xref: /openbmc/qemu/hw/ppc/spapr.c (revision c81219a7)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qemu/datadir.h"
30 #include "qapi/error.h"
31 #include "qapi/visitor.h"
32 #include "sysemu/sysemu.h"
33 #include "sysemu/hostmem.h"
34 #include "sysemu/numa.h"
35 #include "sysemu/qtest.h"
36 #include "sysemu/reset.h"
37 #include "sysemu/runstate.h"
38 #include "qemu/log.h"
39 #include "hw/fw-path-provider.h"
40 #include "elf.h"
41 #include "net/net.h"
42 #include "sysemu/device_tree.h"
43 #include "sysemu/cpus.h"
44 #include "sysemu/hw_accel.h"
45 #include "kvm_ppc.h"
46 #include "migration/misc.h"
47 #include "migration/qemu-file-types.h"
48 #include "migration/global_state.h"
49 #include "migration/register.h"
50 #include "migration/blocker.h"
51 #include "mmu-hash64.h"
52 #include "mmu-book3s-v3.h"
53 #include "cpu-models.h"
54 #include "hw/core/cpu.h"
55 
56 #include "hw/boards.h"
57 #include "hw/ppc/ppc.h"
58 #include "hw/loader.h"
59 
60 #include "hw/ppc/fdt.h"
61 #include "hw/ppc/spapr.h"
62 #include "hw/ppc/spapr_vio.h"
63 #include "hw/qdev-properties.h"
64 #include "hw/pci-host/spapr.h"
65 #include "hw/pci/msi.h"
66 
67 #include "hw/pci/pci.h"
68 #include "hw/scsi/scsi.h"
69 #include "hw/virtio/virtio-scsi.h"
70 #include "hw/virtio/vhost-scsi-common.h"
71 
72 #include "exec/address-spaces.h"
73 #include "exec/ram_addr.h"
74 #include "hw/usb.h"
75 #include "qemu/config-file.h"
76 #include "qemu/error-report.h"
77 #include "trace.h"
78 #include "hw/nmi.h"
79 #include "hw/intc/intc.h"
80 
81 #include "hw/ppc/spapr_cpu_core.h"
82 #include "hw/mem/memory-device.h"
83 #include "hw/ppc/spapr_tpm_proxy.h"
84 #include "hw/ppc/spapr_nvdimm.h"
85 #include "hw/ppc/spapr_numa.h"
86 #include "hw/ppc/pef.h"
87 
88 #include "monitor/monitor.h"
89 
90 #include <libfdt.h>
91 
92 /* SLOF memory layout:
93  *
94  * SLOF raw image loaded at 0, copies its romfs right below the flat
95  * device-tree, then position SLOF itself 31M below that
96  *
97  * So we set FW_OVERHEAD to 40MB which should account for all of that
98  * and more
99  *
100  * We load our kernel at 4M, leaving space for SLOF initial image
101  */
102 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
103 #define FW_MAX_SIZE             0x400000
104 #define FW_FILE_NAME            "slof.bin"
105 #define FW_OVERHEAD             0x2800000
106 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
107 
108 #define MIN_RMA_SLOF            (128 * MiB)
109 
110 #define PHANDLE_INTC            0x00001111
111 
112 /* These two functions implement the VCPU id numbering: one to compute them
113  * all and one to identify thread 0 of a VCORE. Any change to the first one
114  * is likely to have an impact on the second one, so let's keep them close.
115  */
116 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
117 {
118     MachineState *ms = MACHINE(spapr);
119     unsigned int smp_threads = ms->smp.threads;
120 
121     assert(spapr->vsmt);
122     return
123         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
124 }
125 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
126                                       PowerPCCPU *cpu)
127 {
128     assert(spapr->vsmt);
129     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
130 }
131 
132 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
133 {
134     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
135      * and newer QEMUs don't even have them. In both cases, we don't want
136      * to send anything on the wire.
137      */
138     return false;
139 }
140 
141 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
142     .name = "icp/server",
143     .version_id = 1,
144     .minimum_version_id = 1,
145     .needed = pre_2_10_vmstate_dummy_icp_needed,
146     .fields = (VMStateField[]) {
147         VMSTATE_UNUSED(4), /* uint32_t xirr */
148         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
149         VMSTATE_UNUSED(1), /* uint8_t mfrr */
150         VMSTATE_END_OF_LIST()
151     },
152 };
153 
154 static void pre_2_10_vmstate_register_dummy_icp(int i)
155 {
156     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
157                      (void *)(uintptr_t) i);
158 }
159 
160 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
161 {
162     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
163                        (void *)(uintptr_t) i);
164 }
165 
166 int spapr_max_server_number(SpaprMachineState *spapr)
167 {
168     MachineState *ms = MACHINE(spapr);
169 
170     assert(spapr->vsmt);
171     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
172 }
173 
174 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
175                                   int smt_threads)
176 {
177     int i, ret = 0;
178     uint32_t servers_prop[smt_threads];
179     uint32_t gservers_prop[smt_threads * 2];
180     int index = spapr_get_vcpu_id(cpu);
181 
182     if (cpu->compat_pvr) {
183         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
184         if (ret < 0) {
185             return ret;
186         }
187     }
188 
189     /* Build interrupt servers and gservers properties */
190     for (i = 0; i < smt_threads; i++) {
191         servers_prop[i] = cpu_to_be32(index + i);
192         /* Hack, direct the group queues back to cpu 0 */
193         gservers_prop[i*2] = cpu_to_be32(index + i);
194         gservers_prop[i*2 + 1] = 0;
195     }
196     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
197                       servers_prop, sizeof(servers_prop));
198     if (ret < 0) {
199         return ret;
200     }
201     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
202                       gservers_prop, sizeof(gservers_prop));
203 
204     return ret;
205 }
206 
207 static void spapr_dt_pa_features(SpaprMachineState *spapr,
208                                  PowerPCCPU *cpu,
209                                  void *fdt, int offset)
210 {
211     uint8_t pa_features_206[] = { 6, 0,
212         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
213     uint8_t pa_features_207[] = { 24, 0,
214         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
215         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
216         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
217         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
218     uint8_t pa_features_300[] = { 66, 0,
219         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
220         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
221         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
222         /* 6: DS207 */
223         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
224         /* 16: Vector */
225         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
226         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
227         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
228         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
229         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
230         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
231         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
232         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
233         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
234         /* 42: PM, 44: PC RA, 46: SC vec'd */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
236         /* 48: SIMD, 50: QP BFP, 52: String */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
238         /* 54: DecFP, 56: DecI, 58: SHA */
239         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
240         /* 60: NM atomic, 62: RNG */
241         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
242     };
243     uint8_t *pa_features = NULL;
244     size_t pa_size;
245 
246     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
247         pa_features = pa_features_206;
248         pa_size = sizeof(pa_features_206);
249     }
250     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
251         pa_features = pa_features_207;
252         pa_size = sizeof(pa_features_207);
253     }
254     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
255         pa_features = pa_features_300;
256         pa_size = sizeof(pa_features_300);
257     }
258     if (!pa_features) {
259         return;
260     }
261 
262     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
263         /*
264          * Note: we keep CI large pages off by default because a 64K capable
265          * guest provisioned with large pages might otherwise try to map a qemu
266          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
267          * even if that qemu runs on a 4k host.
268          * We dd this bit back here if we are confident this is not an issue
269          */
270         pa_features[3] |= 0x20;
271     }
272     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
273         pa_features[24] |= 0x80;    /* Transactional memory support */
274     }
275     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
276         /* Workaround for broken kernels that attempt (guest) radix
277          * mode when they can't handle it, if they see the radix bit set
278          * in pa-features. So hide it from them. */
279         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
280     }
281 
282     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
283 }
284 
285 static hwaddr spapr_node0_size(MachineState *machine)
286 {
287     if (machine->numa_state->num_nodes) {
288         int i;
289         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
290             if (machine->numa_state->nodes[i].node_mem) {
291                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
292                            machine->ram_size);
293             }
294         }
295     }
296     return machine->ram_size;
297 }
298 
299 static void add_str(GString *s, const gchar *s1)
300 {
301     g_string_append_len(s, s1, strlen(s1) + 1);
302 }
303 
304 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
305                                 hwaddr start, hwaddr size)
306 {
307     char mem_name[32];
308     uint64_t mem_reg_property[2];
309     int off;
310 
311     mem_reg_property[0] = cpu_to_be64(start);
312     mem_reg_property[1] = cpu_to_be64(size);
313 
314     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
315     off = fdt_add_subnode(fdt, 0, mem_name);
316     _FDT(off);
317     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
318     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
319                       sizeof(mem_reg_property))));
320     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
321     return off;
322 }
323 
324 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
325 {
326     MemoryDeviceInfoList *info;
327 
328     for (info = list; info; info = info->next) {
329         MemoryDeviceInfo *value = info->value;
330 
331         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
332             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
333 
334             if (addr >= pcdimm_info->addr &&
335                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
336                 return pcdimm_info->node;
337             }
338         }
339     }
340 
341     return -1;
342 }
343 
344 struct sPAPRDrconfCellV2 {
345      uint32_t seq_lmbs;
346      uint64_t base_addr;
347      uint32_t drc_index;
348      uint32_t aa_index;
349      uint32_t flags;
350 } QEMU_PACKED;
351 
352 typedef struct DrconfCellQueue {
353     struct sPAPRDrconfCellV2 cell;
354     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
355 } DrconfCellQueue;
356 
357 static DrconfCellQueue *
358 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
359                       uint32_t drc_index, uint32_t aa_index,
360                       uint32_t flags)
361 {
362     DrconfCellQueue *elem;
363 
364     elem = g_malloc0(sizeof(*elem));
365     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
366     elem->cell.base_addr = cpu_to_be64(base_addr);
367     elem->cell.drc_index = cpu_to_be32(drc_index);
368     elem->cell.aa_index = cpu_to_be32(aa_index);
369     elem->cell.flags = cpu_to_be32(flags);
370 
371     return elem;
372 }
373 
374 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
375                                       int offset, MemoryDeviceInfoList *dimms)
376 {
377     MachineState *machine = MACHINE(spapr);
378     uint8_t *int_buf, *cur_index;
379     int ret;
380     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
381     uint64_t addr, cur_addr, size;
382     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
383     uint64_t mem_end = machine->device_memory->base +
384                        memory_region_size(&machine->device_memory->mr);
385     uint32_t node, buf_len, nr_entries = 0;
386     SpaprDrc *drc;
387     DrconfCellQueue *elem, *next;
388     MemoryDeviceInfoList *info;
389     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
390         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
391 
392     /* Entry to cover RAM and the gap area */
393     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
394                                  SPAPR_LMB_FLAGS_RESERVED |
395                                  SPAPR_LMB_FLAGS_DRC_INVALID);
396     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
397     nr_entries++;
398 
399     cur_addr = machine->device_memory->base;
400     for (info = dimms; info; info = info->next) {
401         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
402 
403         addr = di->addr;
404         size = di->size;
405         node = di->node;
406 
407         /*
408          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
409          * area is marked hotpluggable in the next iteration for the bigger
410          * chunk including the NVDIMM occupied area.
411          */
412         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
413             continue;
414 
415         /* Entry for hot-pluggable area */
416         if (cur_addr < addr) {
417             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
418             g_assert(drc);
419             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
420                                          cur_addr, spapr_drc_index(drc), -1, 0);
421             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
422             nr_entries++;
423         }
424 
425         /* Entry for DIMM */
426         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
427         g_assert(drc);
428         elem = spapr_get_drconf_cell(size / lmb_size, addr,
429                                      spapr_drc_index(drc), node,
430                                      (SPAPR_LMB_FLAGS_ASSIGNED |
431                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
432         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
433         nr_entries++;
434         cur_addr = addr + size;
435     }
436 
437     /* Entry for remaining hotpluggable area */
438     if (cur_addr < mem_end) {
439         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
440         g_assert(drc);
441         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
442                                      cur_addr, spapr_drc_index(drc), -1, 0);
443         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
444         nr_entries++;
445     }
446 
447     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
448     int_buf = cur_index = g_malloc0(buf_len);
449     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
450     cur_index += sizeof(nr_entries);
451 
452     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
453         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
454         cur_index += sizeof(elem->cell);
455         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
456         g_free(elem);
457     }
458 
459     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
460     g_free(int_buf);
461     if (ret < 0) {
462         return -1;
463     }
464     return 0;
465 }
466 
467 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
468                                    int offset, MemoryDeviceInfoList *dimms)
469 {
470     MachineState *machine = MACHINE(spapr);
471     int i, ret;
472     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
473     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
474     uint32_t nr_lmbs = (machine->device_memory->base +
475                        memory_region_size(&machine->device_memory->mr)) /
476                        lmb_size;
477     uint32_t *int_buf, *cur_index, buf_len;
478 
479     /*
480      * Allocate enough buffer size to fit in ibm,dynamic-memory
481      */
482     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
483     cur_index = int_buf = g_malloc0(buf_len);
484     int_buf[0] = cpu_to_be32(nr_lmbs);
485     cur_index++;
486     for (i = 0; i < nr_lmbs; i++) {
487         uint64_t addr = i * lmb_size;
488         uint32_t *dynamic_memory = cur_index;
489 
490         if (i >= device_lmb_start) {
491             SpaprDrc *drc;
492 
493             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
494             g_assert(drc);
495 
496             dynamic_memory[0] = cpu_to_be32(addr >> 32);
497             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
498             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
499             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
500             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
501             if (memory_region_present(get_system_memory(), addr)) {
502                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
503             } else {
504                 dynamic_memory[5] = cpu_to_be32(0);
505             }
506         } else {
507             /*
508              * LMB information for RMA, boot time RAM and gap b/n RAM and
509              * device memory region -- all these are marked as reserved
510              * and as having no valid DRC.
511              */
512             dynamic_memory[0] = cpu_to_be32(addr >> 32);
513             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
514             dynamic_memory[2] = cpu_to_be32(0);
515             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
516             dynamic_memory[4] = cpu_to_be32(-1);
517             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
518                                             SPAPR_LMB_FLAGS_DRC_INVALID);
519         }
520 
521         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
522     }
523     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
524     g_free(int_buf);
525     if (ret < 0) {
526         return -1;
527     }
528     return 0;
529 }
530 
531 /*
532  * Adds ibm,dynamic-reconfiguration-memory node.
533  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
534  * of this device tree node.
