xref: /openbmc/qemu/hw/ppc/spapr.c (revision 284d697c)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 #include "hw/ppc/spapr_numa.h"
85 
86 #include "monitor/monitor.h"
87 
88 #include <libfdt.h>
89 
90 /* SLOF memory layout:
91  *
92  * SLOF raw image loaded at 0, copies its romfs right below the flat
93  * device-tree, then position SLOF itself 31M below that
94  *
95  * So we set FW_OVERHEAD to 40MB which should account for all of that
96  * and more
97  *
98  * We load our kernel at 4M, leaving space for SLOF initial image
99  */
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            (128 * MiB)
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static void spapr_dt_pa_features(SpaprMachineState *spapr,
206                                  PowerPCCPU *cpu,
207                                  void *fdt, int offset)
208 {
209     uint8_t pa_features_206[] = { 6, 0,
210         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
211     uint8_t pa_features_207[] = { 24, 0,
212         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
213         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
214         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
215         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
216     uint8_t pa_features_300[] = { 66, 0,
217         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
218         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
219         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
220         /* 6: DS207 */
221         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
222         /* 16: Vector */
223         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
224         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
225         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
226         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
227         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
228         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
229         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
230         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
231         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
232         /* 42: PM, 44: PC RA, 46: SC vec'd */
233         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
234         /* 48: SIMD, 50: QP BFP, 52: String */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
236         /* 54: DecFP, 56: DecI, 58: SHA */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
238         /* 60: NM atomic, 62: RNG */
239         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
240     };
241     uint8_t *pa_features = NULL;
242     size_t pa_size;
243 
244     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
245         pa_features = pa_features_206;
246         pa_size = sizeof(pa_features_206);
247     }
248     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
249         pa_features = pa_features_207;
250         pa_size = sizeof(pa_features_207);
251     }
252     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
253         pa_features = pa_features_300;
254         pa_size = sizeof(pa_features_300);
255     }
256     if (!pa_features) {
257         return;
258     }
259 
260     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
261         /*
262          * Note: we keep CI large pages off by default because a 64K capable
263          * guest provisioned with large pages might otherwise try to map a qemu
264          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
265          * even if that qemu runs on a 4k host.
266          * We dd this bit back here if we are confident this is not an issue
267          */
268         pa_features[3] |= 0x20;
269     }
270     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
271         pa_features[24] |= 0x80;    /* Transactional memory support */
272     }
273     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
274         /* Workaround for broken kernels that attempt (guest) radix
275          * mode when they can't handle it, if they see the radix bit set
276          * in pa-features. So hide it from them. */
277         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
278     }
279 
280     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
281 }
282 
283 static hwaddr spapr_node0_size(MachineState *machine)
284 {
285     if (machine->numa_state->num_nodes) {
286         int i;
287         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
288             if (machine->numa_state->nodes[i].node_mem) {
289                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
290                            machine->ram_size);
291             }
292         }
293     }
294     return machine->ram_size;
295 }
296 
297 bool spapr_machine_using_legacy_numa(SpaprMachineState *spapr)
298 {
299     MachineState *machine = MACHINE(spapr);
300     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
301 
302     return smc->pre_5_2_numa_associativity ||
303            machine->numa_state->num_nodes <= 1;
304 }
305 
306 static void add_str(GString *s, const gchar *s1)
307 {
308     g_string_append_len(s, s1, strlen(s1) + 1);
309 }
310 
311 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
312                                 hwaddr start, hwaddr size)
313 {
314     char mem_name[32];
315     uint64_t mem_reg_property[2];
316     int off;
317 
318     mem_reg_property[0] = cpu_to_be64(start);
319     mem_reg_property[1] = cpu_to_be64(size);
320 
321     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
322     off = fdt_add_subnode(fdt, 0, mem_name);
323     _FDT(off);
324     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
325     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
326                       sizeof(mem_reg_property))));
327     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
328     return off;
329 }
330 
331 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
332 {
333     MemoryDeviceInfoList *info;
334 
335     for (info = list; info; info = info->next) {
336         MemoryDeviceInfo *value = info->value;
337 
338         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
339             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
340 
341             if (addr >= pcdimm_info->addr &&
342                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
343                 return pcdimm_info->node;
344             }
345         }
346     }
347 
348     return -1;
349 }
350 
351 struct sPAPRDrconfCellV2 {
352      uint32_t seq_lmbs;
353      uint64_t base_addr;
354      uint32_t drc_index;
355      uint32_t aa_index;
356      uint32_t flags;
357 } QEMU_PACKED;
358 
359 typedef struct DrconfCellQueue {
360     struct sPAPRDrconfCellV2 cell;
361     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
362 } DrconfCellQueue;
363 
364 static DrconfCellQueue *
365 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
366                       uint32_t drc_index, uint32_t aa_index,
367                       uint32_t flags)
368 {
369     DrconfCellQueue *elem;
370 
371     elem = g_malloc0(sizeof(*elem));
372     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
373     elem->cell.base_addr = cpu_to_be64(base_addr);
374     elem->cell.drc_index = cpu_to_be32(drc_index);
375     elem->cell.aa_index = cpu_to_be32(aa_index);
376     elem->cell.flags = cpu_to_be32(flags);
377 
378     return elem;
379 }
380 
381 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
382                                       int offset, MemoryDeviceInfoList *dimms)
383 {
384     MachineState *machine = MACHINE(spapr);
385     uint8_t *int_buf, *cur_index;
386     int ret;
387     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
388     uint64_t addr, cur_addr, size;
389     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
390     uint64_t mem_end = machine->device_memory->base +
391                        memory_region_size(&machine->device_memory->mr);
392     uint32_t node, buf_len, nr_entries = 0;
393     SpaprDrc *drc;
394     DrconfCellQueue *elem, *next;
395     MemoryDeviceInfoList *info;
396     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
397         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
398 
399     /* Entry to cover RAM and the gap area */
400     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
401                                  SPAPR_LMB_FLAGS_RESERVED |
402                                  SPAPR_LMB_FLAGS_DRC_INVALID);
403     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
404     nr_entries++;
405 
406     cur_addr = machine->device_memory->base;
407     for (info = dimms; info; info = info->next) {
408         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
409 
410         addr = di->addr;
411         size = di->size;
412         node = di->node;
413 
414         /*
415          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
416          * area is marked hotpluggable in the next iteration for the bigger
417          * chunk including the NVDIMM occupied area.
418          */
419         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
420             continue;
421 
422         /* Entry for hot-pluggable area */
423         if (cur_addr < addr) {
424             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
425             g_assert(drc);
426             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
427                                          cur_addr, spapr_drc_index(drc), -1, 0);
428             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
429             nr_entries++;
430         }
431 
432         /* Entry for DIMM */
433         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
434         g_assert(drc);
435         elem = spapr_get_drconf_cell(size / lmb_size, addr,
436                                      spapr_drc_index(drc), node,
437                                      (SPAPR_LMB_FLAGS_ASSIGNED |
438                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
439         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
440         nr_entries++;
441         cur_addr = addr + size;
442     }
443 
444     /* Entry for remaining hotpluggable area */
445     if (cur_addr < mem_end) {
446         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
447         g_assert(drc);
448         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
449                                      cur_addr, spapr_drc_index(drc), -1, 0);
450         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
451         nr_entries++;
452     }
453 
454     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
455     int_buf = cur_index = g_malloc0(buf_len);
456     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
457     cur_index += sizeof(nr_entries);
458 
459     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
460         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
461         cur_index += sizeof(elem->cell);
462         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
463         g_free(elem);
464     }
465 
466     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
467     g_free(int_buf);
468     if (ret < 0) {
469         return -1;
470     }
471     return 0;
472 }
473 
474 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
475                                    int offset, MemoryDeviceInfoList *dimms)
476 {
477     MachineState *machine = MACHINE(spapr);
478     int i, ret;
479     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
480     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
481     uint32_t nr_lmbs = (machine->device_memory->base +
482                        memory_region_size(&machine->device_memory->mr)) /
483                        lmb_size;
484     uint32_t *int_buf, *cur_index, buf_len;
485 
486     /*
487      * Allocate enough buffer size to fit in ibm,dynamic-memory
488      */
489     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
490     cur_index = int_buf = g_malloc0(buf_len);
491     int_buf[0] = cpu_to_be32(nr_lmbs);
492     cur_index++;
493     for (i = 0; i < nr_lmbs; i++) {
494         uint64_t addr = i * lmb_size;
495         uint32_t *dynamic_memory = cur_index;
496 
497         if (i >= device_lmb_start) {
498             SpaprDrc *drc;
499 
500             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
501             g_assert(drc);
502 
503             dynamic_memory[0] = cpu_to_be32(addr >> 32);
504             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
505             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
506             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
507             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
508             if (memory_region_present(get_system_memory(), addr)) {
509                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
510             } else {
511                 dynamic_memory[5] = cpu_to_be32(0);
512             }
513         } else {
514             /*
515              * LMB information for RMA, boot time RAM and gap b/n RAM and
516              * device memory region -- all these are marked as reserved
517              * and as having no valid DRC.
518              */
519             dynamic_memory[0] = cpu_to_be32(addr >> 32);
520             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
521             dynamic_memory[2] = cpu_to_be32(0);
522             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
523             dynamic_memory[4] = cpu_to_be32(-1);
524             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
525                                             SPAPR_LMB_FLAGS_DRC_INVALID);
526         }
527 
528         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
529     }
530     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
531     g_free(int_buf);
532     if (ret < 0) {
533         return -1;
534     }
535     return 0;
536 }
537 
538 /*
539  * Adds ibm,dynamic-reconfiguration-memory node.
540  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
541  * of this device tree node.
542  */
543 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
544                                                    void *fdt)
545 {
546     MachineState *machine = MACHINE(spapr);
547     int ret, offset;
548     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
549     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
550                                 cpu_to_be32(lmb_size & 0xffffffff)};
551     MemoryDeviceInfoList *dimms = NULL;
552 
553     /*
554      * Don't create the node if there is no device memory
555      */
556     if (machine->ram_size == machine->maxram_size) {
557         return 0;
558     }
559 
560     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
561 
562     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
563                     sizeof(prop_lmb_size));
564     if (ret < 0) {
565         return ret;
566     }
567 
568     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
569     if (ret < 0) {
570         return ret;
571     }
572 
573     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
574     if (ret < 0) {
575         return ret;
576     }
577 
578     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
579     dimms = qmp_memory_device_list();
580     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
581         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
582     } else {
583         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
584     }
585     qapi_free_MemoryDeviceInfoList(dimms);
586 
587     if (ret < 0) {
588         return ret;
589     }
590 
591     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
592 
593     return ret;
594 }
595 
596 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
597 {
598     MachineState *machine = MACHINE(spapr);
599     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
600     hwaddr mem_start, node_size;
601     int i, nb_nodes = machine->numa_state->num_nodes;
602     NodeInfo *nodes = machine->numa_state->nodes;
603 
604     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
605         if (!nodes[i].node_mem) {
606             continue;
607         }
608         if (mem_start >= machine->ram_size) {
609             node_size = 0;
610         } else {
611             node_size = nodes[i].node_mem;
612             if (node_size > machine->ram_size - mem_start) {
613                 node_size = machine->ram_size - mem_start;
614             }
615         }
616         if (!mem_start) {
617             /* spapr_machine_init() checks for rma_size <= node0_size
618              * already */
619             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
620             mem_start += spapr->rma_size;
621             node_size -= spapr->rma_size;
622         }
623         for ( ; node_size; ) {
624             hwaddr sizetmp = pow2floor(node_size);
625 
626             /* mem_start != 0 here */
627             if (ctzl(mem_start) < ctzl(sizetmp)) {
628                 sizetmp = 1ULL << ctzl(mem_start);
629             }
630 
631             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
632             node_size -= sizetmp;
633             mem_start += sizetmp;
634         }
635     }
636 
637     /* Generate ibm,dynamic-reconfiguration-memory node if required */
638     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
639         int ret;
640 
641         g_assert(smc->dr_lmb_enabled);
642         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
643         if (ret) {
644             return ret;
645         }
646     }
647 
648     return 0;
649 }
650 
651 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
652                          SpaprMachineState *spapr)
653 {
654     MachineState *ms = MACHINE(spapr);
655     PowerPCCPU *cpu = POWERPC_CPU(cs);
656     CPUPPCState *env = &cpu->env;
657     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
658     int index = spapr_get_vcpu_id(cpu);
659     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
660                        0xffffffff, 0xffffffff};
661     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
662         : SPAPR_TIMEBASE_FREQ;
663     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
664     uint32_t page_sizes_prop[64];
665     size_t page_sizes_prop_size;
666     unsigned int smp_threads = ms->smp.threads;
667     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
668     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
669     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
670     SpaprDrc *drc;
671     int drc_index;
672     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
673     int i;
674 
675     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
676     if (drc) {
677         drc_index = spapr_drc_index(drc);
678         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
679     }
680 
681     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
682     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
683 
684     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
685     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
686                            env->dcache_line_size)));
687     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
688                            env->dcache_line_size)));
689     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
690                            env->icache_line_size)));
691     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
692                            env->icache_line_size)));
693 
694     if (pcc->l1_dcache_size) {
695         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
696                                pcc->l1_dcache_size)));
697     } else {
698         warn_report("Unknown L1 dcache size for cpu");
699     }
700     if (pcc->l1_icache_size) {
701         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
702                                pcc->l1_icache_size)));
703     } else {
704         warn_report("Unknown L1 icache size for cpu");
705     }
706 
707     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
708     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
709     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
710     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
711     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
712     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
713 
714     if (env->spr_cb[SPR_PURR].oea_read) {
715         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
716     }
717     if (env->spr_cb[SPR_SPURR].oea_read) {
718         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
719     }
720 
721     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
722         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
723                           segs, sizeof(segs))));
724     }
725 
726     /* Advertise VSX (vector extensions) if available
727      *   1               == VMX / Altivec available
728      *   2               == VSX available
729      *
730      * Only CPUs for which we create core types in spapr_cpu_core.c
731      * are possible, and all of those have VMX */
732     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
733         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
734     } else {
735         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
736     }
737 
738     /* Advertise DFP (Decimal Floating Point) if available
739      *   0 / no property == no DFP
740      *   1               == DFP available */
741     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
742         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
743     }
744 
745     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
746                                                       sizeof(page_sizes_prop));
747     if (page_sizes_prop_size) {
748         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
749                           page_sizes_prop, page_sizes_prop_size)));
750     }
751 
752     spapr_dt_pa_features(spapr, cpu, fdt, offset);
753 
754     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
755                            cs->cpu_index / vcpus_per_socket)));
756 
757     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
758                       pft_size_prop, sizeof(pft_size_prop))));
759 
760     if (ms->numa_state->num_nodes > 1) {
761         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
762     }
763 
764     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
765 
766     if (pcc->radix_page_info) {
767         for (i = 0; i < pcc->radix_page_info->count; i++) {
768             radix_AP_encodings[i] =
769                 cpu_to_be32(pcc->radix_page_info->entries[i]);
770         }
771         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
772                           radix_AP_encodings,
773                           pcc->radix_page_info->count *
774                           sizeof(radix_AP_encodings[0]))));
775     }
776 
777     /*
778      * We set this property to let the guest know that it can use the large
779      * decrementer and its width in bits.
780      */
781     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
782         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
783                               pcc->lrg_decr_bits)));
784 }
785 
786 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
787 {
788     CPUState **rev;
789     CPUState *cs;
790     int n_cpus;
791     int cpus_offset;
792     char *nodename;
793     int i;
794 
795     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
796     _FDT(cpus_offset);
797     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
798     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
799 
800     /*
801      * We walk the CPUs in reverse order to ensure that CPU DT nodes
802      * created by fdt_add_subnode() end up in the right order in FDT
803      * for the guest kernel the enumerate the CPUs correctly.
804      *
805      * The CPU list cannot be traversed in reverse order, so we need
806      * to do extra work.
