xref: /openbmc/qemu/hw/ppc/spapr.c (revision 60e58bd9f08a3b91a35850f7501a0a1bcf912b6f)
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 "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/block-backend.h"
39 #include "sysemu/cpus.h"
40 #include "sysemu/hw_accel.h"
41 #include "kvm_ppc.h"
42 #include "migration/misc.h"
43 #include "migration/global_state.h"
44 #include "migration/register.h"
45 #include "mmu-hash64.h"
46 #include "mmu-book3s-v3.h"
47 #include "cpu-models.h"
48 #include "qom/cpu.h"
49 
50 #include "hw/boards.h"
51 #include "hw/ppc/ppc.h"
52 #include "hw/loader.h"
53 
54 #include "hw/ppc/fdt.h"
55 #include "hw/ppc/spapr.h"
56 #include "hw/ppc/spapr_vio.h"
57 #include "hw/pci-host/spapr.h"
58 #include "hw/ppc/xics.h"
59 #include "hw/pci/msi.h"
60 
61 #include "hw/pci/pci.h"
62 #include "hw/scsi/scsi.h"
63 #include "hw/virtio/virtio-scsi.h"
64 #include "hw/virtio/vhost-scsi-common.h"
65 
66 #include "exec/address-spaces.h"
67 #include "hw/usb.h"
68 #include "qemu/config-file.h"
69 #include "qemu/error-report.h"
70 #include "trace.h"
71 #include "hw/nmi.h"
72 #include "hw/intc/intc.h"
73 
74 #include "hw/compat.h"
75 #include "qemu/cutils.h"
76 #include "hw/ppc/spapr_cpu_core.h"
77 #include "qmp-commands.h"
78 
79 #include <libfdt.h>
80 
81 /* SLOF memory layout:
82  *
83  * SLOF raw image loaded at 0, copies its romfs right below the flat
84  * device-tree, then position SLOF itself 31M below that
85  *
86  * So we set FW_OVERHEAD to 40MB which should account for all of that
87  * and more
88  *
89  * We load our kernel at 4M, leaving space for SLOF initial image
90  */
91 #define FDT_MAX_SIZE            0x100000
92 #define RTAS_MAX_SIZE           0x10000
93 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
94 #define FW_MAX_SIZE             0x400000
95 #define FW_FILE_NAME            "slof.bin"
96 #define FW_OVERHEAD             0x2800000
97 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
98 
99 #define MIN_RMA_SLOF            128UL
100 
101 #define PHANDLE_XICP            0x00001111
102 
103 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
104                                   const char *type_ics,
105                                   int nr_irqs, Error **errp)
106 {
107     Error *local_err = NULL;
108     Object *obj;
109 
110     obj = object_new(type_ics);
111     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
112     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
113                                    &error_abort);
114     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
115     if (local_err) {
116         goto error;
117     }
118     object_property_set_bool(obj, true, "realized", &local_err);
119     if (local_err) {
120         goto error;
121     }
122 
123     return ICS_SIMPLE(obj);
124 
125 error:
126     error_propagate(errp, local_err);
127     return NULL;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 static inline int xics_max_server_number(void)
165 {
166     return DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(), smp_threads);
167 }
168 
169 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
170 {
171     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
172     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
173 
174     if (kvm_enabled()) {
175         if (machine_kernel_irqchip_allowed(machine) &&
176             !xics_kvm_init(spapr, errp)) {
177             spapr->icp_type = TYPE_KVM_ICP;
178             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
179         }
180         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
181             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
182             return;
183         }
184     }
185 
186     if (!spapr->ics) {
187         xics_spapr_init(spapr);
188         spapr->icp_type = TYPE_ICP;
189         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
190         if (!spapr->ics) {
191             return;
192         }
193     }
194 
195     if (smc->pre_2_10_has_unused_icps) {
196         int i;
197 
198         for (i = 0; i < xics_max_server_number(); i++) {
199             /* Dummy entries get deregistered when real ICPState objects
200              * are registered during CPU core hotplug.
201              */
202             pre_2_10_vmstate_register_dummy_icp(i);
203         }
204     }
205 }
206 
207 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
208                                   int smt_threads)
209 {
210     int i, ret = 0;
211     uint32_t servers_prop[smt_threads];
212     uint32_t gservers_prop[smt_threads * 2];
213     int index = spapr_vcpu_id(cpu);
214 
215     if (cpu->compat_pvr) {
216         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
217         if (ret < 0) {
218             return ret;
219         }
220     }
221 
222     /* Build interrupt servers and gservers properties */
223     for (i = 0; i < smt_threads; i++) {
224         servers_prop[i] = cpu_to_be32(index + i);
225         /* Hack, direct the group queues back to cpu 0 */
226         gservers_prop[i*2] = cpu_to_be32(index + i);
227         gservers_prop[i*2 + 1] = 0;
228     }
229     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
230                       servers_prop, sizeof(servers_prop));
231     if (ret < 0) {
232         return ret;
233     }
234     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
235                       gservers_prop, sizeof(gservers_prop));
236 
237     return ret;
238 }
239 
240 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
241 {
242     int index = spapr_vcpu_id(cpu);
243     uint32_t associativity[] = {cpu_to_be32(0x5),
244                                 cpu_to_be32(0x0),
245                                 cpu_to_be32(0x0),
246                                 cpu_to_be32(0x0),
247                                 cpu_to_be32(cpu->node_id),
248                                 cpu_to_be32(index)};
249 
250     /* Advertise NUMA via ibm,associativity */
251     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
252                           sizeof(associativity));
253 }
254 
255 /* Populate the "ibm,pa-features" property */
256 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
257                                        PowerPCCPU *cpu,
258                                        void *fdt, int offset,
259                                        bool legacy_guest)
260 {
261     CPUPPCState *env = &cpu->env;
262     uint8_t pa_features_206[] = { 6, 0,
263         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
264     uint8_t pa_features_207[] = { 24, 0,
265         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
266         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
267         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
268         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
269     uint8_t pa_features_300[] = { 66, 0,
270         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
271         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
272         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
273         /* 6: DS207 */
274         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
275         /* 16: Vector */
276         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
277         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
278         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
279         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
280         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
281         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
282         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
283         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
284         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
285         /* 42: PM, 44: PC RA, 46: SC vec'd */
286         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
287         /* 48: SIMD, 50: QP BFP, 52: String */
288         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
289         /* 54: DecFP, 56: DecI, 58: SHA */
290         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
291         /* 60: NM atomic, 62: RNG */
292         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
293     };
294     uint8_t *pa_features = NULL;
295     size_t pa_size;
296 
297     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
298         pa_features = pa_features_206;
299         pa_size = sizeof(pa_features_206);
300     }
301     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
302         pa_features = pa_features_207;
303         pa_size = sizeof(pa_features_207);
304     }
305     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
306         pa_features = pa_features_300;
307         pa_size = sizeof(pa_features_300);
308     }
309     if (!pa_features) {
310         return;
311     }
312 
313     if (env->ci_large_pages) {
314         /*
315          * Note: we keep CI large pages off by default because a 64K capable
316          * guest provisioned with large pages might otherwise try to map a qemu
317          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
318          * even if that qemu runs on a 4k host.
319          * We dd this bit back here if we are confident this is not an issue
320          */
321         pa_features[3] |= 0x20;
322     }
323     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
324         pa_features[24] |= 0x80;    /* Transactional memory support */
325     }
326     if (legacy_guest && pa_size > 40) {
327         /* Workaround for broken kernels that attempt (guest) radix
328          * mode when they can't handle it, if they see the radix bit set
329          * in pa-features. So hide it from them. */
330         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
331     }
332 
333     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
334 }
335 
336 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
337 {
338     int ret = 0, offset, cpus_offset;
339     CPUState *cs;
340     char cpu_model[32];
341     int smt = kvmppc_smt_threads();
342     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
343 
344     CPU_FOREACH(cs) {
345         PowerPCCPU *cpu = POWERPC_CPU(cs);
346         DeviceClass *dc = DEVICE_GET_CLASS(cs);
347         int index = spapr_vcpu_id(cpu);
348         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
349 
350         if ((index % smt) != 0) {
351             continue;
352         }
353 
354         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
355 
356         cpus_offset = fdt_path_offset(fdt, "/cpus");
357         if (cpus_offset < 0) {
358             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
359             if (cpus_offset < 0) {
360                 return cpus_offset;
361             }
362         }
363         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
364         if (offset < 0) {
365             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
366             if (offset < 0) {
367                 return offset;
368             }
369         }
370 
371         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
372                           pft_size_prop, sizeof(pft_size_prop));
373         if (ret < 0) {
374             return ret;
375         }
376 
377         if (nb_numa_nodes > 1) {
378             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
379             if (ret < 0) {
380                 return ret;
381             }
382         }
383 
384         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
385         if (ret < 0) {
386             return ret;
387         }
388 
389         spapr_populate_pa_features(spapr, cpu, fdt, offset,
390                                    spapr->cas_legacy_guest_workaround);
391     }
392     return ret;
393 }
394 
395 static hwaddr spapr_node0_size(MachineState *machine)
396 {
397     if (nb_numa_nodes) {
398         int i;
399         for (i = 0; i < nb_numa_nodes; ++i) {
400             if (numa_info[i].node_mem) {
401                 return MIN(pow2floor(numa_info[i].node_mem),
402                            machine->ram_size);
403             }
404         }
405     }
406     return machine->ram_size;
407 }
408 
409 static void add_str(GString *s, const gchar *s1)
410 {
411     g_string_append_len(s, s1, strlen(s1) + 1);
412 }
413 
414 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
415                                        hwaddr size)
416 {
417     uint32_t associativity[] = {
418         cpu_to_be32(0x4), /* length */
419         cpu_to_be32(0x0), cpu_to_be32(0x0),
420         cpu_to_be32(0x0), cpu_to_be32(nodeid)
421     };
422     char mem_name[32];
423     uint64_t mem_reg_property[2];
424     int off;
425 
426     mem_reg_property[0] = cpu_to_be64(start);
427     mem_reg_property[1] = cpu_to_be64(size);
428 
429     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
430     off = fdt_add_subnode(fdt, 0, mem_name);
431     _FDT(off);
432     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
433     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
434                       sizeof(mem_reg_property))));
435     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
436                       sizeof(associativity))));
437     return off;
438 }
439 
440 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
441 {
442     MachineState *machine = MACHINE(spapr);
443     hwaddr mem_start, node_size;
444     int i, nb_nodes = nb_numa_nodes;
445     NodeInfo *nodes = numa_info;
446     NodeInfo ramnode;
447 
448     /* No NUMA nodes, assume there is just one node with whole RAM */
449     if (!nb_numa_nodes) {
450         nb_nodes = 1;
451         ramnode.node_mem = machine->ram_size;
452         nodes = &ramnode;
453     }
454 
455     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
456         if (!nodes[i].node_mem) {
457             continue;
458         }
459         if (mem_start >= machine->ram_size) {
460             node_size = 0;
461         } else {
462             node_size = nodes[i].node_mem;
463             if (node_size > machine->ram_size - mem_start) {
464                 node_size = machine->ram_size - mem_start;
465             }
466         }
467         if (!mem_start) {
468             /* ppc_spapr_init() checks for rma_size <= node0_size already */
469             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
470             mem_start += spapr->rma_size;
471             node_size -= spapr->rma_size;
472         }
473         for ( ; node_size; ) {
474             hwaddr sizetmp = pow2floor(node_size);
475 
476             /* mem_start != 0 here */
477             if (ctzl(mem_start) < ctzl(sizetmp)) {
478                 sizetmp = 1ULL << ctzl(mem_start);
479             }
480 
481             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
482             node_size -= sizetmp;
483             mem_start += sizetmp;
484         }
485     }
486 
487     return 0;
488 }
489 
490 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
491                                   sPAPRMachineState *spapr)
492 {
493     PowerPCCPU *cpu = POWERPC_CPU(cs);
494     CPUPPCState *env = &cpu->env;
495     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
496     int index = spapr_vcpu_id(cpu);
497     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
498                        0xffffffff, 0xffffffff};
499     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
500         : SPAPR_TIMEBASE_FREQ;
501     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
502     uint32_t page_sizes_prop[64];
503     size_t page_sizes_prop_size;
504     uint32_t vcpus_per_socket = smp_threads * smp_cores;
505     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
506     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
507     sPAPRDRConnector *drc;
508     int drc_index;
509     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
510     int i;
511 
512     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
513     if (drc) {
514         drc_index = spapr_drc_index(drc);
515         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
516     }
517 
518     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
519     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
520 
521     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
522     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
523                            env->dcache_line_size)));
524     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
525                            env->dcache_line_size)));
526     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
527                            env->icache_line_size)));
528     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
529                            env->icache_line_size)));
530 
531     if (pcc->l1_dcache_size) {
532         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
533                                pcc->l1_dcache_size)));
534     } else {
535         warn_report("Unknown L1 dcache size for cpu");
536     }
537     if (pcc->l1_icache_size) {
538         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
539                                pcc->l1_icache_size)));
540     } else {
541         warn_report("Unknown L1 icache size for cpu");
542     }
543 
544     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
545     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
546     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
547     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
548     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
549     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
550 
551     if (env->spr_cb[SPR_PURR].oea_read) {
552         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
553     }
554 
555     if (env->mmu_model & POWERPC_MMU_1TSEG) {
556         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
557                           segs, sizeof(segs))));
558     }
559 
560     /* Advertise VSX (vector extensions) if available
561      *   1               == VMX / Altivec available
562      *   2               == VSX available
563      *
564      * Only CPUs for which we create core types in spapr_cpu_core.c
565      * are possible, and all of those have VMX */
566     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
567         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
568     } else {
569         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
570     }
571 
572     /* Advertise DFP (Decimal Floating Point) if available
573      *   0 / no property == no DFP
574      *   1               == DFP available */
575     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
576         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
577     }
578 
579     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
580                                                   sizeof(page_sizes_prop));
581     if (page_sizes_prop_size) {
582         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
583                           page_sizes_prop, page_sizes_prop_size)));
584     }
585 
586     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
587 
588     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
589                            cs->cpu_index / vcpus_per_socket)));
590 
591     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
592                       pft_size_prop, sizeof(pft_size_prop))));
593 
594     if (nb_numa_nodes > 1) {
595         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
596     }
597 
598     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
599 
600     if (pcc->radix_page_info) {
601         for (i = 0; i < pcc->radix_page_info->count; i++) {
602             radix_AP_encodings[i] =
603                 cpu_to_be32(pcc->radix_page_info->entries[i]);
604         }
605         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
606                           radix_AP_encodings,
607                           pcc->radix_page_info->count *
608                           sizeof(radix_AP_encodings[0]))));
609     }
610 }
611 
612 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
613 {
614     CPUState *cs;
615     int cpus_offset;
616     char *nodename;
617     int smt = kvmppc_smt_threads();
618 
619     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
620     _FDT(cpus_offset);
621     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
622     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
623 
624     /*
625      * We walk the CPUs in reverse order to ensure that CPU DT nodes
626      * created by fdt_add_subnode() end up in the right order in FDT
627      * for the guest kernel the enumerate the CPUs correctly.