535  */
536 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
537                                                    void *fdt)
538 {
539     MachineState *machine = MACHINE(spapr);
540     int ret, offset;
541     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
542     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
543                                 cpu_to_be32(lmb_size & 0xffffffff)};
544     MemoryDeviceInfoList *dimms = NULL;
545 
546     /*
547      * Don't create the node if there is no device memory
548      */
549     if (machine->ram_size == machine->maxram_size) {
550         return 0;
551     }
552 
553     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
554 
555     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
556                     sizeof(prop_lmb_size));
557     if (ret < 0) {
558         return ret;
559     }
560 
561     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
562     if (ret < 0) {
563         return ret;
564     }
565 
566     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
567     if (ret < 0) {
568         return ret;
569     }
570 
571     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
572     dimms = qmp_memory_device_list();
573     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
574         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
575     } else {
576         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
577     }
578     qapi_free_MemoryDeviceInfoList(dimms);
579 
580     if (ret < 0) {
581         return ret;
582     }
583 
584     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
585 
586     return ret;
587 }
588 
589 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
590 {
591     MachineState *machine = MACHINE(spapr);
592     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
593     hwaddr mem_start, node_size;
594     int i, nb_nodes = machine->numa_state->num_nodes;
595     NodeInfo *nodes = machine->numa_state->nodes;
596 
597     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
598         if (!nodes[i].node_mem) {
599             continue;
600         }
601         if (mem_start >= machine->ram_size) {
602             node_size = 0;
603         } else {
604             node_size = nodes[i].node_mem;
605             if (node_size > machine->ram_size - mem_start) {
606                 node_size = machine->ram_size - mem_start;
607             }
608         }
609         if (!mem_start) {
610             /* spapr_machine_init() checks for rma_size <= node0_size
611              * already */
612             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
613             mem_start += spapr->rma_size;
614             node_size -= spapr->rma_size;
615         }
616         for ( ; node_size; ) {
617             hwaddr sizetmp = pow2floor(node_size);
618 
619             /* mem_start != 0 here */
620             if (ctzl(mem_start) < ctzl(sizetmp)) {
621                 sizetmp = 1ULL << ctzl(mem_start);
622             }
623 
624             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
625             node_size -= sizetmp;
626             mem_start += sizetmp;
627         }
628     }
629 
630     /* Generate ibm,dynamic-reconfiguration-memory node if required */
631     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
632         int ret;
633 
634         g_assert(smc->dr_lmb_enabled);
635         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
636         if (ret) {
637             return ret;
638         }
639     }
640 
641     return 0;
642 }
643 
644 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
645                          SpaprMachineState *spapr)
646 {
647     MachineState *ms = MACHINE(spapr);
648     PowerPCCPU *cpu = POWERPC_CPU(cs);
649     CPUPPCState *env = &cpu->env;
650     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
651     int index = spapr_get_vcpu_id(cpu);
652     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
653                        0xffffffff, 0xffffffff};
654     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
655         : SPAPR_TIMEBASE_FREQ;
656     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
657     uint32_t page_sizes_prop[64];
658     size_t page_sizes_prop_size;
659     unsigned int smp_threads = ms->smp.threads;
660     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
661     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
662     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
663     SpaprDrc *drc;
664     int drc_index;
665     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
666     int i;
667 
668     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
669     if (drc) {
670         drc_index = spapr_drc_index(drc);
671         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
672     }
673 
674     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
675     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
676 
677     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
678     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
679                            env->dcache_line_size)));
680     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
681                            env->dcache_line_size)));
682     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
683                            env->icache_line_size)));
684     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
685                            env->icache_line_size)));
686 
687     if (pcc->l1_dcache_size) {
688         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
689                                pcc->l1_dcache_size)));
690     } else {
691         warn_report("Unknown L1 dcache size for cpu");
692     }
693     if (pcc->l1_icache_size) {
694         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
695                                pcc->l1_icache_size)));
696     } else {
697         warn_report("Unknown L1 icache size for cpu");
698     }
699 
700     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
701     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
702     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
703     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
704     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
705     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
706 
707     if (env->spr_cb[SPR_PURR].oea_read) {
708         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
709     }
710     if (env->spr_cb[SPR_SPURR].oea_read) {
711         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
712     }
713 
714     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
715         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
716                           segs, sizeof(segs))));
717     }
718 
719     /* Advertise VSX (vector extensions) if available
720      *   1               == VMX / Altivec available
721      *   2               == VSX available
722      *
723      * Only CPUs for which we create core types in spapr_cpu_core.c
724      * are possible, and all of those have VMX */
725     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
726         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
727     } else {
728         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
729     }
730 
731     /* Advertise DFP (Decimal Floating Point) if available
732      *   0 / no property == no DFP
733      *   1               == DFP available */
734     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
735         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
736     }
737 
738     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
739                                                       sizeof(page_sizes_prop));
740     if (page_sizes_prop_size) {
741         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
742                           page_sizes_prop, page_sizes_prop_size)));
743     }
744 
745     spapr_dt_pa_features(spapr, cpu, fdt, offset);
746 
747     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
748                            cs->cpu_index / vcpus_per_socket)));
749 
750     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
751                       pft_size_prop, sizeof(pft_size_prop))));
752 
753     if (ms->numa_state->num_nodes > 1) {
754         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
755     }
756 
757     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
758 
759     if (pcc->radix_page_info) {
760         for (i = 0; i < pcc->radix_page_info->count; i++) {
761             radix_AP_encodings[i] =
762                 cpu_to_be32(pcc->radix_page_info->entries[i]);
763         }
764         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
765                           radix_AP_encodings,
766                           pcc->radix_page_info->count *
767                           sizeof(radix_AP_encodings[0]))));
768     }
769 
770     /*
771      * We set this property to let the guest know that it can use the large
772      * decrementer and its width in bits.
773      */
774     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
775         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
776                               pcc->lrg_decr_bits)));
777 }
778 
779 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
780 {
781     CPUState **rev;
782     CPUState *cs;
783     int n_cpus;
784     int cpus_offset;
785     int i;
786 
787     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
788     _FDT(cpus_offset);
789     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
790     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
791 
792     /*
793      * We walk the CPUs in reverse order to ensure that CPU DT nodes
794      * created by fdt_add_subnode() end up in the right order in FDT
795      * for the guest kernel the enumerate the CPUs correctly.
796      *
797      * The CPU list cannot be traversed in reverse order, so we need
798      * to do extra work.
799      */
800     n_cpus = 0;
801     rev = NULL;
802     CPU_FOREACH(cs) {
803         rev = g_renew(CPUState *, rev, n_cpus + 1);
804         rev[n_cpus++] = cs;
805     }
806 
807     for (i = n_cpus - 1; i >= 0; i--) {
808         CPUState *cs = rev[i];
809         PowerPCCPU *cpu = POWERPC_CPU(cs);
810         int index = spapr_get_vcpu_id(cpu);
811         DeviceClass *dc = DEVICE_GET_CLASS(cs);
812         g_autofree char *nodename = NULL;
813         int offset;
814 
815         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
816             continue;
817         }
818 
819         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
820         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
821         _FDT(offset);
822         spapr_dt_cpu(cs, fdt, offset, spapr);
823     }
824 
825     g_free(rev);
826 }
827 
828 static int spapr_dt_rng(void *fdt)
829 {
830     int node;
831     int ret;
832 
833     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
834     if (node <= 0) {
835         return -1;
836     }
837     ret = fdt_setprop_string(fdt, node, "device_type",
838                              "ibm,platform-facilities");
839     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
840     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
841 
842     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
843     if (node <= 0) {
844         return -1;
845     }
846     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
847 
848     return ret ? -1 : 0;
849 }
850 
851 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
852 {
853     MachineState *ms = MACHINE(spapr);
854     int rtas;
855     GString *hypertas = g_string_sized_new(256);
856     GString *qemu_hypertas = g_string_sized_new(256);
857     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
858         memory_region_size(&MACHINE(spapr)->device_memory->mr);
859     uint32_t lrdr_capacity[] = {
860         cpu_to_be32(max_device_addr >> 32),
861         cpu_to_be32(max_device_addr & 0xffffffff),
862         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
863         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
864         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
865     };
866 
867     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
868 
869     /* hypertas */
870     add_str(hypertas, "hcall-pft");
871     add_str(hypertas, "hcall-term");
872     add_str(hypertas, "hcall-dabr");
873     add_str(hypertas, "hcall-interrupt");
874     add_str(hypertas, "hcall-tce");
875     add_str(hypertas, "hcall-vio");
876     add_str(hypertas, "hcall-splpar");
877     add_str(hypertas, "hcall-join");
878     add_str(hypertas, "hcall-bulk");
879     add_str(hypertas, "hcall-set-mode");
880     add_str(hypertas, "hcall-sprg0");
881     add_str(hypertas, "hcall-copy");
882     add_str(hypertas, "hcall-debug");
883     add_str(hypertas, "hcall-vphn");
884     add_str(qemu_hypertas, "hcall-memop1");
885 
886     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
887         add_str(hypertas, "hcall-multi-tce");
888     }
889 
890     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
891         add_str(hypertas, "hcall-hpt-resize");
892     }
893 
894     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
895                      hypertas->str, hypertas->len));
896     g_string_free(hypertas, TRUE);
897     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
898                      qemu_hypertas->str, qemu_hypertas->len));
899     g_string_free(qemu_hypertas, TRUE);
900 
901     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
902 
903     /*
904      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
905      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
906      *
907      * The system reset requirements are driven by existing Linux and PowerVM
908      * implementation which (contrary to PAPR) saves r3 in the error log
909      * structure like machine check, so Linux expects to find the saved r3
910      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
911      * does not look at the error value).
912      *
913      * System reset interrupts are not subject to interlock like machine
914      * check, so this memory area could be corrupted if the sreset is
915      * interrupted by a machine check (or vice versa) if it was shared. To
916      * prevent this, system reset uses per-CPU areas for the sreset save
917      * area. A system reset that interrupts a system reset handler could
918      * still overwrite this area, but Linux doesn't try to recover in that
919      * case anyway.
920      *
921      * The extra 8 bytes is required because Linux's FWNMI error log check
922      * is off-by-one.
923      */
924     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
925 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
926     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
927                           RTAS_ERROR_LOG_MAX));
928     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
929                           RTAS_EVENT_SCAN_RATE));
930 
931     g_assert(msi_nonbroken);
932     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
933 
934     /*
935      * According to PAPR, rtas ibm,os-term does not guarantee a return
936      * back to the guest cpu.
937      *
938      * While an additional ibm,extended-os-term property indicates
939      * that rtas call return will always occur. Set this property.
940      */
941     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
942 
943     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
944                      lrdr_capacity, sizeof(lrdr_capacity)));
945 
946     spapr_dt_rtas_tokens(fdt, rtas);
947 }
948 
949 /*
950  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
951  * and the XIVE features that the guest may request and thus the valid
952  * values for bytes 23..26 of option vector 5:
953  */
954 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
955                                           int chosen)
956 {
957     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
958 
959     char val[2 * 4] = {
960         23, 0x00, /* XICS / XIVE mode */
961         24, 0x00, /* Hash/Radix, filled in below. */
962         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
963         26, 0x40, /* Radix options: GTSE == yes. */
964     };
965 
966     if (spapr->irq->xics && spapr->irq->xive) {
967         val[1] = SPAPR_OV5_XIVE_BOTH;
968     } else if (spapr->irq->xive) {
969         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
970     } else {
971         assert(spapr->irq->xics);
972         val[1] = SPAPR_OV5_XIVE_LEGACY;
973     }
974 
975     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
976                           first_ppc_cpu->compat_pvr)) {
977         /*
978          * If we're in a pre POWER9 compat mode then the guest should
979          * do hash and use the legacy interrupt mode
980          */
981         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
982         val[3] = 0x00; /* Hash */
983     } else if (kvm_enabled()) {
984         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
985             val[3] = 0x80; /* OV5_MMU_BOTH */
986         } else if (kvmppc_has_cap_mmu_radix()) {
987             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
988         } else {
989             val[3] = 0x00; /* Hash */
990         }
991     } else {
992         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
993         val[3] = 0xC0;
994     }
995     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
996                      val, sizeof(val)));
997 }
998 
999 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1000 {
1001     MachineState *machine = MACHINE(spapr);
1002     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1003     int chosen;
1004 
1005     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1006 
1007     if (reset) {
1008         const char *boot_device = machine->boot_order;
1009         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1010         size_t cb = 0;
1011         char *bootlist = get_boot_devices_list(&cb);
1012 
1013         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1014             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1015                                     machine->kernel_cmdline));
1016         }
1017 
1018         if (spapr->initrd_size) {
1019             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1020                                   spapr->initrd_base));
1021             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1022                                   spapr->initrd_base + spapr->initrd_size));
1023         }
1024 
1025         if (spapr->kernel_size) {
1026             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1027                                   cpu_to_be64(spapr->kernel_size) };
1028 
1029             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1030                          &kprop, sizeof(kprop)));
1031             if (spapr->kernel_le) {
1032                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1033             }
1034         }
1035         if (boot_menu) {
1036             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1037         }
1038         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1039         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1040         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1041 
1042         if (cb && bootlist) {
1043             int i;
1044 
1045             for (i = 0; i < cb; i++) {
1046                 if (bootlist[i] == '\n') {
1047                     bootlist[i] = ' ';
1048                 }
1049             }
1050             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1051         }
1052 
1053         if (boot_device && strlen(boot_device)) {
1054             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1055         }
1056 
1057         if (!spapr->has_graphics && stdout_path) {
1058             /*
1059              * "linux,stdout-path" and "stdout" properties are
1060              * deprecated by linux kernel. New platforms should only
1061              * use the "stdout-path" property. Set the new property
1062              * and continue using older property to remain compatible
1063              * with the existing firmware.
1064              */
1065             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1066             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1067         }
1068 
1069         /*
1070          * We can deal with BAR reallocation just fine, advertise it
1071          * to the guest
1072          */
1073         if (smc->linux_pci_probe) {
1074             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1075         }
1076 
1077         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1078 
1079         g_free(stdout_path);
1080         g_free(bootlist);
1081     }
1082 
1083     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1084 }
1085 
1086 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1087 {
1088     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1089      * KVM to work under pHyp with some guest co-operation */
1090     int hypervisor;
1091     uint8_t hypercall[16];
1092 
1093     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1094     /* indicate KVM hypercall interface */
1095     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1096     if (kvmppc_has_cap_fixup_hcalls()) {
1097         /*
1098          * Older KVM versions with older guest kernels were broken
1099          * with the magic page, don't allow the guest to map it.
1100          */
1101         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1102                                   sizeof(hypercall))) {
1103             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1104                              hypercall, sizeof(hypercall)));
1105         }
1106     }
1107 }
1108 
1109 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1110 {
1111     MachineState *machine = MACHINE(spapr);
1112     MachineClass *mc = MACHINE_GET_CLASS(machine);
1113     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1114     uint32_t root_drc_type_mask = 0;
1115     int ret;
1116     void *fdt;
1117     SpaprPhbState *phb;
1118     char *buf;
1119 
1120     fdt = g_malloc0(space);
1121     _FDT((fdt_create_empty_tree(fdt, space)));
1122 
1123     /* Root node */
1124     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1125     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1126     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1127 
1128     /* Guest UUID & Name*/
1129     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1130     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1131     if (qemu_uuid_set) {
1132         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1133     }
1134     g_free(buf);
1135 
1136     if (qemu_get_vm_name()) {
1137         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1138                                 qemu_get_vm_name()));
1139     }
1140 
1141     /* Host Model & Serial Number */
1142     if (spapr->host_model) {
1143         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1144     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1145         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1146         g_free(buf);
1147     }
1148 
1149     if (spapr->host_serial) {
1150         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1151     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1152         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1153         g_free(buf);
1154     }
1155 
1156     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1157     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1158 
1159     /* /interrupt controller */
1160     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1161 
1162     ret = spapr_dt_memory(spapr, fdt);
1163     if (ret < 0) {
1164         error_report("couldn't setup memory nodes in fdt");
1165         exit(1);
1166     }
1167 
1168     /* /vdevice */
1169     spapr_dt_vdevice(spapr->vio_bus, fdt);
1170 
1171     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1172         ret = spapr_dt_rng(fdt);
1173         if (ret < 0) {
1174             error_report("could not set up rng device in the fdt");
1175             exit(1);
1176         }
1177     }
1178 
1179     QLIST_FOREACH(phb, &spapr->phbs, list) {
1180         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1181         if (ret < 0) {
1182             error_report("couldn't setup PCI devices in fdt");
1183             exit(1);
1184         }
1185     }
1186 
1187     spapr_dt_cpus(fdt, spapr);
1188 
1189     /* ibm,drc-indexes and friends */
1190     if (smc->dr_lmb_enabled) {
1191         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1192     }
1193     if (smc->dr_phb_enabled) {
1194         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1195     }
1196     if (mc->nvdimm_supported) {
1197         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1198     }
1199     if (root_drc_type_mask) {
1200         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1201     }
1202 
1203     if (mc->has_hotpluggable_cpus) {
1204         int offset = fdt_path_offset(fdt, "/cpus");
1205         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1206         if (ret < 0) {
1207             error_report("Couldn't set up CPU DR device tree properties");
1208             exit(1);
1209         }
1210     }
1211 
1212     /* /event-sources */
1213     spapr_dt_events(spapr, fdt);
1214 
1215     /* /rtas */
1216     spapr_dt_rtas(spapr, fdt);
1217 
1218     /* /chosen */
1219     spapr_dt_chosen(spapr, fdt, reset);
1220 
1221     /* /hypervisor */
1222     if (kvm_enabled()) {
1223         spapr_dt_hypervisor(spapr, fdt);
1224     }
1225 
1226     /* Build memory reserve map */
1227     if (reset) {
1228         if (spapr->kernel_size) {
1229             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1230                                   spapr->kernel_size)));
1231         }
1232         if (spapr->initrd_size) {
1233             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1234                                   spapr->initrd_size)));
1235         }
1236     }
1237 
1238     /* NVDIMM devices */
1239     if (mc->nvdimm_supported) {
1240         spapr_dt_persistent_memory(spapr, fdt);
1241     }
1242 
1243     return fdt;
1244 }
1245 
1246 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1247 {
1248     SpaprMachineState *spapr = opaque;
1249 
1250     return (addr & 0x0fffffff) + spapr->kernel_addr;
1251 }
1252 
1253 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1254                                     PowerPCCPU *cpu)
1255 {
1256     CPUPPCState *env = &cpu->env;
1257 
1258     /* The TCG path should also be holding the BQL at this point */
1259     g_assert(qemu_mutex_iothread_locked());
1260 
1261     if (msr_pr) {
1262         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1263         env->gpr[3] = H_PRIVILEGE;
1264     } else {
1265         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1266     }
1267 }
1268 
1269 struct LPCRSyncState {
1270     target_ulong value;
1271     target_ulong mask;
1272 };
1273 
1274 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1275 {
1276     struct LPCRSyncState *s = arg.host_ptr;
1277     PowerPCCPU *cpu = POWERPC_CPU(cs);
1278     CPUPPCState *env = &cpu->env;
1279     target_ulong lpcr;
1280 
1281     cpu_synchronize_state(cs);
1282     lpcr = env->spr[SPR_LPCR];
1283     lpcr &= ~s->mask;
1284     lpcr |= s->value;
1285     ppc_store_lpcr(cpu, lpcr);
1286 }
1287 
1288 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1289 {
1290     CPUState *cs;
1291     struct LPCRSyncState s = {
1292         .value = value,
1293         .mask = mask
1294     };
1295     CPU_FOREACH(cs) {
1296         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1297     }
1298 }
1299 
1300 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1301 {
1302     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1303 
1304     /* Copy PATE1:GR into PATE0:HR */
1305     entry->dw0 = spapr->patb_entry & PATE0_HR;
1306     entry->dw1 = spapr->patb_entry;
1307 }
1308 
1309 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1310 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1311 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1312 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1313 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1314 
1315 /*
1316  * Get the fd to access the kernel htab, re-opening it if necessary
1317  */
1318 static int get_htab_fd(SpaprMachineState *spapr)
1319 {
1320     Error *local_err = NULL;
1321 
1322     if (spapr->htab_fd >= 0) {
1323         return spapr->htab_fd;
1324     }
1325 
1326     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1327     if (spapr->htab_fd < 0) {
1328         error_report_err(local_err);
1329     }
1330 
1331     return spapr->htab_fd;
1332 }
1333 
1334 void close_htab_fd(SpaprMachineState *spapr)
1335 {
1336     if (spapr->htab_fd >= 0) {
1337         close(spapr->htab_fd);
1338     }
1339     spapr->htab_fd = -1;
1340 }
1341 
1342 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1343 {
1344     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1345 
1346     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1347 }
1348 
1349 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1350 {
1351     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1352 
1353     assert(kvm_enabled());
1354 
1355     if (!spapr->htab) {
1356         return 0;
1357     }
1358 
1359     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1360 }
1361 
1362 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1363                                                 hwaddr ptex, int n)
1364 {
1365     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1366     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1367 
1368     if (!spapr->htab) {
1369         /*
1370          * HTAB is controlled by KVM. Fetch into temporary buffer
1371          */
1372         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1373         kvmppc_read_hptes(hptes, ptex, n);
1374         return hptes;
1375     }
1376 
1377     /*
1378      * HTAB is controlled by QEMU. Just point to the internally
1379      * accessible PTEG.