807      */
808     n_cpus = 0;
809     rev = NULL;
810     CPU_FOREACH(cs) {
811         rev = g_renew(CPUState *, rev, n_cpus + 1);
812         rev[n_cpus++] = cs;
813     }
814 
815     for (i = n_cpus - 1; i >= 0; i--) {
816         CPUState *cs = rev[i];
817         PowerPCCPU *cpu = POWERPC_CPU(cs);
818         int index = spapr_get_vcpu_id(cpu);
819         DeviceClass *dc = DEVICE_GET_CLASS(cs);
820         int offset;
821 
822         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
823             continue;
824         }
825 
826         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
827         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
828         g_free(nodename);
829         _FDT(offset);
830         spapr_dt_cpu(cs, fdt, offset, spapr);
831     }
832 
833     g_free(rev);
834 }
835 
836 static int spapr_dt_rng(void *fdt)
837 {
838     int node;
839     int ret;
840 
841     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
842     if (node <= 0) {
843         return -1;
844     }
845     ret = fdt_setprop_string(fdt, node, "device_type",
846                              "ibm,platform-facilities");
847     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
848     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
849 
850     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
851     if (node <= 0) {
852         return -1;
853     }
854     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
855 
856     return ret ? -1 : 0;
857 }
858 
859 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
860 {
861     MachineState *ms = MACHINE(spapr);
862     int rtas;
863     GString *hypertas = g_string_sized_new(256);
864     GString *qemu_hypertas = g_string_sized_new(256);
865     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
866         memory_region_size(&MACHINE(spapr)->device_memory->mr);
867     uint32_t lrdr_capacity[] = {
868         cpu_to_be32(max_device_addr >> 32),
869         cpu_to_be32(max_device_addr & 0xffffffff),
870         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
871         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
872         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
873     };
874 
875     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
876 
877     /* hypertas */
878     add_str(hypertas, "hcall-pft");
879     add_str(hypertas, "hcall-term");
880     add_str(hypertas, "hcall-dabr");
881     add_str(hypertas, "hcall-interrupt");
882     add_str(hypertas, "hcall-tce");
883     add_str(hypertas, "hcall-vio");
884     add_str(hypertas, "hcall-splpar");
885     add_str(hypertas, "hcall-join");
886     add_str(hypertas, "hcall-bulk");
887     add_str(hypertas, "hcall-set-mode");
888     add_str(hypertas, "hcall-sprg0");
889     add_str(hypertas, "hcall-copy");
890     add_str(hypertas, "hcall-debug");
891     add_str(hypertas, "hcall-vphn");
892     add_str(qemu_hypertas, "hcall-memop1");
893 
894     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
895         add_str(hypertas, "hcall-multi-tce");
896     }
897 
898     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
899         add_str(hypertas, "hcall-hpt-resize");
900     }
901 
902     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
903                      hypertas->str, hypertas->len));
904     g_string_free(hypertas, TRUE);
905     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
906                      qemu_hypertas->str, qemu_hypertas->len));
907     g_string_free(qemu_hypertas, TRUE);
908 
909     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
910 
911     /*
912      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
913      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
914      *
915      * The system reset requirements are driven by existing Linux and PowerVM
916      * implementation which (contrary to PAPR) saves r3 in the error log
917      * structure like machine check, so Linux expects to find the saved r3
918      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
919      * does not look at the error value).
920      *
921      * System reset interrupts are not subject to interlock like machine
922      * check, so this memory area could be corrupted if the sreset is
923      * interrupted by a machine check (or vice versa) if it was shared. To
924      * prevent this, system reset uses per-CPU areas for the sreset save
925      * area. A system reset that interrupts a system reset handler could
926      * still overwrite this area, but Linux doesn't try to recover in that
927      * case anyway.
928      *
929      * The extra 8 bytes is required because Linux's FWNMI error log check
930      * is off-by-one.
931      */
932     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
933 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
934     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
935                           RTAS_ERROR_LOG_MAX));
936     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
937                           RTAS_EVENT_SCAN_RATE));
938 
939     g_assert(msi_nonbroken);
940     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
941 
942     /*
943      * According to PAPR, rtas ibm,os-term does not guarantee a return
944      * back to the guest cpu.
945      *
946      * While an additional ibm,extended-os-term property indicates
947      * that rtas call return will always occur. Set this property.
948      */
949     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
950 
951     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
952                      lrdr_capacity, sizeof(lrdr_capacity)));
953 
954     spapr_dt_rtas_tokens(fdt, rtas);
955 }
956 
957 /*
958  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
959  * and the XIVE features that the guest may request and thus the valid
960  * values for bytes 23..26 of option vector 5:
961  */
962 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
963                                           int chosen)
964 {
965     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
966 
967     char val[2 * 4] = {
968         23, 0x00, /* XICS / XIVE mode */
969         24, 0x00, /* Hash/Radix, filled in below. */
970         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
971         26, 0x40, /* Radix options: GTSE == yes. */
972     };
973 
974     if (spapr->irq->xics && spapr->irq->xive) {
975         val[1] = SPAPR_OV5_XIVE_BOTH;
976     } else if (spapr->irq->xive) {
977         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
978     } else {
979         assert(spapr->irq->xics);
980         val[1] = SPAPR_OV5_XIVE_LEGACY;
981     }
982 
983     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
984                           first_ppc_cpu->compat_pvr)) {
985         /*
986          * If we're in a pre POWER9 compat mode then the guest should
987          * do hash and use the legacy interrupt mode
988          */
989         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
990         val[3] = 0x00; /* Hash */
991     } else if (kvm_enabled()) {
992         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
993             val[3] = 0x80; /* OV5_MMU_BOTH */
994         } else if (kvmppc_has_cap_mmu_radix()) {
995             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
996         } else {
997             val[3] = 0x00; /* Hash */
998         }
999     } else {
1000         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1001         val[3] = 0xC0;
1002     }
1003     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1004                      val, sizeof(val)));
1005 }
1006 
1007 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1008 {
1009     MachineState *machine = MACHINE(spapr);
1010     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1011     int chosen;
1012 
1013     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1014 
1015     if (reset) {
1016         const char *boot_device = machine->boot_order;
1017         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1018         size_t cb = 0;
1019         char *bootlist = get_boot_devices_list(&cb);
1020 
1021         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1022             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1023                                     machine->kernel_cmdline));
1024         }
1025 
1026         if (spapr->initrd_size) {
1027             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1028                                   spapr->initrd_base));
1029             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1030                                   spapr->initrd_base + spapr->initrd_size));
1031         }
1032 
1033         if (spapr->kernel_size) {
1034             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1035                                   cpu_to_be64(spapr->kernel_size) };
1036 
1037             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1038                          &kprop, sizeof(kprop)));
1039             if (spapr->kernel_le) {
1040                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1041             }
1042         }
1043         if (boot_menu) {
1044             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1045         }
1046         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1047         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1048         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1049 
1050         if (cb && bootlist) {
1051             int i;
1052 
1053             for (i = 0; i < cb; i++) {
1054                 if (bootlist[i] == '\n') {
1055                     bootlist[i] = ' ';
1056                 }
1057             }
1058             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1059         }
1060 
1061         if (boot_device && strlen(boot_device)) {
1062             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1063         }
1064 
1065         if (!spapr->has_graphics && stdout_path) {
1066             /*
1067              * "linux,stdout-path" and "stdout" properties are
1068              * deprecated by linux kernel. New platforms should only
1069              * use the "stdout-path" property. Set the new property
1070              * and continue using older property to remain compatible
1071              * with the existing firmware.
1072              */
1073             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1074             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1075         }
1076 
1077         /*
1078          * We can deal with BAR reallocation just fine, advertise it
1079          * to the guest
1080          */
1081         if (smc->linux_pci_probe) {
1082             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1083         }
1084 
1085         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1086 
1087         g_free(stdout_path);
1088         g_free(bootlist);
1089     }
1090 
1091     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1092 }
1093 
1094 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1095 {
1096     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1097      * KVM to work under pHyp with some guest co-operation */
1098     int hypervisor;
1099     uint8_t hypercall[16];
1100 
1101     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1102     /* indicate KVM hypercall interface */
1103     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1104     if (kvmppc_has_cap_fixup_hcalls()) {
1105         /*
1106          * Older KVM versions with older guest kernels were broken
1107          * with the magic page, don't allow the guest to map it.
1108          */
1109         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1110                                   sizeof(hypercall))) {
1111             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1112                              hypercall, sizeof(hypercall)));
1113         }
1114     }
1115 }
1116 
1117 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1118 {
1119     MachineState *machine = MACHINE(spapr);
1120     MachineClass *mc = MACHINE_GET_CLASS(machine);
1121     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1122     int ret;
1123     void *fdt;
1124     SpaprPhbState *phb;
1125     char *buf;
1126 
1127     fdt = g_malloc0(space);
1128     _FDT((fdt_create_empty_tree(fdt, space)));
1129 
1130     /* Root node */
1131     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1132     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1133     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1134 
1135     /* Guest UUID & Name*/
1136     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1137     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1138     if (qemu_uuid_set) {
1139         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1140     }
1141     g_free(buf);
1142 
1143     if (qemu_get_vm_name()) {
1144         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1145                                 qemu_get_vm_name()));
1146     }
1147 
1148     /* Host Model & Serial Number */
1149     if (spapr->host_model) {
1150         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1151     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1152         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1153         g_free(buf);
1154     }
1155 
1156     if (spapr->host_serial) {
1157         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1158     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1159         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1160         g_free(buf);
1161     }
1162 
1163     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1164     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1165 
1166     /* /interrupt controller */
1167     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1168 
1169     ret = spapr_dt_memory(spapr, fdt);
1170     if (ret < 0) {
1171         error_report("couldn't setup memory nodes in fdt");
1172         exit(1);
1173     }
1174 
1175     /* /vdevice */
1176     spapr_dt_vdevice(spapr->vio_bus, fdt);
1177 
1178     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1179         ret = spapr_dt_rng(fdt);
1180         if (ret < 0) {
1181             error_report("could not set up rng device in the fdt");
1182             exit(1);
1183         }
1184     }
1185 
1186     QLIST_FOREACH(phb, &spapr->phbs, list) {
1187         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1188         if (ret < 0) {
1189             error_report("couldn't setup PCI devices in fdt");
1190             exit(1);
1191         }
1192     }
1193 
1194     spapr_dt_cpus(fdt, spapr);
1195 
1196     if (smc->dr_lmb_enabled) {
1197         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1198     }
1199 
1200     if (mc->has_hotpluggable_cpus) {
1201         int offset = fdt_path_offset(fdt, "/cpus");
1202         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1203         if (ret < 0) {
1204             error_report("Couldn't set up CPU DR device tree properties");
1205             exit(1);
1206         }
1207     }
1208 
1209     /* /event-sources */
1210     spapr_dt_events(spapr, fdt);
1211 
1212     /* /rtas */
1213     spapr_dt_rtas(spapr, fdt);
1214 
1215     /* /chosen */
1216     spapr_dt_chosen(spapr, fdt, reset);
1217 
1218     /* /hypervisor */
1219     if (kvm_enabled()) {
1220         spapr_dt_hypervisor(spapr, fdt);
1221     }
1222 
1223     /* Build memory reserve map */
1224     if (reset) {
1225         if (spapr->kernel_size) {
1226             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1227                                   spapr->kernel_size)));
1228         }
1229         if (spapr->initrd_size) {
1230             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1231                                   spapr->initrd_size)));
1232         }
1233     }
1234 
1235     if (smc->dr_phb_enabled) {
1236         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1237         if (ret < 0) {
1238             error_report("Couldn't set up PHB DR device tree properties");
1239             exit(1);
1240         }
1241     }
1242 
1243     /* NVDIMM devices */
1244     if (mc->nvdimm_supported) {
1245         spapr_dt_persistent_memory(spapr, fdt);
1246     }
1247 
1248     return fdt;
1249 }
1250 
1251 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1252 {
1253     SpaprMachineState *spapr = opaque;
1254 
1255     return (addr & 0x0fffffff) + spapr->kernel_addr;
1256 }
1257 
1258 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1259                                     PowerPCCPU *cpu)
1260 {
1261     CPUPPCState *env = &cpu->env;
1262 
1263     /* The TCG path should also be holding the BQL at this point */
1264     g_assert(qemu_mutex_iothread_locked());
1265 
1266     if (msr_pr) {
1267         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1268         env->gpr[3] = H_PRIVILEGE;
1269     } else {
1270         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1271     }
1272 }
1273 
1274 struct LPCRSyncState {
1275     target_ulong value;
1276     target_ulong mask;
1277 };
1278 
1279 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1280 {
1281     struct LPCRSyncState *s = arg.host_ptr;
1282     PowerPCCPU *cpu = POWERPC_CPU(cs);
1283     CPUPPCState *env = &cpu->env;
1284     target_ulong lpcr;
1285 
1286     cpu_synchronize_state(cs);
1287     lpcr = env->spr[SPR_LPCR];
1288     lpcr &= ~s->mask;
1289     lpcr |= s->value;
1290     ppc_store_lpcr(cpu, lpcr);
1291 }
1292 
1293 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1294 {
1295     CPUState *cs;
1296     struct LPCRSyncState s = {
1297         .value = value,
1298         .mask = mask
1299     };
1300     CPU_FOREACH(cs) {
1301         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1302     }
1303 }
1304 
1305 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1306 {
1307     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1308 
1309     /* Copy PATE1:GR into PATE0:HR */
1310     entry->dw0 = spapr->patb_entry & PATE0_HR;
1311     entry->dw1 = spapr->patb_entry;
1312 }
1313 
1314 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1315 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1316 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1317 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1318 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1319 
1320 /*
1321  * Get the fd to access the kernel htab, re-opening it if necessary
1322  */
1323 static int get_htab_fd(SpaprMachineState *spapr)
1324 {
1325     Error *local_err = NULL;
1326 
1327     if (spapr->htab_fd >= 0) {
1328         return spapr->htab_fd;
1329     }
1330 
1331     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1332     if (spapr->htab_fd < 0) {
1333         error_report_err(local_err);
1334     }
1335 
1336     return spapr->htab_fd;
1337 }
1338 
1339 void close_htab_fd(SpaprMachineState *spapr)
1340 {
1341     if (spapr->htab_fd >= 0) {
1342         close(spapr->htab_fd);
1343     }
1344     spapr->htab_fd = -1;
1345 }
1346 
1347 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1348 {
1349     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1350 
1351     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1352 }
1353 
1354 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1355 {
1356     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1357 
1358     assert(kvm_enabled());
1359 
1360     if (!spapr->htab) {
1361         return 0;
1362     }
1363 
1364     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1365 }
1366 
1367 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1368                                                 hwaddr ptex, int n)
1369 {
1370     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1371     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1372 
1373     if (!spapr->htab) {
1374         /*
1375          * HTAB is controlled by KVM. Fetch into temporary buffer
1376          */
1377         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1378         kvmppc_read_hptes(hptes, ptex, n);
1379         return hptes;
1380     }
1381 
1382     /*
1383      * HTAB is controlled by QEMU. Just point to the internally
1384      * accessible PTEG.