628      */
629     CPU_FOREACH_REVERSE(cs) {
630         PowerPCCPU *cpu = POWERPC_CPU(cs);
631         int index = spapr_vcpu_id(cpu);
632         DeviceClass *dc = DEVICE_GET_CLASS(cs);
633         int offset;
634 
635         if ((index % smt) != 0) {
636             continue;
637         }
638 
639         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
640         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
641         g_free(nodename);
642         _FDT(offset);
643         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
644     }
645 
646 }
647 
648 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
649 {
650     MemoryDeviceInfoList *info;
651 
652     for (info = list; info; info = info->next) {
653         MemoryDeviceInfo *value = info->value;
654 
655         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
656             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
657 
658             if (pcdimm_info->addr >= addr &&
659                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
660                 return pcdimm_info->node;
661             }
662         }
663     }
664 
665     return -1;
666 }
667 
668 /*
669  * Adds ibm,dynamic-reconfiguration-memory node.
670  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
671  * of this device tree node.
672  */
673 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
674 {
675     MachineState *machine = MACHINE(spapr);
676     int ret, i, offset;
677     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
678     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
679     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
680     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
681                        memory_region_size(&spapr->hotplug_memory.mr)) /
682                        lmb_size;
683     uint32_t *int_buf, *cur_index, buf_len;
684     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
685     MemoryDeviceInfoList *dimms = NULL;
686 
687     /*
688      * Don't create the node if there is no hotpluggable memory
689      */
690     if (machine->ram_size == machine->maxram_size) {
691         return 0;
692     }
693 
694     /*
695      * Allocate enough buffer size to fit in ibm,dynamic-memory
696      * or ibm,associativity-lookup-arrays
697      */
698     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
699               * sizeof(uint32_t);
700     cur_index = int_buf = g_malloc0(buf_len);
701 
702     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
703 
704     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
705                     sizeof(prop_lmb_size));
706     if (ret < 0) {
707         goto out;
708     }
709 
710     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
711     if (ret < 0) {
712         goto out;
713     }
714 
715     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
716     if (ret < 0) {
717         goto out;
718     }
719 
720     if (hotplug_lmb_start) {
721         MemoryDeviceInfoList **prev = &dimms;
722         qmp_pc_dimm_device_list(qdev_get_machine(), &prev);
723     }
724 
725     /* ibm,dynamic-memory */
726     int_buf[0] = cpu_to_be32(nr_lmbs);
727     cur_index++;
728     for (i = 0; i < nr_lmbs; i++) {
729         uint64_t addr = i * lmb_size;
730         uint32_t *dynamic_memory = cur_index;
731 
732         if (i >= hotplug_lmb_start) {
733             sPAPRDRConnector *drc;
734 
735             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
736             g_assert(drc);
737 
738             dynamic_memory[0] = cpu_to_be32(addr >> 32);
739             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
740             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
741             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
742             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
743             if (memory_region_present(get_system_memory(), addr)) {
744                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
745             } else {
746                 dynamic_memory[5] = cpu_to_be32(0);
747             }
748         } else {
749             /*
750              * LMB information for RMA, boot time RAM and gap b/n RAM and
751              * hotplug memory region -- all these are marked as reserved
752              * and as having no valid DRC.
753              */
754             dynamic_memory[0] = cpu_to_be32(addr >> 32);
755             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
756             dynamic_memory[2] = cpu_to_be32(0);
757             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
758             dynamic_memory[4] = cpu_to_be32(-1);
759             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
760                                             SPAPR_LMB_FLAGS_DRC_INVALID);
761         }
762 
763         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
764     }
765     qapi_free_MemoryDeviceInfoList(dimms);
766     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
767     if (ret < 0) {
768         goto out;
769     }
770 
771     /* ibm,associativity-lookup-arrays */
772     cur_index = int_buf;
773     int_buf[0] = cpu_to_be32(nr_nodes);
774     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
775     cur_index += 2;
776     for (i = 0; i < nr_nodes; i++) {
777         uint32_t associativity[] = {
778             cpu_to_be32(0x0),
779             cpu_to_be32(0x0),
780             cpu_to_be32(0x0),
781             cpu_to_be32(i)
782         };
783         memcpy(cur_index, associativity, sizeof(associativity));
784         cur_index += 4;
785     }
786     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
787             (cur_index - int_buf) * sizeof(uint32_t));
788 out:
789     g_free(int_buf);
790     return ret;
791 }
792 
793 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
794                                 sPAPROptionVector *ov5_updates)
795 {
796     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
797     int ret = 0, offset;
798 
799     /* Generate ibm,dynamic-reconfiguration-memory node if required */
800     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
801         g_assert(smc->dr_lmb_enabled);
802         ret = spapr_populate_drconf_memory(spapr, fdt);
803         if (ret) {
804             goto out;
805         }
806     }
807 
808     offset = fdt_path_offset(fdt, "/chosen");
809     if (offset < 0) {
810         offset = fdt_add_subnode(fdt, 0, "chosen");
811         if (offset < 0) {
812             return offset;
813         }
814     }
815     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
816                                  "ibm,architecture-vec-5");
817 
818 out:
819     return ret;
820 }
821 
822 static bool spapr_hotplugged_dev_before_cas(void)
823 {
824     Object *drc_container, *obj;
825     ObjectProperty *prop;
826     ObjectPropertyIterator iter;
827 
828     drc_container = container_get(object_get_root(), "/dr-connector");
829     object_property_iter_init(&iter, drc_container);
830     while ((prop = object_property_iter_next(&iter))) {
831         if (!strstart(prop->type, "link<", NULL)) {
832             continue;
833         }
834         obj = object_property_get_link(drc_container, prop->name, NULL);
835         if (spapr_drc_needed(obj)) {
836             return true;
837         }
838     }
839     return false;
840 }
841 
842 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
843                                  target_ulong addr, target_ulong size,
844                                  sPAPROptionVector *ov5_updates)
845 {
846     void *fdt, *fdt_skel;
847     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
848 
849     if (spapr_hotplugged_dev_before_cas()) {
850         return 1;
851     }
852 
853     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
854         error_report("SLOF provided an unexpected CAS buffer size "
855                      TARGET_FMT_lu " (min: %zu, max: %u)",
856                      size, sizeof(hdr), FW_MAX_SIZE);
857         exit(EXIT_FAILURE);
858     }
859 
860     size -= sizeof(hdr);
861 
862     /* Create skeleton */
863     fdt_skel = g_malloc0(size);
864     _FDT((fdt_create(fdt_skel, size)));
865     _FDT((fdt_begin_node(fdt_skel, "")));
866     _FDT((fdt_end_node(fdt_skel)));
867     _FDT((fdt_finish(fdt_skel)));
868     fdt = g_malloc0(size);
869     _FDT((fdt_open_into(fdt_skel, fdt, size)));
870     g_free(fdt_skel);
871 
872     /* Fixup cpu nodes */
873     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
874 
875     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
876         return -1;
877     }
878 
879     /* Pack resulting tree */
880     _FDT((fdt_pack(fdt)));
881 
882     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
883         trace_spapr_cas_failed(size);
884         return -1;
885     }
886 
887     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
888     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
889     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
890     g_free(fdt);
891 
892     return 0;
893 }
894 
895 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
896 {
897     int rtas;
898     GString *hypertas = g_string_sized_new(256);
899     GString *qemu_hypertas = g_string_sized_new(256);
900     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
901     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
902         memory_region_size(&spapr->hotplug_memory.mr);
903     uint32_t lrdr_capacity[] = {
904         cpu_to_be32(max_hotplug_addr >> 32),
905         cpu_to_be32(max_hotplug_addr & 0xffffffff),
906         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
907         cpu_to_be32(max_cpus / smp_threads),
908     };
909 
910     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
911 
912     /* hypertas */
913     add_str(hypertas, "hcall-pft");
914     add_str(hypertas, "hcall-term");
915     add_str(hypertas, "hcall-dabr");
916     add_str(hypertas, "hcall-interrupt");
917     add_str(hypertas, "hcall-tce");
918     add_str(hypertas, "hcall-vio");
919     add_str(hypertas, "hcall-splpar");
920     add_str(hypertas, "hcall-bulk");
921     add_str(hypertas, "hcall-set-mode");
922     add_str(hypertas, "hcall-sprg0");
923     add_str(hypertas, "hcall-copy");
924     add_str(hypertas, "hcall-debug");
925     add_str(qemu_hypertas, "hcall-memop1");
926 
927     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
928         add_str(hypertas, "hcall-multi-tce");
929     }
930 
931     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
932         add_str(hypertas, "hcall-hpt-resize");
933     }
934 
935     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
936                      hypertas->str, hypertas->len));
937     g_string_free(hypertas, TRUE);
938     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
939                      qemu_hypertas->str, qemu_hypertas->len));
940     g_string_free(qemu_hypertas, TRUE);
941 
942     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
943                      refpoints, sizeof(refpoints)));
944 
945     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
946                           RTAS_ERROR_LOG_MAX));
947     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
948                           RTAS_EVENT_SCAN_RATE));
949 
950     g_assert(msi_nonbroken);
951     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
952 
953     /*
954      * According to PAPR, rtas ibm,os-term does not guarantee a return
955      * back to the guest cpu.
956      *
957      * While an additional ibm,extended-os-term property indicates
958      * that rtas call return will always occur. Set this property.
959      */
960     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
961 
962     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
963                      lrdr_capacity, sizeof(lrdr_capacity)));
964 
965     spapr_dt_rtas_tokens(fdt, rtas);
966 }
967 
968 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
969  * that the guest may request and thus the valid values for bytes 24..26 of
970  * option vector 5: */
971 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
972 {
973     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
974 
975     char val[2 * 4] = {
976         23, 0x00, /* Xive mode, filled in below. */
977         24, 0x00, /* Hash/Radix, filled in below. */
978         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
979         26, 0x40, /* Radix options: GTSE == yes. */
980     };
981 
982     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
983                           first_ppc_cpu->compat_pvr)) {
984         /* If we're in a pre POWER9 compat mode then the guest should do hash */
985         val[3] = 0x00; /* Hash */
986     } else if (kvm_enabled()) {
987         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
988             val[3] = 0x80; /* OV5_MMU_BOTH */
989         } else if (kvmppc_has_cap_mmu_radix()) {
990             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
991         } else {
992             val[3] = 0x00; /* Hash */
993         }
994     } else {
995         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
996         val[3] = 0xC0;
997     }
998     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
999                      val, sizeof(val)));
1000 }
1001 
1002 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1003 {
1004     MachineState *machine = MACHINE(spapr);
1005     int chosen;
1006     const char *boot_device = machine->boot_order;
1007     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1008     size_t cb = 0;
1009     char *bootlist = get_boot_devices_list(&cb, true);
1010 
1011     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1012 
1013     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1014     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1015                           spapr->initrd_base));
1016     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1017                           spapr->initrd_base + spapr->initrd_size));
1018 
1019     if (spapr->kernel_size) {
1020         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1021                               cpu_to_be64(spapr->kernel_size) };
1022 
1023         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1024                          &kprop, sizeof(kprop)));
1025         if (spapr->kernel_le) {
1026             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1027         }
1028     }
1029     if (boot_menu) {
1030         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1031     }
1032     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1033     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1034     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1035 
1036     if (cb && bootlist) {
1037         int i;
1038 
1039         for (i = 0; i < cb; i++) {
1040             if (bootlist[i] == '\n') {
1041                 bootlist[i] = ' ';
1042             }
1043         }
1044         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1045     }
1046 
1047     if (boot_device && strlen(boot_device)) {
1048         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1049     }
1050 
1051     if (!spapr->has_graphics && stdout_path) {
1052         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1053     }
1054 
1055     spapr_dt_ov5_platform_support(fdt, chosen);
1056 
1057     g_free(stdout_path);
1058     g_free(bootlist);
1059 }
1060 
1061 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1062 {
1063     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1064      * KVM to work under pHyp with some guest co-operation */
1065     int hypervisor;
1066     uint8_t hypercall[16];
1067 
1068     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1069     /* indicate KVM hypercall interface */
1070     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1071     if (kvmppc_has_cap_fixup_hcalls()) {
1072         /*
1073          * Older KVM versions with older guest kernels were broken
1074          * with the magic page, don't allow the guest to map it.