1380      */
1381     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1382 }
1383 
1384 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1385                               const ppc_hash_pte64_t *hptes,
1386                               hwaddr ptex, int n)
1387 {
1388     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1389 
1390     if (!spapr->htab) {
1391         g_free((void *)hptes);
1392     }
1393 
1394     /* Nothing to do for qemu managed HPT */
1395 }
1396 
1397 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1398                       uint64_t pte0, uint64_t pte1)
1399 {
1400     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1401     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1402 
1403     if (!spapr->htab) {
1404         kvmppc_write_hpte(ptex, pte0, pte1);
1405     } else {
1406         if (pte0 & HPTE64_V_VALID) {
1407             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1408             /*
1409              * When setting valid, we write PTE1 first. This ensures
1410              * proper synchronization with the reading code in
1411              * ppc_hash64_pteg_search()
1412              */
1413             smp_wmb();
1414             stq_p(spapr->htab + offset, pte0);
1415         } else {
1416             stq_p(spapr->htab + offset, pte0);
1417             /*
1418              * When clearing it we set PTE0 first. This ensures proper
1419              * synchronization with the reading code in
1420              * ppc_hash64_pteg_search()
1421              */
1422             smp_wmb();
1423             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1424         }
1425     }
1426 }
1427 
1428 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1429                              uint64_t pte1)
1430 {
1431     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1432     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1433 
1434     if (!spapr->htab) {
1435         /* There should always be a hash table when this is called */
1436         error_report("spapr_hpte_set_c called with no hash table !");
1437         return;
1438     }
1439 
1440     /* The HW performs a non-atomic byte update */
1441     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1442 }
1443 
1444 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1445                              uint64_t pte1)
1446 {
1447     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1448     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1449 
1450     if (!spapr->htab) {
1451         /* There should always be a hash table when this is called */
1452         error_report("spapr_hpte_set_r called with no hash table !");
1453         return;
1454     }
1455 
1456     /* The HW performs a non-atomic byte update */
1457     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1458 }
1459 
1460 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1461 {
1462     int shift;
1463 
1464     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1465      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1466      * that's much more than is needed for Linux guests */
1467     shift = ctz64(pow2ceil(ramsize)) - 7;
1468     shift = MAX(shift, 18); /* Minimum architected size */
1469     shift = MIN(shift, 46); /* Maximum architected size */
1470     return shift;
1471 }
1472 
1473 void spapr_free_hpt(SpaprMachineState *spapr)
1474 {
1475     g_free(spapr->htab);
1476     spapr->htab = NULL;
1477     spapr->htab_shift = 0;
1478     close_htab_fd(spapr);
1479 }
1480 
1481 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1482 {
1483     ERRP_GUARD();
1484     long rc;
1485 
1486     /* Clean up any HPT info from a previous boot */
1487     spapr_free_hpt(spapr);
1488 
1489     rc = kvmppc_reset_htab(shift);
1490 
1491     if (rc == -EOPNOTSUPP) {
1492         error_setg(errp, "HPT not supported in nested guests");
1493         return -EOPNOTSUPP;
1494     }
1495 
1496     if (rc < 0) {
1497         /* kernel-side HPT needed, but couldn't allocate one */
1498         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1499                          shift);
1500         error_append_hint(errp, "Try smaller maxmem?\n");
1501         return -errno;
1502     } else if (rc > 0) {
1503         /* kernel-side HPT allocated */
1504         if (rc != shift) {
1505             error_setg(errp,
1506                        "Requested order %d HPT, but kernel allocated order %ld",
1507                        shift, rc);
1508             error_append_hint(errp, "Try smaller maxmem?\n");
1509             return -ENOSPC;
1510         }
1511 
1512         spapr->htab_shift = shift;
1513         spapr->htab = NULL;
1514     } else {
1515         /* kernel-side HPT not needed, allocate in userspace instead */
1516         size_t size = 1ULL << shift;
1517         int i;
1518 
1519         spapr->htab = qemu_memalign(size, size);
1520         memset(spapr->htab, 0, size);
1521         spapr->htab_shift = shift;
1522 
1523         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1524             DIRTY_HPTE(HPTE(spapr->htab, i));
1525         }
1526     }
1527     /* We're setting up a hash table, so that means we're not radix */
1528     spapr->patb_entry = 0;
1529     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1530     return 0;
1531 }
1532 
1533 void spapr_setup_hpt(SpaprMachineState *spapr)
1534 {
1535     int hpt_shift;
1536 
1537     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1538         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1539     } else {
1540         uint64_t current_ram_size;
1541 
1542         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1543         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1544     }
1545     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1546 
1547     if (kvm_enabled()) {
1548         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1549 
1550         /* Check our RMA fits in the possible VRMA */
1551         if (vrma_limit < spapr->rma_size) {
1552             error_report("Unable to create %" HWADDR_PRIu
1553                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1554                          spapr->rma_size / MiB, vrma_limit / MiB);
1555             exit(EXIT_FAILURE);
1556         }
1557     }
1558 }
1559 
1560 static void spapr_machine_reset(MachineState *machine)
1561 {
1562     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1563     PowerPCCPU *first_ppc_cpu;
1564     hwaddr fdt_addr;
1565     void *fdt;
1566     int rc;
1567 
1568     pef_kvm_reset(machine->cgs, &error_fatal);
1569     spapr_caps_apply(spapr);
1570 
1571     first_ppc_cpu = POWERPC_CPU(first_cpu);
1572     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1573         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1574                               spapr->max_compat_pvr)) {
1575         /*
1576          * If using KVM with radix mode available, VCPUs can be started
1577          * without a HPT because KVM will start them in radix mode.
1578          * Set the GR bit in PATE so that we know there is no HPT.
1579          */
1580         spapr->patb_entry = PATE1_GR;
1581         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1582     } else {
1583         spapr_setup_hpt(spapr);
1584     }
1585 
1586     qemu_devices_reset();
1587 
1588     spapr_ovec_cleanup(spapr->ov5_cas);
1589     spapr->ov5_cas = spapr_ovec_new();
1590 
1591     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1592 
1593     /*
1594      * This is fixing some of the default configuration of the XIVE
1595      * devices. To be called after the reset of the machine devices.
1596      */
1597     spapr_irq_reset(spapr, &error_fatal);
1598 
1599     /*
1600      * There is no CAS under qtest. Simulate one to please the code that
1601      * depends on spapr->ov5_cas. This is especially needed to test device
1602      * unplug, so we do that before resetting the DRCs.
1603      */
1604     if (qtest_enabled()) {
1605         spapr_ovec_cleanup(spapr->ov5_cas);
1606         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1607     }
1608 
1609     /* DRC reset may cause a device to be unplugged. This will cause troubles
1610      * if this device is used by another device (eg, a running vhost backend
1611      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1612      * situations, we reset DRCs after all devices have been reset.
1613      */
1614     spapr_drc_reset_all(spapr);
1615 
1616     spapr_clear_pending_events(spapr);
1617 
1618     /*
1619      * We place the device tree and RTAS just below either the top of the RMA,
1620      * or just below 2GB, whichever is lower, so that it can be
1621      * processed with 32-bit real mode code if necessary
1622      */
1623     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1624 
1625     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1626 
1627     rc = fdt_pack(fdt);
1628 
1629     /* Should only fail if we've built a corrupted tree */
1630     assert(rc == 0);
1631 
1632     /* Load the fdt */
1633     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1634     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1635     g_free(spapr->fdt_blob);
1636     spapr->fdt_size = fdt_totalsize(fdt);
1637     spapr->fdt_initial_size = spapr->fdt_size;
1638     spapr->fdt_blob = fdt;
1639 
1640     /* Set up the entry state */
1641     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1642     first_ppc_cpu->env.gpr[5] = 0;
1643 
1644     spapr->fwnmi_system_reset_addr = -1;
1645     spapr->fwnmi_machine_check_addr = -1;
1646     spapr->fwnmi_machine_check_interlock = -1;
1647 
1648     /* Signal all vCPUs waiting on this condition */
1649     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1650 
1651     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1652 }
1653 
1654 static void spapr_create_nvram(SpaprMachineState *spapr)
1655 {
1656     DeviceState *dev = qdev_new("spapr-nvram");
1657     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1658 
1659     if (dinfo) {
1660         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1661                                 &error_fatal);
1662     }
1663 
1664     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1665 
1666     spapr->nvram = (struct SpaprNvram *)dev;
1667 }
1668 
1669 static void spapr_rtc_create(SpaprMachineState *spapr)
1670 {
1671     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1672                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1673                                        &error_fatal, NULL);
1674     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1675     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1676                               "date");
1677 }
1678 
1679 /* Returns whether we want to use VGA or not */
1680 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1681 {
1682     switch (vga_interface_type) {
1683     case VGA_NONE:
1684         return false;
1685     case VGA_DEVICE:
1686         return true;
1687     case VGA_STD:
1688     case VGA_VIRTIO:
1689     case VGA_CIRRUS:
1690         return pci_vga_init(pci_bus) != NULL;
1691     default:
1692         error_setg(errp,
1693                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1694         return false;
1695     }
1696 }
1697 
1698 static int spapr_pre_load(void *opaque)
1699 {
1700     int rc;
1701 
1702     rc = spapr_caps_pre_load(opaque);
1703     if (rc) {
1704         return rc;
1705     }
1706 
1707     return 0;
1708 }
1709 
1710 static int spapr_post_load(void *opaque, int version_id)
1711 {
1712     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1713     int err = 0;
1714 
1715     err = spapr_caps_post_migration(spapr);
1716     if (err) {
1717         return err;
1718     }
1719 
1720     /*
1721      * In earlier versions, there was no separate qdev for the PAPR
1722      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1723      * So when migrating from those versions, poke the incoming offset
1724      * value into the RTC device
1725      */
1726     if (version_id < 3) {
1727         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1728         if (err) {
1729             return err;
1730         }
1731     }
1732 
1733     if (kvm_enabled() && spapr->patb_entry) {
1734         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1735         bool radix = !!(spapr->patb_entry & PATE1_GR);
1736         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1737 
1738         /*
1739          * Update LPCR:HR and UPRT as they may not be set properly in
1740          * the stream
1741          */
1742         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1743                             LPCR_HR | LPCR_UPRT);
1744 
1745         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1746         if (err) {
1747             error_report("Process table config unsupported by the host");
1748             return -EINVAL;
1749         }
1750     }
1751 
1752     err = spapr_irq_post_load(spapr, version_id);
1753     if (err) {
1754         return err;
1755     }
1756 
1757     return err;
1758 }
1759 
1760 static int spapr_pre_save(void *opaque)
1761 {
1762     int rc;
1763 
1764     rc = spapr_caps_pre_save(opaque);
1765     if (rc) {
1766         return rc;
1767     }
1768 
1769     return 0;
1770 }
1771 
1772 static bool version_before_3(void *opaque, int version_id)
1773 {
1774     return version_id < 3;
1775 }
1776 
1777 static bool spapr_pending_events_needed(void *opaque)
1778 {
1779     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1780     return !QTAILQ_EMPTY(&spapr->pending_events);
1781 }
1782 
1783 static const VMStateDescription vmstate_spapr_event_entry = {
1784     .name = "spapr_event_log_entry",
1785     .version_id = 1,
1786     .minimum_version_id = 1,
1787     .fields = (VMStateField[]) {
1788         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1789         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1790         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1791                                      NULL, extended_length),
1792         VMSTATE_END_OF_LIST()
1793     },
1794 };
1795 
1796 static const VMStateDescription vmstate_spapr_pending_events = {
1797     .name = "spapr_pending_events",
1798     .version_id = 1,
1799     .minimum_version_id = 1,
1800     .needed = spapr_pending_events_needed,
1801     .fields = (VMStateField[]) {
1802         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1803                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1804         VMSTATE_END_OF_LIST()
1805     },
1806 };
1807 
1808 static bool spapr_ov5_cas_needed(void *opaque)
1809 {
1810     SpaprMachineState *spapr = opaque;
1811     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1812     bool cas_needed;
1813 
1814     /* Prior to the introduction of SpaprOptionVector, we had two option
1815      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1816      * Both of these options encode machine topology into the device-tree
1817      * in such a way that the now-booted OS should still be able to interact
1818      * appropriately with QEMU regardless of what options were actually
1819      * negotiatied on the source side.
1820      *
1821      * As such, we can avoid migrating the CAS-negotiated options if these
1822      * are the only options available on the current machine/platform.
1823      * Since these are the only options available for pseries-2.7 and
1824      * earlier, this allows us to maintain old->new/new->old migration
1825      * compatibility.
1826      *
1827      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1828      * via default pseries-2.8 machines and explicit command-line parameters.
1829      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1830      * of the actual CAS-negotiated values to continue working properly. For
1831      * example, availability of memory unplug depends on knowing whether
1832      * OV5_HP_EVT was negotiated via CAS.
1833      *
1834      * Thus, for any cases where the set of available CAS-negotiatable
1835      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1836      * include the CAS-negotiated options in the migration stream, unless
1837      * if they affect boot time behaviour only.
1838      */
1839     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1840     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1841     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1842 
1843     /* We need extra information if we have any bits outside the mask
1844      * defined above */
1845     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1846 
1847     spapr_ovec_cleanup(ov5_mask);
1848 
1849     return cas_needed;
1850 }
1851 
1852 static const VMStateDescription vmstate_spapr_ov5_cas = {
1853     .name = "spapr_option_vector_ov5_cas",
1854     .version_id = 1,
1855     .minimum_version_id = 1,
1856     .needed = spapr_ov5_cas_needed,
1857     .fields = (VMStateField[]) {
1858         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1859                                  vmstate_spapr_ovec, SpaprOptionVector),
1860         VMSTATE_END_OF_LIST()
1861     },
1862 };
1863 
1864 static bool spapr_patb_entry_needed(void *opaque)
1865 {
1866     SpaprMachineState *spapr = opaque;
1867 
1868     return !!spapr->patb_entry;
1869 }
1870 
1871 static const VMStateDescription vmstate_spapr_patb_entry = {
1872     .name = "spapr_patb_entry",
1873     .version_id = 1,
1874     .minimum_version_id = 1,
1875     .needed = spapr_patb_entry_needed,
1876     .fields = (VMStateField[]) {
1877         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1878         VMSTATE_END_OF_LIST()
1879     },
1880 };
1881 
1882 static bool spapr_irq_map_needed(void *opaque)
1883 {
1884     SpaprMachineState *spapr = opaque;
1885 
1886     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1887 }
1888 
1889 static const VMStateDescription vmstate_spapr_irq_map = {
1890     .name = "spapr_irq_map",
1891     .version_id = 1,
1892     .minimum_version_id = 1,
1893     .needed = spapr_irq_map_needed,
1894     .fields = (VMStateField[]) {
1895         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1896         VMSTATE_END_OF_LIST()
1897     },
1898 };
1899 
1900 static bool spapr_dtb_needed(void *opaque)
1901 {
1902     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1903 
1904     return smc->update_dt_enabled;
1905 }
1906 
1907 static int spapr_dtb_pre_load(void *opaque)
1908 {
1909     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1910 
1911     g_free(spapr->fdt_blob);
1912     spapr->fdt_blob = NULL;
1913     spapr->fdt_size = 0;
1914 
1915     return 0;
1916 }
1917 
1918 static const VMStateDescription vmstate_spapr_dtb = {
1919     .name = "spapr_dtb",
1920     .version_id = 1,
1921     .minimum_version_id = 1,
1922     .needed = spapr_dtb_needed,
1923     .pre_load = spapr_dtb_pre_load,
1924     .fields = (VMStateField[]) {
1925         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1926         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1927         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1928                                      fdt_size),
1929         VMSTATE_END_OF_LIST()
1930     },
1931 };
1932 
1933 static bool spapr_fwnmi_needed(void *opaque)
1934 {
1935     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1936 
1937     return spapr->fwnmi_machine_check_addr != -1;
1938 }
1939 
1940 static int spapr_fwnmi_pre_save(void *opaque)
1941 {
1942     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1943 
1944     /*
1945      * Check if machine check handling is in progress and print a
1946      * warning message.