1385      */
1386     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1387 }
1388 
1389 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1390                               const ppc_hash_pte64_t *hptes,
1391                               hwaddr ptex, int n)
1392 {
1393     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1394 
1395     if (!spapr->htab) {
1396         g_free((void *)hptes);
1397     }
1398 
1399     /* Nothing to do for qemu managed HPT */
1400 }
1401 
1402 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1403                       uint64_t pte0, uint64_t pte1)
1404 {
1405     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1406     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1407 
1408     if (!spapr->htab) {
1409         kvmppc_write_hpte(ptex, pte0, pte1);
1410     } else {
1411         if (pte0 & HPTE64_V_VALID) {
1412             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1413             /*
1414              * When setting valid, we write PTE1 first. This ensures
1415              * proper synchronization with the reading code in
1416              * ppc_hash64_pteg_search()
1417              */
1418             smp_wmb();
1419             stq_p(spapr->htab + offset, pte0);
1420         } else {
1421             stq_p(spapr->htab + offset, pte0);
1422             /*
1423              * When clearing it we set PTE0 first. This ensures proper
1424              * synchronization with the reading code in
1425              * ppc_hash64_pteg_search()
1426              */
1427             smp_wmb();
1428             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1429         }
1430     }
1431 }
1432 
1433 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1434                              uint64_t pte1)
1435 {
1436     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1437     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1438 
1439     if (!spapr->htab) {
1440         /* There should always be a hash table when this is called */
1441         error_report("spapr_hpte_set_c called with no hash table !");
1442         return;
1443     }
1444 
1445     /* The HW performs a non-atomic byte update */
1446     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1447 }
1448 
1449 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1450                              uint64_t pte1)
1451 {
1452     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1453     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1454 
1455     if (!spapr->htab) {
1456         /* There should always be a hash table when this is called */
1457         error_report("spapr_hpte_set_r called with no hash table !");
1458         return;
1459     }
1460 
1461     /* The HW performs a non-atomic byte update */
1462     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1463 }
1464 
1465 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1466 {
1467     int shift;
1468 
1469     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1470      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1471      * that's much more than is needed for Linux guests */
1472     shift = ctz64(pow2ceil(ramsize)) - 7;
1473     shift = MAX(shift, 18); /* Minimum architected size */
1474     shift = MIN(shift, 46); /* Maximum architected size */
1475     return shift;
1476 }
1477 
1478 void spapr_free_hpt(SpaprMachineState *spapr)
1479 {
1480     g_free(spapr->htab);
1481     spapr->htab = NULL;
1482     spapr->htab_shift = 0;
1483     close_htab_fd(spapr);
1484 }
1485 
1486 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1487 {
1488     ERRP_GUARD();
1489     long rc;
1490 
1491     /* Clean up any HPT info from a previous boot */
1492     spapr_free_hpt(spapr);
1493 
1494     rc = kvmppc_reset_htab(shift);
1495 
1496     if (rc == -EOPNOTSUPP) {
1497         error_setg(errp, "HPT not supported in nested guests");
1498         return -EOPNOTSUPP;
1499     }
1500 
1501     if (rc < 0) {
1502         /* kernel-side HPT needed, but couldn't allocate one */
1503         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1504                          shift);
1505         error_append_hint(errp, "Try smaller maxmem?\n");
1506         return -errno;
1507     } else if (rc > 0) {
1508         /* kernel-side HPT allocated */
1509         if (rc != shift) {
1510             error_setg(errp,
1511                        "Requested order %d HPT, but kernel allocated order %ld",
1512                        shift, rc);
1513             error_append_hint(errp, "Try smaller maxmem?\n");
1514             return -ENOSPC;
1515         }
1516 
1517         spapr->htab_shift = shift;
1518         spapr->htab = NULL;
1519     } else {
1520         /* kernel-side HPT not needed, allocate in userspace instead */
1521         size_t size = 1ULL << shift;
1522         int i;
1523 
1524         spapr->htab = qemu_memalign(size, size);
1525         if (!spapr->htab) {
1526             error_setg_errno(errp, errno,
1527                              "Could not allocate HPT of order %d", shift);
1528             return -ENOMEM;
1529         }
1530 
1531         memset(spapr->htab, 0, size);
1532         spapr->htab_shift = shift;
1533 
1534         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1535             DIRTY_HPTE(HPTE(spapr->htab, i));
1536         }
1537     }
1538     /* We're setting up a hash table, so that means we're not radix */
1539     spapr->patb_entry = 0;
1540     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1541     return 0;
1542 }
1543 
1544 void spapr_setup_hpt(SpaprMachineState *spapr)
1545 {
1546     int hpt_shift;
1547 
1548     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1549         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1550     } else {
1551         uint64_t current_ram_size;
1552 
1553         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1554         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1555     }
1556     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1557 
1558     if (kvm_enabled()) {
1559         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1560 
1561         /* Check our RMA fits in the possible VRMA */
1562         if (vrma_limit < spapr->rma_size) {
1563             error_report("Unable to create %" HWADDR_PRIu
1564                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1565                          spapr->rma_size / MiB, vrma_limit / MiB);
1566             exit(EXIT_FAILURE);
1567         }
1568     }
1569 }
1570 
1571 static int spapr_reset_drcs(Object *child, void *opaque)
1572 {
1573     SpaprDrc *drc =
1574         (SpaprDrc *) object_dynamic_cast(child,
1575                                                  TYPE_SPAPR_DR_CONNECTOR);
1576 
1577     if (drc) {
1578         spapr_drc_reset(drc);
1579     }
1580 
1581     return 0;
1582 }
1583 
1584 static void spapr_machine_reset(MachineState *machine)
1585 {
1586     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1587     PowerPCCPU *first_ppc_cpu;
1588     hwaddr fdt_addr;
1589     void *fdt;
1590     int rc;
1591 
1592     kvmppc_svm_off(&error_fatal);
1593     spapr_caps_apply(spapr);
1594 
1595     first_ppc_cpu = POWERPC_CPU(first_cpu);
1596     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1597         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1598                               spapr->max_compat_pvr)) {
1599         /*
1600          * If using KVM with radix mode available, VCPUs can be started
1601          * without a HPT because KVM will start them in radix mode.
1602          * Set the GR bit in PATE so that we know there is no HPT.
1603          */
1604         spapr->patb_entry = PATE1_GR;
1605         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1606     } else {
1607         spapr_setup_hpt(spapr);
1608     }
1609 
1610     qemu_devices_reset();
1611 
1612     spapr_ovec_cleanup(spapr->ov5_cas);
1613     spapr->ov5_cas = spapr_ovec_new();
1614 
1615     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1616 
1617     /*
1618      * This is fixing some of the default configuration of the XIVE
1619      * devices. To be called after the reset of the machine devices.
1620      */
1621     spapr_irq_reset(spapr, &error_fatal);
1622 
1623     /*
1624      * There is no CAS under qtest. Simulate one to please the code that
1625      * depends on spapr->ov5_cas. This is especially needed to test device
1626      * unplug, so we do that before resetting the DRCs.
1627      */
1628     if (qtest_enabled()) {
1629         spapr_ovec_cleanup(spapr->ov5_cas);
1630         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1631     }
1632 
1633     /* DRC reset may cause a device to be unplugged. This will cause troubles
1634      * if this device is used by another device (eg, a running vhost backend
1635      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1636      * situations, we reset DRCs after all devices have been reset.
1637      */
1638     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1639 
1640     spapr_clear_pending_events(spapr);
1641 
1642     /*
1643      * We place the device tree and RTAS just below either the top of the RMA,
1644      * or just below 2GB, whichever is lower, so that it can be
1645      * processed with 32-bit real mode code if necessary
1646      */
1647     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1648 
1649     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1650 
1651     rc = fdt_pack(fdt);
1652 
1653     /* Should only fail if we've built a corrupted tree */
1654     assert(rc == 0);
1655 
1656     /* Load the fdt */
1657     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1658     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1659     g_free(spapr->fdt_blob);
1660     spapr->fdt_size = fdt_totalsize(fdt);
1661     spapr->fdt_initial_size = spapr->fdt_size;
1662     spapr->fdt_blob = fdt;
1663 
1664     /* Set up the entry state */
1665     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1666     first_ppc_cpu->env.gpr[5] = 0;
1667 
1668     spapr->fwnmi_system_reset_addr = -1;
1669     spapr->fwnmi_machine_check_addr = -1;
1670     spapr->fwnmi_machine_check_interlock = -1;
1671 
1672     /* Signal all vCPUs waiting on this condition */
1673     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1674 
1675     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1676 }
1677 
1678 static void spapr_create_nvram(SpaprMachineState *spapr)
1679 {
1680     DeviceState *dev = qdev_new("spapr-nvram");
1681     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1682 
1683     if (dinfo) {
1684         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1685                                 &error_fatal);
1686     }
1687 
1688     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1689 
1690     spapr->nvram = (struct SpaprNvram *)dev;
1691 }
1692 
1693 static void spapr_rtc_create(SpaprMachineState *spapr)
1694 {
1695     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1696                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1697                                        &error_fatal, NULL);
1698     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1699     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1700                               "date");
1701 }
1702 
1703 /* Returns whether we want to use VGA or not */
1704 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1705 {
1706     switch (vga_interface_type) {
1707     case VGA_NONE:
1708         return false;
1709     case VGA_DEVICE:
1710         return true;
1711     case VGA_STD:
1712     case VGA_VIRTIO:
1713     case VGA_CIRRUS:
1714         return pci_vga_init(pci_bus) != NULL;
1715     default:
1716         error_setg(errp,
1717                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1718         return false;
1719     }
1720 }
1721 
1722 static int spapr_pre_load(void *opaque)
1723 {
1724     int rc;
1725 
1726     rc = spapr_caps_pre_load(opaque);
1727     if (rc) {
1728         return rc;
1729     }
1730 
1731     return 0;
1732 }
1733 
1734 static int spapr_post_load(void *opaque, int version_id)
1735 {
1736     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1737     int err = 0;
1738 
1739     err = spapr_caps_post_migration(spapr);
1740     if (err) {
1741         return err;
1742     }
1743 
1744     /*
1745      * In earlier versions, there was no separate qdev for the PAPR
1746      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1747      * So when migrating from those versions, poke the incoming offset
1748      * value into the RTC device
1749      */
1750     if (version_id < 3) {
1751         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1752         if (err) {
1753             return err;
1754         }
1755     }
1756 
1757     if (kvm_enabled() && spapr->patb_entry) {
1758         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1759         bool radix = !!(spapr->patb_entry & PATE1_GR);
1760         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1761 
1762         /*
1763          * Update LPCR:HR and UPRT as they may not be set properly in
1764          * the stream
1765          */
1766         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1767                             LPCR_HR | LPCR_UPRT);
1768 
1769         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1770         if (err) {
1771             error_report("Process table config unsupported by the host");
1772             return -EINVAL;
1773         }
1774     }
1775 
1776     err = spapr_irq_post_load(spapr, version_id);
1777     if (err) {
1778         return err;
1779     }
1780 
1781     return err;
1782 }
1783 
1784 static int spapr_pre_save(void *opaque)
1785 {
1786     int rc;
1787 
1788     rc = spapr_caps_pre_save(opaque);
1789     if (rc) {
1790         return rc;
1791     }
1792 
1793     return 0;
1794 }
1795 
1796 static bool version_before_3(void *opaque, int version_id)
1797 {
1798     return version_id < 3;
1799 }
1800 
1801 static bool spapr_pending_events_needed(void *opaque)
1802 {
1803     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1804     return !QTAILQ_EMPTY(&spapr->pending_events);
1805 }
1806 
1807 static const VMStateDescription vmstate_spapr_event_entry = {
1808     .name = "spapr_event_log_entry",
1809     .version_id = 1,
1810     .minimum_version_id = 1,
1811     .fields = (VMStateField[]) {
1812         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1813         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1814         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1815                                      NULL, extended_length),
1816         VMSTATE_END_OF_LIST()
1817     },
1818 };
1819 
1820 static const VMStateDescription vmstate_spapr_pending_events = {
1821     .name = "spapr_pending_events",
1822     .version_id = 1,
1823     .minimum_version_id = 1,
1824     .needed = spapr_pending_events_needed,
1825     .fields = (VMStateField[]) {
1826         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1827                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1828         VMSTATE_END_OF_LIST()
1829     },
1830 };
1831 
1832 static bool spapr_ov5_cas_needed(void *opaque)
1833 {
1834     SpaprMachineState *spapr = opaque;
1835     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1836     bool cas_needed;
1837 
1838     /* Prior to the introduction of SpaprOptionVector, we had two option
1839      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1840      * Both of these options encode machine topology into the device-tree
1841      * in such a way that the now-booted OS should still be able to interact
1842      * appropriately with QEMU regardless of what options were actually
1843      * negotiatied on the source side.
1844      *
1845      * As such, we can avoid migrating the CAS-negotiated options if these
1846      * are the only options available on the current machine/platform.
1847      * Since these are the only options available for pseries-2.7 and
1848      * earlier, this allows us to maintain old->new/new->old migration
1849      * compatibility.
1850      *
1851      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1852      * via default pseries-2.8 machines and explicit command-line parameters.
1853      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1854      * of the actual CAS-negotiated values to continue working properly. For
1855      * example, availability of memory unplug depends on knowing whether
1856      * OV5_HP_EVT was negotiated via CAS.
1857      *
1858      * Thus, for any cases where the set of available CAS-negotiatable
1859      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1860      * include the CAS-negotiated options in the migration stream, unless
1861      * if they affect boot time behaviour only.
1862      */
1863     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1864     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1865     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1866 
1867     /* We need extra information if we have any bits outside the mask
1868      * defined above */
1869     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1870 
1871     spapr_ovec_cleanup(ov5_mask);
1872 
1873     return cas_needed;
1874 }
1875 
1876 static const VMStateDescription vmstate_spapr_ov5_cas = {
1877     .name = "spapr_option_vector_ov5_cas",
1878     .version_id = 1,
1879     .minimum_version_id = 1,
1880     .needed = spapr_ov5_cas_needed,
1881     .fields = (VMStateField[]) {
1882         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1883                                  vmstate_spapr_ovec, SpaprOptionVector),
1884         VMSTATE_END_OF_LIST()
1885     },
1886 };
1887 
1888 static bool spapr_patb_entry_needed(void *opaque)
1889 {
1890     SpaprMachineState *spapr = opaque;
1891 
1892     return !!spapr->patb_entry;
1893 }
1894 
1895 static const VMStateDescription vmstate_spapr_patb_entry = {
1896     .name = "spapr_patb_entry",
1897     .version_id = 1,
1898     .minimum_version_id = 1,
1899     .needed = spapr_patb_entry_needed,
1900     .fields = (VMStateField[]) {
1901         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1902         VMSTATE_END_OF_LIST()
1903     },
1904 };
1905 
1906 static bool spapr_irq_map_needed(void *opaque)
1907 {
1908     SpaprMachineState *spapr = opaque;
1909 
1910     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1911 }
1912 
1913 static const VMStateDescription vmstate_spapr_irq_map = {
1914     .name = "spapr_irq_map",
1915     .version_id = 1,
1916     .minimum_version_id = 1,
1917     .needed = spapr_irq_map_needed,
1918     .fields = (VMStateField[]) {
1919         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1920         VMSTATE_END_OF_LIST()
1921     },
1922 };
1923 
1924 static bool spapr_dtb_needed(void *opaque)
1925 {
1926     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1927 
1928     return smc->update_dt_enabled;
1929 }
1930 
1931 static int spapr_dtb_pre_load(void *opaque)
1932 {
1933     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1934 
1935     g_free(spapr->fdt_blob);
1936     spapr->fdt_blob = NULL;
1937     spapr->fdt_size = 0;
1938 
1939     return 0;
1940 }
1941 
1942 static const VMStateDescription vmstate_spapr_dtb = {
1943     .name = "spapr_dtb",
1944     .version_id = 1,
1945     .minimum_version_id = 1,
1946     .needed = spapr_dtb_needed,
1947     .pre_load = spapr_dtb_pre_load,
1948     .fields = (VMStateField[]) {
1949         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1950         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1951         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1952                                      fdt_size),
1953         VMSTATE_END_OF_LIST()
1954     },
1955 };
1956 
1957 static bool spapr_fwnmi_needed(void *opaque)
1958 {
1959     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1960 
1961     return spapr->fwnmi_machine_check_addr != -1;
1962 }
1963 
1964 static int spapr_fwnmi_pre_save(void *opaque)
1965 {
1966     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1967 
1968     /*
1969      * Check if machine check handling is in progress and print a
1970      * warning message.
