xref: /qemu/hw/ppc/spapr.c (revision aebb9b9cb2738c2a0b21c7315822fac7d9bb5004)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qemu/datadir.h"
30 #include "qapi/error.h"
31 #include "qapi/qapi-events-machine.h"
32 #include "qapi/qapi-events-qdev.h"
33 #include "qapi/visitor.h"
34 #include "sysemu/sysemu.h"
35 #include "sysemu/hostmem.h"
36 #include "sysemu/numa.h"
37 #include "sysemu/qtest.h"
38 #include "sysemu/reset.h"
39 #include "sysemu/runstate.h"
40 #include "qemu/log.h"
41 #include "hw/fw-path-provider.h"
42 #include "elf.h"
43 #include "net/net.h"
44 #include "sysemu/device_tree.h"
45 #include "sysemu/cpus.h"
46 #include "sysemu/hw_accel.h"
47 #include "kvm_ppc.h"
48 #include "migration/misc.h"
49 #include "migration/qemu-file-types.h"
50 #include "migration/global_state.h"
51 #include "migration/register.h"
52 #include "migration/blocker.h"
53 #include "mmu-hash64.h"
54 #include "mmu-book3s-v3.h"
55 #include "cpu-models.h"
56 #include "hw/core/cpu.h"
57 
58 #include "hw/ppc/ppc.h"
59 #include "hw/loader.h"
60 
61 #include "hw/ppc/fdt.h"
62 #include "hw/ppc/spapr.h"
63 #include "hw/ppc/spapr_vio.h"
64 #include "hw/qdev-properties.h"
65 #include "hw/pci-host/spapr.h"
66 #include "hw/pci/msi.h"
67 
68 #include "hw/pci/pci.h"
69 #include "hw/scsi/scsi.h"
70 #include "hw/virtio/virtio-scsi.h"
71 #include "hw/virtio/vhost-scsi-common.h"
72 
73 #include "exec/ram_addr.h"
74 #include "hw/usb.h"
75 #include "qemu/config-file.h"
76 #include "qemu/error-report.h"
77 #include "trace.h"
78 #include "hw/nmi.h"
79 #include "hw/intc/intc.h"
80 
81 #include "hw/ppc/spapr_cpu_core.h"
82 #include "hw/mem/memory-device.h"
83 #include "hw/ppc/spapr_tpm_proxy.h"
84 #include "hw/ppc/spapr_nvdimm.h"
85 #include "hw/ppc/spapr_numa.h"
86 #include "hw/ppc/pef.h"
87 
88 #include "monitor/monitor.h"
89 
90 #include <libfdt.h>
91 
92 /* SLOF memory layout:
93  *
94  * SLOF raw image loaded at 0, copies its romfs right below the flat
95  * device-tree, then position SLOF itself 31M below that
96  *
97  * So we set FW_OVERHEAD to 40MB which should account for all of that
98  * and more
99  *
100  * We load our kernel at 4M, leaving space for SLOF initial image
101  */
102 #define FDT_MAX_ADDR            0x80000000 /* FDT must stay below that */
103 #define FW_MAX_SIZE             0x400000
104 #define FW_FILE_NAME            "slof.bin"
105 #define FW_FILE_NAME_VOF        "vof.bin"
106 #define FW_OVERHEAD             0x2800000
107 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
108 
109 #define MIN_RMA_SLOF            (128 * MiB)
110 
111 #define PHANDLE_INTC            0x00001111
112 
113 /* These two functions implement the VCPU id numbering: one to compute them
114  * all and one to identify thread 0 of a VCORE. Any change to the first one
115  * is likely to have an impact on the second one, so let's keep them close.
116  */
117 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
118 {
119     MachineState *ms = MACHINE(spapr);
120     unsigned int smp_threads = ms->smp.threads;
121 
122     assert(spapr->vsmt);
123     return
124         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
125 }
126 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
127                                       PowerPCCPU *cpu)
128 {
129     assert(spapr->vsmt);
130     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
131 }
132 
133 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
134 {
135     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
136      * and newer QEMUs don't even have them. In both cases, we don't want
137      * to send anything on the wire.
138      */
139     return false;
140 }
141 
142 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
143     .name = "icp/server",
144     .version_id = 1,
145     .minimum_version_id = 1,
146     .needed = pre_2_10_vmstate_dummy_icp_needed,
147     .fields = (VMStateField[]) {
148         VMSTATE_UNUSED(4), /* uint32_t xirr */
149         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
150         VMSTATE_UNUSED(1), /* uint8_t mfrr */
151         VMSTATE_END_OF_LIST()
152     },
153 };
154 
155 static void pre_2_10_vmstate_register_dummy_icp(int i)
156 {
157     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
158                      (void *)(uintptr_t) i);
159 }
160 
161 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
162 {
163     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
164                        (void *)(uintptr_t) i);
165 }
166 
167 int spapr_max_server_number(SpaprMachineState *spapr)
168 {
169     MachineState *ms = MACHINE(spapr);
170 
171     assert(spapr->vsmt);
172     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
173 }
174 
175 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
176                                   int smt_threads)
177 {
178     int i, ret = 0;
179     uint32_t servers_prop[smt_threads];
180     uint32_t gservers_prop[smt_threads * 2];
181     int index = spapr_get_vcpu_id(cpu);
182 
183     if (cpu->compat_pvr) {
184         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
185         if (ret < 0) {
186             return ret;
187         }
188     }
189 
190     /* Build interrupt servers and gservers properties */
191     for (i = 0; i < smt_threads; i++) {
192         servers_prop[i] = cpu_to_be32(index + i);
193         /* Hack, direct the group queues back to cpu 0 */
194         gservers_prop[i*2] = cpu_to_be32(index + i);
195         gservers_prop[i*2 + 1] = 0;
196     }
197     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
198                       servers_prop, sizeof(servers_prop));
199     if (ret < 0) {
200         return ret;
201     }
202     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
203                       gservers_prop, sizeof(gservers_prop));
204 
205     return ret;
206 }
207 
208 static void spapr_dt_pa_features(SpaprMachineState *spapr,
209                                  PowerPCCPU *cpu,
210                                  void *fdt, int offset)
211 {
212     uint8_t pa_features_206[] = { 6, 0,
213         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214     uint8_t pa_features_207[] = { 24, 0,
215         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219     uint8_t pa_features_300[] = { 66, 0,
220         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
223         /* 6: DS207 */
224         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
225         /* 16: Vector */
226         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235         /* 42: PM, 44: PC RA, 46: SC vec'd */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237         /* 48: SIMD, 50: QP BFP, 52: String */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239         /* 54: DecFP, 56: DecI, 58: SHA */
240         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241         /* 60: NM atomic, 62: RNG */
242         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
243     };
244     uint8_t *pa_features = NULL;
245     size_t pa_size;
246 
247     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
248         pa_features = pa_features_206;
249         pa_size = sizeof(pa_features_206);
250     }
251     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
252         pa_features = pa_features_207;
253         pa_size = sizeof(pa_features_207);
254     }
255     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
256         pa_features = pa_features_300;
257         pa_size = sizeof(pa_features_300);
258     }
259     if (!pa_features) {
260         return;
261     }
262 
263     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
264         /*
265          * Note: we keep CI large pages off by default because a 64K capable
266          * guest provisioned with large pages might otherwise try to map a qemu
267          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268          * even if that qemu runs on a 4k host.
269          * We dd this bit back here if we are confident this is not an issue
270          */
271         pa_features[3] |= 0x20;
272     }
273     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
274         pa_features[24] |= 0x80;    /* Transactional memory support */
275     }
276     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
277         /* Workaround for broken kernels that attempt (guest) radix
278          * mode when they can't handle it, if they see the radix bit set
279          * in pa-features. So hide it from them. */
280         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
281     }
282 
283     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
284 }
285 
286 static hwaddr spapr_node0_size(MachineState *machine)
287 {
288     if (machine->numa_state->num_nodes) {
289         int i;
290         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
291             if (machine->numa_state->nodes[i].node_mem) {
292                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
293                            machine->ram_size);
294             }
295         }
296     }
297     return machine->ram_size;
298 }
299 
300 static void add_str(GString *s, const gchar *s1)
301 {
302     g_string_append_len(s, s1, strlen(s1) + 1);
303 }
304 
305 static int spapr_dt_memory_node(SpaprMachineState *spapr, void *fdt, int nodeid,
306                                 hwaddr start, hwaddr size)
307 {
308     char mem_name[32];
309     uint64_t mem_reg_property[2];
310     int off;
311 
312     mem_reg_property[0] = cpu_to_be64(start);
313     mem_reg_property[1] = cpu_to_be64(size);
314 
315     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
316     off = fdt_add_subnode(fdt, 0, mem_name);
317     _FDT(off);
318     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
319     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
320                       sizeof(mem_reg_property))));
321     spapr_numa_write_associativity_dt(spapr, fdt, off, nodeid);
322     return off;
323 }
324 
325 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
326 {
327     MemoryDeviceInfoList *info;
328 
329     for (info = list; info; info = info->next) {
330         MemoryDeviceInfo *value = info->value;
331 
332         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
333             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
334 
335             if (addr >= pcdimm_info->addr &&
336                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
337                 return pcdimm_info->node;
338             }
339         }
340     }
341 
342     return -1;
343 }
344 
345 struct sPAPRDrconfCellV2 {
346      uint32_t seq_lmbs;
347      uint64_t base_addr;
348      uint32_t drc_index;
349      uint32_t aa_index;
350      uint32_t flags;
351 } QEMU_PACKED;
352 
353 typedef struct DrconfCellQueue {
354     struct sPAPRDrconfCellV2 cell;
355     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
356 } DrconfCellQueue;
357 
358 static DrconfCellQueue *
359 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
360                       uint32_t drc_index, uint32_t aa_index,
361                       uint32_t flags)
362 {
363     DrconfCellQueue *elem;
364 
365     elem = g_malloc0(sizeof(*elem));
366     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
367     elem->cell.base_addr = cpu_to_be64(base_addr);
368     elem->cell.drc_index = cpu_to_be32(drc_index);
369     elem->cell.aa_index = cpu_to_be32(aa_index);
370     elem->cell.flags = cpu_to_be32(flags);
371 
372     return elem;
373 }
374 
375 static int spapr_dt_dynamic_memory_v2(SpaprMachineState *spapr, void *fdt,
376                                       int offset, MemoryDeviceInfoList *dimms)
377 {
378     MachineState *machine = MACHINE(spapr);
379     uint8_t *int_buf, *cur_index;
380     int ret;
381     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
382     uint64_t addr, cur_addr, size;
383     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
384     uint64_t mem_end = machine->device_memory->base +
385                        memory_region_size(&machine->device_memory->mr);
386     uint32_t node, buf_len, nr_entries = 0;
387     SpaprDrc *drc;
388     DrconfCellQueue *elem, *next;
389     MemoryDeviceInfoList *info;
390     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
391         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
392 
393     /* Entry to cover RAM and the gap area */
394     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
395                                  SPAPR_LMB_FLAGS_RESERVED |
396                                  SPAPR_LMB_FLAGS_DRC_INVALID);
397     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
398     nr_entries++;
399 
400     cur_addr = machine->device_memory->base;
401     for (info = dimms; info; info = info->next) {
402         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
403 
404         addr = di->addr;
405         size = di->size;
406         node = di->node;
407 
408         /*
409          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
410          * area is marked hotpluggable in the next iteration for the bigger
411          * chunk including the NVDIMM occupied area.
412          */
413         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
414             continue;
415 
416         /* Entry for hot-pluggable area */
417         if (cur_addr < addr) {
418             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
419             g_assert(drc);
420             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
421                                          cur_addr, spapr_drc_index(drc), -1, 0);
422             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
423             nr_entries++;
424         }
425 
426         /* Entry for DIMM */
427         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
428         g_assert(drc);
429         elem = spapr_get_drconf_cell(size / lmb_size, addr,
430                                      spapr_drc_index(drc), node,
431                                      (SPAPR_LMB_FLAGS_ASSIGNED |
432                                       SPAPR_LMB_FLAGS_HOTREMOVABLE));
433         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
434         nr_entries++;
435         cur_addr = addr + size;
436     }
437 
438     /* Entry for remaining hotpluggable area */
439     if (cur_addr < mem_end) {
440         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
441         g_assert(drc);
442         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
443                                      cur_addr, spapr_drc_index(drc), -1, 0);
444         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
445         nr_entries++;
446     }
447 
448     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
449     int_buf = cur_index = g_malloc0(buf_len);
450     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
451     cur_index += sizeof(nr_entries);
452 
453     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
454         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
455         cur_index += sizeof(elem->cell);
456         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
457         g_free(elem);
458     }
459 
460     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
461     g_free(int_buf);
462     if (ret < 0) {
463         return -1;
464     }
465     return 0;
466 }
467 
468 static int spapr_dt_dynamic_memory(SpaprMachineState *spapr, void *fdt,
469                                    int offset, MemoryDeviceInfoList *dimms)
470 {
471     MachineState *machine = MACHINE(spapr);
472     int i, ret;
473     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
474     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
475     uint32_t nr_lmbs = (machine->device_memory->base +
476                        memory_region_size(&machine->device_memory->mr)) /
477                        lmb_size;
478     uint32_t *int_buf, *cur_index, buf_len;
479 
480     /*
481      * Allocate enough buffer size to fit in ibm,dynamic-memory
482      */
483     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
484     cur_index = int_buf = g_malloc0(buf_len);
485     int_buf[0] = cpu_to_be32(nr_lmbs);
486     cur_index++;
487     for (i = 0; i < nr_lmbs; i++) {
488         uint64_t addr = i * lmb_size;
489         uint32_t *dynamic_memory = cur_index;
490 
491         if (i >= device_lmb_start) {
492             SpaprDrc *drc;
493 
494             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
495             g_assert(drc);
496 
497             dynamic_memory[0] = cpu_to_be32(addr >> 32);
498             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
499             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
500             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
501             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
502             if (memory_region_present(get_system_memory(), addr)) {
503                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
504             } else {
505                 dynamic_memory[5] = cpu_to_be32(0);
506             }
507         } else {
508             /*
509              * LMB information for RMA, boot time RAM and gap b/n RAM and
510              * device memory region -- all these are marked as reserved
511              * and as having no valid DRC.
512              */
513             dynamic_memory[0] = cpu_to_be32(addr >> 32);
514             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
515             dynamic_memory[2] = cpu_to_be32(0);
516             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
517             dynamic_memory[4] = cpu_to_be32(-1);
518             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
519                                             SPAPR_LMB_FLAGS_DRC_INVALID);
520         }
521 
522         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
523     }
524     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
525     g_free(int_buf);
526     if (ret < 0) {
527         return -1;
528     }
529     return 0;
530 }
531 
532 /*
533  * Adds ibm,dynamic-reconfiguration-memory node.
534  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
535  * of this device tree node.
536  */
537 static int spapr_dt_dynamic_reconfiguration_memory(SpaprMachineState *spapr,
538                                                    void *fdt)
539 {
540     MachineState *machine = MACHINE(spapr);
541     int ret, offset;
542     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
543     uint32_t prop_lmb_size[] = {cpu_to_be32(lmb_size >> 32),
544                                 cpu_to_be32(lmb_size & 0xffffffff)};
545     MemoryDeviceInfoList *dimms = NULL;
546 
547     /*
548      * Don't create the node if there is no device memory
549      */
550     if (machine->ram_size == machine->maxram_size) {
551         return 0;
552     }
553 
554     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
555 
556     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
557                     sizeof(prop_lmb_size));
558     if (ret < 0) {
559         return ret;
560     }
561 
562     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
563     if (ret < 0) {
564         return ret;
565     }
566 
567     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
568     if (ret < 0) {
569         return ret;
570     }
571 
572     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
573     dimms = qmp_memory_device_list();
574     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
575         ret = spapr_dt_dynamic_memory_v2(spapr, fdt, offset, dimms);
576     } else {
577         ret = spapr_dt_dynamic_memory(spapr, fdt, offset, dimms);
578     }
579     qapi_free_MemoryDeviceInfoList(dimms);
580 
581     if (ret < 0) {
582         return ret;
583     }
584 
585     ret = spapr_numa_write_assoc_lookup_arrays(spapr, fdt, offset);
586 
587     return ret;
588 }
589 
590 static int spapr_dt_memory(SpaprMachineState *spapr, void *fdt)
591 {
592     MachineState *machine = MACHINE(spapr);
593     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
594     hwaddr mem_start, node_size;
595     int i, nb_nodes = machine->numa_state->num_nodes;
596     NodeInfo *nodes = machine->numa_state->nodes;
597 
598     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
599         if (!nodes[i].node_mem) {
600             continue;
601         }
602         if (mem_start >= machine->ram_size) {
603             node_size = 0;
604         } else {
605             node_size = nodes[i].node_mem;
606             if (node_size > machine->ram_size - mem_start) {
607                 node_size = machine->ram_size - mem_start;
608             }
609         }
610         if (!mem_start) {
611             /* spapr_machine_init() checks for rma_size <= node0_size
612              * already */
613             spapr_dt_memory_node(spapr, fdt, i, 0, spapr->rma_size);
614             mem_start += spapr->rma_size;
615             node_size -= spapr->rma_size;
616         }
617         for ( ; node_size; ) {
618             hwaddr sizetmp = pow2floor(node_size);
619 
620             /* mem_start != 0 here */
621             if (ctzl(mem_start) < ctzl(sizetmp)) {
622                 sizetmp = 1ULL << ctzl(mem_start);
623             }
624 
625             spapr_dt_memory_node(spapr, fdt, i, mem_start, sizetmp);
626             node_size -= sizetmp;
627             mem_start += sizetmp;
628         }
629     }
630 
631     /* Generate ibm,dynamic-reconfiguration-memory node if required */
632     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRCONF_MEMORY)) {
633         int ret;
634 
635         g_assert(smc->dr_lmb_enabled);
636         ret = spapr_dt_dynamic_reconfiguration_memory(spapr, fdt);
637         if (ret) {
638             return ret;
639         }
640     }
641 
642     return 0;
643 }
644 
645 static void spapr_dt_cpu(CPUState *cs, void *fdt, int offset,
646                          SpaprMachineState *spapr)
647 {
648     MachineState *ms = MACHINE(spapr);
649     PowerPCCPU *cpu = POWERPC_CPU(cs);
650     CPUPPCState *env = &cpu->env;
651     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
652     int index = spapr_get_vcpu_id(cpu);
653     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
654                        0xffffffff, 0xffffffff};
655     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
656         : SPAPR_TIMEBASE_FREQ;
657     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
658     uint32_t page_sizes_prop[64];
659     size_t page_sizes_prop_size;
660     unsigned int smp_threads = ms->smp.threads;
661     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
662     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
663     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
664     SpaprDrc *drc;
665     int drc_index;
666     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
667     int i;
668 
669     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
670     if (drc) {
671         drc_index = spapr_drc_index(drc);
672         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
673     }
674 
675     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
676     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
677 
678     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
679     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
680                            env->dcache_line_size)));
681     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
682                            env->dcache_line_size)));
683     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
684                            env->icache_line_size)));
685     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
686                            env->icache_line_size)));
687 
688     if (pcc->l1_dcache_size) {
689         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
690                                pcc->l1_dcache_size)));
691     } else {
692         warn_report("Unknown L1 dcache size for cpu");
693     }
694     if (pcc->l1_icache_size) {
695         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
696                                pcc->l1_icache_size)));
697     } else {
698         warn_report("Unknown L1 icache size for cpu");
699     }
700 
701     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
702     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
703     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
704     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
705     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
706     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
707 
708     if (ppc_has_spr(cpu, SPR_PURR)) {
709         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
710     }
711     if (ppc_has_spr(cpu, SPR_PURR)) {
712         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
713     }
714 
715     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
716         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
717                           segs, sizeof(segs))));
718     }
719 
720     /* Advertise VSX (vector extensions) if available
721      *   1               == VMX / Altivec available
722      *   2               == VSX available
723      *
724      * Only CPUs for which we create core types in spapr_cpu_core.c
725      * are possible, and all of those have VMX */
726     if (env->insns_flags & PPC_ALTIVEC) {
727         if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
728             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
729         } else {
730             _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
731         }
732     }
733 
734     /* Advertise DFP (Decimal Floating Point) if available
735      *   0 / no property == no DFP
736      *   1               == DFP available */
737     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
738         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
739     }
740 
741     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
742                                                       sizeof(page_sizes_prop));
743     if (page_sizes_prop_size) {
744         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
745                           page_sizes_prop, page_sizes_prop_size)));
746     }
747 
748     spapr_dt_pa_features(spapr, cpu, fdt, offset);
749 
750     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
751                            cs->cpu_index / vcpus_per_socket)));
752 
753     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
754                       pft_size_prop, sizeof(pft_size_prop))));
755 
756     if (ms->numa_state->num_nodes > 1) {
757         _FDT(spapr_numa_fixup_cpu_dt(spapr, fdt, offset, cpu));
758     }
759 
760     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
761 
762     if (pcc->radix_page_info) {
763         for (i = 0; i < pcc->radix_page_info->count; i++) {
764             radix_AP_encodings[i] =
765                 cpu_to_be32(pcc->radix_page_info->entries[i]);
766         }
767         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
768                           radix_AP_encodings,
769                           pcc->radix_page_info->count *
770                           sizeof(radix_AP_encodings[0]))));
771     }
772 
773     /*
774      * We set this property to let the guest know that it can use the large
775      * decrementer and its width in bits.