1947      */
1948     if (spapr->fwnmi_machine_check_interlock != -1) {
1949         warn_report("A machine check is being handled during migration. The"
1950                 "handler may run and log hardware error on the destination");
1951     }
1952 
1953     return 0;
1954 }
1955 
1956 static const VMStateDescription vmstate_spapr_fwnmi = {
1957     .name = "spapr_fwnmi",
1958     .version_id = 1,
1959     .minimum_version_id = 1,
1960     .needed = spapr_fwnmi_needed,
1961     .pre_save = spapr_fwnmi_pre_save,
1962     .fields = (VMStateField[]) {
1963         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1964         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1965         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1966         VMSTATE_END_OF_LIST()
1967     },
1968 };
1969 
1970 static const VMStateDescription vmstate_spapr = {
1971     .name = "spapr",
1972     .version_id = 3,
1973     .minimum_version_id = 1,
1974     .pre_load = spapr_pre_load,
1975     .post_load = spapr_post_load,
1976     .pre_save = spapr_pre_save,
1977     .fields = (VMStateField[]) {
1978         /* used to be @next_irq */
1979         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1980 
1981         /* RTC offset */
1982         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
1983 
1984         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
1985         VMSTATE_END_OF_LIST()
1986     },
1987     .subsections = (const VMStateDescription*[]) {
1988         &vmstate_spapr_ov5_cas,
1989         &vmstate_spapr_patb_entry,
1990         &vmstate_spapr_pending_events,
1991         &vmstate_spapr_cap_htm,
1992         &vmstate_spapr_cap_vsx,
1993         &vmstate_spapr_cap_dfp,
1994         &vmstate_spapr_cap_cfpc,
1995         &vmstate_spapr_cap_sbbc,
1996         &vmstate_spapr_cap_ibs,
1997         &vmstate_spapr_cap_hpt_maxpagesize,
1998         &vmstate_spapr_irq_map,
1999         &vmstate_spapr_cap_nested_kvm_hv,
2000         &vmstate_spapr_dtb,
2001         &vmstate_spapr_cap_large_decr,
2002         &vmstate_spapr_cap_ccf_assist,
2003         &vmstate_spapr_cap_fwnmi,
2004         &vmstate_spapr_fwnmi,
2005         NULL
2006     }
2007 };
2008 
2009 static int htab_save_setup(QEMUFile *f, void *opaque)
2010 {
2011     SpaprMachineState *spapr = opaque;
2012 
2013     /* "Iteration" header */
2014     if (!spapr->htab_shift) {
2015         qemu_put_be32(f, -1);
2016     } else {
2017         qemu_put_be32(f, spapr->htab_shift);
2018     }
2019 
2020     if (spapr->htab) {
2021         spapr->htab_save_index = 0;
2022         spapr->htab_first_pass = true;
2023     } else {
2024         if (spapr->htab_shift) {
2025             assert(kvm_enabled());
2026         }
2027     }
2028 
2029 
2030     return 0;
2031 }
2032 
2033 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2034                             int chunkstart, int n_valid, int n_invalid)
2035 {
2036     qemu_put_be32(f, chunkstart);
2037     qemu_put_be16(f, n_valid);
2038     qemu_put_be16(f, n_invalid);
2039     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2040                     HASH_PTE_SIZE_64 * n_valid);
2041 }
2042 
2043 static void htab_save_end_marker(QEMUFile *f)
2044 {
2045     qemu_put_be32(f, 0);
2046     qemu_put_be16(f, 0);
2047     qemu_put_be16(f, 0);
2048 }
2049 
2050 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2051                                  int64_t max_ns)
2052 {
2053     bool has_timeout = max_ns != -1;
2054     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2055     int index = spapr->htab_save_index;
2056     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2057 
2058     assert(spapr->htab_first_pass);
2059 
2060     do {
2061         int chunkstart;
2062 
2063         /* Consume invalid HPTEs */
2064         while ((index < htabslots)
2065                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2066             CLEAN_HPTE(HPTE(spapr->htab, index));
2067             index++;
2068         }
2069 
2070         /* Consume valid HPTEs */
2071         chunkstart = index;
2072         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2073                && HPTE_VALID(HPTE(spapr->htab, index))) {
2074             CLEAN_HPTE(HPTE(spapr->htab, index));
2075             index++;
2076         }
2077 
2078         if (index > chunkstart) {
2079             int n_valid = index - chunkstart;
2080 
2081             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2082 
2083             if (has_timeout &&
2084                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2085                 break;
2086             }
2087         }
2088     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2089 
2090     if (index >= htabslots) {
2091         assert(index == htabslots);
2092         index = 0;
2093         spapr->htab_first_pass = false;
2094     }
2095     spapr->htab_save_index = index;
2096 }
2097 
2098 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2099                                 int64_t max_ns)
2100 {
2101     bool final = max_ns < 0;
2102     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2103     int examined = 0, sent = 0;
2104     int index = spapr->htab_save_index;
2105     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2106 
2107     assert(!spapr->htab_first_pass);
2108 
2109     do {
2110         int chunkstart, invalidstart;
2111 
2112         /* Consume non-dirty HPTEs */
2113         while ((index < htabslots)
2114                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2115             index++;
2116             examined++;
2117         }
2118 
2119         chunkstart = index;
2120         /* Consume valid dirty HPTEs */
2121         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2122                && HPTE_DIRTY(HPTE(spapr->htab, index))
2123                && HPTE_VALID(HPTE(spapr->htab, index))) {
2124             CLEAN_HPTE(HPTE(spapr->htab, index));
2125             index++;
2126             examined++;
2127         }
2128 
2129         invalidstart = index;
2130         /* Consume invalid dirty HPTEs */
2131         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2132                && HPTE_DIRTY(HPTE(spapr->htab, index))
2133                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2134             CLEAN_HPTE(HPTE(spapr->htab, index));
2135             index++;
2136             examined++;
2137         }
2138 
2139         if (index > chunkstart) {
2140             int n_valid = invalidstart - chunkstart;
2141             int n_invalid = index - invalidstart;
2142 
2143             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2144             sent += index - chunkstart;
2145 
2146             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2147                 break;
2148             }
2149         }
2150 
2151         if (examined >= htabslots) {
2152             break;
2153         }
2154 
2155         if (index >= htabslots) {
2156             assert(index == htabslots);
2157             index = 0;
2158         }
2159     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2160 
2161     if (index >= htabslots) {
2162         assert(index == htabslots);
2163         index = 0;
2164     }
2165 
2166     spapr->htab_save_index = index;
2167 
2168     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2169 }
2170 
2171 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2172 #define MAX_KVM_BUF_SIZE    2048
2173 
2174 static int htab_save_iterate(QEMUFile *f, void *opaque)
2175 {
2176     SpaprMachineState *spapr = opaque;
2177     int fd;
2178     int rc = 0;
2179 
2180     /* Iteration header */
2181     if (!spapr->htab_shift) {
2182         qemu_put_be32(f, -1);
2183         return 1;
2184     } else {
2185         qemu_put_be32(f, 0);
2186     }
2187 
2188     if (!spapr->htab) {
2189         assert(kvm_enabled());
2190 
2191         fd = get_htab_fd(spapr);
2192         if (fd < 0) {
2193             return fd;
2194         }
2195 
2196         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2197         if (rc < 0) {
2198             return rc;
2199         }
2200     } else  if (spapr->htab_first_pass) {
2201         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2202     } else {
2203         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2204     }
2205 
2206     htab_save_end_marker(f);
2207 
2208     return rc;
2209 }
2210 
2211 static int htab_save_complete(QEMUFile *f, void *opaque)
2212 {
2213     SpaprMachineState *spapr = opaque;
2214     int fd;
2215 
2216     /* Iteration header */
2217     if (!spapr->htab_shift) {
2218         qemu_put_be32(f, -1);
2219         return 0;
2220     } else {
2221         qemu_put_be32(f, 0);
2222     }
2223 
2224     if (!spapr->htab) {
2225         int rc;
2226 
2227         assert(kvm_enabled());
2228 
2229         fd = get_htab_fd(spapr);
2230         if (fd < 0) {
2231             return fd;
2232         }
2233 
2234         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2235         if (rc < 0) {
2236             return rc;
2237         }
2238     } else {
2239         if (spapr->htab_first_pass) {
2240             htab_save_first_pass(f, spapr, -1);
2241         }
2242         htab_save_later_pass(f, spapr, -1);
2243     }
2244 
2245     /* End marker */
2246     htab_save_end_marker(f);
2247 
2248     return 0;
2249 }
2250 
2251 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2252 {
2253     SpaprMachineState *spapr = opaque;
2254     uint32_t section_hdr;
2255     int fd = -1;
2256     Error *local_err = NULL;
2257 
2258     if (version_id < 1 || version_id > 1) {
2259         error_report("htab_load() bad version");
2260         return -EINVAL;
2261     }
2262 
2263     section_hdr = qemu_get_be32(f);
2264 
2265     if (section_hdr == -1) {
2266         spapr_free_hpt(spapr);
2267         return 0;
2268     }
2269 
2270     if (section_hdr) {
2271         int ret;
2272 
2273         /* First section gives the htab size */
2274         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2275         if (ret < 0) {
2276             error_report_err(local_err);
2277             return ret;
2278         }
2279         return 0;
2280     }
2281 
2282     if (!spapr->htab) {
2283         assert(kvm_enabled());
2284 
2285         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2286         if (fd < 0) {
2287             error_report_err(local_err);
2288             return fd;
2289         }
2290     }
2291 
2292     while (true) {
2293         uint32_t index;
2294         uint16_t n_valid, n_invalid;
2295 
2296         index = qemu_get_be32(f);
2297         n_valid = qemu_get_be16(f);
2298         n_invalid = qemu_get_be16(f);
2299 
2300         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2301             /* End of Stream */
2302             break;
2303         }
2304 
2305         if ((index + n_valid + n_invalid) >
2306             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2307             /* Bad index in stream */
2308             error_report(
2309                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2310                 index, n_valid, n_invalid, spapr->htab_shift);
2311             return -EINVAL;
2312         }
2313 
2314         if (spapr->htab) {
2315             if (n_valid) {
2316                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2317                                 HASH_PTE_SIZE_64 * n_valid);
2318             }
2319             if (n_invalid) {
2320                 memset(HPTE(spapr->htab, index + n_valid), 0,
2321                        HASH_PTE_SIZE_64 * n_invalid);
2322             }
2323         } else {
2324             int rc;
2325 
2326             assert(fd >= 0);
2327 
2328             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2329                                         &local_err);
2330             if (rc < 0) {
2331                 error_report_err(local_err);
2332                 return rc;
2333             }
2334         }
2335     }
2336 
2337     if (!spapr->htab) {
2338         assert(fd >= 0);
2339         close(fd);
2340     }
2341 
2342     return 0;
2343 }
2344 
2345 static void htab_save_cleanup(void *opaque)
2346 {
2347     SpaprMachineState *spapr = opaque;
2348 
2349     close_htab_fd(spapr);
2350 }
2351 
2352 static SaveVMHandlers savevm_htab_handlers = {
2353     .save_setup = htab_save_setup,
2354     .save_live_iterate = htab_save_iterate,
2355     .save_live_complete_precopy = htab_save_complete,
2356     .save_cleanup = htab_save_cleanup,
2357     .load_state = htab_load,
2358 };
2359 
2360 static void spapr_boot_set(void *opaque, const char *boot_device,
2361                            Error **errp)
2362 {
2363     MachineState *machine = MACHINE(opaque);
2364     machine->boot_order = g_strdup(boot_device);
2365 }
2366 
2367 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2368 {
2369     MachineState *machine = MACHINE(spapr);
2370     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2371     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2372     int i;
2373 
2374     for (i = 0; i < nr_lmbs; i++) {
2375         uint64_t addr;
2376 
2377         addr = i * lmb_size + machine->device_memory->base;
2378         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2379                                addr / lmb_size);
2380     }
2381 }
2382 
2383 /*
2384  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2385  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2386  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2387  */
2388 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2389 {
2390     int i;
2391 
2392     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2393         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2394                    " is not aligned to %" PRIu64 " MiB",
2395                    machine->ram_size,
2396                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2397         return;
2398     }
2399 
2400     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2401         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2402                    " is not aligned to %" PRIu64 " MiB",
2403                    machine->ram_size,
2404                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2405         return;
2406     }
2407 
2408     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2409         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2410             error_setg(errp,
2411                        "Node %d memory size 0x%" PRIx64
2412                        " is not aligned to %" PRIu64 " MiB",
2413                        i, machine->numa_state->nodes[i].node_mem,
2414                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2415             return;
2416         }
2417     }
2418 }
2419 
2420 /* find cpu slot in machine->possible_cpus by core_id */
2421 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2422 {
2423     int index = id / ms->smp.threads;
2424 
2425     if (index >= ms->possible_cpus->len) {
2426         return NULL;
2427     }
2428     if (idx) {
2429         *idx = index;
2430     }
2431     return &ms->possible_cpus->cpus[index];
2432 }
2433 
2434 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2435 {
2436     MachineState *ms = MACHINE(spapr);
2437     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2438     Error *local_err = NULL;
2439     bool vsmt_user = !!spapr->vsmt;
2440     int kvm_smt = kvmppc_smt_threads();
2441     int ret;
2442     unsigned int smp_threads = ms->smp.threads;
2443 
2444     if (!kvm_enabled() && (smp_threads > 1)) {
2445         error_setg(errp, "TCG cannot support more than 1 thread/core "
2446                    "on a pseries machine");
2447         return;
2448     }
2449     if (!is_power_of_2(smp_threads)) {
2450         error_setg(errp, "Cannot support %d threads/core on a pseries "
2451                    "machine because it must be a power of 2", smp_threads);
2452         return;
2453     }
2454 
2455     /* Detemine the VSMT mode to use: */
2456     if (vsmt_user) {
2457         if (spapr->vsmt < smp_threads) {
2458             error_setg(errp, "Cannot support VSMT mode %d"
2459                        " because it must be >= threads/core (%d)",
2460                        spapr->vsmt, smp_threads);
2461             return;
2462         }
2463         /* In this case, spapr->vsmt has been set by the command line */
2464     } else if (!smc->smp_threads_vsmt) {
2465         /*
2466          * Default VSMT value is tricky, because we need it to be as
2467          * consistent as possible (for migration), but this requires
2468          * changing it for at least some existing cases.  We pick 8 as
2469          * the value that we'd get with KVM on POWER8, the
2470          * overwhelmingly common case in production systems.
2471          */
2472         spapr->vsmt = MAX(8, smp_threads);
2473     } else {
2474         spapr->vsmt = smp_threads;
2475     }
2476 
2477     /* KVM: If necessary, set the SMT mode: */
2478     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2479         ret = kvmppc_set_smt_threads(spapr->vsmt);
2480         if (ret) {
2481             /* Looks like KVM isn't able to change VSMT mode */
2482             error_setg(&local_err,
2483                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2484                        spapr->vsmt, ret);
2485             /* We can live with that if the default one is big enough
2486              * for the number of threads, and a submultiple of the one
2487              * we want.  In this case we'll waste some vcpu ids, but
2488              * behaviour will be correct */
2489             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2490                 warn_report_err(local_err);
2491             } else {
2492                 if (!vsmt_user) {
2493                     error_append_hint(&local_err,
2494                                       "On PPC, a VM with %d threads/core"
2495                                       " on a host with %d threads/core"
2496                                       " requires the use of VSMT mode %d.\n",
2497                                       smp_threads, kvm_smt, spapr->vsmt);
2498                 }
2499                 kvmppc_error_append_smt_possible_hint(&local_err);
2500                 error_propagate(errp, local_err);
2501             }
2502         }
2503     }
2504     /* else TCG: nothing to do currently */
2505 }
2506 
2507 static void spapr_init_cpus(SpaprMachineState *spapr)
2508 {
2509     MachineState *machine = MACHINE(spapr);
2510     MachineClass *mc = MACHINE_GET_CLASS(machine);
2511     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2512     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2513     const CPUArchIdList *possible_cpus;
2514     unsigned int smp_cpus = machine->smp.cpus;
2515     unsigned int smp_threads = machine->smp.threads;
2516     unsigned int max_cpus = machine->smp.max_cpus;
2517     int boot_cores_nr = smp_cpus / smp_threads;
2518     int i;
2519 
2520     possible_cpus = mc->possible_cpu_arch_ids(machine);
2521     if (mc->has_hotpluggable_cpus) {
2522         if (smp_cpus % smp_threads) {
2523             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2524                          smp_cpus, smp_threads);
2525             exit(1);
2526         }
2527         if (max_cpus % smp_threads) {
2528             error_report("max_cpus (%u) must be multiple of threads (%u)",
2529                          max_cpus, smp_threads);
2530             exit(1);
2531         }
2532     } else {
2533         if (max_cpus != smp_cpus) {
2534             error_report("This machine version does not support CPU hotplug");
2535             exit(1);
2536         }
2537         boot_cores_nr = possible_cpus->len;
2538     }
2539 
2540     if (smc->pre_2_10_has_unused_icps) {
2541         int i;
2542 
2543         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2544             /* Dummy entries get deregistered when real ICPState objects
2545              * are registered during CPU core hotplug.