1971      */
1972     if (spapr->fwnmi_machine_check_interlock != -1) {
1973         warn_report("A machine check is being handled during migration. The"
1974                 "handler may run and log hardware error on the destination");
1975     }
1976 
1977     return 0;
1978 }
1979 
1980 static const VMStateDescription vmstate_spapr_fwnmi = {
1981     .name = "spapr_fwnmi",
1982     .version_id = 1,
1983     .minimum_version_id = 1,
1984     .needed = spapr_fwnmi_needed,
1985     .pre_save = spapr_fwnmi_pre_save,
1986     .fields = (VMStateField[]) {
1987         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
1988         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
1989         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
1990         VMSTATE_END_OF_LIST()
1991     },
1992 };
1993 
1994 static const VMStateDescription vmstate_spapr = {
1995     .name = "spapr",
1996     .version_id = 3,
1997     .minimum_version_id = 1,
1998     .pre_load = spapr_pre_load,
1999     .post_load = spapr_post_load,
2000     .pre_save = spapr_pre_save,
2001     .fields = (VMStateField[]) {
2002         /* used to be @next_irq */
2003         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2004 
2005         /* RTC offset */
2006         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2007 
2008         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2009         VMSTATE_END_OF_LIST()
2010     },
2011     .subsections = (const VMStateDescription*[]) {
2012         &vmstate_spapr_ov5_cas,
2013         &vmstate_spapr_patb_entry,
2014         &vmstate_spapr_pending_events,
2015         &vmstate_spapr_cap_htm,
2016         &vmstate_spapr_cap_vsx,
2017         &vmstate_spapr_cap_dfp,
2018         &vmstate_spapr_cap_cfpc,
2019         &vmstate_spapr_cap_sbbc,
2020         &vmstate_spapr_cap_ibs,
2021         &vmstate_spapr_cap_hpt_maxpagesize,
2022         &vmstate_spapr_irq_map,
2023         &vmstate_spapr_cap_nested_kvm_hv,
2024         &vmstate_spapr_dtb,
2025         &vmstate_spapr_cap_large_decr,
2026         &vmstate_spapr_cap_ccf_assist,
2027         &vmstate_spapr_cap_fwnmi,
2028         &vmstate_spapr_fwnmi,
2029         NULL
2030     }
2031 };
2032 
2033 static int htab_save_setup(QEMUFile *f, void *opaque)
2034 {
2035     SpaprMachineState *spapr = opaque;
2036 
2037     /* "Iteration" header */
2038     if (!spapr->htab_shift) {
2039         qemu_put_be32(f, -1);
2040     } else {
2041         qemu_put_be32(f, spapr->htab_shift);
2042     }
2043 
2044     if (spapr->htab) {
2045         spapr->htab_save_index = 0;
2046         spapr->htab_first_pass = true;
2047     } else {
2048         if (spapr->htab_shift) {
2049             assert(kvm_enabled());
2050         }
2051     }
2052 
2053 
2054     return 0;
2055 }
2056 
2057 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2058                             int chunkstart, int n_valid, int n_invalid)
2059 {
2060     qemu_put_be32(f, chunkstart);
2061     qemu_put_be16(f, n_valid);
2062     qemu_put_be16(f, n_invalid);
2063     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2064                     HASH_PTE_SIZE_64 * n_valid);
2065 }
2066 
2067 static void htab_save_end_marker(QEMUFile *f)
2068 {
2069     qemu_put_be32(f, 0);
2070     qemu_put_be16(f, 0);
2071     qemu_put_be16(f, 0);
2072 }
2073 
2074 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2075                                  int64_t max_ns)
2076 {
2077     bool has_timeout = max_ns != -1;
2078     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2079     int index = spapr->htab_save_index;
2080     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2081 
2082     assert(spapr->htab_first_pass);
2083 
2084     do {
2085         int chunkstart;
2086 
2087         /* Consume invalid HPTEs */
2088         while ((index < htabslots)
2089                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2090             CLEAN_HPTE(HPTE(spapr->htab, index));
2091             index++;
2092         }
2093 
2094         /* Consume valid HPTEs */
2095         chunkstart = index;
2096         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2097                && HPTE_VALID(HPTE(spapr->htab, index))) {
2098             CLEAN_HPTE(HPTE(spapr->htab, index));
2099             index++;
2100         }
2101 
2102         if (index > chunkstart) {
2103             int n_valid = index - chunkstart;
2104 
2105             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2106 
2107             if (has_timeout &&
2108                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2109                 break;
2110             }
2111         }
2112     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2113 
2114     if (index >= htabslots) {
2115         assert(index == htabslots);
2116         index = 0;
2117         spapr->htab_first_pass = false;
2118     }
2119     spapr->htab_save_index = index;
2120 }
2121 
2122 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2123                                 int64_t max_ns)
2124 {
2125     bool final = max_ns < 0;
2126     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2127     int examined = 0, sent = 0;
2128     int index = spapr->htab_save_index;
2129     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2130 
2131     assert(!spapr->htab_first_pass);
2132 
2133     do {
2134         int chunkstart, invalidstart;
2135 
2136         /* Consume non-dirty HPTEs */
2137         while ((index < htabslots)
2138                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2139             index++;
2140             examined++;
2141         }
2142 
2143         chunkstart = index;
2144         /* Consume valid dirty HPTEs */
2145         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2146                && HPTE_DIRTY(HPTE(spapr->htab, index))
2147                && HPTE_VALID(HPTE(spapr->htab, index))) {
2148             CLEAN_HPTE(HPTE(spapr->htab, index));
2149             index++;
2150             examined++;
2151         }
2152 
2153         invalidstart = index;
2154         /* Consume invalid dirty HPTEs */
2155         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2156                && HPTE_DIRTY(HPTE(spapr->htab, index))
2157                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2158             CLEAN_HPTE(HPTE(spapr->htab, index));
2159             index++;
2160             examined++;
2161         }
2162 
2163         if (index > chunkstart) {
2164             int n_valid = invalidstart - chunkstart;
2165             int n_invalid = index - invalidstart;
2166 
2167             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2168             sent += index - chunkstart;
2169 
2170             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2171                 break;
2172             }
2173         }
2174 
2175         if (examined >= htabslots) {
2176             break;
2177         }
2178 
2179         if (index >= htabslots) {
2180             assert(index == htabslots);
2181             index = 0;
2182         }
2183     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2184 
2185     if (index >= htabslots) {
2186         assert(index == htabslots);
2187         index = 0;
2188     }
2189 
2190     spapr->htab_save_index = index;
2191 
2192     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2193 }
2194 
2195 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2196 #define MAX_KVM_BUF_SIZE    2048
2197 
2198 static int htab_save_iterate(QEMUFile *f, void *opaque)
2199 {
2200     SpaprMachineState *spapr = opaque;
2201     int fd;
2202     int rc = 0;
2203 
2204     /* Iteration header */
2205     if (!spapr->htab_shift) {
2206         qemu_put_be32(f, -1);
2207         return 1;
2208     } else {
2209         qemu_put_be32(f, 0);
2210     }
2211 
2212     if (!spapr->htab) {
2213         assert(kvm_enabled());
2214 
2215         fd = get_htab_fd(spapr);
2216         if (fd < 0) {
2217             return fd;
2218         }
2219 
2220         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2221         if (rc < 0) {
2222             return rc;
2223         }
2224     } else  if (spapr->htab_first_pass) {
2225         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2226     } else {
2227         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2228     }
2229 
2230     htab_save_end_marker(f);
2231 
2232     return rc;
2233 }
2234 
2235 static int htab_save_complete(QEMUFile *f, void *opaque)
2236 {
2237     SpaprMachineState *spapr = opaque;
2238     int fd;
2239 
2240     /* Iteration header */
2241     if (!spapr->htab_shift) {
2242         qemu_put_be32(f, -1);
2243         return 0;
2244     } else {
2245         qemu_put_be32(f, 0);
2246     }
2247 
2248     if (!spapr->htab) {
2249         int rc;
2250 
2251         assert(kvm_enabled());
2252 
2253         fd = get_htab_fd(spapr);
2254         if (fd < 0) {
2255             return fd;
2256         }
2257 
2258         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2259         if (rc < 0) {
2260             return rc;
2261         }
2262     } else {
2263         if (spapr->htab_first_pass) {
2264             htab_save_first_pass(f, spapr, -1);
2265         }
2266         htab_save_later_pass(f, spapr, -1);
2267     }
2268 
2269     /* End marker */
2270     htab_save_end_marker(f);
2271 
2272     return 0;
2273 }
2274 
2275 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2276 {
2277     SpaprMachineState *spapr = opaque;
2278     uint32_t section_hdr;
2279     int fd = -1;
2280     Error *local_err = NULL;
2281 
2282     if (version_id < 1 || version_id > 1) {
2283         error_report("htab_load() bad version");
2284         return -EINVAL;
2285     }
2286 
2287     section_hdr = qemu_get_be32(f);
2288 
2289     if (section_hdr == -1) {
2290         spapr_free_hpt(spapr);
2291         return 0;
2292     }
2293 
2294     if (section_hdr) {
2295         int ret;
2296 
2297         /* First section gives the htab size */
2298         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2299         if (ret < 0) {
2300             error_report_err(local_err);
2301             return ret;
2302         }
2303         return 0;
2304     }
2305 
2306     if (!spapr->htab) {
2307         assert(kvm_enabled());
2308 
2309         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2310         if (fd < 0) {
2311             error_report_err(local_err);
2312             return fd;
2313         }
2314     }
2315 
2316     while (true) {
2317         uint32_t index;
2318         uint16_t n_valid, n_invalid;
2319 
2320         index = qemu_get_be32(f);
2321         n_valid = qemu_get_be16(f);
2322         n_invalid = qemu_get_be16(f);
2323 
2324         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2325             /* End of Stream */
2326             break;
2327         }
2328 
2329         if ((index + n_valid + n_invalid) >
2330             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2331             /* Bad index in stream */
2332             error_report(
2333                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2334                 index, n_valid, n_invalid, spapr->htab_shift);
2335             return -EINVAL;
2336         }
2337 
2338         if (spapr->htab) {
2339             if (n_valid) {
2340                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2341                                 HASH_PTE_SIZE_64 * n_valid);
2342             }
2343             if (n_invalid) {
2344                 memset(HPTE(spapr->htab, index + n_valid), 0,
2345                        HASH_PTE_SIZE_64 * n_invalid);
2346             }
2347         } else {
2348             int rc;
2349 
2350             assert(fd >= 0);
2351 
2352             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2353                                         &local_err);
2354             if (rc < 0) {
2355                 error_report_err(local_err);
2356                 return rc;
2357             }
2358         }
2359     }
2360 
2361     if (!spapr->htab) {
2362         assert(fd >= 0);
2363         close(fd);
2364     }
2365 
2366     return 0;
2367 }
2368 
2369 static void htab_save_cleanup(void *opaque)
2370 {
2371     SpaprMachineState *spapr = opaque;
2372 
2373     close_htab_fd(spapr);
2374 }
2375 
2376 static SaveVMHandlers savevm_htab_handlers = {
2377     .save_setup = htab_save_setup,
2378     .save_live_iterate = htab_save_iterate,
2379     .save_live_complete_precopy = htab_save_complete,
2380     .save_cleanup = htab_save_cleanup,
2381     .load_state = htab_load,
2382 };
2383 
2384 static void spapr_boot_set(void *opaque, const char *boot_device,
2385                            Error **errp)
2386 {
2387     MachineState *machine = MACHINE(opaque);
2388     machine->boot_order = g_strdup(boot_device);
2389 }
2390 
2391 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2392 {
2393     MachineState *machine = MACHINE(spapr);
2394     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2395     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2396     int i;
2397 
2398     for (i = 0; i < nr_lmbs; i++) {
2399         uint64_t addr;
2400 
2401         addr = i * lmb_size + machine->device_memory->base;
2402         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2403                                addr / lmb_size);
2404     }
2405 }
2406 
2407 /*
2408  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2409  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2410  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2411  */
2412 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2413 {
2414     int i;
2415 
2416     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2417         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2418                    " is not aligned to %" PRIu64 " MiB",
2419                    machine->ram_size,
2420                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2421         return;
2422     }
2423 
2424     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2425         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2426                    " is not aligned to %" PRIu64 " MiB",
2427                    machine->ram_size,
2428                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2429         return;
2430     }
2431 
2432     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2433         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2434             error_setg(errp,
2435                        "Node %d memory size 0x%" PRIx64
2436                        " is not aligned to %" PRIu64 " MiB",
2437                        i, machine->numa_state->nodes[i].node_mem,
2438                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2439             return;
2440         }
2441     }
2442 }
2443 
2444 /* find cpu slot in machine->possible_cpus by core_id */
2445 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2446 {
2447     int index = id / ms->smp.threads;
2448 
2449     if (index >= ms->possible_cpus->len) {
2450         return NULL;
2451     }
2452     if (idx) {
2453         *idx = index;
2454     }
2455     return &ms->possible_cpus->cpus[index];
2456 }
2457 
2458 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2459 {
2460     MachineState *ms = MACHINE(spapr);
2461     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2462     Error *local_err = NULL;
2463     bool vsmt_user = !!spapr->vsmt;
2464     int kvm_smt = kvmppc_smt_threads();
2465     int ret;
2466     unsigned int smp_threads = ms->smp.threads;
2467 
2468     if (!kvm_enabled() && (smp_threads > 1)) {
2469         error_setg(errp, "TCG cannot support more than 1 thread/core "
2470                    "on a pseries machine");
2471         return;
2472     }
2473     if (!is_power_of_2(smp_threads)) {
2474         error_setg(errp, "Cannot support %d threads/core on a pseries "
2475                    "machine because it must be a power of 2", smp_threads);
2476         return;
2477     }
2478 
2479     /* Detemine the VSMT mode to use: */
2480     if (vsmt_user) {
2481         if (spapr->vsmt < smp_threads) {
2482             error_setg(errp, "Cannot support VSMT mode %d"
2483                        " because it must be >= threads/core (%d)",
2484                        spapr->vsmt, smp_threads);
2485             return;
2486         }
2487         /* In this case, spapr->vsmt has been set by the command line */
2488     } else if (!smc->smp_threads_vsmt) {
2489         /*
2490          * Default VSMT value is tricky, because we need it to be as
2491          * consistent as possible (for migration), but this requires
2492          * changing it for at least some existing cases.  We pick 8 as
2493          * the value that we'd get with KVM on POWER8, the
2494          * overwhelmingly common case in production systems.
2495          */
2496         spapr->vsmt = MAX(8, smp_threads);
2497     } else {
2498         spapr->vsmt = smp_threads;
2499     }
2500 
2501     /* KVM: If necessary, set the SMT mode: */
2502     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2503         ret = kvmppc_set_smt_threads(spapr->vsmt);
2504         if (ret) {
2505             /* Looks like KVM isn't able to change VSMT mode */
2506             error_setg(&local_err,
2507                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2508                        spapr->vsmt, ret);
2509             /* We can live with that if the default one is big enough
2510              * for the number of threads, and a submultiple of the one
2511              * we want.  In this case we'll waste some vcpu ids, but
2512              * behaviour will be correct */
2513             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2514                 warn_report_err(local_err);
2515             } else {
2516                 if (!vsmt_user) {
2517                     error_append_hint(&local_err,
2518                                       "On PPC, a VM with %d threads/core"
2519                                       " on a host with %d threads/core"
2520                                       " requires the use of VSMT mode %d.\n",
2521                                       smp_threads, kvm_smt, spapr->vsmt);
2522                 }
2523                 kvmppc_error_append_smt_possible_hint(&local_err);
2524                 error_propagate(errp, local_err);
2525             }
2526         }
2527     }
2528     /* else TCG: nothing to do currently */
2529 }
2530 
2531 static void spapr_init_cpus(SpaprMachineState *spapr)
2532 {
2533     MachineState *machine = MACHINE(spapr);
2534     MachineClass *mc = MACHINE_GET_CLASS(machine);
2535     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2536     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2537     const CPUArchIdList *possible_cpus;
2538     unsigned int smp_cpus = machine->smp.cpus;
2539     unsigned int smp_threads = machine->smp.threads;
2540     unsigned int max_cpus = machine->smp.max_cpus;
2541     int boot_cores_nr = smp_cpus / smp_threads;
2542     int i;
2543 
2544     possible_cpus = mc->possible_cpu_arch_ids(machine);
2545     if (mc->has_hotpluggable_cpus) {
2546         if (smp_cpus % smp_threads) {
2547             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2548                          smp_cpus, smp_threads);
2549             exit(1);
2550         }
2551         if (max_cpus % smp_threads) {
2552             error_report("max_cpus (%u) must be multiple of threads (%u)",
2553                          max_cpus, smp_threads);
2554             exit(1);
2555         }
2556     } else {
2557         if (max_cpus != smp_cpus) {
2558             error_report("This machine version does not support CPU hotplug");
2559             exit(1);
2560         }
2561         boot_cores_nr = possible_cpus->len;
2562     }
2563 
2564     if (smc->pre_2_10_has_unused_icps) {
2565         int i;
2566 
2567         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2568             /* Dummy entries get deregistered when real ICPState objects
2569              * are registered during CPU core hotplug.