776      */
777     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
778         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
779                               pcc->lrg_decr_bits)));
780 }
781 
782 static void spapr_dt_cpus(void *fdt, SpaprMachineState *spapr)
783 {
784     CPUState **rev;
785     CPUState *cs;
786     int n_cpus;
787     int cpus_offset;
788     int i;
789 
790     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
791     _FDT(cpus_offset);
792     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
793     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
794 
795     /*
796      * We walk the CPUs in reverse order to ensure that CPU DT nodes
797      * created by fdt_add_subnode() end up in the right order in FDT
798      * for the guest kernel the enumerate the CPUs correctly.
799      *
800      * The CPU list cannot be traversed in reverse order, so we need
801      * to do extra work.
802      */
803     n_cpus = 0;
804     rev = NULL;
805     CPU_FOREACH(cs) {
806         rev = g_renew(CPUState *, rev, n_cpus + 1);
807         rev[n_cpus++] = cs;
808     }
809 
810     for (i = n_cpus - 1; i >= 0; i--) {
811         CPUState *cs = rev[i];
812         PowerPCCPU *cpu = POWERPC_CPU(cs);
813         int index = spapr_get_vcpu_id(cpu);
814         DeviceClass *dc = DEVICE_GET_CLASS(cs);
815         g_autofree char *nodename = NULL;
816         int offset;
817 
818         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
819             continue;
820         }
821 
822         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
823         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
824         _FDT(offset);
825         spapr_dt_cpu(cs, fdt, offset, spapr);
826     }
827 
828     g_free(rev);
829 }
830 
831 static int spapr_dt_rng(void *fdt)
832 {
833     int node;
834     int ret;
835 
836     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
837     if (node <= 0) {
838         return -1;
839     }
840     ret = fdt_setprop_string(fdt, node, "device_type",
841                              "ibm,platform-facilities");
842     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
843     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
844 
845     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
846     if (node <= 0) {
847         return -1;
848     }
849     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
850 
851     return ret ? -1 : 0;
852 }
853 
854 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
855 {
856     MachineState *ms = MACHINE(spapr);
857     int rtas;
858     GString *hypertas = g_string_sized_new(256);
859     GString *qemu_hypertas = g_string_sized_new(256);
860     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
861         memory_region_size(&MACHINE(spapr)->device_memory->mr);
862     uint32_t lrdr_capacity[] = {
863         cpu_to_be32(max_device_addr >> 32),
864         cpu_to_be32(max_device_addr & 0xffffffff),
865         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE >> 32),
866         cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE & 0xffffffff),
867         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
868     };
869 
870     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
871 
872     /* hypertas */
873     add_str(hypertas, "hcall-pft");
874     add_str(hypertas, "hcall-term");
875     add_str(hypertas, "hcall-dabr");
876     add_str(hypertas, "hcall-interrupt");
877     add_str(hypertas, "hcall-tce");
878     add_str(hypertas, "hcall-vio");
879     add_str(hypertas, "hcall-splpar");
880     add_str(hypertas, "hcall-join");
881     add_str(hypertas, "hcall-bulk");
882     add_str(hypertas, "hcall-set-mode");
883     add_str(hypertas, "hcall-sprg0");
884     add_str(hypertas, "hcall-copy");
885     add_str(hypertas, "hcall-debug");
886     add_str(hypertas, "hcall-vphn");
887     if (spapr_get_cap(spapr, SPAPR_CAP_RPT_INVALIDATE) == SPAPR_CAP_ON) {
888         add_str(hypertas, "hcall-rpt-invalidate");
889     }
890 
891     add_str(qemu_hypertas, "hcall-memop1");
892 
893     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
894         add_str(hypertas, "hcall-multi-tce");
895     }
896 
897     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
898         add_str(hypertas, "hcall-hpt-resize");
899     }
900 
901     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
902                      hypertas->str, hypertas->len));
903     g_string_free(hypertas, TRUE);
904     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
905                      qemu_hypertas->str, qemu_hypertas->len));
906     g_string_free(qemu_hypertas, TRUE);
907 
908     spapr_numa_write_rtas_dt(spapr, fdt, rtas);
909 
910     /*
911      * FWNMI reserves RTAS_ERROR_LOG_MAX for the machine check error log,
912      * and 16 bytes per CPU for system reset error log plus an extra 8 bytes.
913      *
914      * The system reset requirements are driven by existing Linux and PowerVM
915      * implementation which (contrary to PAPR) saves r3 in the error log
916      * structure like machine check, so Linux expects to find the saved r3
917      * value at the address in r3 upon FWNMI-enabled sreset interrupt (and
918      * does not look at the error value).
919      *
920      * System reset interrupts are not subject to interlock like machine
921      * check, so this memory area could be corrupted if the sreset is
922      * interrupted by a machine check (or vice versa) if it was shared. To
923      * prevent this, system reset uses per-CPU areas for the sreset save
924      * area. A system reset that interrupts a system reset handler could
925      * still overwrite this area, but Linux doesn't try to recover in that
926      * case anyway.
927      *
928      * The extra 8 bytes is required because Linux's FWNMI error log check
929      * is off-by-one.
930      *
931      * RTAS_MIN_SIZE is required for the RTAS blob itself.
932      */
933     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_MIN_SIZE +
934                           RTAS_ERROR_LOG_MAX +
935                           ms->smp.max_cpus * sizeof(uint64_t) * 2 +
936                           sizeof(uint64_t)));
937     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
938                           RTAS_ERROR_LOG_MAX));
939     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
940                           RTAS_EVENT_SCAN_RATE));
941 
942     g_assert(msi_nonbroken);
943     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
944 
945     /*
946      * According to PAPR, rtas ibm,os-term does not guarantee a return
947      * back to the guest cpu.
948      *
949      * While an additional ibm,extended-os-term property indicates
950      * that rtas call return will always occur. Set this property.
951      */
952     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
953 
954     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
955                      lrdr_capacity, sizeof(lrdr_capacity)));
956 
957     spapr_dt_rtas_tokens(fdt, rtas);
958 }
959 
960 /*
961  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
962  * and the XIVE features that the guest may request and thus the valid
963  * values for bytes 23..26 of option vector 5:
964  */
965 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
966                                           int chosen)
967 {
968     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
969 
970     char val[2 * 4] = {
971         23, 0x00, /* XICS / XIVE mode */
972         24, 0x00, /* Hash/Radix, filled in below. */
973         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
974         26, 0x40, /* Radix options: GTSE == yes. */
975     };
976 
977     if (spapr->irq->xics && spapr->irq->xive) {
978         val[1] = SPAPR_OV5_XIVE_BOTH;
979     } else if (spapr->irq->xive) {
980         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
981     } else {
982         assert(spapr->irq->xics);
983         val[1] = SPAPR_OV5_XIVE_LEGACY;
984     }
985 
986     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
987                           first_ppc_cpu->compat_pvr)) {
988         /*
989          * If we're in a pre POWER9 compat mode then the guest should
990          * do hash and use the legacy interrupt mode
991          */
992         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
993         val[3] = 0x00; /* Hash */
994         spapr_check_mmu_mode(false);
995     } else if (kvm_enabled()) {
996         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
997             val[3] = 0x80; /* OV5_MMU_BOTH */
998         } else if (kvmppc_has_cap_mmu_radix()) {
999             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1000         } else {
1001             val[3] = 0x00; /* Hash */
1002         }
1003     } else {
1004         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1005         val[3] = 0xC0;
1006     }
1007     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1008                      val, sizeof(val)));
1009 }
1010 
1011 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt, bool reset)
1012 {
1013     MachineState *machine = MACHINE(spapr);
1014     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1015     int chosen;
1016 
1017     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1018 
1019     if (reset) {
1020         const char *boot_device = spapr->boot_device;
1021         g_autofree char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1022         size_t cb = 0;
1023         g_autofree char *bootlist = get_boot_devices_list(&cb);
1024 
1025         if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1026             _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1027                                     machine->kernel_cmdline));
1028         }
1029 
1030         if (spapr->initrd_size) {
1031             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1032                                   spapr->initrd_base));
1033             _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1034                                   spapr->initrd_base + spapr->initrd_size));
1035         }
1036 
1037         if (spapr->kernel_size) {
1038             uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1039                                   cpu_to_be64(spapr->kernel_size) };
1040 
1041             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1042                          &kprop, sizeof(kprop)));
1043             if (spapr->kernel_le) {
1044                 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1045             }
1046         }
1047         if (boot_menu) {
1048             _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1049         }
1050         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1051         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1052         _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1053 
1054         if (cb && bootlist) {
1055             int i;
1056 
1057             for (i = 0; i < cb; i++) {
1058                 if (bootlist[i] == '\n') {
1059                     bootlist[i] = ' ';
1060                 }
1061             }
1062             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1063         }
1064 
1065         if (boot_device && strlen(boot_device)) {
1066             _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1067         }
1068 
1069         if (!spapr->has_graphics && stdout_path) {
1070             /*
1071              * "linux,stdout-path" and "stdout" properties are
1072              * deprecated by linux kernel. New platforms should only
1073              * use the "stdout-path" property. Set the new property
1074              * and continue using older property to remain compatible
1075              * with the existing firmware.
1076              */
1077             _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1078             _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1079         }
1080 
1081         /*
1082          * We can deal with BAR reallocation just fine, advertise it
1083          * to the guest
1084          */
1085         if (smc->linux_pci_probe) {
1086             _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1087         }
1088 
1089         spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1090     }
1091 
1092     _FDT(spapr_dt_ovec(fdt, chosen, spapr->ov5_cas, "ibm,architecture-vec-5"));
1093 }
1094 
1095 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1096 {
1097     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1098      * KVM to work under pHyp with some guest co-operation */
1099     int hypervisor;
1100     uint8_t hypercall[16];
1101 
1102     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1103     /* indicate KVM hypercall interface */
1104     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1105     if (kvmppc_has_cap_fixup_hcalls()) {
1106         /*
1107          * Older KVM versions with older guest kernels were broken
1108          * with the magic page, don't allow the guest to map it.
1109          */
1110         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1111                                   sizeof(hypercall))) {
1112             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1113                              hypercall, sizeof(hypercall)));
1114         }
1115     }
1116 }
1117 
1118 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1119 {
1120     MachineState *machine = MACHINE(spapr);
1121     MachineClass *mc = MACHINE_GET_CLASS(machine);
1122     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1123     uint32_t root_drc_type_mask = 0;
1124     int ret;
1125     void *fdt;
1126     SpaprPhbState *phb;
1127     char *buf;
1128 
1129     fdt = g_malloc0(space);
1130     _FDT((fdt_create_empty_tree(fdt, space)));
1131 
1132     /* Root node */
1133     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1134     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1135     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1136 
1137     /* Guest UUID & Name*/
1138     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1139     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1140     if (qemu_uuid_set) {
1141         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1142     }
1143     g_free(buf);
1144 
1145     if (qemu_get_vm_name()) {
1146         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1147                                 qemu_get_vm_name()));
1148     }
1149 
1150     /* Host Model & Serial Number */
1151     if (spapr->host_model) {
1152         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1153     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1154         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1155         g_free(buf);
1156     }
1157 
1158     if (spapr->host_serial) {
1159         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1160     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1161         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1162         g_free(buf);
1163     }
1164 
1165     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1166     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1167 
1168     /* /interrupt controller */
1169     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1170 
1171     ret = spapr_dt_memory(spapr, fdt);
1172     if (ret < 0) {
1173         error_report("couldn't setup memory nodes in fdt");
1174         exit(1);
1175     }
1176 
1177     /* /vdevice */
1178     spapr_dt_vdevice(spapr->vio_bus, fdt);
1179 
1180     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1181         ret = spapr_dt_rng(fdt);
1182         if (ret < 0) {
1183             error_report("could not set up rng device in the fdt");
1184             exit(1);
1185         }
1186     }
1187 
1188     QLIST_FOREACH(phb, &spapr->phbs, list) {
1189         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1190         if (ret < 0) {
1191             error_report("couldn't setup PCI devices in fdt");
1192             exit(1);
1193         }
1194     }
1195 
1196     spapr_dt_cpus(fdt, spapr);
1197 
1198     /* ibm,drc-indexes and friends */
1199     if (smc->dr_lmb_enabled) {
1200         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_LMB;
1201     }
1202     if (smc->dr_phb_enabled) {
1203         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PHB;
1204     }
1205     if (mc->nvdimm_supported) {
1206         root_drc_type_mask |= SPAPR_DR_CONNECTOR_TYPE_PMEM;
1207     }
1208     if (root_drc_type_mask) {
1209         _FDT(spapr_dt_drc(fdt, 0, NULL, root_drc_type_mask));
1210     }
1211 
1212     if (mc->has_hotpluggable_cpus) {
1213         int offset = fdt_path_offset(fdt, "/cpus");
1214         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1215         if (ret < 0) {
1216             error_report("Couldn't set up CPU DR device tree properties");
1217             exit(1);
1218         }
1219     }
1220 
1221     /* /event-sources */
1222     spapr_dt_events(spapr, fdt);
1223 
1224     /* /rtas */
1225     spapr_dt_rtas(spapr, fdt);
1226 
1227     /* /chosen */
1228     spapr_dt_chosen(spapr, fdt, reset);
1229 
1230     /* /hypervisor */
1231     if (kvm_enabled()) {
1232         spapr_dt_hypervisor(spapr, fdt);
1233     }
1234 
1235     /* Build memory reserve map */
1236     if (reset) {
1237         if (spapr->kernel_size) {
1238             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1239                                   spapr->kernel_size)));
1240         }
1241         if (spapr->initrd_size) {
1242             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1243                                   spapr->initrd_size)));
1244         }
1245     }
1246 
1247     /* NVDIMM devices */
1248     if (mc->nvdimm_supported) {
1249         spapr_dt_persistent_memory(spapr, fdt);
1250     }
1251 
1252     return fdt;
1253 }
1254 
1255 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1256 {
1257     SpaprMachineState *spapr = opaque;
1258 
1259     return (addr & 0x0fffffff) + spapr->kernel_addr;
1260 }
1261 
1262 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1263                                     PowerPCCPU *cpu)
1264 {
1265     CPUPPCState *env = &cpu->env;
1266 
1267     /* The TCG path should also be holding the BQL at this point */
1268     g_assert(qemu_mutex_iothread_locked());
1269 
1270     g_assert(!vhyp_cpu_in_nested(cpu));
1271 
1272     if (msr_pr) {
1273         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1274         env->gpr[3] = H_PRIVILEGE;
1275     } else {
1276         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1277     }
1278 }
1279 
1280 struct LPCRSyncState {
1281     target_ulong value;
1282     target_ulong mask;
1283 };
1284 
1285 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1286 {
1287     struct LPCRSyncState *s = arg.host_ptr;
1288     PowerPCCPU *cpu = POWERPC_CPU(cs);
1289     CPUPPCState *env = &cpu->env;
1290     target_ulong lpcr;
1291 
1292     cpu_synchronize_state(cs);
1293     lpcr = env->spr[SPR_LPCR];
1294     lpcr &= ~s->mask;
1295     lpcr |= s->value;
1296     ppc_store_lpcr(cpu, lpcr);
1297 }
1298 
1299 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1300 {
1301     CPUState *cs;
1302     struct LPCRSyncState s = {
1303         .value = value,
1304         .mask = mask
1305     };
1306     CPU_FOREACH(cs) {
1307         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1308     }
1309 }
1310 
1311 static bool spapr_get_pate(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu,
1312                            target_ulong lpid, ppc_v3_pate_t *entry)
1313 {
1314     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1315     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
1316 
1317     if (!spapr_cpu->in_nested) {
1318         assert(lpid == 0);
1319 
1320         /* Copy PATE1:GR into PATE0:HR */
1321         entry->dw0 = spapr->patb_entry & PATE0_HR;
1322         entry->dw1 = spapr->patb_entry;
1323 
1324     } else {
1325         uint64_t patb, pats;
1326 
1327         assert(lpid != 0);
1328 
1329         patb = spapr->nested_ptcr & PTCR_PATB;
1330         pats = spapr->nested_ptcr & PTCR_PATS;
1331 
1332         /* Calculate number of entries */
1333         pats = 1ull << (pats + 12 - 4);
1334         if (pats <= lpid) {
1335             return false;
1336         }
1337 
1338         /* Grab entry */
1339         patb += 16 * lpid;
1340         entry->dw0 = ldq_phys(CPU(cpu)->as, patb);
1341         entry->dw1 = ldq_phys(CPU(cpu)->as, patb + 8);
1342     }
1343 
1344     return true;
1345 }
1346 
1347 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1348 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1349 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1350 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1351 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1352 
1353 /*
1354  * Get the fd to access the kernel htab, re-opening it if necessary
1355  */
1356 static int get_htab_fd(SpaprMachineState *spapr)
1357 {
1358     Error *local_err = NULL;
1359 
1360     if (spapr->htab_fd >= 0) {
1361         return spapr->htab_fd;
1362     }
1363 
1364     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1365     if (spapr->htab_fd < 0) {
1366         error_report_err(local_err);
1367     }
1368 
1369     return spapr->htab_fd;
1370 }
1371 
1372 void close_htab_fd(SpaprMachineState *spapr)
1373 {
1374     if (spapr->htab_fd >= 0) {
1375         close(spapr->htab_fd);
1376     }
1377     spapr->htab_fd = -1;
1378 }
1379 
1380 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1381 {
1382     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1383 
1384     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1385 }
1386 
1387 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1388 {
1389     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1390 
1391     assert(kvm_enabled());
1392 
1393     if (!spapr->htab) {
1394         return 0;
1395     }
1396 
1397     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1398 }
1399 
1400 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1401                                                 hwaddr ptex, int n)
1402 {
1403     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1404     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1405 
1406     if (!spapr->htab) {
1407         /*
1408          * HTAB is controlled by KVM. Fetch into temporary buffer
1409          */
1410         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1411         kvmppc_read_hptes(hptes, ptex, n);
1412         return hptes;
1413     }
1414 
1415     /*
1416      * HTAB is controlled by QEMU. Just point to the internally
1417      * accessible PTEG.