2546              */
2547             pre_2_10_vmstate_register_dummy_icp(i);
2548         }
2549     }
2550 
2551     for (i = 0; i < possible_cpus->len; i++) {
2552         int core_id = i * smp_threads;
2553 
2554         if (mc->has_hotpluggable_cpus) {
2555             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2556                                    spapr_vcpu_id(spapr, core_id));
2557         }
2558 
2559         if (i < boot_cores_nr) {
2560             Object *core  = object_new(type);
2561             int nr_threads = smp_threads;
2562 
2563             /* Handle the partially filled core for older machine types */
2564             if ((i + 1) * smp_threads >= smp_cpus) {
2565                 nr_threads = smp_cpus - i * smp_threads;
2566             }
2567 
2568             object_property_set_int(core, "nr-threads", nr_threads,
2569                                     &error_fatal);
2570             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2571                                     &error_fatal);
2572             qdev_realize(DEVICE(core), NULL, &error_fatal);
2573 
2574             object_unref(core);
2575         }
2576     }
2577 }
2578 
2579 static PCIHostState *spapr_create_default_phb(void)
2580 {
2581     DeviceState *dev;
2582 
2583     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2584     qdev_prop_set_uint32(dev, "index", 0);
2585     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2586 
2587     return PCI_HOST_BRIDGE(dev);
2588 }
2589 
2590 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2591 {
2592     MachineState *machine = MACHINE(spapr);
2593     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2594     hwaddr rma_size = machine->ram_size;
2595     hwaddr node0_size = spapr_node0_size(machine);
2596 
2597     /* RMA has to fit in the first NUMA node */
2598     rma_size = MIN(rma_size, node0_size);
2599 
2600     /*
2601      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2602      * never exceed that
2603      */
2604     rma_size = MIN(rma_size, 1 * TiB);
2605 
2606     /*
2607      * Clamp the RMA size based on machine type.  This is for
2608      * migration compatibility with older qemu versions, which limited
2609      * the RMA size for complicated and mostly bad reasons.
2610      */
2611     if (smc->rma_limit) {
2612         rma_size = MIN(rma_size, smc->rma_limit);
2613     }
2614 
2615     if (rma_size < MIN_RMA_SLOF) {
2616         error_setg(errp,
2617                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2618                    "ldMiB guest RMA (Real Mode Area memory)",
2619                    MIN_RMA_SLOF / MiB);
2620         return 0;
2621     }
2622 
2623     return rma_size;
2624 }
2625 
2626 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2627 {
2628     MachineState *machine = MACHINE(spapr);
2629     int i;
2630 
2631     for (i = 0; i < machine->ram_slots; i++) {
2632         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2633     }
2634 }
2635 
2636 /* pSeries LPAR / sPAPR hardware init */
2637 static void spapr_machine_init(MachineState *machine)
2638 {
2639     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2640     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2641     MachineClass *mc = MACHINE_GET_CLASS(machine);
2642     const char *bios_name = machine->firmware ?: FW_FILE_NAME;
2643     const char *kernel_filename = machine->kernel_filename;
2644     const char *initrd_filename = machine->initrd_filename;
2645     PCIHostState *phb;
2646     int i;
2647     MemoryRegion *sysmem = get_system_memory();
2648     long load_limit, fw_size;
2649     char *filename;
2650     Error *resize_hpt_err = NULL;
2651 
2652     /*
2653      * if Secure VM (PEF) support is configured, then initialize it
2654      */
2655     pef_kvm_init(machine->cgs, &error_fatal);
2656 
2657     msi_nonbroken = true;
2658 
2659     QLIST_INIT(&spapr->phbs);
2660     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2661 
2662     /* Determine capabilities to run with */
2663     spapr_caps_init(spapr);
2664 
2665     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2666     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2667         /*
2668          * If the user explicitly requested a mode we should either
2669          * supply it, or fail completely (which we do below).  But if
2670          * it's not set explicitly, we reset our mode to something
2671          * that works
2672          */
2673         if (resize_hpt_err) {
2674             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2675             error_free(resize_hpt_err);
2676             resize_hpt_err = NULL;
2677         } else {
2678             spapr->resize_hpt = smc->resize_hpt_default;
2679         }
2680     }
2681 
2682     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2683 
2684     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2685         /*
2686          * User requested HPT resize, but this host can't supply it.  Bail out
2687          */
2688         error_report_err(resize_hpt_err);
2689         exit(1);
2690     }
2691     error_free(resize_hpt_err);
2692 
2693     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2694 
2695     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2696     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2697 
2698     /*
2699      * VSMT must be set in order to be able to compute VCPU ids, ie to
2700      * call spapr_max_server_number() or spapr_vcpu_id().
2701      */
2702     spapr_set_vsmt_mode(spapr, &error_fatal);
2703 
2704     /* Set up Interrupt Controller before we create the VCPUs */
2705     spapr_irq_init(spapr, &error_fatal);
2706 
2707     /* Set up containers for ibm,client-architecture-support negotiated options
2708      */
2709     spapr->ov5 = spapr_ovec_new();
2710     spapr->ov5_cas = spapr_ovec_new();
2711 
2712     if (smc->dr_lmb_enabled) {
2713         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2714         spapr_validate_node_memory(machine, &error_fatal);
2715     }
2716 
2717     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2718 
2719     /* advertise support for dedicated HP event source to guests */
2720     if (spapr->use_hotplug_event_source) {
2721         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2722     }
2723 
2724     /* advertise support for HPT resizing */
2725     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2726         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2727     }
2728 
2729     /* advertise support for ibm,dyamic-memory-v2 */
2730     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2731 
2732     /* advertise XIVE on POWER9 machines */
2733     if (spapr->irq->xive) {
2734         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2735     }
2736 
2737     /* init CPUs */
2738     spapr_init_cpus(spapr);
2739 
2740     /*
2741      * check we don't have a memory-less/cpu-less NUMA node
2742      * Firmware relies on the existing memory/cpu topology to provide the
2743      * NUMA topology to the kernel.
2744      * And the linux kernel needs to know the NUMA topology at start
2745      * to be able to hotplug CPUs later.
2746      */
2747     if (machine->numa_state->num_nodes) {
2748         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2749             /* check for memory-less node */
2750             if (machine->numa_state->nodes[i].node_mem == 0) {
2751                 CPUState *cs;
2752                 int found = 0;
2753                 /* check for cpu-less node */
2754                 CPU_FOREACH(cs) {
2755                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2756                     if (cpu->node_id == i) {
2757                         found = 1;
2758                         break;
2759                     }
2760                 }
2761                 /* memory-less and cpu-less node */
2762                 if (!found) {
2763                     error_report(
2764                        "Memory-less/cpu-less nodes are not supported (node %d)",
2765                                  i);
2766                     exit(1);
2767                 }
2768             }
2769         }
2770 
2771     }
2772 
2773     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2774 
2775     /* Init numa_assoc_array */
2776     spapr_numa_associativity_init(spapr, machine);
2777 
2778     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2779         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2780                               spapr->max_compat_pvr)) {
2781         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2782         /* KVM and TCG always allow GTSE with radix... */
2783         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2784     }
2785     /* ... but not with hash (currently). */
2786 
2787     if (kvm_enabled()) {
2788         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2789         kvmppc_enable_logical_ci_hcalls();
2790         kvmppc_enable_set_mode_hcall();
2791 
2792         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2793         kvmppc_enable_clear_ref_mod_hcalls();
2794 
2795         /* Enable H_PAGE_INIT */
2796         kvmppc_enable_h_page_init();
2797     }
2798 
2799     /* map RAM */
2800     memory_region_add_subregion(sysmem, 0, machine->ram);
2801 
2802     /* always allocate the device memory information */
2803     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2804 
2805     /* initialize hotplug memory address space */
2806     if (machine->ram_size < machine->maxram_size) {
2807         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2808         /*
2809          * Limit the number of hotpluggable memory slots to half the number
2810          * slots that KVM supports, leaving the other half for PCI and other
2811          * devices. However ensure that number of slots doesn't drop below 32.
2812          */
2813         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2814                            SPAPR_MAX_RAM_SLOTS;
2815 
2816         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2817             max_memslots = SPAPR_MAX_RAM_SLOTS;
2818         }
2819         if (machine->ram_slots > max_memslots) {
2820             error_report("Specified number of memory slots %"
2821                          PRIu64" exceeds max supported %d",
2822                          machine->ram_slots, max_memslots);
2823             exit(1);
2824         }
2825 
2826         machine->device_memory->base = ROUND_UP(machine->ram_size,
2827                                                 SPAPR_DEVICE_MEM_ALIGN);
2828         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2829                            "device-memory", device_mem_size);
2830         memory_region_add_subregion(sysmem, machine->device_memory->base,
2831                                     &machine->device_memory->mr);
2832     }
2833 
2834     if (smc->dr_lmb_enabled) {
2835         spapr_create_lmb_dr_connectors(spapr);
2836     }
2837 
2838     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2839         /* Create the error string for live migration blocker */
2840         error_setg(&spapr->fwnmi_migration_blocker,
2841             "A machine check is being handled during migration. The handler"
2842             "may run and log hardware error on the destination");
2843     }
2844 
2845     if (mc->nvdimm_supported) {
2846         spapr_create_nvdimm_dr_connectors(spapr);
2847     }
2848 
2849     /* Set up RTAS event infrastructure */
2850     spapr_events_init(spapr);
2851 
2852     /* Set up the RTC RTAS interfaces */
2853     spapr_rtc_create(spapr);
2854 
2855     /* Set up VIO bus */
2856     spapr->vio_bus = spapr_vio_bus_init();
2857 
2858     for (i = 0; serial_hd(i); i++) {
2859         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2860     }
2861 
2862     /* We always have at least the nvram device on VIO */
2863     spapr_create_nvram(spapr);
2864 
2865     /*
2866      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2867      * connectors (described in root DT node's "ibm,drc-types" property)
2868      * are pre-initialized here. additional child connectors (such as
2869      * connectors for a PHBs PCI slots) are added as needed during their
2870      * parent's realization.
2871      */
2872     if (smc->dr_phb_enabled) {
2873         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2874             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2875         }
2876     }
2877 
2878     /* Set up PCI */
2879     spapr_pci_rtas_init();
2880 
2881     phb = spapr_create_default_phb();
2882 
2883     for (i = 0; i < nb_nics; i++) {
2884         NICInfo *nd = &nd_table[i];
2885 
2886         if (!nd->model) {
2887             nd->model = g_strdup("spapr-vlan");
2888         }
2889 
2890         if (g_str_equal(nd->model, "spapr-vlan") ||
2891             g_str_equal(nd->model, "ibmveth")) {
2892             spapr_vlan_create(spapr->vio_bus, nd);
2893         } else {
2894             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2895         }
2896     }
2897 
2898     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2899         spapr_vscsi_create(spapr->vio_bus);
2900     }
2901 
2902     /* Graphics */
2903     if (spapr_vga_init(phb->bus, &error_fatal)) {
2904         spapr->has_graphics = true;
2905         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2906     }
2907 
2908     if (machine->usb) {
2909         if (smc->use_ohci_by_default) {
2910             pci_create_simple(phb->bus, -1, "pci-ohci");
2911         } else {
2912             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2913         }
2914 
2915         if (spapr->has_graphics) {
2916             USBBus *usb_bus = usb_bus_find(-1);
2917 
2918             usb_create_simple(usb_bus, "usb-kbd");
2919             usb_create_simple(usb_bus, "usb-mouse");
2920         }
2921     }
2922 
2923     if (kernel_filename) {
2924         spapr->kernel_size = load_elf(kernel_filename, NULL,
2925                                       translate_kernel_address, spapr,
2926                                       NULL, NULL, NULL, NULL, 1,
2927                                       PPC_ELF_MACHINE, 0, 0);
2928         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2929             spapr->kernel_size = load_elf(kernel_filename, NULL,
2930                                           translate_kernel_address, spapr,
2931                                           NULL, NULL, NULL, NULL, 0,
2932                                           PPC_ELF_MACHINE, 0, 0);
2933             spapr->kernel_le = spapr->kernel_size > 0;
2934         }
2935         if (spapr->kernel_size < 0) {
2936             error_report("error loading %s: %s", kernel_filename,
2937                          load_elf_strerror(spapr->kernel_size));
2938             exit(1);
2939         }
2940 
2941         /* load initrd */
2942         if (initrd_filename) {
2943             /* Try to locate the initrd in the gap between the kernel
2944              * and the firmware. Add a bit of space just in case
2945              */
2946             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2947                                   + 0x1ffff) & ~0xffff;
2948             spapr->initrd_size = load_image_targphys(initrd_filename,
2949                                                      spapr->initrd_base,
2950                                                      load_limit
2951                                                      - spapr->initrd_base);
2952             if (spapr->initrd_size < 0) {
2953                 error_report("could not load initial ram disk '%s'",
2954                              initrd_filename);
2955                 exit(1);
2956             }
2957         }
2958     }
2959 
2960     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2961     if (!filename) {
2962         error_report("Could not find LPAR firmware '%s'", bios_name);
2963         exit(1);
2964     }
2965     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2966     if (fw_size <= 0) {
2967         error_report("Could not load LPAR firmware '%s'", filename);
2968         exit(1);
2969     }
2970     g_free(filename);
2971 
2972     /* FIXME: Should register things through the MachineState's qdev
2973      * interface, this is a legacy from the sPAPREnvironment structure
2974      * which predated MachineState but had a similar function */
2975     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2976     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
2977                          &savevm_htab_handlers, spapr);
2978 
2979     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
2980 
2981     qemu_register_boot_set(spapr_boot_set, spapr);
2982 
2983     /*
2984      * Nothing needs to be done to resume a suspended guest because
2985      * suspending does not change the machine state, so no need for
2986      * a ->wakeup method.
2987      */
2988     qemu_register_wakeup_support();
2989 
2990     if (kvm_enabled()) {
2991         /* to stop and start vmclock */
2992         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2993                                          &spapr->tb);
2994 
2995         kvmppc_spapr_enable_inkernel_multitce();
2996     }
2997 
2998     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
2999 }
3000 
3001 #define DEFAULT_KVM_TYPE "auto"
3002 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3003 {
3004     /*
3005      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3006      * accomodate the 'HV' and 'PV' formats that exists in the
3007      * wild. The 'auto' mode is being introduced already as
3008      * lower-case, thus we don't need to bother checking for
3009      * "AUTO".
3010      */
3011     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3012         return 0;
3013     }
3014 
3015     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3016         return 1;
3017     }
3018 
3019     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3020         return 2;
3021     }
3022 
3023     error_report("Unknown kvm-type specified '%s'", vm_type);
3024     exit(1);
3025 }
3026 
3027 /*
3028  * Implementation of an interface to adjust firmware path
3029  * for the bootindex property handling.
