xref: /openbmc/qemu/hw/ppc/spapr.c (revision f76b348e)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 
85 #include "monitor/monitor.h"
86 
87 #include <libfdt.h>
88 
89 /* SLOF memory layout:
90  *
91  * SLOF raw image loaded at 0, copies its romfs right below the flat
92  * device-tree, then position SLOF itself 31M below that
93  *
94  * So we set FW_OVERHEAD to 40MB which should account for all of that
95  * and more
96  *
97  * We load our kernel at 4M, leaving space for SLOF initial image
98  */
99 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
100 #define FW_MAX_SIZE             0x400000
101 #define FW_FILE_NAME            "slof.bin"
102 #define FW_OVERHEAD             0x2800000
103 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
104 
105 #define MIN_RMA_SLOF            (128 * MiB)
106 
107 #define PHANDLE_INTC            0x00001111
108 
109 /* These two functions implement the VCPU id numbering: one to compute them
110  * all and one to identify thread 0 of a VCORE. Any change to the first one
111  * is likely to have an impact on the second one, so let's keep them close.
112  */
113 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
114 {
115     MachineState *ms = MACHINE(spapr);
116     unsigned int smp_threads = ms->smp.threads;
117 
118     assert(spapr->vsmt);
119     return
120         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
121 }
122 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
123                                       PowerPCCPU *cpu)
124 {
125     assert(spapr->vsmt);
126     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 int spapr_max_server_number(SpaprMachineState *spapr)
164 {
165     MachineState *ms = MACHINE(spapr);
166 
167     assert(spapr->vsmt);
168     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
169 }
170 
171 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
172                                   int smt_threads)
173 {
174     int i, ret = 0;
175     uint32_t servers_prop[smt_threads];
176     uint32_t gservers_prop[smt_threads * 2];
177     int index = spapr_get_vcpu_id(cpu);
178 
179     if (cpu->compat_pvr) {
180         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
181         if (ret < 0) {
182             return ret;
183         }
184     }
185 
186     /* Build interrupt servers and gservers properties */
187     for (i = 0; i < smt_threads; i++) {
188         servers_prop[i] = cpu_to_be32(index + i);
189         /* Hack, direct the group queues back to cpu 0 */
190         gservers_prop[i*2] = cpu_to_be32(index + i);
191         gservers_prop[i*2 + 1] = 0;
192     }
193     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
194                       servers_prop, sizeof(servers_prop));
195     if (ret < 0) {
196         return ret;
197     }
198     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
199                       gservers_prop, sizeof(gservers_prop));
200 
201     return ret;
202 }
203 
204 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
205 {
206     int index = spapr_get_vcpu_id(cpu);
207     uint32_t associativity[] = {cpu_to_be32(0x5),
208                                 cpu_to_be32(0x0),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(cpu->node_id),
212                                 cpu_to_be32(index)};
213 
214     /* Advertise NUMA via ibm,associativity */
215     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
216                           sizeof(associativity));
217 }
218 
219 static void spapr_dt_pa_features(SpaprMachineState *spapr,
220                                  PowerPCCPU *cpu,
221                                  void *fdt, int offset)
222 {
223     uint8_t pa_features_206[] = { 6, 0,
224         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
225     uint8_t pa_features_207[] = { 24, 0,
226         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
227         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
228         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
229         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
230     uint8_t pa_features_300[] = { 66, 0,
231         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
232         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
233         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
234         /* 6: DS207 */
235         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
236         /* 16: Vector */
237         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
238         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
239         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
240         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
242         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
243         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
244         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
245         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
246         /* 42: PM, 44: PC RA, 46: SC vec'd */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
248         /* 48: SIMD, 50: QP BFP, 52: String */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
250         /* 54: DecFP, 56: DecI, 58: SHA */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
252         /* 60: NM atomic, 62: RNG */
253         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
254     };
255     uint8_t *pa_features = NULL;
256     size_t pa_size;
257 
258     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
259         pa_features = pa_features_206;
260         pa_size = sizeof(pa_features_206);
261     }
262     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
263         pa_features = pa_features_207;
264         pa_size = sizeof(pa_features_207);
265     }
266     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
267         pa_features = pa_features_300;
268         pa_size = sizeof(pa_features_300);
269     }
270     if (!pa_features) {
271         return;
272     }
273 
274     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
275         /*
276          * Note: we keep CI large pages off by default because a 64K capable
277          * guest provisioned with large pages might otherwise try to map a qemu
278          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
279          * even if that qemu runs on a 4k host.
280          * We dd this bit back here if we are confident this is not an issue
281          */
282         pa_features[3] |= 0x20;
283     }
284     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
285         pa_features[24] |= 0x80;    /* Transactional memory support */
286     }
287     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
288         /* Workaround for broken kernels that attempt (guest) radix
289          * mode when they can't handle it, if they see the radix bit set
290          * in pa-features. So hide it from them. */
291         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
292     }
293 
294     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
295 }
296 
297 static hwaddr spapr_node0_size(MachineState *machine)
298 {
299     if (machine->numa_state->num_nodes) {
300         int i;
301         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
302             if (machine->numa_state->nodes[i].node_mem) {
303                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
304                            machine->ram_size);
305             }
306         }
307     }
308     return machine->ram_size;
309 }
310 
311 static void add_str(GString *s, const gchar *s1)
312 {
313     g_string_append_len(s, s1, strlen(s1) + 1);
314 }
315 
316 static int spapr_dt_memory_node(void *fdt, int nodeid, hwaddr start,
317                                 hwaddr size)
318 {
319     uint32_t associativity[] = {
320         cpu_to_be32(0x4), /* length */
321         cpu_to_be32(0x0), cpu_to_be32(0x0),
322         cpu_to_be32(0x0), cpu_to_be32(nodeid)
323     };
324     char mem_name[32];
325     uint64_t mem_reg_property[2];
326     int off;
327 
328     mem_reg_property[0] = cpu_to_be64(start);
329     mem_reg_property[1] = cpu_to_be64(size);
330 
331     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
332     off = fdt_add_subnode(fdt, 0, mem_name);
333     _FDT(off);
334     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
335     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
336                       sizeof(mem_reg_property))));
337     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
338                       sizeof(associativity))));
339     return off;
340 }
341 
342 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
343 {
344     MemoryDeviceInfoList *info;
345 
346     for (info = list; info; info = info->next) {
347         MemoryDeviceInfo *value = info->value;
348 
349         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
350             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
351 
352             if (addr >= pcdimm_info->addr &&
353                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
354                 return pcdimm_info->node;
355             }
356         }
357     }
358 
359     return -1;
360 }
361 
362 struct sPAPRDrconfCellV2 {
363      uint32_t seq_lmbs;
364      uint64_t base_addr;
365      uint32_t drc_index;
366      uint32_t aa_index;
367      uint32_t flags;
368 } QEMU_PACKED;
369 
370 typedef struct DrconfCellQueue {
371     struct sPAPRDrconfCellV2 cell;
372     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
373 } DrconfCellQueue;
374 
375 static DrconfCellQueue *
376 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
377                       uint32_t drc_index, uint32_t aa_index,
378                       uint32_t flags)
379 {
380     DrconfCellQueue *elem;
381 
382     elem = g_malloc0(sizeof(*elem));
383     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
384     elem->cell.base_addr = cpu_to_be64(base_addr);
385     elem->cell.drc_index = cpu_to_be32(drc_index);
386     elem->cell.aa_index = cpu_to_be32(aa_index);
387     elem->cell.flags = cpu_to_be32(flags);
388 
389     return elem;
390 }
391 
392 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
393                                       int offset, MemoryDeviceInfoList *dimms)
394 {
395     MachineState *machine = MACHINE(spapr);
396     uint8_t *int_buf, *cur_index;
397     int ret;
398     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
399     uint64_t addr, cur_addr, size;
400     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
401     uint64_t mem_end = machine->device_memory->base +
402                        memory_region_size(&machine->device_memory->mr);
403     uint32_t node, buf_len, nr_entries = 0;
404     SpaprDrc *drc;
405     DrconfCellQueue *elem, *next;
406     MemoryDeviceInfoList *info;
407     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
408         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
409 
410     /* Entry to cover RAM and the gap area */
411     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
412                                  SPAPR_LMB_FLAGS_RESERVED |
413                                  SPAPR_LMB_FLAGS_DRC_INVALID);
414     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
415     nr_entries++;
416 
417     cur_addr = machine->device_memory->base;
418     for (info = dimms; info; info = info->next) {
419         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
420 
421         addr = di->addr;
422         size = di->size;
423         node = di->node;
424 
425         /*
426          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
427          * area is marked hotpluggable in the next iteration for the bigger
428          * chunk including the NVDIMM occupied area.
429          */
430         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
431             continue;
432 
433         /* Entry for hot-pluggable area */
434         if (cur_addr < addr) {
435             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
436             g_assert(drc);
437             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
438                                          cur_addr, spapr_drc_index(drc), -1, 0);
439             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
440             nr_entries++;
441         }
442 
443         /* Entry for DIMM */
444         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
445         g_assert(drc);
446         elem = spapr_get_drconf_cell(size / lmb_size, addr,
447                                      spapr_drc_index(drc), node,
448                                      (SPAPR_LMB_FLAGS_ASSIGNED |
449                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
450         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
451         nr_entries++;
452         cur_addr = addr + size;
453     }
454 
455     /* Entry for remaining hotpluggable area */
456     if (cur_addr < mem_end) {
457         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
458         g_assert(drc);
459         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
460                                      cur_addr, spapr_drc_index(drc), -1, 0);
461         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
462         nr_entries++;
463     }
464 
465     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
466     int_buf = cur_index = g_malloc0(buf_len);
467     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
468     cur_index += sizeof(nr_entries);
469 
470     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
471         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
472         cur_index += sizeof(elem->cell);
473         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
474         g_free(elem);
475     }
476 
477     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
478     g_free(int_buf);
479     if (ret < 0) {
480         return -1;
481     }
482     return 0;
483 }
484 
485 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
486                                    int offset, MemoryDeviceInfoList *dimms)
487 {
488     MachineState *machine = MACHINE(spapr);
489     int i, ret;
490     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
491     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
492     uint32_t nr_lmbs = (machine->device_memory->base +
493                        memory_region_size(&machine->device_memory->mr)) /
494                        lmb_size;
495     uint32_t *int_buf, *cur_index, buf_len;
496 
497     /*
498      * Allocate enough buffer size to fit in ibm,dynamic-memory
499      */
500     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
501     cur_index = int_buf = g_malloc0(buf_len);
502     int_buf[0] = cpu_to_be32(nr_lmbs);
503     cur_index++;
504     for (i = 0; i < nr_lmbs; i++) {
505         uint64_t addr = i * lmb_size;
506         uint32_t *dynamic_memory = cur_index;
507 
508         if (i >= device_lmb_start) {
509             SpaprDrc *drc;
510 
511             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
512             g_assert(drc);
513 
514             dynamic_memory[0] = cpu_to_be32(addr >> 32);
515             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
516             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
517             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
518             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
519             if (memory_region_present(get_system_memory(), addr)) {
520                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
521             } else {
522                 dynamic_memory[5] = cpu_to_be32(0);
523             }
524         } else {
525             /*
526              * LMB information for RMA, boot time RAM and gap b/n RAM and
527              * device memory region -- all these are marked as reserved
528              * and as having no valid DRC.
529              */
530             dynamic_memory[0] = cpu_to_be32(addr >> 32);
531             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
532             dynamic_memory[2] = cpu_to_be32(0);
533             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
534             dynamic_memory[4] = cpu_to_be32(-1);
535             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
536                                             SPAPR_LMB_FLAGS_DRC_INVALID);
537         }
538 
539         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
540     }
541     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
542     g_free(int_buf);
543     if (ret < 0) {
544         return -1;
545     }
546     return 0;
547 }
548 
549 /*
550  * Adds ibm,dynamic-reconfiguration-memory node.
551  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
552  * of this device tree node.
553  */
554 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
555                                                    void *fdt)
556 {
557     MachineState *machine = MACHINE(spapr);
558     int nb_numa_nodes = machine->numa_state->num_nodes;
559     int ret, i, offset;
560     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
561     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
562     uint32_t *int_buf, *cur_index, buf_len;
563     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
564     MemoryDeviceInfoList *dimms = NULL;
565 
566     /*
567      * Don't create the node if there is no device memory
568      */
569     if (machine->ram_size == machine->maxram_size) {
570         return 0;
571     }
572 
573     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
574 
575     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
576                     sizeof(prop_lmb_size));
577     if (ret < 0) {
578         return ret;
579     }
580 
581     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
582     if (ret < 0) {
583         return ret;
584     }
585 
586     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
587     if (ret < 0) {
588         return ret;
589     }
590 
591     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
592     dimms = qmp_memory_device_list();
593     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
594         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
595     } else {
596         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
597     }
598     qapi_free_MemoryDeviceInfoList(dimms);
599 
600     if (ret < 0) {
601         return ret;
602     }
603 
604     /* ibm,associativity-lookup-arrays */
605     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
606     cur_index = int_buf = g_malloc0(buf_len);
607     int_buf[0] = cpu_to_be32(nr_nodes);
608     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
609     cur_index += 2;
610     for (i = 0; i < nr_nodes; i++) {
611         uint32_t associativity[] = {
612             cpu_to_be32(0x0),
613             cpu_to_be32(0x0),
614             cpu_to_be32(0x0),
615             cpu_to_be32(i)
616         };
617         memcpy(cur_index, associativity, sizeof(associativity));
618         cur_index += 4;
619     }
620     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
621             (cur_index - int_buf) * sizeof(uint32_t));
622     g_free(int_buf);
623 
624     return ret;
625 }
626 
627 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
628 {
629     MachineState *machine = MACHINE(spapr);
630     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
631     hwaddr mem_start, node_size;
632     int i, nb_nodes = machine->numa_state->num_nodes;
633     NodeInfo *nodes = machine->numa_state->nodes;
634 
635     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
636         if (!nodes[i].node_mem) {
637             continue;
638         }
639         if (mem_start >= machine->ram_size) {
640             node_size = 0;
641         } else {
642             node_size = nodes[i].node_mem;
643             if (node_size > machine->ram_size - mem_start) {
644                 node_size = machine->ram_size - mem_start;
645             }
646         }
647         if (!mem_start) {
648             /* spapr_machine_init() checks for rma_size <= node0_size
649              * already */
650             spapr_dt_memory_node(fdt, i, 0, spapr->rma_size);
651             mem_start += spapr->rma_size;
652             node_size -= spapr->rma_size;
653         }
654         for ( ; node_size; ) {
655             hwaddr sizetmp = pow2floor(node_size);
656 
657             /* mem_start != 0 here */
658             if (ctzl(mem_start) < ctzl(sizetmp)) {
659                 sizetmp = 1ULL << ctzl(mem_start);
660             }
661 
662             spapr_dt_memory_node(fdt, i, mem_start, sizetmp);
663             node_size -= sizetmp;
664             mem_start += sizetmp;
665         }
666     }
667 
668     /* Generate ibm,dynamic-reconfiguration-memory node if required */
669     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
670         int ret;
671 
672         g_assert(smc->dr_lmb_enabled);
673         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
674         if (ret) {
675             return ret;
676         }
677     }
678 
679     return 0;
680 }
681 
682 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
683                          SpaprMachineState *spapr)
684 {
685     MachineState *ms = MACHINE(spapr);
686     PowerPCCPU *cpu = POWERPC_CPU(cs);
687     CPUPPCState *env = &cpu->env;
688     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
689     int index = spapr_get_vcpu_id(cpu);
690     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
691                        0xffffffff, 0xffffffff};
692     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
693         : SPAPR_TIMEBASE_FREQ;
694     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
695     uint32_t page_sizes_prop[64];
696     size_t page_sizes_prop_size;
697     unsigned int smp_threads = ms->smp.threads;
698     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
699     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
700     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
701     SpaprDrc *drc;
702     int drc_index;
703     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
704     int i;
705 
706     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
707     if (drc) {
708         drc_index = spapr_drc_index(drc);
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
710     }
711 
712     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
713     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
714 
715     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
716     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
717                            env->dcache_line_size)));
718     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
719                            env->dcache_line_size)));
720     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
721                            env->icache_line_size)));
722     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
723                            env->icache_line_size)));
724 
725     if (pcc->l1_dcache_size) {
726         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
727                                pcc->l1_dcache_size)));
728     } else {
729         warn_report("Unknown L1 dcache size for cpu");
730     }
731     if (pcc->l1_icache_size) {
732         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
733                                pcc->l1_icache_size)));
734     } else {
735         warn_report("Unknown L1 icache size for cpu");
736     }
737 
738     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
739     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
740     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
741     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
742     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
743     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
744 
745     if (env->spr_cb[SPR_PURR].oea_read) {
746         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
747     }
748     if (env->spr_cb[SPR_SPURR].oea_read) {
749         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
750     }
751 
752     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
753         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
754                           segs, sizeof(segs))));
755     }
756 
757     /* Advertise VSX (vector extensions) if available
758      *   1               == VMX / Altivec available
759      *   2               == VSX available
760      *
761      * Only CPUs for which we create core types in spapr_cpu_core.c
762      * are possible, and all of those have VMX */
763     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
764         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
765     } else {
766         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
767     }
768 
769     /* Advertise DFP (Decimal Floating Point) if available
770      *   0 / no property == no DFP
771      *   1               == DFP available */
772     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
773         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
774     }
775 
776     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
777                                                       sizeof(page_sizes_prop));
778     if (page_sizes_prop_size) {
779         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
780                           page_sizes_prop, page_sizes_prop_size)));
781     }
782 
783     spapr_dt_pa_features(spapr, cpu, fdt, offset);
784 
785     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
786                            cs->cpu_index / vcpus_per_socket)));
787 
788     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
789                       pft_size_prop, sizeof(pft_size_prop))));
790 
791     if (ms->numa_state->num_nodes > 1) {
792         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
793     }
794 
795     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
796 
797     if (pcc->radix_page_info) {
798         for (i = 0; i < pcc->radix_page_info->count; i++) {
799             radix_AP_encodings[i] =
800                 cpu_to_be32(pcc->radix_page_info->entries[i]);
801         }
802         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
803                           radix_AP_encodings,
804                           pcc->radix_page_info->count *
805                           sizeof(radix_AP_encodings[0]))));
806     }
807 
808     /*
809      * We set this property to let the guest know that it can use the large
810      * decrementer and its width in bits.