2570              */
2571             pre_2_10_vmstate_register_dummy_icp(i);
2572         }
2573     }
2574 
2575     for (i = 0; i < possible_cpus->len; i++) {
2576         int core_id = i * smp_threads;
2577 
2578         if (mc->has_hotpluggable_cpus) {
2579             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2580                                    spapr_vcpu_id(spapr, core_id));
2581         }
2582 
2583         if (i < boot_cores_nr) {
2584             Object *core  = object_new(type);
2585             int nr_threads = smp_threads;
2586 
2587             /* Handle the partially filled core for older machine types */
2588             if ((i + 1) * smp_threads >= smp_cpus) {
2589                 nr_threads = smp_cpus - i * smp_threads;
2590             }
2591 
2592             object_property_set_int(core, "nr-threads", nr_threads,
2593                                     &error_fatal);
2594             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2595                                     &error_fatal);
2596             qdev_realize(DEVICE(core), NULL, &error_fatal);
2597 
2598             object_unref(core);
2599         }
2600     }
2601 }
2602 
2603 static PCIHostState *spapr_create_default_phb(void)
2604 {
2605     DeviceState *dev;
2606 
2607     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2608     qdev_prop_set_uint32(dev, "index", 0);
2609     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2610 
2611     return PCI_HOST_BRIDGE(dev);
2612 }
2613 
2614 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2615 {
2616     MachineState *machine = MACHINE(spapr);
2617     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2618     hwaddr rma_size = machine->ram_size;
2619     hwaddr node0_size = spapr_node0_size(machine);
2620 
2621     /* RMA has to fit in the first NUMA node */
2622     rma_size = MIN(rma_size, node0_size);
2623 
2624     /*
2625      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2626      * never exceed that
2627      */
2628     rma_size = MIN(rma_size, 1 * TiB);
2629 
2630     /*
2631      * Clamp the RMA size based on machine type.  This is for
2632      * migration compatibility with older qemu versions, which limited
2633      * the RMA size for complicated and mostly bad reasons.
2634      */
2635     if (smc->rma_limit) {
2636         rma_size = MIN(rma_size, smc->rma_limit);
2637     }
2638 
2639     if (rma_size < MIN_RMA_SLOF) {
2640         error_setg(errp,
2641                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2642                    "ldMiB guest RMA (Real Mode Area memory)",
2643                    MIN_RMA_SLOF / MiB);
2644         return 0;
2645     }
2646 
2647     return rma_size;
2648 }
2649 
2650 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2651 {
2652     MachineState *machine = MACHINE(spapr);
2653     int i;
2654 
2655     for (i = 0; i < machine->ram_slots; i++) {
2656         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2657     }
2658 }
2659 
2660 /* pSeries LPAR / sPAPR hardware init */
2661 static void spapr_machine_init(MachineState *machine)
2662 {
2663     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2664     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2665     MachineClass *mc = MACHINE_GET_CLASS(machine);
2666     const char *kernel_filename = machine->kernel_filename;
2667     const char *initrd_filename = machine->initrd_filename;
2668     PCIHostState *phb;
2669     int i;
2670     MemoryRegion *sysmem = get_system_memory();
2671     long load_limit, fw_size;
2672     char *filename;
2673     Error *resize_hpt_err = NULL;
2674 
2675     msi_nonbroken = true;
2676 
2677     QLIST_INIT(&spapr->phbs);
2678     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2679 
2680     /* Determine capabilities to run with */
2681     spapr_caps_init(spapr);
2682 
2683     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2684     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2685         /*
2686          * If the user explicitly requested a mode we should either
2687          * supply it, or fail completely (which we do below).  But if
2688          * it's not set explicitly, we reset our mode to something
2689          * that works
2690          */
2691         if (resize_hpt_err) {
2692             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2693             error_free(resize_hpt_err);
2694             resize_hpt_err = NULL;
2695         } else {
2696             spapr->resize_hpt = smc->resize_hpt_default;
2697         }
2698     }
2699 
2700     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2701 
2702     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2703         /*
2704          * User requested HPT resize, but this host can't supply it.  Bail out
2705          */
2706         error_report_err(resize_hpt_err);
2707         exit(1);
2708     }
2709     error_free(resize_hpt_err);
2710 
2711     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2712 
2713     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2714     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2715 
2716     /*
2717      * VSMT must be set in order to be able to compute VCPU ids, ie to
2718      * call spapr_max_server_number() or spapr_vcpu_id().
2719      */
2720     spapr_set_vsmt_mode(spapr, &error_fatal);
2721 
2722     /* Set up Interrupt Controller before we create the VCPUs */
2723     spapr_irq_init(spapr, &error_fatal);
2724 
2725     /* Set up containers for ibm,client-architecture-support negotiated options
2726      */
2727     spapr->ov5 = spapr_ovec_new();
2728     spapr->ov5_cas = spapr_ovec_new();
2729 
2730     if (smc->dr_lmb_enabled) {
2731         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2732         spapr_validate_node_memory(machine, &error_fatal);
2733     }
2734 
2735     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2736 
2737     /* advertise support for dedicated HP event source to guests */
2738     if (spapr->use_hotplug_event_source) {
2739         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2740     }
2741 
2742     /* advertise support for HPT resizing */
2743     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2744         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2745     }
2746 
2747     /* advertise support for ibm,dyamic-memory-v2 */
2748     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2749 
2750     /* advertise XIVE on POWER9 machines */
2751     if (spapr->irq->xive) {
2752         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2753     }
2754 
2755     /* init CPUs */
2756     spapr_init_cpus(spapr);
2757 
2758     /*
2759      * check we don't have a memory-less/cpu-less NUMA node
2760      * Firmware relies on the existing memory/cpu topology to provide the
2761      * NUMA topology to the kernel.
2762      * And the linux kernel needs to know the NUMA topology at start
2763      * to be able to hotplug CPUs later.
2764      */
2765     if (machine->numa_state->num_nodes) {
2766         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2767             /* check for memory-less node */
2768             if (machine->numa_state->nodes[i].node_mem == 0) {
2769                 CPUState *cs;
2770                 int found = 0;
2771                 /* check for cpu-less node */
2772                 CPU_FOREACH(cs) {
2773                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2774                     if (cpu->node_id == i) {
2775                         found = 1;
2776                         break;
2777                     }
2778                 }
2779                 /* memory-less and cpu-less node */
2780                 if (!found) {
2781                     error_report(
2782                        "Memory-less/cpu-less nodes are not supported (node %d)",
2783                                  i);
2784                     exit(1);
2785                 }
2786             }
2787         }
2788 
2789     }
2790 
2791     /*
2792      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2793      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2794      * called from vPHB reset handler so we initialize the counter here.
2795      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2796      * must be equally distant from any other node.
2797      * The final value of spapr->gpu_numa_id is going to be written to
2798      * max-associativity-domains in spapr_build_fdt().
2799      */
2800     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2801 
2802     /* Init numa_assoc_array */
2803     spapr_numa_associativity_init(spapr, machine);
2804 
2805     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2806         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2807                               spapr->max_compat_pvr)) {
2808         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2809         /* KVM and TCG always allow GTSE with radix... */
2810         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2811     }
2812     /* ... but not with hash (currently). */
2813 
2814     if (kvm_enabled()) {
2815         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2816         kvmppc_enable_logical_ci_hcalls();
2817         kvmppc_enable_set_mode_hcall();
2818 
2819         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2820         kvmppc_enable_clear_ref_mod_hcalls();
2821 
2822         /* Enable H_PAGE_INIT */
2823         kvmppc_enable_h_page_init();
2824     }
2825 
2826     /* map RAM */
2827     memory_region_add_subregion(sysmem, 0, machine->ram);
2828 
2829     /* always allocate the device memory information */
2830     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2831 
2832     /* initialize hotplug memory address space */
2833     if (machine->ram_size < machine->maxram_size) {
2834         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2835         /*
2836          * Limit the number of hotpluggable memory slots to half the number
2837          * slots that KVM supports, leaving the other half for PCI and other
2838          * devices. However ensure that number of slots doesn't drop below 32.
2839          */
2840         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2841                            SPAPR_MAX_RAM_SLOTS;
2842 
2843         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2844             max_memslots = SPAPR_MAX_RAM_SLOTS;
2845         }
2846         if (machine->ram_slots > max_memslots) {
2847             error_report("Specified number of memory slots %"
2848                          PRIu64" exceeds max supported %d",
2849                          machine->ram_slots, max_memslots);
2850             exit(1);
2851         }
2852 
2853         machine->device_memory->base = ROUND_UP(machine->ram_size,
2854                                                 SPAPR_DEVICE_MEM_ALIGN);
2855         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2856                            "device-memory", device_mem_size);
2857         memory_region_add_subregion(sysmem, machine->device_memory->base,
2858                                     &machine->device_memory->mr);
2859     }
2860 
2861     if (smc->dr_lmb_enabled) {
2862         spapr_create_lmb_dr_connectors(spapr);
2863     }
2864 
2865     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2866         /* Create the error string for live migration blocker */
2867         error_setg(&spapr->fwnmi_migration_blocker,
2868             "A machine check is being handled during migration. The handler"
2869             "may run and log hardware error on the destination");
2870     }
2871 
2872     if (mc->nvdimm_supported) {
2873         spapr_create_nvdimm_dr_connectors(spapr);
2874     }
2875 
2876     /* Set up RTAS event infrastructure */
2877     spapr_events_init(spapr);
2878 
2879     /* Set up the RTC RTAS interfaces */
2880     spapr_rtc_create(spapr);
2881 
2882     /* Set up VIO bus */
2883     spapr->vio_bus = spapr_vio_bus_init();
2884 
2885     for (i = 0; i < serial_max_hds(); i++) {
2886         if (serial_hd(i)) {
2887             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2888         }
2889     }
2890 
2891     /* We always have at least the nvram device on VIO */
2892     spapr_create_nvram(spapr);
2893 
2894     /*
2895      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2896      * connectors (described in root DT node's "ibm,drc-types" property)
2897      * are pre-initialized here. additional child connectors (such as
2898      * connectors for a PHBs PCI slots) are added as needed during their
2899      * parent's realization.
2900      */
2901     if (smc->dr_phb_enabled) {
2902         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2903             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2904         }
2905     }
2906 
2907     /* Set up PCI */
2908     spapr_pci_rtas_init();
2909 
2910     phb = spapr_create_default_phb();
2911 
2912     for (i = 0; i < nb_nics; i++) {
2913         NICInfo *nd = &nd_table[i];
2914 
2915         if (!nd->model) {
2916             nd->model = g_strdup("spapr-vlan");
2917         }
2918 
2919         if (g_str_equal(nd->model, "spapr-vlan") ||
2920             g_str_equal(nd->model, "ibmveth")) {
2921             spapr_vlan_create(spapr->vio_bus, nd);
2922         } else {
2923             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2924         }
2925     }
2926 
2927     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2928         spapr_vscsi_create(spapr->vio_bus);
2929     }
2930 
2931     /* Graphics */
2932     if (spapr_vga_init(phb->bus, &error_fatal)) {
2933         spapr->has_graphics = true;
2934         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2935     }
2936 
2937     if (machine->usb) {
2938         if (smc->use_ohci_by_default) {
2939             pci_create_simple(phb->bus, -1, "pci-ohci");
2940         } else {
2941             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2942         }
2943 
2944         if (spapr->has_graphics) {
2945             USBBus *usb_bus = usb_bus_find(-1);
2946 
2947             usb_create_simple(usb_bus, "usb-kbd");
2948             usb_create_simple(usb_bus, "usb-mouse");
2949         }
2950     }
2951 
2952     if (kernel_filename) {
2953         spapr->kernel_size = load_elf(kernel_filename, NULL,
2954                                       translate_kernel_address, spapr,
2955                                       NULL, NULL, NULL, NULL, 1,
2956                                       PPC_ELF_MACHINE, 0, 0);
2957         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2958             spapr->kernel_size = load_elf(kernel_filename, NULL,
2959                                           translate_kernel_address, spapr,
2960                                           NULL, NULL, NULL, NULL, 0,
2961                                           PPC_ELF_MACHINE, 0, 0);
2962             spapr->kernel_le = spapr->kernel_size > 0;
2963         }
2964         if (spapr->kernel_size < 0) {
2965             error_report("error loading %s: %s", kernel_filename,
2966                          load_elf_strerror(spapr->kernel_size));
2967             exit(1);
2968         }
2969 
2970         /* load initrd */
2971         if (initrd_filename) {
2972             /* Try to locate the initrd in the gap between the kernel
2973              * and the firmware. Add a bit of space just in case
2974              */
2975             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2976                                   + 0x1ffff) & ~0xffff;
2977             spapr->initrd_size = load_image_targphys(initrd_filename,
2978                                                      spapr->initrd_base,
2979                                                      load_limit
2980                                                      - spapr->initrd_base);
2981             if (spapr->initrd_size < 0) {
2982                 error_report("could not load initial ram disk '%s'",
2983                              initrd_filename);
2984                 exit(1);
2985             }
2986         }
2987     }
2988 
2989     if (bios_name == NULL) {
2990         bios_name = FW_FILE_NAME;
2991     }
2992     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2993     if (!filename) {
2994         error_report("Could not find LPAR firmware '%s'", bios_name);
2995         exit(1);
2996     }
2997     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2998     if (fw_size <= 0) {
2999         error_report("Could not load LPAR firmware '%s'", filename);
3000         exit(1);
3001     }
3002     g_free(filename);
3003 
3004     /* FIXME: Should register things through the MachineState's qdev
3005      * interface, this is a legacy from the sPAPREnvironment structure
3006      * which predated MachineState but had a similar function */
3007     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3008     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3009                          &savevm_htab_handlers, spapr);
3010 
3011     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3012 
3013     qemu_register_boot_set(spapr_boot_set, spapr);
3014 
3015     /*
3016      * Nothing needs to be done to resume a suspended guest because
3017      * suspending does not change the machine state, so no need for
3018      * a ->wakeup method.
3019      */
3020     qemu_register_wakeup_support();
3021 
3022     if (kvm_enabled()) {
3023         /* to stop and start vmclock */
3024         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3025                                          &spapr->tb);
3026 
3027         kvmppc_spapr_enable_inkernel_multitce();
3028     }
3029 
3030     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3031 }
3032 
3033 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3034 {
3035     if (!vm_type) {
3036         return 0;
3037     }
3038 
3039     if (!strcmp(vm_type, "HV")) {
3040         return 1;
3041     }
3042 
3043     if (!strcmp(vm_type, "PR")) {
3044         return 2;
3045     }
3046 
3047     error_report("Unknown kvm-type specified '%s'", vm_type);
3048     exit(1);
3049 }
3050 
3051 /*
3052  * Implementation of an interface to adjust firmware path
3053  * for the bootindex property handling.
