   Error **errp)
3530 {
3531     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3532     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3533     const MachineClass *mc = MACHINE_CLASS(smc);
3534     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3535     PCDIMMDevice *dimm = PC_DIMM(dev);
3536     Error *local_err = NULL;
3537     uint64_t size;
3538     Object *memdev;
3539     hwaddr pagesize;
3540 
3541     if (!smc->dr_lmb_enabled) {
3542         error_setg(errp, "Memory hotplug not supported for this machine");
3543         return;
3544     }
3545 
3546     if (is_nvdimm && !mc->nvdimm_supported) {
3547         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3548         return;
3549     }
3550 
3551     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3552     if (local_err) {
3553         error_propagate(errp, local_err);
3554         return;
3555     }
3556 
3557     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3558         error_setg(errp, "Hotplugged memory size must be a multiple of "
3559                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3560         return;
3561     } else if (is_nvdimm) {
3562         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3563         if (local_err) {
3564             error_propagate(errp, local_err);
3565             return;
3566         }
3567     }
3568 
3569     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3570                                       &error_abort);
3571     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3572     spapr_check_pagesize(spapr, pagesize, &local_err);
3573     if (local_err) {
3574         error_propagate(errp, local_err);
3575         return;
3576     }
3577 
3578     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3579 }
3580 
3581 struct SpaprDimmState {
3582     PCDIMMDevice *dimm;
3583     uint32_t nr_lmbs;
3584     QTAILQ_ENTRY(SpaprDimmState) next;
3585 };
3586 
3587 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3588                                                        PCDIMMDevice *dimm)
3589 {
3590     SpaprDimmState *dimm_state = NULL;
3591 
3592     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3593         if (dimm_state->dimm == dimm) {
3594             break;
3595         }
3596     }
3597     return dimm_state;
3598 }
3599 
3600 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3601                                                       uint32_t nr_lmbs,
3602                                                       PCDIMMDevice *dimm)
3603 {
3604     SpaprDimmState *ds = NULL;
3605 
3606     /*
3607      * If this request is for a DIMM whose removal had failed earlier
3608      * (due to guest's refusal to remove the LMBs), we would have this
3609      * dimm already in the pending_dimm_unplugs list. In that
3610      * case don't add again.
3611      */
3612     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3613     if (!ds) {
3614         ds = g_malloc0(sizeof(SpaprDimmState));
3615         ds->nr_lmbs = nr_lmbs;
3616         ds->dimm = dimm;
3617         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3618     }
3619     return ds;
3620 }
3621 
3622 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3623                                               SpaprDimmState *dimm_state)
3624 {
3625     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3626     g_free(dimm_state);
3627 }
3628 
3629 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3630                                                         PCDIMMDevice *dimm)
3631 {
3632     SpaprDrc *drc;
3633     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3634                                                   &error_abort);
3635     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3636     uint32_t avail_lmbs = 0;
3637     uint64_t addr_start, addr;
3638     int i;
3639 
3640     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3641                                          &error_abort);
3642 
3643     addr = addr_start;
3644     for (i = 0; i < nr_lmbs; i++) {
3645         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3646                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3647         g_assert(drc);
3648         if (drc->dev) {
3649             avail_lmbs++;
3650         }
3651         addr += SPAPR_MEMORY_BLOCK_SIZE;
3652     }
3653 
3654     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3655 }
3656 
3657 /* Callback to be called during DRC release. */
3658 void spapr_lmb_release(DeviceState *dev)
3659 {
3660     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3661     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3662     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3663 
3664     /* This information will get lost if a migration occurs
3665      * during the unplug process. In this case recover it. */
3666     if (ds == NULL) {
3667         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3668         g_assert(ds);
3669         /* The DRC being examined by the caller at least must be counted */
3670         g_assert(ds->nr_lmbs);
3671     }
3672 
3673     if (--ds->nr_lmbs) {
3674         return;
3675     }
3676 
3677     /*
3678      * Now that all the LMBs have been removed by the guest, call the
3679      * unplug handler chain. This can never fail.
3680      */
3681     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3682     object_unparent(OBJECT(dev));
3683 }
3684 
3685 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3686 {
3687     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3688     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3689 
3690     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3691     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3692     spapr_pending_dimm_unplugs_remove(spapr, ds);
3693 }
3694 
3695 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3696                                         DeviceState *dev, Error **errp)
3697 {
3698     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3699     Error *local_err = NULL;
3700     PCDIMMDevice *dimm = PC_DIMM(dev);
3701     uint32_t nr_lmbs;
3702     uint64_t size, addr_start, addr;
3703     int i;
3704     SpaprDrc *drc;
3705 
3706     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3707         error_setg(&local_err,
3708                    "nvdimm device hot unplug is not supported yet.");
3709         goto out;
3710     }
3711 
3712     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3713     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3714 
3715     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3716                                          &local_err);
3717     if (local_err) {
3718         goto out;
3719     }
3720 
3721     /*
3722      * An existing pending dimm state for this DIMM means that there is an
3723      * unplug operation in progress, waiting for the spapr_lmb_release
3724      * callback to complete the job (BQL can't cover that far). In this case,
3725      * bail out to avoid detaching DRCs that were already released.