1075          */
1076         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1077                                   sizeof(hypercall))) {
1078             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1079                              hypercall, sizeof(hypercall)));
1080         }
1081     }
1082 }
1083 
1084 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1085                              hwaddr rtas_addr,
1086                              hwaddr rtas_size)
1087 {
1088     MachineState *machine = MACHINE(spapr);
1089     MachineClass *mc = MACHINE_GET_CLASS(machine);
1090     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1091     int ret;
1092     void *fdt;
1093     sPAPRPHBState *phb;
1094     char *buf;
1095 
1096     fdt = g_malloc0(FDT_MAX_SIZE);
1097     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1098 
1099     /* Root node */
1100     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1101     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1102     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1103 
1104     /*
1105      * Add info to guest to indentify which host is it being run on
1106      * and what is the uuid of the guest
1107      */
1108     if (kvmppc_get_host_model(&buf)) {
1109         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1110         g_free(buf);
1111     }
1112     if (kvmppc_get_host_serial(&buf)) {
1113         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1114         g_free(buf);
1115     }
1116 
1117     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1118 
1119     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1120     if (qemu_uuid_set) {
1121         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1122     }
1123     g_free(buf);
1124 
1125     if (qemu_get_vm_name()) {
1126         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1127                                 qemu_get_vm_name()));
1128     }
1129 
1130     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1131     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1132 
1133     /* /interrupt controller */
1134     spapr_dt_xics(xics_max_server_number(), fdt, PHANDLE_XICP);
1135 
1136     ret = spapr_populate_memory(spapr, fdt);
1137     if (ret < 0) {
1138         error_report("couldn't setup memory nodes in fdt");
1139         exit(1);
1140     }
1141 
1142     /* /vdevice */
1143     spapr_dt_vdevice(spapr->vio_bus, fdt);
1144 
1145     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1146         ret = spapr_rng_populate_dt(fdt);
1147         if (ret < 0) {
1148             error_report("could not set up rng device in the fdt");
1149             exit(1);
1150         }
1151     }
1152 
1153     QLIST_FOREACH(phb, &spapr->phbs, list) {
1154         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1155         if (ret < 0) {
1156             error_report("couldn't setup PCI devices in fdt");
1157             exit(1);
1158         }
1159     }
1160 
1161     /* cpus */
1162     spapr_populate_cpus_dt_node(fdt, spapr);
1163 
1164     if (smc->dr_lmb_enabled) {
1165         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1166     }
1167 
1168     if (mc->has_hotpluggable_cpus) {
1169         int offset = fdt_path_offset(fdt, "/cpus");
1170         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1171                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1172         if (ret < 0) {
1173             error_report("Couldn't set up CPU DR device tree properties");
1174             exit(1);
1175         }
1176     }
1177 
1178     /* /event-sources */
1179     spapr_dt_events(spapr, fdt);
1180 
1181     /* /rtas */
1182     spapr_dt_rtas(spapr, fdt);
1183 
1184     /* /chosen */
1185     spapr_dt_chosen(spapr, fdt);
1186 
1187     /* /hypervisor */
1188     if (kvm_enabled()) {
1189         spapr_dt_hypervisor(spapr, fdt);
1190     }
1191 
1192     /* Build memory reserve map */
1193     if (spapr->kernel_size) {
1194         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1195     }
1196     if (spapr->initrd_size) {
1197         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1198     }
1199 
1200     /* ibm,client-architecture-support updates */
1201     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1202     if (ret < 0) {
1203         error_report("couldn't setup CAS properties fdt");
1204         exit(1);
1205     }
1206 
1207     return fdt;
1208 }
1209 
1210 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1211 {
1212     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1213 }
1214 
1215 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1216                                     PowerPCCPU *cpu)
1217 {
1218     CPUPPCState *env = &cpu->env;
1219 
1220     /* The TCG path should also be holding the BQL at this point */
1221     g_assert(qemu_mutex_iothread_locked());
1222 
1223     if (msr_pr) {
1224         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1225         env->gpr[3] = H_PRIVILEGE;
1226     } else {
1227         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1228     }
1229 }
1230 
1231 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1232 {
1233     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1234 
1235     return spapr->patb_entry;
1236 }
1237 
1238 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1239 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1240 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1241 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1242 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1243 
1244 /*
1245  * Get the fd to access the kernel htab, re-opening it if necessary
1246  */
1247 static int get_htab_fd(sPAPRMachineState *spapr)
1248 {
1249     Error *local_err = NULL;
1250 
1251     if (spapr->htab_fd >= 0) {
1252         return spapr->htab_fd;
1253     }
1254 
1255     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1256     if (spapr->htab_fd < 0) {
1257         error_report_err(local_err);
1258     }
1259 
1260     return spapr->htab_fd;
1261 }
1262 
1263 void close_htab_fd(sPAPRMachineState *spapr)
1264 {
1265     if (spapr->htab_fd >= 0) {
1266         close(spapr->htab_fd);
1267     }
1268     spapr->htab_fd = -1;
1269 }
1270 
1271 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1272 {
1273     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1274 
1275     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1276 }
1277 
1278 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1279 {
1280     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1281 
1282     assert(kvm_enabled());
1283 
1284     if (!spapr->htab) {
1285         return 0;
1286     }
1287 
1288     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1289 }
1290 
1291 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1292                                                 hwaddr ptex, int n)
1293 {
1294     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1295     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1296 
1297     if (!spapr->htab) {
1298         /*
1299          * HTAB is controlled by KVM. Fetch into temporary buffer
1300          */
1301         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1302         kvmppc_read_hptes(hptes, ptex, n);
1303         return hptes;
1304     }
1305 
1306     /*
1307      * HTAB is controlled by QEMU. Just point to the internally
1308      * accessible PTEG.
1309      */
1310     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1311 }
1312 
1313 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1314                               const ppc_hash_pte64_t *hptes,
1315                               hwaddr ptex, int n)
1316 {
1317     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1318 
1319     if (!spapr->htab) {
1320         g_free((void *)hptes);
1321     }
1322 
1323     /* Nothing to do for qemu managed HPT */
1324 }
1325 
1326 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1327                              uint64_t pte0, uint64_t pte1)
1328 {
1329     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1330     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1331 
1332     if (!spapr->htab) {
1333         kvmppc_write_hpte(ptex, pte0, pte1);
1334     } else {
1335         stq_p(spapr->htab + offset, pte0);
1336         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1337     }
1338 }
1339 
1340 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1341 {
1342     int shift;
1343 
1344     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1345      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1346      * that's much more than is needed for Linux guests */
1347     shift = ctz64(pow2ceil(ramsize)) - 7;
1348     shift = MAX(shift, 18); /* Minimum architected size */
1349     shift = MIN(shift, 46); /* Maximum architected size */
1350     return shift;
1351 }
1352 
1353 void spapr_free_hpt(sPAPRMachineState *spapr)
1354 {
1355     g_free(spapr->htab);
1356     spapr->htab = NULL;
1357     spapr->htab_shift = 0;
1358     close_htab_fd(spapr);
1359 }
1360 
1361 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1362                           Error **errp)
1363 {
1364     long rc;
1365 
1366     /* Clean up any HPT info from a previous boot */
1367     spapr_free_hpt(spapr);
1368 
1369     rc = kvmppc_reset_htab(shift);
1370     if (rc < 0) {
1371         /* kernel-side HPT needed, but couldn't allocate one */
1372         error_setg_errno(errp, errno,
1373                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1374                          shift);
1375         /* This is almost certainly fatal, but if the caller really
1376          * wants to carry on with shift == 0, it's welcome to try */
1377     } else if (rc > 0) {
1378         /* kernel-side HPT allocated */
1379         if (rc != shift) {
1380             error_setg(errp,
1381                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1382                        shift, rc);
1383         }
1384 
1385         spapr->htab_shift = shift;
1386         spapr->htab = NULL;
1387     } else {
1388         /* kernel-side HPT not needed, allocate in userspace instead */
1389         size_t size = 1ULL << shift;
1390         int i;
1391 
1392         spapr->htab = qemu_memalign(size, size);
1393         if (!spapr->htab) {
1394             error_setg_errno(errp, errno,
1395                              "Could not allocate HPT of order %d", shift);
1396             return;
1397         }
1398 
1399         memset(spapr->htab, 0, size);
1400         spapr->htab_shift = shift;
1401 
1402         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1403             DIRTY_HPTE(HPTE(spapr->htab, i));
1404         }
1405     }
1406     /* We're setting up a hash table, so that means we're not radix */
1407     spapr->patb_entry = 0;
1408 }
1409 
1410 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1411 {
1412     int hpt_shift;
1413 
1414     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1415         || (spapr->cas_reboot
1416             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1417         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1418     } else {
1419         uint64_t current_ram_size;
1420 
1421         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1422         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1423     }
1424     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1425 
1426     if (spapr->vrma_adjust) {
1427         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1428                                           spapr->htab_shift);
1429     }
1430 }
1431 
1432 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1433 {
1434     bool matched = false;
1435 
1436     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1437         matched = true;
1438     }
1439 
1440     if (!matched) {
1441         error_report("Device %s is not supported by this machine yet.",
1442                      qdev_fw_name(DEVICE(sbdev)));
1443         exit(1);
1444     }
1445 }
1446 
1447 static int spapr_reset_drcs(Object *child, void *opaque)
1448 {
1449     sPAPRDRConnector *drc =
1450         (sPAPRDRConnector *) object_dynamic_cast(child,
1451                                                  TYPE_SPAPR_DR_CONNECTOR);
1452 
1453     if (drc) {
1454         spapr_drc_reset(drc);
1455     }
1456 
1457     return 0;
1458 }
1459 
1460 static void spapr_machine_reset(void)
1461 {
1462     MachineState *machine = MACHINE(qdev_get_machine());
1463     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1464     PowerPCCPU *first_ppc_cpu;
1465     uint32_t rtas_limit;
1466     hwaddr rtas_addr, fdt_addr;
1467     void *fdt;
1468     int rc;
1469 
1470     /* Check for unknown sysbus devices */
1471     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1472 
1473     spapr_caps_reset(spapr);
1474 
1475     first_ppc_cpu = POWERPC_CPU(first_cpu);
1476     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1477         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1478                          spapr->max_compat_pvr)) {
1479         /* If using KVM with radix mode available, VCPUs can be started
1480          * without a HPT because KVM will start them in radix mode.
1481          * Set the GR bit in PATB so that we know there is no HPT. */
1482         spapr->patb_entry = PATBE1_GR;
1483     } else {
1484         spapr_setup_hpt_and_vrma(spapr);
1485     }
1486 
1487     qemu_devices_reset();
1488 
1489     /* DRC reset may cause a device to be unplugged. This will cause troubles
1490      * if this device is used by another device (eg, a running vhost backend
1491      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1492      * situations, we reset DRCs after all devices have been reset.
1493      */
1494     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1495 
1496     spapr_clear_pending_events(spapr);
1497 
1498     /*
1499      * We place the device tree and RTAS just below either the top of the RMA,
1500      * or just below 2GB, whichever is lowere, so that it can be
1501      * processed with 32-bit real mode code if necessary
1502      */
1503     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1504     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1505     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1506 
1507     /* if this reset wasn't generated by CAS, we should reset our
1508      * negotiated options and start from scratch */
1509     if (!spapr->cas_reboot) {
1510         spapr_ovec_cleanup(spapr->ov5_cas);
1511         spapr->ov5_cas = spapr_ovec_new();
1512 
1513         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1514     }
1515 
1516     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1517 
1518     spapr_load_rtas(spapr, fdt, rtas_addr);
1519 
1520     rc = fdt_pack(fdt);
1521 
1522     /* Should only fail if we've built a corrupted tree */
1523     assert(rc == 0);
1524 
1525     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1526         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1527                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1528         exit(1);
1529     }
1530 
1531     /* Load the fdt */
1532     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1533     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1534     g_free(fdt);
1535 
1536     /* Set up the entry state */
1537     first_ppc_cpu->env.gpr[3] = fdt_addr;
1538     first_ppc_cpu->env.gpr[5] = 0;
1539     first_cpu->halted = 0;
1540     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1541 
1542     spapr->cas_reboot = false;
1543 }
1544 
1545 static void spapr_create_nvram(sPAPRMachineState *spapr)
1546 {
1547     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1548     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1549 
1550     if (dinfo) {
1551         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1552                             &error_fatal);
1553     }
1554 
1555     qdev_init_nofail(dev);
1556 
1557     spapr->nvram = (struct sPAPRNVRAM *)dev;
1558 }
1559 
1560 static void spapr_rtc_create(sPAPRMachineState *spapr)
1561 {
1562     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1563     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1564                               &error_fatal);
1565     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1566                               &error_fatal);
1567     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1568                               "date", &error_fatal);
1569 }
1570 
1571 /* Returns whether we want to use VGA or not */
1572 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1573 {
1574     switch (vga_interface_type) {
1575     case VGA_NONE:
1576         return false;
1577     case VGA_DEVICE:
1578         return true;
1579     case VGA_STD:
1580     case VGA_VIRTIO:
1581         return pci_vga_init(pci_bus) != NULL;
1582     default:
1583         error_setg(errp,
1584                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1585         return false;
1586     }
1587 }
1588 
1589 static int spapr_pre_load(void *opaque)
1590 {
1591     int rc;
1592 
1593     rc = spapr_caps_pre_load(opaque);
1594     if (rc) {
1595         return rc;
1596     }
1597 
1598     return 0;
1599 }
1600 
1601 static int spapr_post_load(void *opaque, int version_id)
1602 {
1603     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1604     int err = 0;
1605 
1606     err = spapr_caps_post_migration(spapr);
1607     if (err) {
1608         return err;
1609     }
1610 
1611     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1612         CPUState *cs;
1613         CPU_FOREACH(cs) {
1614             PowerPCCPU *cpu = POWERPC_CPU(cs);
1615             icp_resend(ICP(cpu->intc));
1616         }
1617     }
1618 
1619     /* In earlier versions, there was no separate qdev for the PAPR
1620      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1621      * So when migrating from those versions, poke the incoming offset
1622      * value into the RTC device */
1623     if (version_id < 3) {
1624         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1625     }
1626 
1627     if (kvm_enabled() && spapr->patb_entry) {
1628         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1629         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1630         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1631 
1632         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1633         if (err) {
1634             error_report("Process table config unsupported by the host");
1635             return -EINVAL;
1636         }
1637     }
1638 
1639     return err;
1640 }
1641 
1642 static int spapr_pre_save(void *opaque)
1643 {
1644     int rc;
1645 
1646     rc = spapr_caps_pre_save(opaque);
1647     if (rc) {
1648         return rc;
1649     }
1650 
1651     return 0;
1652 }
1653 
1654 static bool version_before_3(void *opaque, int version_id)
1655 {
1656     return version_id < 3;
1657 }
1658 
1659 static bool spapr_pending_events_needed(void *opaque)
1660 {
1661     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1662     return !QTAILQ_EMPTY(&spapr->pending_events);
1663 }
1664 
1665 static const VMStateDescription vmstate_spapr_event_entry = {
1666     .name = "spapr_event_log_entry",
1667     .version_id = 1,
1668     .minimum_version_id = 1,
1669     .fields = (VMStateField[]) {
1670         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1671         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1672         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1673                                      NULL, extended_length),
1674         VMSTATE_END_OF_LIST()
1675     },
1676 };
1677 
1678 static const VMStateDescription vmstate_spapr_pending_events = {
1679     .name = "spapr_pending_events",
1680     .version_id = 1,
1681     .minimum_version_id = 1,
1682     .needed = spapr_pending_events_needed,
1683     .fields = (VMStateField[]) {
1684         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1685                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1686         VMSTATE_END_OF_LIST()
1687     },
1688 };
1689 
1690 static bool spapr_ov5_cas_needed(void *opaque)
1691 {
1692     sPAPRMachineState *spapr = opaque;
1693     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1694     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1695     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1696     bool cas_needed;
1697 
1698     /* Prior to the introduction of sPAPROptionVector, we had two option
1699      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1700      * Both of these options encode machine topology into the device-tree
1701      * in such a way that the now-booted OS should still be able to interact
1702      * appropriately with QEMU regardless of what options were actually
1703      * negotiatied on the source side.
1704      *
1705      * As such, we can avoid migrating the CAS-negotiated options if these
1706      * are the only options available on the current machine/platform.
1707      * Since these are the only options available for pseries-2.7 and
1708      * earlier, this allows us to maintain old->new/new->old migration
1709      * compatibility.
1710      *
1711      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1712      * via default pseries-2.8 machines and explicit command-line parameters.
1713      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1714      * of the actual CAS-negotiated values to continue working properly. For
1715      * example, availability of memory unplug depends on knowing whether
1716      * OV5_HP_EVT was negotiated via CAS.
