xref: /openbmc/qemu/hw/ppc/spapr.c (revision 89854803)
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 
78 #include <libfdt.h>
79 
80 /* SLOF memory layout:
81  *
82  * SLOF raw image loaded at 0, copies its romfs right below the flat
83  * device-tree, then position SLOF itself 31M below that
84  *
85  * So we set FW_OVERHEAD to 40MB which should account for all of that
86  * and more
87  *
88  * We load our kernel at 4M, leaving space for SLOF initial image
89  */
90 #define FDT_MAX_SIZE            0x100000
91 #define RTAS_MAX_SIZE           0x10000
92 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
93 #define FW_MAX_SIZE             0x400000
94 #define FW_FILE_NAME            "slof.bin"
95 #define FW_OVERHEAD             0x2800000
96 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
97 
98 #define MIN_RMA_SLOF            128UL
99 
100 #define PHANDLE_XICP            0x00001111
101 
102 /* These two functions implement the VCPU id numbering: one to compute them
103  * all and one to identify thread 0 of a VCORE. Any change to the first one
104  * is likely to have an impact on the second one, so let's keep them close.
105  */
106 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
107 {
108     assert(spapr->vsmt);
109     return
110         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
111 }
112 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
113                                       PowerPCCPU *cpu)
114 {
115     assert(spapr->vsmt);
116     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
117 }
118 
119 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
120                                   const char *type_ics,
121                                   int nr_irqs, Error **errp)
122 {
123     Error *local_err = NULL;
124     Object *obj;
125 
126     obj = object_new(type_ics);
127     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
128     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
129                                    &error_abort);
130     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
131     if (local_err) {
132         goto error;
133     }
134     object_property_set_bool(obj, true, "realized", &local_err);
135     if (local_err) {
136         goto error;
137     }
138 
139     return ICS_SIMPLE(obj);
140 
141 error:
142     error_propagate(errp, local_err);
143     return NULL;
144 }
145 
146 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
147 {
148     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
149      * and newer QEMUs don't even have them. In both cases, we don't want
150      * to send anything on the wire.
151      */
152     return false;
153 }
154 
155 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
156     .name = "icp/server",
157     .version_id = 1,
158     .minimum_version_id = 1,
159     .needed = pre_2_10_vmstate_dummy_icp_needed,
160     .fields = (VMStateField[]) {
161         VMSTATE_UNUSED(4), /* uint32_t xirr */
162         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
163         VMSTATE_UNUSED(1), /* uint8_t mfrr */
164         VMSTATE_END_OF_LIST()
165     },
166 };
167 
168 static void pre_2_10_vmstate_register_dummy_icp(int i)
169 {
170     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
171                      (void *)(uintptr_t) i);
172 }
173 
174 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
175 {
176     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
177                        (void *)(uintptr_t) i);
178 }
179 
180 static int xics_max_server_number(sPAPRMachineState *spapr)
181 {
182     assert(spapr->vsmt);
183     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
184 }
185 
186 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
187 {
188     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
189 
190     if (kvm_enabled()) {
191         if (machine_kernel_irqchip_allowed(machine) &&
192             !xics_kvm_init(spapr, errp)) {
193             spapr->icp_type = TYPE_KVM_ICP;
194             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
195         }
196         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
197             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
198             return;
199         }
200     }
201 
202     if (!spapr->ics) {
203         xics_spapr_init(spapr);
204         spapr->icp_type = TYPE_ICP;
205         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
206         if (!spapr->ics) {
207             return;
208         }
209     }
210 }
211 
212 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
213                                   int smt_threads)
214 {
215     int i, ret = 0;
216     uint32_t servers_prop[smt_threads];
217     uint32_t gservers_prop[smt_threads * 2];
218     int index = spapr_get_vcpu_id(cpu);
219 
220     if (cpu->compat_pvr) {
221         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
222         if (ret < 0) {
223             return ret;
224         }
225     }
226 
227     /* Build interrupt servers and gservers properties */
228     for (i = 0; i < smt_threads; i++) {
229         servers_prop[i] = cpu_to_be32(index + i);
230         /* Hack, direct the group queues back to cpu 0 */
231         gservers_prop[i*2] = cpu_to_be32(index + i);