3726      */
3727     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3728         error_setg(&local_err,
3729                    "Memory unplug already in progress for device %s",
3730                    dev->id);
3731         goto out;
3732     }
3733 
3734     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3735 
3736     addr = addr_start;
3737     for (i = 0; i < nr_lmbs; i++) {
3738         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3739                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3740         g_assert(drc);
3741 
3742         spapr_drc_detach(drc);
3743         addr += SPAPR_MEMORY_BLOCK_SIZE;
3744     }
3745 
3746     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3747                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3748     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3749                                               nr_lmbs, spapr_drc_index(drc));
3750 out:
3751     error_propagate(errp, local_err);
3752 }
3753 
3754 /* Callback to be called during DRC release. */
3755 void spapr_core_release(DeviceState *dev)
3756 {
3757     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3758 
3759     /* Call the unplug handler chain. This can never fail. */
3760     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3761     object_unparent(OBJECT(dev));
3762 }
3763 
3764 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3765 {
3766     MachineState *ms = MACHINE(hotplug_dev);
3767     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3768     CPUCore *cc = CPU_CORE(dev);
3769     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3770 
3771     if (smc->pre_2_10_has_unused_icps) {
3772         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3773         int i;
3774 
3775         for (i = 0; i < cc->nr_threads; i++) {
3776             CPUState *cs = CPU(sc->threads[i]);
3777 
3778             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3779         }
3780     }
3781 
3782     assert(core_slot);
3783     core_slot->cpu = NULL;
3784     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3785 }
3786 
3787 static
3788 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3789                                Error **errp)
3790 {
3791     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3792     int index;
3793     SpaprDrc *drc;
3794     CPUCore *cc = CPU_CORE(dev);
3795 
3796     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3797         error_setg(errp, "Unable to find CPU core with core-id: %d",
3798                    cc->core_id);
3799         return;
3800     }
3801     if (index == 0) {
3802         error_setg(errp, "Boot CPU core may not be unplugged");
3803         return;
3804     }
3805 
3806     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3807                           spapr_vcpu_id(spapr, cc->core_id));
3808     g_assert(drc);
3809 
3810     if (!spapr_drc_unplug_requested(drc)) {
3811         spapr_drc_detach(drc);
3812         spapr_hotplug_req_remove_by_index(drc);
3813     }
3814 }
3815 
3816 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3817                            void *fdt, int *fdt_start_offset, Error **errp)
3818 {
3819     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3820     CPUState *cs = CPU(core->threads[0]);
3821     PowerPCCPU *cpu = POWERPC_CPU(cs);
3822     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3823     int id = spapr_get_vcpu_id(cpu);
3824     char *nodename;
3825     int offset;
3826 
3827     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3828     offset = fdt_add_subnode(fdt, 0, nodename);
3829     g_free(nodename);
3830 
3831     spapr_dt_cpu(cs, fdt, offset, spapr);
3832 
3833     *fdt_start_offset = offset;
3834     return 0;
3835 }
3836 
3837 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3838                             Error **errp)
3839 {
3840     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3841     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3842     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3843     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3844     CPUCore *cc = CPU_CORE(dev);
3845     CPUState *cs;
3846     SpaprDrc *drc;
3847     Error *local_err = NULL;
3848     CPUArchId *core_slot;
3849     int index;
3850     bool hotplugged = spapr_drc_hotplugged(dev);
3851     int i;
3852 
3853     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3854     if (!core_slot) {
3855         error_setg(errp, "Unable to find CPU core with core-id: %d",
3856                    cc->core_id);
3857         return;
3858     }
3859     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3860                           spapr_vcpu_id(spapr, cc->core_id));
3861 
3862     g_assert(drc || !mc->has_hotpluggable_cpus);
3863 
3864     if (drc) {
3865         spapr_drc_attach(drc, dev, &local_err);
3866         if (local_err) {
3867             error_propagate(errp, local_err);
3868             return;
3869         }
3870 
3871         if (hotplugged) {
3872             /*
3873              * Send hotplug notification interrupt to the guest only
3874              * in case of hotplugged CPUs.
3875              */
3876             spapr_hotplug_req_add_by_index(drc);
3877         } else {
3878             spapr_drc_reset(drc);
3879         }
3880     }
3881 
3882     core_slot->cpu = OBJECT(dev);
3883 
3884     if (smc->pre_2_10_has_unused_icps) {
3885         for (i = 0; i < cc->nr_threads; i++) {
3886             cs = CPU(core->threads[i]);
3887             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3888         }
3889     }
3890 
3891     /*
3892      * Set compatibility mode to match the boot CPU, which was either set
3893      * by the machine reset code or by CAS.