1418      */
1419     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1420 }
1421 
1422 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1423                               const ppc_hash_pte64_t *hptes,
1424                               hwaddr ptex, int n)
1425 {
1426     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1427 
1428     if (!spapr->htab) {
1429         g_free((void *)hptes);
1430     }
1431 
1432     /* Nothing to do for qemu managed HPT */
1433 }
1434 
1435 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1436                       uint64_t pte0, uint64_t pte1)
1437 {
1438     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1439     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1440 
1441     if (!spapr->htab) {
1442         kvmppc_write_hpte(ptex, pte0, pte1);
1443     } else {
1444         if (pte0 & HPTE64_V_VALID) {
1445             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1446             /*
1447              * When setting valid, we write PTE1 first. This ensures
1448              * proper synchronization with the reading code in
1449              * ppc_hash64_pteg_search()
1450              */
1451             smp_wmb();
1452             stq_p(spapr->htab + offset, pte0);
1453         } else {
1454             stq_p(spapr->htab + offset, pte0);
1455             /*
1456              * When clearing it we set PTE0 first. This ensures proper
1457              * synchronization with the reading code in
1458              * ppc_hash64_pteg_search()
1459              */
1460             smp_wmb();
1461             stq_p(spapr->htab + offset + HPTE64_DW1, pte1);
1462         }
1463     }
1464 }
1465 
1466 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1467                              uint64_t pte1)
1468 {
1469     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_C;
1470     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1471 
1472     if (!spapr->htab) {
1473         /* There should always be a hash table when this is called */
1474         error_report("spapr_hpte_set_c called with no hash table !");
1475         return;
1476     }
1477 
1478     /* The HW performs a non-atomic byte update */
1479     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1480 }
1481 
1482 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1483                              uint64_t pte1)
1484 {
1485     hwaddr offset = ptex * HASH_PTE_SIZE_64 + HPTE64_DW1_R;
1486     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1487 
1488     if (!spapr->htab) {
1489         /* There should always be a hash table when this is called */
1490         error_report("spapr_hpte_set_r called with no hash table !");
1491         return;
1492     }
1493 
1494     /* The HW performs a non-atomic byte update */
1495     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1496 }
1497 
1498 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1499 {
1500     int shift;
1501 
1502     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1503      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1504      * that's much more than is needed for Linux guests */
1505     shift = ctz64(pow2ceil(ramsize)) - 7;
1506     shift = MAX(shift, 18); /* Minimum architected size */
1507     shift = MIN(shift, 46); /* Maximum architected size */
1508     return shift;
1509 }
1510 
1511 void spapr_free_hpt(SpaprMachineState *spapr)
1512 {
1513     g_free(spapr->htab);
1514     spapr->htab = NULL;
1515     spapr->htab_shift = 0;
1516     close_htab_fd(spapr);
1517 }
1518 
1519 int spapr_reallocate_hpt(SpaprMachineState *spapr, int shift, Error **errp)
1520 {
1521     ERRP_GUARD();
1522     long rc;
1523 
1524     /* Clean up any HPT info from a previous boot */
1525     spapr_free_hpt(spapr);
1526 
1527     rc = kvmppc_reset_htab(shift);
1528 
1529     if (rc == -EOPNOTSUPP) {
1530         error_setg(errp, "HPT not supported in nested guests");
1531         return -EOPNOTSUPP;
1532     }
1533 
1534     if (rc < 0) {
1535         /* kernel-side HPT needed, but couldn't allocate one */
1536         error_setg_errno(errp, errno, "Failed to allocate KVM HPT of order %d",
1537                          shift);
1538         error_append_hint(errp, "Try smaller maxmem?\n");
1539         return -errno;
1540     } else if (rc > 0) {
1541         /* kernel-side HPT allocated */
1542         if (rc != shift) {
1543             error_setg(errp,
1544                        "Requested order %d HPT, but kernel allocated order %ld",
1545                        shift, rc);
1546             error_append_hint(errp, "Try smaller maxmem?\n");
1547             return -ENOSPC;
1548         }
1549 
1550         spapr->htab_shift = shift;
1551         spapr->htab = NULL;
1552     } else {
1553         /* kernel-side HPT not needed, allocate in userspace instead */
1554         size_t size = 1ULL << shift;
1555         int i;
1556 
1557         spapr->htab = qemu_memalign(size, size);
1558         memset(spapr->htab, 0, size);
1559         spapr->htab_shift = shift;
1560 
1561         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1562             DIRTY_HPTE(HPTE(spapr->htab, i));
1563         }
1564     }
1565     /* We're setting up a hash table, so that means we're not radix */
1566     spapr->patb_entry = 0;
1567     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1568     return 0;
1569 }
1570 
1571 void spapr_setup_hpt(SpaprMachineState *spapr)
1572 {
1573     int hpt_shift;
1574 
1575     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED) {
1576         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1577     } else {
1578         uint64_t current_ram_size;
1579 
1580         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1581         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1582     }
1583     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1584 
1585     if (kvm_enabled()) {
1586         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1587 
1588         /* Check our RMA fits in the possible VRMA */
1589         if (vrma_limit < spapr->rma_size) {
1590             error_report("Unable to create %" HWADDR_PRIu
1591                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1592                          spapr->rma_size / MiB, vrma_limit / MiB);
1593             exit(EXIT_FAILURE);
1594         }
1595     }
1596 }
1597 
1598 void spapr_check_mmu_mode(bool guest_radix)
1599 {
1600     if (guest_radix) {
1601         if (kvm_enabled() && !kvmppc_has_cap_mmu_radix()) {
1602             error_report("Guest requested unavailable MMU mode (radix).");
1603             exit(EXIT_FAILURE);
1604         }
1605     } else {
1606         if (kvm_enabled() && kvmppc_has_cap_mmu_radix()
1607             && !kvmppc_has_cap_mmu_hash_v3()) {
1608             error_report("Guest requested unavailable MMU mode (hash).");
1609             exit(EXIT_FAILURE);
1610         }
1611     }
1612 }
1613 
1614 static void spapr_machine_reset(MachineState *machine)
1615 {
1616     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1617     PowerPCCPU *first_ppc_cpu;
1618     hwaddr fdt_addr;
1619     void *fdt;
1620     int rc;
1621 
1622     pef_kvm_reset(machine->cgs, &error_fatal);
1623     spapr_caps_apply(spapr);
1624 
1625     first_ppc_cpu = POWERPC_CPU(first_cpu);
1626     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1627         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1628                               spapr->max_compat_pvr)) {
1629         /*
1630          * If using KVM with radix mode available, VCPUs can be started
1631          * without a HPT because KVM will start them in radix mode.
1632          * Set the GR bit in PATE so that we know there is no HPT.
1633          */
1634         spapr->patb_entry = PATE1_GR;
1635         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1636     } else {
1637         spapr_setup_hpt(spapr);
1638     }
1639 
1640     qemu_devices_reset();
1641 
1642     spapr_ovec_cleanup(spapr->ov5_cas);
1643     spapr->ov5_cas = spapr_ovec_new();
1644 
1645     ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1646 
1647     /*
1648      * This is fixing some of the default configuration of the XIVE
1649      * devices. To be called after the reset of the machine devices.
1650      */
1651     spapr_irq_reset(spapr, &error_fatal);
1652 
1653     /*
1654      * There is no CAS under qtest. Simulate one to please the code that
1655      * depends on spapr->ov5_cas. This is especially needed to test device
1656      * unplug, so we do that before resetting the DRCs.
1657      */
1658     if (qtest_enabled()) {
1659         spapr_ovec_cleanup(spapr->ov5_cas);
1660         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1661     }
1662 
1663     spapr_nvdimm_finish_flushes();
1664 
1665     /* DRC reset may cause a device to be unplugged. This will cause troubles
1666      * if this device is used by another device (eg, a running vhost backend
1667      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1668      * situations, we reset DRCs after all devices have been reset.
1669      */
1670     spapr_drc_reset_all(spapr);
1671 
1672     spapr_clear_pending_events(spapr);
1673 
1674     /*
1675      * We place the device tree just below either the top of the RMA,
1676      * or just below 2GB, whichever is lower, so that it can be
1677      * processed with 32-bit real mode code if necessary
1678      */
1679     fdt_addr = MIN(spapr->rma_size, FDT_MAX_ADDR) - FDT_MAX_SIZE;
1680 
1681     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1682     if (spapr->vof) {
1683         spapr_vof_reset(spapr, fdt, &error_fatal);
1684         /*
1685          * Do not pack the FDT as the client may change properties.
1686          * VOF client does not expect the FDT so we do not load it to the VM.
1687          */
1688     } else {
1689         rc = fdt_pack(fdt);
1690         /* Should only fail if we've built a corrupted tree */
1691         assert(rc == 0);
1692 
1693         spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT,
1694                                   0, fdt_addr, 0);
1695         cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1696     }
1697     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1698 
1699     g_free(spapr->fdt_blob);
1700     spapr->fdt_size = fdt_totalsize(fdt);
1701     spapr->fdt_initial_size = spapr->fdt_size;
1702     spapr->fdt_blob = fdt;
1703 
1704     /* Set up the entry state */
1705     first_ppc_cpu->env.gpr[5] = 0;
1706 
1707     spapr->fwnmi_system_reset_addr = -1;
1708     spapr->fwnmi_machine_check_addr = -1;
1709     spapr->fwnmi_machine_check_interlock = -1;
1710 
1711     /* Signal all vCPUs waiting on this condition */
1712     qemu_cond_broadcast(&spapr->fwnmi_machine_check_interlock_cond);
1713 
1714     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1715 }
1716 
1717 static void spapr_create_nvram(SpaprMachineState *spapr)
1718 {
1719     DeviceState *dev = qdev_new("spapr-nvram");
1720     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1721 
1722     if (dinfo) {
1723         qdev_prop_set_drive_err(dev, "drive", blk_by_legacy_dinfo(dinfo),
1724                                 &error_fatal);
1725     }
1726 
1727     qdev_realize_and_unref(dev, &spapr->vio_bus->bus, &error_fatal);
1728 
1729     spapr->nvram = (struct SpaprNvram *)dev;
1730 }
1731 
1732 static void spapr_rtc_create(SpaprMachineState *spapr)
1733 {
1734     object_initialize_child_with_props(OBJECT(spapr), "rtc", &spapr->rtc,
1735                                        sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1736                                        &error_fatal, NULL);
1737     qdev_realize(DEVICE(&spapr->rtc), NULL, &error_fatal);
1738     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1739                               "date");
1740 }
1741 
1742 /* Returns whether we want to use VGA or not */
1743 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1744 {
1745     switch (vga_interface_type) {
1746     case VGA_NONE:
1747         return false;
1748     case VGA_DEVICE:
1749         return true;
1750     case VGA_STD:
1751     case VGA_VIRTIO:
1752     case VGA_CIRRUS:
1753         return pci_vga_init(pci_bus) != NULL;
1754     default:
1755         error_setg(errp,
1756                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1757         return false;
1758     }
1759 }
1760 
1761 static int spapr_pre_load(void *opaque)
1762 {
1763     int rc;
1764 
1765     rc = spapr_caps_pre_load(opaque);
1766     if (rc) {
1767         return rc;
1768     }
1769 
1770     return 0;
1771 }
1772 
1773 static int spapr_post_load(void *opaque, int version_id)
1774 {
1775     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1776     int err = 0;
1777 
1778     err = spapr_caps_post_migration(spapr);
1779     if (err) {
1780         return err;
1781     }
1782 
1783     /*
1784      * In earlier versions, there was no separate qdev for the PAPR
1785      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1786      * So when migrating from those versions, poke the incoming offset
1787      * value into the RTC device
1788      */
1789     if (version_id < 3) {
1790         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1791         if (err) {
1792             return err;
1793         }
1794     }
1795 
1796     if (kvm_enabled() && spapr->patb_entry) {
1797         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1798         bool radix = !!(spapr->patb_entry & PATE1_GR);
1799         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1800 
1801         /*
1802          * Update LPCR:HR and UPRT as they may not be set properly in
1803          * the stream
1804          */
1805         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1806                             LPCR_HR | LPCR_UPRT);
1807 
1808         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1809         if (err) {
1810             error_report("Process table config unsupported by the host");
1811             return -EINVAL;
1812         }
1813     }
1814 
1815     err = spapr_irq_post_load(spapr, version_id);
1816     if (err) {
1817         return err;
1818     }
1819 
1820     return err;
1821 }
1822 
1823 static int spapr_pre_save(void *opaque)
1824 {
1825     int rc;
1826 
1827     rc = spapr_caps_pre_save(opaque);
1828     if (rc) {
1829         return rc;
1830     }
1831 
1832     return 0;
1833 }
1834 
1835 static bool version_before_3(void *opaque, int version_id)
1836 {
1837     return version_id < 3;
1838 }
1839 
1840 static bool spapr_pending_events_needed(void *opaque)
1841 {
1842     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1843     return !QTAILQ_EMPTY(&spapr->pending_events);
1844 }
1845 
1846 static const VMStateDescription vmstate_spapr_event_entry = {
1847     .name = "spapr_event_log_entry",
1848     .version_id = 1,
1849     .minimum_version_id = 1,
1850     .fields = (VMStateField[]) {
1851         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1852         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1853         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1854                                      NULL, extended_length),
1855         VMSTATE_END_OF_LIST()
1856     },
1857 };
1858 
1859 static const VMStateDescription vmstate_spapr_pending_events = {
1860     .name = "spapr_pending_events",
1861     .version_id = 1,
1862     .minimum_version_id = 1,
1863     .needed = spapr_pending_events_needed,
1864     .fields = (VMStateField[]) {
1865         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1866                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1867         VMSTATE_END_OF_LIST()
1868     },
1869 };
1870 
1871 static bool spapr_ov5_cas_needed(void *opaque)
1872 {
1873     SpaprMachineState *spapr = opaque;
1874     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1875     bool cas_needed;
1876 
1877     /* Prior to the introduction of SpaprOptionVector, we had two option
1878      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1879      * Both of these options encode machine topology into the device-tree
1880      * in such a way that the now-booted OS should still be able to interact
1881      * appropriately with QEMU regardless of what options were actually
1882      * negotiatied on the source side.
1883      *
1884      * As such, we can avoid migrating the CAS-negotiated options if these
1885      * are the only options available on the current machine/platform.
1886      * Since these are the only options available for pseries-2.7 and
1887      * earlier, this allows us to maintain old->new/new->old migration
1888      * compatibility.
1889      *
1890      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1891      * via default pseries-2.8 machines and explicit command-line parameters.
1892      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1893      * of the actual CAS-negotiated values to continue working properly. For
1894      * example, availability of memory unplug depends on knowing whether
1895      * OV5_HP_EVT was negotiated via CAS.
1896      *
1897      * Thus, for any cases where the set of available CAS-negotiatable
1898      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1899      * include the CAS-negotiated options in the migration stream, unless
1900      * if they affect boot time behaviour only.
1901      */
1902     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1903     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1904     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1905 
1906     /* We need extra information if we have any bits outside the mask
1907      * defined above */
1908     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1909 
1910     spapr_ovec_cleanup(ov5_mask);
1911 
1912     return cas_needed;
1913 }
1914 
1915 static const VMStateDescription vmstate_spapr_ov5_cas = {
1916     .name = "spapr_option_vector_ov5_cas",
1917     .version_id = 1,
1918     .minimum_version_id = 1,
1919     .needed = spapr_ov5_cas_needed,
1920     .fields = (VMStateField[]) {
1921         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1922                                  vmstate_spapr_ovec, SpaprOptionVector),
1923         VMSTATE_END_OF_LIST()
1924     },
1925 };
1926 
1927 static bool spapr_patb_entry_needed(void *opaque)
1928 {
1929     SpaprMachineState *spapr = opaque;
1930 
1931     return !!spapr->patb_entry;
1932 }
1933 
1934 static const VMStateDescription vmstate_spapr_patb_entry = {
1935     .name = "spapr_patb_entry",
1936     .version_id = 1,
1937     .minimum_version_id = 1,
1938     .needed = spapr_patb_entry_needed,
1939     .fields = (VMStateField[]) {
1940         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1941         VMSTATE_END_OF_LIST()
1942     },
1943 };
1944 
1945 static bool spapr_irq_map_needed(void *opaque)
1946 {
1947     SpaprMachineState *spapr = opaque;
1948 
1949     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1950 }
1951 
1952 static const VMStateDescription vmstate_spapr_irq_map = {
1953     .name = "spapr_irq_map",
1954     .version_id = 1,
1955     .minimum_version_id = 1,
1956     .needed = spapr_irq_map_needed,
1957     .fields = (VMStateField[]) {
1958         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1959         VMSTATE_END_OF_LIST()
1960     },
1961 };
1962 
1963 static bool spapr_dtb_needed(void *opaque)
1964 {
1965     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1966 
1967     return smc->update_dt_enabled;
1968 }
1969 
1970 static int spapr_dtb_pre_load(void *opaque)
1971 {
1972     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1973 
1974     g_free(spapr->fdt_blob);
1975     spapr->fdt_blob = NULL;
1976     spapr->fdt_size = 0;
1977 
1978     return 0;
1979 }
1980 
1981 static const VMStateDescription vmstate_spapr_dtb = {
1982     .name = "spapr_dtb",
1983     .version_id = 1,
1984     .minimum_version_id = 1,
1985     .needed = spapr_dtb_needed,
1986     .pre_load = spapr_dtb_pre_load,
1987     .fields = (VMStateField[]) {
1988         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1989         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1990         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1991                                      fdt_size),
1992         VMSTATE_END_OF_LIST()
1993     },
1994 };
1995 
1996 static bool spapr_fwnmi_needed(void *opaque)
1997 {
1998     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1999 
2000     return spapr->fwnmi_machine_check_addr != -1;
2001 }
2002 
2003 static int spapr_fwnmi_pre_save(void *opaque)
2004 {
2005     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2006 
2007     /*
2008      * Check if machine check handling is in progress and print a
2009      * warning message.