3030  */
3031 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3032                                    DeviceState *dev)
3033 {
3034 #define CAST(type, obj, name) \
3035     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3036     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3037     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3038     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3039     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3040 
3041     if (d) {
3042         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3043         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3044         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3045 
3046         if (spapr) {
3047             /*
3048              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3049              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3050              * 0x8000 | (target << 8) | (bus << 5) | lun
3051              * (see the "Logical unit addressing format" table in SAM5)
3052              */
3053             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3054             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3055                                    (uint64_t)id << 48);
3056         } else if (virtio) {
3057             /*
3058              * We use SRP luns of the form 01000000 | (target << 8) | lun
3059              * in the top 32 bits of the 64-bit LUN
3060              * Note: the quote above is from SLOF and it is wrong,
3061              * the actual binding is:
3062              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3063              */
3064             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3065             if (d->lun >= 256) {
3066                 /* Use the LUN "flat space addressing method" */
3067                 id |= 0x4000;
3068             }
3069             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3070                                    (uint64_t)id << 32);
3071         } else if (usb) {
3072             /*
3073              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3074              * in the top 32 bits of the 64-bit LUN
3075              */
3076             unsigned usb_port = atoi(usb->port->path);
3077             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3078             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3079                                    (uint64_t)id << 32);
3080         }
3081     }
3082 
3083     /*
3084      * SLOF probes the USB devices, and if it recognizes that the device is a
3085      * storage device, it changes its name to "storage" instead of "usb-host",
3086      * and additionally adds a child node for the SCSI LUN, so the correct
3087      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3088      */
3089     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3090         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3091         if (usb_host_dev_is_scsi_storage(usbdev)) {
3092             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3093         }
3094     }
3095 
3096     if (phb) {
3097         /* Replace "pci" with "pci@800000020000000" */
3098         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3099     }
3100 
3101     if (vsc) {
3102         /* Same logic as virtio above */
3103         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3104         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3105     }
3106 
3107     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3108         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3109         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3110         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3111     }
3112 
3113     if (pcidev) {
3114         return spapr_pci_fw_dev_name(pcidev);
3115     }
3116 
3117     return NULL;
3118 }
3119 
3120 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3121 {
3122     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3123 
3124     return g_strdup(spapr->kvm_type);
3125 }
3126 
3127 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3128 {
3129     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3130 
3131     g_free(spapr->kvm_type);
3132     spapr->kvm_type = g_strdup(value);
3133 }
3134 
3135 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3136 {
3137     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3138 
3139     return spapr->use_hotplug_event_source;
3140 }
3141 
3142 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3143                                             Error **errp)
3144 {
3145     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3146 
3147     spapr->use_hotplug_event_source = value;
3148 }
3149 
3150 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3151 {
3152     return true;
3153 }
3154 
3155 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3156 {
3157     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3158 
3159     switch (spapr->resize_hpt) {
3160     case SPAPR_RESIZE_HPT_DEFAULT:
3161         return g_strdup("default");
3162     case SPAPR_RESIZE_HPT_DISABLED:
3163         return g_strdup("disabled");
3164     case SPAPR_RESIZE_HPT_ENABLED:
3165         return g_strdup("enabled");
3166     case SPAPR_RESIZE_HPT_REQUIRED:
3167         return g_strdup("required");
3168     }
3169     g_assert_not_reached();
3170 }
3171 
3172 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3173 {
3174     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3175 
3176     if (strcmp(value, "default") == 0) {
3177         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3178     } else if (strcmp(value, "disabled") == 0) {
3179         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3180     } else if (strcmp(value, "enabled") == 0) {
3181         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3182     } else if (strcmp(value, "required") == 0) {
3183         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3184     } else {
3185         error_setg(errp, "Bad value for \"resize-hpt\" property");
3186     }
3187 }
3188 
3189 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3190 {
3191     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3192 
3193     if (spapr->irq == &spapr_irq_xics_legacy) {
3194         return g_strdup("legacy");
3195     } else if (spapr->irq == &spapr_irq_xics) {
3196         return g_strdup("xics");
3197     } else if (spapr->irq == &spapr_irq_xive) {
3198         return g_strdup("xive");
3199     } else if (spapr->irq == &spapr_irq_dual) {
3200         return g_strdup("dual");
3201     }
3202     g_assert_not_reached();
3203 }
3204 
3205 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3206 {
3207     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3208 
3209     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3210         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3211         return;
3212     }
3213 
3214     /* The legacy IRQ backend can not be set */
3215     if (strcmp(value, "xics") == 0) {
3216         spapr->irq = &spapr_irq_xics;
3217     } else if (strcmp(value, "xive") == 0) {
3218         spapr->irq = &spapr_irq_xive;
3219     } else if (strcmp(value, "dual") == 0) {
3220         spapr->irq = &spapr_irq_dual;
3221     } else {
3222         error_setg(errp, "Bad value for \"ic-mode\" property");
3223     }
3224 }
3225 
3226 static char *spapr_get_host_model(Object *obj, Error **errp)
3227 {
3228     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3229 
3230     return g_strdup(spapr->host_model);
3231 }
3232 
3233 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3234 {
3235     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3236 
3237     g_free(spapr->host_model);
3238     spapr->host_model = g_strdup(value);
3239 }
3240 
3241 static char *spapr_get_host_serial(Object *obj, Error **errp)
3242 {
3243     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3244 
3245     return g_strdup(spapr->host_serial);
3246 }
3247 
3248 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3249 {
3250     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3251 
3252     g_free(spapr->host_serial);
3253     spapr->host_serial = g_strdup(value);
3254 }
3255 
3256 static void spapr_instance_init(Object *obj)
3257 {
3258     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3259     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3260     MachineState *ms = MACHINE(spapr);
3261     MachineClass *mc = MACHINE_GET_CLASS(ms);
3262 
3263     /*
3264      * NVDIMM support went live in 5.1 without considering that, in
3265      * other archs, the user needs to enable NVDIMM support with the
3266      * 'nvdimm' machine option and the default behavior is NVDIMM
3267      * support disabled. It is too late to roll back to the standard
3268      * behavior without breaking 5.1 guests.
3269      */
3270     if (mc->nvdimm_supported) {
3271         ms->nvdimms_state->is_enabled = true;
3272     }
3273 
3274     spapr->htab_fd = -1;
3275     spapr->use_hotplug_event_source = true;
3276     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3277     object_property_add_str(obj, "kvm-type",
3278                             spapr_get_kvm_type, spapr_set_kvm_type);
3279     object_property_set_description(obj, "kvm-type",
3280                                     "Specifies the KVM virtualization mode (auto,"
3281                                     " hv, pr). Defaults to 'auto'. This mode will use"
3282                                     " any available KVM module loaded in the host,"
3283                                     " where kvm_hv takes precedence if both kvm_hv and"
3284                                     " kvm_pr are loaded.");
3285     object_property_add_bool(obj, "modern-hotplug-events",
3286                             spapr_get_modern_hotplug_events,
3287                             spapr_set_modern_hotplug_events);
3288     object_property_set_description(obj, "modern-hotplug-events",
3289                                     "Use dedicated hotplug event mechanism in"
3290                                     " place of standard EPOW events when possible"
3291                                     " (required for memory hot-unplug support)");
3292     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3293                             "Maximum permitted CPU compatibility mode");
3294 
3295     object_property_add_str(obj, "resize-hpt",
3296                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3297     object_property_set_description(obj, "resize-hpt",
3298                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3299     object_property_add_uint32_ptr(obj, "vsmt",
3300                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3301     object_property_set_description(obj, "vsmt",
3302                                     "Virtual SMT: KVM behaves as if this were"
3303                                     " the host's SMT mode");
3304 
3305     object_property_add_bool(obj, "vfio-no-msix-emulation",
3306                              spapr_get_msix_emulation, NULL);
3307 
3308     object_property_add_uint64_ptr(obj, "kernel-addr",
3309                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3310     object_property_set_description(obj, "kernel-addr",
3311                                     stringify(KERNEL_LOAD_ADDR)
3312                                     " for -kernel is the default");
3313     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3314     /* The machine class defines the default interrupt controller mode */
3315     spapr->irq = smc->irq;
3316     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3317                             spapr_set_ic_mode);
3318     object_property_set_description(obj, "ic-mode",
3319                  "Specifies the interrupt controller mode (xics, xive, dual)");
3320 
3321     object_property_add_str(obj, "host-model",
3322         spapr_get_host_model, spapr_set_host_model);
3323     object_property_set_description(obj, "host-model",
3324         "Host model to advertise in guest device tree");
3325     object_property_add_str(obj, "host-serial",
3326         spapr_get_host_serial, spapr_set_host_serial);
3327     object_property_set_description(obj, "host-serial",
3328         "Host serial number to advertise in guest device tree");
3329 }
3330 
3331 static void spapr_machine_finalizefn(Object *obj)
3332 {
3333     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3334 
3335     g_free(spapr->kvm_type);
3336 }
3337 
3338 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3339 {
3340     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3341     PowerPCCPU *cpu = POWERPC_CPU(cs);
3342     CPUPPCState *env = &cpu->env;
3343 
3344     cpu_synchronize_state(cs);
3345     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3346     if (spapr->fwnmi_system_reset_addr != -1) {
3347         uint64_t rtas_addr, addr;
3348 
3349         /* get rtas addr from fdt */
3350         rtas_addr = spapr_get_rtas_addr();
3351         if (!rtas_addr) {
3352             qemu_system_guest_panicked(NULL);
3353             return;
3354         }
3355 
3356         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3357         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3358         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3359         env->gpr[3] = addr;
3360     }
3361     ppc_cpu_do_system_reset(cs);
3362     if (spapr->fwnmi_system_reset_addr != -1) {
3363         env->nip = spapr->fwnmi_system_reset_addr;
3364     }
3365 }
3366 
3367 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3368 {
3369     CPUState *cs;
3370 
3371     CPU_FOREACH(cs) {
3372         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3373     }
3374 }
3375 
3376 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3377                           void *fdt, int *fdt_start_offset, Error **errp)
3378 {
3379     uint64_t addr;
3380     uint32_t node;
3381 
3382     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3383     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3384                                     &error_abort);
3385     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3386                                              SPAPR_MEMORY_BLOCK_SIZE);
3387     return 0;
3388 }
3389 
3390 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3391                            bool dedicated_hp_event_source)
3392 {
3393     SpaprDrc *drc;
3394     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3395     int i;
3396     uint64_t addr = addr_start;
3397     bool hotplugged = spapr_drc_hotplugged(dev);
3398 
3399     for (i = 0; i < nr_lmbs; i++) {
3400         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3401                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3402         g_assert(drc);
3403 
3404         /*
3405          * memory_device_get_free_addr() provided a range of free addresses
3406          * that doesn't overlap with any existing mapping at pre-plug. The
3407          * corresponding LMB DRCs are thus assumed to be all attachable.
3408          */
3409         spapr_drc_attach(drc, dev);
3410         if (!hotplugged) {
3411             spapr_drc_reset(drc);
3412         }
3413         addr += SPAPR_MEMORY_BLOCK_SIZE;
3414     }
3415     /* send hotplug notification to the
3416      * guest only in case of hotplugged memory
3417      */
3418     if (hotplugged) {
3419         if (dedicated_hp_event_source) {
3420             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3421                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3422             g_assert(drc);
3423             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3424                                                    nr_lmbs,
3425                                                    spapr_drc_index(drc));
3426         } else {
3427             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3428                                            nr_lmbs);
3429         }
3430     }
3431 }
3432 
3433 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3434 {
3435     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3436     PCDIMMDevice *dimm = PC_DIMM(dev);
3437     uint64_t size, addr;
3438     int64_t slot;
3439     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3440 
3441     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3442 
3443     pc_dimm_plug(dimm, MACHINE(ms));
3444 
3445     if (!is_nvdimm) {
3446         addr = object_property_get_uint(OBJECT(dimm),
3447                                         PC_DIMM_ADDR_PROP, &error_abort);
3448         spapr_add_lmbs(dev, addr, size,
3449                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3450     } else {
3451         slot = object_property_get_int(OBJECT(dimm),
3452                                        PC_DIMM_SLOT_PROP, &error_abort);
3453         /* We should have valid slot number at this point */
3454         g_assert(slot >= 0);
3455         spapr_add_nvdimm(dev, slot);
3456     }
3457 }
3458 
3459 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3460                                   Error **errp)
3461 {
3462     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3463     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3464     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3465     PCDIMMDevice *dimm = PC_DIMM(dev);
3466     Error *local_err = NULL;
3467     uint64_t size;
3468     Object *memdev;
3469     hwaddr pagesize;
3470 
3471     if (!smc->dr_lmb_enabled) {
3472         error_setg(errp, "Memory hotplug not supported for this machine");
3473         return;
3474     }
3475 
3476     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3477     if (local_err) {
3478         error_propagate(errp, local_err);
3479         return;
3480     }
3481 
3482     if (is_nvdimm) {
3483         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3484             return;
3485         }
3486     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3487         error_setg(errp, "Hotplugged memory size must be a multiple of "
3488                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3489         return;
3490     }
3491 
3492     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3493                                       &error_abort);
3494     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3495     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3496         return;
3497     }
3498 
3499     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3500 }
3501 
3502 struct SpaprDimmState {
3503     PCDIMMDevice *dimm;
3504     uint32_t nr_lmbs;
3505     QTAILQ_ENTRY(SpaprDimmState) next;
3506 };
3507 
3508 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3509                                                        PCDIMMDevice *dimm)
3510 {
3511     SpaprDimmState *dimm_state = NULL;
3512 
3513     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3514         if (dimm_state->dimm == dimm) {
3515             break;
3516         }
3517     }
3518     return dimm_state;
3519 }
3520 
3521 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3522                                                       uint32_t nr_lmbs,
3523                                                       PCDIMMDevice *dimm)
3524 {
3525     SpaprDimmState *ds = NULL;
3526 
3527     /*
3528      * If this request is for a DIMM whose removal had failed earlier
3529      * (due to guest's refusal to remove the LMBs), we would have this
3530      * dimm already in the pending_dimm_unplugs list. In that
3531      * case don't add again.
3532      */
3533     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3534     if (!ds) {
3535         ds = g_malloc0(sizeof(SpaprDimmState));
3536         ds->nr_lmbs = nr_lmbs;
3537         ds->dimm = dimm;
3538         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3539     }
3540     return ds;
3541 }
3542 
3543 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3544                                               SpaprDimmState *dimm_state)
3545 {
3546     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3547     g_free(dimm_state);
3548 }
3549 
3550 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3551                                                         PCDIMMDevice *dimm)
3552 {
3553     SpaprDrc *drc;
3554     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3555                                                   &error_abort);
3556     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3557     uint32_t avail_lmbs = 0;
3558     uint64_t addr_start, addr;
3559     int i;
3560 
3561     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3562                                           &error_abort);
3563 
3564     addr = addr_start;
3565     for (i = 0; i < nr_lmbs; i++) {
3566         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3567                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3568         g_assert(drc);
3569         if (drc->dev) {
3570             avail_lmbs++;
3571         }
3572         addr += SPAPR_MEMORY_BLOCK_SIZE;
3573     }
3574 
3575     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3576 }
3577 
3578 /* Callback to be called during DRC release. */
3579 void spapr_lmb_release(DeviceState *dev)
3580 {
3581     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3582     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3583     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3584 
3585     /* This information will get lost if a migration occurs
3586      * during the unplug process. In this case recover it. */
3587     if (ds == NULL) {
3588         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3589         g_assert(ds);
3590         /* The DRC being examined by the caller at least must be counted */
3591         g_assert(ds->nr_lmbs);
3592     }
3593 
3594     if (--ds->nr_lmbs) {
3595         return;
3596     }
3597 
3598     /*
3599      * Now that all the LMBs have been removed by the guest, call the
3600      * unplug handler chain. This can never fail.
3601      */
3602     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3603     object_unparent(OBJECT(dev));
3604 }
3605 
3606 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3607 {
3608     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3609     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3610 
3611     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3612     qdev_unrealize(dev);
3613     spapr_pending_dimm_unplugs_remove(spapr, ds);
3614 }
3615 
3616 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3617                                         DeviceState *dev, Error **errp)
3618 {
3619     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3620     PCDIMMDevice *dimm = PC_DIMM(dev);
3621     uint32_t nr_lmbs;
3622     uint64_t size, addr_start, addr;
3623     int i;
3624     SpaprDrc *drc;
3625 
3626     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3627         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3628         return;
3629     }
3630 
3631     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3632     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3633 
3634     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3635                                           &error_abort);
3636 
3637     /*
3638      * An existing pending dimm state for this DIMM means that there is an
3639      * unplug operation in progress, waiting for the spapr_lmb_release
3640      * callback to complete the job (BQL can't cover that far). In this case,
3641      * bail out to avoid detaching DRCs that were already released.