3894      */
3895     if (hotplugged) {
3896         for (i = 0; i < cc->nr_threads; i++) {
3897             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3898                            &local_err);
3899             if (local_err) {
3900                 error_propagate(errp, local_err);
3901                 return;
3902             }
3903         }
3904     }
3905 }
3906 
3907 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3908                                 Error **errp)
3909 {
3910     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3911     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3912     Error *local_err = NULL;
3913     CPUCore *cc = CPU_CORE(dev);
3914     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3915     const char *type = object_get_typename(OBJECT(dev));
3916     CPUArchId *core_slot;
3917     int index;
3918     unsigned int smp_threads = machine->smp.threads;
3919 
3920     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3921         error_setg(&local_err, "CPU hotplug not supported for this machine");
3922         goto out;
3923     }
3924 
3925     if (strcmp(base_core_type, type)) {
3926         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3927         goto out;
3928     }
3929 
3930     if (cc->core_id % smp_threads) {
3931         error_setg(&local_err, "invalid core id %d", cc->core_id);
3932         goto out;
3933     }
3934 
3935     /*
3936      * In general we should have homogeneous threads-per-core, but old
3937      * (pre hotplug support) machine types allow the last core to have
3938      * reduced threads as a compatibility hack for when we allowed
3939      * total vcpus not a multiple of threads-per-core.
3940      */
3941     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3942         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3943                    cc->nr_threads, smp_threads);
3944         goto out;
3945     }
3946 
3947     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3948     if (!core_slot) {
3949         error_setg(&local_err, "core id %d out of range", cc->core_id);
3950         goto out;
3951     }
3952 
3953     if (core_slot->cpu) {
3954         error_setg(&local_err, "core %d already populated", cc->core_id);
3955         goto out;
3956     }
3957 
3958     numa_cpu_pre_plug(core_slot, dev, &local_err);
3959 
3960 out:
3961     error_propagate(errp, local_err);
3962 }
3963 
3964 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3965                           void *fdt, int *fdt_start_offset, Error **errp)
3966 {
3967     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3968     int intc_phandle;
3969 
3970     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3971     if (intc_phandle <= 0) {
3972         return -1;
3973     }
3974 
3975     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3976         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3977         return -1;
3978     }
3979 
3980     /* generally SLOF creates these, for hotplug it's up to QEMU */
3981     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3982 
3983     return 0;
3984 }
3985 
3986 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3987                                Error **errp)
3988 {
3989     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3990     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3991     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3992     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3993 
3994     if (dev->hotplugged && !smc->dr_phb_enabled) {
3995         error_setg(errp, "PHB hotplug not supported for this machine");
3996         return;
3997     }
3998 
3999     if (sphb->index == (uint32_t)-1) {
4000         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4001         return;
4002     }
4003 
4004     /*
4005      * This will check that sphb->index doesn't exceed the maximum number of
4006      * PHBs for the current machine type.
4007      */
4008     smc->phb_placement(spapr, sphb->index,
4009                        &sphb->buid, &sphb->io_win_addr,
4010                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
4011                        windows_supported, sphb->dma_liobn,
4012                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4013                        errp);
4014 }
4015 
4016 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4017                            Error **errp)
4018 {
4019     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4020     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4021     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4022     SpaprDrc *drc;
4023     bool hotplugged = spapr_drc_hotplugged(dev);
4024     Error *local_err = NULL;
4025 
4026     if (!smc->dr_phb_enabled) {
4027         return;
4028     }
4029 
4030     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4031     /* hotplug hooks should check it's enabled before getting this far */
4032     assert(drc);
4033 
4034     spapr_drc_attach(drc, DEVICE(dev), &local_err);
4035     if (local_err) {
4036         error_propagate(errp, local_err);
4037         return;
4038     }
4039 
4040     if (hotplugged) {
4041         spapr_hotplug_req_add_by_index(drc);
4042     } else {
4043         spapr_drc_reset(drc);
4044     }
4045 }
4046 
4047 void spapr_phb_release(DeviceState *dev)
4048 {
4049     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4050 
4051     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4052     object_unparent(OBJECT(dev));
4053 }
4054 
4055 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4056 {
4057     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4058 }
4059 
4060 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4061                                      DeviceState *dev, Error **errp)
4062 {
4063     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4064     SpaprDrc *drc;
4065 
4066     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4067     assert(drc);
4068 
4069     if (!spapr_drc_unplug_requested(drc)) {
4070         spapr_drc_detach(drc);
4071         spapr_hotplug_req_remove_by_index(drc);
4072     }
4073 }
4074 
4075 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4076                                  Error **errp)
4077 {
4078     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4079     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4080 
4081     if (spapr->tpm_proxy != NULL) {
4082         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4083         return;
4084     }
4085 
4086     spapr->tpm_proxy = tpm_proxy;
4087 }
4088 
4089 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4090 {
4091     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4092 
4093     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4094     object_unparent(OBJECT(dev));
4095     spapr->tpm_proxy = NULL;
4096 }
4097 
4098 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4099                                       DeviceState *dev, Error **errp)