3054  */
3055 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3056                                    DeviceState *dev)
3057 {
3058 #define CAST(type, obj, name) \
3059     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3060     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3061     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3062     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3063 
3064     if (d) {
3065         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3066         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3067         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3068 
3069         if (spapr) {
3070             /*
3071              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3072              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3073              * 0x8000 | (target << 8) | (bus << 5) | lun
3074              * (see the "Logical unit addressing format" table in SAM5)
3075              */
3076             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3077             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3078                                    (uint64_t)id << 48);
3079         } else if (virtio) {
3080             /*
3081              * We use SRP luns of the form 01000000 | (target << 8) | lun
3082              * in the top 32 bits of the 64-bit LUN
3083              * Note: the quote above is from SLOF and it is wrong,
3084              * the actual binding is:
3085              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3086              */
3087             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3088             if (d->lun >= 256) {
3089                 /* Use the LUN "flat space addressing method" */
3090                 id |= 0x4000;
3091             }
3092             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3093                                    (uint64_t)id << 32);
3094         } else if (usb) {
3095             /*
3096              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3097              * in the top 32 bits of the 64-bit LUN
3098              */
3099             unsigned usb_port = atoi(usb->port->path);
3100             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3101             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3102                                    (uint64_t)id << 32);
3103         }
3104     }
3105 
3106     /*
3107      * SLOF probes the USB devices, and if it recognizes that the device is a
3108      * storage device, it changes its name to "storage" instead of "usb-host",
3109      * and additionally adds a child node for the SCSI LUN, so the correct
3110      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3111      */
3112     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3113         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3114         if (usb_host_dev_is_scsi_storage(usbdev)) {
3115             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3116         }
3117     }
3118 
3119     if (phb) {
3120         /* Replace "pci" with "pci@800000020000000" */
3121         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3122     }
3123 
3124     if (vsc) {
3125         /* Same logic as virtio above */
3126         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3127         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3128     }
3129 
3130     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3131         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3132         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3133         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3134     }
3135 
3136     return NULL;
3137 }
3138 
3139 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3140 {
3141     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3142 
3143     return g_strdup(spapr->kvm_type);
3144 }
3145 
3146 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3147 {
3148     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3149 
3150     g_free(spapr->kvm_type);
3151     spapr->kvm_type = g_strdup(value);
3152 }
3153 
3154 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3155 {
3156     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3157 
3158     return spapr->use_hotplug_event_source;
3159 }
3160 
3161 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3162                                             Error **errp)
3163 {
3164     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3165 
3166     spapr->use_hotplug_event_source = value;
3167 }
3168 
3169 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3170 {
3171     return true;
3172 }
3173 
3174 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3175 {
3176     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3177 
3178     switch (spapr->resize_hpt) {
3179     case SPAPR_RESIZE_HPT_DEFAULT:
3180         return g_strdup("default");
3181     case SPAPR_RESIZE_HPT_DISABLED:
3182         return g_strdup("disabled");
3183     case SPAPR_RESIZE_HPT_ENABLED:
3184         return g_strdup("enabled");
3185     case SPAPR_RESIZE_HPT_REQUIRED:
3186         return g_strdup("required");
3187     }
3188     g_assert_not_reached();
3189 }
3190 
3191 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3192 {
3193     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3194 
3195     if (strcmp(value, "default") == 0) {
3196         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3197     } else if (strcmp(value, "disabled") == 0) {
3198         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3199     } else if (strcmp(value, "enabled") == 0) {
3200         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3201     } else if (strcmp(value, "required") == 0) {
3202         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3203     } else {
3204         error_setg(errp, "Bad value for \"resize-hpt\" property");
3205     }
3206 }
3207 
3208 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3209 {
3210     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3211 
3212     if (spapr->irq == &spapr_irq_xics_legacy) {
3213         return g_strdup("legacy");
3214     } else if (spapr->irq == &spapr_irq_xics) {
3215         return g_strdup("xics");
3216     } else if (spapr->irq == &spapr_irq_xive) {
3217         return g_strdup("xive");
3218     } else if (spapr->irq == &spapr_irq_dual) {
3219         return g_strdup("dual");
3220     }
3221     g_assert_not_reached();
3222 }
3223 
3224 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3225 {
3226     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3227 
3228     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3229         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3230         return;
3231     }
3232 
3233     /* The legacy IRQ backend can not be set */
3234     if (strcmp(value, "xics") == 0) {
3235         spapr->irq = &spapr_irq_xics;
3236     } else if (strcmp(value, "xive") == 0) {
3237         spapr->irq = &spapr_irq_xive;
3238     } else if (strcmp(value, "dual") == 0) {
3239         spapr->irq = &spapr_irq_dual;
3240     } else {
3241         error_setg(errp, "Bad value for \"ic-mode\" property");
3242     }
3243 }
3244 
3245 static char *spapr_get_host_model(Object *obj, Error **errp)
3246 {
3247     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3248 
3249     return g_strdup(spapr->host_model);
3250 }
3251 
3252 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3253 {
3254     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3255 
3256     g_free(spapr->host_model);
3257     spapr->host_model = g_strdup(value);
3258 }
3259 
3260 static char *spapr_get_host_serial(Object *obj, Error **errp)
3261 {
3262     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3263 
3264     return g_strdup(spapr->host_serial);
3265 }
3266 
3267 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3268 {
3269     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3270 
3271     g_free(spapr->host_serial);
3272     spapr->host_serial = g_strdup(value);
3273 }
3274 
3275 static void spapr_instance_init(Object *obj)
3276 {
3277     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3278     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3279 
3280     spapr->htab_fd = -1;
3281     spapr->use_hotplug_event_source = true;
3282     object_property_add_str(obj, "kvm-type",
3283                             spapr_get_kvm_type, spapr_set_kvm_type);
3284     object_property_set_description(obj, "kvm-type",
3285                                     "Specifies the KVM virtualization mode (HV, PR)");
3286     object_property_add_bool(obj, "modern-hotplug-events",
3287                             spapr_get_modern_hotplug_events,
3288                             spapr_set_modern_hotplug_events);
3289     object_property_set_description(obj, "modern-hotplug-events",
3290                                     "Use dedicated hotplug event mechanism in"
3291                                     " place of standard EPOW events when possible"
3292                                     " (required for memory hot-unplug support)");
3293     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3294                             "Maximum permitted CPU compatibility mode");
3295 
3296     object_property_add_str(obj, "resize-hpt",
3297                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3298     object_property_set_description(obj, "resize-hpt",
3299                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3300     object_property_add_uint32_ptr(obj, "vsmt",
3301                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3302     object_property_set_description(obj, "vsmt",
3303                                     "Virtual SMT: KVM behaves as if this were"
3304                                     " the host's SMT mode");
3305 
3306     object_property_add_bool(obj, "vfio-no-msix-emulation",
3307                              spapr_get_msix_emulation, NULL);
3308 
3309     object_property_add_uint64_ptr(obj, "kernel-addr",
3310                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3311     object_property_set_description(obj, "kernel-addr",
3312                                     stringify(KERNEL_LOAD_ADDR)
3313                                     " for -kernel is the default");
3314     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3315     /* The machine class defines the default interrupt controller mode */
3316     spapr->irq = smc->irq;
3317     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3318                             spapr_set_ic_mode);
3319     object_property_set_description(obj, "ic-mode",
3320                  "Specifies the interrupt controller mode (xics, xive, dual)");
3321 
3322     object_property_add_str(obj, "host-model",
3323         spapr_get_host_model, spapr_set_host_model);
3324     object_property_set_description(obj, "host-model",
3325         "Host model to advertise in guest device tree");
3326     object_property_add_str(obj, "host-serial",
3327         spapr_get_host_serial, spapr_set_host_serial);
3328     object_property_set_description(obj, "host-serial",
3329         "Host serial number to advertise in guest device tree");
3330 }
3331 
3332 static void spapr_machine_finalizefn(Object *obj)
3333 {
3334     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3335 
3336     g_free(spapr->kvm_type);
3337 }
3338 
3339 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3340 {
3341     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3342     PowerPCCPU *cpu = POWERPC_CPU(cs);
3343     CPUPPCState *env = &cpu->env;
3344 
3345     cpu_synchronize_state(cs);
3346     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3347     if (spapr->fwnmi_system_reset_addr != -1) {
3348         uint64_t rtas_addr, addr;
3349 
3350         /* get rtas addr from fdt */
3351         rtas_addr = spapr_get_rtas_addr();
3352         if (!rtas_addr) {
3353             qemu_system_guest_panicked(NULL);
3354             return;
3355         }
3356 
3357         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3358         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3359         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3360         env->gpr[3] = addr;
3361     }
3362     ppc_cpu_do_system_reset(cs);
3363     if (spapr->fwnmi_system_reset_addr != -1) {
3364         env->nip = spapr->fwnmi_system_reset_addr;
3365     }
3366 }
3367 
3368 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3369 {
3370     CPUState *cs;
3371 
3372     CPU_FOREACH(cs) {
3373         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3374     }
3375 }
3376 
3377 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3378                           void *fdt, int *fdt_start_offset, Error **errp)
3379 {
3380     uint64_t addr;
3381     uint32_t node;
3382 
3383     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3384     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3385                                     &error_abort);
3386     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3387                                              SPAPR_MEMORY_BLOCK_SIZE);
3388     return 0;
3389 }
3390 
3391 static bool spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3392                            bool dedicated_hp_event_source, Error **errp)
3393 {
3394     SpaprDrc *drc;
3395     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3396     int i;
3397     uint64_t addr = addr_start;
3398     bool hotplugged = spapr_drc_hotplugged(dev);
3399 
3400     for (i = 0; i < nr_lmbs; i++) {
3401         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3402                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3403         g_assert(drc);
3404 
3405         if (!spapr_drc_attach(drc, dev, errp)) {
3406             while (addr > addr_start) {
3407                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3408                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3409                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3410                 spapr_drc_detach(drc);
3411             }
3412             return false;
3413         }
3414         if (!hotplugged) {
3415             spapr_drc_reset(drc);
3416         }
3417         addr += SPAPR_MEMORY_BLOCK_SIZE;
3418     }
3419     /* send hotplug notification to the
3420      * guest only in case of hotplugged memory
3421      */
3422     if (hotplugged) {
3423         if (dedicated_hp_event_source) {
3424             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3425                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3426             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3427                                                    nr_lmbs,
3428                                                    spapr_drc_index(drc));
3429         } else {
3430             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3431                                            nr_lmbs);
3432         }
3433     }
3434     return true;
3435 }
3436 
3437 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3438                               Error **errp)
3439 {
3440     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3441     PCDIMMDevice *dimm = PC_DIMM(dev);
3442     uint64_t size, addr;
3443     int64_t slot;
3444     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3445 
3446     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3447 
3448     pc_dimm_plug(dimm, MACHINE(ms));
3449 
3450     if (!is_nvdimm) {
3451         addr = object_property_get_uint(OBJECT(dimm),
3452                                         PC_DIMM_ADDR_PROP, &error_abort);
3453         if (!spapr_add_lmbs(dev, addr, size,
3454                             spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT), errp)) {
3455             goto out_unplug;
3456         }
3457     } else {
3458         slot = object_property_get_int(OBJECT(dimm),
3459                                        PC_DIMM_SLOT_PROP, &error_abort);
3460         /* We should have valid slot number at this point */
3461         g_assert(slot >= 0);
3462         if (!spapr_add_nvdimm(dev, slot, errp)) {
3463             goto out_unplug;
3464         }
3465     }
3466 
3467     return;
3468 
3469 out_unplug:
3470     pc_dimm_unplug(dimm, MACHINE(ms));
3471 }
3472 
3473 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3474                                   Error **errp)
3475 {
3476     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3477     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3478     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3479     PCDIMMDevice *dimm = PC_DIMM(dev);
3480     Error *local_err = NULL;
3481     uint64_t size;
3482     Object *memdev;
3483     hwaddr pagesize;
3484 
3485     if (!smc->dr_lmb_enabled) {
3486         error_setg(errp, "Memory hotplug not supported for this machine");
3487         return;
3488     }
3489 
3490     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3491     if (local_err) {
3492         error_propagate(errp, local_err);
3493         return;
3494     }
3495 
3496     if (is_nvdimm) {
3497         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3498             return;
3499         }
3500     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3501         error_setg(errp, "Hotplugged memory size must be a multiple of "
3502                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3503         return;
3504     }
3505 
3506     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3507                                       &error_abort);
3508     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3509     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3510         return;
3511     }
3512 
3513     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3514 }
3515 
3516 struct SpaprDimmState {
3517     PCDIMMDevice *dimm;
3518     uint32_t nr_lmbs;
3519     QTAILQ_ENTRY(SpaprDimmState) next;
3520 };
3521 
3522 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3523                                                        PCDIMMDevice *dimm)
3524 {
3525     SpaprDimmState *dimm_state = NULL;
3526 
3527     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3528         if (dimm_state->dimm == dimm) {
3529             break;
3530         }
3531     }
3532     return dimm_state;
3533 }
3534 
3535 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3536                                                       uint32_t nr_lmbs,
3537                                                       PCDIMMDevice *dimm)
3538 {
3539     SpaprDimmState *ds = NULL;
3540 
3541     /*
3542      * If this request is for a DIMM whose removal had failed earlier
3543      * (due to guest's refusal to remove the LMBs), we would have this
3544      * dimm already in the pending_dimm_unplugs list. In that
3545      * case don't add again.
3546      */
3547     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3548     if (!ds) {
3549         ds = g_malloc0(sizeof(SpaprDimmState));
3550         ds->nr_lmbs = nr_lmbs;
3551         ds->dimm = dimm;
3552         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3553     }
3554     return ds;
3555 }
3556 
3557 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3558                                               SpaprDimmState *dimm_state)
3559 {
3560     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3561     g_free(dimm_state);
3562 }
3563 
3564 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3565                                                         PCDIMMDevice *dimm)
3566 {
3567     SpaprDrc *drc;
3568     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3569                                                   &error_abort);
3570     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3571     uint32_t avail_lmbs = 0;
3572     uint64_t addr_start, addr;
3573     int i;
3574 
3575     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3576                                           &error_abort);
3577 
3578     addr = addr_start;
3579     for (i = 0; i < nr_lmbs; i++) {
3580         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3581                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3582         g_assert(drc);
3583         if (drc->dev) {
3584             avail_lmbs++;
3585         }
3586         addr += SPAPR_MEMORY_BLOCK_SIZE;
3587     }
3588 
3589     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3590 }
3591 
3592 /* Callback to be called during DRC release. */
3593 void spapr_lmb_release(DeviceState *dev)
3594 {
3595     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3596     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3597     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3598 
3599     /* This information will get lost if a migration occurs
3600      * during the unplug process. In this case recover it. */
3601     if (ds == NULL) {
3602         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3603         g_assert(ds);
3604         /* The DRC being examined by the caller at least must be counted */
3605         g_assert(ds->nr_lmbs);
3606     }
3607 
3608     if (--ds->nr_lmbs) {
3609         return;
3610     }
3611 
3612     /*
3613      * Now that all the LMBs have been removed by the guest, call the
3614      * unplug handler chain. This can never fail.
3615      */
3616     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3617     object_unparent(OBJECT(dev));
3618 }
3619 
3620 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3621 {
3622     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3623     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3624 
3625     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3626     qdev_unrealize(dev);
3627     spapr_pending_dimm_unplugs_remove(spapr, ds);
3628 }
3629 
3630 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3631                                         DeviceState *dev, Error **errp)
3632 {
3633     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3634     PCDIMMDevice *dimm = PC_DIMM(dev);
3635     uint32_t nr_lmbs;
3636     uint64_t size, addr_start, addr;
3637     int i;
3638     SpaprDrc *drc;
3639 
3640     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3641         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3642         return;
3643     }
3644 
3645     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3646     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3647 
3648     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3649                                           &error_abort);
3650 
3651     /*
3652      * An existing pending dimm state for this DIMM means that there is an
3653      * unplug operation in progress, waiting for the spapr_lmb_release
3654      * callback to complete the job (BQL can't cover that far). In this case,
3655      * bail out to avoid detaching DRCs that were already released.