1717      *
1718      * Thus, for any cases where the set of available CAS-negotiatable
1719      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1720      * include the CAS-negotiated options in the migration stream.
1721      */
1722     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1723     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1724 
1725     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1726      * the mask itself since in the future it's possible "legacy" bits may be
1727      * removed via machine options, which could generate a false positive
1728      * that breaks migration.
1729      */
1730     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1731     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1732 
1733     spapr_ovec_cleanup(ov5_mask);
1734     spapr_ovec_cleanup(ov5_legacy);
1735     spapr_ovec_cleanup(ov5_removed);
1736 
1737     return cas_needed;
1738 }
1739 
1740 static const VMStateDescription vmstate_spapr_ov5_cas = {
1741     .name = "spapr_option_vector_ov5_cas",
1742     .version_id = 1,
1743     .minimum_version_id = 1,
1744     .needed = spapr_ov5_cas_needed,
1745     .fields = (VMStateField[]) {
1746         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1747                                  vmstate_spapr_ovec, sPAPROptionVector),
1748         VMSTATE_END_OF_LIST()
1749     },
1750 };
1751 
1752 static bool spapr_patb_entry_needed(void *opaque)
1753 {
1754     sPAPRMachineState *spapr = opaque;
1755 
1756     return !!spapr->patb_entry;
1757 }
1758 
1759 static const VMStateDescription vmstate_spapr_patb_entry = {
1760     .name = "spapr_patb_entry",
1761     .version_id = 1,
1762     .minimum_version_id = 1,
1763     .needed = spapr_patb_entry_needed,
1764     .fields = (VMStateField[]) {
1765         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1766         VMSTATE_END_OF_LIST()
1767     },
1768 };
1769 
1770 static const VMStateDescription vmstate_spapr = {
1771     .name = "spapr",
1772     .version_id = 3,
1773     .minimum_version_id = 1,
1774     .pre_load = spapr_pre_load,
1775     .post_load = spapr_post_load,
1776     .pre_save = spapr_pre_save,
1777     .fields = (VMStateField[]) {
1778         /* used to be @next_irq */
1779         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1780 
1781         /* RTC offset */
1782         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1783 
1784         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1785         VMSTATE_END_OF_LIST()
1786     },
1787     .subsections = (const VMStateDescription*[]) {
1788         &vmstate_spapr_ov5_cas,
1789         &vmstate_spapr_patb_entry,
1790         &vmstate_spapr_pending_events,
1791         &vmstate_spapr_cap_htm,
1792         &vmstate_spapr_cap_vsx,
1793         &vmstate_spapr_cap_dfp,
1794         NULL
1795     }
1796 };
1797 
1798 static int htab_save_setup(QEMUFile *f, void *opaque)
1799 {
1800     sPAPRMachineState *spapr = opaque;
1801 
1802     /* "Iteration" header */
1803     if (!spapr->htab_shift) {
1804         qemu_put_be32(f, -1);
1805     } else {
1806         qemu_put_be32(f, spapr->htab_shift);
1807     }
1808 
1809     if (spapr->htab) {
1810         spapr->htab_save_index = 0;
1811         spapr->htab_first_pass = true;
1812     } else {
1813         if (spapr->htab_shift) {
1814             assert(kvm_enabled());
1815         }
1816     }
1817 
1818 
1819     return 0;
1820 }
1821 
1822 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1823                             int chunkstart, int n_valid, int n_invalid)
1824 {
1825     qemu_put_be32(f, chunkstart);
1826     qemu_put_be16(f, n_valid);
1827     qemu_put_be16(f, n_invalid);
1828     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1829                     HASH_PTE_SIZE_64 * n_valid);
1830 }
1831 
1832 static void htab_save_end_marker(QEMUFile *f)
1833 {
1834     qemu_put_be32(f, 0);
1835     qemu_put_be16(f, 0);
1836     qemu_put_be16(f, 0);
1837 }
1838 
1839 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1840                                  int64_t max_ns)
1841 {
1842     bool has_timeout = max_ns != -1;
1843     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1844     int index = spapr->htab_save_index;
1845     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1846 
1847     assert(spapr->htab_first_pass);
1848 
1849     do {
1850         int chunkstart;
1851 
1852         /* Consume invalid HPTEs */
1853         while ((index < htabslots)
1854                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1855             CLEAN_HPTE(HPTE(spapr->htab, index));
1856             index++;
1857         }
1858 
1859         /* Consume valid HPTEs */
1860         chunkstart = index;
1861         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1862                && HPTE_VALID(HPTE(spapr->htab, index))) {
1863             CLEAN_HPTE(HPTE(spapr->htab, index));
1864             index++;
1865         }
1866 
1867         if (index > chunkstart) {
1868             int n_valid = index - chunkstart;
1869 
1870             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
1871 
1872             if (has_timeout &&
1873                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1874                 break;
1875             }
1876         }
1877     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1878 
1879     if (index >= htabslots) {
1880         assert(index == htabslots);
1881         index = 0;
1882         spapr->htab_first_pass = false;
1883     }
1884     spapr->htab_save_index = index;
1885 }
1886 
1887 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1888                                 int64_t max_ns)
1889 {
1890     bool final = max_ns < 0;
1891     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1892     int examined = 0, sent = 0;
1893     int index = spapr->htab_save_index;
1894     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1895 
1896     assert(!spapr->htab_first_pass);
1897 
1898     do {
1899         int chunkstart, invalidstart;
1900 
1901         /* Consume non-dirty HPTEs */
1902         while ((index < htabslots)
1903                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1904             index++;
1905             examined++;
1906         }
1907 
1908         chunkstart = index;
1909         /* Consume valid dirty HPTEs */
1910         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1911                && HPTE_DIRTY(HPTE(spapr->htab, index))
1912                && HPTE_VALID(HPTE(spapr->htab, index))) {
1913             CLEAN_HPTE(HPTE(spapr->htab, index));
1914             index++;
1915             examined++;
1916         }
1917 
1918         invalidstart = index;
1919         /* Consume invalid dirty HPTEs */
1920         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1921                && HPTE_DIRTY(HPTE(spapr->htab, index))
1922                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1923             CLEAN_HPTE(HPTE(spapr->htab, index));
1924             index++;
1925             examined++;
1926         }
1927 
1928         if (index > chunkstart) {
1929             int n_valid = invalidstart - chunkstart;
1930             int n_invalid = index - invalidstart;
1931 
1932             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
1933             sent += index - chunkstart;
1934 
1935             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1936                 break;
1937             }
1938         }
1939 
1940         if (examined >= htabslots) {
1941             break;
1942         }
1943 
1944         if (index >= htabslots) {
1945             assert(index == htabslots);
1946             index = 0;
1947         }
1948     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1949 
1950     if (index >= htabslots) {
1951         assert(index == htabslots);
1952         index = 0;
1953     }
1954 
1955     spapr->htab_save_index = index;
1956 
1957     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1958 }
1959 
1960 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1961 #define MAX_KVM_BUF_SIZE    2048
1962 
1963 static int htab_save_iterate(QEMUFile *f, void *opaque)
1964 {
1965     sPAPRMachineState *spapr = opaque;
1966     int fd;
1967     int rc = 0;
1968 
1969     /* Iteration header */
1970     if (!spapr->htab_shift) {
1971         qemu_put_be32(f, -1);
1972         return 1;
1973     } else {
1974         qemu_put_be32(f, 0);
1975     }
1976 
1977     if (!spapr->htab) {
1978         assert(kvm_enabled());
1979 
1980         fd = get_htab_fd(spapr);
1981         if (fd < 0) {
1982             return fd;
1983         }
1984 
1985         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1986         if (rc < 0) {
1987             return rc;
1988         }
1989     } else  if (spapr->htab_first_pass) {
1990         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1991     } else {
1992         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1993     }
1994 
1995     htab_save_end_marker(f);
1996 
1997     return rc;
1998 }
1999 
2000 static int htab_save_complete(QEMUFile *f, void *opaque)
2001 {
2002     sPAPRMachineState *spapr = opaque;
2003     int fd;
2004 
2005     /* Iteration header */
2006     if (!spapr->htab_shift) {
2007         qemu_put_be32(f, -1);
2008         return 0;
2009     } else {
2010         qemu_put_be32(f, 0);
2011     }
2012 
2013     if (!spapr->htab) {
2014         int rc;
2015 
2016         assert(kvm_enabled());
2017 
2018         fd = get_htab_fd(spapr);
2019         if (fd < 0) {
2020             return fd;
2021         }
2022 
2023         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2024         if (rc < 0) {
2025             return rc;
2026         }
2027     } else {
2028         if (spapr->htab_first_pass) {
2029             htab_save_first_pass(f, spapr, -1);
2030         }
2031         htab_save_later_pass(f, spapr, -1);
2032     }
2033 
2034     /* End marker */
2035     htab_save_end_marker(f);
2036 
2037     return 0;
2038 }
2039 
2040 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2041 {
2042     sPAPRMachineState *spapr = opaque;
2043     uint32_t section_hdr;
2044     int fd = -1;
2045     Error *local_err = NULL;
2046 
2047     if (version_id < 1 || version_id > 1) {
2048         error_report("htab_load() bad version");
2049         return -EINVAL;
2050     }
2051 
2052     section_hdr = qemu_get_be32(f);
2053 
2054     if (section_hdr == -1) {
2055         spapr_free_hpt(spapr);
2056         return 0;
2057     }
2058 
2059     if (section_hdr) {
2060         /* First section gives the htab size */
2061         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2062         if (local_err) {
2063             error_report_err(local_err);
2064             return -EINVAL;
2065         }
2066         return 0;
2067     }
2068 
2069     if (!spapr->htab) {
2070         assert(kvm_enabled());
2071 
2072         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2073         if (fd < 0) {
2074             error_report_err(local_err);
2075             return fd;
2076         }
2077     }
2078 
2079     while (true) {
2080         uint32_t index;
2081         uint16_t n_valid, n_invalid;
2082 
2083         index = qemu_get_be32(f);
2084         n_valid = qemu_get_be16(f);
2085         n_invalid = qemu_get_be16(f);
2086 
2087         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2088             /* End of Stream */
2089             break;
2090         }
2091 
2092         if ((index + n_valid + n_invalid) >
2093             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2094             /* Bad index in stream */
2095             error_report(
2096                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2097                 index, n_valid, n_invalid, spapr->htab_shift);
2098             return -EINVAL;
2099         }
2100 
2101         if (spapr->htab) {
2102             if (n_valid) {
2103                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2104                                 HASH_PTE_SIZE_64 * n_valid);
2105             }
2106             if (n_invalid) {
2107                 memset(HPTE(spapr->htab, index + n_valid), 0,
2108                        HASH_PTE_SIZE_64 * n_invalid);
2109             }
2110         } else {
2111             int rc;
2112 
2113             assert(fd >= 0);
2114 
2115             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2116             if (rc < 0) {
2117                 return rc;
2118             }
2119         }
2120     }
2121 
2122     if (!spapr->htab) {
2123         assert(fd >= 0);
2124         close(fd);
2125     }
2126 
2127     return 0;
2128 }
2129 
2130 static void htab_save_cleanup(void *opaque)
2131 {
2132     sPAPRMachineState *spapr = opaque;
2133 
2134     close_htab_fd(spapr);
2135 }
2136 
2137 static SaveVMHandlers savevm_htab_handlers = {
2138     .save_setup = htab_save_setup,
2139     .save_live_iterate = htab_save_iterate,
2140     .save_live_complete_precopy = htab_save_complete,
2141     .save_cleanup = htab_save_cleanup,
2142     .load_state = htab_load,
2143 };
2144 
2145 static void spapr_boot_set(void *opaque, const char *boot_device,
2146                            Error **errp)
2147 {
2148     MachineState *machine = MACHINE(opaque);
2149     machine->boot_order = g_strdup(boot_device);
2150 }
2151 
2152 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2153 {
2154     MachineState *machine = MACHINE(spapr);
2155     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2156     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2157     int i;
2158 
2159     for (i = 0; i < nr_lmbs; i++) {
2160         uint64_t addr;
2161 
2162         addr = i * lmb_size + spapr->hotplug_memory.base;
2163         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2164                                addr / lmb_size);
2165     }
2166 }
2167 
2168 /*
2169  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2170  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2171  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2172  */
2173 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2174 {
2175     int i;
2176 
2177     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2178         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2179                    " is not aligned to %llu MiB",
2180                    machine->ram_size,
2181                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2182         return;
2183     }
2184 
2185     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2186         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2187                    " is not aligned to %llu MiB",
2188                    machine->ram_size,
2189                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2190         return;
2191     }
2192 
2193     for (i = 0; i < nb_numa_nodes; i++) {
2194         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2195             error_setg(errp,
2196                        "Node %d memory size 0x%" PRIx64
2197                        " is not aligned to %llu MiB",
2198                        i, numa_info[i].node_mem,
2199                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2200             return;
2201         }
2202     }
2203 }
2204 
2205 /* find cpu slot in machine->possible_cpus by core_id */
2206 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2207 {
2208     int index = id / smp_threads;
2209 
2210     if (index >= ms->possible_cpus->len) {
2211         return NULL;
2212     }
2213     if (idx) {
2214         *idx = index;
2215     }
2216     return &ms->possible_cpus->cpus[index];
2217 }
2218 
2219 static void spapr_init_cpus(sPAPRMachineState *spapr)
2220 {
2221     MachineState *machine = MACHINE(spapr);
2222     MachineClass *mc = MACHINE_GET_CLASS(machine);
2223     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2224     int smt = kvmppc_smt_threads();
2225     const CPUArchIdList *possible_cpus;
2226     int boot_cores_nr = smp_cpus / smp_threads;
2227     int i;
2228 
2229     if (!type) {
2230         error_report("Unable to find sPAPR CPU Core definition");
2231         exit(1);
2232     }
2233 
2234     possible_cpus = mc->possible_cpu_arch_ids(machine);
2235     if (mc->has_hotpluggable_cpus) {
2236         if (smp_cpus % smp_threads) {
2237             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2238                          smp_cpus, smp_threads);
2239             exit(1);
2240         }
2241         if (max_cpus % smp_threads) {
2242             error_report("max_cpus (%u) must be multiple of threads (%u)",
2243                          max_cpus, smp_threads);
2244             exit(1);
2245         }
2246     } else {
2247         if (max_cpus != smp_cpus) {
2248             error_report("This machine version does not support CPU hotplug");
2249             exit(1);
2250         }
2251         boot_cores_nr = possible_cpus->len;
2252     }
2253 
2254     for (i = 0; i < possible_cpus->len; i++) {
2255         int core_id = i * smp_threads;
2256 
2257         if (mc->has_hotpluggable_cpus) {
2258             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2259                                    (core_id / smp_threads) * smt);
2260         }
2261 
2262         if (i < boot_cores_nr) {
2263             Object *core  = object_new(type);
2264             int nr_threads = smp_threads;
2265 
2266             /* Handle the partially filled core for older machine types */
2267             if ((i + 1) * smp_threads >= smp_cpus) {
2268                 nr_threads = smp_cpus - i * smp_threads;
2269             }
2270 
2271             object_property_set_int(core, nr_threads, "nr-threads",
2272                                     &error_fatal);
2273             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2274                                     &error_fatal);
2275             object_property_set_bool(core, true, "realized", &error_fatal);
2276         }
2277     }
2278 }
2279 
2280 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2281 {
2282     Error *local_err = NULL;
2283     bool vsmt_user = !!spapr->vsmt;
2284     int kvm_smt = kvmppc_smt_threads();
2285     int ret;
2286 
2287     if (!kvm_enabled() && (smp_threads > 1)) {
2288         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2289                      "on a pseries machine");
2290         goto out;
2291     }
2292     if (!is_power_of_2(smp_threads)) {
2293         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2294                      "machine because it must be a power of 2", smp_threads);
2295         goto out;
2296     }
2297 
2298     /* Detemine the VSMT mode to use: */
2299     if (vsmt_user) {
2300         if (spapr->vsmt < smp_threads) {
2301             error_setg(&local_err, "Cannot support VSMT mode %d"
2302                          " because it must be >= threads/core (%d)",
2303                          spapr->vsmt, smp_threads);
2304             goto out;
2305         }
2306         /* In this case, spapr->vsmt has been set by the command line */
2307     } else {
2308         /*
2309          * Default VSMT value is tricky, because we need it to be as
2310          * consistent as possible (for migration), but this requires
2311          * changing it for at least some existing cases.  We pick 8 as
2312          * the value that we'd get with KVM on POWER8, the
2313          * overwhelmingly common case in production systems.