4100 {
4101     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4102         spapr_memory_plug(hotplug_dev, dev, errp);
4103     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4104         spapr_core_plug(hotplug_dev, dev, errp);
4105     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4106         spapr_phb_plug(hotplug_dev, dev, errp);
4107     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4108         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4109     }
4110 }
4111 
4112 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4113                                         DeviceState *dev, Error **errp)
4114 {
4115     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4116         spapr_memory_unplug(hotplug_dev, dev);
4117     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4118         spapr_core_unplug(hotplug_dev, dev);
4119     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4120         spapr_phb_unplug(hotplug_dev, dev);
4121     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4122         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4123     }
4124 }
4125 
4126 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4127                                                 DeviceState *dev, Error **errp)
4128 {
4129     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4130     MachineClass *mc = MACHINE_GET_CLASS(sms);
4131     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4132 
4133     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4134         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4135             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4136         } else {
4137             /* NOTE: this means there is a window after guest reset, prior to
4138              * CAS negotiation, where unplug requests will fail due to the
4139              * capability not being detected yet. This is a bit different than
4140              * the case with PCI unplug, where the events will be queued and
4141              * eventually handled by the guest after boot
4142              */
4143             error_setg(errp, "Memory hot unplug not supported for this guest");
4144         }
4145     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4146         if (!mc->has_hotpluggable_cpus) {
4147             error_setg(errp, "CPU hot unplug not supported on this machine");
4148             return;
4149         }
4150         spapr_core_unplug_request(hotplug_dev, dev, errp);
4151     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4152         if (!smc->dr_phb_enabled) {
4153             error_setg(errp, "PHB hot unplug not supported on this machine");
4154             return;
4155         }
4156         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4157     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4158         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4159     }
4160 }
4161 
4162 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4163                                           DeviceState *dev, Error **errp)
4164 {
4165     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4166         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4167     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4168         spapr_core_pre_plug(hotplug_dev, dev, errp);
4169     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4170         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4171     }
4172 }
4173 
4174 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4175                                                  DeviceState *dev)
4176 {
4177     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4178         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4179         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4180         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4181         return HOTPLUG_HANDLER(machine);
4182     }
4183     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4184         PCIDevice *pcidev = PCI_DEVICE(dev);
4185         PCIBus *root = pci_device_root_bus(pcidev);
4186         SpaprPhbState *phb =
4187             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4188                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4189 
4190         if (phb) {
4191             return HOTPLUG_HANDLER(phb);
4192         }
4193     }
4194     return NULL;
4195 }
4196 
4197 static CpuInstanceProperties
4198 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4199 {
4200     CPUArchId *core_slot;
4201     MachineClass *mc = MACHINE_GET_CLASS(machine);
4202 
4203     /* make sure possible_cpu are intialized */
4204     mc->possible_cpu_arch_ids(machine);
4205     /* get CPU core slot containing thread that matches cpu_index */
4206     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4207     assert(core_slot);
4208     return core_slot->props;
4209 }
4210 
4211 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4212 {
4213     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4214 }
4215 
4216 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4217 {
4218     int i;
4219     unsigned int smp_threads = machine->smp.threads;
4220     unsigned int smp_cpus = machine->smp.cpus;
4221     const char *core_type;
4222     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4223     MachineClass *mc = MACHINE_GET_CLASS(machine);
4224 
4225     if (!mc->has_hotpluggable_cpus) {
4226         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4227     }
4228     if (machine->possible_cpus) {
4229         assert(machine->possible_cpus->len == spapr_max_cores);
4230         return machine->possible_cpus;
4231     }
4232 
4233     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4234     if (!core_type) {
4235         error_report("Unable to find sPAPR CPU Core definition");
4236         exit(1);
4237     }
4238 
4239     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4240                              sizeof(CPUArchId) * spapr_max_cores);
4241     machine->possible_cpus->len = spapr_max_cores;
4242     for (i = 0; i < machine->possible_cpus->len; i++) {
4243         int core_id = i * smp_threads;
4244 
4245         machine->possible_cpus->cpus[i].type = core_type;
4246         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4247         machine->possible_cpus->cpus[i].arch_id = core_id;
4248         machine->possible_cpus->cpus[i].props.has_core_id = true;
4249         machine->possible_cpus->cpus[i].props.core_id = core_id;
4250     }
4251     return machine->possible_cpus;
4252 }
4253 
4254 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4255                                 uint64_t *buid, hwaddr *pio,
4256                                 hwaddr *mmio32, hwaddr *mmio64,
4257                                 unsigned n_dma, uint32_t *liobns,
4258                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4259 {
4260     /*
4261      * New-style PHB window placement.