811      */
812     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
813         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
814                               pcc->lrg_decr_bits)));
815 }
816 
817 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
818 {
819     CPUState **rev;
820     CPUState *cs;
821     int n_cpus;
822     int cpus_offset;
823     char *nodename;
824     int i;
825 
826     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
827     _FDT(cpus_offset);
828     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
829     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
830 
831     /*
832      * We walk the CPUs in reverse order to ensure that CPU DT nodes
833      * created by fdt_add_subnode() end up in the right order in FDT
834      * for the guest kernel the enumerate the CPUs correctly.
835      *
836      * The CPU list cannot be traversed in reverse order, so we need
837      * to do extra work.
838      */
839     n_cpus = 0;
840     rev = NULL;
841     CPU_FOREACH(cs) {
842         rev = g_renew(CPUState *, rev, n_cpus + 1);
843         rev[n_cpus++] = cs;
844     }
845 
846     for (i = n_cpus - 1; i >= 0; i--) {
847         CPUState *cs = rev[i];
848         PowerPCCPU *cpu = POWERPC_CPU(cs);
849         int index = spapr_get_vcpu_id(cpu);
850         DeviceClass *dc = DEVICE_GET_CLASS(cs);
851         int offset;
852 
853         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
854             continue;
855         }
856 
857         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
858         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
859         g_free(nodename);
860         _FDT(offset);
861         spapr_dt_cpu(cs, fdt, offset, spapr);
862     }
863 
864     g_free(rev);
865 }
866 
867 static int spapr_dt_rng(void *fdt)
868 {
869     int node;
870     int ret;
871 
872     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
873     if (node <= 0) {
874         return -1;
875     }
876     ret = fdt_setprop_string(fdt, node, "device_type",
877                              "ibm,platform-facilities");
878     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
879     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
880 
881     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
882     if (node <= 0) {
883         return -1;
884     }
885     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
886 
887     return ret ? -1 : 0;
888 }
889 
890 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
891 {
892     MachineState *ms = MACHINE(spapr);
893     int rtas;
894     GString *hypertas = g_string_sized_new(256);
895     GString *qemu_hypertas = g_string_sized_new(256);
896     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
897     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
898         memory_region_size(&MACHINE(spapr)->device_memory->mr);
899     uint32_t lrdr_capacity[] = {
900         cpu_to_be32(max_device_addr >> 32),
901         cpu_to_be32(max_device_addr & 0xffffffff),
902         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
903         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
904     };
905     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
906     uint32_t maxdomains[] = {
907         cpu_to_be32(4),
908         maxdomain,
909         maxdomain,
910         maxdomain,
911         cpu_to_be32(spapr->gpu_numa_id),
912     };
913 
914     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
915 
916     /* hypertas */
917     add_str(hypertas, "hcall-pft");
918     add_str(hypertas, "hcall-term");
919     add_str(hypertas, "hcall-dabr");
920     add_str(hypertas, "hcall-interrupt");
921     add_str(hypertas, "hcall-tce");
922     add_str(hypertas, "hcall-vio");
923     add_str(hypertas, "hcall-splpar");
924     add_str(hypertas, "hcall-join");
925     add_str(hypertas, "hcall-bulk");
926     add_str(hypertas, "hcall-set-mode");
927     add_str(hypertas, "hcall-sprg0");
928     add_str(hypertas, "hcall-copy");
929     add_str(hypertas, "hcall-debug");
930     add_str(hypertas, "hcall-vphn");
931     add_str(qemu_hypertas, "hcall-memop1");
932 
933     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
934         add_str(hypertas, "hcall-multi-tce");
935     }
936 
937     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
938         add_str(hypertas, "hcall-hpt-resize");
939     }
940 
941     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
942                      hypertas->str, hypertas->len));
943     g_string_free(hypertas, TRUE);
944     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
945                      qemu_hypertas->str, qemu_hypertas->len));
946     g_string_free(qemu_hypertas, TRUE);
947 
948     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
949                      refpoints, sizeof(refpoints)));
950 
951     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
952                      maxdomains, sizeof(maxdomains)));
953 
954     /*
955      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
956      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
957      *
958      * The system reset requirements are driven by existing Linux and PowerVM
959      * implementation which (contrary to PAPR) saves r3 in the error log
960      * structure like machine check, so Linux expects to find the saved r3
961      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
962      * does not look at the error value).
963      *
964      * System reset interrupts are not subject to interlock like machine
965      * check, so this memory area could be corrupted if the sreset is
966      * interrupted by a machine check (or vice versa) if it was shared. To
967      * prevent this, system reset uses per-CPU areas for the sreset save
968      * area. A system reset that interrupts a system reset handler could
969      * still overwrite this area, but Linux doesn't try to recover in that
970      * case anyway.
971      *
972      * The extra 8 bytes is required because Linux's FWNMI error log check
973      * is off-by-one.
974      */
975     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_ERROR_LOG_MAX +
976 			  ms->smp.max_cpus * sizeof(uint64_t)*2 + sizeof(uint64_t)));
977     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
978                           RTAS_ERROR_LOG_MAX));
979     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
980                           RTAS_EVENT_SCAN_RATE));
981 
982     g_assert(msi_nonbroken);
983     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
984 
985     /*
986      * According to PAPR, rtas ibm,os-term does not guarantee a return
987      * back to the guest cpu.
988      *
989      * While an additional ibm,extended-os-term property indicates
990      * that rtas call return will always occur. Set this property.
991      */
992     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
993 
994     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
995                      lrdr_capacity, sizeof(lrdr_capacity)));
996 
997     spapr_dt_rtas_tokens(fdt, rtas);
998 }
999 
1000 /*
1001  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1002  * and the XIVE features that the guest may request and thus the valid
1003  * values for bytes 23..26 of option vector 5:
1004  */
1005 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1006                                           int chosen)
1007 {
1008     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1009 
1010     char val[2 * 4] = {
1011         23, 0x00, /* XICS / XIVE mode */
1012         24, 0x00, /* Hash/Radix, filled in below. */
1013         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1014         26, 0x40, /* Radix options: GTSE == yes. */
1015     };
1016 
1017     if (spapr->irq->xics && spapr->irq->xive) {
1018         val[1] = SPAPR_OV5_XIVE_BOTH;
1019     } else if (spapr->irq->xive) {
1020         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1021     } else {
1022         assert(spapr->irq->xics);
1023         val[1] = SPAPR_OV5_XIVE_LEGACY;
1024     }
1025 
1026     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1027                           first_ppc_cpu->compat_pvr)) {
1028         /*
1029          * If we're in a pre POWER9 compat mode then the guest should
1030          * do hash and use the legacy interrupt mode
1031          */
1032         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1033         val[3] = 0x00; /* Hash */
1034     } else if (kvm_enabled()) {
1035         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1036             val[3] = 0x80; /* OV5_MMU_BOTH */
1037         } else if (kvmppc_has_cap_mmu_radix()) {
1038             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1039         } else {
1040             val[3] = 0x00; /* Hash */
1041         }
1042     } else {
1043         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1044         val[3] = 0xC0;
1045     }
1046     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1047                      val, sizeof(val)));
1048 }
1049 
1050 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1051 {
1052     MachineState *machine = MACHINE(spapr);
1053     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1054     int chosen;
1055 
1056     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1057 
1058     if (reset) {
1059         const char *boot_device = machine->boot_order;
1060         char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1061         size_t cb = 0;
1062         char *bootlist = get_boot_devices_list(&cb);
1063 
1064         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1065             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1066                                     machine->kernel_cmdline));
1067         }
1068 
1069         if (spapr->initrd_size) {
1070             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1071                                   spapr->initrd_base));
1072             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1073                                   spapr->initrd_base + spapr->initrd_size));
1074         }
1075 
1076         if (spapr->kernel_size) {
1077             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1078                                   cpu_to_be64(spapr->kernel_size) };
1079 
1080             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1081                          &kprop, sizeof(kprop)));
1082             if (spapr->kernel_le) {
1083                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1084             }
1085         }
1086         if (boot_menu) {
1087             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1088         }
1089         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1090         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1091         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1092 
1093         if (cb && bootlist) {
1094             int i;
1095 
1096             for (i = 0; i < cb; i++) {
1097                 if (bootlist[i] == '\n') {
1098                     bootlist[i] = ' ';
1099                 }
1100             }
1101             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1102         }
1103 
1104         if (boot_device && strlen(boot_device)) {
1105             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1106         }
1107 
1108         if (!spapr->has_graphics && stdout_path) {
1109             /*
1110              * "linux,stdout-path" and "stdout" properties are
1111              * deprecated by linux kernel. New platforms should only
1112              * use the "stdout-path" property. Set the new property
1113              * and continue using older property to remain compatible
1114              * with the existing firmware.
1115              */
1116             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1117             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1118         }
1119 
1120         /*
1121          * We can deal with BAR reallocation just fine, advertise it
1122          * to the guest
1123          */
1124         if (smc->linux_pci_probe) {
1125             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1126         }
1127 
1128         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1129 
1130         g_free(stdout_path);
1131         g_free(bootlist);
1132     }
1133 
1134     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1135 }
1136 
1137 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1138 {
1139     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1140      * KVM to work under pHyp with some guest co-operation */
1141     int hypervisor;
1142     uint8_t hypercall[16];
1143 
1144     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1145     /* indicate KVM hypercall interface */
1146     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1147     if (kvmppc_has_cap_fixup_hcalls()) {
1148         /*
1149          * Older KVM versions with older guest kernels were broken
1150          * with the magic page, don't allow the guest to map it.
1151          */
1152         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1153                                   sizeof(hypercall))) {
1154             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1155                              hypercall, sizeof(hypercall)));
1156         }
1157     }
1158 }
1159 
1160 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1161 {
1162     MachineState *machine = MACHINE(spapr);
1163     MachineClass *mc = MACHINE_GET_CLASS(machine);
1164     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1165     int ret;
1166     void *fdt;
1167     SpaprPhbState *phb;
1168     char *buf;
1169 
1170     fdt = g_malloc0(space);
1171     _FDT((fdt_create_empty_tree(fdt, space)));
1172 
1173     /* Root node */
1174     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1175     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1176     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1177 
1178     /* Guest UUID & Name*/
1179     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1180     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1181     if (qemu_uuid_set) {
1182         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1183     }
1184     g_free(buf);
1185 
1186     if (qemu_get_vm_name()) {
1187         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1188                                 qemu_get_vm_name()));
1189     }
1190 
1191     /* Host Model & Serial Number */
1192     if (spapr->host_model) {
1193         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1194     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1195         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1196         g_free(buf);
1197     }
1198 
1199     if (spapr->host_serial) {
1200         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1201     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1202         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1203         g_free(buf);
1204     }
1205 
1206     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1207     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1208 
1209     /* /interrupt controller */
1210     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1211 
1212     ret = spapr_dt_memory(spapr, fdt);
1213     if (ret < 0) {
1214         error_report("couldn't setup memory nodes in fdt");
1215         exit(1);
1216     }
1217 
1218     /* /vdevice */
1219     spapr_dt_vdevice(spapr->vio_bus, fdt);
1220 
1221     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1222         ret = spapr_dt_rng(fdt);
1223         if (ret < 0) {
1224             error_report("could not set up rng device in the fdt");
1225             exit(1);
1226         }
1227     }
1228 
1229     QLIST_FOREACH(phb, &spapr->phbs, list) {
1230         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1231         if (ret < 0) {
1232             error_report("couldn't setup PCI devices in fdt");
1233             exit(1);
1234         }
1235     }
1236 
1237     spapr_dt_cpus(fdt, spapr);
1238 
1239     if (smc->dr_lmb_enabled) {
1240         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1241     }
1242 
1243     if (mc->has_hotpluggable_cpus) {
1244         int offset = fdt_path_offset(fdt, "/cpus");
1245         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1246         if (ret < 0) {
1247             error_report("Couldn't set up CPU DR device tree properties");
1248             exit(1);
1249         }
1250     }
1251 
1252     /* /event-sources */
1253     spapr_dt_events(spapr, fdt);
1254 
1255     /* /rtas */
1256     spapr_dt_rtas(spapr, fdt);
1257 
1258     /* /chosen */
1259     spapr_dt_chosen(spapr, fdt, reset);
1260 
1261     /* /hypervisor */
1262     if (kvm_enabled()) {
1263         spapr_dt_hypervisor(spapr, fdt);
1264     }
1265 
1266     /* Build memory reserve map */
1267     if (reset) {
1268         if (spapr->kernel_size) {
1269             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1270                                   spapr->kernel_size)));
1271         }
1272         if (spapr->initrd_size) {
1273             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1274                                   spapr->initrd_size)));
1275         }
1276     }
1277 
1278     if (smc->dr_phb_enabled) {
1279         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1280         if (ret < 0) {
1281             error_report("Couldn't set up PHB DR device tree properties");
1282             exit(1);
1283         }
1284     }
1285 
1286     /* NVDIMM devices */
1287     if (mc->nvdimm_supported) {
1288         spapr_dt_persistent_memory(fdt);
1289     }
1290 
1291     return fdt;
1292 }
1293 
1294 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1295 {
1296     SpaprMachineState *spapr = opaque;
1297 
1298     return (addr & 0x0fffffff) + spapr->kernel_addr;
1299 }
1300 
1301 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1302                                     PowerPCCPU *cpu)
1303 {
1304     CPUPPCState *env = &cpu->env;
1305 
1306     /* The TCG path should also be holding the BQL at this point */
1307     g_assert(qemu_mutex_iothread_locked());
1308 
1309     if (msr_pr) {
1310         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1311         env->gpr[3] = H_PRIVILEGE;
1312     } else {
1313         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1314     }
1315 }
1316 
1317 struct LPCRSyncState {
1318     target_ulong value;
1319     target_ulong mask;
1320 };
1321 
1322 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1323 {
1324     struct LPCRSyncState *s = arg.host_ptr;
1325     PowerPCCPU *cpu = POWERPC_CPU(cs);
1326     CPUPPCState *env = &cpu->env;
1327     target_ulong lpcr;
1328 
1329     cpu_synchronize_state(cs);
1330     lpcr = env->spr[SPR_LPCR];
1331     lpcr &= ~s->mask;
1332     lpcr |= s->value;
1333     ppc_store_lpcr(cpu, lpcr);
1334 }
1335 
1336 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1337 {
1338     CPUState *cs;
1339     struct LPCRSyncState s = {
1340         .value = value,
1341         .mask = mask
1342     };
1343     CPU_FOREACH(cs) {
1344         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1345     }
1346 }
1347 
1348 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1349 {
1350     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1351 
1352     /* Copy PATE1:GR into PATE0:HR */
1353     entry->dw0 = spapr->patb_entry & PATE0_HR;
1354     entry->dw1 = spapr->patb_entry;
1355 }
1356 
1357 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1358 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1359 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1360 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1361 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1362 
1363 /*
1364  * Get the fd to access the kernel htab, re-opening it if necessary
1365  */
1366 static int get_htab_fd(SpaprMachineState *spapr)
1367 {
1368     Error *local_err = NULL;
1369 
1370     if (spapr->htab_fd >= 0) {
1371         return spapr->htab_fd;
1372     }
1373 
1374     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1375     if (spapr->htab_fd < 0) {
1376         error_report_err(local_err);
1377     }
1378 
1379     return spapr->htab_fd;
1380 }
1381 
1382 void close_htab_fd(SpaprMachineState *spapr)
1383 {
1384     if (spapr->htab_fd >= 0) {
1385         close(spapr->htab_fd);
1386     }
1387     spapr->htab_fd = -1;
1388 }
1389 
1390 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1391 {
1392     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1393 
1394     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1395 }
1396 
1397 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1398 {
1399     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1400 
1401     assert(kvm_enabled());
1402 
1403     if (!spapr->htab) {
1404         return 0;
1405     }
1406 
1407     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1408 }
1409 
1410 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1411                                                 hwaddr ptex, int n)
1412 {
1413     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1414     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1415 
1416     if (!spapr->htab) {
1417         /*
1418          * HTAB is controlled by KVM. Fetch into temporary buffer
1419          */
1420         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1421         kvmppc_read_hptes(hptes, ptex, n);
1422         return hptes;
1423     }
1424 
1425     /*
1426      * HTAB is controlled by QEMU. Just point to the internally
1427      * accessible PTEG.