2314          */
2315         spapr->vsmt = 8;
2316     }
2317 
2318     /* KVM: If necessary, set the SMT mode: */
2319     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2320         ret = kvmppc_set_smt_threads(spapr->vsmt);
2321         if (ret) {
2322             /* Looks like KVM isn't able to change VSMT mode */
2323             error_setg(&local_err,
2324                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2325                        spapr->vsmt, ret);
2326             /* We can live with that if the default one is big enough
2327              * for the number of threads, and a submultiple of the one
2328              * we want.  In this case we'll waste some vcpu ids, but
2329              * behaviour will be correct */
2330             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2331                 warn_report_err(local_err);
2332                 local_err = NULL;
2333                 goto out;
2334             } else {
2335                 if (!vsmt_user) {
2336                     error_append_hint(&local_err,
2337                                       "On PPC, a VM with %d threads/core"
2338                                       " on a host with %d threads/core"
2339                                       " requires the use of VSMT mode %d.\n",
2340                                       smp_threads, kvm_smt, spapr->vsmt);
2341                 }
2342                 kvmppc_hint_smt_possible(&local_err);
2343                 goto out;
2344             }
2345         }
2346     }
2347     /* else TCG: nothing to do currently */
2348 out:
2349     error_propagate(errp, local_err);
2350 }
2351 
2352 /* pSeries LPAR / sPAPR hardware init */
2353 static void spapr_machine_init(MachineState *machine)
2354 {
2355     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2356     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2357     const char *kernel_filename = machine->kernel_filename;
2358     const char *initrd_filename = machine->initrd_filename;
2359     PCIHostState *phb;
2360     int i;
2361     MemoryRegion *sysmem = get_system_memory();
2362     MemoryRegion *ram = g_new(MemoryRegion, 1);
2363     MemoryRegion *rma_region;
2364     void *rma = NULL;
2365     hwaddr rma_alloc_size;
2366     hwaddr node0_size = spapr_node0_size(machine);
2367     long load_limit, fw_size;
2368     char *filename;
2369     Error *resize_hpt_err = NULL;
2370 
2371     msi_nonbroken = true;
2372 
2373     QLIST_INIT(&spapr->phbs);
2374     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2375 
2376     /* Check HPT resizing availability */
2377     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2378     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2379         /*
2380          * If the user explicitly requested a mode we should either
2381          * supply it, or fail completely (which we do below).  But if
2382          * it's not set explicitly, we reset our mode to something
2383          * that works
2384          */
2385         if (resize_hpt_err) {
2386             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2387             error_free(resize_hpt_err);
2388             resize_hpt_err = NULL;
2389         } else {
2390             spapr->resize_hpt = smc->resize_hpt_default;
2391         }
2392     }
2393 
2394     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2395 
2396     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2397         /*
2398          * User requested HPT resize, but this host can't supply it.  Bail out
2399          */
2400         error_report_err(resize_hpt_err);
2401         exit(1);
2402     }
2403 
2404     /* Allocate RMA if necessary */
2405     rma_alloc_size = kvmppc_alloc_rma(&rma);
2406 
2407     if (rma_alloc_size == -1) {
2408         error_report("Unable to create RMA");
2409         exit(1);
2410     }
2411 
2412     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2413         spapr->rma_size = rma_alloc_size;
2414     } else {
2415         spapr->rma_size = node0_size;
2416 
2417         /* With KVM, we don't actually know whether KVM supports an
2418          * unbounded RMA (PR KVM) or is limited by the hash table size
2419          * (HV KVM using VRMA), so we always assume the latter
2420          *
2421          * In that case, we also limit the initial allocations for RTAS
2422          * etc... to 256M since we have no way to know what the VRMA size
2423          * is going to be as it depends on the size of the hash table
2424          * isn't determined yet.
2425          */
2426         if (kvm_enabled()) {
2427             spapr->vrma_adjust = 1;
2428             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2429         }
2430 
2431         /* Actually we don't support unbounded RMA anymore since we
2432          * added proper emulation of HV mode. The max we can get is
2433          * 16G which also happens to be what we configure for PAPR
2434          * mode so make sure we don't do anything bigger than that
2435          */
2436         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2437     }
2438 
2439     if (spapr->rma_size > node0_size) {
2440         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2441                      spapr->rma_size);
2442         exit(1);
2443     }
2444 
2445     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2446     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2447 
2448     /* Set up Interrupt Controller before we create the VCPUs */
2449     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2450 
2451     /* Set up containers for ibm,client-architecture-support negotiated options
2452      */
2453     spapr->ov5 = spapr_ovec_new();
2454     spapr->ov5_cas = spapr_ovec_new();
2455 
2456     if (smc->dr_lmb_enabled) {
2457         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2458         spapr_validate_node_memory(machine, &error_fatal);
2459     }
2460 
2461     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2462     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2463         /* KVM and TCG always allow GTSE with radix... */
2464         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2465     }
2466     /* ... but not with hash (currently). */
2467 
2468     /* advertise support for dedicated HP event source to guests */
2469     if (spapr->use_hotplug_event_source) {
2470         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2471     }
2472 
2473     /* advertise support for HPT resizing */
2474     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2475         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2476     }
2477 
2478     /* init CPUs */
2479     spapr_set_vsmt_mode(spapr, &error_fatal);
2480 
2481     spapr_init_cpus(spapr);
2482 
2483     if (kvm_enabled()) {
2484         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2485         kvmppc_enable_logical_ci_hcalls();
2486         kvmppc_enable_set_mode_hcall();
2487 
2488         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2489         kvmppc_enable_clear_ref_mod_hcalls();
2490     }
2491 
2492     /* allocate RAM */
2493     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2494                                          machine->ram_size);
2495     memory_region_add_subregion(sysmem, 0, ram);
2496 
2497     if (rma_alloc_size && rma) {
2498         rma_region = g_new(MemoryRegion, 1);
2499         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2500                                    rma_alloc_size, rma);
2501         vmstate_register_ram_global(rma_region);
2502         memory_region_add_subregion(sysmem, 0, rma_region);
2503     }
2504 
2505     /* initialize hotplug memory address space */
2506     if (machine->ram_size < machine->maxram_size) {
2507         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2508         /*
2509          * Limit the number of hotpluggable memory slots to half the number
2510          * slots that KVM supports, leaving the other half for PCI and other
2511          * devices. However ensure that number of slots doesn't drop below 32.
2512          */
2513         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2514                            SPAPR_MAX_RAM_SLOTS;
2515 
2516         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2517             max_memslots = SPAPR_MAX_RAM_SLOTS;
2518         }
2519         if (machine->ram_slots > max_memslots) {
2520             error_report("Specified number of memory slots %"
2521                          PRIu64" exceeds max supported %d",
2522                          machine->ram_slots, max_memslots);
2523             exit(1);
2524         }
2525 
2526         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2527                                               SPAPR_HOTPLUG_MEM_ALIGN);
2528         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2529                            "hotplug-memory", hotplug_mem_size);
2530         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2531                                     &spapr->hotplug_memory.mr);
2532     }
2533 
2534     if (smc->dr_lmb_enabled) {
2535         spapr_create_lmb_dr_connectors(spapr);
2536     }
2537 
2538     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2539     if (!filename) {
2540         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2541         exit(1);
2542     }
2543     spapr->rtas_size = get_image_size(filename);
2544     if (spapr->rtas_size < 0) {
2545         error_report("Could not get size of LPAR rtas '%s'", filename);
2546         exit(1);
2547     }
2548     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2549     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2550         error_report("Could not load LPAR rtas '%s'", filename);
2551         exit(1);
2552     }
2553     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2554         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2555                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2556         exit(1);
2557     }
2558     g_free(filename);
2559 
2560     /* Set up RTAS event infrastructure */
2561     spapr_events_init(spapr);
2562 
2563     /* Set up the RTC RTAS interfaces */
2564     spapr_rtc_create(spapr);
2565 
2566     /* Set up VIO bus */
2567     spapr->vio_bus = spapr_vio_bus_init();
2568 
2569     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2570         if (serial_hds[i]) {
2571             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2572         }
2573     }
2574 
2575     /* We always have at least the nvram device on VIO */
2576     spapr_create_nvram(spapr);
2577 
2578     /* Set up PCI */
2579     spapr_pci_rtas_init();
2580 
2581     phb = spapr_create_phb(spapr, 0);
2582 
2583     for (i = 0; i < nb_nics; i++) {
2584         NICInfo *nd = &nd_table[i];
2585 
2586         if (!nd->model) {
2587             nd->model = g_strdup("ibmveth");
2588         }
2589 
2590         if (strcmp(nd->model, "ibmveth") == 0) {
2591             spapr_vlan_create(spapr->vio_bus, nd);
2592         } else {
2593             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2594         }
2595     }
2596 
2597     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2598         spapr_vscsi_create(spapr->vio_bus);
2599     }
2600 
2601     /* Graphics */
2602     if (spapr_vga_init(phb->bus, &error_fatal)) {
2603         spapr->has_graphics = true;
2604         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2605     }
2606 
2607     if (machine->usb) {
2608         if (smc->use_ohci_by_default) {
2609             pci_create_simple(phb->bus, -1, "pci-ohci");
2610         } else {
2611             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2612         }
2613 
2614         if (spapr->has_graphics) {
2615             USBBus *usb_bus = usb_bus_find(-1);
2616 
2617             usb_create_simple(usb_bus, "usb-kbd");
2618             usb_create_simple(usb_bus, "usb-mouse");
2619         }
2620     }
2621 
2622     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2623         error_report(
2624             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2625             MIN_RMA_SLOF);
2626         exit(1);
2627     }
2628 
2629     if (kernel_filename) {
2630         uint64_t lowaddr = 0;
2631 
2632         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2633                                       NULL, NULL, &lowaddr, NULL, 1,
2634                                       PPC_ELF_MACHINE, 0, 0);
2635         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2636             spapr->kernel_size = load_elf(kernel_filename,
2637                                           translate_kernel_address, NULL, NULL,
2638                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2639                                           0, 0);
2640             spapr->kernel_le = spapr->kernel_size > 0;
2641         }
2642         if (spapr->kernel_size < 0) {
2643             error_report("error loading %s: %s", kernel_filename,
2644                          load_elf_strerror(spapr->kernel_size));
2645             exit(1);
2646         }
2647 
2648         /* load initrd */
2649         if (initrd_filename) {
2650             /* Try to locate the initrd in the gap between the kernel
2651              * and the firmware. Add a bit of space just in case
2652              */
2653             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2654                                   + 0x1ffff) & ~0xffff;
2655             spapr->initrd_size = load_image_targphys(initrd_filename,
2656                                                      spapr->initrd_base,
2657                                                      load_limit
2658                                                      - spapr->initrd_base);
2659             if (spapr->initrd_size < 0) {
2660                 error_report("could not load initial ram disk '%s'",
2661                              initrd_filename);
2662                 exit(1);
2663             }
2664         }
2665     }
2666 
2667     if (bios_name == NULL) {
2668         bios_name = FW_FILE_NAME;
2669     }
2670     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2671     if (!filename) {
2672         error_report("Could not find LPAR firmware '%s'", bios_name);
2673         exit(1);
2674     }
2675     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2676     if (fw_size <= 0) {
2677         error_report("Could not load LPAR firmware '%s'", filename);
2678         exit(1);
2679     }
2680     g_free(filename);
2681 
2682     /* FIXME: Should register things through the MachineState's qdev
2683      * interface, this is a legacy from the sPAPREnvironment structure
2684      * which predated MachineState but had a similar function */
2685     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2686     register_savevm_live(NULL, "spapr/htab", -1, 1,
2687                          &savevm_htab_handlers, spapr);
2688 
2689     qemu_register_boot_set(spapr_boot_set, spapr);
2690 
2691     if (kvm_enabled()) {
2692         /* to stop and start vmclock */
2693         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2694                                          &spapr->tb);
2695 
2696         kvmppc_spapr_enable_inkernel_multitce();
2697     }
2698 }
2699 
2700 static int spapr_kvm_type(const char *vm_type)
2701 {
2702     if (!vm_type) {
2703         return 0;
2704     }
2705 
2706     if (!strcmp(vm_type, "HV")) {
2707         return 1;
2708     }
2709 
2710     if (!strcmp(vm_type, "PR")) {
2711         return 2;
2712     }
2713 
2714     error_report("Unknown kvm-type specified '%s'", vm_type);
2715     exit(1);
2716 }
2717 
2718 /*
2719  * Implementation of an interface to adjust firmware path
2720  * for the bootindex property handling.