4262      *
4263      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4264      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4265      * windows.
4266      *
4267      * Some guest kernels can't work with MMIO windows above 1<<46
4268      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4269      *
4270      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4271      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4272      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4273      * 1TiB 64-bit MMIO windows for each PHB.
4274      */
4275     const uint64_t base_buid = 0x800000020000000ULL;
4276     int i;
4277 
4278     /* Sanity check natural alignments */
4279     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4280     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4281     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4282     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4283     /* Sanity check bounds */
4284     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4285                       SPAPR_PCI_MEM32_WIN_SIZE);
4286     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4287                       SPAPR_PCI_MEM64_WIN_SIZE);
4288 
4289     if (index >= SPAPR_MAX_PHBS) {
4290         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4291                    SPAPR_MAX_PHBS - 1);
4292         return;
4293     }
4294 
4295     *buid = base_buid + index;
4296     for (i = 0; i < n_dma; ++i) {
4297         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4298     }
4299 
4300     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4301     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4302     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4303 
4304     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4305     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4306 }
4307 
4308 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4309 {
4310     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4311 
4312     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4313 }
4314 
4315 static void spapr_ics_resend(XICSFabric *dev)
4316 {
4317     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4318 
4319     ics_resend(spapr->ics);
4320 }
4321 
4322 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4323 {
4324     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4325 
4326     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4327 }
4328 
4329 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4330                                  Monitor *mon)
4331 {
4332     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4333 
4334     spapr_irq_print_info(spapr, mon);
4335     monitor_printf(mon, "irqchip: %s\n",
4336                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4337 }
4338 
4339 /*
4340  * This is a XIVE only operation
4341  */
4342 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4343                            uint8_t nvt_blk, uint32_t nvt_idx,
4344                            bool cam_ignore, uint8_t priority,
4345                            uint32_t logic_serv, XiveTCTXMatch *match)
4346 {
4347     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4348     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4349     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4350     int count;
4351 
4352     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4353                            priority, logic_serv, match);
4354     if (count < 0) {
4355         return count;
4356     }
4357 
4358     /*
4359      * When we implement the save and restore of the thread interrupt
4360      * contexts in the enter/exit CPU handlers of the machine and the
4361      * escalations in QEMU, we should be able to handle non dispatched
4362      * vCPUs.
4363      *
4364      * Until this is done, the sPAPR machine should find at least one
4365      * matching context always.
4366      */
4367     if (count == 0) {
4368         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4369                       nvt_blk, nvt_idx);
4370     }
4371 
4372     return count;
4373 }
4374 
4375 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4376 {
4377     return cpu->vcpu_id;
4378 }
4379 
4380 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4381 {
4382     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4383     MachineState *ms = MACHINE(spapr);
4384     int vcpu_id;
4385 
4386     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4387 
4388     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4389         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4390         error_append_hint(errp, "Adjust the number of cpus to %d "
4391                           "or try to raise the number of threads per core\n",
4392                           vcpu_id * ms->smp.threads / spapr->vsmt);
4393         return;
4394     }
4395 
4396     cpu->vcpu_id = vcpu_id;
4397 }
4398 
4399 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4400 {
4401     CPUState *cs;
4402 
4403     CPU_FOREACH(cs) {
4404         PowerPCCPU *cpu = POWERPC_CPU(cs);
4405 
4406         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4407             return cpu;
4408         }
4409     }
4410 
4411     return NULL;
4412 }
4413 
4414 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4415 {
4416     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4417 
4418     /* These are only called by TCG, KVM maintains dispatch state */
4419 
4420     spapr_cpu->prod = false;
4421     if (spapr_cpu->vpa_addr) {
4422         CPUState *cs = CPU(cpu);
4423         uint32_t dispatch;
4424 
4425         dispatch = ldl_be_phys(cs->as,
4426                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4427         dispatch++;
4428         if ((dispatch & 1) != 0) {
4429             qemu_log_mask(LOG_GUEST_ERROR,
4430                           "VPA: incorrect dispatch counter value for "
4431                           "dispatched partition %u, correcting.\n", dispatch);
4432             dispatch++;
4433         }
4434         stl_be_phys(cs->as,
4435                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4436     }
4437 }
4438 
4439 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4440 {
4441     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4442 
4443     if (spapr_cpu->vpa_addr) {
4444         CPUState *cs = CPU(cpu);
4445         uint32_t dispatch;
4446 
4447         dispatch = ldl_be_phys(cs->as,
4448                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4449         dispatch++;
4450         if ((dispatch & 1) != 1) {
4451             qemu_log_mask(LOG_GUEST_ERROR,
4452                           "VPA: incorrect dispatch counter value for "