2010      */
2011     if (spapr->fwnmi_machine_check_interlock != -1) {
2012         warn_report("A machine check is being handled during migration. The"
2013                 "handler may run and log hardware error on the destination");
2014     }
2015 
2016     return 0;
2017 }
2018 
2019 static const VMStateDescription vmstate_spapr_fwnmi = {
2020     .name = "spapr_fwnmi",
2021     .version_id = 1,
2022     .minimum_version_id = 1,
2023     .needed = spapr_fwnmi_needed,
2024     .pre_save = spapr_fwnmi_pre_save,
2025     .fields = (VMStateField[]) {
2026         VMSTATE_UINT64(fwnmi_system_reset_addr, SpaprMachineState),
2027         VMSTATE_UINT64(fwnmi_machine_check_addr, SpaprMachineState),
2028         VMSTATE_INT32(fwnmi_machine_check_interlock, SpaprMachineState),
2029         VMSTATE_END_OF_LIST()
2030     },
2031 };
2032 
2033 static const VMStateDescription vmstate_spapr = {
2034     .name = "spapr",
2035     .version_id = 3,
2036     .minimum_version_id = 1,
2037     .pre_load = spapr_pre_load,
2038     .post_load = spapr_post_load,
2039     .pre_save = spapr_pre_save,
2040     .fields = (VMStateField[]) {
2041         /* used to be @next_irq */
2042         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2043 
2044         /* RTC offset */
2045         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2046 
2047         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2048         VMSTATE_END_OF_LIST()
2049     },
2050     .subsections = (const VMStateDescription*[]) {
2051         &vmstate_spapr_ov5_cas,
2052         &vmstate_spapr_patb_entry,
2053         &vmstate_spapr_pending_events,
2054         &vmstate_spapr_cap_htm,
2055         &vmstate_spapr_cap_vsx,
2056         &vmstate_spapr_cap_dfp,
2057         &vmstate_spapr_cap_cfpc,
2058         &vmstate_spapr_cap_sbbc,
2059         &vmstate_spapr_cap_ibs,
2060         &vmstate_spapr_cap_hpt_maxpagesize,
2061         &vmstate_spapr_irq_map,
2062         &vmstate_spapr_cap_nested_kvm_hv,
2063         &vmstate_spapr_dtb,
2064         &vmstate_spapr_cap_large_decr,
2065         &vmstate_spapr_cap_ccf_assist,
2066         &vmstate_spapr_cap_fwnmi,
2067         &vmstate_spapr_fwnmi,
2068         &vmstate_spapr_cap_rpt_invalidate,
2069         NULL
2070     }
2071 };
2072 
2073 static int htab_save_setup(QEMUFile *f, void *opaque)
2074 {
2075     SpaprMachineState *spapr = opaque;
2076 
2077     /* "Iteration" header */
2078     if (!spapr->htab_shift) {
2079         qemu_put_be32(f, -1);
2080     } else {
2081         qemu_put_be32(f, spapr->htab_shift);
2082     }
2083 
2084     if (spapr->htab) {
2085         spapr->htab_save_index = 0;
2086         spapr->htab_first_pass = true;
2087     } else {
2088         if (spapr->htab_shift) {
2089             assert(kvm_enabled());
2090         }
2091     }
2092 
2093 
2094     return 0;
2095 }
2096 
2097 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2098                             int chunkstart, int n_valid, int n_invalid)
2099 {
2100     qemu_put_be32(f, chunkstart);
2101     qemu_put_be16(f, n_valid);
2102     qemu_put_be16(f, n_invalid);
2103     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2104                     HASH_PTE_SIZE_64 * n_valid);
2105 }
2106 
2107 static void htab_save_end_marker(QEMUFile *f)
2108 {
2109     qemu_put_be32(f, 0);
2110     qemu_put_be16(f, 0);
2111     qemu_put_be16(f, 0);
2112 }
2113 
2114 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2115                                  int64_t max_ns)
2116 {
2117     bool has_timeout = max_ns != -1;
2118     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2119     int index = spapr->htab_save_index;
2120     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2121 
2122     assert(spapr->htab_first_pass);
2123 
2124     do {
2125         int chunkstart;
2126 
2127         /* Consume invalid HPTEs */
2128         while ((index < htabslots)
2129                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2130             CLEAN_HPTE(HPTE(spapr->htab, index));
2131             index++;
2132         }
2133 
2134         /* Consume valid HPTEs */
2135         chunkstart = index;
2136         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2137                && HPTE_VALID(HPTE(spapr->htab, index))) {
2138             CLEAN_HPTE(HPTE(spapr->htab, index));
2139             index++;
2140         }
2141 
2142         if (index > chunkstart) {
2143             int n_valid = index - chunkstart;
2144 
2145             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2146 
2147             if (has_timeout &&
2148                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2149                 break;
2150             }
2151         }
2152     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2153 
2154     if (index >= htabslots) {
2155         assert(index == htabslots);
2156         index = 0;
2157         spapr->htab_first_pass = false;
2158     }
2159     spapr->htab_save_index = index;
2160 }
2161 
2162 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2163                                 int64_t max_ns)
2164 {
2165     bool final = max_ns < 0;
2166     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2167     int examined = 0, sent = 0;
2168     int index = spapr->htab_save_index;
2169     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2170 
2171     assert(!spapr->htab_first_pass);
2172 
2173     do {
2174         int chunkstart, invalidstart;
2175 
2176         /* Consume non-dirty HPTEs */
2177         while ((index < htabslots)
2178                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2179             index++;
2180             examined++;
2181         }
2182 
2183         chunkstart = index;
2184         /* Consume valid dirty HPTEs */
2185         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2186                && HPTE_DIRTY(HPTE(spapr->htab, index))
2187                && HPTE_VALID(HPTE(spapr->htab, index))) {
2188             CLEAN_HPTE(HPTE(spapr->htab, index));
2189             index++;
2190             examined++;
2191         }
2192 
2193         invalidstart = index;
2194         /* Consume invalid dirty HPTEs */
2195         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2196                && HPTE_DIRTY(HPTE(spapr->htab, index))
2197                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2198             CLEAN_HPTE(HPTE(spapr->htab, index));
2199             index++;
2200             examined++;
2201         }
2202 
2203         if (index > chunkstart) {
2204             int n_valid = invalidstart - chunkstart;
2205             int n_invalid = index - invalidstart;
2206 
2207             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2208             sent += index - chunkstart;
2209 
2210             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2211                 break;
2212             }
2213         }
2214 
2215         if (examined >= htabslots) {
2216             break;
2217         }
2218 
2219         if (index >= htabslots) {
2220             assert(index == htabslots);
2221             index = 0;
2222         }
2223     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2224 
2225     if (index >= htabslots) {
2226         assert(index == htabslots);
2227         index = 0;
2228     }
2229 
2230     spapr->htab_save_index = index;
2231 
2232     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2233 }
2234 
2235 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2236 #define MAX_KVM_BUF_SIZE    2048
2237 
2238 static int htab_save_iterate(QEMUFile *f, void *opaque)
2239 {
2240     SpaprMachineState *spapr = opaque;
2241     int fd;
2242     int rc = 0;
2243 
2244     /* Iteration header */
2245     if (!spapr->htab_shift) {
2246         qemu_put_be32(f, -1);
2247         return 1;
2248     } else {
2249         qemu_put_be32(f, 0);
2250     }
2251 
2252     if (!spapr->htab) {
2253         assert(kvm_enabled());
2254 
2255         fd = get_htab_fd(spapr);
2256         if (fd < 0) {
2257             return fd;
2258         }
2259 
2260         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2261         if (rc < 0) {
2262             return rc;
2263         }
2264     } else  if (spapr->htab_first_pass) {
2265         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2266     } else {
2267         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2268     }
2269 
2270     htab_save_end_marker(f);
2271 
2272     return rc;
2273 }
2274 
2275 static int htab_save_complete(QEMUFile *f, void *opaque)
2276 {
2277     SpaprMachineState *spapr = opaque;
2278     int fd;
2279 
2280     /* Iteration header */
2281     if (!spapr->htab_shift) {
2282         qemu_put_be32(f, -1);
2283         return 0;
2284     } else {
2285         qemu_put_be32(f, 0);
2286     }
2287 
2288     if (!spapr->htab) {
2289         int rc;
2290 
2291         assert(kvm_enabled());
2292 
2293         fd = get_htab_fd(spapr);
2294         if (fd < 0) {
2295             return fd;
2296         }
2297 
2298         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2299         if (rc < 0) {
2300             return rc;
2301         }
2302     } else {
2303         if (spapr->htab_first_pass) {
2304             htab_save_first_pass(f, spapr, -1);
2305         }
2306         htab_save_later_pass(f, spapr, -1);
2307     }
2308 
2309     /* End marker */
2310     htab_save_end_marker(f);
2311 
2312     return 0;
2313 }
2314 
2315 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2316 {
2317     SpaprMachineState *spapr = opaque;
2318     uint32_t section_hdr;
2319     int fd = -1;
2320     Error *local_err = NULL;
2321 
2322     if (version_id < 1 || version_id > 1) {
2323         error_report("htab_load() bad version");
2324         return -EINVAL;
2325     }
2326 
2327     section_hdr = qemu_get_be32(f);
2328 
2329     if (section_hdr == -1) {
2330         spapr_free_hpt(spapr);
2331         return 0;
2332     }
2333 
2334     if (section_hdr) {
2335         int ret;
2336 
2337         /* First section gives the htab size */
2338         ret = spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2339         if (ret < 0) {
2340             error_report_err(local_err);
2341             return ret;
2342         }
2343         return 0;
2344     }
2345 
2346     if (!spapr->htab) {
2347         assert(kvm_enabled());
2348 
2349         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2350         if (fd < 0) {
2351             error_report_err(local_err);
2352             return fd;
2353         }
2354     }
2355 
2356     while (true) {
2357         uint32_t index;
2358         uint16_t n_valid, n_invalid;
2359 
2360         index = qemu_get_be32(f);
2361         n_valid = qemu_get_be16(f);
2362         n_invalid = qemu_get_be16(f);
2363 
2364         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2365             /* End of Stream */
2366             break;
2367         }
2368 
2369         if ((index + n_valid + n_invalid) >
2370             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2371             /* Bad index in stream */
2372             error_report(
2373                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2374                 index, n_valid, n_invalid, spapr->htab_shift);
2375             return -EINVAL;
2376         }
2377 
2378         if (spapr->htab) {
2379             if (n_valid) {
2380                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2381                                 HASH_PTE_SIZE_64 * n_valid);
2382             }
2383             if (n_invalid) {
2384                 memset(HPTE(spapr->htab, index + n_valid), 0,
2385                        HASH_PTE_SIZE_64 * n_invalid);
2386             }
2387         } else {
2388             int rc;
2389 
2390             assert(fd >= 0);
2391 
2392             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid,
2393                                         &local_err);
2394             if (rc < 0) {
2395                 error_report_err(local_err);
2396                 return rc;
2397             }
2398         }
2399     }
2400 
2401     if (!spapr->htab) {
2402         assert(fd >= 0);
2403         close(fd);
2404     }
2405 
2406     return 0;
2407 }
2408 
2409 static void htab_save_cleanup(void *opaque)
2410 {
2411     SpaprMachineState *spapr = opaque;
2412 
2413     close_htab_fd(spapr);
2414 }
2415 
2416 static SaveVMHandlers savevm_htab_handlers = {
2417     .save_setup = htab_save_setup,
2418     .save_live_iterate = htab_save_iterate,
2419     .save_live_complete_precopy = htab_save_complete,
2420     .save_cleanup = htab_save_cleanup,
2421     .load_state = htab_load,
2422 };
2423 
2424 static void spapr_boot_set(void *opaque, const char *boot_device,
2425                            Error **errp)
2426 {
2427     SpaprMachineState *spapr = SPAPR_MACHINE(opaque);
2428 
2429     g_free(spapr->boot_device);
2430     spapr->boot_device = g_strdup(boot_device);
2431 }
2432 
2433 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2434 {
2435     MachineState *machine = MACHINE(spapr);
2436     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2437     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2438     int i;
2439 
2440     for (i = 0; i < nr_lmbs; i++) {
2441         uint64_t addr;
2442 
2443         addr = i * lmb_size + machine->device_memory->base;
2444         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2445                                addr / lmb_size);
2446     }
2447 }
2448 
2449 /*
2450  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2451  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2452  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2453  */
2454 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2455 {
2456     int i;
2457 
2458     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2459         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2460                    " is not aligned to %" PRIu64 " MiB",
2461                    machine->ram_size,
2462                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2463         return;
2464     }
2465 
2466     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2467         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2468                    " is not aligned to %" PRIu64 " MiB",
2469                    machine->ram_size,
2470                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2471         return;
2472     }
2473 
2474     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2475         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2476             error_setg(errp,
2477                        "Node %d memory size 0x%" PRIx64
2478                        " is not aligned to %" PRIu64 " MiB",
2479                        i, machine->numa_state->nodes[i].node_mem,
2480                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2481             return;
2482         }
2483     }
2484 }
2485 
2486 /* find cpu slot in machine->possible_cpus by core_id */
2487 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2488 {
2489     int index = id / ms->smp.threads;
2490 
2491     if (index >= ms->possible_cpus->len) {
2492         return NULL;
2493     }
2494     if (idx) {
2495         *idx = index;
2496     }
2497     return &ms->possible_cpus->cpus[index];
2498 }
2499 
2500 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2501 {
2502     MachineState *ms = MACHINE(spapr);
2503     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2504     Error *local_err = NULL;
2505     bool vsmt_user = !!spapr->vsmt;
2506     int kvm_smt = kvmppc_smt_threads();
2507     int ret;
2508     unsigned int smp_threads = ms->smp.threads;
2509 
2510     if (!kvm_enabled() && (smp_threads > 1)) {
2511         error_setg(errp, "TCG cannot support more than 1 thread/core "
2512                    "on a pseries machine");
2513         return;
2514     }
2515     if (!is_power_of_2(smp_threads)) {
2516         error_setg(errp, "Cannot support %d threads/core on a pseries "
2517                    "machine because it must be a power of 2", smp_threads);
2518         return;
2519     }
2520 
2521     /* Detemine the VSMT mode to use: */
2522     if (vsmt_user) {
2523         if (spapr->vsmt < smp_threads) {
2524             error_setg(errp, "Cannot support VSMT mode %d"
2525                        " because it must be >= threads/core (%d)",
2526                        spapr->vsmt, smp_threads);
2527             return;
2528         }
2529         /* In this case, spapr->vsmt has been set by the command line */
2530     } else if (!smc->smp_threads_vsmt) {
2531         /*
2532          * Default VSMT value is tricky, because we need it to be as
2533          * consistent as possible (for migration), but this requires
2534          * changing it for at least some existing cases.  We pick 8 as
2535          * the value that we'd get with KVM on POWER8, the
2536          * overwhelmingly common case in production systems.
2537          */
2538         spapr->vsmt = MAX(8, smp_threads);
2539     } else {
2540         spapr->vsmt = smp_threads;
2541     }
2542 
2543     /* KVM: If necessary, set the SMT mode: */
2544     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2545         ret = kvmppc_set_smt_threads(spapr->vsmt);
2546         if (ret) {
2547             /* Looks like KVM isn't able to change VSMT mode */
2548             error_setg(&local_err,
2549                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2550                        spapr->vsmt, ret);
2551             /* We can live with that if the default one is big enough
2552              * for the number of threads, and a submultiple of the one
2553              * we want.  In this case we'll waste some vcpu ids, but
2554              * behaviour will be correct */
2555             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2556                 warn_report_err(local_err);
2557             } else {
2558                 if (!vsmt_user) {
2559                     error_append_hint(&local_err,
2560                                       "On PPC, a VM with %d threads/core"
2561                                       " on a host with %d threads/core"
2562                                       " requires the use of VSMT mode %d.\n",
2563                                       smp_threads, kvm_smt, spapr->vsmt);
2564                 }
2565                 kvmppc_error_append_smt_possible_hint(&local_err);
2566                 error_propagate(errp, local_err);
2567             }
2568         }
2569     }
2570     /* else TCG: nothing to do currently */
2571 }
2572 
2573 static void spapr_init_cpus(SpaprMachineState *spapr)
2574 {
2575     MachineState *machine = MACHINE(spapr);
2576     MachineClass *mc = MACHINE_GET_CLASS(machine);
2577     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2578     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2579     const CPUArchIdList *possible_cpus;
2580     unsigned int smp_cpus = machine->smp.cpus;
2581     unsigned int smp_threads = machine->smp.threads;
2582     unsigned int max_cpus = machine->smp.max_cpus;
2583     int boot_cores_nr = smp_cpus / smp_threads;
2584     int i;
2585 
2586     possible_cpus = mc->possible_cpu_arch_ids(machine);
2587     if (mc->has_hotpluggable_cpus) {
2588         if (smp_cpus % smp_threads) {
2589             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2590                          smp_cpus, smp_threads);
2591             exit(1);
2592         }
2593         if (max_cpus % smp_threads) {
2594             error_report("max_cpus (%u) must be multiple of threads (%u)",
2595                          max_cpus, smp_threads);
2596             exit(1);
2597         }
2598     } else {
2599         if (max_cpus != smp_cpus) {
2600             error_report("This machine version does not support CPU hotplug");
2601             exit(1);
2602         }
2603         boot_cores_nr = possible_cpus->len;
2604     }
2605 
2606     if (smc->pre_2_10_has_unused_icps) {
2607         int i;
2608 
2609         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2610             /* Dummy entries get deregistered when real ICPState objects
2611              * are registered during CPU core hotplug.