2721  */
2722 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2723                                    DeviceState *dev)
2724 {
2725 #define CAST(type, obj, name) \
2726     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2727     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2728     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2729     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2730 
2731     if (d) {
2732         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2733         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2734         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2735 
2736         if (spapr) {
2737             /*
2738              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2739              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2740              * in the top 16 bits of the 64-bit LUN
2741              */
2742             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2743             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2744                                    (uint64_t)id << 48);
2745         } else if (virtio) {
2746             /*
2747              * We use SRP luns of the form 01000000 | (target << 8) | lun
2748              * in the top 32 bits of the 64-bit LUN
2749              * Note: the quote above is from SLOF and it is wrong,
2750              * the actual binding is:
2751              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2752              */
2753             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2754             if (d->lun >= 256) {
2755                 /* Use the LUN "flat space addressing method" */
2756                 id |= 0x4000;
2757             }
2758             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2759                                    (uint64_t)id << 32);
2760         } else if (usb) {
2761             /*
2762              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2763              * in the top 32 bits of the 64-bit LUN
2764              */
2765             unsigned usb_port = atoi(usb->port->path);
2766             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2767             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2768                                    (uint64_t)id << 32);
2769         }
2770     }
2771 
2772     /*
2773      * SLOF probes the USB devices, and if it recognizes that the device is a
2774      * storage device, it changes its name to "storage" instead of "usb-host",
2775      * and additionally adds a child node for the SCSI LUN, so the correct
2776      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2777      */
2778     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2779         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2780         if (usb_host_dev_is_scsi_storage(usbdev)) {
2781             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2782         }
2783     }
2784 
2785     if (phb) {
2786         /* Replace "pci" with "pci@800000020000000" */
2787         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2788     }
2789 
2790     if (vsc) {
2791         /* Same logic as virtio above */
2792         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2793         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2794     }
2795 
2796     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2797         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2798         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2799         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2800     }
2801 
2802     return NULL;
2803 }
2804 
2805 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2806 {
2807     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2808 
2809     return g_strdup(spapr->kvm_type);
2810 }
2811 
2812 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2813 {
2814     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2815 
2816     g_free(spapr->kvm_type);
2817     spapr->kvm_type = g_strdup(value);
2818 }
2819 
2820 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2821 {
2822     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2823 
2824     return spapr->use_hotplug_event_source;
2825 }
2826 
2827 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2828                                             Error **errp)
2829 {
2830     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2831 
2832     spapr->use_hotplug_event_source = value;
2833 }
2834 
2835 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2836 {
2837     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2838 
2839     switch (spapr->resize_hpt) {
2840     case SPAPR_RESIZE_HPT_DEFAULT:
2841         return g_strdup("default");
2842     case SPAPR_RESIZE_HPT_DISABLED:
2843         return g_strdup("disabled");
2844     case SPAPR_RESIZE_HPT_ENABLED:
2845         return g_strdup("enabled");
2846     case SPAPR_RESIZE_HPT_REQUIRED:
2847         return g_strdup("required");
2848     }
2849     g_assert_not_reached();
2850 }
2851 
2852 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2853 {
2854     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2855 
2856     if (strcmp(value, "default") == 0) {
2857         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2858     } else if (strcmp(value, "disabled") == 0) {
2859         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2860     } else if (strcmp(value, "enabled") == 0) {
2861         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2862     } else if (strcmp(value, "required") == 0) {
2863         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2864     } else {
2865         error_setg(errp, "Bad value for \"resize-hpt\" property");
2866     }
2867 }
2868 
2869 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
2870                                    void *opaque, Error **errp)
2871 {
2872     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2873 }
2874 
2875 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
2876                                    void *opaque, Error **errp)
2877 {
2878     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2879 }
2880 
2881 static void spapr_instance_init(Object *obj)
2882 {
2883     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2884 
2885     spapr->htab_fd = -1;
2886     spapr->use_hotplug_event_source = true;
2887     object_property_add_str(obj, "kvm-type",
2888                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2889     object_property_set_description(obj, "kvm-type",
2890                                     "Specifies the KVM virtualization mode (HV, PR)",
2891                                     NULL);
2892     object_property_add_bool(obj, "modern-hotplug-events",
2893                             spapr_get_modern_hotplug_events,
2894                             spapr_set_modern_hotplug_events,
2895                             NULL);
2896     object_property_set_description(obj, "modern-hotplug-events",
2897                                     "Use dedicated hotplug event mechanism in"
2898                                     " place of standard EPOW events when possible"
2899                                     " (required for memory hot-unplug support)",
2900                                     NULL);
2901 
2902     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
2903                             "Maximum permitted CPU compatibility mode",
2904                             &error_fatal);
2905 
2906     object_property_add_str(obj, "resize-hpt",
2907                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
2908     object_property_set_description(obj, "resize-hpt",
2909                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
2910                                     NULL);
2911     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
2912                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
2913     object_property_set_description(obj, "vsmt",
2914                                     "Virtual SMT: KVM behaves as if this were"
2915                                     " the host's SMT mode", &error_abort);
2916 }
2917 
2918 static void spapr_machine_finalizefn(Object *obj)
2919 {
2920     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2921 
2922     g_free(spapr->kvm_type);
2923 }
2924 
2925 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2926 {
2927     cpu_synchronize_state(cs);
2928     ppc_cpu_do_system_reset(cs);
2929 }
2930 
2931 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2932 {
2933     CPUState *cs;
2934 
2935     CPU_FOREACH(cs) {
2936         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2937     }
2938 }
2939 
2940 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2941                            uint32_t node, bool dedicated_hp_event_source,
2942                            Error **errp)
2943 {
2944     sPAPRDRConnector *drc;
2945     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2946     int i, fdt_offset, fdt_size;
2947     void *fdt;
2948     uint64_t addr = addr_start;
2949     bool hotplugged = spapr_drc_hotplugged(dev);
2950     Error *local_err = NULL;
2951 
2952     for (i = 0; i < nr_lmbs; i++) {
2953         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2954                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2955         g_assert(drc);
2956 
2957         fdt = create_device_tree(&fdt_size);
2958         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2959                                                 SPAPR_MEMORY_BLOCK_SIZE);
2960 
2961         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
2962         if (local_err) {
2963             while (addr > addr_start) {
2964                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
2965                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2966                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
2967                 spapr_drc_detach(drc);
2968             }
2969             g_free(fdt);
2970             error_propagate(errp, local_err);
2971             return;
2972         }
2973         if (!hotplugged) {
2974             spapr_drc_reset(drc);
2975         }
2976         addr += SPAPR_MEMORY_BLOCK_SIZE;
2977     }
2978     /* send hotplug notification to the
2979      * guest only in case of hotplugged memory
2980      */
2981     if (hotplugged) {
2982         if (dedicated_hp_event_source) {
2983             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2984                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2985             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2986                                                    nr_lmbs,
2987                                                    spapr_drc_index(drc));
2988         } else {
2989             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2990                                            nr_lmbs);
2991         }
2992     }
2993 }
2994 
2995 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2996                               uint32_t node, Error **errp)
2997 {
2998     Error *local_err = NULL;
2999     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3000     PCDIMMDevice *dimm = PC_DIMM(dev);
3001     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3002     MemoryRegion *mr;
3003     uint64_t align, size, addr;
3004 
3005     mr = ddc->get_memory_region(dimm, &local_err);
3006     if (local_err) {
3007         goto out;
3008     }
3009     align = memory_region_get_alignment(mr);
3010     size = memory_region_size(mr);
3011 
3012     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
3013     if (local_err) {
3014         goto out;
3015     }
3016 
3017     addr = object_property_get_uint(OBJECT(dimm),
3018                                     PC_DIMM_ADDR_PROP, &local_err);
3019     if (local_err) {
3020         goto out_unplug;
3021     }
3022 
3023     spapr_add_lmbs(dev, addr, size, node,
3024                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3025                    &local_err);
3026     if (local_err) {
3027         goto out_unplug;
3028     }
3029 
3030     return;
3031 
3032 out_unplug:
3033     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
3034 out:
3035     error_propagate(errp, local_err);
3036 }
3037 
3038 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3039                                   Error **errp)
3040 {
3041     PCDIMMDevice *dimm = PC_DIMM(dev);
3042     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3043     MemoryRegion *mr;
3044     uint64_t size;
3045     char *mem_dev;
3046 
3047     mr = ddc->get_memory_region(dimm, errp);
3048     if (!mr) {
3049         return;
3050     }
3051     size = memory_region_size(mr);
3052 
3053     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3054         error_setg(errp, "Hotplugged memory size must be a multiple of "
3055                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
3056         return;
3057     }
3058 
3059     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
3060     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
3061         error_setg(errp, "Memory backend has bad page size. "
3062                    "Use 'memory-backend-file' with correct mem-path.");
3063         goto out;
3064     }
3065 
3066 out:
3067     g_free(mem_dev);
3068 }
3069 
3070 struct sPAPRDIMMState {
3071     PCDIMMDevice *dimm;
3072     uint32_t nr_lmbs;
3073     QTAILQ_ENTRY(sPAPRDIMMState) next;
3074 };
3075 
3076 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3077                                                        PCDIMMDevice *dimm)
3078 {
3079     sPAPRDIMMState *dimm_state = NULL;
3080 
3081     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3082         if (dimm_state->dimm == dimm) {
3083             break;
3084         }
3085     }
3086     return dimm_state;
3087 }
3088 
3089 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3090                                                       uint32_t nr_lmbs,
3091                                                       PCDIMMDevice *dimm)
3092 {
3093     sPAPRDIMMState *ds = NULL;
3094 
3095     /*
3096      * If this request is for a DIMM whose removal had failed earlier
3097      * (due to guest's refusal to remove the LMBs), we would have this
3098      * dimm already in the pending_dimm_unplugs list. In that
3099      * case don't add again.
3100      */
3101     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3102     if (!ds) {
3103         ds = g_malloc0(sizeof(sPAPRDIMMState));
3104         ds->nr_lmbs = nr_lmbs;
3105         ds->dimm = dimm;
3106         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3107     }
3108     return ds;
3109 }
3110 
3111 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3112                                               sPAPRDIMMState *dimm_state)
3113 {
3114     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3115     g_free(dimm_state);
3116 }
3117 
3118 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3119                                                         PCDIMMDevice *dimm)
3120 {
3121     sPAPRDRConnector *drc;
3122     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3123     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3124     uint64_t size = memory_region_size(mr);
3125     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3126     uint32_t avail_lmbs = 0;
3127     uint64_t addr_start, addr;
3128     int i;
3129 
3130     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3131                                          &error_abort);
3132 
3133     addr = addr_start;
3134     for (i = 0; i < nr_lmbs; i++) {
3135         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3136                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3137         g_assert(drc);
3138         if (drc->dev) {
3139             avail_lmbs++;
3140         }
3141         addr += SPAPR_MEMORY_BLOCK_SIZE;
3142     }
3143 
3144     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3145 }
3146 
3147 /* Callback to be called during DRC release. */
3148 void spapr_lmb_release(DeviceState *dev)
3149 {
3150     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_hotplug_handler(dev));
3151     PCDIMMDevice *dimm = PC_DIMM(dev);
3152     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3153     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3154     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3155 
3156     /* This information will get lost if a migration occurs
3157      * during the unplug process. In this case recover it. */
3158     if (ds == NULL) {
3159         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3160         g_assert(ds);
3161         /* The DRC being examined by the caller at least must be counted */
3162         g_assert(ds->nr_lmbs);
3163     }
3164 
3165     if (--ds->nr_lmbs) {
3166         return;
3167     }
3168 
3169     /*
3170      * Now that all the LMBs have been removed by the guest, call the
3171      * pc-dimm unplug handler to cleanup up the pc-dimm device.
3172      */
3173     pc_dimm_memory_unplug(dev, &spapr->hotplug_memory, mr);
3174     object_unparent(OBJECT(dev));
3175     spapr_pending_dimm_unplugs_remove(spapr, ds);
3176 }
3177 
3178 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3179                                         DeviceState *dev, Error **errp)
3180 {
3181     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3182     Error *local_err = NULL;
3183     PCDIMMDevice *dimm = PC_DIMM(dev);
3184     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3185     MemoryRegion *mr;
3186     uint32_t nr_lmbs;
3187     uint64_t size, addr_start, addr;
3188     int i;
3189     sPAPRDRConnector *drc;
3190 
3191     mr = ddc->get_memory_region(dimm, &local_err);
3192     if (local_err) {
3193         goto out;
3194     }
3195     size = memory_region_size(mr);
3196     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3197 
3198     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3199                                          &local_err);
3200     if (local_err) {
3201         goto out;
3202     }
3203 
3204     /*
3205      * An existing pending dimm state for this DIMM means that there is an
3206      * unplug operation in progress, waiting for the spapr_lmb_release
3207      * callback to complete the job (BQL can't cover that far). In this case,
3208      * bail out to avoid detaching DRCs that were already released.
3209      */
3210     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3211         error_setg(&local_err,
3212                    "Memory unplug already in progress for device %s",
3213                    dev->id);
3214         goto out;
3215     }
3216 
3217     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3218 
3219     addr = addr_start;
3220     for (i = 0; i < nr_lmbs; i++) {
3221         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3222                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3223         g_assert(drc);
3224 
3225         spapr_drc_detach(drc);
3226         addr += SPAPR_MEMORY_BLOCK_SIZE;
3227     }
3228 
3229     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3230                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3231     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3232                                               nr_lmbs, spapr_drc_index(drc));
3233 out:
3234     error_propagate(errp, local_err);
3235 }
3236 
3237 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3238                                            sPAPRMachineState *spapr)
3239 {
3240     PowerPCCPU *cpu = POWERPC_CPU(cs);
3241     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3242     int id = spapr_vcpu_id(cpu);
3243     void *fdt;
3244     int offset, fdt_size;
3245     char *nodename;
3246 
3247     fdt = create_device_tree(&fdt_size);
3248     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3249     offset = fdt_add_subnode(fdt, 0, nodename);
3250 
3251     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3252     g_free(nodename);
3253 
3254     *fdt_offset = offset;
3255     return fdt;
3256 }
3257 
3258 /* Callback to be called during DRC release. */
3259 void spapr_core_release(DeviceState *dev)
3260 {
3261     MachineState *ms = MACHINE(qdev_get_hotplug_handler(dev));
3262     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3263     CPUCore *cc = CPU_CORE(dev);
3264     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3265 
3266     if (smc->pre_2_10_has_unused_icps) {
3267         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3268         int i;
3269 
3270         for (i = 0; i < cc->nr_threads; i++) {
3271             CPUState *cs = CPU(sc->threads[i]);
3272 
3273             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3274         }
3275     }
3276 
3277     assert(core_slot);
3278     core_slot->cpu = NULL;
3279     object_unparent(OBJECT(dev));
3280 }
3281 
3282 static
3283 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3284                                Error **errp)
3285 {
3286     int index;
3287     sPAPRDRConnector *drc;
3288     CPUCore *cc = CPU_CORE(dev);
3289     int smt = kvmppc_smt_threads();
3290 
3291     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3292         error_setg(errp, "Unable to find CPU core with core-id: %d",
3293                    cc->core_id);
3294         return;
3295     }
3296     if (index == 0) {
3297         error_setg(errp, "Boot CPU core may not be unplugged");
3298         return;
3299     }
3300 
3301     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3302     g_assert(drc);
3303 
3304     spapr_drc_detach(drc);
3305 
3306     spapr_hotplug_req_remove_by_index(drc);
3307 }
3308 
3309 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3310                             Error **errp)
3311 {
3312     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3313     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3314     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3315     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3316     CPUCore *cc = CPU_CORE(dev);
3317     CPUState *cs = CPU(core->threads[0]);
3318     sPAPRDRConnector *drc;
3319     Error *local_err = NULL;
3320     int smt = kvmppc_smt_threads();
3321     CPUArchId *core_slot;
3322     int index;
3323     bool hotplugged = spapr_drc_hotplugged(dev);
3324 
3325     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3326     if (!core_slot) {
3327         error_setg(errp, "Unable to find CPU core with core-id: %d",
3328                    cc->core_id);
3329         return;
3330     }
3331     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3332 
3333     g_assert(drc || !mc->has_hotpluggable_cpus);
3334 
3335     if (drc) {
3336         void *fdt;
3337         int fdt_offset;
3338 
3339         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3340 
3341         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3342         if (local_err) {
3343             g_free(fdt);
3344             error_propagate(errp, local_err);
3345             return;
3346         }
3347 
3348         if (hotplugged) {
3349             /*
3350              * Send hotplug notification interrupt to the guest only
3351              * in case of hotplugged CPUs.