4453                           "preempted partition %u, correcting.\n", dispatch);
4454             dispatch++;
4455         }
4456         stl_be_phys(cs->as,
4457                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4458     }
4459 }
4460 
4461 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4462 {
4463     MachineClass *mc = MACHINE_CLASS(oc);
4464     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4465     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4466     NMIClass *nc = NMI_CLASS(oc);
4467     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4468     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4469     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4470     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4471     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4472 
4473     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4474     mc->ignore_boot_device_suffixes = true;
4475 
4476     /*
4477      * We set up the default / latest behaviour here.  The class_init
4478      * functions for the specific versioned machine types can override
4479      * these details for backwards compatibility
4480      */
4481     mc->init = spapr_machine_init;
4482     mc->reset = spapr_machine_reset;
4483     mc->block_default_type = IF_SCSI;
4484     mc->max_cpus = 1024;
4485     mc->no_parallel = 1;
4486     mc->default_boot_order = "";
4487     mc->default_ram_size = 512 * MiB;
4488     mc->default_ram_id = "ppc_spapr.ram";
4489     mc->default_display = "std";
4490     mc->kvm_type = spapr_kvm_type;
4491     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4492     mc->pci_allow_0_address = true;
4493     assert(!mc->get_hotplug_handler);
4494     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4495     hc->pre_plug = spapr_machine_device_pre_plug;
4496     hc->plug = spapr_machine_device_plug;
4497     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4498     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4499     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4500     hc->unplug_request = spapr_machine_device_unplug_request;
4501     hc->unplug = spapr_machine_device_unplug;
4502 
4503     smc->dr_lmb_enabled = true;
4504     smc->update_dt_enabled = true;
4505     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4506     mc->has_hotpluggable_cpus = true;
4507     mc->nvdimm_supported = true;
4508     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4509     fwc->get_dev_path = spapr_get_fw_dev_path;
4510     nc->nmi_monitor_handler = spapr_nmi;
4511     smc->phb_placement = spapr_phb_placement;
4512     vhc->hypercall = emulate_spapr_hypercall;
4513     vhc->hpt_mask = spapr_hpt_mask;
4514     vhc->map_hptes = spapr_map_hptes;
4515     vhc->unmap_hptes = spapr_unmap_hptes;
4516     vhc->hpte_set_c = spapr_hpte_set_c;
4517     vhc->hpte_set_r = spapr_hpte_set_r;
4518     vhc->get_pate = spapr_get_pate;
4519     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4520     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4521     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4522     xic->ics_get = spapr_ics_get;
4523     xic->ics_resend = spapr_ics_resend;
4524     xic->icp_get = spapr_icp_get;
4525     ispc->print_info = spapr_pic_print_info;
4526     /* Force NUMA node memory size to be a multiple of
4527      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4528      * in which LMBs are represented and hot-added
4529      */
4530     mc->numa_mem_align_shift = 28;
4531     mc->numa_mem_supported = true;
4532     mc->auto_enable_numa = true;
4533 
4534     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4535     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4536     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4537     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4538     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4539     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4540     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4541     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4542     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4543     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4544     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4545     spapr_caps_add_properties(smc, &error_abort);
4546     smc->irq = &spapr_irq_dual;
4547     smc->dr_phb_enabled = true;
4548     smc->linux_pci_probe = true;
4549     smc->smp_threads_vsmt = true;
4550     smc->nr_xirqs = SPAPR_NR_XIRQS;
4551     xfc->match_nvt = spapr_match_nvt;
4552 }
4553 
4554 static const TypeInfo spapr_machine_info = {
4555     .name          = TYPE_SPAPR_MACHINE,
4556     .parent        = TYPE_MACHINE,
4557     .abstract      = true,
4558     .instance_size = sizeof(SpaprMachineState),
4559     .instance_init = spapr_instance_init,
4560     .instance_finalize = spapr_machine_finalizefn,
4561     .class_size    = sizeof(SpaprMachineClass),
4562     .class_init    = spapr_machine_class_init,
4563     .interfaces = (InterfaceInfo[]) {
4564         { TYPE_FW_PATH_PROVIDER },
4565         { TYPE_NMI },
4566         { TYPE_HOTPLUG_HANDLER },
4567         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4568         { TYPE_XICS_FABRIC },
4569         { TYPE_INTERRUPT_STATS_PROVIDER },
4570         { TYPE_XIVE_FABRIC },
4571         { }
4572     },
4573 };
4574 
4575 static void spapr_machine_latest_class_options(MachineClass *mc)
4576 {
4577     mc->alias = "pseries";
4578     mc->is_default = true;
4579 }
4580 
4581 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4582     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4583                                                     void *data)      \
4584     {                                                                \
4585         MachineClass *mc = MACHINE_CLASS(oc);                        \
4586         spapr_machine_##suffix##_class_options(mc);                  \
4587         if (latest) {                                                \
4588             spapr_machine_latest_class_options(mc);                  \
4589         }                                                            \
4590     }                                                                \
4591     static const TypeInfo spapr_machine_##suffix##_info = {          \
4592         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4593         .parent = TYPE_SPAPR_MACHINE,                                \
4594         .class_init = spapr_machine_##suffix##_class_init,           \
4595     };                                                               \
4596     static void spapr_machine_register_##suffix(void)                \