3642      */
3643     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3644         error_setg(errp, "Memory unplug already in progress for device %s",
3645                    dev->id);
3646         return;
3647     }
3648 
3649     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3650 
3651     addr = addr_start;
3652     for (i = 0; i < nr_lmbs; i++) {
3653         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3654                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3655         g_assert(drc);
3656 
3657         spapr_drc_detach(drc);
3658         addr += SPAPR_MEMORY_BLOCK_SIZE;
3659     }
3660 
3661     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3662                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3663     g_assert(drc);
3664     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3665                                               nr_lmbs, spapr_drc_index(drc));
3666 }
3667 
3668 /* Callback to be called during DRC release. */
3669 void spapr_core_release(DeviceState *dev)
3670 {
3671     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3672 
3673     /* Call the unplug handler chain. This can never fail. */
3674     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3675     object_unparent(OBJECT(dev));
3676 }
3677 
3678 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3679 {
3680     MachineState *ms = MACHINE(hotplug_dev);
3681     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3682     CPUCore *cc = CPU_CORE(dev);
3683     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3684 
3685     if (smc->pre_2_10_has_unused_icps) {
3686         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3687         int i;
3688 
3689         for (i = 0; i < cc->nr_threads; i++) {
3690             CPUState *cs = CPU(sc->threads[i]);
3691 
3692             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3693         }
3694     }
3695 
3696     assert(core_slot);
3697     core_slot->cpu = NULL;
3698     qdev_unrealize(dev);
3699 }
3700 
3701 static
3702 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3703                                Error **errp)
3704 {
3705     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3706     int index;
3707     SpaprDrc *drc;
3708     CPUCore *cc = CPU_CORE(dev);
3709 
3710     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3711         error_setg(errp, "Unable to find CPU core with core-id: %d",
3712                    cc->core_id);
3713         return;
3714     }
3715     if (index == 0) {
3716         error_setg(errp, "Boot CPU core may not be unplugged");
3717         return;
3718     }
3719 
3720     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3721                           spapr_vcpu_id(spapr, cc->core_id));
3722     g_assert(drc);
3723 
3724     if (!spapr_drc_unplug_requested(drc)) {
3725         spapr_drc_detach(drc);
3726         spapr_hotplug_req_remove_by_index(drc);
3727     }
3728 }
3729 
3730 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3731                            void *fdt, int *fdt_start_offset, Error **errp)
3732 {
3733     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3734     CPUState *cs = CPU(core->threads[0]);
3735     PowerPCCPU *cpu = POWERPC_CPU(cs);
3736     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3737     int id = spapr_get_vcpu_id(cpu);
3738     g_autofree char *nodename = NULL;
3739     int offset;
3740 
3741     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3742     offset = fdt_add_subnode(fdt, 0, nodename);
3743 
3744     spapr_dt_cpu(cs, fdt, offset, spapr);
3745 
3746     /*
3747      * spapr_dt_cpu() does not fill the 'name' property in the
3748      * CPU node. The function is called during boot process, before
3749      * and after CAS, and overwriting the 'name' property written
3750      * by SLOF is not allowed.
3751      *
3752      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3753      * CPUs more compatible with the coldplugged ones, which have
3754      * the 'name' property. Linux Kernel also relies on this
3755      * property to identify CPU nodes.
3756      */
3757     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3758 
3759     *fdt_start_offset = offset;
3760     return 0;
3761 }
3762 
3763 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3764 {
3765     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3766     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3767     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3768     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3769     CPUCore *cc = CPU_CORE(dev);
3770     CPUState *cs;
3771     SpaprDrc *drc;
3772     CPUArchId *core_slot;
3773     int index;
3774     bool hotplugged = spapr_drc_hotplugged(dev);
3775     int i;
3776 
3777     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3778     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3779 
3780     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3781                           spapr_vcpu_id(spapr, cc->core_id));
3782 
3783     g_assert(drc || !mc->has_hotpluggable_cpus);
3784 
3785     if (drc) {
3786         /*
3787          * spapr_core_pre_plug() already buys us this is a brand new
3788          * core being plugged into a free slot. Nothing should already
3789          * be attached to the corresponding DRC.
3790          */
3791         spapr_drc_attach(drc, dev);
3792 
3793         if (hotplugged) {
3794             /*
3795              * Send hotplug notification interrupt to the guest only
3796              * in case of hotplugged CPUs.
3797              */
3798             spapr_hotplug_req_add_by_index(drc);
3799         } else {
3800             spapr_drc_reset(drc);
3801         }
3802     }
3803 
3804     core_slot->cpu = OBJECT(dev);
3805 
3806     /*
3807      * Set compatibility mode to match the boot CPU, which was either set
3808      * by the machine reset code or by CAS. This really shouldn't fail at
3809      * this point.
3810      */
3811     if (hotplugged) {
3812         for (i = 0; i < cc->nr_threads; i++) {
3813             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3814                            &error_abort);
3815         }
3816     }
3817 
3818     if (smc->pre_2_10_has_unused_icps) {
3819         for (i = 0; i < cc->nr_threads; i++) {
3820             cs = CPU(core->threads[i]);
3821             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3822         }
3823     }
3824 }
3825 
3826 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3827                                 Error **errp)
3828 {
3829     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3830     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3831     CPUCore *cc = CPU_CORE(dev);
3832     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3833     const char *type = object_get_typename(OBJECT(dev));
3834     CPUArchId *core_slot;
3835     int index;
3836     unsigned int smp_threads = machine->smp.threads;
3837 
3838     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3839         error_setg(errp, "CPU hotplug not supported for this machine");
3840         return;
3841     }
3842 
3843     if (strcmp(base_core_type, type)) {
3844         error_setg(errp, "CPU core type should be %s", base_core_type);
3845         return;
3846     }
3847 
3848     if (cc->core_id % smp_threads) {
3849         error_setg(errp, "invalid core id %d", cc->core_id);
3850         return;
3851     }
3852 
3853     /*
3854      * In general we should have homogeneous threads-per-core, but old
3855      * (pre hotplug support) machine types allow the last core to have
3856      * reduced threads as a compatibility hack for when we allowed
3857      * total vcpus not a multiple of threads-per-core.
3858      */
3859     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3860         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3861                    smp_threads);
3862         return;
3863     }
3864 
3865     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3866     if (!core_slot) {
3867         error_setg(errp, "core id %d out of range", cc->core_id);
3868         return;
3869     }
3870 
3871     if (core_slot->cpu) {
3872         error_setg(errp, "core %d already populated", cc->core_id);
3873         return;
3874     }
3875 
3876     numa_cpu_pre_plug(core_slot, dev, errp);
3877 }
3878 
3879 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3880                           void *fdt, int *fdt_start_offset, Error **errp)
3881 {
3882     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3883     int intc_phandle;
3884 
3885     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3886     if (intc_phandle <= 0) {
3887         return -1;
3888     }
3889 
3890     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3891         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3892         return -1;
3893     }
3894 
3895     /* generally SLOF creates these, for hotplug it's up to QEMU */
3896     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3897 
3898     return 0;
3899 }
3900 
3901 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3902                                Error **errp)
3903 {
3904     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3905     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3906     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3907     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3908     SpaprDrc *drc;
3909 
3910     if (dev->hotplugged && !smc->dr_phb_enabled) {
3911         error_setg(errp, "PHB hotplug not supported for this machine");
3912         return false;
3913     }
3914 
3915     if (sphb->index == (uint32_t)-1) {
3916         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3917         return false;
3918     }
3919 
3920     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3921     if (drc && drc->dev) {
3922         error_setg(errp, "PHB %d already attached", sphb->index);
3923         return false;
3924     }
3925 
3926     /*
3927      * This will check that sphb->index doesn't exceed the maximum number of
3928      * PHBs for the current machine type.
3929      */
3930     return
3931         smc->phb_placement(spapr, sphb->index,
3932                            &sphb->buid, &sphb->io_win_addr,
3933                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
3934                            windows_supported, sphb->dma_liobn,
3935                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3936                            errp);
3937 }
3938 
3939 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3940 {
3941     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3942     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3943     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3944     SpaprDrc *drc;
3945     bool hotplugged = spapr_drc_hotplugged(dev);
3946 
3947     if (!smc->dr_phb_enabled) {
3948         return;
3949     }
3950 
3951     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3952     /* hotplug hooks should check it's enabled before getting this far */
3953     assert(drc);
3954 
3955     /* spapr_phb_pre_plug() already checked the DRC is attachable */
3956     spapr_drc_attach(drc, dev);
3957 
3958     if (hotplugged) {
3959         spapr_hotplug_req_add_by_index(drc);
3960     } else {
3961         spapr_drc_reset(drc);
3962     }
3963 }
3964 
3965 void spapr_phb_release(DeviceState *dev)
3966 {
3967     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3968 
3969     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3970     object_unparent(OBJECT(dev));
3971 }
3972 
3973 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3974 {
3975     qdev_unrealize(dev);
3976 }
3977 
3978 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3979                                      DeviceState *dev, Error **errp)
3980 {
3981     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3982     SpaprDrc *drc;
3983 
3984     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3985     assert(drc);
3986 
3987     if (!spapr_drc_unplug_requested(drc)) {
3988         spapr_drc_detach(drc);
3989         spapr_hotplug_req_remove_by_index(drc);
3990     }
3991 }
3992 
3993 static
3994 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3995                               Error **errp)
3996 {
3997     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3998 
3999     if (spapr->tpm_proxy != NULL) {
4000         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4001         return false;
4002     }
4003 
4004     return true;
4005 }
4006 
4007 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4008 {
4009     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4010     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4011 
4012     /* Already checked in spapr_tpm_proxy_pre_plug() */
4013     g_assert(spapr->tpm_proxy == NULL);
4014 
4015     spapr->tpm_proxy = tpm_proxy;
4016 }
4017 
4018 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4019 {
4020     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4021 
4022     qdev_unrealize(dev);
4023     object_unparent(OBJECT(dev));
4024     spapr->tpm_proxy = NULL;
4025 }
4026 
4027 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4028                                       DeviceState *dev, Error **errp)
4029 {
4030     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4031         spapr_memory_plug(hotplug_dev, dev);
4032     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4033         spapr_core_plug(hotplug_dev, dev);
4034     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4035         spapr_phb_plug(hotplug_dev, dev);
4036     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4037         spapr_tpm_proxy_plug(hotplug_dev, dev);
4038     }
4039 }
4040 
4041 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4042                                         DeviceState *dev, Error **errp)
4043 {
4044     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4045         spapr_memory_unplug(hotplug_dev, dev);
4046     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4047         spapr_core_unplug(hotplug_dev, dev);
4048     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4049         spapr_phb_unplug(hotplug_dev, dev);
4050     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4051         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4052     }
4053 }
4054 
4055 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4056 {
4057     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4058         /*
4059          * CAS will process all pending unplug requests.
4060          *
4061          * HACK: a guest could theoretically have cleared all bits in OV5,
4062          * but none of the guests we care for do.
4063          */
4064         spapr_ovec_empty(spapr->ov5_cas);
4065 }
4066 
4067 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4068                                                 DeviceState *dev, Error **errp)
4069 {
4070     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4071     MachineClass *mc = MACHINE_GET_CLASS(sms);
4072     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4073 
4074     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4075         if (spapr_memory_hot_unplug_supported(sms)) {
4076             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4077         } else {
4078             error_setg(errp, "Memory hot unplug not supported for this guest");
4079         }
4080     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4081         if (!mc->has_hotpluggable_cpus) {
4082             error_setg(errp, "CPU hot unplug not supported on this machine");
4083             return;
4084         }
4085         spapr_core_unplug_request(hotplug_dev, dev, errp);
4086     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4087         if (!smc->dr_phb_enabled) {
4088             error_setg(errp, "PHB hot unplug not supported on this machine");
4089             return;
4090         }
4091         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4092     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4093         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4094     }
4095 }
4096 
4097 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4098                                           DeviceState *dev, Error **errp)
4099 {
4100     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4101         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4102     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4103         spapr_core_pre_plug(hotplug_dev, dev, errp);
4104     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4105         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4106     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4107         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4108     }
4109 }
4110 
4111 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4112                                                  DeviceState *dev)
4113 {
4114     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4115         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4116         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4117         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4118         return HOTPLUG_HANDLER(machine);
4119     }
4120     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4121         PCIDevice *pcidev = PCI_DEVICE(dev);
4122         PCIBus *root = pci_device_root_bus(pcidev);
4123         SpaprPhbState *phb =
4124             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4125                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4126 
4127         if (phb) {
4128             return HOTPLUG_HANDLER(phb);
4129         }
4130     }
4131     return NULL;
4132 }
4133 
4134 static CpuInstanceProperties
4135 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4136 {
4137     CPUArchId *core_slot;
4138     MachineClass *mc = MACHINE_GET_CLASS(machine);
4139 
4140     /* make sure possible_cpu are intialized */
4141     mc->possible_cpu_arch_ids(machine);
4142     /* get CPU core slot containing thread that matches cpu_index */
4143     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4144     assert(core_slot);
4145     return core_slot->props;
4146 }
4147 
4148 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4149 {
4150     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4151 }
4152 
4153 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4154 {
4155     int i;
4156     unsigned int smp_threads = machine->smp.threads;
4157     unsigned int smp_cpus = machine->smp.cpus;
4158     const char *core_type;
4159     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4160     MachineClass *mc = MACHINE_GET_CLASS(machine);
4161 
4162     if (!mc->has_hotpluggable_cpus) {
4163         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4164     }
4165     if (machine->possible_cpus) {
4166         assert(machine->possible_cpus->len == spapr_max_cores);
4167         return machine->possible_cpus;
4168     }
4169 
4170     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4171     if (!core_type) {
4172         error_report("Unable to find sPAPR CPU Core definition");
4173         exit(1);
4174     }
4175 
4176     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4177                              sizeof(CPUArchId) * spapr_max_cores);
4178     machine->possible_cpus->len = spapr_max_cores;
4179     for (i = 0; i < machine->possible_cpus->len; i++) {
4180         int core_id = i * smp_threads;
4181 
4182         machine->possible_cpus->cpus[i].type = core_type;
4183         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4184         machine->possible_cpus->cpus[i].arch_id = core_id;
4185         machine->possible_cpus->cpus[i].props.has_core_id = true;
4186         machine->possible_cpus->cpus[i].props.core_id = core_id;
4187     }
4188     return machine->possible_cpus;
4189 }
4190 
4191 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4192                                 uint64_t *buid, hwaddr *pio,
4193                                 hwaddr *mmio32, hwaddr *mmio64,
4194                                 unsigned n_dma, uint32_t *liobns,
4195                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4196 {
4197     /*
4198      * New-style PHB window placement.
4199      *
4200      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4201      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4202      * windows.
4203      *
4204      * Some guest kernels can't work with MMIO windows above 1<<46
4205      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4206      *
4207      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4208      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4209      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4210      * 1TiB 64-bit MMIO windows for each PHB.
4211      */
4212     const uint64_t base_buid = 0x800000020000000ULL;
4213     int i;
4214 
4215     /* Sanity check natural alignments */
4216     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4217     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4218     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4219     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4220     /* Sanity check bounds */
4221     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4222                       SPAPR_PCI_MEM32_WIN_SIZE);
4223     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4224                       SPAPR_PCI_MEM64_WIN_SIZE);
4225 
4226     if (index >= SPAPR_MAX_PHBS) {
4227         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4228                    SPAPR_MAX_PHBS - 1);
4229         return false;
4230     }
4231 
4232     *buid = base_buid + index;
4233     for (i = 0; i < n_dma; ++i) {
4234         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4235     }
4236 
4237     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4238     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4239     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4240 
4241     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4242     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4243     return true;
4244 }
4245 
4246 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4247 {
4248     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4249 
4250     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4251 }
4252 
4253 static void spapr_ics_resend(XICSFabric *dev)
4254 {
4255     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4256 
4257     ics_resend(spapr->ics);
4258 }
4259 
4260 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4261 {
4262     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4263 
4264     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4265 }
4266 
4267 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4268                                  Monitor *mon)
4269 {
4270     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4271 
4272     spapr_irq_print_info(spapr, mon);
4273     monitor_printf(mon, "irqchip: %s\n",
4274                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4275 }
4276 
4277 /*
4278  * This is a XIVE only operation
4279  */
4280 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4281                            uint8_t nvt_blk, uint32_t nvt_idx,
4282                            bool cam_ignore, uint8_t priority,
4283                            uint32_t logic_serv, XiveTCTXMatch *match)
4284 {
4285     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4286     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4287     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4288     int count;
4289 
4290     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4291                            priority, logic_serv, match);
4292     if (count < 0) {
4293         return count;
4294     }
4295 
4296     /*
4297      * When we implement the save and restore of the thread interrupt
4298      * contexts in the enter/exit CPU handlers of the machine and the
4299      * escalations in QEMU, we should be able to handle non dispatched
4300      * vCPUs.
4301      *
4302      * Until this is done, the sPAPR machine should find at least one
4303      * matching context always.