232         gservers_prop[i*2 + 1] = 0;
233     }
234     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
235                       servers_prop, sizeof(servers_prop));
236     if (ret < 0) {
237         return ret;
238     }
239     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
240                       gservers_prop, sizeof(gservers_prop));
241 
242     return ret;
243 }
244 
245 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
246 {
247     int index = spapr_get_vcpu_id(cpu);
248     uint32_t associativity[] = {cpu_to_be32(0x5),
249                                 cpu_to_be32(0x0),
250                                 cpu_to_be32(0x0),
251                                 cpu_to_be32(0x0),
252                                 cpu_to_be32(cpu->node_id),
253                                 cpu_to_be32(index)};
254 
255     /* Advertise NUMA via ibm,associativity */
256     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
257                           sizeof(associativity));
258 }
259 
260 /* Populate the "ibm,pa-features" property */
261 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
262                                        PowerPCCPU *cpu,
263                                        void *fdt, int offset,
264                                        bool legacy_guest)
265 {
266     uint8_t pa_features_206[] = { 6, 0,
267         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
268     uint8_t pa_features_207[] = { 24, 0,
269         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
270         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
271         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
272         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
273     uint8_t pa_features_300[] = { 66, 0,
274         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
275         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
276         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
277         /* 6: DS207 */
278         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
279         /* 16: Vector */
280         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
281         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
282         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
283         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
284         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
285         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
286         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
287         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
288         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
289         /* 42: PM, 44: PC RA, 46: SC vec'd */
290         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
291         /* 48: SIMD, 50: QP BFP, 52: String */
292         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
293         /* 54: DecFP, 56: DecI, 58: SHA */
294         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
295         /* 60: NM atomic, 62: RNG */
296         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
297     };
298     uint8_t *pa_features = NULL;
299     size_t pa_size;
300 
301     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
302         pa_features = pa_features_206;
303         pa_size = sizeof(pa_features_206);
304     }
305     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
306         pa_features = pa_features_207;
307         pa_size = sizeof(pa_features_207);
308     }
309     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
310         pa_features = pa_features_300;
311         pa_size = sizeof(pa_features_300);
312     }
313     if (!pa_features) {
314         return;
315     }
316 
317     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
318         /*
319          * Note: we keep CI large pages off by default because a 64K capable
320          * guest provisioned with large pages might otherwise try to map a qemu
321          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
322          * even if that qemu runs on a 4k host.
323          * We dd this bit back here if we are confident this is not an issue
324          */
325         pa_features[3] |= 0x20;
326     }
327     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
328         pa_features[24] |= 0x80;    /* Transactional memory support */
329     }
330     if (legacy_guest && pa_size > 40) {
331         /* Workaround for broken kernels that attempt (guest) radix
332          * mode when they can't handle it, if they see the radix bit set
333          * in pa-features. So hide it from them. */
334         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
335     }
336 
337     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
338 }
339 
340 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
341 {
342     int ret = 0, offset, cpus_offset;
343     CPUState *cs;
344     char cpu_model[32];
345     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
346 
347     CPU_FOREACH(cs) {
348         PowerPCCPU *cpu = POWERPC_CPU(cs);
349         DeviceClass *dc = DEVICE_GET_CLASS(cs);
350         int index = spapr_get_vcpu_id(cpu);
351         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
352 