4597     {                                                                \
4598         type_register(&spapr_machine_##suffix##_info);               \
4599     }                                                                \
4600     type_init(spapr_machine_register_##suffix)
4601 
4602 /*
4603  * pseries-5.0
4604  */
4605 static void spapr_machine_5_0_class_options(MachineClass *mc)
4606 {
4607     /* Defaults for the latest behaviour inherited from the base class */
4608 }
4609 
4610 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4611 
4612 /*
4613  * pseries-4.2
4614  */
4615 static void spapr_machine_4_2_class_options(MachineClass *mc)
4616 {
4617     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4618 
4619     spapr_machine_5_0_class_options(mc);
4620     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4621     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4622     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4623     smc->rma_limit = 16 * GiB;
4624     mc->nvdimm_supported = false;
4625 }
4626 
4627 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4628 
4629 /*
4630  * pseries-4.1
4631  */
4632 static void spapr_machine_4_1_class_options(MachineClass *mc)
4633 {
4634     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4635     static GlobalProperty compat[] = {
4636         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4637         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4638     };
4639 
4640     spapr_machine_4_2_class_options(mc);
4641     smc->linux_pci_probe = false;
4642     smc->smp_threads_vsmt = false;
4643     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4644     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4645 }
4646 
4647 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4648 
4649 /*
4650  * pseries-4.0
4651  */
4652 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4653                               uint64_t *buid, hwaddr *pio,
4654                               hwaddr *mmio32, hwaddr *mmio64,
4655                               unsigned n_dma, uint32_t *liobns,
4656                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4657 {
4658     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4659                         nv2gpa, nv2atsd, errp);
4660     *nv2gpa = 0;
4661     *nv2atsd = 0;
4662 }
4663 
4664 static void spapr_machine_4_0_class_options(MachineClass *mc)
4665 {
4666     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4667 
4668     spapr_machine_4_1_class_options(mc);
4669     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4670     smc->phb_placement = phb_placement_4_0;
4671     smc->irq = &spapr_irq_xics;
4672     smc->pre_4_1_migration = true;
4673 }
4674 
4675 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4676 
4677 /*
4678  * pseries-3.1
4679  */
4680 static void spapr_machine_3_1_class_options(MachineClass *mc)
4681 {
4682     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4683 
4684     spapr_machine_4_0_class_options(mc);
4685     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4686 
4687     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4688     smc->update_dt_enabled = false;
4689     smc->dr_phb_enabled = false;
4690     smc->broken_host_serial_model = true;
4691     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4692     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4693     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4694     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4695 }
4696 
4697 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4698 
4699 /*
4700  * pseries-3.0
4701  */
4702 
4703 static void spapr_machine_3_0_class_options(MachineClass *mc)
4704 {
4705     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4706 
4707     spapr_machine_3_1_class_options(mc);
4708     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4709 
4710     smc->legacy_irq_allocation = true;
4711     smc->nr_xirqs = 0x400;
4712     smc->irq = &spapr_irq_xics_legacy;
4713 }
4714 
4715 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4716 
4717 /*
4718  * pseries-2.12
4719  */
4720 static void spapr_machine_2_12_class_options(MachineClass *mc)
4721 {
4722     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4723     static GlobalProperty compat[] = {
4724         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4725         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4726     };
4727 
4728     spapr_machine_3_0_class_options(mc);
4729     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4730     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4731 
4732     /* We depend on kvm_enabled() to choose a default value for the
4733      * hpt-max-page-size capability. Of course we can't do it here
4734      * because this is too early and the HW accelerator isn't initialzed
4735      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4736      */
4737     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4738 }
4739 
4740 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4741 
4742 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4743 {
4744     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4745 
4746     spapr_machine_2_12_class_options(mc);
4747     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4748     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4749     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4750 }
4751 
4752 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4753 
4754 /*
4755  * pseries-2.11
4756  */
4757 
4758 static void spapr_machine_2_11_class_options(MachineClass *mc)
4759 {
4760     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4761 
4762     spapr_machine_2_12_class_options(mc);
4763     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4764     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4765 }
4766 
4767 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4768 
4769 /*
4770  * pseries-2.10
4771  */
4772 
4773 static void spapr_machine_2_10_class_options(MachineClass *mc)
4774 {
4775     spapr_machine_2_11_class_options(mc);
4776     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4777 }
4778 
4779 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4780 
4781 /*
4782  * pseries-2.9
4783  */
4784 
4785 static void spapr_machine_2_9_class_options(MachineClass *mc)
4786 {
4787     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4788     static GlobalProperty compat[] = {
4789         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4790     };
4791 
4792     spapr_machine_2_10_class_options(mc);
4793     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4794     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4795     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4796     smc->pre_2_10_has_unused_icps = true;