1428      */
1429     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1430 }
1431 
1432 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1433                               const ppc_hash_pte64_t *hptes,
1434                               hwaddr ptex, int n)
1435 {
1436     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1437 
1438     if (!spapr->htab) {
1439         g_free((void *)hptes);
1440     }
1441 
1442     /* Nothing to do for qemu managed HPT */
1443 }
1444 
1445 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1446                       uint64_t pte0, uint64_t pte1)
1447 {
1448     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1449     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1450 
1451     if (!spapr->htab) {
1452         kvmppc_write_hpte(ptex, pte0, pte1);
1453     } else {
1454         if (pte0 & HPTE64_V_VALID) {
1455             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1456             /*
1457              * When setting valid, we write PTE1 first. This ensures
1458              * proper synchronization with the reading code in
1459              * ppc_hash64_pteg_search()
1460              */
1461             smp_wmb();
1462             stq_p(spapr->htab + offset, pte0);
1463         } else {
1464             stq_p(spapr->htab + offset, pte0);
1465             /*
1466              * When clearing it we set PTE0 first. This ensures proper
1467              * synchronization with the reading code in
1468              * ppc_hash64_pteg_search()
1469              */
1470             smp_wmb();
1471             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1472         }
1473     }
1474 }
1475 
1476 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1477                              uint64_t pte1)
1478 {
1479     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1480     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1481 
1482     if (!spapr->htab) {
1483         /* There should always be a hash table when this is called */
1484         error_report("spapr_hpte_set_c called with no hash table !");
1485         return;
1486     }
1487 
1488     /* The HW performs a non-atomic byte update */
1489     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1490 }
1491 
1492 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1493                              uint64_t pte1)
1494 {
1495     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1496     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1497 
1498     if (!spapr->htab) {
1499         /* There should always be a hash table when this is called */
1500         error_report("spapr_hpte_set_r called with no hash table !");
1501         return;
1502     }
1503 
1504     /* The HW performs a non-atomic byte update */
1505     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1506 }
1507 
1508 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1509 {
1510     int shift;
1511 
1512     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1513      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1514      * that's much more than is needed for Linux guests */
1515     shift = ctz64(pow2ceil(ramsize)) - 7;
1516     shift = MAX(shift, 18); /* Minimum architected size */
1517     shift = MIN(shift, 46); /* Maximum architected size */
1518     return shift;
1519 }
1520 
1521 void spapr_free_hpt(SpaprMachineState *spapr)
1522 {
1523     g_free(spapr->htab);
1524     spapr->htab = NULL;
1525     spapr->htab_shift = 0;
1526     close_htab_fd(spapr);
1527 }
1528 
1529 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1530                           Error **errp)
1531 {
1532     long rc;
1533 
1534     /* Clean up any HPT info from a previous boot */
1535     spapr_free_hpt(spapr);
1536 
1537     rc = kvmppc_reset_htab(shift);
1538     if (rc < 0) {
1539         /* kernel-side HPT needed, but couldn't allocate one */
1540         error_setg_errno(errp, errno,
1541                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1542                          shift);
1543         /* This is almost certainly fatal, but if the caller really
1544          * wants to carry on with shift == 0, it's welcome to try */
1545     } else if (rc > 0) {
1546         /* kernel-side HPT allocated */
1547         if (rc != shift) {
1548             error_setg(errp,
1549                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1550                        shift, rc);
1551         }
1552 
1553         spapr->htab_shift = shift;
1554         spapr->htab = NULL;
1555     } else {
1556         /* kernel-side HPT not needed, allocate in userspace instead */
1557         size_t size = 1ULL << shift;
1558         int i;
1559 
1560         spapr->htab = qemu_memalign(size, size);
1561         if (!spapr->htab) {
1562             error_setg_errno(errp, errno,
1563                              "Could not allocate HPT of order %d", shift);
1564             return;
1565         }
1566 
1567         memset(spapr->htab, 0, size);
1568         spapr->htab_shift = shift;
1569 
1570         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1571             DIRTY_HPTE(HPTE(spapr->htab, i));
1572         }
1573     }
1574     /* We're setting up a hash table, so that means we're not radix */
1575     spapr->patb_entry = 0;
1576     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1577 }
1578 
1579 void spapr_setup_hpt(SpaprMachineState *spapr)
1580 {
1581     int hpt_shift;
1582 
1583     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1584         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1585     } else {
1586         uint64_t current_ram_size;
1587 
1588         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1589         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1590     }
1591     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1592 
1593     if (kvm_enabled()) {
1594         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1595 
1596         /* Check our RMA fits in the possible VRMA */
1597         if (vrma_limit < spapr->rma_size) {
1598             error_report("Unable to create %" HWADDR_PRIu
1599                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1600                          spapr->rma_size / MiB, vrma_limit / MiB);
1601             exit(EXIT_FAILURE);
1602         }
1603     }
1604 }
1605 
1606 static int spapr_reset_drcs(Object *child, void *opaque)
1607 {
1608     SpaprDrc *drc =
1609         (SpaprDrc *) object_dynamic_cast(child,
1610                                                  TYPE_SPAPR_DR_CONNECTOR);
1611 
1612     if (drc) {
1613         spapr_drc_reset(drc);
1614     }
1615 
1616     return 0;
1617 }
1618 
1619 static void spapr_machine_reset(MachineState *machine)
1620 {
1621     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1622     PowerPCCPU *first_ppc_cpu;
1623     hwaddr fdt_addr;
1624     void *fdt;
1625     int rc;
1626 
1627     kvmppc_svm_off(&error_fatal);
1628     spapr_caps_apply(spapr);
1629 
1630     first_ppc_cpu = POWERPC_CPU(first_cpu);
1631     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1632         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1633                               spapr->max_compat_pvr)) {
1634         /*
1635          * If using KVM with radix mode available, VCPUs can be started
1636          * without a HPT because KVM will start them in radix mode.
1637          * Set the GR bit in PATE so that we know there is no HPT.
1638          */
1639         spapr->patb_entry = PATE1_GR;
1640         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1641     } else {
1642         spapr_setup_hpt(spapr);
1643     }
1644 
1645     qemu_devices_reset();
1646 
1647     spapr_ovec_cleanup(spapr->ov5_cas);
1648     spapr->ov5_cas = spapr_ovec_new();
1649 
1650     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1651 
1652     /*
1653      * This is fixing some of the default configuration of the XIVE
1654      * devices. To be called after the reset of the machine devices.
1655      */
1656     spapr_irq_reset(spapr, &error_fatal);
1657 
1658     /*
1659      * There is no CAS under qtest. Simulate one to please the code that
1660      * depends on spapr->ov5_cas. This is especially needed to test device
1661      * unplug, so we do that before resetting the DRCs.
1662      */
1663     if (qtest_enabled()) {
1664         spapr_ovec_cleanup(spapr->ov5_cas);
1665         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1666     }
1667 
1668     /* DRC reset may cause a device to be unplugged. This will cause troubles
1669      * if this device is used by another device (eg, a running vhost backend
1670      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1671      * situations, we reset DRCs after all devices have been reset.
1672      */
1673     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1674 
1675     spapr_clear_pending_events(spapr);
1676 
1677     /*
1678      * We place the device tree and RTAS just below either the top of the RMA,
1679      * or just below 2GB, whichever is lower, so that it can be
1680      * processed with 32-bit real mode code if necessary
1681      */
1682     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1683 
1684     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1685 
1686     rc = fdt_pack(fdt);
1687 
1688     /* Should only fail if we've built a corrupted tree */
1689     assert(rc == 0);
1690 
1691     /* Load the fdt */
1692     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1693     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1694     g_free(spapr->fdt_blob);
1695     spapr->fdt_size = fdt_totalsize(fdt);
1696     spapr->fdt_initial_size = spapr->fdt_size;
1697     spapr->fdt_blob = fdt;
1698 
1699     /* Set up the entry state */
1700     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1701     first_ppc_cpu->env.gpr[5] = 0;
1702 
1703     spapr->fwnmi_system_reset_addr = -1;
1704     spapr->fwnmi_machine_check_addr = -1;
1705     spapr->fwnmi_machine_check_interlock = -1;
1706 
1707     /* Signal all vCPUs waiting on this condition */
1708     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1709 
1710     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1711 }
1712 
1713 static void spapr_create_nvram(SpaprMachineState *spapr)
1714 {
1715     DeviceState *dev = qdev_new("spapr-nvram");
1716     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1717 
1718     if (dinfo) {
1719         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1720                             &error_fatal);
1721     }
1722 
1723     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1724 
1725     spapr->nvram = (struct SpaprNvram *)dev;
1726 }
1727 
1728 static void spapr_rtc_create(SpaprMachineState *spapr)
1729 {
1730     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1731                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1732                                        &error_fatal, NULL);
1733     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1734     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1735                               "date");
1736 }
1737 
1738 /* Returns whether we want to use VGA or not */
1739 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1740 {
1741     switch (vga_interface_type) {
1742     case VGA_NONE:
1743         return false;
1744     case VGA_DEVICE:
1745         return true;
1746     case VGA_STD:
1747     case VGA_VIRTIO:
1748     case VGA_CIRRUS:
1749         return pci_vga_init(pci_bus) != NULL;
1750     default:
1751         error_setg(errp,
1752                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1753         return false;
1754     }
1755 }
1756 
1757 static int spapr_pre_load(void *opaque)
1758 {
1759     int rc;
1760 
1761     rc = spapr_caps_pre_load(opaque);
1762     if (rc) {
1763         return rc;
1764     }
1765 
1766     return 0;
1767 }
1768 
1769 static int spapr_post_load(void *opaque, int version_id)
1770 {
1771     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1772     int err = 0;
1773 
1774     err = spapr_caps_post_migration(spapr);
1775     if (err) {
1776         return err;
1777     }
1778 
1779     /*
1780      * In earlier versions, there was no separate qdev for the PAPR
1781      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1782      * So when migrating from those versions, poke the incoming offset
1783      * value into the RTC device
1784      */
1785     if (version_id < 3) {
1786         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1787         if (err) {
1788             return err;
1789         }
1790     }
1791 
1792     if (kvm_enabled() && spapr->patb_entry) {
1793         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1794         bool radix = !!(spapr->patb_entry & PATE1_GR);
1795         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1796 
1797         /*
1798          * Update LPCR:HR and UPRT as they may not be set properly in
1799          * the stream
1800          */
1801         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1802                             LPCR_HR | LPCR_UPRT);
1803 
1804         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1805         if (err) {
1806             error_report("Process table config unsupported by the host");
1807             return -EINVAL;
1808         }
1809     }
1810 
1811     err = spapr_irq_post_load(spapr, version_id);
1812     if (err) {
1813         return err;
1814     }
1815 
1816     return err;
1817 }
1818 
1819 static int spapr_pre_save(void *opaque)
1820 {
1821     int rc;
1822 
1823     rc = spapr_caps_pre_save(opaque);
1824     if (rc) {
1825         return rc;
1826     }
1827 
1828     return 0;
1829 }
1830 
1831 static bool version_before_3(void *opaque, int version_id)
1832 {
1833     return version_id < 3;
1834 }
1835 
1836 static bool spapr_pending_events_needed(void *opaque)
1837 {
1838     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1839     return !QTAILQ_EMPTY(&spapr->pending_events);
1840 }
1841 
1842 static const VMStateDescription vmstate_spapr_event_entry = {
1843     .name = "spapr_event_log_entry",
1844     .version_id = 1,
1845     .minimum_version_id = 1,
1846     .fields = (VMStateField[]) {
1847         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1848         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1849         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1850                                      NULL, extended_length),
1851         VMSTATE_END_OF_LIST()
1852     },
1853 };
1854 
1855 static const VMStateDescription vmstate_spapr_pending_events = {
1856     .name = "spapr_pending_events",
1857     .version_id = 1,
1858     .minimum_version_id = 1,
1859     .needed = spapr_pending_events_needed,
1860     .fields = (VMStateField[]) {
1861         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1862                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1863         VMSTATE_END_OF_LIST()
1864     },
1865 };
1866 
1867 static bool spapr_ov5_cas_needed(void *opaque)
1868 {
1869     SpaprMachineState *spapr = opaque;
1870     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1871     bool cas_needed;
1872 
1873     /* Prior to the introduction of SpaprOptionVector, we had two option
1874      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1875      * Both of these options encode machine topology into the device-tree
1876      * in such a way that the now-booted OS should still be able to interact
1877      * appropriately with QEMU regardless of what options were actually
1878      * negotiatied on the source side.
1879      *
1880      * As such, we can avoid migrating the CAS-negotiated options if these
1881      * are the only options available on the current machine/platform.
1882      * Since these are the only options available for pseries-2.7 and
1883      * earlier, this allows us to maintain old->new/new->old migration
1884      * compatibility.
1885      *
1886      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1887      * via default pseries-2.8 machines and explicit command-line parameters.
1888      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1889      * of the actual CAS-negotiated values to continue working properly. For
1890      * example, availability of memory unplug depends on knowing whether
1891      * OV5_HP_EVT was negotiated via CAS.
1892      *
1893      * Thus, for any cases where the set of available CAS-negotiatable
1894      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1895      * include the CAS-negotiated options in the migration stream, unless
1896      * if they affect boot time behaviour only.
1897      */
1898     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1899     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1900     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1901 
1902     /* We need extra information if we have any bits outside the mask
1903      * defined above */
1904     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1905 
1906     spapr_ovec_cleanup(ov5_mask);
1907 
1908     return cas_needed;
1909 }
1910 
1911 static const VMStateDescription vmstate_spapr_ov5_cas = {
1912     .name = "spapr_option_vector_ov5_cas",
1913     .version_id = 1,
1914     .minimum_version_id = 1,
1915     .needed = spapr_ov5_cas_needed,
1916     .fields = (VMStateField[]) {
1917         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1918                                  vmstate_spapr_ovec, SpaprOptionVector),
1919         VMSTATE_END_OF_LIST()
1920     },
1921 };
1922 
1923 static bool spapr_patb_entry_needed(void *opaque)
1924 {
1925     SpaprMachineState *spapr = opaque;
1926 
1927     return !!spapr->patb_entry;
1928 }
1929 
1930 static const VMStateDescription vmstate_spapr_patb_entry = {
1931     .name = "spapr_patb_entry",
1932     .version_id = 1,
1933     .minimum_version_id = 1,
1934     .needed = spapr_patb_entry_needed,
1935     .fields = (VMStateField[]) {
1936         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1937         VMSTATE_END_OF_LIST()
1938     },
1939 };
1940 
1941 static bool spapr_irq_map_needed(void *opaque)
1942 {
1943     SpaprMachineState *spapr = opaque;
1944 
1945     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1946 }
1947 
1948 static const VMStateDescription vmstate_spapr_irq_map = {
1949     .name = "spapr_irq_map",
1950     .version_id = 1,
1951     .minimum_version_id = 1,
1952     .needed = spapr_irq_map_needed,
1953     .fields = (VMStateField[]) {
1954         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1955         VMSTATE_END_OF_LIST()
1956     },
1957 };
1958 
1959 static bool spapr_dtb_needed(void *opaque)
1960 {
1961     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1962 
1963     return smc->update_dt_enabled;
1964 }
1965 
1966 static int spapr_dtb_pre_load(void *opaque)
1967 {
1968     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1969 
1970     g_free(spapr->fdt_blob);
1971     spapr->fdt_blob = NULL;
1972     spapr->fdt_size = 0;
1973 
1974     return 0;
1975 }
1976 
1977 static const VMStateDescription vmstate_spapr_dtb = {
1978     .name = "spapr_dtb",
1979     .version_id = 1,
1980     .minimum_version_id = 1,
1981     .needed = spapr_dtb_needed,
1982     .pre_load = spapr_dtb_pre_load,
1983     .fields = (VMStateField[]) {
1984         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1985         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1986         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1987                                      fdt_size),
1988         VMSTATE_END_OF_LIST()
1989     },
1990 };
1991 
1992 static bool spapr_fwnmi_needed(void *opaque)
1993 {
1994     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1995 
1996     return spapr->fwnmi_machine_check_addr != -1;
1997 }
1998 
1999 static int spapr_fwnmi_pre_save(void *opaque)
2000 {
2001     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2002 
2003     /*
2004      * Check if machine check handling is in progress and print a
2005      * warning message.