2612              */
2613             pre_2_10_vmstate_register_dummy_icp(i);
2614         }
2615     }
2616 
2617     for (i = 0; i < possible_cpus->len; i++) {
2618         int core_id = i * smp_threads;
2619 
2620         if (mc->has_hotpluggable_cpus) {
2621             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2622                                    spapr_vcpu_id(spapr, core_id));
2623         }
2624 
2625         if (i < boot_cores_nr) {
2626             Object *core  = object_new(type);
2627             int nr_threads = smp_threads;
2628 
2629             /* Handle the partially filled core for older machine types */
2630             if ((i + 1) * smp_threads >= smp_cpus) {
2631                 nr_threads = smp_cpus - i * smp_threads;
2632             }
2633 
2634             object_property_set_int(core, "nr-threads", nr_threads,
2635                                     &error_fatal);
2636             object_property_set_int(core, CPU_CORE_PROP_CORE_ID, core_id,
2637                                     &error_fatal);
2638             qdev_realize(DEVICE(core), NULL, &error_fatal);
2639 
2640             object_unref(core);
2641         }
2642     }
2643 }
2644 
2645 static PCIHostState *spapr_create_default_phb(void)
2646 {
2647     DeviceState *dev;
2648 
2649     dev = qdev_new(TYPE_SPAPR_PCI_HOST_BRIDGE);
2650     qdev_prop_set_uint32(dev, "index", 0);
2651     sysbus_realize_and_unref(SYS_BUS_DEVICE(dev), &error_fatal);
2652 
2653     return PCI_HOST_BRIDGE(dev);
2654 }
2655 
2656 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2657 {
2658     MachineState *machine = MACHINE(spapr);
2659     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2660     hwaddr rma_size = machine->ram_size;
2661     hwaddr node0_size = spapr_node0_size(machine);
2662 
2663     /* RMA has to fit in the first NUMA node */
2664     rma_size = MIN(rma_size, node0_size);
2665 
2666     /*
2667      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2668      * never exceed that
2669      */
2670     rma_size = MIN(rma_size, 1 * TiB);
2671 
2672     /*
2673      * Clamp the RMA size based on machine type.  This is for
2674      * migration compatibility with older qemu versions, which limited
2675      * the RMA size for complicated and mostly bad reasons.
2676      */
2677     if (smc->rma_limit) {
2678         rma_size = MIN(rma_size, smc->rma_limit);
2679     }
2680 
2681     if (rma_size < MIN_RMA_SLOF) {
2682         error_setg(errp,
2683                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2684                    "ldMiB guest RMA (Real Mode Area memory)",
2685                    MIN_RMA_SLOF / MiB);
2686         return 0;
2687     }
2688 
2689     return rma_size;
2690 }
2691 
2692 static void spapr_create_nvdimm_dr_connectors(SpaprMachineState *spapr)
2693 {
2694     MachineState *machine = MACHINE(spapr);
2695     int i;
2696 
2697     for (i = 0; i < machine->ram_slots; i++) {
2698         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_PMEM, i);
2699     }
2700 }
2701 
2702 /* pSeries LPAR / sPAPR hardware init */
2703 static void spapr_machine_init(MachineState *machine)
2704 {
2705     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2706     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2707     MachineClass *mc = MACHINE_GET_CLASS(machine);
2708     const char *bios_default = spapr->vof ? FW_FILE_NAME_VOF : FW_FILE_NAME;
2709     const char *bios_name = machine->firmware ?: bios_default;
2710     const char *kernel_filename = machine->kernel_filename;
2711     const char *initrd_filename = machine->initrd_filename;
2712     PCIHostState *phb;
2713     int i;
2714     MemoryRegion *sysmem = get_system_memory();
2715     long load_limit, fw_size;
2716     char *filename;
2717     Error *resize_hpt_err = NULL;
2718 
2719     /*
2720      * if Secure VM (PEF) support is configured, then initialize it
2721      */
2722     pef_kvm_init(machine->cgs, &error_fatal);
2723 
2724     msi_nonbroken = true;
2725 
2726     QLIST_INIT(&spapr->phbs);
2727     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2728 
2729     /* Determine capabilities to run with */
2730     spapr_caps_init(spapr);
2731 
2732     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2733     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2734         /*
2735          * If the user explicitly requested a mode we should either
2736          * supply it, or fail completely (which we do below).  But if
2737          * it's not set explicitly, we reset our mode to something
2738          * that works
2739          */
2740         if (resize_hpt_err) {
2741             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2742             error_free(resize_hpt_err);
2743             resize_hpt_err = NULL;
2744         } else {
2745             spapr->resize_hpt = smc->resize_hpt_default;
2746         }
2747     }
2748 
2749     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2750 
2751     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2752         /*
2753          * User requested HPT resize, but this host can't supply it.  Bail out
2754          */
2755         error_report_err(resize_hpt_err);
2756         exit(1);
2757     }
2758     error_free(resize_hpt_err);
2759 
2760     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2761 
2762     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2763     load_limit = MIN(spapr->rma_size, FDT_MAX_ADDR) - FW_OVERHEAD;
2764 
2765     /*
2766      * VSMT must be set in order to be able to compute VCPU ids, ie to
2767      * call spapr_max_server_number() or spapr_vcpu_id().
2768      */
2769     spapr_set_vsmt_mode(spapr, &error_fatal);
2770 
2771     /* Set up Interrupt Controller before we create the VCPUs */
2772     spapr_irq_init(spapr, &error_fatal);
2773 
2774     /* Set up containers for ibm,client-architecture-support negotiated options
2775      */
2776     spapr->ov5 = spapr_ovec_new();
2777     spapr->ov5_cas = spapr_ovec_new();
2778 
2779     if (smc->dr_lmb_enabled) {
2780         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2781         spapr_validate_node_memory(machine, &error_fatal);
2782     }
2783 
2784     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2785 
2786     /* Do not advertise FORM2 NUMA support for pseries-6.1 and older */
2787     if (!smc->pre_6_2_numa_affinity) {
2788         spapr_ovec_set(spapr->ov5, OV5_FORM2_AFFINITY);
2789     }
2790 
2791     /* advertise support for dedicated HP event source to guests */
2792     if (spapr->use_hotplug_event_source) {
2793         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2794     }
2795 
2796     /* advertise support for HPT resizing */
2797     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2798         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2799     }
2800 
2801     /* advertise support for ibm,dyamic-memory-v2 */
2802     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2803 
2804     /* advertise XIVE on POWER9 machines */
2805     if (spapr->irq->xive) {
2806         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2807     }
2808 
2809     /* init CPUs */
2810     spapr_init_cpus(spapr);
2811 
2812     spapr->gpu_numa_id = spapr_numa_initial_nvgpu_numa_id(machine);
2813 
2814     /* Init numa_assoc_array */
2815     spapr_numa_associativity_init(spapr, machine);
2816 
2817     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2818         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2819                               spapr->max_compat_pvr)) {
2820         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_300);
2821         /* KVM and TCG always allow GTSE with radix... */
2822         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2823     }
2824     /* ... but not with hash (currently). */
2825 
2826     if (kvm_enabled()) {
2827         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2828         kvmppc_enable_logical_ci_hcalls();
2829         kvmppc_enable_set_mode_hcall();
2830 
2831         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2832         kvmppc_enable_clear_ref_mod_hcalls();
2833 
2834         /* Enable H_PAGE_INIT */
2835         kvmppc_enable_h_page_init();
2836     }
2837 
2838     /* map RAM */
2839     memory_region_add_subregion(sysmem, 0, machine->ram);
2840 
2841     /* always allocate the device memory information */
2842     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2843 
2844     /* initialize hotplug memory address space */
2845     if (machine->ram_size < machine->maxram_size) {
2846         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2847         /*
2848          * Limit the number of hotpluggable memory slots to half the number
2849          * slots that KVM supports, leaving the other half for PCI and other
2850          * devices. However ensure that number of slots doesn't drop below 32.
2851          */
2852         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2853                            SPAPR_MAX_RAM_SLOTS;
2854 
2855         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2856             max_memslots = SPAPR_MAX_RAM_SLOTS;
2857         }
2858         if (machine->ram_slots > max_memslots) {
2859             error_report("Specified number of memory slots %"
2860                          PRIu64" exceeds max supported %d",
2861                          machine->ram_slots, max_memslots);
2862             exit(1);
2863         }
2864 
2865         machine->device_memory->base = ROUND_UP(machine->ram_size,
2866                                                 SPAPR_DEVICE_MEM_ALIGN);
2867         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2868                            "device-memory", device_mem_size);
2869         memory_region_add_subregion(sysmem, machine->device_memory->base,
2870                                     &machine->device_memory->mr);
2871     }
2872 
2873     if (smc->dr_lmb_enabled) {
2874         spapr_create_lmb_dr_connectors(spapr);
2875     }
2876 
2877     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI) == SPAPR_CAP_ON) {
2878         /* Create the error string for live migration blocker */
2879         error_setg(&spapr->fwnmi_migration_blocker,
2880             "A machine check is being handled during migration. The handler"
2881             "may run and log hardware error on the destination");
2882     }
2883 
2884     if (mc->nvdimm_supported) {
2885         spapr_create_nvdimm_dr_connectors(spapr);
2886     }
2887 
2888     /* Set up RTAS event infrastructure */
2889     spapr_events_init(spapr);
2890 
2891     /* Set up the RTC RTAS interfaces */
2892     spapr_rtc_create(spapr);
2893 
2894     /* Set up VIO bus */
2895     spapr->vio_bus = spapr_vio_bus_init();
2896 
2897     for (i = 0; serial_hd(i); i++) {
2898         spapr_vty_create(spapr->vio_bus, serial_hd(i));
2899     }
2900 
2901     /* We always have at least the nvram device on VIO */
2902     spapr_create_nvram(spapr);
2903 
2904     /*
2905      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2906      * connectors (described in root DT node's "ibm,drc-types" property)
2907      * are pre-initialized here. additional child connectors (such as
2908      * connectors for a PHBs PCI slots) are added as needed during their
2909      * parent's realization.
2910      */
2911     if (smc->dr_phb_enabled) {
2912         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2913             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2914         }
2915     }
2916 
2917     /* Set up PCI */
2918     spapr_pci_rtas_init();
2919 
2920     phb = spapr_create_default_phb();
2921 
2922     for (i = 0; i < nb_nics; i++) {
2923         NICInfo *nd = &nd_table[i];
2924 
2925         if (!nd->model) {
2926             nd->model = g_strdup("spapr-vlan");
2927         }
2928 
2929         if (g_str_equal(nd->model, "spapr-vlan") ||
2930             g_str_equal(nd->model, "ibmveth")) {
2931             spapr_vlan_create(spapr->vio_bus, nd);
2932         } else {
2933             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2934         }
2935     }
2936 
2937     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2938         spapr_vscsi_create(spapr->vio_bus);
2939     }
2940 
2941     /* Graphics */
2942     if (spapr_vga_init(phb->bus, &error_fatal)) {
2943         spapr->has_graphics = true;
2944         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2945     }
2946 
2947     if (machine->usb) {
2948         if (smc->use_ohci_by_default) {
2949             pci_create_simple(phb->bus, -1, "pci-ohci");
2950         } else {
2951             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2952         }
2953 
2954         if (spapr->has_graphics) {
2955             USBBus *usb_bus = usb_bus_find(-1);
2956 
2957             usb_create_simple(usb_bus, "usb-kbd");
2958             usb_create_simple(usb_bus, "usb-mouse");
2959         }
2960     }
2961 
2962     if (kernel_filename) {
2963         spapr->kernel_size = load_elf(kernel_filename, NULL,
2964                                       translate_kernel_address, spapr,
2965                                       NULL, NULL, NULL, NULL, 1,
2966                                       PPC_ELF_MACHINE, 0, 0);
2967         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2968             spapr->kernel_size = load_elf(kernel_filename, NULL,
2969                                           translate_kernel_address, spapr,
2970                                           NULL, NULL, NULL, NULL, 0,
2971                                           PPC_ELF_MACHINE, 0, 0);
2972             spapr->kernel_le = spapr->kernel_size > 0;
2973         }
2974         if (spapr->kernel_size < 0) {
2975             error_report("error loading %s: %s", kernel_filename,
2976                          load_elf_strerror(spapr->kernel_size));
2977             exit(1);
2978         }
2979 
2980         /* load initrd */
2981         if (initrd_filename) {
2982             /* Try to locate the initrd in the gap between the kernel
2983              * and the firmware. Add a bit of space just in case
2984              */
2985             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2986                                   + 0x1ffff) & ~0xffff;
2987             spapr->initrd_size = load_image_targphys(initrd_filename,
2988                                                      spapr->initrd_base,
2989                                                      load_limit
2990                                                      - spapr->initrd_base);
2991             if (spapr->initrd_size < 0) {
2992                 error_report("could not load initial ram disk '%s'",
2993                              initrd_filename);
2994                 exit(1);
2995             }
2996         }
2997     }
2998 
2999     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3000     if (!filename) {
3001         error_report("Could not find LPAR firmware '%s'", bios_name);
3002         exit(1);
3003     }
3004     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3005     if (fw_size <= 0) {
3006         error_report("Could not load LPAR firmware '%s'", filename);
3007         exit(1);
3008     }
3009     g_free(filename);
3010 
3011     /* FIXME: Should register things through the MachineState's qdev
3012      * interface, this is a legacy from the sPAPREnvironment structure
3013      * which predated MachineState but had a similar function */
3014     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3015     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3016                          &savevm_htab_handlers, spapr);
3017 
3018     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine));
3019 
3020     qemu_register_boot_set(spapr_boot_set, spapr);
3021 
3022     /*
3023      * Nothing needs to be done to resume a suspended guest because
3024      * suspending does not change the machine state, so no need for
3025      * a ->wakeup method.
3026      */
3027     qemu_register_wakeup_support();
3028 
3029     if (kvm_enabled()) {
3030         /* to stop and start vmclock */
3031         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3032                                          &spapr->tb);
3033 
3034         kvmppc_spapr_enable_inkernel_multitce();
3035     }
3036 
3037     qemu_cond_init(&spapr->fwnmi_machine_check_interlock_cond);
3038     if (spapr->vof) {
3039         spapr->vof->fw_size = fw_size; /* for claim() on itself */
3040         spapr_register_hypercall(KVMPPC_H_VOF_CLIENT, spapr_h_vof_client);
3041     }
3042 }
3043 
3044 #define DEFAULT_KVM_TYPE "auto"
3045 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3046 {
3047     /*
3048      * The use of g_ascii_strcasecmp() for 'hv' and 'pr' is to
3049      * accomodate the 'HV' and 'PV' formats that exists in the
3050      * wild. The 'auto' mode is being introduced already as
3051      * lower-case, thus we don't need to bother checking for
3052      * "AUTO".
3053      */
3054     if (!vm_type || !strcmp(vm_type, DEFAULT_KVM_TYPE)) {
3055         return 0;
3056     }
3057 
3058     if (!g_ascii_strcasecmp(vm_type, "hv")) {
3059         return 1;
3060     }
3061 
3062     if (!g_ascii_strcasecmp(vm_type, "pr")) {
3063         return 2;
3064     }
3065 
3066     error_report("Unknown kvm-type specified '%s'", vm_type);
3067     exit(1);
3068 }
3069 
3070 /*
3071  * Implementation of an interface to adjust firmware path
3072  * for the bootindex property handling.
3073  */
3074 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3075                                    DeviceState *dev)
3076 {
3077 #define CAST(type, obj, name) \
3078     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3079     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3080     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3081     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3082     PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3083 
3084     if (d && bus) {
3085         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3086         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3087         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3088 
3089         if (spapr) {
3090             /*
3091              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3092              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3093              * 0x8000 | (target << 8) | (bus << 5) | lun
3094              * (see the "Logical unit addressing format" table in SAM5)
3095              */
3096             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3097             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3098                                    (uint64_t)id << 48);
3099         } else if (virtio) {
3100             /*
3101              * We use SRP luns of the form 01000000 | (target << 8) | lun
3102              * in the top 32 bits of the 64-bit LUN
3103              * Note: the quote above is from SLOF and it is wrong,
3104              * the actual binding is:
3105              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3106              */
3107             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3108             if (d->lun >= 256) {
3109                 /* Use the LUN "flat space addressing method" */
3110                 id |= 0x4000;
3111             }
3112             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3113                                    (uint64_t)id << 32);
3114         } else if (usb) {
3115             /*
3116              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3117              * in the top 32 bits of the 64-bit LUN
3118              */
3119             unsigned usb_port = atoi(usb->port->path);
3120             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3121             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3122                                    (uint64_t)id << 32);
3123         }
3124     }
3125 
3126     /*
3127      * SLOF probes the USB devices, and if it recognizes that the device is a
3128      * storage device, it changes its name to "storage" instead of "usb-host",
3129      * and additionally adds a child node for the SCSI LUN, so the correct
3130      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3131      */
3132     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3133         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3134         if (usb_device_is_scsi_storage(usbdev)) {
3135             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3136         }
3137     }
3138 
3139     if (phb) {
3140         /* Replace "pci" with "pci@800000020000000" */
3141         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3142     }
3143 
3144     if (vsc) {
3145         /* Same logic as virtio above */
3146         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3147         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3148     }
3149 
3150     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3151         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3152         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3153         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3154     }
3155 
3156     if (pcidev) {
3157         return spapr_pci_fw_dev_name(pcidev);
3158     }
3159 
3160     return NULL;
3161 }
3162 
3163 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3164 {
3165     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3166 
3167     return g_strdup(spapr->kvm_type);
3168 }
3169 
3170 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3171 {
3172     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3173 
3174     g_free(spapr->kvm_type);
3175     spapr->kvm_type = g_strdup(value);
3176 }
3177 
3178 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3179 {
3180     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3181 
3182     return spapr->use_hotplug_event_source;
3183 }
3184 
3185 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3186                                             Error **errp)
3187 {
3188     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3189 
3190     spapr->use_hotplug_event_source = value;
3191 }
3192 
3193 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3194 {
3195     return true;
3196 }
3197 
3198 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3199 {
3200     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3201 
3202     switch (spapr->resize_hpt) {
3203     case SPAPR_RESIZE_HPT_DEFAULT:
3204         return g_strdup("default");
3205     case SPAPR_RESIZE_HPT_DISABLED:
3206         return g_strdup("disabled");
3207     case SPAPR_RESIZE_HPT_ENABLED:
3208         return g_strdup("enabled");
3209     case SPAPR_RESIZE_HPT_REQUIRED:
3210         return g_strdup("required");
3211     }
3212     g_assert_not_reached();
3213 }
3214 
3215 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3216 {
3217     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3218 
3219     if (strcmp(value, "default") == 0) {
3220         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3221     } else if (strcmp(value, "disabled") == 0) {
3222         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3223     } else if (strcmp(value, "enabled") == 0) {
3224         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3225     } else if (strcmp(value, "required") == 0) {
3226         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3227     } else {
3228         error_setg(errp, "Bad value for \"resize-hpt\" property");
3229     }
3230 }
3231 
3232 static bool spapr_get_vof(Object *obj, Error **errp)
3233 {
3234     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3235 
3236     return spapr->vof != NULL;
3237 }
3238 
3239 static void spapr_set_vof(Object *obj, bool value, Error **errp)
3240 {
3241     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3242 
3243     if (spapr->vof) {
3244         vof_cleanup(spapr->vof);
3245         g_free(spapr->vof);
3246         spapr->vof = NULL;
3247     }
3248     if (!value) {
3249         return;
3250     }
3251     spapr->vof = g_malloc0(sizeof(*spapr->vof));
3252 }
3253 
3254 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3255 {
3256     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3257 
3258     if (spapr->irq == &spapr_irq_xics_legacy) {
3259         return g_strdup("legacy");
3260     } else if (spapr->irq == &spapr_irq_xics) {
3261         return g_strdup("xics");
3262     } else if (spapr->irq == &spapr_irq_xive) {
3263         return g_strdup("xive");
3264     } else if (spapr->irq == &spapr_irq_dual) {
3265         return g_strdup("dual");
3266     }
3267     g_assert_not_reached();
3268 }
3269 
3270 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3271 {
3272     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3273 
3274     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3275         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3276         return;
3277     }
3278 
3279     /* The legacy IRQ backend can not be set */
3280     if (strcmp(value, "xics") == 0) {
3281         spapr->irq = &spapr_irq_xics;
3282     } else if (strcmp(value, "xive") == 0) {
3283         spapr->irq = &spapr_irq_xive;
3284     } else if (strcmp(value, "dual") == 0) {
3285         spapr->irq = &spapr_irq_dual;
3286     } else {
3287         error_setg(errp, "Bad value for \"ic-mode\" property");
3288     }
3289 }
3290 
3291 static char *spapr_get_host_model(Object *obj, Error **errp)
3292 {
3293     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3294 
3295     return g_strdup(spapr->host_model);
3296 }
3297 
3298 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3299 {
3300     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3301 
3302     g_free(spapr->host_model);
3303     spapr->host_model = g_strdup(value);
3304 }
3305 
3306 static char *spapr_get_host_serial(Object *obj, Error **errp)
3307 {
3308     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3309 
3310     return g_strdup(spapr->host_serial);
3311 }
3312 
3313 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3314 {
3315     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3316 
3317     g_free(spapr->host_serial);
3318     spapr->host_serial = g_strdup(value);
3319 }
3320 
3321 static void spapr_instance_init(Object *obj)
3322 {
3323     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3324     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3325     MachineState *ms = MACHINE(spapr);
3326     MachineClass *mc = MACHINE_GET_CLASS(ms);
3327 
3328     /*
3329      * NVDIMM support went live in 5.1 without considering that, in
3330      * other archs, the user needs to enable NVDIMM support with the
3331      * 'nvdimm' machine option and the default behavior is NVDIMM
3332      * support disabled. It is too late to roll back to the standard
3333      * behavior without breaking 5.1 guests.