4304      */
4305     if (count == 0) {
4306         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4307                       nvt_blk, nvt_idx);
4308     }
4309 
4310     return count;
4311 }
4312 
4313 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4314 {
4315     return cpu->vcpu_id;
4316 }
4317 
4318 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4319 {
4320     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4321     MachineState *ms = MACHINE(spapr);
4322     int vcpu_id;
4323 
4324     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4325 
4326     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4327         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4328         error_append_hint(errp, "Adjust the number of cpus to %d "
4329                           "or try to raise the number of threads per core\n",
4330                           vcpu_id * ms->smp.threads / spapr->vsmt);
4331         return false;
4332     }
4333 
4334     cpu->vcpu_id = vcpu_id;
4335     return true;
4336 }
4337 
4338 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4339 {
4340     CPUState *cs;
4341 
4342     CPU_FOREACH(cs) {
4343         PowerPCCPU *cpu = POWERPC_CPU(cs);
4344 
4345         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4346             return cpu;
4347         }
4348     }
4349 
4350     return NULL;
4351 }
4352 
4353 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4354 {
4355     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4356 
4357     /* These are only called by TCG, KVM maintains dispatch state */
4358 
4359     spapr_cpu->prod = false;
4360     if (spapr_cpu->vpa_addr) {
4361         CPUState *cs = CPU(cpu);
4362         uint32_t dispatch;
4363 
4364         dispatch = ldl_be_phys(cs->as,
4365                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4366         dispatch++;
4367         if ((dispatch & 1) != 0) {
4368             qemu_log_mask(LOG_GUEST_ERROR,
4369                           "VPA: incorrect dispatch counter value for "
4370                           "dispatched partition %u, correcting.\n", dispatch);
4371             dispatch++;
4372         }
4373         stl_be_phys(cs->as,
4374                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4375     }
4376 }
4377 
4378 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4379 {
4380     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4381 
4382     if (spapr_cpu->vpa_addr) {
4383         CPUState *cs = CPU(cpu);
4384         uint32_t dispatch;
4385 
4386         dispatch = ldl_be_phys(cs->as,
4387                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4388         dispatch++;
4389         if ((dispatch & 1) != 1) {
4390             qemu_log_mask(LOG_GUEST_ERROR,
4391                           "VPA: incorrect dispatch counter value for "
4392                           "preempted partition %u, correcting.\n", dispatch);
4393             dispatch++;
4394         }
4395         stl_be_phys(cs->as,
4396                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4397     }
4398 }
4399 
4400 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4401 {
4402     MachineClass *mc = MACHINE_CLASS(oc);
4403     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4404     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4405     NMIClass *nc = NMI_CLASS(oc);
4406     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4407     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4408     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4409     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4410     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4411 
4412     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4413     mc->ignore_boot_device_suffixes = true;
4414 
4415     /*
4416      * We set up the default / latest behaviour here.  The class_init
4417      * functions for the specific versioned machine types can override
4418      * these details for backwards compatibility
4419      */
4420     mc->init = spapr_machine_init;
4421     mc->reset = spapr_machine_reset;
4422     mc->block_default_type = IF_SCSI;
4423     mc->max_cpus = 1024;
4424     mc->no_parallel = 1;
4425     mc->default_boot_order = "";
4426     mc->default_ram_size = 512 * MiB;
4427     mc->default_ram_id = "ppc_spapr.ram";
4428     mc->default_display = "std";
4429     mc->kvm_type = spapr_kvm_type;
4430     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4431     mc->pci_allow_0_address = true;
4432     assert(!mc->get_hotplug_handler);
4433     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4434     hc->pre_plug = spapr_machine_device_pre_plug;
4435     hc->plug = spapr_machine_device_plug;
4436     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4437     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4438     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4439     hc->unplug_request = spapr_machine_device_unplug_request;
4440     hc->unplug = spapr_machine_device_unplug;
4441 
4442     smc->dr_lmb_enabled = true;
4443     smc->update_dt_enabled = true;
4444     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4445     mc->has_hotpluggable_cpus = true;
4446     mc->nvdimm_supported = true;
4447     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4448     fwc->get_dev_path = spapr_get_fw_dev_path;
4449     nc->nmi_monitor_handler = spapr_nmi;
4450     smc->phb_placement = spapr_phb_placement;
4451     vhc->hypercall = emulate_spapr_hypercall;
4452     vhc->hpt_mask = spapr_hpt_mask;
4453     vhc->map_hptes = spapr_map_hptes;
4454     vhc->unmap_hptes = spapr_unmap_hptes;
4455     vhc->hpte_set_c = spapr_hpte_set_c;
4456     vhc->hpte_set_r = spapr_hpte_set_r;
4457     vhc->get_pate = spapr_get_pate;
4458     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4459     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4460     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4461     xic->ics_get = spapr_ics_get;
4462     xic->ics_resend = spapr_ics_resend;
4463     xic->icp_get = spapr_icp_get;
4464     ispc->print_info = spapr_pic_print_info;
4465     /* Force NUMA node memory size to be a multiple of
4466      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4467      * in which LMBs are represented and hot-added
4468      */
4469     mc->numa_mem_align_shift = 28;
4470     mc->auto_enable_numa = true;
4471 
4472     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4473     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4474     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4475     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4476     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4477     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4478     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4479     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4480     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4481     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4482     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4483     spapr_caps_add_properties(smc);
4484     smc->irq = &spapr_irq_dual;
4485     smc->dr_phb_enabled = true;
4486     smc->linux_pci_probe = true;
4487     smc->smp_threads_vsmt = true;
4488     smc->nr_xirqs = SPAPR_NR_XIRQS;
4489     xfc->match_nvt = spapr_match_nvt;
4490 }
4491 
4492 static const TypeInfo spapr_machine_info = {
4493     .name          = TYPE_SPAPR_MACHINE,
4494     .parent        = TYPE_MACHINE,
4495     .abstract      = true,
4496     .instance_size = sizeof(SpaprMachineState),
4497     .instance_init = spapr_instance_init,
4498     .instance_finalize = spapr_machine_finalizefn,
4499     .class_size    = sizeof(SpaprMachineClass),
4500     .class_init    = spapr_machine_class_init,
4501     .interfaces = (InterfaceInfo[]) {
4502         { TYPE_FW_PATH_PROVIDER },
4503         { TYPE_NMI },
4504         { TYPE_HOTPLUG_HANDLER },
4505         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4506         { TYPE_XICS_FABRIC },
4507         { TYPE_INTERRUPT_STATS_PROVIDER },
4508         { TYPE_XIVE_FABRIC },
4509         { }
4510     },
4511 };
4512 
4513 static void spapr_machine_latest_class_options(MachineClass *mc)
4514 {
4515     mc->alias = "pseries";
4516     mc->is_default = true;
4517 }
4518 
4519 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4520     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4521                                                     void *data)      \
4522     {                                                                \
4523         MachineClass *mc = MACHINE_CLASS(oc);                        \
4524         spapr_machine_##suffix##_class_options(mc);                  \
4525         if (latest) {                                                \
4526             spapr_machine_latest_class_options(mc);                  \
4527         }                                                            \
4528     }                                                                \
4529     static const TypeInfo spapr_machine_##suffix##_info = {          \
4530         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4531         .parent = TYPE_SPAPR_MACHINE,                                \
4532         .class_init = spapr_machine_##suffix##_class_init,           \
4533     };                                                               \
4534     static void spapr_machine_register_##suffix(void)                \
4535     {                                                                \
4536         type_register(&spapr_machine_##suffix##_info);               \
4537     }                                                                \
4538     type_init(spapr_machine_register_##suffix)
4539 
4540 /*
4541  * pseries-6.0
4542  */
4543 static void spapr_machine_6_0_class_options(MachineClass *mc)
4544 {
4545     /* Defaults for the latest behaviour inherited from the base class */
4546 }
4547 
4548 DEFINE_SPAPR_MACHINE(6_0, "6.0", true);
4549 
4550 /*
4551  * pseries-5.2
4552  */
4553 static void spapr_machine_5_2_class_options(MachineClass *mc)
4554 {
4555     spapr_machine_6_0_class_options(mc);
4556     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4557 }
4558 
4559 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4560 
4561 /*
4562  * pseries-5.1
4563  */
4564 static void spapr_machine_5_1_class_options(MachineClass *mc)
4565 {
4566     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4567 
4568     spapr_machine_5_2_class_options(mc);
4569     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4570     smc->pre_5_2_numa_associativity = true;
4571 }
4572 
4573 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4574 
4575 /*
4576  * pseries-5.0
4577  */
4578 static void spapr_machine_5_0_class_options(MachineClass *mc)
4579 {
4580     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4581     static GlobalProperty compat[] = {
4582         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4583     };
4584 
4585     spapr_machine_5_1_class_options(mc);
4586     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4587     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4588     mc->numa_mem_supported = true;
4589     smc->pre_5_1_assoc_refpoints = true;
4590 }
4591 
4592 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4593 
4594 /*
4595  * pseries-4.2
4596  */
4597 static void spapr_machine_4_2_class_options(MachineClass *mc)
4598 {
4599     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4600 
4601     spapr_machine_5_0_class_options(mc);
4602     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4603     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4604     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4605     smc->rma_limit = 16 * GiB;
4606     mc->nvdimm_supported = false;
4607 }
4608 
4609 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4610 
4611 /*
4612  * pseries-4.1
4613  */
4614 static void spapr_machine_4_1_class_options(MachineClass *mc)
4615 {
4616     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4617     static GlobalProperty compat[] = {
4618         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4619         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4620     };
4621 
4622     spapr_machine_4_2_class_options(mc);
4623     smc->linux_pci_probe = false;
4624     smc->smp_threads_vsmt = false;
4625     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4626     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4627 }
4628 
4629 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4630 
4631 /*
4632  * pseries-4.0
4633  */
4634 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4635                               uint64_t *buid, hwaddr *pio,
4636                               hwaddr *mmio32, hwaddr *mmio64,
4637                               unsigned n_dma, uint32_t *liobns,
4638                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4639 {
4640     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4641                              liobns, nv2gpa, nv2atsd, errp)) {
4642         return false;
4643     }
4644 
4645     *nv2gpa = 0;
4646     *nv2atsd = 0;
4647     return true;
4648 }
4649 static void spapr_machine_4_0_class_options(MachineClass *mc)
4650 {
4651     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4652 
4653     spapr_machine_4_1_class_options(mc);
4654     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4655     smc->phb_placement = phb_placement_4_0;
4656     smc->irq = &spapr_irq_xics;
4657     smc->pre_4_1_migration = true;
4658 }
4659 
4660 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4661 
4662 /*
4663  * pseries-3.1
4664  */
4665 static void spapr_machine_3_1_class_options(MachineClass *mc)
4666 {
4667     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4668 
4669     spapr_machine_4_0_class_options(mc);
4670     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4671 
4672     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4673     smc->update_dt_enabled = false;
4674     smc->dr_phb_enabled = false;
4675     smc->broken_host_serial_model = true;
4676     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4677     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4678     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4679     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4680 }
4681 
4682 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4683 
4684 /*
4685  * pseries-3.0
4686  */
4687 
4688 static void spapr_machine_3_0_class_options(MachineClass *mc)
4689 {
4690     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4691 
4692     spapr_machine_3_1_class_options(mc);
4693     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4694 
4695     smc->legacy_irq_allocation = true;
4696     smc->nr_xirqs = 0x400;
4697     smc->irq = &spapr_irq_xics_legacy;
4698 }
4699 
4700 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4701 
4702 /*
4703  * pseries-2.12
4704  */
4705 static void spapr_machine_2_12_class_options(MachineClass *mc)
4706 {
4707     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4708     static GlobalProperty compat[] = {
4709         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4710         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4711     };
4712 
4713     spapr_machine_3_0_class_options(mc);
4714     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4715     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4716 
4717     /* We depend on kvm_enabled() to choose a default value for the
4718      * hpt-max-page-size capability. Of course we can't do it here
4719      * because this is too early and the HW accelerator isn't initialzed
4720      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4721      */
4722     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4723 }
4724 
4725 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4726 
4727 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4728 {
4729     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4730 
4731     spapr_machine_2_12_class_options(mc);
4732     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4733     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4734     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4735 }
4736 
4737 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4738 
4739 /*
4740  * pseries-2.11
4741  */
4742 
4743 static void spapr_machine_2_11_class_options(MachineClass *mc)
4744 {
4745     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4746 
4747     spapr_machine_2_12_class_options(mc);
4748     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4749     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4750 }
4751 
4752 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4753 
4754 /*
4755  * pseries-2.10
4756  */
4757 
4758 static void spapr_machine_2_10_class_options(MachineClass *mc)
4759 {
4760     spapr_machine_2_11_class_options(mc);
4761     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4762 }
4763 
4764 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4765 
4766 /*
4767  * pseries-2.9
4768  */
4769 
4770 static void spapr_machine_2_9_class_options(MachineClass *mc)
4771 {
4772     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4773     static GlobalProperty compat[] = {
4774         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4775     };
4776 
4777     spapr_machine_2_10_class_options(mc);
4778     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4779     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4780     smc->pre_2_10_has_unused_icps = true;
4781     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4782 }
4783 
4784 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4785 
4786 /*
4787  * pseries-2.8
4788  */
4789 
4790 static void spapr_machine_2_8_class_options(MachineClass *mc)
4791 {
4792     static GlobalProperty compat[] = {
4793         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4794     };
4795 
4796     spapr_machine_2_9_class_options(mc);
4797     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4798     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4799     mc->numa_mem_align_shift = 23;
4800 }
4801 
4802 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4803 
4804 /*
4805  * pseries-2.7
4806  */
4807 
4808 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4809                               uint64_t *buid, hwaddr *pio,
4810                               hwaddr *mmio32, hwaddr *mmio64,
4811                               unsigned n_dma, uint32_t *liobns,
4812                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4813 {
4814     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4815     const uint64_t base_buid = 0x800000020000000ULL;
4816     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4817     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4818     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4819     const uint32_t max_index = 255;
4820     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4821 
4822     uint64_t ram_top = MACHINE(spapr)->ram_size;
4823     hwaddr phb0_base, phb_base;
4824     int i;
4825 
4826     /* Do we have device memory? */
4827     if (MACHINE(spapr)->maxram_size > ram_top) {
4828         /* Can't just use maxram_size, because there may be an
4829          * alignment gap between normal and device memory regions
4830          */
4831         ram_top = MACHINE(spapr)->device_memory->base +
4832             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4833     }
4834 
4835     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4836 
4837     if (index > max_index) {
4838         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4839                    max_index);
4840         return false;
4841     }
4842 
4843     *buid = base_buid + index;
4844     for (i = 0; i < n_dma; ++i) {
4845         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4846     }
4847 
4848     phb_base = phb0_base + index * phb_spacing;
4849     *pio = phb_base + pio_offset;
4850     *mmio32 = phb_base + mmio_offset;
4851     /*
4852      * We don't set the 64-bit MMIO window, relying on the PHB's
4853      * fallback behaviour of automatically splitting a large "32-bit"
4854      * window into contiguous 32-bit and 64-bit windows
4855      */
4856 
4857     *nv2gpa = 0;
4858     *nv2atsd = 0;
4859     return true;
4860 }
4861 
4862 static void spapr_machine_2_7_class_options(MachineClass *mc)
4863 {
4864     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4865     static GlobalProperty compat[] = {
4866         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4867         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4868         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4869         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4870     };
4871 
4872     spapr_machine_2_8_class_options(mc);
4873     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4874     mc->default_machine_opts = "modern-hotplug-events=off";
4875     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4876     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4877     smc->phb_placement = phb_placement_2_7;
4878 }
4879 
4880 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4881 
4882 /*
4883  * pseries-2.6
4884  */
4885 
4886 static void spapr_machine_2_6_class_options(MachineClass *mc)
4887 {
4888     static GlobalProperty compat[] = {
4889         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4890     };
4891 
4892     spapr_machine_2_7_class_options(mc);
4893     mc->has_hotpluggable_cpus = false;
4894     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4895     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4896 }
4897 
4898 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4899 
4900 /*
4901  * pseries-2.5
4902  */
4903 
4904 static void spapr_machine_2_5_class_options(MachineClass *mc)
4905 {
4906     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4907     static GlobalProperty compat[] = {
4908         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4909     };
4910 
4911     spapr_machine_2_6_class_options(mc);
4912     smc->use_ohci_by_default = true;
4913     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4914     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4915 }
4916 
4917 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4918 
4919 /*
4920  * pseries-2.4
4921  */
4922 
4923 static void spapr_machine_2_4_class_options(MachineClass *mc)
4924 {
4925     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4926 
4927     spapr_machine_2_5_class_options(mc);
4928     smc->dr_lmb_enabled = false;
4929     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4930 }
4931 
4932 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4933 
4934 /*
4935  * pseries-2.3
4936  */
4937 
4938 static void spapr_machine_2_3_class_options(MachineClass *mc)
4939 {
4940     static GlobalProperty compat[] = {
4941         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4942     };
4943     spapr_machine_2_4_class_options(mc);
4944     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4945     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4946 }
4947 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4948 
4949 /*
4950  * pseries-2.2
4951  */
4952 
4953 static void spapr_machine_2_2_class_options(MachineClass *mc)
4954 {
4955     static GlobalProperty compat[] = {
4956         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4957     };
4958 
4959     spapr_machine_2_3_class_options(mc);
4960     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4961     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4962     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4963 }
4964 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4965 
4966 /*
4967  * pseries-2.1
4968  */
4969 
4970 static void spapr_machine_2_1_class_options(MachineClass *mc)
4971 {
4972     spapr_machine_2_2_class_options(mc);
4973     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4974 }
4975 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4976 
4977 static void spapr_machine_register_types(void)
4978 {
4979     type_register_static(&spapr_machine_info);
4980 }
4981 
4982 type_init(spapr_machine_register_types)
4983