3656      */
3657     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3658         error_setg(errp, "Memory unplug already in progress for device %s",
3659                    dev->id);
3660         return;
3661     }
3662 
3663     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3664 
3665     addr = addr_start;
3666     for (i = 0; i < nr_lmbs; i++) {
3667         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3668                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3669         g_assert(drc);
3670 
3671         spapr_drc_detach(drc);
3672         addr += SPAPR_MEMORY_BLOCK_SIZE;
3673     }
3674 
3675     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3676                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3677     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3678                                               nr_lmbs, spapr_drc_index(drc));
3679 }
3680 
3681 /* Callback to be called during DRC release. */
3682 void spapr_core_release(DeviceState *dev)
3683 {
3684     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3685 
3686     /* Call the unplug handler chain. This can never fail. */
3687     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3688     object_unparent(OBJECT(dev));
3689 }
3690 
3691 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3692 {
3693     MachineState *ms = MACHINE(hotplug_dev);
3694     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3695     CPUCore *cc = CPU_CORE(dev);
3696     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3697 
3698     if (smc->pre_2_10_has_unused_icps) {
3699         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3700         int i;
3701 
3702         for (i = 0; i < cc->nr_threads; i++) {
3703             CPUState *cs = CPU(sc->threads[i]);
3704 
3705             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3706         }
3707     }
3708 
3709     assert(core_slot);
3710     core_slot->cpu = NULL;
3711     qdev_unrealize(dev);
3712 }
3713 
3714 static
3715 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3716                                Error **errp)
3717 {
3718     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3719     int index;
3720     SpaprDrc *drc;
3721     CPUCore *cc = CPU_CORE(dev);
3722 
3723     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3724         error_setg(errp, "Unable to find CPU core with core-id: %d",
3725                    cc->core_id);
3726         return;
3727     }
3728     if (index == 0) {
3729         error_setg(errp, "Boot CPU core may not be unplugged");
3730         return;
3731     }
3732 
3733     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3734                           spapr_vcpu_id(spapr, cc->core_id));
3735     g_assert(drc);
3736 
3737     if (!spapr_drc_unplug_requested(drc)) {
3738         spapr_drc_detach(drc);
3739         spapr_hotplug_req_remove_by_index(drc);
3740     }
3741 }
3742 
3743 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3744                            void *fdt, int *fdt_start_offset, Error **errp)
3745 {
3746     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3747     CPUState *cs = CPU(core->threads[0]);
3748     PowerPCCPU *cpu = POWERPC_CPU(cs);
3749     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3750     int id = spapr_get_vcpu_id(cpu);
3751     char *nodename;
3752     int offset;
3753 
3754     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3755     offset = fdt_add_subnode(fdt, 0, nodename);
3756     g_free(nodename);
3757 
3758     spapr_dt_cpu(cs, fdt, offset, spapr);
3759 
3760     *fdt_start_offset = offset;
3761     return 0;
3762 }
3763 
3764 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3765                             Error **errp)
3766 {
3767     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3768     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3769     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3770     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3771     CPUCore *cc = CPU_CORE(dev);
3772     CPUState *cs;
3773     SpaprDrc *drc;
3774     CPUArchId *core_slot;
3775     int index;
3776     bool hotplugged = spapr_drc_hotplugged(dev);
3777     int i;
3778 
3779     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3780     if (!core_slot) {
3781         error_setg(errp, "Unable to find CPU core with core-id: %d",
3782                    cc->core_id);
3783         return;
3784     }
3785     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3786                           spapr_vcpu_id(spapr, cc->core_id));
3787 
3788     g_assert(drc || !mc->has_hotpluggable_cpus);
3789 
3790     if (drc) {
3791         if (!spapr_drc_attach(drc, dev, errp)) {
3792             return;
3793         }
3794 
3795         if (hotplugged) {
3796             /*
3797              * Send hotplug notification interrupt to the guest only
3798              * in case of hotplugged CPUs.
3799              */
3800             spapr_hotplug_req_add_by_index(drc);
3801         } else {
3802             spapr_drc_reset(drc);
3803         }
3804     }
3805 
3806     core_slot->cpu = OBJECT(dev);
3807 
3808     if (smc->pre_2_10_has_unused_icps) {
3809         for (i = 0; i < cc->nr_threads; i++) {
3810             cs = CPU(core->threads[i]);
3811             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3812         }
3813     }
3814 
3815     /*
3816      * Set compatibility mode to match the boot CPU, which was either set
3817      * by the machine reset code or by CAS.
3818      */
3819     if (hotplugged) {
3820         for (i = 0; i < cc->nr_threads; i++) {
3821             if (ppc_set_compat(core->threads[i],
3822                                POWERPC_CPU(first_cpu)->compat_pvr,
3823                                errp) < 0) {
3824                 return;
3825             }
3826         }
3827     }
3828 }
3829 
3830 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3831                                 Error **errp)
3832 {
3833     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3834     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3835     CPUCore *cc = CPU_CORE(dev);
3836     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3837     const char *type = object_get_typename(OBJECT(dev));
3838     CPUArchId *core_slot;
3839     int index;
3840     unsigned int smp_threads = machine->smp.threads;
3841 
3842     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3843         error_setg(errp, "CPU hotplug not supported for this machine");
3844         return;
3845     }
3846 
3847     if (strcmp(base_core_type, type)) {
3848         error_setg(errp, "CPU core type should be %s", base_core_type);
3849         return;
3850     }
3851 
3852     if (cc->core_id % smp_threads) {
3853         error_setg(errp, "invalid core id %d", cc->core_id);
3854         return;
3855     }
3856 
3857     /*
3858      * In general we should have homogeneous threads-per-core, but old
3859      * (pre hotplug support) machine types allow the last core to have
3860      * reduced threads as a compatibility hack for when we allowed
3861      * total vcpus not a multiple of threads-per-core.
3862      */
3863     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3864         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
3865                    smp_threads);
3866         return;
3867     }
3868 
3869     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3870     if (!core_slot) {
3871         error_setg(errp, "core id %d out of range", cc->core_id);
3872         return;
3873     }
3874 
3875     if (core_slot->cpu) {
3876         error_setg(errp, "core %d already populated", cc->core_id);
3877         return;
3878     }
3879 
3880     numa_cpu_pre_plug(core_slot, dev, errp);
3881 }
3882 
3883 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3884                           void *fdt, int *fdt_start_offset, Error **errp)
3885 {
3886     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3887     int intc_phandle;
3888 
3889     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3890     if (intc_phandle <= 0) {
3891         return -1;
3892     }
3893 
3894     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3895         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3896         return -1;
3897     }
3898 
3899     /* generally SLOF creates these, for hotplug it's up to QEMU */
3900     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3901 
3902     return 0;
3903 }
3904 
3905 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3906                                Error **errp)
3907 {
3908     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3909     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3910     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3911     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3912 
3913     if (dev->hotplugged && !smc->dr_phb_enabled) {
3914         error_setg(errp, "PHB hotplug not supported for this machine");
3915         return;
3916     }
3917 
3918     if (sphb->index == (uint32_t)-1) {
3919         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3920         return;
3921     }
3922 
3923     /*
3924      * This will check that sphb->index doesn't exceed the maximum number of
3925      * PHBs for the current machine type.
3926      */
3927     smc->phb_placement(spapr, sphb->index,
3928                        &sphb->buid, &sphb->io_win_addr,
3929                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3930                        windows_supported, sphb->dma_liobn,
3931                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3932                        errp);
3933 }
3934 
3935 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3936                            Error **errp)
3937 {
3938     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3939     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3940     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3941     SpaprDrc *drc;
3942     bool hotplugged = spapr_drc_hotplugged(dev);
3943 
3944     if (!smc->dr_phb_enabled) {
3945         return;
3946     }
3947 
3948     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3949     /* hotplug hooks should check it's enabled before getting this far */
3950     assert(drc);
3951 
3952     if (!spapr_drc_attach(drc, dev, errp)) {
3953         return;
3954     }
3955 
3956     if (hotplugged) {
3957         spapr_hotplug_req_add_by_index(drc);
3958     } else {
3959         spapr_drc_reset(drc);
3960     }
3961 }
3962 
3963 void spapr_phb_release(DeviceState *dev)
3964 {
3965     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3966 
3967     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3968     object_unparent(OBJECT(dev));
3969 }
3970 
3971 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3972 {
3973     qdev_unrealize(dev);
3974 }
3975 
3976 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3977                                      DeviceState *dev, Error **errp)
3978 {
3979     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3980     SpaprDrc *drc;
3981 
3982     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3983     assert(drc);
3984 
3985     if (!spapr_drc_unplug_requested(drc)) {
3986         spapr_drc_detach(drc);
3987         spapr_hotplug_req_remove_by_index(drc);
3988     }
3989 }
3990 
3991 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3992                                  Error **errp)
3993 {
3994     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3995     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
3996 
3997     if (spapr->tpm_proxy != NULL) {
3998         error_setg(errp, "Only one TPM proxy can be specified for this machine");
3999         return;
4000     }
4001 
4002     spapr->tpm_proxy = tpm_proxy;
4003 }
4004 
4005 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4006 {
4007     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4008 
4009     qdev_unrealize(dev);
4010     object_unparent(OBJECT(dev));
4011     spapr->tpm_proxy = NULL;
4012 }
4013 
4014 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4015                                       DeviceState *dev, Error **errp)
4016 {
4017     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4018         spapr_memory_plug(hotplug_dev, dev, errp);
4019     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4020         spapr_core_plug(hotplug_dev, dev, errp);
4021     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4022         spapr_phb_plug(hotplug_dev, dev, errp);
4023     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4024         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4025     }
4026 }
4027 
4028 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4029                                         DeviceState *dev, Error **errp)
4030 {
4031     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4032         spapr_memory_unplug(hotplug_dev, dev);
4033     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4034         spapr_core_unplug(hotplug_dev, dev);
4035     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4036         spapr_phb_unplug(hotplug_dev, dev);
4037     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4038         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4039     }
4040 }
4041 
4042 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4043                                                 DeviceState *dev, Error **errp)
4044 {
4045     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4046     MachineClass *mc = MACHINE_GET_CLASS(sms);
4047     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4048 
4049     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4050         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4051             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4052         } else {
4053             /* NOTE: this means there is a window after guest reset, prior to
4054              * CAS negotiation, where unplug requests will fail due to the
4055              * capability not being detected yet. This is a bit different than
4056              * the case with PCI unplug, where the events will be queued and
4057              * eventually handled by the guest after boot
4058              */
4059             error_setg(errp, "Memory hot unplug not supported for this guest");
4060         }
4061     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4062         if (!mc->has_hotpluggable_cpus) {
4063             error_setg(errp, "CPU hot unplug not supported on this machine");
4064             return;
4065         }
4066         spapr_core_unplug_request(hotplug_dev, dev, errp);
4067     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4068         if (!smc->dr_phb_enabled) {
4069             error_setg(errp, "PHB hot unplug not supported on this machine");
4070             return;
4071         }
4072         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4073     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4074         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4075     }
4076 }
4077 
4078 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4079                                           DeviceState *dev, Error **errp)
4080 {
4081     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4082         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4083     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4084         spapr_core_pre_plug(hotplug_dev, dev, errp);
4085     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4086         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4087     }
4088 }
4089 
4090 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4091                                                  DeviceState *dev)
4092 {
4093     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4094         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4095         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4096         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4097         return HOTPLUG_HANDLER(machine);
4098     }
4099     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4100         PCIDevice *pcidev = PCI_DEVICE(dev);
4101         PCIBus *root = pci_device_root_bus(pcidev);
4102         SpaprPhbState *phb =
4103             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4104                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4105 
4106         if (phb) {
4107             return HOTPLUG_HANDLER(phb);
4108         }
4109     }
4110     return NULL;
4111 }
4112 
4113 static CpuInstanceProperties
4114 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4115 {
4116     CPUArchId *core_slot;
4117     MachineClass *mc = MACHINE_GET_CLASS(machine);
4118 
4119     /* make sure possible_cpu are intialized */
4120     mc->possible_cpu_arch_ids(machine);
4121     /* get CPU core slot containing thread that matches cpu_index */
4122     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4123     assert(core_slot);
4124     return core_slot->props;
4125 }
4126 
4127 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4128 {
4129     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4130 }
4131 
4132 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4133 {
4134     int i;
4135     unsigned int smp_threads = machine->smp.threads;
4136     unsigned int smp_cpus = machine->smp.cpus;
4137     const char *core_type;
4138     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4139     MachineClass *mc = MACHINE_GET_CLASS(machine);
4140 
4141     if (!mc->has_hotpluggable_cpus) {
4142         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4143     }
4144     if (machine->possible_cpus) {
4145         assert(machine->possible_cpus->len == spapr_max_cores);
4146         return machine->possible_cpus;
4147     }
4148 
4149     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4150     if (!core_type) {
4151         error_report("Unable to find sPAPR CPU Core definition");
4152         exit(1);
4153     }
4154 
4155     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4156                              sizeof(CPUArchId) * spapr_max_cores);
4157     machine->possible_cpus->len = spapr_max_cores;
4158     for (i = 0; i < machine->possible_cpus->len; i++) {
4159         int core_id = i * smp_threads;
4160 
4161         machine->possible_cpus->cpus[i].type = core_type;
4162         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4163         machine->possible_cpus->cpus[i].arch_id = core_id;
4164         machine->possible_cpus->cpus[i].props.has_core_id = true;
4165         machine->possible_cpus->cpus[i].props.core_id = core_id;
4166     }
4167     return machine->possible_cpus;
4168 }
4169 
4170 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4171                                 uint64_t *buid, hwaddr *pio,
4172                                 hwaddr *mmio32, hwaddr *mmio64,
4173                                 unsigned n_dma, uint32_t *liobns,
4174                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4175 {
4176     /*
4177      * New-style PHB window placement.
4178      *
4179      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4180      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4181      * windows.
4182      *
4183      * Some guest kernels can't work with MMIO windows above 1<<46
4184      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4185      *
4186      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4187      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4188      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4189      * 1TiB 64-bit MMIO windows for each PHB.
4190      */
4191     const uint64_t base_buid = 0x800000020000000ULL;
4192     int i;
4193 
4194     /* Sanity check natural alignments */
4195     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4196     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4197     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4198     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4199     /* Sanity check bounds */
4200     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4201                       SPAPR_PCI_MEM32_WIN_SIZE);
4202     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4203                       SPAPR_PCI_MEM64_WIN_SIZE);
4204 
4205     if (index >= SPAPR_MAX_PHBS) {
4206         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4207                    SPAPR_MAX_PHBS - 1);
4208         return;
4209     }
4210 
4211     *buid = base_buid + index;
4212     for (i = 0; i < n_dma; ++i) {
4213         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4214     }
4215 
4216     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4217     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4218     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4219 
4220     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4221     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4222 }
4223 
4224 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4225 {
4226     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4227 
4228     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4229 }
4230 
4231 static void spapr_ics_resend(XICSFabric *dev)
4232 {
4233     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4234 
4235     ics_resend(spapr->ics);
4236 }
4237 
4238 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4239 {
4240     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4241 
4242     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4243 }
4244 
4245 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4246                                  Monitor *mon)
4247 {
4248     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4249 
4250     spapr_irq_print_info(spapr, mon);
4251     monitor_printf(mon, "irqchip: %s\n",
4252                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4253 }
4254 
4255 /*
4256  * This is a XIVE only operation
4257  */
4258 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4259                            uint8_t nvt_blk, uint32_t nvt_idx,
4260                            bool cam_ignore, uint8_t priority,
4261                            uint32_t logic_serv, XiveTCTXMatch *match)
4262 {
4263     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4264     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4265     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4266     int count;
4267 
4268     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4269                            priority, logic_serv, match);
4270     if (count < 0) {
4271         return count;
4272     }
4273 
4274     /*
4275      * When we implement the save and restore of the thread interrupt
4276      * contexts in the enter/exit CPU handlers of the machine and the
4277      * escalations in QEMU, we should be able to handle non dispatched
4278      * vCPUs.
4279      *
4280      * Until this is done, the sPAPR machine should find at least one
4281      * matching context always.