2006      */
2007     if (spapr->fwnmi_machine_check_interlock != -1) {
2008         warn_report("A machine check is being handled during migration. The"
2009                 "handler may run and log hardware error on the destination");
2010     }
2011 
2012     return 0;
2013 }
2014 
2015 static const VMStateDescription vmstate_spapr_fwnmi = {
2016     .name = "spapr_fwnmi",
2017     .version_id = 1,
2018     .minimum_version_id = 1,
2019     .needed = spapr_fwnmi_needed,
2020     .pre_save = spapr_fwnmi_pre_save,
2021     .fields = (VMStateField[]) {
2022         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2023         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2024         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2025         VMSTATE_END_OF_LIST()
2026     },
2027 };
2028 
2029 static const VMStateDescription vmstate_spapr = {
2030     .name = "spapr",
2031     .version_id = 3,
2032     .minimum_version_id = 1,
2033     .pre_load = spapr_pre_load,
2034     .post_load = spapr_post_load,
2035     .pre_save = spapr_pre_save,
2036     .fields = (VMStateField[]) {
2037         /* used to be @next_irq */
2038         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2039 
2040         /* RTC offset */
2041         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2042 
2043         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2044         VMSTATE_END_OF_LIST()
2045     },
2046     .subsections = (const VMStateDescription*[]) {
2047         &vmstate_spapr_ov5_cas,
2048         &vmstate_spapr_patb_entry,
2049         &vmstate_spapr_pending_events,
2050         &vmstate_spapr_cap_htm,
2051         &vmstate_spapr_cap_vsx,
2052         &vmstate_spapr_cap_dfp,
2053         &vmstate_spapr_cap_cfpc,
2054         &vmstate_spapr_cap_sbbc,
2055         &vmstate_spapr_cap_ibs,
2056         &vmstate_spapr_cap_hpt_maxpagesize,
2057         &vmstate_spapr_irq_map,
2058         &vmstate_spapr_cap_nested_kvm_hv,
2059         &vmstate_spapr_dtb,
2060         &vmstate_spapr_cap_large_decr,
2061         &vmstate_spapr_cap_ccf_assist,
2062         &vmstate_spapr_cap_fwnmi,
2063         &vmstate_spapr_fwnmi,
2064         NULL
2065     }
2066 };
2067 
2068 static int htab_save_setup(QEMUFile *f, void *opaque)
2069 {
2070     SpaprMachineState *spapr = opaque;
2071 
2072     /* "Iteration" header */
2073     if (!spapr->htab_shift) {
2074         qemu_put_be32(f, -1);
2075     } else {
2076         qemu_put_be32(f, spapr->htab_shift);
2077     }
2078 
2079     if (spapr->htab) {
2080         spapr->htab_save_index = 0;
2081         spapr->htab_first_pass = true;
2082     } else {
2083         if (spapr->htab_shift) {
2084             assert(kvm_enabled());
2085         }
2086     }
2087 
2088 
2089     return 0;
2090 }
2091 
2092 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2093                             int chunkstart, int n_valid, int n_invalid)
2094 {
2095     qemu_put_be32(f, chunkstart);
2096     qemu_put_be16(f, n_valid);
2097     qemu_put_be16(f, n_invalid);
2098     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2099                     HASH_PTE_SIZE_64 * n_valid);
2100 }
2101 
2102 static void htab_save_end_marker(QEMUFile *f)
2103 {
2104     qemu_put_be32(f, 0);
2105     qemu_put_be16(f, 0);
2106     qemu_put_be16(f, 0);
2107 }
2108 
2109 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2110                                  int64_t max_ns)
2111 {
2112     bool has_timeout = max_ns != -1;
2113     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2114     int index = spapr->htab_save_index;
2115     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2116 
2117     assert(spapr->htab_first_pass);
2118 
2119     do {
2120         int chunkstart;
2121 
2122         /* Consume invalid HPTEs */
2123         while ((index < htabslots)
2124                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2125             CLEAN_HPTE(HPTE(spapr->htab, index));
2126             index++;
2127         }
2128 
2129         /* Consume valid HPTEs */
2130         chunkstart = index;
2131         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2132                && HPTE_VALID(HPTE(spapr->htab, index))) {
2133             CLEAN_HPTE(HPTE(spapr->htab, index));
2134             index++;
2135         }
2136 
2137         if (index > chunkstart) {
2138             int n_valid = index - chunkstart;
2139 
2140             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2141 
2142             if (has_timeout &&
2143                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2144                 break;
2145             }
2146         }
2147     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2148 
2149     if (index >= htabslots) {
2150         assert(index == htabslots);
2151         index = 0;
2152         spapr->htab_first_pass = false;
2153     }
2154     spapr->htab_save_index = index;
2155 }
2156 
2157 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2158                                 int64_t max_ns)
2159 {
2160     bool final = max_ns < 0;
2161     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2162     int examined = 0, sent = 0;
2163     int index = spapr->htab_save_index;
2164     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2165 
2166     assert(!spapr->htab_first_pass);
2167 
2168     do {
2169         int chunkstart, invalidstart;
2170 
2171         /* Consume non-dirty HPTEs */
2172         while ((index < htabslots)
2173                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2174             index++;
2175             examined++;
2176         }
2177 
2178         chunkstart = index;
2179         /* Consume valid dirty HPTEs */
2180         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2181                && HPTE_DIRTY(HPTE(spapr->htab, index))
2182                && HPTE_VALID(HPTE(spapr->htab, index))) {
2183             CLEAN_HPTE(HPTE(spapr->htab, index));
2184             index++;
2185             examined++;
2186         }
2187 
2188         invalidstart = index;
2189         /* Consume invalid dirty HPTEs */
2190         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2191                && HPTE_DIRTY(HPTE(spapr->htab, index))
2192                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2193             CLEAN_HPTE(HPTE(spapr->htab, index));
2194             index++;
2195             examined++;
2196         }
2197 
2198         if (index > chunkstart) {
2199             int n_valid = invalidstart - chunkstart;
2200             int n_invalid = index - invalidstart;
2201 
2202             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2203             sent += index - chunkstart;
2204 
2205             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2206                 break;
2207             }
2208         }
2209 
2210         if (examined >= htabslots) {
2211             break;
2212         }
2213 
2214         if (index >= htabslots) {
2215             assert(index == htabslots);
2216             index = 0;
2217         }
2218     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2219 
2220     if (index >= htabslots) {
2221         assert(index == htabslots);
2222         index = 0;
2223     }
2224 
2225     spapr->htab_save_index = index;
2226 
2227     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2228 }
2229 
2230 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2231 #define MAX_KVM_BUF_SIZE    2048
2232 
2233 static int htab_save_iterate(QEMUFile *f, void *opaque)
2234 {
2235     SpaprMachineState *spapr = opaque;
2236     int fd;
2237     int rc = 0;
2238 
2239     /* Iteration header */
2240     if (!spapr->htab_shift) {
2241         qemu_put_be32(f, -1);
2242         return 1;
2243     } else {
2244         qemu_put_be32(f, 0);
2245     }
2246 
2247     if (!spapr->htab) {
2248         assert(kvm_enabled());
2249 
2250         fd = get_htab_fd(spapr);
2251         if (fd < 0) {
2252             return fd;
2253         }
2254 
2255         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2256         if (rc < 0) {
2257             return rc;
2258         }
2259     } else  if (spapr->htab_first_pass) {
2260         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2261     } else {
2262         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2263     }
2264 
2265     htab_save_end_marker(f);
2266 
2267     return rc;
2268 }
2269 
2270 static int htab_save_complete(QEMUFile *f, void *opaque)
2271 {
2272     SpaprMachineState *spapr = opaque;
2273     int fd;
2274 
2275     /* Iteration header */
2276     if (!spapr->htab_shift) {
2277         qemu_put_be32(f, -1);
2278         return 0;
2279     } else {
2280         qemu_put_be32(f, 0);
2281     }
2282 
2283     if (!spapr->htab) {
2284         int rc;
2285 
2286         assert(kvm_enabled());
2287 
2288         fd = get_htab_fd(spapr);
2289         if (fd < 0) {
2290             return fd;
2291         }
2292 
2293         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2294         if (rc < 0) {
2295             return rc;
2296         }
2297     } else {
2298         if (spapr->htab_first_pass) {
2299             htab_save_first_pass(f, spapr, -1);
2300         }
2301         htab_save_later_pass(f, spapr, -1);
2302     }
2303 
2304     /* End marker */
2305     htab_save_end_marker(f);
2306 
2307     return 0;
2308 }
2309 
2310 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2311 {
2312     SpaprMachineState *spapr = opaque;
2313     uint32_t section_hdr;
2314     int fd = -1;
2315     Error *local_err = NULL;
2316 
2317     if (version_id < 1 || version_id > 1) {
2318         error_report("htab_load() bad version");
2319         return -EINVAL;
2320     }
2321 
2322     section_hdr = qemu_get_be32(f);
2323 
2324     if (section_hdr == -1) {
2325         spapr_free_hpt(spapr);
2326         return 0;
2327     }
2328 
2329     if (section_hdr) {
2330         /* First section gives the htab size */
2331         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2332         if (local_err) {
2333             error_report_err(local_err);
2334             return -EINVAL;
2335         }
2336         return 0;
2337     }
2338 
2339     if (!spapr->htab) {
2340         assert(kvm_enabled());
2341 
2342         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2343         if (fd < 0) {
2344             error_report_err(local_err);
2345             return fd;
2346         }
2347     }
2348 
2349     while (true) {
2350         uint32_t index;
2351         uint16_t n_valid, n_invalid;
2352 
2353         index = qemu_get_be32(f);
2354         n_valid = qemu_get_be16(f);
2355         n_invalid = qemu_get_be16(f);
2356 
2357         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2358             /* End of Stream */
2359             break;
2360         }
2361 
2362         if ((index + n_valid + n_invalid) >
2363             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2364             /* Bad index in stream */
2365             error_report(
2366                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2367                 index, n_valid, n_invalid, spapr->htab_shift);
2368             return -EINVAL;
2369         }
2370 
2371         if (spapr->htab) {
2372             if (n_valid) {
2373                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2374                                 HASH_PTE_SIZE_64 * n_valid);
2375             }
2376             if (n_invalid) {
2377                 memset(HPTE(spapr->htab, index + n_valid), 0,
2378                        HASH_PTE_SIZE_64 * n_invalid);
2379             }
2380         } else {
2381             int rc;
2382 
2383             assert(fd >= 0);
2384 
2385             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2386             if (rc < 0) {
2387                 return rc;
2388             }
2389         }
2390     }
2391 
2392     if (!spapr->htab) {
2393         assert(fd >= 0);
2394         close(fd);
2395     }
2396 
2397     return 0;
2398 }
2399 
2400 static void htab_save_cleanup(void *opaque)
2401 {
2402     SpaprMachineState *spapr = opaque;
2403 
2404     close_htab_fd(spapr);
2405 }
2406 
2407 static SaveVMHandlers savevm_htab_handlers = {
2408     .save_setup = htab_save_setup,
2409     .save_live_iterate = htab_save_iterate,
2410     .save_live_complete_precopy = htab_save_complete,
2411     .save_cleanup = htab_save_cleanup,
2412     .load_state = htab_load,
2413 };
2414 
2415 static void spapr_boot_set(void *opaque, const char *boot_device,
2416                            Error **errp)
2417 {
2418     MachineState *machine = MACHINE(opaque);
2419     machine->boot_order = g_strdup(boot_device);
2420 }
2421 
2422 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2423 {
2424     MachineState *machine = MACHINE(spapr);
2425     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2426     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2427     int i;
2428 
2429     for (i = 0; i < nr_lmbs; i++) {
2430         uint64_t addr;
2431 
2432         addr = i * lmb_size + machine->device_memory->base;
2433         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2434                                addr / lmb_size);
2435     }
2436 }
2437 
2438 /*
2439  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2440  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2441  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2442  */
2443 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2444 {
2445     int i;
2446 
2447     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2448         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2449                    " is not aligned to %" PRIu64 " MiB",
2450                    machine->ram_size,
2451                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2452         return;
2453     }
2454 
2455     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2456         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2457                    " is not aligned to %" PRIu64 " MiB",
2458                    machine->ram_size,
2459                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2460         return;
2461     }
2462 
2463     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2464         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2465             error_setg(errp,
2466                        "Node %d memory size 0x%" PRIx64
2467                        " is not aligned to %" PRIu64 " MiB",
2468                        i, machine->numa_state->nodes[i].node_mem,
2469                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2470             return;
2471         }
2472     }
2473 }
2474 
2475 /* find cpu slot in machine->possible_cpus by core_id */
2476 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2477 {
2478     int index = id / ms->smp.threads;
2479 
2480     if (index >= ms->possible_cpus->len) {
2481         return NULL;
2482     }
2483     if (idx) {
2484         *idx = index;
2485     }
2486     return &ms->possible_cpus->cpus[index];
2487 }
2488 
2489 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2490 {
2491     MachineState *ms = MACHINE(spapr);
2492     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2493     Error *local_err = NULL;
2494     bool vsmt_user = !!spapr->vsmt;
2495     int kvm_smt = kvmppc_smt_threads();
2496     int ret;
2497     unsigned int smp_threads = ms->smp.threads;
2498 
2499     if (!kvm_enabled() && (smp_threads > 1)) {
2500         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2501                      "on a pseries machine");
2502         goto out;
2503     }
2504     if (!is_power_of_2(smp_threads)) {
2505         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2506                      "machine because it must be a power of 2", smp_threads);
2507         goto out;
2508     }
2509 
2510     /* Detemine the VSMT mode to use: */
2511     if (vsmt_user) {
2512         if (spapr->vsmt < smp_threads) {
2513             error_setg(&local_err, "Cannot support VSMT mode %d"
2514                          " because it must be >= threads/core (%d)",
2515                          spapr->vsmt, smp_threads);
2516             goto out;
2517         }
2518         /* In this case, spapr->vsmt has been set by the command line */
2519     } else if (!smc->smp_threads_vsmt) {
2520         /*
2521          * Default VSMT value is tricky, because we need it to be as
2522          * consistent as possible (for migration), but this requires
2523          * changing it for at least some existing cases.  We pick 8 as
2524          * the value that we'd get with KVM on POWER8, the
2525          * overwhelmingly common case in production systems.