3334      */
3335     if (mc->nvdimm_supported) {
3336         ms->nvdimms_state->is_enabled = true;
3337     }
3338 
3339     spapr->htab_fd = -1;
3340     spapr->use_hotplug_event_source = true;
3341     spapr->kvm_type = g_strdup(DEFAULT_KVM_TYPE);
3342     object_property_add_str(obj, "kvm-type",
3343                             spapr_get_kvm_type, spapr_set_kvm_type);
3344     object_property_set_description(obj, "kvm-type",
3345                                     "Specifies the KVM virtualization mode (auto,"
3346                                     " hv, pr). Defaults to 'auto'. This mode will use"
3347                                     " any available KVM module loaded in the host,"
3348                                     " where kvm_hv takes precedence if both kvm_hv and"
3349                                     " kvm_pr are loaded.");
3350     object_property_add_bool(obj, "modern-hotplug-events",
3351                             spapr_get_modern_hotplug_events,
3352                             spapr_set_modern_hotplug_events);
3353     object_property_set_description(obj, "modern-hotplug-events",
3354                                     "Use dedicated hotplug event mechanism in"
3355                                     " place of standard EPOW events when possible"
3356                                     " (required for memory hot-unplug support)");
3357     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3358                             "Maximum permitted CPU compatibility mode");
3359 
3360     object_property_add_str(obj, "resize-hpt",
3361                             spapr_get_resize_hpt, spapr_set_resize_hpt);
3362     object_property_set_description(obj, "resize-hpt",
3363                                     "Resizing of the Hash Page Table (enabled, disabled, required)");
3364     object_property_add_uint32_ptr(obj, "vsmt",
3365                                    &spapr->vsmt, OBJ_PROP_FLAG_READWRITE);
3366     object_property_set_description(obj, "vsmt",
3367                                     "Virtual SMT: KVM behaves as if this were"
3368                                     " the host's SMT mode");
3369 
3370     object_property_add_bool(obj, "vfio-no-msix-emulation",
3371                              spapr_get_msix_emulation, NULL);
3372 
3373     object_property_add_uint64_ptr(obj, "kernel-addr",
3374                                    &spapr->kernel_addr, OBJ_PROP_FLAG_READWRITE);
3375     object_property_set_description(obj, "kernel-addr",
3376                                     stringify(KERNEL_LOAD_ADDR)
3377                                     " for -kernel is the default");
3378     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3379 
3380     object_property_add_bool(obj, "x-vof", spapr_get_vof, spapr_set_vof);
3381     object_property_set_description(obj, "x-vof",
3382                                     "Enable Virtual Open Firmware (experimental)");
3383 
3384     /* The machine class defines the default interrupt controller mode */
3385     spapr->irq = smc->irq;
3386     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3387                             spapr_set_ic_mode);
3388     object_property_set_description(obj, "ic-mode",
3389                  "Specifies the interrupt controller mode (xics, xive, dual)");
3390 
3391     object_property_add_str(obj, "host-model",
3392         spapr_get_host_model, spapr_set_host_model);
3393     object_property_set_description(obj, "host-model",
3394         "Host model to advertise in guest device tree");
3395     object_property_add_str(obj, "host-serial",
3396         spapr_get_host_serial, spapr_set_host_serial);
3397     object_property_set_description(obj, "host-serial",
3398         "Host serial number to advertise in guest device tree");
3399 }
3400 
3401 static void spapr_machine_finalizefn(Object *obj)
3402 {
3403     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3404 
3405     g_free(spapr->kvm_type);
3406 }
3407 
3408 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3409 {
3410     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3411     PowerPCCPU *cpu = POWERPC_CPU(cs);
3412     CPUPPCState *env = &cpu->env;
3413 
3414     cpu_synchronize_state(cs);
3415     /* If FWNMI is inactive, addr will be -1, which will deliver to 0x100 */
3416     if (spapr->fwnmi_system_reset_addr != -1) {
3417         uint64_t rtas_addr, addr;
3418 
3419         /* get rtas addr from fdt */
3420         rtas_addr = spapr_get_rtas_addr();
3421         if (!rtas_addr) {
3422             qemu_system_guest_panicked(NULL);
3423             return;
3424         }
3425 
3426         addr = rtas_addr + RTAS_ERROR_LOG_MAX + cs->cpu_index * sizeof(uint64_t)*2;
3427         stq_be_phys(&address_space_memory, addr, env->gpr[3]);
3428         stq_be_phys(&address_space_memory, addr + sizeof(uint64_t), 0);
3429         env->gpr[3] = addr;
3430     }
3431     ppc_cpu_do_system_reset(cs);
3432     if (spapr->fwnmi_system_reset_addr != -1) {
3433         env->nip = spapr->fwnmi_system_reset_addr;
3434     }
3435 }
3436 
3437 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3438 {
3439     CPUState *cs;
3440 
3441     CPU_FOREACH(cs) {
3442         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3443     }
3444 }
3445 
3446 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3447                           void *fdt, int *fdt_start_offset, Error **errp)
3448 {
3449     uint64_t addr;
3450     uint32_t node;
3451 
3452     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3453     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3454                                     &error_abort);
3455     *fdt_start_offset = spapr_dt_memory_node(spapr, fdt, node, addr,
3456                                              SPAPR_MEMORY_BLOCK_SIZE);
3457     return 0;
3458 }
3459 
3460 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3461                            bool dedicated_hp_event_source)
3462 {
3463     SpaprDrc *drc;
3464     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3465     int i;
3466     uint64_t addr = addr_start;
3467     bool hotplugged = spapr_drc_hotplugged(dev);
3468 
3469     for (i = 0; i < nr_lmbs; i++) {
3470         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3471                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3472         g_assert(drc);
3473 
3474         /*
3475          * memory_device_get_free_addr() provided a range of free addresses
3476          * that doesn't overlap with any existing mapping at pre-plug. The
3477          * corresponding LMB DRCs are thus assumed to be all attachable.
3478          */
3479         spapr_drc_attach(drc, dev);
3480         if (!hotplugged) {
3481             spapr_drc_reset(drc);
3482         }
3483         addr += SPAPR_MEMORY_BLOCK_SIZE;
3484     }
3485     /* send hotplug notification to the
3486      * guest only in case of hotplugged memory
3487      */
3488     if (hotplugged) {
3489         if (dedicated_hp_event_source) {
3490             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3491                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3492             g_assert(drc);
3493             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3494                                                    nr_lmbs,
3495                                                    spapr_drc_index(drc));
3496         } else {
3497             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3498                                            nr_lmbs);
3499         }
3500     }
3501 }
3502 
3503 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3504 {
3505     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3506     PCDIMMDevice *dimm = PC_DIMM(dev);
3507     uint64_t size, addr;
3508     int64_t slot;
3509     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3510 
3511     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3512 
3513     pc_dimm_plug(dimm, MACHINE(ms));
3514 
3515     if (!is_nvdimm) {
3516         addr = object_property_get_uint(OBJECT(dimm),
3517                                         PC_DIMM_ADDR_PROP, &error_abort);
3518         spapr_add_lmbs(dev, addr, size,
3519                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT));
3520     } else {
3521         slot = object_property_get_int(OBJECT(dimm),
3522                                        PC_DIMM_SLOT_PROP, &error_abort);
3523         /* We should have valid slot number at this point */
3524         g_assert(slot >= 0);
3525         spapr_add_nvdimm(dev, slot);
3526     }
3527 }
3528 
3529 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3530                                   Error **errp)
3531 {
3532     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3533     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3534     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3535     PCDIMMDevice *dimm = PC_DIMM(dev);
3536     Error *local_err = NULL;
3537     uint64_t size;
3538     Object *memdev;
3539     hwaddr pagesize;
3540 
3541     if (!smc->dr_lmb_enabled) {
3542         error_setg(errp, "Memory hotplug not supported for this machine");
3543         return;
3544     }
3545 
3546     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3547     if (local_err) {
3548         error_propagate(errp, local_err);
3549         return;
3550     }
3551 
3552     if (is_nvdimm) {
3553         if (!spapr_nvdimm_validate(hotplug_dev, NVDIMM(dev), size, errp)) {
3554             return;
3555         }
3556     } else if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3557         error_setg(errp, "Hotplugged memory size must be a multiple of "
3558                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3559         return;
3560     }
3561 
3562     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3563                                       &error_abort);
3564     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3565     if (!spapr_check_pagesize(spapr, pagesize, errp)) {
3566         return;
3567     }
3568 
3569     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3570 }
3571 
3572 struct SpaprDimmState {
3573     PCDIMMDevice *dimm;
3574     uint32_t nr_lmbs;
3575     QTAILQ_ENTRY(SpaprDimmState) next;
3576 };
3577 
3578 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3579                                                        PCDIMMDevice *dimm)
3580 {
3581     SpaprDimmState *dimm_state = NULL;
3582 
3583     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3584         if (dimm_state->dimm == dimm) {
3585             break;
3586         }
3587     }
3588     return dimm_state;
3589 }
3590 
3591 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3592                                                       uint32_t nr_lmbs,
3593                                                       PCDIMMDevice *dimm)
3594 {
3595     SpaprDimmState *ds = NULL;
3596 
3597     /*
3598      * If this request is for a DIMM whose removal had failed earlier
3599      * (due to guest's refusal to remove the LMBs), we would have this
3600      * dimm already in the pending_dimm_unplugs list. In that
3601      * case don't add again.
3602      */
3603     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3604     if (!ds) {
3605         ds = g_malloc0(sizeof(SpaprDimmState));
3606         ds->nr_lmbs = nr_lmbs;
3607         ds->dimm = dimm;
3608         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3609     }
3610     return ds;
3611 }
3612 
3613 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3614                                               SpaprDimmState *dimm_state)
3615 {
3616     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3617     g_free(dimm_state);
3618 }
3619 
3620 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3621                                                         PCDIMMDevice *dimm)
3622 {
3623     SpaprDrc *drc;
3624     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3625                                                   &error_abort);
3626     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3627     uint32_t avail_lmbs = 0;
3628     uint64_t addr_start, addr;
3629     int i;
3630 
3631     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3632                                           &error_abort);
3633 
3634     addr = addr_start;
3635     for (i = 0; i < nr_lmbs; i++) {
3636         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3637                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3638         g_assert(drc);
3639         if (drc->dev) {
3640             avail_lmbs++;
3641         }
3642         addr += SPAPR_MEMORY_BLOCK_SIZE;
3643     }
3644 
3645     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3646 }
3647 
3648 void spapr_memory_unplug_rollback(SpaprMachineState *spapr, DeviceState *dev)
3649 {
3650     SpaprDimmState *ds;
3651     PCDIMMDevice *dimm;
3652     SpaprDrc *drc;
3653     uint32_t nr_lmbs;
3654     uint64_t size, addr_start, addr;
3655     g_autofree char *qapi_error = NULL;
3656     int i;
3657 
3658     if (!dev) {
3659         return;
3660     }
3661 
3662     dimm = PC_DIMM(dev);
3663     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3664 
3665     /*
3666      * 'ds == NULL' would mean that the DIMM doesn't have a pending
3667      * unplug state, but one of its DRC is marked as unplug_requested.
3668      * This is bad and weird enough to g_assert() out.
3669      */
3670     g_assert(ds);
3671 
3672     spapr_pending_dimm_unplugs_remove(spapr, ds);
3673 
3674     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3675     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3676 
3677     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3678                                           &error_abort);
3679 
3680     addr = addr_start;
3681     for (i = 0; i < nr_lmbs; i++) {
3682         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3683                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3684         g_assert(drc);
3685 
3686         drc->unplug_requested = false;
3687         addr += SPAPR_MEMORY_BLOCK_SIZE;
3688     }
3689 
3690     /*
3691      * Tell QAPI that something happened and the memory
3692      * hotunplug wasn't successful. Keep sending
3693      * MEM_UNPLUG_ERROR even while sending
3694      * DEVICE_UNPLUG_GUEST_ERROR until the deprecation of
3695      * MEM_UNPLUG_ERROR is due.
3696      */
3697     qapi_error = g_strdup_printf("Memory hotunplug rejected by the guest "
3698                                  "for device %s", dev->id);
3699 
3700     qapi_event_send_mem_unplug_error(dev->id ? : "", qapi_error);
3701 
3702     qapi_event_send_device_unplug_guest_error(!!dev->id, dev->id,
3703                                               dev->canonical_path);
3704 }
3705 
3706 /* Callback to be called during DRC release. */
3707 void spapr_lmb_release(DeviceState *dev)
3708 {
3709     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3710     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3711     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3712 
3713     /* This information will get lost if a migration occurs
3714      * during the unplug process. In this case recover it. */
3715     if (ds == NULL) {
3716         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3717         g_assert(ds);
3718         /* The DRC being examined by the caller at least must be counted */
3719         g_assert(ds->nr_lmbs);
3720     }
3721 
3722     if (--ds->nr_lmbs) {
3723         return;
3724     }
3725 
3726     /*
3727      * Now that all the LMBs have been removed by the guest, call the
3728      * unplug handler chain. This can never fail.
3729      */
3730     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3731     object_unparent(OBJECT(dev));
3732 }
3733 
3734 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3735 {
3736     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3737     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3738 
3739     /* We really shouldn't get this far without anything to unplug */
3740     g_assert(ds);
3741 
3742     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3743     qdev_unrealize(dev);
3744     spapr_pending_dimm_unplugs_remove(spapr, ds);
3745 }
3746 
3747 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3748                                         DeviceState *dev, Error **errp)
3749 {
3750     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3751     PCDIMMDevice *dimm = PC_DIMM(dev);
3752     uint32_t nr_lmbs;
3753     uint64_t size, addr_start, addr;
3754     int i;
3755     SpaprDrc *drc;
3756 
3757     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3758         error_setg(errp, "nvdimm device hot unplug is not supported yet.");
3759         return;
3760     }
3761 
3762     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3763     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3764 
3765     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3766                                           &error_abort);
3767 
3768     /*
3769      * An existing pending dimm state for this DIMM means that there is an
3770      * unplug operation in progress, waiting for the spapr_lmb_release
3771      * callback to complete the job (BQL can't cover that far). In this case,
3772      * bail out to avoid detaching DRCs that were already released.
3773      */
3774     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3775         error_setg(errp, "Memory unplug already in progress for device %s",
3776                    dev->id);
3777         return;
3778     }
3779 
3780     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3781 
3782     addr = addr_start;
3783     for (i = 0; i < nr_lmbs; i++) {
3784         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3785                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3786         g_assert(drc);
3787 
3788         spapr_drc_unplug_request(drc);
3789         addr += SPAPR_MEMORY_BLOCK_SIZE;
3790     }
3791 
3792     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3793                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3794     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3795                                               nr_lmbs, spapr_drc_index(drc));
3796 }
3797 
3798 /* Callback to be called during DRC release. */
3799 void spapr_core_release(DeviceState *dev)
3800 {
3801     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3802 
3803     /* Call the unplug handler chain. This can never fail. */
3804     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3805     object_unparent(OBJECT(dev));
3806 }
3807 
3808 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3809 {
3810     MachineState *ms = MACHINE(hotplug_dev);
3811     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3812     CPUCore *cc = CPU_CORE(dev);
3813     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3814 
3815     if (smc->pre_2_10_has_unused_icps) {
3816         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3817         int i;
3818 
3819         for (i = 0; i < cc->nr_threads; i++) {
3820             CPUState *cs = CPU(sc->threads[i]);
3821 
3822             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3823         }
3824     }
3825 
3826     assert(core_slot);
3827     core_slot->cpu = NULL;
3828     qdev_unrealize(dev);
3829 }
3830 
3831 static
3832 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3833                                Error **errp)
3834 {
3835     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3836     int index;
3837     SpaprDrc *drc;
3838     CPUCore *cc = CPU_CORE(dev);
3839 
3840     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3841         error_setg(errp, "Unable to find CPU core with core-id: %d",
3842                    cc->core_id);
3843         return;
3844     }
3845     if (index == 0) {
3846         error_setg(errp, "Boot CPU core may not be unplugged");
3847         return;
3848     }
3849 
3850     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3851                           spapr_vcpu_id(spapr, cc->core_id));
3852     g_assert(drc);
3853 
3854     if (!spapr_drc_unplug_requested(drc)) {
3855         spapr_drc_unplug_request(drc);
3856     }
3857 
3858     /*
3859      * spapr_hotplug_req_remove_by_index is left unguarded, out of the
3860      * "!spapr_drc_unplug_requested" check, to allow for multiple IRQ
3861      * pulses removing the same CPU. Otherwise, in an failed hotunplug
3862      * attempt (e.g. the kernel will refuse to remove the last online
3863      * CPU), we will never attempt it again because unplug_requested
3864      * will still be 'true' in that case.