4282      */
4283     if (count == 0) {
4284         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4285                       nvt_blk, nvt_idx);
4286     }
4287 
4288     return count;
4289 }
4290 
4291 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4292 {
4293     return cpu->vcpu_id;
4294 }
4295 
4296 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4297 {
4298     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4299     MachineState *ms = MACHINE(spapr);
4300     int vcpu_id;
4301 
4302     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4303 
4304     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4305         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4306         error_append_hint(errp, "Adjust the number of cpus to %d "
4307                           "or try to raise the number of threads per core\n",
4308                           vcpu_id * ms->smp.threads / spapr->vsmt);
4309         return false;
4310     }
4311 
4312     cpu->vcpu_id = vcpu_id;
4313     return true;
4314 }
4315 
4316 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4317 {
4318     CPUState *cs;
4319 
4320     CPU_FOREACH(cs) {
4321         PowerPCCPU *cpu = POWERPC_CPU(cs);
4322 
4323         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4324             return cpu;
4325         }
4326     }
4327 
4328     return NULL;
4329 }
4330 
4331 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4332 {
4333     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4334 
4335     /* These are only called by TCG, KVM maintains dispatch state */
4336 
4337     spapr_cpu->prod = false;
4338     if (spapr_cpu->vpa_addr) {
4339         CPUState *cs = CPU(cpu);
4340         uint32_t dispatch;
4341 
4342         dispatch = ldl_be_phys(cs->as,
4343                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4344         dispatch++;
4345         if ((dispatch & 1) != 0) {
4346             qemu_log_mask(LOG_GUEST_ERROR,
4347                           "VPA: incorrect dispatch counter value for "
4348                           "dispatched partition %u, correcting.\n", dispatch);
4349             dispatch++;
4350         }
4351         stl_be_phys(cs->as,
4352                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4353     }
4354 }
4355 
4356 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4357 {
4358     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4359 
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) != 1) {
4368             qemu_log_mask(LOG_GUEST_ERROR,
4369                           "VPA: incorrect dispatch counter value for "
4370                           "preempted 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_machine_class_init(ObjectClass *oc, void *data)
4379 {
4380     MachineClass *mc = MACHINE_CLASS(oc);
4381     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4382     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4383     NMIClass *nc = NMI_CLASS(oc);
4384     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4385     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4386     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4387     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4388     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4389 
4390     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4391     mc->ignore_boot_device_suffixes = true;
4392 
4393     /*
4394      * We set up the default / latest behaviour here.  The class_init
4395      * functions for the specific versioned machine types can override
4396      * these details for backwards compatibility
4397      */
4398     mc->init = spapr_machine_init;
4399     mc->reset = spapr_machine_reset;
4400     mc->block_default_type = IF_SCSI;
4401     mc->max_cpus = 1024;
4402     mc->no_parallel = 1;
4403     mc->default_boot_order = "";
4404     mc->default_ram_size = 512 * MiB;
4405     mc->default_ram_id = "ppc_spapr.ram";
4406     mc->default_display = "std";
4407     mc->kvm_type = spapr_kvm_type;
4408     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4409     mc->pci_allow_0_address = true;
4410     assert(!mc->get_hotplug_handler);
4411     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4412     hc->pre_plug = spapr_machine_device_pre_plug;
4413     hc->plug = spapr_machine_device_plug;
4414     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4415     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4416     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4417     hc->unplug_request = spapr_machine_device_unplug_request;
4418     hc->unplug = spapr_machine_device_unplug;
4419 
4420     smc->dr_lmb_enabled = true;
4421     smc->update_dt_enabled = true;
4422     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4423     mc->has_hotpluggable_cpus = true;
4424     mc->nvdimm_supported = true;
4425     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4426     fwc->get_dev_path = spapr_get_fw_dev_path;
4427     nc->nmi_monitor_handler = spapr_nmi;
4428     smc->phb_placement = spapr_phb_placement;
4429     vhc->hypercall = emulate_spapr_hypercall;
4430     vhc->hpt_mask = spapr_hpt_mask;
4431     vhc->map_hptes = spapr_map_hptes;
4432     vhc->unmap_hptes = spapr_unmap_hptes;
4433     vhc->hpte_set_c = spapr_hpte_set_c;
4434     vhc->hpte_set_r = spapr_hpte_set_r;
4435     vhc->get_pate = spapr_get_pate;
4436     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4437     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4438     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4439     xic->ics_get = spapr_ics_get;
4440     xic->ics_resend = spapr_ics_resend;
4441     xic->icp_get = spapr_icp_get;
4442     ispc->print_info = spapr_pic_print_info;
4443     /* Force NUMA node memory size to be a multiple of
4444      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4445      * in which LMBs are represented and hot-added
4446      */
4447     mc->numa_mem_align_shift = 28;
4448     mc->auto_enable_numa = true;
4449 
4450     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4451     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4452     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4453     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4454     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4455     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4456     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4457     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4458     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4459     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4460     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4461     spapr_caps_add_properties(smc);
4462     smc->irq = &spapr_irq_dual;
4463     smc->dr_phb_enabled = true;
4464     smc->linux_pci_probe = true;
4465     smc->smp_threads_vsmt = true;
4466     smc->nr_xirqs = SPAPR_NR_XIRQS;
4467     xfc->match_nvt = spapr_match_nvt;
4468 }
4469 
4470 static const TypeInfo spapr_machine_info = {
4471     .name          = TYPE_SPAPR_MACHINE,
4472     .parent        = TYPE_MACHINE,
4473     .abstract      = true,
4474     .instance_size = sizeof(SpaprMachineState),
4475     .instance_init = spapr_instance_init,
4476     .instance_finalize = spapr_machine_finalizefn,
4477     .class_size    = sizeof(SpaprMachineClass),
4478     .class_init    = spapr_machine_class_init,
4479     .interfaces = (InterfaceInfo[]) {
4480         { TYPE_FW_PATH_PROVIDER },
4481         { TYPE_NMI },
4482         { TYPE_HOTPLUG_HANDLER },
4483         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4484         { TYPE_XICS_FABRIC },
4485         { TYPE_INTERRUPT_STATS_PROVIDER },
4486         { TYPE_XIVE_FABRIC },
4487         { }
4488     },
4489 };
4490 
4491 static void spapr_machine_latest_class_options(MachineClass *mc)
4492 {
4493     mc->alias = "pseries";
4494     mc->is_default = true;
4495 }
4496 
4497 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4498     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4499                                                     void *data)      \
4500     {                                                                \
4501         MachineClass *mc = MACHINE_CLASS(oc);                        \
4502         spapr_machine_##suffix##_class_options(mc);                  \
4503         if (latest) {                                                \
4504             spapr_machine_latest_class_options(mc);                  \
4505         }                                                            \
4506     }                                                                \
4507     static const TypeInfo spapr_machine_##suffix##_info = {          \
4508         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4509         .parent = TYPE_SPAPR_MACHINE,                                \
4510         .class_init = spapr_machine_##suffix##_class_init,           \
4511     };                                                               \
4512     static void spapr_machine_register_##suffix(void)                \
4513     {                                                                \
4514         type_register(&spapr_machine_##suffix##_info);               \
4515     }                                                                \
4516     type_init(spapr_machine_register_##suffix)
4517 
4518 /*
4519  * pseries-5.2
4520  */
4521 static void spapr_machine_5_2_class_options(MachineClass *mc)
4522 {
4523     /* Defaults for the latest behaviour inherited from the base class */
4524 }
4525 
4526 DEFINE_SPAPR_MACHINE(5_2, "5.2", true);
4527 
4528 /*
4529  * pseries-5.1
4530  */
4531 static void spapr_machine_5_1_class_options(MachineClass *mc)
4532 {
4533     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4534 
4535     spapr_machine_5_2_class_options(mc);
4536     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4537     smc->pre_5_2_numa_associativity = true;
4538 }
4539 
4540 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4541 
4542 /*
4543  * pseries-5.0
4544  */
4545 static void spapr_machine_5_0_class_options(MachineClass *mc)
4546 {
4547     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4548     static GlobalProperty compat[] = {
4549         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4550     };
4551 
4552     spapr_machine_5_1_class_options(mc);
4553     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4554     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4555     mc->numa_mem_supported = true;
4556     smc->pre_5_1_assoc_refpoints = true;
4557 }
4558 
4559 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4560 
4561 /*
4562  * pseries-4.2
4563  */
4564 static void spapr_machine_4_2_class_options(MachineClass *mc)
4565 {
4566     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4567 
4568     spapr_machine_5_0_class_options(mc);
4569     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4570     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4571     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4572     smc->rma_limit = 16 * GiB;
4573     mc->nvdimm_supported = false;
4574 }
4575 
4576 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4577 
4578 /*
4579  * pseries-4.1
4580  */
4581 static void spapr_machine_4_1_class_options(MachineClass *mc)
4582 {
4583     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4584     static GlobalProperty compat[] = {
4585         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4586         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4587     };
4588 
4589     spapr_machine_4_2_class_options(mc);
4590     smc->linux_pci_probe = false;
4591     smc->smp_threads_vsmt = false;
4592     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4593     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4594 }
4595 
4596 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4597 
4598 /*
4599  * pseries-4.0
4600  */
4601 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4602                               uint64_t *buid, hwaddr *pio,
4603                               hwaddr *mmio32, hwaddr *mmio64,
4604                               unsigned n_dma, uint32_t *liobns,
4605                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4606 {
4607     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4608                         nv2gpa, nv2atsd, errp);
4609     *nv2gpa = 0;
4610     *nv2atsd = 0;
4611 }
4612 
4613 static void spapr_machine_4_0_class_options(MachineClass *mc)
4614 {
4615     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4616 
4617     spapr_machine_4_1_class_options(mc);
4618     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4619     smc->phb_placement = phb_placement_4_0;
4620     smc->irq = &spapr_irq_xics;
4621     smc->pre_4_1_migration = true;
4622 }
4623 
4624 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4625 
4626 /*
4627  * pseries-3.1
4628  */
4629 static void spapr_machine_3_1_class_options(MachineClass *mc)
4630 {
4631     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4632 
4633     spapr_machine_4_0_class_options(mc);
4634     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4635 
4636     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4637     smc->update_dt_enabled = false;
4638     smc->dr_phb_enabled = false;
4639     smc->broken_host_serial_model = true;
4640     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4641     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4642     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4643     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4644 }
4645 
4646 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4647 
4648 /*
4649  * pseries-3.0
4650  */
4651 
4652 static void spapr_machine_3_0_class_options(MachineClass *mc)
4653 {
4654     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4655 
4656     spapr_machine_3_1_class_options(mc);
4657     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4658 
4659     smc->legacy_irq_allocation = true;
4660     smc->nr_xirqs = 0x400;
4661     smc->irq = &spapr_irq_xics_legacy;
4662 }
4663 
4664 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4665 
4666 /*
4667  * pseries-2.12
4668  */
4669 static void spapr_machine_2_12_class_options(MachineClass *mc)
4670 {
4671     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4672     static GlobalProperty compat[] = {
4673         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4674         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4675     };
4676 
4677     spapr_machine_3_0_class_options(mc);
4678     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4679     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4680 
4681     /* We depend on kvm_enabled() to choose a default value for the
4682      * hpt-max-page-size capability. Of course we can't do it here
4683      * because this is too early and the HW accelerator isn't initialzed
4684      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4685      */
4686     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4687 }
4688 
4689 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4690 
4691 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4692 {
4693     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4694 
4695     spapr_machine_2_12_class_options(mc);
4696     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4697     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4698     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4699 }
4700 
4701 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4702 
4703 /*
4704  * pseries-2.11
4705  */
4706 
4707 static void spapr_machine_2_11_class_options(MachineClass *mc)
4708 {
4709     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4710 
4711     spapr_machine_2_12_class_options(mc);
4712     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4713     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4714 }
4715 
4716 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4717 
4718 /*
4719  * pseries-2.10
4720  */
4721 
4722 static void spapr_machine_2_10_class_options(MachineClass *mc)
4723 {
4724     spapr_machine_2_11_class_options(mc);
4725     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4726 }
4727 
4728 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4729 
4730 /*
4731  * pseries-2.9
4732  */
4733 
4734 static void spapr_machine_2_9_class_options(MachineClass *mc)
4735 {
4736     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4737     static GlobalProperty compat[] = {
4738         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4739     };
4740 
4741     spapr_machine_2_10_class_options(mc);
4742     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4743     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4744     smc->pre_2_10_has_unused_icps = true;
4745     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4746 }
4747 
4748 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4749 
4750 /*
4751  * pseries-2.8
4752  */
4753 
4754 static void spapr_machine_2_8_class_options(MachineClass *mc)
4755 {
4756     static GlobalProperty compat[] = {
4757         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4758     };
4759 
4760     spapr_machine_2_9_class_options(mc);
4761     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4762     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4763     mc->numa_mem_align_shift = 23;
4764 }
4765 
4766 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4767 
4768 /*
4769  * pseries-2.7
4770  */
4771 
4772 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4773                               uint64_t *buid, hwaddr *pio,
4774                               hwaddr *mmio32, hwaddr *mmio64,
4775                               unsigned n_dma, uint32_t *liobns,
4776                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4777 {
4778     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4779     const uint64_t base_buid = 0x800000020000000ULL;
4780     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4781     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4782     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4783     const uint32_t max_index = 255;
4784     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4785 
4786     uint64_t ram_top = MACHINE(spapr)->ram_size;
4787     hwaddr phb0_base, phb_base;
4788     int i;
4789 
4790     /* Do we have device memory? */
4791     if (MACHINE(spapr)->maxram_size > ram_top) {
4792         /* Can't just use maxram_size, because there may be an
4793          * alignment gap between normal and device memory regions
4794          */
4795         ram_top = MACHINE(spapr)->device_memory->base +
4796             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4797     }
4798 
4799     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4800 
4801     if (index > max_index) {
4802         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4803                    max_index);
4804         return;
4805     }
4806 
4807     *buid = base_buid + index;
4808     for (i = 0; i < n_dma; ++i) {
4809         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4810     }
4811 
4812     phb_base = phb0_base + index * phb_spacing;
4813     *pio = phb_base + pio_offset;
4814     *mmio32 = phb_base + mmio_offset;
4815     /*
4816      * We don't set the 64-bit MMIO window, relying on the PHB's
4817      * fallback behaviour of automatically splitting a large "32-bit"
4818      * window into contiguous 32-bit and 64-bit windows
4819      */
4820 
4821     *nv2gpa = 0;
4822     *nv2atsd = 0;
4823 }
4824 
4825 static void spapr_machine_2_7_class_options(MachineClass *mc)
4826 {
4827     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4828     static GlobalProperty compat[] = {
4829         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4830         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4831         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4832         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4833     };
4834 
4835     spapr_machine_2_8_class_options(mc);
4836     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4837     mc->default_machine_opts = "modern-hotplug-events=off";
4838     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4839     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4840     smc->phb_placement = phb_placement_2_7;
4841 }
4842 
4843 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4844 
4845 /*
4846  * pseries-2.6
4847  */
4848 
4849 static void spapr_machine_2_6_class_options(MachineClass *mc)
4850 {
4851     static GlobalProperty compat[] = {
4852         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4853     };
4854 
4855     spapr_machine_2_7_class_options(mc);
4856     mc->has_hotpluggable_cpus = false;
4857     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4858     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4859 }
4860 
4861 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4862 
4863 /*
4864  * pseries-2.5
4865  */
4866 
4867 static void spapr_machine_2_5_class_options(MachineClass *mc)
4868 {
4869     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4870     static GlobalProperty compat[] = {
4871         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4872     };
4873 
4874     spapr_machine_2_6_class_options(mc);
4875     smc->use_ohci_by_default = true;
4876     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4877     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4878 }
4879 
4880 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4881 
4882 /*
4883  * pseries-2.4
4884  */
4885 
4886 static void spapr_machine_2_4_class_options(MachineClass *mc)
4887 {
4888     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4889 
4890     spapr_machine_2_5_class_options(mc);
4891     smc->dr_lmb_enabled = false;
4892     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4893 }
4894 
4895 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4896 
4897 /*
4898  * pseries-2.3
4899  */
4900 
4901 static void spapr_machine_2_3_class_options(MachineClass *mc)
4902 {
4903     static GlobalProperty compat[] = {
4904         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4905     };
4906     spapr_machine_2_4_class_options(mc);
4907     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4908     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4909 }
4910 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4911 
4912 /*
4913  * pseries-2.2
4914  */
4915 
4916 static void spapr_machine_2_2_class_options(MachineClass *mc)
4917 {
4918     static GlobalProperty compat[] = {
4919         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4920     };
4921 
4922     spapr_machine_2_3_class_options(mc);
4923     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4924     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4925     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4926 }
4927 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4928 
4929 /*
4930  * pseries-2.1
4931  */
4932 
4933 static void spapr_machine_2_1_class_options(MachineClass *mc)
4934 {
4935     spapr_machine_2_2_class_options(mc);
4936     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4937 }
4938 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4939 
4940 static void spapr_machine_register_types(void)
4941 {
4942     type_register_static(&spapr_machine_info);
4943 }
4944 
4945 type_init(spapr_machine_register_types)
4946