2526          */
2527         spapr->vsmt = MAX(8, smp_threads);
2528     } else {
2529         spapr->vsmt = smp_threads;
2530     }
2531 
2532     /* KVM: If necessary, set the SMT mode: */
2533     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2534         ret = kvmppc_set_smt_threads(spapr->vsmt);
2535         if (ret) {
2536             /* Looks like KVM isn't able to change VSMT mode */
2537             error_setg(&local_err,
2538                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2539                        spapr->vsmt, ret);
2540             /* We can live with that if the default one is big enough
2541              * for the number of threads, and a submultiple of the one
2542              * we want.  In this case we'll waste some vcpu ids, but
2543              * behaviour will be correct */
2544             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2545                 warn_report_err(local_err);
2546                 local_err = NULL;
2547                 goto out;
2548             } else {
2549                 if (!vsmt_user) {
2550                     error_append_hint(&local_err,
2551                                       "On PPC, a VM with %d threads/core"
2552                                       " on a host with %d threads/core"
2553                                       " requires the use of VSMT mode %d.\n",
2554                                       smp_threads, kvm_smt, spapr->vsmt);
2555                 }
2556                 kvmppc_error_append_smt_possible_hint(&local_err);
2557                 goto out;
2558             }
2559         }
2560     }
2561     /* else TCG: nothing to do currently */
2562 out:
2563     error_propagate(errp, local_err);
2564 }
2565 
2566 static void spapr_init_cpus(SpaprMachineState *spapr)
2567 {
2568     MachineState *machine = MACHINE(spapr);
2569     MachineClass *mc = MACHINE_GET_CLASS(machine);
2570     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2571     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2572     const CPUArchIdList *possible_cpus;
2573     unsigned int smp_cpus = machine->smp.cpus;
2574     unsigned int smp_threads = machine->smp.threads;
2575     unsigned int max_cpus = machine->smp.max_cpus;
2576     int boot_cores_nr = smp_cpus / smp_threads;
2577     int i;
2578 
2579     possible_cpus = mc->possible_cpu_arch_ids(machine);
2580     if (mc->has_hotpluggable_cpus) {
2581         if (smp_cpus % smp_threads) {
2582             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2583                          smp_cpus, smp_threads);
2584             exit(1);
2585         }
2586         if (max_cpus % smp_threads) {
2587             error_report("max_cpus (%u) must be multiple of threads (%u)",
2588                          max_cpus, smp_threads);
2589             exit(1);
2590         }
2591     } else {
2592         if (max_cpus != smp_cpus) {
2593             error_report("This machine version does not support CPU hotplug");
2594             exit(1);
2595         }
2596         boot_cores_nr = possible_cpus->len;
2597     }
2598 
2599     if (smc->pre_2_10_has_unused_icps) {
2600         int i;
2601 
2602         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2603             /* Dummy entries get deregistered when real ICPState objects
2604              * are registered during CPU core hotplug.
2605              */
2606             pre_2_10_vmstate_register_dummy_icp(i);
2607         }
2608     }
2609 
2610     for (i = 0; i < possible_cpus->len; i++) {
2611         int core_id = i * smp_threads;
2612 
2613         if (mc->has_hotpluggable_cpus) {
2614             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2615                                    spapr_vcpu_id(spapr, core_id));
2616         }
2617 
2618         if (i < boot_cores_nr) {
2619             Object *core  = object_new(type);
2620             int nr_threads = smp_threads;
2621 
2622             /* Handle the partially filled core for older machine types */
2623             if ((i + 1) * smp_threads >= smp_cpus) {
2624                 nr_threads = smp_cpus - i * smp_threads;
2625             }
2626 
2627             object_property_set_int(core, nr_threads, "nr-threads",
2628                                     &error_fatal);
2629             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2630                                     &error_fatal);
2631             qdev_realize(DEVICE(core), NULL, &error_fatal);
2632 
2633             object_unref(core);
2634         }
2635     }
2636 }
2637 
2638 static PCIHostState *spapr_create_default_phb(void)
2639 {
2640     DeviceState *dev;
2641 
2642     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2643     qdev_prop_set_uint32(dev, "index", 0);
2644     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2645 
2646     return PCI_HOST_BRIDGE(dev);
2647 }
2648 
2649 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2650 {
2651     MachineState *machine = MACHINE(spapr);
2652     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2653     hwaddr rma_size = machine->ram_size;
2654     hwaddr node0_size = spapr_node0_size(machine);
2655 
2656     /* RMA has to fit in the first NUMA node */
2657     rma_size = MIN(rma_size, node0_size);
2658 
2659     /*
2660      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2661      * never exceed that
2662      */
2663     rma_size = MIN(rma_size, 1 * TiB);
2664 
2665     /*
2666      * Clamp the RMA size based on machine type.  This is for
2667      * migration compatibility with older qemu versions, which limited
2668      * the RMA size for complicated and mostly bad reasons.
2669      */
2670     if (smc->rma_limit) {
2671         rma_size = MIN(rma_size, smc->rma_limit);
2672     }
2673 
2674     if (rma_size < MIN_RMA_SLOF) {
2675         error_setg(errp,
2676                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2677                    "ldMiB guest RMA (Real Mode Area memory)",
2678                    MIN_RMA_SLOF / MiB);
2679         return 0;
2680     }
2681 
2682     return rma_size;
2683 }
2684 
2685 /* pSeries LPAR / sPAPR hardware init */
2686 static void spapr_machine_init(MachineState *machine)
2687 {
2688     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2689     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2690     MachineClass *mc = MACHINE_GET_CLASS(machine);
2691     const char *kernel_filename = machine->kernel_filename;
2692     const char *initrd_filename = machine->initrd_filename;
2693     PCIHostState *phb;
2694     int i;
2695     MemoryRegion *sysmem = get_system_memory();
2696     long load_limit, fw_size;
2697     char *filename;
2698     Error *resize_hpt_err = NULL;
2699 
2700     msi_nonbroken = true;
2701 
2702     QLIST_INIT(&spapr->phbs);
2703     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2704 
2705     /* Determine capabilities to run with */
2706     spapr_caps_init(spapr);
2707 
2708     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2709     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2710         /*
2711          * If the user explicitly requested a mode we should either
2712          * supply it, or fail completely (which we do below).  But if
2713          * it's not set explicitly, we reset our mode to something
2714          * that works
2715          */
2716         if (resize_hpt_err) {
2717             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2718             error_free(resize_hpt_err);
2719             resize_hpt_err = NULL;
2720         } else {
2721             spapr->resize_hpt = smc->resize_hpt_default;
2722         }
2723     }
2724 
2725     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2726 
2727     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2728         /*
2729          * User requested HPT resize, but this host can't supply it.  Bail out
2730          */
2731         error_report_err(resize_hpt_err);
2732         exit(1);
2733     }
2734 
2735     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2736 
2737     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2738     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2739 
2740     /*
2741      * VSMT must be set in order to be able to compute VCPU ids, ie to
2742      * call spapr_max_server_number() or spapr_vcpu_id().
2743      */
2744     spapr_set_vsmt_mode(spapr, &error_fatal);
2745 
2746     /* Set up Interrupt Controller before we create the VCPUs */
2747     spapr_irq_init(spapr, &error_fatal);
2748 
2749     /* Set up containers for ibm,client-architecture-support negotiated options
2750      */
2751     spapr->ov5 = spapr_ovec_new();
2752     spapr->ov5_cas = spapr_ovec_new();
2753 
2754     if (smc->dr_lmb_enabled) {
2755         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2756         spapr_validate_node_memory(machine, &error_fatal);
2757     }
2758 
2759     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2760 
2761     /* advertise support for dedicated HP event source to guests */
2762     if (spapr->use_hotplug_event_source) {
2763         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2764     }
2765 
2766     /* advertise support for HPT resizing */
2767     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2768         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2769     }
2770 
2771     /* advertise support for ibm,dyamic-memory-v2 */
2772     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2773 
2774     /* advertise XIVE on POWER9 machines */
2775     if (spapr->irq->xive) {
2776         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2777     }
2778 
2779     /* init CPUs */
2780     spapr_init_cpus(spapr);
2781 
2782     /*
2783      * check we don't have a memory-less/cpu-less NUMA node
2784      * Firmware relies on the existing memory/cpu topology to provide the
2785      * NUMA topology to the kernel.
2786      * And the linux kernel needs to know the NUMA topology at start
2787      * to be able to hotplug CPUs later.
2788      */
2789     if (machine->numa_state->num_nodes) {
2790         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2791             /* check for memory-less node */
2792             if (machine->numa_state->nodes[i].node_mem == 0) {
2793                 CPUState *cs;
2794                 int found = 0;
2795                 /* check for cpu-less node */
2796                 CPU_FOREACH(cs) {
2797                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2798                     if (cpu->node_id == i) {
2799                         found = 1;
2800                         break;
2801                     }
2802                 }
2803                 /* memory-less and cpu-less node */
2804                 if (!found) {
2805                     error_report(
2806                        "Memory-less/cpu-less nodes are not supported (node %d)",
2807                                  i);
2808                     exit(1);
2809                 }
2810             }
2811         }
2812 
2813     }
2814 
2815     /*
2816      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2817      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2818      * called from vPHB reset handler so we initialize the counter here.
2819      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2820      * must be equally distant from any other node.
2821      * The final value of spapr->gpu_numa_id is going to be written to
2822      * max-associativity-domains in spapr_build_fdt().
2823      */
2824     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2825 
2826     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2827         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2828                               spapr->max_compat_pvr)) {
2829         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2830         /* KVM and TCG always allow GTSE with radix... */
2831         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2832     }
2833     /* ... but not with hash (currently). */
2834 
2835     if (kvm_enabled()) {
2836         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2837         kvmppc_enable_logical_ci_hcalls();
2838         kvmppc_enable_set_mode_hcall();
2839 
2840         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2841         kvmppc_enable_clear_ref_mod_hcalls();
2842 
2843         /* Enable H_PAGE_INIT */
2844         kvmppc_enable_h_page_init();
2845     }
2846 
2847     /* map RAM */
2848     memory_region_add_subregion(sysmem, 0, machine->ram);
2849 
2850     /* always allocate the device memory information */
2851     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2852 
2853     /* initialize hotplug memory address space */
2854     if (machine->ram_size < machine->maxram_size) {
2855         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2856         /*
2857          * Limit the number of hotpluggable memory slots to half the number
2858          * slots that KVM supports, leaving the other half for PCI and other
2859          * devices. However ensure that number of slots doesn't drop below 32.
2860          */
2861         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2862                            SPAPR_MAX_RAM_SLOTS;
2863 
2864         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2865             max_memslots = SPAPR_MAX_RAM_SLOTS;
2866         }
2867         if (machine->ram_slots > max_memslots) {
2868             error_report("Specified number of memory slots %"
2869                          PRIu64" exceeds max supported %d",
2870                          machine->ram_slots, max_memslots);
2871             exit(1);
2872         }
2873 
2874         machine->device_memory->base = ROUND_UP(machine->ram_size,
2875                                                 SPAPR_DEVICE_MEM_ALIGN);
2876         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2877                            "device-memory", device_mem_size);
2878         memory_region_add_subregion(sysmem, machine->device_memory->base,
2879                                     &machine->device_memory->mr);
2880     }
2881 
2882     if (smc->dr_lmb_enabled) {
2883         spapr_create_lmb_dr_connectors(spapr);
2884     }
2885 
2886     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2887         /* Create the error string for live migration blocker */
2888         error_setg(&spapr->fwnmi_migration_blocker,
2889             "A machine check is being handled during migration. The handler"
2890             "may run and log hardware error on the destination");
2891     }
2892 
2893     if (mc->nvdimm_supported) {
2894         spapr_create_nvdimm_dr_connectors(spapr);
2895     }
2896 
2897     /* Set up RTAS event infrastructure */
2898     spapr_events_init(spapr);
2899 
2900     /* Set up the RTC RTAS interfaces */
2901     spapr_rtc_create(spapr);
2902 
2903     /* Set up VIO bus */
2904     spapr->vio_bus = spapr_vio_bus_init();
2905 
2906     for (i = 0; i < serial_max_hds(); i++) {
2907         if (serial_hd(i)) {
2908             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2909         }
2910     }
2911 
2912     /* We always have at least the nvram device on VIO */
2913     spapr_create_nvram(spapr);
2914 
2915     /*
2916      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2917      * connectors (described in root DT node's "ibm,drc-types" property)
2918      * are pre-initialized here. additional child connectors (such as
2919      * connectors for a PHBs PCI slots) are added as needed during their
2920      * parent's realization.
2921      */
2922     if (smc->dr_phb_enabled) {
2923         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2924             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2925         }
2926     }
2927 
2928     /* Set up PCI */
2929     spapr_pci_rtas_init();
2930 
2931     phb = spapr_create_default_phb();
2932 
2933     for (i = 0; i < nb_nics; i++) {
2934         NICInfo *nd = &nd_table[i];
2935 
2936         if (!nd->model) {
2937             nd->model = g_strdup("spapr-vlan");
2938         }
2939 
2940         if (g_str_equal(nd->model, "spapr-vlan") ||
2941             g_str_equal(nd->model, "ibmveth")) {
2942             spapr_vlan_create(spapr->vio_bus, nd);
2943         } else {
2944             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2945         }
2946     }
2947 
2948     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2949         spapr_vscsi_create(spapr->vio_bus);
2950     }
2951 
2952     /* Graphics */
2953     if (spapr_vga_init(phb->bus, &error_fatal)) {
2954         spapr->has_graphics = true;
2955         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2956     }
2957 
2958     if (machine->usb) {
2959         if (smc->use_ohci_by_default) {
2960             pci_create_simple(phb->bus, -1, "pci-ohci");
2961         } else {
2962             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2963         }
2964 
2965         if (spapr->has_graphics) {
2966             USBBus *usb_bus = usb_bus_find(-1);
2967 
2968             usb_create_simple(usb_bus, "usb-kbd");
2969             usb_create_simple(usb_bus, "usb-mouse");
2970         }
2971     }
2972 
2973     if (kernel_filename) {
2974         uint64_t lowaddr = 0;
2975 
2976         spapr->kernel_size = load_elf(kernel_filename, NULL,
2977                                       translate_kernel_address, spapr,
2978                                       NULL, &lowaddr, NULL, NULL, 1,
2979                                       PPC_ELF_MACHINE, 0, 0);
2980         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2981             spapr->kernel_size = load_elf(kernel_filename, NULL,
2982                                           translate_kernel_address, spapr, NULL,
2983                                           &lowaddr, NULL, NULL, 0,
2984                                           PPC_ELF_MACHINE,
2985                                           0, 0);
2986             spapr->kernel_le = spapr->kernel_size > 0;
2987         }
2988         if (spapr->kernel_size < 0) {
2989             error_report("error loading %s: %s", kernel_filename,
2990                          load_elf_strerror(spapr->kernel_size));
2991             exit(1);
2992         }
2993 
2994         /* load initrd */
2995         if (initrd_filename) {
2996             /* Try to locate the initrd in the gap between the kernel
2997              * and the firmware. Add a bit of space just in case
2998              */
2999             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3000                                   + 0x1ffff) & ~0xffff;
3001             spapr->initrd_size = load_image_targphys(initrd_filename,
3002                                                      spapr->initrd_base,
3003                                                      load_limit
3004                                                      - spapr->initrd_base);
3005             if (spapr->initrd_size < 0) {
3006                 error_report("could not load initial ram disk '%s'",
3007                              initrd_filename);
3008                 exit(1);
3009             }
3010         }
3011     }
3012 
3013     if (bios_name == NULL) {
3014         bios_name = FW_FILE_NAME;
3015     }
3016     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3017     if (!filename) {
3018         error_report("Could not find LPAR firmware '%s'", bios_name);
3019         exit(1);
3020     }
3021     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3022     if (fw_size <= 0) {
3023         error_report("Could not load LPAR firmware '%s'", filename);
3024         exit(1);
3025     }
3026     g_free(filename);
3027 
3028     /* FIXME: Should register things through the MachineState's qdev
3029      * interface, this is a legacy from the sPAPREnvironment structure
3030      * which predated MachineState but had a similar function */
3031     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3032     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3033                          &savevm_htab_handlers, spapr);
3034 
3035     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3036                              &error_fatal);
3037 
3038     qemu_register_boot_set(spapr_boot_set, spapr);
3039 
3040     /*
3041      * Nothing needs to be done to resume a suspended guest because
3042      * suspending does not change the machine state, so no need for
3043      * a ->wakeup method.
3044      */
3045     qemu_register_wakeup_support();
3046 
3047     if (kvm_enabled()) {
3048         /* to stop and start vmclock */
3049         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3050                                          &spapr->tb);
3051 
3052         kvmppc_spapr_enable_inkernel_multitce();
3053     }
3054 
3055     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3056 }
3057 
3058 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3059 {
3060     if (!vm_type) {
3061         return 0;
3062     }
3063 
3064     if (!strcmp(vm_type, "HV")) {
3065         return 1;
3066     }
3067 
3068     if (!strcmp(vm_type, "PR")) {
3069         return 2;
3070     }
3071 
3072     error_report("Unknown kvm-type specified '%s'", vm_type);
3073     exit(1);
3074 }
3075 
3076 /*
3077  * Implementation of an interface to adjust firmware path
3078  * for the bootindex property handling.