3352              */
3353             spapr_hotplug_req_add_by_index(drc);
3354         } else {
3355             spapr_drc_reset(drc);
3356         }
3357     }
3358 
3359     core_slot->cpu = OBJECT(dev);
3360 
3361     if (smc->pre_2_10_has_unused_icps) {
3362         int i;
3363 
3364         for (i = 0; i < cc->nr_threads; i++) {
3365             sPAPRCPUCore *sc = SPAPR_CPU_CORE(dev);
3366 
3367             cs = CPU(sc->threads[i]);
3368             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3369         }
3370     }
3371 }
3372 
3373 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3374                                 Error **errp)
3375 {
3376     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3377     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3378     Error *local_err = NULL;
3379     CPUCore *cc = CPU_CORE(dev);
3380     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3381     const char *type = object_get_typename(OBJECT(dev));
3382     CPUArchId *core_slot;
3383     int index;
3384 
3385     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3386         error_setg(&local_err, "CPU hotplug not supported for this machine");
3387         goto out;
3388     }
3389 
3390     if (strcmp(base_core_type, type)) {
3391         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3392         goto out;
3393     }
3394 
3395     if (cc->core_id % smp_threads) {
3396         error_setg(&local_err, "invalid core id %d", cc->core_id);
3397         goto out;
3398     }
3399 
3400     /*
3401      * In general we should have homogeneous threads-per-core, but old
3402      * (pre hotplug support) machine types allow the last core to have
3403      * reduced threads as a compatibility hack for when we allowed
3404      * total vcpus not a multiple of threads-per-core.
3405      */
3406     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3407         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3408                    cc->nr_threads, smp_threads);
3409         goto out;
3410     }
3411 
3412     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3413     if (!core_slot) {
3414         error_setg(&local_err, "core id %d out of range", cc->core_id);
3415         goto out;
3416     }
3417 
3418     if (core_slot->cpu) {
3419         error_setg(&local_err, "core %d already populated", cc->core_id);
3420         goto out;
3421     }
3422 
3423     numa_cpu_pre_plug(core_slot, dev, &local_err);
3424 
3425 out:
3426     error_propagate(errp, local_err);
3427 }
3428 
3429 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3430                                       DeviceState *dev, Error **errp)
3431 {
3432     MachineState *ms = MACHINE(hotplug_dev);
3433     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3434 
3435     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3436         int node;
3437 
3438         if (!smc->dr_lmb_enabled) {
3439             error_setg(errp, "Memory hotplug not supported for this machine");
3440             return;
3441         }
3442         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3443         if (*errp) {
3444             return;
3445         }
3446         if (node < 0 || node >= MAX_NODES) {
3447             error_setg(errp, "Invaild node %d", node);
3448             return;
3449         }
3450 
3451         /*
3452          * Currently PowerPC kernel doesn't allow hot-adding memory to
3453          * memory-less node, but instead will silently add the memory
3454          * to the first node that has some memory. This causes two
3455          * unexpected behaviours for the user.
3456          *
3457          * - Memory gets hotplugged to a different node than what the user
3458          *   specified.
3459          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3460          *   to memory-less node, a reboot will set things accordingly
3461          *   and the previously hotplugged memory now ends in the right node.
3462          *   This appears as if some memory moved from one node to another.
3463          *
3464          * So until kernel starts supporting memory hotplug to memory-less
3465          * nodes, just prevent such attempts upfront in QEMU.
3466          */
3467         if (nb_numa_nodes && !numa_info[node].node_mem) {
3468             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3469                        node);
3470             return;
3471         }
3472 
3473         spapr_memory_plug(hotplug_dev, dev, node, errp);
3474     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3475         spapr_core_plug(hotplug_dev, dev, errp);
3476     }
3477 }
3478 
3479 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3480                                                 DeviceState *dev, Error **errp)
3481 {
3482     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3483     MachineClass *mc = MACHINE_GET_CLASS(sms);
3484 
3485     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3486         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3487             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3488         } else {
3489             /* NOTE: this means there is a window after guest reset, prior to
3490              * CAS negotiation, where unplug requests will fail due to the
3491              * capability not being detected yet. This is a bit different than
3492              * the case with PCI unplug, where the events will be queued and
3493              * eventually handled by the guest after boot
3494              */
3495             error_setg(errp, "Memory hot unplug not supported for this guest");
3496         }
3497     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3498         if (!mc->has_hotpluggable_cpus) {
3499             error_setg(errp, "CPU hot unplug not supported on this machine");
3500             return;
3501         }
3502         spapr_core_unplug_request(hotplug_dev, dev, errp);
3503     }
3504 }
3505 
3506 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3507                                           DeviceState *dev, Error **errp)
3508 {
3509     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3510         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3511     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3512         spapr_core_pre_plug(hotplug_dev, dev, errp);
3513     }
3514 }
3515 
3516 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3517                                                  DeviceState *dev)
3518 {
3519     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3520         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3521         return HOTPLUG_HANDLER(machine);
3522     }
3523     return NULL;
3524 }
3525 
3526 static CpuInstanceProperties
3527 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3528 {
3529     CPUArchId *core_slot;
3530     MachineClass *mc = MACHINE_GET_CLASS(machine);
3531 
3532     /* make sure possible_cpu are intialized */
3533     mc->possible_cpu_arch_ids(machine);
3534     /* get CPU core slot containing thread that matches cpu_index */
3535     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3536     assert(core_slot);
3537     return core_slot->props;
3538 }
3539 
3540 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3541 {
3542     return idx / smp_cores % nb_numa_nodes;
3543 }
3544 
3545 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3546 {
3547     int i;
3548     int spapr_max_cores = max_cpus / smp_threads;
3549     MachineClass *mc = MACHINE_GET_CLASS(machine);
3550 
3551     if (!mc->has_hotpluggable_cpus) {
3552         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3553     }
3554     if (machine->possible_cpus) {
3555         assert(machine->possible_cpus->len == spapr_max_cores);
3556         return machine->possible_cpus;
3557     }
3558 
3559     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3560                              sizeof(CPUArchId) * spapr_max_cores);
3561     machine->possible_cpus->len = spapr_max_cores;
3562     for (i = 0; i < machine->possible_cpus->len; i++) {
3563         int core_id = i * smp_threads;
3564 
3565         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3566         machine->possible_cpus->cpus[i].arch_id = core_id;
3567         machine->possible_cpus->cpus[i].props.has_core_id = true;
3568         machine->possible_cpus->cpus[i].props.core_id = core_id;
3569     }
3570     return machine->possible_cpus;
3571 }
3572 
3573 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3574                                 uint64_t *buid, hwaddr *pio,
3575                                 hwaddr *mmio32, hwaddr *mmio64,
3576                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3577 {
3578     /*
3579      * New-style PHB window placement.
3580      *
3581      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3582      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3583      * windows.
3584      *
3585      * Some guest kernels can't work with MMIO windows above 1<<46
3586      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3587      *
3588      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3589      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3590      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3591      * 1TiB 64-bit MMIO windows for each PHB.
3592      */
3593     const uint64_t base_buid = 0x800000020000000ULL;
3594 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3595                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3596     int i;
3597 
3598     /* Sanity check natural alignments */
3599     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3600     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3601     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3602     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3603     /* Sanity check bounds */
3604     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3605                       SPAPR_PCI_MEM32_WIN_SIZE);
3606     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3607                       SPAPR_PCI_MEM64_WIN_SIZE);
3608 
3609     if (index >= SPAPR_MAX_PHBS) {
3610         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3611                    SPAPR_MAX_PHBS - 1);
3612         return;
3613     }
3614 
3615     *buid = base_buid + index;
3616     for (i = 0; i < n_dma; ++i) {
3617         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3618     }
3619 
3620     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3621     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3622     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3623 }
3624 
3625 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3626 {
3627     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3628 
3629     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3630 }
3631 
3632 static void spapr_ics_resend(XICSFabric *dev)
3633 {
3634     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3635 
3636     ics_resend(spapr->ics);
3637 }
3638 
3639 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3640 {
3641     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3642 
3643     return cpu ? ICP(cpu->intc) : NULL;
3644 }
3645 
3646 #define ICS_IRQ_FREE(ics, srcno)   \
3647     (!((ics)->irqs[(srcno)].flags & (XICS_FLAGS_IRQ_MASK)))
3648 
3649 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
3650 {
3651     int first, i;
3652 
3653     for (first = 0; first < ics->nr_irqs; first += alignnum) {
3654         if (num > (ics->nr_irqs - first)) {
3655             return -1;
3656         }
3657         for (i = first; i < first + num; ++i) {
3658             if (!ICS_IRQ_FREE(ics, i)) {
3659                 break;
3660             }
3661         }
3662         if (i == (first + num)) {
3663             return first;
3664         }
3665     }
3666 
3667     return -1;
3668 }
3669 
3670 /*
3671  * Allocate the IRQ number and set the IRQ type, LSI or MSI
3672  */
3673 static void spapr_irq_set_lsi(sPAPRMachineState *spapr, int irq, bool lsi)
3674 {
3675     ics_set_irq_type(spapr->ics, irq - spapr->ics->offset, lsi);
3676 }
3677 
3678 int spapr_irq_alloc(sPAPRMachineState *spapr, int irq_hint, bool lsi,
3679                     Error **errp)
3680 {
3681     ICSState *ics = spapr->ics;
3682     int irq;
3683 
3684     if (!ics) {
3685         return -1;
3686     }
3687     if (irq_hint) {
3688         if (!ICS_IRQ_FREE(ics, irq_hint - ics->offset)) {
3689             error_setg(errp, "can't allocate IRQ %d: already in use", irq_hint);
3690             return -1;
3691         }
3692         irq = irq_hint;
3693     } else {
3694         irq = ics_find_free_block(ics, 1, 1);
3695         if (irq < 0) {
3696             error_setg(errp, "can't allocate IRQ: no IRQ left");
3697             return -1;
3698         }
3699         irq += ics->offset;
3700     }
3701 
3702     spapr_irq_set_lsi(spapr, irq, lsi);
3703     trace_spapr_irq_alloc(irq);
3704 
3705     return irq;
3706 }
3707 
3708 /*
3709  * Allocate block of consecutive IRQs, and return the number of the first IRQ in
3710  * the block. If align==true, aligns the first IRQ number to num.
3711  */
3712 int spapr_irq_alloc_block(sPAPRMachineState *spapr, int num, bool lsi,
3713                           bool align, Error **errp)
3714 {
3715     ICSState *ics = spapr->ics;
3716     int i, first = -1;
3717 
3718     if (!ics) {
3719         return -1;
3720     }
3721 
3722     /*
3723      * MSIMesage::data is used for storing VIRQ so
3724      * it has to be aligned to num to support multiple
3725      * MSI vectors. MSI-X is not affected by this.
3726      * The hint is used for the first IRQ, the rest should
3727      * be allocated continuously.