3865      */
3866     spapr_hotplug_req_remove_by_index(drc);
3867 }
3868 
3869 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3870                            void *fdt, int *fdt_start_offset, Error **errp)
3871 {
3872     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3873     CPUState *cs = CPU(core->threads[0]);
3874     PowerPCCPU *cpu = POWERPC_CPU(cs);
3875     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3876     int id = spapr_get_vcpu_id(cpu);
3877     g_autofree char *nodename = NULL;
3878     int offset;
3879 
3880     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3881     offset = fdt_add_subnode(fdt, 0, nodename);
3882 
3883     spapr_dt_cpu(cs, fdt, offset, spapr);
3884 
3885     /*
3886      * spapr_dt_cpu() does not fill the 'name' property in the
3887      * CPU node. The function is called during boot process, before
3888      * and after CAS, and overwriting the 'name' property written
3889      * by SLOF is not allowed.
3890      *
3891      * Write it manually after spapr_dt_cpu(). This makes the hotplug
3892      * CPUs more compatible with the coldplugged ones, which have
3893      * the 'name' property. Linux Kernel also relies on this
3894      * property to identify CPU nodes.
3895      */
3896     _FDT((fdt_setprop_string(fdt, offset, "name", nodename)));
3897 
3898     *fdt_start_offset = offset;
3899     return 0;
3900 }
3901 
3902 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
3903 {
3904     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3905     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3906     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3907     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3908     CPUCore *cc = CPU_CORE(dev);
3909     CPUState *cs;
3910     SpaprDrc *drc;
3911     CPUArchId *core_slot;
3912     int index;
3913     bool hotplugged = spapr_drc_hotplugged(dev);
3914     int i;
3915 
3916     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3917     g_assert(core_slot); /* Already checked in spapr_core_pre_plug() */
3918 
3919     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3920                           spapr_vcpu_id(spapr, cc->core_id));
3921 
3922     g_assert(drc || !mc->has_hotpluggable_cpus);
3923 
3924     if (drc) {
3925         /*
3926          * spapr_core_pre_plug() already buys us this is a brand new
3927          * core being plugged into a free slot. Nothing should already
3928          * be attached to the corresponding DRC.
3929          */
3930         spapr_drc_attach(drc, dev);
3931 
3932         if (hotplugged) {
3933             /*
3934              * Send hotplug notification interrupt to the guest only
3935              * in case of hotplugged CPUs.
3936              */
3937             spapr_hotplug_req_add_by_index(drc);
3938         } else {
3939             spapr_drc_reset(drc);
3940         }
3941     }
3942 
3943     core_slot->cpu = OBJECT(dev);
3944 
3945     /*
3946      * Set compatibility mode to match the boot CPU, which was either set
3947      * by the machine reset code or by CAS. This really shouldn't fail at
3948      * this point.
3949      */
3950     if (hotplugged) {
3951         for (i = 0; i < cc->nr_threads; i++) {
3952             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3953                            &error_abort);
3954         }
3955     }
3956 
3957     if (smc->pre_2_10_has_unused_icps) {
3958         for (i = 0; i < cc->nr_threads; i++) {
3959             cs = CPU(core->threads[i]);
3960             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3961         }
3962     }
3963 }
3964 
3965 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3966                                 Error **errp)
3967 {
3968     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3969     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3970     CPUCore *cc = CPU_CORE(dev);
3971     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3972     const char *type = object_get_typename(OBJECT(dev));
3973     CPUArchId *core_slot;
3974     int index;
3975     unsigned int smp_threads = machine->smp.threads;
3976 
3977     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3978         error_setg(errp, "CPU hotplug not supported for this machine");
3979         return;
3980     }
3981 
3982     if (strcmp(base_core_type, type)) {
3983         error_setg(errp, "CPU core type should be %s", base_core_type);
3984         return;
3985     }
3986 
3987     if (cc->core_id % smp_threads) {
3988         error_setg(errp, "invalid core id %d", cc->core_id);
3989         return;
3990     }
3991 
3992     /*
3993      * In general we should have homogeneous threads-per-core, but old
3994      * (pre hotplug support) machine types allow the last core to have
3995      * reduced threads as a compatibility hack for when we allowed
3996      * total vcpus not a multiple of threads-per-core.
3997      */
3998     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3999         error_setg(errp, "invalid nr-threads %d, must be %d", cc->nr_threads,
4000                    smp_threads);
4001         return;
4002     }
4003 
4004     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
4005     if (!core_slot) {
4006         error_setg(errp, "core id %d out of range", cc->core_id);
4007         return;
4008     }
4009 
4010     if (core_slot->cpu) {
4011         error_setg(errp, "core %d already populated", cc->core_id);
4012         return;
4013     }
4014 
4015     numa_cpu_pre_plug(core_slot, dev, errp);
4016 }
4017 
4018 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
4019                           void *fdt, int *fdt_start_offset, Error **errp)
4020 {
4021     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
4022     int intc_phandle;
4023 
4024     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
4025     if (intc_phandle <= 0) {
4026         return -1;
4027     }
4028 
4029     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
4030         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
4031         return -1;
4032     }
4033 
4034     /* generally SLOF creates these, for hotplug it's up to QEMU */
4035     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
4036 
4037     return 0;
4038 }
4039 
4040 static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4041                                Error **errp)
4042 {
4043     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4044     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4045     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4046     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
4047     SpaprDrc *drc;
4048 
4049     if (dev->hotplugged && !smc->dr_phb_enabled) {
4050         error_setg(errp, "PHB hotplug not supported for this machine");
4051         return false;
4052     }
4053 
4054     if (sphb->index == (uint32_t)-1) {
4055         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
4056         return false;
4057     }
4058 
4059     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4060     if (drc && drc->dev) {
4061         error_setg(errp, "PHB %d already attached", sphb->index);
4062         return false;
4063     }
4064 
4065     /*
4066      * This will check that sphb->index doesn't exceed the maximum number of
4067      * PHBs for the current machine type.
4068      */
4069     return
4070         smc->phb_placement(spapr, sphb->index,
4071                            &sphb->buid, &sphb->io_win_addr,
4072                            &sphb->mem_win_addr, &sphb->mem64_win_addr,
4073                            windows_supported, sphb->dma_liobn,
4074                            &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4075                            errp);
4076 }
4077 
4078 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4079 {
4080     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4081     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4082     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4083     SpaprDrc *drc;
4084     bool hotplugged = spapr_drc_hotplugged(dev);
4085 
4086     if (!smc->dr_phb_enabled) {
4087         return;
4088     }
4089 
4090     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4091     /* hotplug hooks should check it's enabled before getting this far */
4092     assert(drc);
4093 
4094     /* spapr_phb_pre_plug() already checked the DRC is attachable */
4095     spapr_drc_attach(drc, dev);
4096 
4097     if (hotplugged) {
4098         spapr_hotplug_req_add_by_index(drc);
4099     } else {
4100         spapr_drc_reset(drc);
4101     }
4102 }
4103 
4104 void spapr_phb_release(DeviceState *dev)
4105 {
4106     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4107 
4108     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4109     object_unparent(OBJECT(dev));
4110 }
4111 
4112 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4113 {
4114     qdev_unrealize(dev);
4115 }
4116 
4117 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4118                                      DeviceState *dev, Error **errp)
4119 {
4120     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4121     SpaprDrc *drc;
4122 
4123     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4124     assert(drc);
4125 
4126     if (!spapr_drc_unplug_requested(drc)) {
4127         spapr_drc_unplug_request(drc);
4128         spapr_hotplug_req_remove_by_index(drc);
4129     } else {
4130         error_setg(errp,
4131                    "PCI Host Bridge unplug already in progress for device %s",
4132                    dev->id);
4133     }
4134 }
4135 
4136 static
4137 bool spapr_tpm_proxy_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4138                               Error **errp)
4139 {
4140     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4141 
4142     if (spapr->tpm_proxy != NULL) {
4143         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4144         return false;
4145     }
4146 
4147     return true;
4148 }
4149 
4150 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
4151 {
4152     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4153     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4154 
4155     /* Already checked in spapr_tpm_proxy_pre_plug() */
4156     g_assert(spapr->tpm_proxy == NULL);
4157 
4158     spapr->tpm_proxy = tpm_proxy;
4159 }
4160 
4161 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4162 {
4163     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4164 
4165     qdev_unrealize(dev);
4166     object_unparent(OBJECT(dev));
4167     spapr->tpm_proxy = NULL;
4168 }
4169 
4170 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4171                                       DeviceState *dev, Error **errp)
4172 {
4173     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4174         spapr_memory_plug(hotplug_dev, dev);
4175     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4176         spapr_core_plug(hotplug_dev, dev);
4177     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4178         spapr_phb_plug(hotplug_dev, dev);
4179     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4180         spapr_tpm_proxy_plug(hotplug_dev, dev);
4181     }
4182 }
4183 
4184 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4185                                         DeviceState *dev, Error **errp)
4186 {
4187     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4188         spapr_memory_unplug(hotplug_dev, dev);
4189     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4190         spapr_core_unplug(hotplug_dev, dev);
4191     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4192         spapr_phb_unplug(hotplug_dev, dev);
4193     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4194         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4195     }
4196 }
4197 
4198 bool spapr_memory_hot_unplug_supported(SpaprMachineState *spapr)
4199 {
4200     return spapr_ovec_test(spapr->ov5_cas, OV5_HP_EVT) ||
4201         /*
4202          * CAS will process all pending unplug requests.
4203          *
4204          * HACK: a guest could theoretically have cleared all bits in OV5,
4205          * but none of the guests we care for do.
4206          */
4207         spapr_ovec_empty(spapr->ov5_cas);
4208 }
4209 
4210 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4211                                                 DeviceState *dev, Error **errp)
4212 {
4213     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4214     MachineClass *mc = MACHINE_GET_CLASS(sms);
4215     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4216 
4217     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4218         if (spapr_memory_hot_unplug_supported(sms)) {
4219             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4220         } else {
4221             error_setg(errp, "Memory hot unplug not supported for this guest");
4222         }
4223     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4224         if (!mc->has_hotpluggable_cpus) {
4225             error_setg(errp, "CPU hot unplug not supported on this machine");
4226             return;
4227         }
4228         spapr_core_unplug_request(hotplug_dev, dev, errp);
4229     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4230         if (!smc->dr_phb_enabled) {
4231             error_setg(errp, "PHB hot unplug not supported on this machine");
4232             return;
4233         }
4234         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4235     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4236         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4237     }
4238 }
4239 
4240 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4241                                           DeviceState *dev, Error **errp)
4242 {
4243     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4244         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4245     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4246         spapr_core_pre_plug(hotplug_dev, dev, errp);
4247     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4248         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4249     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4250         spapr_tpm_proxy_pre_plug(hotplug_dev, dev, errp);
4251     }
4252 }
4253 
4254 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4255                                                  DeviceState *dev)
4256 {
4257     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4258         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4259         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4260         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4261         return HOTPLUG_HANDLER(machine);
4262     }
4263     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4264         PCIDevice *pcidev = PCI_DEVICE(dev);
4265         PCIBus *root = pci_device_root_bus(pcidev);
4266         SpaprPhbState *phb =
4267             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4268                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4269 
4270         if (phb) {
4271             return HOTPLUG_HANDLER(phb);
4272         }
4273     }
4274     return NULL;
4275 }
4276 
4277 static CpuInstanceProperties
4278 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4279 {
4280     CPUArchId *core_slot;
4281     MachineClass *mc = MACHINE_GET_CLASS(machine);
4282 
4283     /* make sure possible_cpu are intialized */
4284     mc->possible_cpu_arch_ids(machine);
4285     /* get CPU core slot containing thread that matches cpu_index */
4286     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4287     assert(core_slot);
4288     return core_slot->props;
4289 }
4290 
4291 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4292 {
4293     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4294 }
4295 
4296 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4297 {
4298     int i;
4299     unsigned int smp_threads = machine->smp.threads;
4300     unsigned int smp_cpus = machine->smp.cpus;
4301     const char *core_type;
4302     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4303     MachineClass *mc = MACHINE_GET_CLASS(machine);
4304 
4305     if (!mc->has_hotpluggable_cpus) {
4306         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4307     }
4308     if (machine->possible_cpus) {
4309         assert(machine->possible_cpus->len == spapr_max_cores);
4310         return machine->possible_cpus;
4311     }
4312 
4313     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4314     if (!core_type) {
4315         error_report("Unable to find sPAPR CPU Core definition");
4316         exit(1);
4317     }
4318 
4319     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4320                              sizeof(CPUArchId) * spapr_max_cores);
4321     machine->possible_cpus->len = spapr_max_cores;
4322     for (i = 0; i < machine->possible_cpus->len; i++) {
4323         int core_id = i * smp_threads;
4324 
4325         machine->possible_cpus->cpus[i].type = core_type;
4326         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4327         machine->possible_cpus->cpus[i].arch_id = core_id;
4328         machine->possible_cpus->cpus[i].props.has_core_id = true;
4329         machine->possible_cpus->cpus[i].props.core_id = core_id;
4330     }
4331     return machine->possible_cpus;
4332 }
4333 
4334 static bool spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4335                                 uint64_t *buid, hwaddr *pio,
4336                                 hwaddr *mmio32, hwaddr *mmio64,
4337                                 unsigned n_dma, uint32_t *liobns,
4338                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4339 {
4340     /*
4341      * New-style PHB window placement.
4342      *
4343      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4344      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4345      * windows.
4346      *
4347      * Some guest kernels can't work with MMIO windows above 1<<46
4348      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4349      *
4350      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4351      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4352      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4353      * 1TiB 64-bit MMIO windows for each PHB.
4354      */
4355     const uint64_t base_buid = 0x800000020000000ULL;
4356     int i;
4357 
4358     /* Sanity check natural alignments */
4359     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4360     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4361     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4362     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4363     /* Sanity check bounds */
4364     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4365                       SPAPR_PCI_MEM32_WIN_SIZE);
4366     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4367                       SPAPR_PCI_MEM64_WIN_SIZE);
4368 
4369     if (index >= SPAPR_MAX_PHBS) {
4370         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4371                    SPAPR_MAX_PHBS - 1);
4372         return false;
4373     }
4374 
4375     *buid = base_buid + index;
4376     for (i = 0; i < n_dma; ++i) {
4377         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4378     }
4379 
4380     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4381     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4382     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4383 
4384     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4385     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4386     return true;
4387 }
4388 
4389 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4390 {
4391     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4392 
4393     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4394 }
4395 
4396 static void spapr_ics_resend(XICSFabric *dev)
4397 {
4398     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4399 
4400     ics_resend(spapr->ics);
4401 }
4402 
4403 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4404 {
4405     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4406 
4407     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4408 }
4409 
4410 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4411                                  Monitor *mon)
4412 {
4413     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4414 
4415     spapr_irq_print_info(spapr, mon);
4416     monitor_printf(mon, "irqchip: %s\n",
4417                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4418 }
4419 
4420 /*
4421  * This is a XIVE only operation
4422  */
4423 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4424                            uint8_t nvt_blk, uint32_t nvt_idx,
4425                            bool cam_ignore, uint8_t priority,
4426                            uint32_t logic_serv, XiveTCTXMatch *match)
4427 {
4428     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4429     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4430     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4431     int count;
4432 
4433     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4434                            priority, logic_serv, match);
4435     if (count < 0) {
4436         return count;
4437     }
4438 
4439     /*
4440      * When we implement the save and restore of the thread interrupt
4441      * contexts in the enter/exit CPU handlers of the machine and the
4442      * escalations in QEMU, we should be able to handle non dispatched
4443      * vCPUs.
4444      *
4445      * Until this is done, the sPAPR machine should find at least one
4446      * matching context always.
4447      */
4448     if (count == 0) {
4449         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4450                       nvt_blk, nvt_idx);
4451     }
4452 
4453     return count;
4454 }
4455 
4456 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4457 {
4458     return cpu->vcpu_id;
4459 }
4460 
4461 bool spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4462 {
4463     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4464     MachineState *ms = MACHINE(spapr);
4465     int vcpu_id;
4466 
4467     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4468 
4469     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4470         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4471         error_append_hint(errp, "Adjust the number of cpus to %d "
4472                           "or try to raise the number of threads per core\n",
4473                           vcpu_id * ms->smp.threads / spapr->vsmt);
4474         return false;
4475     }
4476 
4477     cpu->vcpu_id = vcpu_id;
4478     return true;
4479 }
4480 
4481 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4482 {
4483     CPUState *cs;
4484 
4485     CPU_FOREACH(cs) {
4486         PowerPCCPU *cpu = POWERPC_CPU(cs);
4487 
4488         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4489             return cpu;
4490         }
4491     }
4492 
4493     return NULL;
4494 }
4495 
4496 static bool spapr_cpu_in_nested(PowerPCCPU *cpu)
4497 {
4498     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4499 
4500     return spapr_cpu->in_nested;
4501 }
4502 
4503 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4504 {
4505     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4506 
4507     /* These are only called by TCG, KVM maintains dispatch state */
4508 
4509     spapr_cpu->prod = false;
4510     if (spapr_cpu->vpa_addr) {
4511         CPUState *cs = CPU(cpu);
4512         uint32_t dispatch;
4513 
4514         dispatch = ldl_be_phys(cs->as,
4515                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4516         dispatch++;
4517         if ((dispatch & 1) != 0) {
4518             qemu_log_mask(LOG_GUEST_ERROR,
4519                           "VPA: incorrect dispatch counter value for "
4520                           "dispatched partition %u, correcting.\n", dispatch);
4521             dispatch++;
4522         }
4523         stl_be_phys(cs->as,
4524                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4525     }
4526 }
4527 
4528 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4529 {
4530     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4531 
4532     if (spapr_cpu->vpa_addr) {
4533         CPUState *cs = CPU(cpu);
4534         uint32_t dispatch;
4535 
4536         dispatch = ldl_be_phys(cs->as,
4537                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4538         dispatch++;
4539         if ((dispatch & 1) != 1) {
4540             qemu_log_mask(LOG_GUEST_ERROR,
4541                           "VPA: incorrect dispatch counter value for "
4542                           "preempted partition %u, correcting.\n", dispatch);
4543             dispatch++;
4544         }
4545         stl_be_phys(cs->as,
4546                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4547     }
4548 }
4549 
4550 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4551 {
4552     MachineClass *mc = MACHINE_CLASS(oc);
4553     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4554     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4555     NMIClass *nc = NMI_CLASS(oc);
4556     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4557     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4558     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4559     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4560     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4561     VofMachineIfClass *vmc = VOF_MACHINE_CLASS(oc);
4562 
4563     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4564     mc->ignore_boot_device_suffixes = true;
4565 
4566     /*
4567      * We set up the default / latest behaviour here.  The class_init
4568      * functions for the specific versioned machine types can override
4569      * these details for backwards compatibility
4570      */
4571     mc->init = spapr_machine_init;
4572     mc->reset = spapr_machine_reset;
4573     mc->block_default_type = IF_SCSI;
4574 
4575     /*
4576      * Setting max_cpus to INT32_MAX. Both KVM and TCG max_cpus values
4577      * should be limited by the host capability instead of hardcoded.