3079  */
3080 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3081                                    DeviceState *dev)
3082 {
3083 #define CAST(type, obj, name) \
3084     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3085     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3086     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3087     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3088 
3089     if (d) {
3090         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3091         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3092         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3093 
3094         if (spapr) {
3095             /*
3096              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3097              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3098              * 0x8000 | (target << 8) | (bus << 5) | lun
3099              * (see the "Logical unit addressing format" table in SAM5)
3100              */
3101             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3102             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3103                                    (uint64_t)id << 48);
3104         } else if (virtio) {
3105             /*
3106              * We use SRP luns of the form 01000000 | (target << 8) | lun
3107              * in the top 32 bits of the 64-bit LUN
3108              * Note: the quote above is from SLOF and it is wrong,
3109              * the actual binding is:
3110              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3111              */
3112             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3113             if (d->lun >= 256) {
3114                 /* Use the LUN "flat space addressing method" */
3115                 id |= 0x4000;
3116             }
3117             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3118                                    (uint64_t)id << 32);
3119         } else if (usb) {
3120             /*
3121              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3122              * in the top 32 bits of the 64-bit LUN
3123              */
3124             unsigned usb_port = atoi(usb->port->path);
3125             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3126             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3127                                    (uint64_t)id << 32);
3128         }
3129     }
3130 
3131     /*
3132      * SLOF probes the USB devices, and if it recognizes that the device is a
3133      * storage device, it changes its name to "storage" instead of "usb-host",
3134      * and additionally adds a child node for the SCSI LUN, so the correct
3135      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3136      */
3137     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3138         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3139         if (usb_host_dev_is_scsi_storage(usbdev)) {
3140             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3141         }
3142     }
3143 
3144     if (phb) {
3145         /* Replace "pci" with "pci@800000020000000" */
3146         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3147     }
3148 
3149     if (vsc) {
3150         /* Same logic as virtio above */
3151         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3152         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3153     }
3154 
3155     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3156         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3157         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3158         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3159     }
3160 
3161     return NULL;
3162 }
3163 
3164 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3165 {
3166     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3167 
3168     return g_strdup(spapr->kvm_type);
3169 }
3170 
3171 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3172 {
3173     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3174 
3175     g_free(spapr->kvm_type);
3176     spapr->kvm_type = g_strdup(value);
3177 }
3178 
3179 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3180 {
3181     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3182 
3183     return spapr->use_hotplug_event_source;
3184 }
3185 
3186 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3187                                             Error **errp)
3188 {
3189     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3190 
3191     spapr->use_hotplug_event_source = value;
3192 }
3193 
3194 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3195 {
3196     return true;
3197 }
3198 
3199 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3200 {
3201     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3202 
3203     switch (spapr->resize_hpt) {
3204     case SPAPR_RESIZE_HPT_DEFAULT:
3205         return g_strdup("default");
3206     case SPAPR_RESIZE_HPT_DISABLED:
3207         return g_strdup("disabled");
3208     case SPAPR_RESIZE_HPT_ENABLED:
3209         return g_strdup("enabled");
3210     case SPAPR_RESIZE_HPT_REQUIRED:
3211         return g_strdup("required");
3212     }
3213     g_assert_not_reached();
3214 }
3215 
3216 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3217 {
3218     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3219 
3220     if (strcmp(value, "default") == 0) {
3221         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3222     } else if (strcmp(value, "disabled") == 0) {
3223         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3224     } else if (strcmp(value, "enabled") == 0) {
3225         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3226     } else if (strcmp(value, "required") == 0) {
3227         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3228     } else {
3229         error_setg(errp, "Bad value for \"resize-hpt\" property");
3230     }
3231 }
3232 
3233 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3234 {
3235     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3236 
3237     if (spapr->irq == &spapr_irq_xics_legacy) {
3238         return g_strdup("legacy");
3239     } else if (spapr->irq == &spapr_irq_xics) {
3240         return g_strdup("xics");
3241     } else if (spapr->irq == &spapr_irq_xive) {
3242         return g_strdup("xive");
3243     } else if (spapr->irq == &spapr_irq_dual) {
3244         return g_strdup("dual");
3245     }
3246     g_assert_not_reached();
3247 }
3248 
3249 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3250 {
3251     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3252 
3253     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3254         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3255         return;
3256     }
3257 
3258     /* The legacy IRQ backend can not be set */
3259     if (strcmp(value, "xics") == 0) {
3260         spapr->irq = &spapr_irq_xics;
3261     } else if (strcmp(value, "xive") == 0) {
3262         spapr->irq = &spapr_irq_xive;
3263     } else if (strcmp(value, "dual") == 0) {
3264         spapr->irq = &spapr_irq_dual;
3265     } else {
3266         error_setg(errp, "Bad value for \"ic-mode\" property");
3267     }
3268 }
3269 
3270 static char *spapr_get_host_model(Object *obj, Error **errp)
3271 {
3272     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3273 
3274     return g_strdup(spapr->host_model);
3275 }
3276 
3277 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3278 {
3279     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3280 
3281     g_free(spapr->host_model);
3282     spapr->host_model = g_strdup(value);
3283 }
3284 
3285 static char *spapr_get_host_serial(Object *obj, Error **errp)
3286 {
3287     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3288 
3289     return g_strdup(spapr->host_serial);
3290 }
3291 
3292 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3293 {
3294     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3295 
3296     g_free(spapr->host_serial);
3297     spapr->host_serial = g_strdup(value);
3298 }
3299 
3300 static void spapr_instance_init(Object *obj)
3301 {
3302     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3303     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3304 
3305     spapr->htab_fd = -1;
3306     spapr->use_hotplug_event_source = true;
3307     object_property_add_str(obj, "kvm-type",
3308                             spapr_get_kvm_type, spapr_set_kvm_type);
3309     object_property_set_description(obj, "kvm-type",
3310                                     "Specifies the KVM virtualization mode (HV, PR)");
3311     object_property_add_bool(obj, "modern-hotplug-events",
3312                             spapr_get_modern_hotplug_events,
3313                             spapr_set_modern_hotplug_events);
3314     object_property_set_description(obj, "modern-hotplug-events",
3315                                     "Use dedicated hotplug event mechanism in"
3316                                     " place of standard EPOW events when possible"
3317                                     " (required for memory hot-unplug support)");
3318     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3319                             "Maximum permitted CPU compatibility mode");
3320 
3321     object_property_add_str(obj, "resize-hpt",
3322                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3323     object_property_set_description(obj, "resize-hpt",
3324                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3325     object_property_add_uint32_ptr(obj, "vsmt",
3326                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3327     object_property_set_description(obj, "vsmt",
3328                                     "Virtual SMT: KVM behaves as if this were"
3329                                     " the host's SMT mode");
3330 
3331     object_property_add_bool(obj, "vfio-no-msix-emulation",
3332                              spapr_get_msix_emulation, NULL);
3333 
3334     object_property_add_uint64_ptr(obj, "kernel-addr",
3335                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3336     object_property_set_description(obj, "kernel-addr",
3337                                     stringify(KERNEL_LOAD_ADDR)
3338                                     " for -kernel is the default");
3339     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3340     /* The machine class defines the default interrupt controller mode */
3341     spapr->irq = smc->irq;
3342     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3343                             spapr_set_ic_mode);
3344     object_property_set_description(obj, "ic-mode",
3345                  "Specifies the interrupt controller mode (xics, xive, dual)");
3346 
3347     object_property_add_str(obj, "host-model",
3348         spapr_get_host_model, spapr_set_host_model);
3349     object_property_set_description(obj, "host-model",
3350         "Host model to advertise in guest device tree");
3351     object_property_add_str(obj, "host-serial",
3352         spapr_get_host_serial, spapr_set_host_serial);
3353     object_property_set_description(obj, "host-serial",
3354         "Host serial number to advertise in guest device tree");
3355 }
3356 
3357 static void spapr_machine_finalizefn(Object *obj)
3358 {
3359     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3360 
3361     g_free(spapr->kvm_type);
3362 }
3363 
3364 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3365 {
3366     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3367     PowerPCCPU *cpu = POWERPC_CPU(cs);
3368     CPUPPCState *env = &cpu->env;
3369 
3370     cpu_synchronize_state(cs);
3371     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3372     if (spapr->fwnmi_system_reset_addr != -1) {
3373         uint64_t rtas_addr, addr;
3374 
3375         /* get rtas addr from fdt */
3376         rtas_addr = spapr_get_rtas_addr();
3377         if (!rtas_addr) {
3378             qemu_system_guest_panicked(NULL);
3379             return;
3380         }
3381 
3382         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3383         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3384         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3385         env->gpr[3] = addr;
3386     }
3387     ppc_cpu_do_system_reset(cs);
3388     if (spapr->fwnmi_system_reset_addr != -1) {
3389         env->nip = spapr->fwnmi_system_reset_addr;
3390     }
3391 }
3392 
3393 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3394 {
3395     CPUState *cs;
3396 
3397     CPU_FOREACH(cs) {
3398         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3399     }
3400 }
3401 
3402 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3403                           void *fdt, int *fdt_start_offset, Error **errp)
3404 {
3405     uint64_t addr;
3406     uint32_t node;
3407 
3408     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3409     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3410                                     &error_abort);
3411     *fdt_start_offset = spapr_dt_memory_node(fdt, node, addr,
3412                                              SPAPR_MEMORY_BLOCK_SIZE);
3413     return 0;
3414 }
3415 
3416 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3417                            bool dedicated_hp_event_source, Error **errp)
3418 {
3419     SpaprDrc *drc;
3420     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3421     int i;
3422     uint64_t addr = addr_start;
3423     bool hotplugged = spapr_drc_hotplugged(dev);
3424     Error *local_err = NULL;
3425 
3426     for (i = 0; i < nr_lmbs; i++) {
3427         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3428                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3429         g_assert(drc);
3430 
3431         spapr_drc_attach(drc, dev, &local_err);
3432         if (local_err) {
3433             while (addr > addr_start) {
3434                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3435                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3436                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3437                 spapr_drc_detach(drc);
3438             }
3439             error_propagate(errp, local_err);
3440             return;
3441         }
3442         if (!hotplugged) {
3443             spapr_drc_reset(drc);
3444         }
3445         addr += SPAPR_MEMORY_BLOCK_SIZE;
3446     }
3447     /* send hotplug notification to the
3448      * guest only in case of hotplugged memory
3449      */
3450     if (hotplugged) {
3451         if (dedicated_hp_event_source) {
3452             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3453                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3454             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3455                                                    nr_lmbs,
3456                                                    spapr_drc_index(drc));
3457         } else {
3458             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3459                                            nr_lmbs);
3460         }
3461     }
3462 }
3463 
3464 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3465                               Error **errp)
3466 {
3467     Error *local_err = NULL;
3468     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3469     PCDIMMDevice *dimm = PC_DIMM(dev);
3470     uint64_t size, addr, slot;
3471     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3472 
3473     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3474 
3475     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3476     if (local_err) {
3477         goto out;
3478     }
3479 
3480     if (!is_nvdimm) {
3481         addr = object_property_get_uint(OBJECT(dimm),
3482                                         PC_DIMM_ADDR_PROP, &local_err);
3483         if (local_err) {
3484             goto out_unplug;
3485         }
3486         spapr_add_lmbs(dev, addr, size,
3487                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3488                        &local_err);
3489     } else {
3490         slot = object_property_get_uint(OBJECT(dimm),
3491                                         PC_DIMM_SLOT_PROP, &local_err);
3492         if (local_err) {
3493             goto out_unplug;
3494         }
3495         spapr_add_nvdimm(dev, slot, &local_err);
3496     }
3497 
3498     if (local_err) {
3499         goto out_unplug;
3500     }
3501 
3502     return;
3503 
3504 out_unplug:
3505     pc_dimm_unplug(dimm, MACHINE(ms));
3506 out:
3507     error_propagate(errp, local_err);
3508 }
3509 
3510 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3511                                   Error **errp)
3512 {
3513     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3514     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3515     const MachineClass *mc = MACHINE_CLASS(smc);
3516     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3517     PCDIMMDevice *dimm = PC_DIMM(dev);
3518     Error *local_err = NULL;
3519     uint64_t size;
3520     Object *memdev;
3521     hwaddr pagesize;
3522 
3523     if (!smc->dr_lmb_enabled) {
3524         error_setg(errp, "Memory hotplug not supported for this machine");
3525         return;
3526     }
3527 
3528     if (is_nvdimm && !mc->nvdimm_supported) {
3529         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3530         return;
3531     }
3532 
3533     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3534     if (local_err) {
3535         error_propagate(errp, local_err);
3536         return;
3537     }
3538 
3539     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3540         error_setg(errp, "Hotplugged memory size must be a multiple of "
3541                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3542         return;
3543     } else if (is_nvdimm) {
3544         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3545         if (local_err) {
3546             error_propagate(errp, local_err);
3547             return;
3548         }
3549     }
3550 
3551     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3552                                       &error_abort);
3553     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3554     spapr_check_pagesize(spapr, pagesize, &local_err);
3555     if (local_err) {
3556         error_propagate(errp, local_err);
3557         return;
3558     }
3559 
3560     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3561 }
3562 
3563 struct SpaprDimmState {
3564     PCDIMMDevice *dimm;
3565     uint32_t nr_lmbs;
3566     QTAILQ_ENTRY(SpaprDimmState) next;
3567 };
3568 
3569 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3570                                                        PCDIMMDevice *dimm)
3571 {
3572     SpaprDimmState *dimm_state = NULL;
3573 
3574     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3575         if (dimm_state->dimm == dimm) {
3576             break;
3577         }
3578     }
3579     return dimm_state;
3580 }
3581 
3582 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3583                                                       uint32_t nr_lmbs,
3584                                                       PCDIMMDevice *dimm)
3585 {
3586     SpaprDimmState *ds = NULL;
3587 
3588     /*
3589      * If this request is for a DIMM whose removal had failed earlier
3590      * (due to guest's refusal to remove the LMBs), we would have this
3591      * dimm already in the pending_dimm_unplugs list. In that
3592      * case don't add again.
3593      */
3594     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3595     if (!ds) {
3596         ds = g_malloc0(sizeof(SpaprDimmState));
3597         ds->nr_lmbs = nr_lmbs;
3598         ds->dimm = dimm;
3599         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3600     }
3601     return ds;
3602 }
3603 
3604 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3605                                               SpaprDimmState *dimm_state)
3606 {
3607     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3608     g_free(dimm_state);
3609 }
3610 
3611 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3612                                                         PCDIMMDevice *dimm)
3613 {
3614     SpaprDrc *drc;
3615     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3616                                                   &error_abort);
3617     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3618     uint32_t avail_lmbs = 0;
3619     uint64_t addr_start, addr;
3620     int i;
3621 
3622     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3623                                          &error_abort);
3624 
3625     addr = addr_start;
3626     for (i = 0; i < nr_lmbs; i++) {
3627         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3628                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3629         g_assert(drc);
3630         if (drc->dev) {
3631             avail_lmbs++;
3632         }
3633         addr += SPAPR_MEMORY_BLOCK_SIZE;
3634     }
3635 
3636     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3637 }
3638 
3639 /* Callback to be called during DRC release. */
3640 void spapr_lmb_release(DeviceState *dev)
3641 {
3642     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3643     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3644     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3645 
3646     /* This information will get lost if a migration occurs
3647      * during the unplug process. In this case recover it. */
3648     if (ds == NULL) {
3649         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3650         g_assert(ds);
3651         /* The DRC being examined by the caller at least must be counted */
3652         g_assert(ds->nr_lmbs);
3653     }
3654 
3655     if (--ds->nr_lmbs) {
3656         return;
3657     }
3658 
3659     /*
3660      * Now that all the LMBs have been removed by the guest, call the
3661      * unplug handler chain. This can never fail.
3662      */
3663     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3664     object_unparent(OBJECT(dev));
3665 }
3666 
3667 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3668 {
3669     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3670     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3671 
3672     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3673     qdev_unrealize(dev);
3674     spapr_pending_dimm_unplugs_remove(spapr, ds);
3675 }
3676 
3677 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3678                                         DeviceState *dev, Error **errp)
3679 {
3680     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3681     Error *local_err = NULL;
3682     PCDIMMDevice *dimm = PC_DIMM(dev);
3683     uint32_t nr_lmbs;
3684     uint64_t size, addr_start, addr;
3685     int i;
3686     SpaprDrc *drc;
3687 
3688     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3689         error_setg(&local_err,
3690                    "nvdimm device hot unplug is not supported yet.");
3691         goto out;
3692     }
3693 
3694     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3695     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3696 
3697     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3698                                          &local_err);
3699     if (local_err) {
3700         goto out;
3701     }
3702 
3703     /*
3704      * An existing pending dimm state for this DIMM means that there is an
3705      * unplug operation in progress, waiting for the spapr_lmb_release
3706      * callback to complete the job (BQL can't cover that far). In this case,
3707      * bail out to avoid detaching DRCs that were already released.