4797     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4798 }
4799 
4800 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4801 
4802 /*
4803  * pseries-2.8
4804  */
4805 
4806 static void spapr_machine_2_8_class_options(MachineClass *mc)
4807 {
4808     static GlobalProperty compat[] = {
4809         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4810     };
4811 
4812     spapr_machine_2_9_class_options(mc);
4813     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4814     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4815     mc->numa_mem_align_shift = 23;
4816 }
4817 
4818 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4819 
4820 /*
4821  * pseries-2.7
4822  */
4823 
4824 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4825                               uint64_t *buid, hwaddr *pio,
4826                               hwaddr *mmio32, hwaddr *mmio64,
4827                               unsigned n_dma, uint32_t *liobns,
4828                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4829 {
4830     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4831     const uint64_t base_buid = 0x800000020000000ULL;
4832     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4833     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4834     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4835     const uint32_t max_index = 255;
4836     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4837 
4838     uint64_t ram_top = MACHINE(spapr)->ram_size;
4839     hwaddr phb0_base, phb_base;
4840     int i;
4841 
4842     /* Do we have device memory? */
4843     if (MACHINE(spapr)->maxram_size > ram_top) {
4844         /* Can't just use maxram_size, because there may be an
4845          * alignment gap between normal and device memory regions
4846          */
4847         ram_top = MACHINE(spapr)->device_memory->base +
4848             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4849     }
4850 
4851     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4852 
4853     if (index > max_index) {
4854         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4855                    max_index);
4856         return;
4857     }
4858 
4859     *buid = base_buid + index;
4860     for (i = 0; i < n_dma; ++i) {
4861         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4862     }
4863 
4864     phb_base = phb0_base + index * phb_spacing;
4865     *pio = phb_base + pio_offset;
4866     *mmio32 = phb_base + mmio_offset;
4867     /*
4868      * We don't set the 64-bit MMIO window, relying on the PHB's
4869      * fallback behaviour of automatically splitting a large "32-bit"
4870      * window into contiguous 32-bit and 64-bit windows
4871      */
4872 
4873     *nv2gpa = 0;
4874     *nv2atsd = 0;
4875 }
4876 
4877 static void spapr_machine_2_7_class_options(MachineClass *mc)
4878 {
4879     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4880     static GlobalProperty compat[] = {
4881         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4882         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4883         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4884         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4885     };
4886 
4887     spapr_machine_2_8_class_options(mc);
4888     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4889     mc->default_machine_opts = "modern-hotplug-events=off";
4890     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4891     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4892     smc->phb_placement = phb_placement_2_7;
4893 }
4894 
4895 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4896 
4897 /*
4898  * pseries-2.6
4899  */
4900 
4901 static void spapr_machine_2_6_class_options(MachineClass *mc)
4902 {
4903     static GlobalProperty compat[] = {
4904         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4905     };
4906 
4907     spapr_machine_2_7_class_options(mc);
4908     mc->has_hotpluggable_cpus = false;
4909     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4910     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4911 }
4912 
4913 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4914 
4915 /*
4916  * pseries-2.5
4917  */
4918 
4919 static void spapr_machine_2_5_class_options(MachineClass *mc)
4920 {
4921     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4922     static GlobalProperty compat[] = {
4923         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4924     };
4925 
4926     spapr_machine_2_6_class_options(mc);
4927     smc->use_ohci_by_default = true;
4928     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4929     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4930 }
4931 
4932 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4933 
4934 /*
4935  * pseries-2.4
4936  */
4937 
4938 static void spapr_machine_2_4_class_options(MachineClass *mc)
4939 {
4940     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4941 
4942     spapr_machine_2_5_class_options(mc);
4943     smc->dr_lmb_enabled = false;
4944     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4945 }
4946 
4947 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4948 
4949 /*
4950  * pseries-2.3
4951  */
4952 
4953 static void spapr_machine_2_3_class_options(MachineClass *mc)
4954 {
4955     static GlobalProperty compat[] = {
4956         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4957     };
4958     spapr_machine_2_4_class_options(mc);
4959     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4960     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4961 }
4962 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4963 
4964 /*
4965  * pseries-2.2
4966  */
4967 
4968 static void spapr_machine_2_2_class_options(MachineClass *mc)
4969 {
4970     static GlobalProperty compat[] = {
4971         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4972     };
4973 
4974     spapr_machine_2_3_class_options(mc);
4975     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4976     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4977     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4978 }
4979 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4980 
4981 /*
4982  * pseries-2.1
4983  */
4984 
4985 static void spapr_machine_2_1_class_options(MachineClass *mc)
4986 {
4987     spapr_machine_2_2_class_options(mc);
4988     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4989 }
4990 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4991 
4992 static void spapr_machine_register_types(void)
4993 {
4994     type_register_static(&spapr_machine_info);
4995 }
4996 
4997 type_init(spapr_machine_register_types)
4998