4578      * max_cpus for KVM guests will be checked in kvm_init(), and TCG
4579      * guests are welcome to have as many CPUs as the host are capable
4580      * of emulate.
4581      */
4582     mc->max_cpus = INT32_MAX;
4583 
4584     mc->no_parallel = 1;
4585     mc->default_boot_order = "";
4586     mc->default_ram_size = 512 * MiB;
4587     mc->default_ram_id = "ppc_spapr.ram";
4588     mc->default_display = "std";
4589     mc->kvm_type = spapr_kvm_type;
4590     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4591     mc->pci_allow_0_address = true;
4592     assert(!mc->get_hotplug_handler);
4593     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4594     hc->pre_plug = spapr_machine_device_pre_plug;
4595     hc->plug = spapr_machine_device_plug;
4596     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4597     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4598     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4599     hc->unplug_request = spapr_machine_device_unplug_request;
4600     hc->unplug = spapr_machine_device_unplug;
4601 
4602     smc->dr_lmb_enabled = true;
4603     smc->update_dt_enabled = true;
4604     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4605     mc->has_hotpluggable_cpus = true;
4606     mc->nvdimm_supported = true;
4607     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4608     fwc->get_dev_path = spapr_get_fw_dev_path;
4609     nc->nmi_monitor_handler = spapr_nmi;
4610     smc->phb_placement = spapr_phb_placement;
4611     vhc->cpu_in_nested = spapr_cpu_in_nested;
4612     vhc->deliver_hv_excp = spapr_exit_nested;
4613     vhc->hypercall = emulate_spapr_hypercall;
4614     vhc->hpt_mask = spapr_hpt_mask;
4615     vhc->map_hptes = spapr_map_hptes;
4616     vhc->unmap_hptes = spapr_unmap_hptes;
4617     vhc->hpte_set_c = spapr_hpte_set_c;
4618     vhc->hpte_set_r = spapr_hpte_set_r;
4619     vhc->get_pate = spapr_get_pate;
4620     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4621     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4622     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4623     xic->ics_get = spapr_ics_get;
4624     xic->ics_resend = spapr_ics_resend;
4625     xic->icp_get = spapr_icp_get;
4626     ispc->print_info = spapr_pic_print_info;
4627     /* Force NUMA node memory size to be a multiple of
4628      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4629      * in which LMBs are represented and hot-added
4630      */
4631     mc->numa_mem_align_shift = 28;
4632     mc->auto_enable_numa = true;
4633 
4634     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4635     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4636     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4637     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4638     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4639     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4640     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4641     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4642     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4643     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4644     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_ON;
4645     smc->default_caps.caps[SPAPR_CAP_RPT_INVALIDATE] = SPAPR_CAP_OFF;
4646     spapr_caps_add_properties(smc);
4647     smc->irq = &spapr_irq_dual;
4648     smc->dr_phb_enabled = true;
4649     smc->linux_pci_probe = true;
4650     smc->smp_threads_vsmt = true;
4651     smc->nr_xirqs = SPAPR_NR_XIRQS;
4652     xfc->match_nvt = spapr_match_nvt;
4653     vmc->client_architecture_support = spapr_vof_client_architecture_support;
4654     vmc->quiesce = spapr_vof_quiesce;
4655     vmc->setprop = spapr_vof_setprop;
4656 }
4657 
4658 static const TypeInfo spapr_machine_info = {
4659     .name          = TYPE_SPAPR_MACHINE,
4660     .parent        = TYPE_MACHINE,
4661     .abstract      = true,
4662     .instance_size = sizeof(SpaprMachineState),
4663     .instance_init = spapr_instance_init,
4664     .instance_finalize = spapr_machine_finalizefn,
4665     .class_size    = sizeof(SpaprMachineClass),
4666     .class_init    = spapr_machine_class_init,
4667     .interfaces = (InterfaceInfo[]) {
4668         { TYPE_FW_PATH_PROVIDER },
4669         { TYPE_NMI },
4670         { TYPE_HOTPLUG_HANDLER },
4671         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4672         { TYPE_XICS_FABRIC },
4673         { TYPE_INTERRUPT_STATS_PROVIDER },
4674         { TYPE_XIVE_FABRIC },
4675         { TYPE_VOF_MACHINE_IF },
4676         { }
4677     },
4678 };
4679 
4680 static void spapr_machine_latest_class_options(MachineClass *mc)
4681 {
4682     mc->alias = "pseries";
4683     mc->is_default = true;
4684 }
4685 
4686 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4687     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4688                                                     void *data)      \
4689     {                                                                \
4690         MachineClass *mc = MACHINE_CLASS(oc);                        \
4691         spapr_machine_##suffix##_class_options(mc);                  \
4692         if (latest) {                                                \
4693             spapr_machine_latest_class_options(mc);                  \
4694         }                                                            \
4695     }                                                                \
4696     static const TypeInfo spapr_machine_##suffix##_info = {          \
4697         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4698         .parent = TYPE_SPAPR_MACHINE,                                \
4699         .class_init = spapr_machine_##suffix##_class_init,           \
4700     };                                                               \
4701     static void spapr_machine_register_##suffix(void)                \
4702     {                                                                \
4703         type_register(&spapr_machine_##suffix##_info);               \
4704     }                                                                \
4705     type_init(spapr_machine_register_##suffix)
4706 
4707 /*
4708  * pseries-7.0
4709  */
4710 static void spapr_machine_7_0_class_options(MachineClass *mc)
4711 {
4712     /* Defaults for the latest behaviour inherited from the base class */
4713 }
4714 
4715 DEFINE_SPAPR_MACHINE(7_0, "7.0", true);
4716 
4717 /*
4718  * pseries-6.2
4719  */
4720 static void spapr_machine_6_2_class_options(MachineClass *mc)
4721 {
4722     spapr_machine_7_0_class_options(mc);
4723     compat_props_add(mc->compat_props, hw_compat_6_2, hw_compat_6_2_len);
4724 }
4725 
4726 DEFINE_SPAPR_MACHINE(6_2, "6.2", false);
4727 
4728 /*
4729  * pseries-6.1
4730  */
4731 static void spapr_machine_6_1_class_options(MachineClass *mc)
4732 {
4733     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4734 
4735     spapr_machine_6_2_class_options(mc);
4736     compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
4737     smc->pre_6_2_numa_affinity = true;
4738     mc->smp_props.prefer_sockets = true;
4739 }
4740 
4741 DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
4742 
4743 /*
4744  * pseries-6.0
4745  */
4746 static void spapr_machine_6_0_class_options(MachineClass *mc)
4747 {
4748     spapr_machine_6_1_class_options(mc);
4749     compat_props_add(mc->compat_props, hw_compat_6_0, hw_compat_6_0_len);
4750 }
4751 
4752 DEFINE_SPAPR_MACHINE(6_0, "6.0", false);
4753 
4754 /*
4755  * pseries-5.2
4756  */
4757 static void spapr_machine_5_2_class_options(MachineClass *mc)
4758 {
4759     spapr_machine_6_0_class_options(mc);
4760     compat_props_add(mc->compat_props, hw_compat_5_2, hw_compat_5_2_len);
4761 }
4762 
4763 DEFINE_SPAPR_MACHINE(5_2, "5.2", false);
4764 
4765 /*
4766  * pseries-5.1
4767  */
4768 static void spapr_machine_5_1_class_options(MachineClass *mc)
4769 {
4770     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4771 
4772     spapr_machine_5_2_class_options(mc);
4773     compat_props_add(mc->compat_props, hw_compat_5_1, hw_compat_5_1_len);
4774     smc->pre_5_2_numa_associativity = true;
4775 }
4776 
4777 DEFINE_SPAPR_MACHINE(5_1, "5.1", false);
4778 
4779 /*
4780  * pseries-5.0
4781  */
4782 static void spapr_machine_5_0_class_options(MachineClass *mc)
4783 {
4784     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4785     static GlobalProperty compat[] = {
4786         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-5.1-associativity", "on" },
4787     };
4788 
4789     spapr_machine_5_1_class_options(mc);
4790     compat_props_add(mc->compat_props, hw_compat_5_0, hw_compat_5_0_len);
4791     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4792     mc->numa_mem_supported = true;
4793     smc->pre_5_1_assoc_refpoints = true;
4794 }
4795 
4796 DEFINE_SPAPR_MACHINE(5_0, "5.0", false);
4797 
4798 /*
4799  * pseries-4.2
4800  */
4801 static void spapr_machine_4_2_class_options(MachineClass *mc)
4802 {
4803     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4804 
4805     spapr_machine_5_0_class_options(mc);
4806     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4807     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4808     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
4809     smc->rma_limit = 16 * GiB;
4810     mc->nvdimm_supported = false;
4811 }
4812 
4813 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4814 
4815 /*
4816  * pseries-4.1
4817  */
4818 static void spapr_machine_4_1_class_options(MachineClass *mc)
4819 {
4820     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4821     static GlobalProperty compat[] = {
4822         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4823         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4824     };
4825 
4826     spapr_machine_4_2_class_options(mc);
4827     smc->linux_pci_probe = false;
4828     smc->smp_threads_vsmt = false;
4829     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4830     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4831 }
4832 
4833 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4834 
4835 /*
4836  * pseries-4.0
4837  */
4838 static bool phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4839                               uint64_t *buid, hwaddr *pio,
4840                               hwaddr *mmio32, hwaddr *mmio64,
4841                               unsigned n_dma, uint32_t *liobns,
4842                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4843 {
4844     if (!spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma,
4845                              liobns, nv2gpa, nv2atsd, errp)) {
4846         return false;
4847     }
4848 
4849     *nv2gpa = 0;
4850     *nv2atsd = 0;
4851     return true;
4852 }
4853 static void spapr_machine_4_0_class_options(MachineClass *mc)
4854 {
4855     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4856 
4857     spapr_machine_4_1_class_options(mc);
4858     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4859     smc->phb_placement = phb_placement_4_0;
4860     smc->irq = &spapr_irq_xics;
4861     smc->pre_4_1_migration = true;
4862 }
4863 
4864 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4865 
4866 /*
4867  * pseries-3.1
4868  */
4869 static void spapr_machine_3_1_class_options(MachineClass *mc)
4870 {
4871     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4872 
4873     spapr_machine_4_0_class_options(mc);
4874     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4875 
4876     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4877     smc->update_dt_enabled = false;
4878     smc->dr_phb_enabled = false;
4879     smc->broken_host_serial_model = true;
4880     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4881     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4882     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4883     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4884 }
4885 
4886 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4887 
4888 /*
4889  * pseries-3.0
4890  */
4891 
4892 static void spapr_machine_3_0_class_options(MachineClass *mc)
4893 {
4894     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4895 
4896     spapr_machine_3_1_class_options(mc);
4897     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4898 
4899     smc->legacy_irq_allocation = true;
4900     smc->nr_xirqs = 0x400;
4901     smc->irq = &spapr_irq_xics_legacy;
4902 }
4903 
4904 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4905 
4906 /*
4907  * pseries-2.12
4908  */
4909 static void spapr_machine_2_12_class_options(MachineClass *mc)
4910 {
4911     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4912     static GlobalProperty compat[] = {
4913         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4914         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4915     };
4916 
4917     spapr_machine_3_0_class_options(mc);
4918     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4919     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4920 
4921     /* We depend on kvm_enabled() to choose a default value for the
4922      * hpt-max-page-size capability. Of course we can't do it here
4923      * because this is too early and the HW accelerator isn't initialzed
4924      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4925      */
4926     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4927 }
4928 
4929 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4930 
4931 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4932 {
4933     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4934 
4935     spapr_machine_2_12_class_options(mc);
4936     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4937     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4938     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4939 }
4940 
4941 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4942 
4943 /*
4944  * pseries-2.11
4945  */
4946 
4947 static void spapr_machine_2_11_class_options(MachineClass *mc)
4948 {
4949     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4950 
4951     spapr_machine_2_12_class_options(mc);
4952     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4953     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4954 }
4955 
4956 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4957 
4958 /*
4959  * pseries-2.10
4960  */
4961 
4962 static void spapr_machine_2_10_class_options(MachineClass *mc)
4963 {
4964     spapr_machine_2_11_class_options(mc);
4965     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4966 }
4967 
4968 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4969 
4970 /*
4971  * pseries-2.9
4972  */
4973 
4974 static void spapr_machine_2_9_class_options(MachineClass *mc)
4975 {
4976     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4977     static GlobalProperty compat[] = {
4978         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4979     };
4980 
4981     spapr_machine_2_10_class_options(mc);
4982     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4983     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4984     smc->pre_2_10_has_unused_icps = true;
4985     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4986 }
4987 
4988 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4989 
4990 /*
4991  * pseries-2.8
4992  */
4993 
4994 static void spapr_machine_2_8_class_options(MachineClass *mc)
4995 {
4996     static GlobalProperty compat[] = {
4997         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4998     };
4999 
5000     spapr_machine_2_9_class_options(mc);
5001     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
5002     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5003     mc->numa_mem_align_shift = 23;
5004 }
5005 
5006 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
5007 
5008 /*
5009  * pseries-2.7
5010  */
5011 
5012 static bool phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
5013                               uint64_t *buid, hwaddr *pio,
5014                               hwaddr *mmio32, hwaddr *mmio64,
5015                               unsigned n_dma, uint32_t *liobns,
5016                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
5017 {
5018     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
5019     const uint64_t base_buid = 0x800000020000000ULL;
5020     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
5021     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
5022     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
5023     const uint32_t max_index = 255;
5024     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
5025 
5026     uint64_t ram_top = MACHINE(spapr)->ram_size;
5027     hwaddr phb0_base, phb_base;
5028     int i;
5029 
5030     /* Do we have device memory? */
5031     if (MACHINE(spapr)->maxram_size > ram_top) {
5032         /* Can't just use maxram_size, because there may be an
5033          * alignment gap between normal and device memory regions
5034          */
5035         ram_top = MACHINE(spapr)->device_memory->base +
5036             memory_region_size(&MACHINE(spapr)->device_memory->mr);
5037     }
5038 
5039     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
5040 
5041     if (index > max_index) {
5042         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
5043                    max_index);
5044         return false;
5045     }
5046 
5047     *buid = base_buid + index;
5048     for (i = 0; i < n_dma; ++i) {
5049         liobns[i] = SPAPR_PCI_LIOBN(index, i);
5050     }
5051 
5052     phb_base = phb0_base + index * phb_spacing;
5053     *pio = phb_base + pio_offset;
5054     *mmio32 = phb_base + mmio_offset;
5055     /*
5056      * We don't set the 64-bit MMIO window, relying on the PHB's
5057      * fallback behaviour of automatically splitting a large "32-bit"
5058      * window into contiguous 32-bit and 64-bit windows
5059      */
5060 
5061     *nv2gpa = 0;
5062     *nv2atsd = 0;
5063     return true;
5064 }
5065 
5066 static void spapr_machine_2_7_class_options(MachineClass *mc)
5067 {
5068     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5069     static GlobalProperty compat[] = {
5070         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
5071         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
5072         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
5073         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
5074     };
5075 
5076     spapr_machine_2_8_class_options(mc);
5077     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
5078     mc->default_machine_opts = "modern-hotplug-events=off";
5079     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
5080     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5081     smc->phb_placement = phb_placement_2_7;
5082 }
5083 
5084 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
5085 
5086 /*
5087  * pseries-2.6
5088  */
5089 
5090 static void spapr_machine_2_6_class_options(MachineClass *mc)
5091 {
5092     static GlobalProperty compat[] = {
5093         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
5094     };
5095 
5096     spapr_machine_2_7_class_options(mc);
5097     mc->has_hotpluggable_cpus = false;
5098     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
5099     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5100 }
5101 
5102 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
5103 
5104 /*
5105  * pseries-2.5
5106  */
5107 
5108 static void spapr_machine_2_5_class_options(MachineClass *mc)
5109 {
5110     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5111     static GlobalProperty compat[] = {
5112         { "spapr-vlan", "use-rx-buffer-pools", "off" },
5113     };
5114 
5115     spapr_machine_2_6_class_options(mc);
5116     smc->use_ohci_by_default = true;
5117     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
5118     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5119 }
5120 
5121 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
5122 
5123 /*
5124  * pseries-2.4
5125  */
5126 
5127 static void spapr_machine_2_4_class_options(MachineClass *mc)
5128 {
5129     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
5130 
5131     spapr_machine_2_5_class_options(mc);
5132     smc->dr_lmb_enabled = false;
5133     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
5134 }
5135 
5136 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
5137 
5138 /*
5139  * pseries-2.3
5140  */
5141 
5142 static void spapr_machine_2_3_class_options(MachineClass *mc)
5143 {
5144     static GlobalProperty compat[] = {
5145         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
5146     };
5147     spapr_machine_2_4_class_options(mc);
5148     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
5149     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5150 }
5151 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
5152 
5153 /*
5154  * pseries-2.2
5155  */
5156 
5157 static void spapr_machine_2_2_class_options(MachineClass *mc)
5158 {
5159     static GlobalProperty compat[] = {
5160         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
5161     };
5162 
5163     spapr_machine_2_3_class_options(mc);
5164     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
5165     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
5166     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
5167 }
5168 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
5169 
5170 /*
5171  * pseries-2.1
5172  */
5173 
5174 static void spapr_machine_2_1_class_options(MachineClass *mc)
5175 {
5176     spapr_machine_2_2_class_options(mc);
5177     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
5178 }
5179 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
5180 
5181 static void spapr_machine_register_types(void)
5182 {
5183     type_register_static(&spapr_machine_info);
5184 }
5185 
5186 type_init(spapr_machine_register_types)
5187