3708      */
3709     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3710         error_setg(&local_err,
3711                    "Memory unplug already in progress for device %s",
3712                    dev->id);
3713         goto out;
3714     }
3715 
3716     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3717 
3718     addr = addr_start;
3719     for (i = 0; i < nr_lmbs; i++) {
3720         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3721                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3722         g_assert(drc);
3723 
3724         spapr_drc_detach(drc);
3725         addr += SPAPR_MEMORY_BLOCK_SIZE;
3726     }
3727 
3728     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3729                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3730     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3731                                               nr_lmbs, spapr_drc_index(drc));
3732 out:
3733     error_propagate(errp, local_err);
3734 }
3735 
3736 /* Callback to be called during DRC release. */
3737 void spapr_core_release(DeviceState *dev)
3738 {
3739     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3740 
3741     /* Call the unplug handler chain. This can never fail. */
3742     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3743     object_unparent(OBJECT(dev));
3744 }
3745 
3746 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3747 {
3748     MachineState *ms = MACHINE(hotplug_dev);
3749     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3750     CPUCore *cc = CPU_CORE(dev);
3751     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3752 
3753     if (smc->pre_2_10_has_unused_icps) {
3754         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3755         int i;
3756 
3757         for (i = 0; i < cc->nr_threads; i++) {
3758             CPUState *cs = CPU(sc->threads[i]);
3759 
3760             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3761         }
3762     }
3763 
3764     assert(core_slot);
3765     core_slot->cpu = NULL;
3766     qdev_unrealize(dev);
3767 }
3768 
3769 static
3770 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3771                                Error **errp)
3772 {
3773     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3774     int index;
3775     SpaprDrc *drc;
3776     CPUCore *cc = CPU_CORE(dev);
3777 
3778     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3779         error_setg(errp, "Unable to find CPU core with core-id: %d",
3780                    cc->core_id);
3781         return;
3782     }
3783     if (index == 0) {
3784         error_setg(errp, "Boot CPU core may not be unplugged");
3785         return;
3786     }
3787 
3788     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3789                           spapr_vcpu_id(spapr, cc->core_id));
3790     g_assert(drc);
3791 
3792     if (!spapr_drc_unplug_requested(drc)) {
3793         spapr_drc_detach(drc);
3794         spapr_hotplug_req_remove_by_index(drc);
3795     }
3796 }
3797 
3798 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3799                            void *fdt, int *fdt_start_offset, Error **errp)
3800 {
3801     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3802     CPUState *cs = CPU(core->threads[0]);
3803     PowerPCCPU *cpu = POWERPC_CPU(cs);
3804     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3805     int id = spapr_get_vcpu_id(cpu);
3806     char *nodename;
3807     int offset;
3808 
3809     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3810     offset = fdt_add_subnode(fdt, 0, nodename);
3811     g_free(nodename);
3812 
3813     spapr_dt_cpu(cs, fdt, offset, spapr);
3814 
3815     *fdt_start_offset = offset;
3816     return 0;
3817 }
3818 
3819 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3820                             Error **errp)
3821 {
3822     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3823     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3824     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3825     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3826     CPUCore *cc = CPU_CORE(dev);
3827     CPUState *cs;
3828     SpaprDrc *drc;
3829     Error *local_err = NULL;
3830     CPUArchId *core_slot;
3831     int index;
3832     bool hotplugged = spapr_drc_hotplugged(dev);
3833     int i;
3834 
3835     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3836     if (!core_slot) {
3837         error_setg(errp, "Unable to find CPU core with core-id: %d",
3838                    cc->core_id);
3839         return;
3840     }
3841     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3842                           spapr_vcpu_id(spapr, cc->core_id));
3843 
3844     g_assert(drc || !mc->has_hotpluggable_cpus);
3845 
3846     if (drc) {
3847         spapr_drc_attach(drc, dev, &local_err);
3848         if (local_err) {
3849             error_propagate(errp, local_err);
3850             return;
3851         }
3852 
3853         if (hotplugged) {
3854             /*
3855              * Send hotplug notification interrupt to the guest only
3856              * in case of hotplugged CPUs.
3857              */
3858             spapr_hotplug_req_add_by_index(drc);
3859         } else {
3860             spapr_drc_reset(drc);
3861         }
3862     }
3863 
3864     core_slot->cpu = OBJECT(dev);
3865 
3866     if (smc->pre_2_10_has_unused_icps) {
3867         for (i = 0; i < cc->nr_threads; i++) {
3868             cs = CPU(core->threads[i]);
3869             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3870         }
3871     }
3872 
3873     /*
3874      * Set compatibility mode to match the boot CPU, which was either set
3875      * by the machine reset code or by CAS.
3876      */
3877     if (hotplugged) {
3878         for (i = 0; i < cc->nr_threads; i++) {
3879             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3880                            &local_err);
3881             if (local_err) {
3882                 error_propagate(errp, local_err);
3883                 return;
3884             }
3885         }
3886     }
3887 }
3888 
3889 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3890                                 Error **errp)
3891 {
3892     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3893     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3894     Error *local_err = NULL;
3895     CPUCore *cc = CPU_CORE(dev);
3896     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3897     const char *type = object_get_typename(OBJECT(dev));
3898     CPUArchId *core_slot;
3899     int index;
3900     unsigned int smp_threads = machine->smp.threads;
3901 
3902     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3903         error_setg(&local_err, "CPU hotplug not supported for this machine");
3904         goto out;
3905     }
3906 
3907     if (strcmp(base_core_type, type)) {
3908         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3909         goto out;
3910     }
3911 
3912     if (cc->core_id % smp_threads) {
3913         error_setg(&local_err, "invalid core id %d", cc->core_id);
3914         goto out;
3915     }
3916 
3917     /*
3918      * In general we should have homogeneous threads-per-core, but old
3919      * (pre hotplug support) machine types allow the last core to have
3920      * reduced threads as a compatibility hack for when we allowed
3921      * total vcpus not a multiple of threads-per-core.
3922      */
3923     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3924         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3925                    cc->nr_threads, smp_threads);
3926         goto out;
3927     }
3928 
3929     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3930     if (!core_slot) {
3931         error_setg(&local_err, "core id %d out of range", cc->core_id);
3932         goto out;
3933     }
3934 
3935     if (core_slot->cpu) {
3936         error_setg(&local_err, "core %d already populated", cc->core_id);
3937         goto out;
3938     }
3939 
3940     numa_cpu_pre_plug(core_slot, dev, &local_err);
3941 
3942 out:
3943     error_propagate(errp, local_err);
3944 }
3945 
3946 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3947                           void *fdt, int *fdt_start_offset, Error **errp)
3948 {
3949     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3950     int intc_phandle;
3951 
3952     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3953     if (intc_phandle <= 0) {
3954         return -1;
3955     }
3956 
3957     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3958         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3959         return -1;
3960     }
3961 
3962     /* generally SLOF creates these, for hotplug it's up to QEMU */
3963     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3964 
3965     return 0;
3966 }
3967 
3968 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3969                                Error **errp)
3970 {
3971     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3972     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3973     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3974     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3975 
3976     if (dev->hotplugged && !smc->dr_phb_enabled) {
3977         error_setg(errp, "PHB hotplug not supported for this machine");
3978         return;
3979     }
3980 
3981     if (sphb->index == (uint32_t)-1) {
3982         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3983         return;
3984     }
3985 
3986     /*
3987      * This will check that sphb->index doesn't exceed the maximum number of
3988      * PHBs for the current machine type.
3989      */
3990     smc->phb_placement(spapr, sphb->index,
3991                        &sphb->buid, &sphb->io_win_addr,
3992                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3993                        windows_supported, sphb->dma_liobn,
3994                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3995                        errp);
3996 }
3997 
3998 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3999                            Error **errp)
4000 {
4001     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4002     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4003     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4004     SpaprDrc *drc;
4005     bool hotplugged = spapr_drc_hotplugged(dev);
4006     Error *local_err = NULL;
4007 
4008     if (!smc->dr_phb_enabled) {
4009         return;
4010     }
4011 
4012     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4013     /* hotplug hooks should check it's enabled before getting this far */
4014     assert(drc);
4015 
4016     spapr_drc_attach(drc, dev, &local_err);
4017     if (local_err) {
4018         error_propagate(errp, local_err);
4019         return;
4020     }
4021 
4022     if (hotplugged) {
4023         spapr_hotplug_req_add_by_index(drc);
4024     } else {
4025         spapr_drc_reset(drc);
4026     }
4027 }
4028 
4029 void spapr_phb_release(DeviceState *dev)
4030 {
4031     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4032 
4033     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4034     object_unparent(OBJECT(dev));
4035 }
4036 
4037 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4038 {
4039     qdev_unrealize(dev);
4040 }
4041 
4042 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4043                                      DeviceState *dev, Error **errp)
4044 {
4045     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4046     SpaprDrc *drc;
4047 
4048     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4049     assert(drc);
4050 
4051     if (!spapr_drc_unplug_requested(drc)) {
4052         spapr_drc_detach(drc);
4053         spapr_hotplug_req_remove_by_index(drc);
4054     }
4055 }
4056 
4057 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4058                                  Error **errp)
4059 {
4060     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4061     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4062 
4063     if (spapr->tpm_proxy != NULL) {
4064         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4065         return;
4066     }
4067 
4068     spapr->tpm_proxy = tpm_proxy;
4069 }
4070 
4071 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4072 {
4073     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4074 
4075     qdev_unrealize(dev);
4076     object_unparent(OBJECT(dev));
4077     spapr->tpm_proxy = NULL;
4078 }
4079 
4080 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4081                                       DeviceState *dev, Error **errp)
4082 {
4083     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4084         spapr_memory_plug(hotplug_dev, dev, errp);
4085     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4086         spapr_core_plug(hotplug_dev, dev, errp);
4087     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4088         spapr_phb_plug(hotplug_dev, dev, errp);
4089     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4090         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4091     }
4092 }
4093 
4094 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4095                                         DeviceState *dev, Error **errp)
4096 {
4097     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4098         spapr_memory_unplug(hotplug_dev, dev);
4099     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4100         spapr_core_unplug(hotplug_dev, dev);
4101     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4102         spapr_phb_unplug(hotplug_dev, dev);
4103     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4104         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4105     }
4106 }
4107 
4108 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4109                                                 DeviceState *dev, Error **errp)
4110 {
4111     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4112     MachineClass *mc = MACHINE_GET_CLASS(sms);
4113     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4114 
4115     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4116         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4117             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4118         } else {
4119             /* NOTE: this means there is a window after guest reset, prior to
4120              * CAS negotiation, where unplug requests will fail due to the
4121              * capability not being detected yet. This is a bit different than
4122              * the case with PCI unplug, where the events will be queued and
4123              * eventually handled by the guest after boot
4124              */
4125             error_setg(errp, "Memory hot unplug not supported for this guest");
4126         }
4127     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4128         if (!mc->has_hotpluggable_cpus) {
4129             error_setg(errp, "CPU hot unplug not supported on this machine");
4130             return;
4131         }
4132         spapr_core_unplug_request(hotplug_dev, dev, errp);
4133     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4134         if (!smc->dr_phb_enabled) {
4135             error_setg(errp, "PHB hot unplug not supported on this machine");
4136             return;
4137         }
4138         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4139     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4140         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4141     }
4142 }
4143 
4144 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4145                                           DeviceState *dev, Error **errp)
4146 {
4147     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4148         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4149     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4150         spapr_core_pre_plug(hotplug_dev, dev, errp);
4151     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4152         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4153     }
4154 }
4155 
4156 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4157                                                  DeviceState *dev)
4158 {
4159     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4160         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4161         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4162         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4163         return HOTPLUG_HANDLER(machine);
4164     }
4165     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4166         PCIDevice *pcidev = PCI_DEVICE(dev);
4167         PCIBus *root = pci_device_root_bus(pcidev);
4168         SpaprPhbState *phb =
4169             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4170                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4171 
4172         if (phb) {
4173             return HOTPLUG_HANDLER(phb);
4174         }
4175     }
4176     return NULL;
4177 }
4178 
4179 static CpuInstanceProperties
4180 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4181 {
4182     CPUArchId *core_slot;
4183     MachineClass *mc = MACHINE_GET_CLASS(machine);
4184 
4185     /* make sure possible_cpu are intialized */
4186     mc->possible_cpu_arch_ids(machine);
4187     /* get CPU core slot containing thread that matches cpu_index */
4188     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4189     assert(core_slot);
4190     return core_slot->props;
4191 }
4192 
4193 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4194 {
4195     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4196 }
4197 
4198 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4199 {
4200     int i;
4201     unsigned int smp_threads = machine->smp.threads;
4202     unsigned int smp_cpus = machine->smp.cpus;
4203     const char *core_type;
4204     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4205     MachineClass *mc = MACHINE_GET_CLASS(machine);
4206 
4207     if (!mc->has_hotpluggable_cpus) {
4208         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4209     }
4210     if (machine->possible_cpus) {
4211         assert(machine->possible_cpus->len == spapr_max_cores);
4212         return machine->possible_cpus;
4213     }
4214 
4215     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4216     if (!core_type) {
4217         error_report("Unable to find sPAPR CPU Core definition");
4218         exit(1);
4219     }
4220 
4221     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4222                              sizeof(CPUArchId) * spapr_max_cores);
4223     machine->possible_cpus->len = spapr_max_cores;
4224     for (i = 0; i < machine->possible_cpus->len; i++) {
4225         int core_id = i * smp_threads;
4226 
4227         machine->possible_cpus->cpus[i].type = core_type;
4228         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4229         machine->possible_cpus->cpus[i].arch_id = core_id;
4230         machine->possible_cpus->cpus[i].props.has_core_id = true;
4231         machine->possible_cpus->cpus[i].props.core_id = core_id;
4232     }
4233     return machine->possible_cpus;
4234 }
4235 
4236 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4237                                 uint64_t *buid, hwaddr *pio,
4238                                 hwaddr *mmio32, hwaddr *mmio64,
4239                                 unsigned n_dma, uint32_t *liobns,
4240                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4241 {
4242     /*
4243      * New-style PHB window placement.
4244      *
4245      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4246      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4247      * windows.
4248      *
4249      * Some guest kernels can't work with MMIO windows above 1<<46
4250      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4251      *
4252      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4253      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4254      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4255      * 1TiB 64-bit MMIO windows for each PHB.
4256      */
4257     const uint64_t base_buid = 0x800000020000000ULL;
4258     int i;
4259 
4260     /* Sanity check natural alignments */
4261     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4262     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4263     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4264     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4265     /* Sanity check bounds */
4266     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4267                       SPAPR_PCI_MEM32_WIN_SIZE);
4268     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4269                       SPAPR_PCI_MEM64_WIN_SIZE);
4270 
4271     if (index >= SPAPR_MAX_PHBS) {
4272         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4273                    SPAPR_MAX_PHBS - 1);
4274         return;
4275     }
4276 
4277     *buid = base_buid + index;
4278     for (i = 0; i < n_dma; ++i) {
4279         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4280     }
4281 
4282     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4283     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4284     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4285 
4286     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4287     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4288 }
4289 
4290 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4291 {
4292     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4293 
4294     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4295 }
4296 
4297 static void spapr_ics_resend(XICSFabric *dev)
4298 {
4299     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4300 
4301     ics_resend(spapr->ics);
4302 }
4303 
4304 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4305 {
4306     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4307 
4308     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4309 }
4310 
4311 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4312                                  Monitor *mon)
4313 {
4314     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4315 
4316     spapr_irq_print_info(spapr, mon);
4317     monitor_printf(mon, "irqchip: %s\n",
4318                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4319 }
4320 
4321 /*
4322  * This is a XIVE only operation
4323  */
4324 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4325                            uint8_t nvt_blk, uint32_t nvt_idx,
4326                            bool cam_ignore, uint8_t priority,
4327                            uint32_t logic_serv, XiveTCTXMatch *match)
4328 {
4329     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4330     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4331     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4332     int count;
4333 
4334     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4335                            priority, logic_serv, match);
4336     if (count < 0) {
4337         return count;
4338     }
4339 
4340     /*
4341      * When we implement the save and restore of the thread interrupt
4342      * contexts in the enter/exit CPU handlers of the machine and the
4343      * escalations in QEMU, we should be able to handle non dispatched
4344      * vCPUs.
4345      *
4346      * Until this is done, the sPAPR machine should find at least one
4347      * matching context always.