3728      */
3729     if (align) {
3730         assert((num == 1) || (num == 2) || (num == 4) ||
3731                (num == 8) || (num == 16) || (num == 32));
3732         first = ics_find_free_block(ics, num, num);
3733     } else {
3734         first = ics_find_free_block(ics, num, 1);
3735     }
3736     if (first < 0) {
3737         error_setg(errp, "can't find a free %d-IRQ block", num);
3738         return -1;
3739     }
3740 
3741     first += ics->offset;
3742     for (i = first; i < first + num; ++i) {
3743         spapr_irq_set_lsi(spapr, i, lsi);
3744     }
3745 
3746     trace_spapr_irq_alloc_block(first, num, lsi, align);
3747 
3748     return first;
3749 }
3750 
3751 void spapr_irq_free(sPAPRMachineState *spapr, int irq, int num)
3752 {
3753     ICSState *ics = spapr->ics;
3754     int srcno = irq - ics->offset;
3755     int i;
3756 
3757     if (ics_valid_irq(ics, irq)) {
3758         trace_spapr_irq_free(0, irq, num);
3759         for (i = srcno; i < srcno + num; ++i) {
3760             if (ICS_IRQ_FREE(ics, i)) {
3761                 trace_spapr_irq_free_warn(0, i + ics->offset);
3762             }
3763             memset(&ics->irqs[i], 0, sizeof(ICSIRQState));
3764         }
3765     }
3766 }
3767 
3768 qemu_irq spapr_qirq(sPAPRMachineState *spapr, int irq)
3769 {
3770     ICSState *ics = spapr->ics;
3771 
3772     if (ics_valid_irq(ics, irq)) {
3773         return ics->qirqs[irq - ics->offset];
3774     }
3775 
3776     return NULL;
3777 }
3778 
3779 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3780                                  Monitor *mon)
3781 {
3782     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3783     CPUState *cs;
3784 
3785     CPU_FOREACH(cs) {
3786         PowerPCCPU *cpu = POWERPC_CPU(cs);
3787 
3788         icp_pic_print_info(ICP(cpu->intc), mon);
3789     }
3790 
3791     ics_pic_print_info(spapr->ics, mon);
3792 }
3793 
3794 int spapr_vcpu_id(PowerPCCPU *cpu)
3795 {
3796     CPUState *cs = CPU(cpu);
3797 
3798     if (kvm_enabled()) {
3799         return kvm_arch_vcpu_id(cs);
3800     } else {
3801         return cs->cpu_index;
3802     }
3803 }
3804 
3805 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3806 {
3807     CPUState *cs;
3808 
3809     CPU_FOREACH(cs) {
3810         PowerPCCPU *cpu = POWERPC_CPU(cs);
3811 
3812         if (spapr_vcpu_id(cpu) == vcpu_id) {
3813             return cpu;
3814         }
3815     }
3816 
3817     return NULL;
3818 }
3819 
3820 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3821 {
3822     MachineClass *mc = MACHINE_CLASS(oc);
3823     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3824     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3825     NMIClass *nc = NMI_CLASS(oc);
3826     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3827     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3828     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3829     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3830 
3831     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3832 
3833     /*
3834      * We set up the default / latest behaviour here.  The class_init
3835      * functions for the specific versioned machine types can override
3836      * these details for backwards compatibility
3837      */
3838     mc->init = spapr_machine_init;
3839     mc->reset = spapr_machine_reset;
3840     mc->block_default_type = IF_SCSI;
3841     mc->max_cpus = 1024;
3842     mc->no_parallel = 1;
3843     mc->default_boot_order = "";
3844     mc->default_ram_size = 512 * M_BYTE;
3845     mc->kvm_type = spapr_kvm_type;
3846     mc->has_dynamic_sysbus = true;
3847     mc->pci_allow_0_address = true;
3848     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3849     hc->pre_plug = spapr_machine_device_pre_plug;
3850     hc->plug = spapr_machine_device_plug;
3851     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3852     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3853     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3854     hc->unplug_request = spapr_machine_device_unplug_request;
3855 
3856     smc->dr_lmb_enabled = true;
3857     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3858     mc->has_hotpluggable_cpus = true;
3859     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3860     fwc->get_dev_path = spapr_get_fw_dev_path;
3861     nc->nmi_monitor_handler = spapr_nmi;
3862     smc->phb_placement = spapr_phb_placement;
3863     vhc->hypercall = emulate_spapr_hypercall;
3864     vhc->hpt_mask = spapr_hpt_mask;
3865     vhc->map_hptes = spapr_map_hptes;
3866     vhc->unmap_hptes = spapr_unmap_hptes;
3867     vhc->store_hpte = spapr_store_hpte;
3868     vhc->get_patbe = spapr_get_patbe;
3869     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3870     xic->ics_get = spapr_ics_get;
3871     xic->ics_resend = spapr_ics_resend;
3872     xic->icp_get = spapr_icp_get;
3873     ispc->print_info = spapr_pic_print_info;
3874     /* Force NUMA node memory size to be a multiple of
3875      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3876      * in which LMBs are represented and hot-added
3877      */
3878     mc->numa_mem_align_shift = 28;
3879 
3880     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
3881     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
3882     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
3883     spapr_caps_add_properties(smc, &error_abort);
3884 }
3885 
3886 static const TypeInfo spapr_machine_info = {
3887     .name          = TYPE_SPAPR_MACHINE,
3888     .parent        = TYPE_MACHINE,
3889     .abstract      = true,
3890     .instance_size = sizeof(sPAPRMachineState),
3891     .instance_init = spapr_instance_init,
3892     .instance_finalize = spapr_machine_finalizefn,
3893     .class_size    = sizeof(sPAPRMachineClass),
3894     .class_init    = spapr_machine_class_init,
3895     .interfaces = (InterfaceInfo[]) {
3896         { TYPE_FW_PATH_PROVIDER },
3897         { TYPE_NMI },
3898         { TYPE_HOTPLUG_HANDLER },
3899         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3900         { TYPE_XICS_FABRIC },
3901         { TYPE_INTERRUPT_STATS_PROVIDER },
3902         { }
3903     },
3904 };
3905 
3906 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3907     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3908                                                     void *data)      \
3909     {                                                                \
3910         MachineClass *mc = MACHINE_CLASS(oc);                        \
3911         spapr_machine_##suffix##_class_options(mc);                  \
3912         if (latest) {                                                \
3913             mc->alias = "pseries";                                   \
3914             mc->is_default = 1;                                      \
3915         }                                                            \
3916     }                                                                \
3917     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3918     {                                                                \
3919         MachineState *machine = MACHINE(obj);                        \
3920         spapr_machine_##suffix##_instance_options(machine);          \
3921     }                                                                \
3922     static const TypeInfo spapr_machine_##suffix##_info = {          \
3923         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3924         .parent = TYPE_SPAPR_MACHINE,                                \
3925         .class_init = spapr_machine_##suffix##_class_init,           \
3926         .instance_init = spapr_machine_##suffix##_instance_init,     \
3927     };                                                               \
3928     static void spapr_machine_register_##suffix(void)                \
3929     {                                                                \
3930         type_register(&spapr_machine_##suffix##_info);               \
3931     }                                                                \
3932     type_init(spapr_machine_register_##suffix)
3933 
3934 /*
3935  * pseries-2.12
3936  */
3937 static void spapr_machine_2_12_instance_options(MachineState *machine)
3938 {
3939 }
3940 
3941 static void spapr_machine_2_12_class_options(MachineClass *mc)
3942 {
3943     /* Defaults for the latest behaviour inherited from the base class */
3944 }
3945 
3946 DEFINE_SPAPR_MACHINE(2_12, "2.12", true);
3947 
3948 /*
3949  * pseries-2.11
3950  */
3951 #define SPAPR_COMPAT_2_11                                              \
3952     HW_COMPAT_2_11
3953 
3954 static void spapr_machine_2_11_instance_options(MachineState *machine)
3955 {
3956     spapr_machine_2_12_instance_options(machine);
3957 }
3958 
3959 static void spapr_machine_2_11_class_options(MachineClass *mc)
3960 {
3961     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3962 
3963     spapr_machine_2_12_class_options(mc);
3964     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
3965     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
3966 }
3967 
3968 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
3969 
3970 /*
3971  * pseries-2.10
3972  */
3973 #define SPAPR_COMPAT_2_10                                              \
3974     HW_COMPAT_2_10
3975 
3976 static void spapr_machine_2_10_instance_options(MachineState *machine)
3977 {
3978     spapr_machine_2_11_instance_options(machine);
3979 }
3980 
3981 static void spapr_machine_2_10_class_options(MachineClass *mc)
3982 {
3983     spapr_machine_2_11_class_options(mc);
3984     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
3985 }
3986 
3987 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
3988 
3989 /*
3990  * pseries-2.9
3991  */
3992 #define SPAPR_COMPAT_2_9                                               \
3993     HW_COMPAT_2_9                                                      \
3994     {                                                                  \
3995         .driver = TYPE_POWERPC_CPU,                                    \
3996         .property = "pre-2.10-migration",                              \
3997         .value    = "on",                                              \
3998     },                                                                 \
3999 
4000 static void spapr_machine_2_9_instance_options(MachineState *machine)
4001 {
4002     spapr_machine_2_10_instance_options(machine);
4003 }
4004 
4005 static void spapr_machine_2_9_class_options(MachineClass *mc)
4006 {
4007     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4008 
4009     spapr_machine_2_10_class_options(mc);
4010     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4011     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4012     smc->pre_2_10_has_unused_icps = true;
4013     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4014 }
4015 
4016 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4017 
4018 /*
4019  * pseries-2.8
4020  */
4021 #define SPAPR_COMPAT_2_8                                        \
4022     HW_COMPAT_2_8                                               \
4023     {                                                           \
4024         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4025         .property = "pcie-extended-configuration-space",        \
4026         .value    = "off",                                      \
4027     },
4028 
4029 static void spapr_machine_2_8_instance_options(MachineState *machine)
4030 {
4031     spapr_machine_2_9_instance_options(machine);
4032 }
4033 
4034 static void spapr_machine_2_8_class_options(MachineClass *mc)
4035 {
4036     spapr_machine_2_9_class_options(mc);
4037     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4038     mc->numa_mem_align_shift = 23;
4039 }
4040 
4041 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4042 
4043 /*
4044  * pseries-2.7
4045  */
4046 #define SPAPR_COMPAT_2_7                            \
4047     HW_COMPAT_2_7                                   \
4048     {                                               \
4049         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4050         .property = "mem_win_size",                 \
4051         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4052     },                                              \
4053     {                                               \
4054         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4055         .property = "mem64_win_size",               \
4056         .value    = "0",                            \
4057     },                                              \
4058     {                                               \
4059         .driver = TYPE_POWERPC_CPU,                 \
4060         .property = "pre-2.8-migration",            \
4061         .value    = "on",                           \
4062     },                                              \
4063     {                                               \
4064         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4065         .property = "pre-2.8-migration",            \
4066         .value    = "on",                           \
4067     },
4068 
4069 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4070                               uint64_t *buid, hwaddr *pio,
4071                               hwaddr *mmio32, hwaddr *mmio64,
4072                               unsigned n_dma, uint32_t *liobns, Error **errp)
4073 {
4074     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4075     const uint64_t base_buid = 0x800000020000000ULL;
4076     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4077     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4078     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4079     const uint32_t max_index = 255;
4080     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4081 
4082     uint64_t ram_top = MACHINE(spapr)->ram_size;
4083     hwaddr phb0_base, phb_base;
4084     int i;
4085 
4086     /* Do we have hotpluggable memory? */
4087     if (MACHINE(spapr)->maxram_size > ram_top) {
4088         /* Can't just use maxram_size, because there may be an
4089          * alignment gap between normal and hotpluggable memory
4090          * regions */
4091         ram_top = spapr->hotplug_memory.base +
4092             memory_region_size(&spapr->hotplug_memory.mr);
4093     }
4094 
4095     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4096 
4097     if (index > max_index) {
4098         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4099                    max_index);
4100         return;
4101     }
4102 
4103     *buid = base_buid + index;
4104     for (i = 0; i < n_dma; ++i) {
4105         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4106     }
4107 
4108     phb_base = phb0_base + index * phb_spacing;
4109     *pio = phb_base + pio_offset;
4110     *mmio32 = phb_base + mmio_offset;
4111     /*
4112      * We don't set the 64-bit MMIO window, relying on the PHB's
4113      * fallback behaviour of automatically splitting a large "32-bit"
4114      * window into contiguous 32-bit and 64-bit windows
4115      */
4116 }
4117 
4118 static void spapr_machine_2_7_instance_options(MachineState *machine)
4119 {
4120     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4121 
4122     spapr_machine_2_8_instance_options(machine);
4123     spapr->use_hotplug_event_source = false;
4124 }
4125 
4126 static void spapr_machine_2_7_class_options(MachineClass *mc)
4127 {
4128     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4129 
4130     spapr_machine_2_8_class_options(mc);
4131     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4132     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4133     smc->phb_placement = phb_placement_2_7;
4134 }
4135 
4136 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4137 
4138 /*
4139  * pseries-2.6
4140  */
4141 #define SPAPR_COMPAT_2_6 \
4142     HW_COMPAT_2_6 \
4143     { \
4144         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4145         .property = "ddw",\
4146         .value    = stringify(off),\
4147     },
4148 
4149 static void spapr_machine_2_6_instance_options(MachineState *machine)
4150 {
4151     spapr_machine_2_7_instance_options(machine);
4152 }
4153 
4154 static void spapr_machine_2_6_class_options(MachineClass *mc)
4155 {
4156     spapr_machine_2_7_class_options(mc);
4157     mc->has_hotpluggable_cpus = false;
4158     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4159 }
4160 
4161 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4162 
4163 /*
4164  * pseries-2.5
4165  */
4166 #define SPAPR_COMPAT_2_5 \
4167     HW_COMPAT_2_5 \
4168     { \
4169         .driver   = "spapr-vlan", \
4170         .property = "use-rx-buffer-pools", \
4171         .value    = "off", \
4172     },
4173 
4174 static void spapr_machine_2_5_instance_options(MachineState *machine)
4175 {
4176     spapr_machine_2_6_instance_options(machine);
4177 }
4178 
4179 static void spapr_machine_2_5_class_options(MachineClass *mc)
4180 {
4181     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4182 
4183     spapr_machine_2_6_class_options(mc);
4184     smc->use_ohci_by_default = true;
4185     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4186 }
4187 
4188 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4189 
4190 /*
4191  * pseries-2.4
4192  */
4193 #define SPAPR_COMPAT_2_4 \
4194         HW_COMPAT_2_4
4195 
4196 static void spapr_machine_2_4_instance_options(MachineState *machine)
4197 {
4198     spapr_machine_2_5_instance_options(machine);
4199 }
4200 
4201 static void spapr_machine_2_4_class_options(MachineClass *mc)
4202 {
4203     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4204 
4205     spapr_machine_2_5_class_options(mc);
4206     smc->dr_lmb_enabled = false;
4207     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4208 }
4209 
4210 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4211 
4212 /*
4213  * pseries-2.3
4214  */
4215 #define SPAPR_COMPAT_2_3 \
4216         HW_COMPAT_2_3 \
4217         {\
4218             .driver   = "spapr-pci-host-bridge",\
4219             .property = "dynamic-reconfiguration",\
4220             .value    = "off",\
4221         },
4222 
4223 static void spapr_machine_2_3_instance_options(MachineState *machine)
4224 {
4225     spapr_machine_2_4_instance_options(machine);
4226 }
4227 
4228 static void spapr_machine_2_3_class_options(MachineClass *mc)
4229 {
4230     spapr_machine_2_4_class_options(mc);
4231     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4232 }
4233 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4234 
4235 /*
4236  * pseries-2.2
4237  */
4238 
4239 #define SPAPR_COMPAT_2_2 \
4240         HW_COMPAT_2_2 \
4241         {\
4242             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4243             .property = "mem_win_size",\
4244             .value    = "0x20000000",\
4245         },
4246 
4247 static void spapr_machine_2_2_instance_options(MachineState *machine)
4248 {
4249     spapr_machine_2_3_instance_options(machine);
4250     machine->suppress_vmdesc = true;
4251 }
4252 
4253 static void spapr_machine_2_2_class_options(MachineClass *mc)
4254 {
4255     spapr_machine_2_3_class_options(mc);
4256     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4257 }
4258 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4259 
4260 /*
4261  * pseries-2.1
4262  */
4263 #define SPAPR_COMPAT_2_1 \
4264         HW_COMPAT_2_1
4265 
4266 static void spapr_machine_2_1_instance_options(MachineState *machine)
4267 {
4268     spapr_machine_2_2_instance_options(machine);
4269 }
4270 
4271 static void spapr_machine_2_1_class_options(MachineClass *mc)
4272 {
4273     spapr_machine_2_2_class_options(mc);
4274     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4275 }
4276 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4277 
4278 static void spapr_machine_register_types(void)
4279 {
4280     type_register_static(&spapr_machine_info);
4281 }
4282 
4283 type_init(spapr_machine_register_types)
4284