4348      */
4349     if (count == 0) {
4350         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4351                       nvt_blk, nvt_idx);
4352     }
4353 
4354     return count;
4355 }
4356 
4357 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4358 {
4359     return cpu->vcpu_id;
4360 }
4361 
4362 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4363 {
4364     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4365     MachineState *ms = MACHINE(spapr);
4366     int vcpu_id;
4367 
4368     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4369 
4370     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4371         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4372         error_append_hint(errp, "Adjust the number of cpus to %d "
4373                           "or try to raise the number of threads per core\n",
4374                           vcpu_id * ms->smp.threads / spapr->vsmt);
4375         return;
4376     }
4377 
4378     cpu->vcpu_id = vcpu_id;
4379 }
4380 
4381 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4382 {
4383     CPUState *cs;
4384 
4385     CPU_FOREACH(cs) {
4386         PowerPCCPU *cpu = POWERPC_CPU(cs);
4387 
4388         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4389             return cpu;
4390         }
4391     }
4392 
4393     return NULL;
4394 }
4395 
4396 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4397 {
4398     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4399 
4400     /* These are only called by TCG, KVM maintains dispatch state */
4401 
4402     spapr_cpu->prod = false;
4403     if (spapr_cpu->vpa_addr) {
4404         CPUState *cs = CPU(cpu);
4405         uint32_t dispatch;
4406 
4407         dispatch = ldl_be_phys(cs->as,
4408                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4409         dispatch++;
4410         if ((dispatch & 1) != 0) {
4411             qemu_log_mask(LOG_GUEST_ERROR,
4412                           "VPA: incorrect dispatch counter value for "
4413                           "dispatched partition %u, correcting.\n", dispatch);
4414             dispatch++;
4415         }
4416         stl_be_phys(cs->as,
4417                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4418     }
4419 }
4420 
4421 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4422 {
4423     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4424 
4425     if (spapr_cpu->vpa_addr) {
4426         CPUState *cs = CPU(cpu);
4427         uint32_t dispatch;
4428 
4429         dispatch = ldl_be_phys(cs->as,
4430                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4431         dispatch++;
4432         if ((dispatch & 1) != 1) {
4433             qemu_log_mask(LOG_GUEST_ERROR,
4434                           "VPA: incorrect dispatch counter value for "
4435                           "preempted partition %u, correcting.\n", dispatch);
4436             dispatch++;
4437         }
4438         stl_be_phys(cs->as,
4439                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4440     }
4441 }
4442 
4443 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4444 {
4445     MachineClass *mc = MACHINE_CLASS(oc);
4446     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4447     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4448     NMIClass *nc = NMI_CLASS(oc);
4449     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4450     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4451     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4452     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4453     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4454 
4455     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4456     mc->ignore_boot_device_suffixes = true;
4457 
4458     /*
4459      * We set up the default / latest behaviour here.  The class_init
4460      * functions for the specific versioned machine types can override
4461      * these details for backwards compatibility
4462      */
4463     mc->init = spapr_machine_init;
4464     mc->reset = spapr_machine_reset;
4465     mc->block_default_type = IF_SCSI;
4466     mc->max_cpus = 1024;
4467     mc->no_parallel = 1;
4468     mc->default_boot_order = "";
4469     mc->default_ram_size = 512 * MiB;
4470     mc->default_ram_id = "ppc_spapr.ram";
4471     mc->default_display = "std";
4472     mc->kvm_type = spapr_kvm_type;
4473     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4474     mc->pci_allow_0_address = true;
4475     assert(!mc->get_hotplug_handler);
4476     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4477     hc->pre_plug = spapr_machine_device_pre_plug;
4478     hc->plug = spapr_machine_device_plug;
4479     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4480     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4481     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4482     hc->unplug_request = spapr_machine_device_unplug_request;
4483     hc->unplug = spapr_machine_device_unplug;
4484 
4485     smc->dr_lmb_enabled = true;
4486     smc->update_dt_enabled = true;
4487     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4488     mc->has_hotpluggable_cpus = true;
4489     mc->nvdimm_supported = true;
4490     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4491     fwc->get_dev_path = spapr_get_fw_dev_path;
4492     nc->nmi_monitor_handler = spapr_nmi;
4493     smc->phb_placement = spapr_phb_placement;
4494     vhc->hypercall = emulate_spapr_hypercall;
4495     vhc->hpt_mask = spapr_hpt_mask;
4496     vhc->map_hptes = spapr_map_hptes;
4497     vhc->unmap_hptes = spapr_unmap_hptes;
4498     vhc->hpte_set_c = spapr_hpte_set_c;
4499     vhc->hpte_set_r = spapr_hpte_set_r;
4500     vhc->get_pate = spapr_get_pate;
4501     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4502     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4503     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4504     xic->ics_get = spapr_ics_get;
4505     xic->ics_resend = spapr_ics_resend;
4506     xic->icp_get = spapr_icp_get;
4507     ispc->print_info = spapr_pic_print_info;
4508     /* Force NUMA node memory size to be a multiple of
4509      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4510      * in which LMBs are represented and hot-added
4511      */
4512     mc->numa_mem_align_shift = 28;
4513     mc->numa_mem_supported = true;
4514     mc->auto_enable_numa = true;
4515 
4516     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4517     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4518     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4519     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4520     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4521     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4522     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4523     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4524     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4525     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4526     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4527     spapr_caps_add_properties(smc);
4528     smc->irq = &spapr_irq_dual;
4529     smc->dr_phb_enabled = true;
4530     smc->linux_pci_probe = true;
4531     smc->smp_threads_vsmt = true;
4532     smc->nr_xirqs = SPAPR_NR_XIRQS;
4533     xfc->match_nvt = spapr_match_nvt;
4534 }
4535 
4536 static const TypeInfo spapr_machine_info = {
4537     .name          = TYPE_SPAPR_MACHINE,
4538     .parent        = TYPE_MACHINE,
4539     .abstract      = true,
4540     .instance_size = sizeof(SpaprMachineState),
4541     .instance_init = spapr_instance_init,
4542     .instance_finalize = spapr_machine_finalizefn,
4543     .class_size    = sizeof(SpaprMachineClass),
4544     .class_init    = spapr_machine_class_init,
4545     .interfaces = (InterfaceInfo[]) {
4546         { TYPE_FW_PATH_PROVIDER },
4547         { TYPE_NMI },
4548         { TYPE_HOTPLUG_HANDLER },
4549         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4550         { TYPE_XICS_FABRIC },
4551         { TYPE_INTERRUPT_STATS_PROVIDER },
4552         { TYPE_XIVE_FABRIC },
4553         { }
4554     },
4555 };
4556 
4557 static void spapr_machine_latest_class_options(MachineClass *mc)
4558 {
4559     mc->alias = "pseries";
4560     mc->is_default = true;
4561 }
4562 
4563 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4564     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4565                                                     void *data)      \
4566     {                                                                \
4567         MachineClass *mc = MACHINE_CLASS(oc);                        \
4568         spapr_machine_##suffix##_class_options(mc);                  \
4569         if (latest) {                                                \
4570             spapr_machine_latest_class_options(mc);                  \
4571         }                                                            \
4572     }                                                                \
4573     static const TypeInfo spapr_machine_##suffix##_info = {          \
4574         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4575         .parent = TYPE_SPAPR_MACHINE,                                \
4576         .class_init = spapr_machine_##suffix##_class_init,           \
4577     };                                                               \
4578     static void spapr_machine_register_##suffix(void)                \
4579     {                                                                \
4580         type_register(&spapr_machine_##suffix##_info);               \
4581     }                                                                \
4582     type_init(spapr_machine_register_##suffix)
4583 
4584 /*
4585  * pseries-5.1
4586  */
4587 static void spapr_machine_5_1_class_options(MachineClass *mc)
4588 {
4589     /* Defaults for the latest behaviour inherited from the base class */
4590 }
4591 
4592 DEFINE_SPAPR_MACHINE(5_1, "5.1", true);
4593 
4594 /*
4595  * pseries-5.0
4596  */
4597 static void spapr_machine_5_0_class_options(MachineClass *mc)
4598 {
4599     spapr_machine_5_1_class_options(mc);
4600     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4601 }
4602 
4603 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4604 
4605 /*
4606  * pseries-4.2
4607  */
4608 static void spapr_machine_4_2_class_options(MachineClass *mc)
4609 {
4610     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4611 
4612     spapr_machine_5_0_class_options(mc);
4613     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4614     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4615     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4616     smc->rma_limit = 16 * GiB;
4617     mc->nvdimm_supported = false;
4618 }
4619 
4620 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4621 
4622 /*
4623  * pseries-4.1
4624  */
4625 static void spapr_machine_4_1_class_options(MachineClass *mc)
4626 {
4627     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4628     static GlobalProperty compat[] = {
4629         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4630         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4631     };
4632 
4633     spapr_machine_4_2_class_options(mc);
4634     smc->linux_pci_probe = false;
4635     smc->smp_threads_vsmt = false;
4636     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4637     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4638 }
4639 
4640 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4641 
4642 /*
4643  * pseries-4.0
4644  */
4645 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4646                               uint64_t *buid, hwaddr *pio,
4647                               hwaddr *mmio32, hwaddr *mmio64,
4648                               unsigned n_dma, uint32_t *liobns,
4649                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4650 {
4651     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4652                         nv2gpa, nv2atsd, errp);
4653     *nv2gpa = 0;
4654     *nv2atsd = 0;
4655 }
4656 
4657 static void spapr_machine_4_0_class_options(MachineClass *mc)
4658 {
4659     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4660 
4661     spapr_machine_4_1_class_options(mc);
4662     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4663     smc->phb_placement = phb_placement_4_0;
4664     smc->irq = &spapr_irq_xics;
4665     smc->pre_4_1_migration = true;
4666 }
4667 
4668 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4669 
4670 /*
4671  * pseries-3.1
4672  */
4673 static void spapr_machine_3_1_class_options(MachineClass *mc)
4674 {
4675     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4676 
4677     spapr_machine_4_0_class_options(mc);
4678     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4679 
4680     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4681     smc->update_dt_enabled = false;
4682     smc->dr_phb_enabled = false;
4683     smc->broken_host_serial_model = true;
4684     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4685     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4686     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4687     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4688 }
4689 
4690 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4691 
4692 /*
4693  * pseries-3.0
4694  */
4695 
4696 static void spapr_machine_3_0_class_options(MachineClass *mc)
4697 {
4698     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4699 
4700     spapr_machine_3_1_class_options(mc);
4701     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4702 
4703     smc->legacy_irq_allocation = true;
4704     smc->nr_xirqs = 0x400;
4705     smc->irq = &spapr_irq_xics_legacy;
4706 }
4707 
4708 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4709 
4710 /*
4711  * pseries-2.12
4712  */
4713 static void spapr_machine_2_12_class_options(MachineClass *mc)
4714 {
4715     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4716     static GlobalProperty compat[] = {
4717         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4718         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4719     };
4720 
4721     spapr_machine_3_0_class_options(mc);
4722     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4723     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4724 
4725     /* We depend on kvm_enabled() to choose a default value for the
4726      * hpt-max-page-size capability. Of course we can't do it here
4727      * because this is too early and the HW accelerator isn't initialzed
4728      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4729      */
4730     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4731 }
4732 
4733 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4734 
4735 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4736 {
4737     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4738 
4739     spapr_machine_2_12_class_options(mc);
4740     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4741     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4742     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4743 }
4744 
4745 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4746 
4747 /*
4748  * pseries-2.11
4749  */
4750 
4751 static void spapr_machine_2_11_class_options(MachineClass *mc)
4752 {
4753     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4754 
4755     spapr_machine_2_12_class_options(mc);
4756     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4757     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4758 }
4759 
4760 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4761 
4762 /*
4763  * pseries-2.10
4764  */
4765 
4766 static void spapr_machine_2_10_class_options(MachineClass *mc)
4767 {
4768     spapr_machine_2_11_class_options(mc);
4769     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4770 }
4771 
4772 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4773 
4774 /*
4775  * pseries-2.9
4776  */
4777 
4778 static void spapr_machine_2_9_class_options(MachineClass *mc)
4779 {
4780     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4781     static GlobalProperty compat[] = {
4782         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4783     };
4784 
4785     spapr_machine_2_10_class_options(mc);
4786     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4787     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4788     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4789     smc->pre_2_10_has_unused_icps = true;
4790     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4791 }
4792 
4793 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4794 
4795 /*
4796  * pseries-2.8
4797  */
4798 
4799 static void spapr_machine_2_8_class_options(MachineClass *mc)
4800 {
4801     static GlobalProperty compat[] = {
4802         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4803     };
4804 
4805     spapr_machine_2_9_class_options(mc);
4806     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4807     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4808     mc->numa_mem_align_shift = 23;
4809 }
4810 
4811 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4812 
4813 /*
4814  * pseries-2.7
4815  */
4816 
4817 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4818                               uint64_t *buid, hwaddr *pio,
4819                               hwaddr *mmio32, hwaddr *mmio64,
4820                               unsigned n_dma, uint32_t *liobns,
4821                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4822 {
4823     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4824     const uint64_t base_buid = 0x800000020000000ULL;
4825     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4826     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4827     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4828     const uint32_t max_index = 255;
4829     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4830 
4831     uint64_t ram_top = MACHINE(spapr)->ram_size;
4832     hwaddr phb0_base, phb_base;
4833     int i;
4834 
4835     /* Do we have device memory? */
4836     if (MACHINE(spapr)->maxram_size > ram_top) {
4837         /* Can't just use maxram_size, because there may be an
4838          * alignment gap between normal and device memory regions
4839          */
4840         ram_top = MACHINE(spapr)->device_memory->base +
4841             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4842     }
4843 
4844     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4845 
4846     if (index > max_index) {
4847         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4848                    max_index);
4849         return;
4850     }
4851 
4852     *buid = base_buid + index;
4853     for (i = 0; i < n_dma; ++i) {
4854         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4855     }
4856 
4857     phb_base = phb0_base + index * phb_spacing;
4858     *pio = phb_base + pio_offset;
4859     *mmio32 = phb_base + mmio_offset;
4860     /*
4861      * We don't set the 64-bit MMIO window, relying on the PHB's
4862      * fallback behaviour of automatically splitting a large "32-bit"
4863      * window into contiguous 32-bit and 64-bit windows
4864      */
4865 
4866     *nv2gpa = 0;
4867     *nv2atsd = 0;
4868 }
4869 
4870 static void spapr_machine_2_7_class_options(MachineClass *mc)
4871 {
4872     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4873     static GlobalProperty compat[] = {
4874         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4875         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4876         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4877         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4878     };
4879 
4880     spapr_machine_2_8_class_options(mc);
4881     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4882     mc->default_machine_opts = "modern-hotplug-events=off";
4883     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4884     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4885     smc->phb_placement = phb_placement_2_7;
4886 }
4887 
4888 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4889 
4890 /*
4891  * pseries-2.6
4892  */
4893 
4894 static void spapr_machine_2_6_class_options(MachineClass *mc)
4895 {
4896     static GlobalProperty compat[] = {
4897         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4898     };
4899 
4900     spapr_machine_2_7_class_options(mc);
4901     mc->has_hotpluggable_cpus = false;
4902     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4903     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4904 }
4905 
4906 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4907 
4908 /*
4909  * pseries-2.5
4910  */
4911 
4912 static void spapr_machine_2_5_class_options(MachineClass *mc)
4913 {
4914     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4915     static GlobalProperty compat[] = {
4916         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4917     };
4918 
4919     spapr_machine_2_6_class_options(mc);
4920     smc->use_ohci_by_default = true;
4921     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4922     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4923 }
4924 
4925 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4926 
4927 /*
4928  * pseries-2.4
4929  */
4930 
4931 static void spapr_machine_2_4_class_options(MachineClass *mc)
4932 {
4933     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4934 
4935     spapr_machine_2_5_class_options(mc);
4936     smc->dr_lmb_enabled = false;
4937     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4938 }
4939 
4940 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4941 
4942 /*
4943  * pseries-2.3
4944  */
4945 
4946 static void spapr_machine_2_3_class_options(MachineClass *mc)
4947 {
4948     static GlobalProperty compat[] = {
4949         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4950     };
4951     spapr_machine_2_4_class_options(mc);
4952     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4953     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4954 }
4955 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4956 
4957 /*
4958  * pseries-2.2
4959  */
4960 
4961 static void spapr_machine_2_2_class_options(MachineClass *mc)
4962 {
4963     static GlobalProperty compat[] = {
4964         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4965     };
4966 
4967     spapr_machine_2_3_class_options(mc);
4968     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4969     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4970     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4971 }
4972 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4973 
4974 /*
4975  * pseries-2.1
4976  */
4977 
4978 static void spapr_machine_2_1_class_options(MachineClass *mc)
4979 {
4980     spapr_machine_2_2_class_options(mc);
4981     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4982 }
4983 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4984 
4985 static void spapr_machine_register_types(void)
4986 {
4987     type_register_static(&spapr_machine_info);
4988 }
4989 
4990 type_init(spapr_machine_register_types)
4991