353         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
354             continue;
355         }
356 
357         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
358 
359         cpus_offset = fdt_path_offset(fdt, "/cpus");
360         if (cpus_offset < 0) {
361             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
362             if (cpus_offset < 0) {
363                 return cpus_offset;
364             }
365         }
366         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
367         if (offset < 0) {
368             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
369             if (offset < 0) {
370                 return offset;
371             }
372         }
373 
374         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
375                           pft_size_prop, sizeof(pft_size_prop));
376         if (ret < 0) {
377             return ret;
378         }
379 
380         if (nb_numa_nodes > 1) {
381             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
382             if (ret < 0) {
383                 return ret;
384             }
385         }
386 
387         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
388         if (ret < 0) {
389             return ret;
390         }
391 
392         spapr_populate_pa_features(spapr, cpu, fdt, offset,
393                                    spapr->cas_legacy_guest_workaround);
394     }
395     return ret;
396 }
397 
398 static hwaddr spapr_node0_size(MachineState *machine)
399 {
400     if (nb_numa_nodes) {
401         int i;
402         for (i = 0; i < nb_numa_nodes; ++i) {
403             if (numa_info[i].node_mem) {
404                 return MIN(pow2floor(numa_info[i].node_mem),
405                            machine->ram_size);
406             }
407         }
408     }
409     return machine->ram_size;
410 }
411 
412 static void add_str(GString *s, const gchar *s1)
413 {
414     g_string_append_len(s, s1, strlen(s1) + 1);
415 }
416 
417 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
418                                        hwaddr size)
419 {
420     uint32_t associativity[] = {
421         cpu_to_be32(0x4), /* length */
422         cpu_to_be32(0x0), cpu_to_be32(0x0),
423         cpu_to_be32(0x0), cpu_to_be32(nodeid)
424     };
425     char mem_name[32];
426     uint64_t mem_reg_property[2];
427     int off;
428 
429     mem_reg_property[0] = cpu_to_be64(start);
430     mem_reg_property[1] = cpu_to_be64(size);
431 
432     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
433     off = fdt_add_subnode(fdt, 0, mem_name);
434     _FDT(off);
435     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
436     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
437                       sizeof(mem_reg_property))));
438     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
439                       sizeof(associativity))));
440     return off;
441 }
442 
443 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
444 {
445     MachineState *machine = MACHINE(spapr);
446     hwaddr mem_start, node_size;
447     int i, nb_nodes = nb_numa_nodes;
448     NodeInfo *nodes = numa_info;
449     NodeInfo ramnode;
450 
451     /* No NUMA nodes, assume there is just one node with whole RAM */
452     if (!nb_numa_nodes) {
453         nb_nodes = 1;
454         ramnode.node_mem = machine->ram_size;
455         nodes = &ramnode;
456     }
457 
458     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
459         if (!nodes[i].node_mem) {
460             continue;
461         }
462         if (mem_start >= machine->ram_size) {
463             node_size = 0;
464         } else {
465             node_size = nodes[i].node_mem;
466             if (node_size > machine->ram_size - mem_start) {
467                 node_size = machine->ram_size - mem_start;
468             }
469         }
470         if (!mem_start) {
471             /* spapr_machine_init() checks for rma_size <= node0_size
472              * already */
473             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
474             mem_start += spapr->rma_size;
475             node_size -= spapr->rma_size;
476         }
477         for ( ; node_size; ) {
478             hwaddr sizetmp = pow2floor(node_size);
479 
480             /* mem_start != 0 here */
481             if (ctzl(mem_start) < ctzl(sizetmp)) {
482                 sizetmp = 1ULL << ctzl(mem_start);
483             }
484 
485             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
486             node_size -= sizetmp;
487             mem_start += sizetmp;
488         }
489     }
490 
491     return 0;
492 }
493 
494 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
495                                   sPAPRMachineState *spapr)
496 {
497     PowerPCCPU *cpu = POWERPC_CPU(cs);
498     CPUPPCState *env = &cpu->env;
499     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
500     int index = spapr_get_vcpu_id(cpu);
501     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
502                        0xffffffff, 0xffffffff};
503     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
504         : SPAPR_TIMEBASE_FREQ;
505     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
506     uint32_t page_sizes_prop[64];
507     size_t page_sizes_prop_size;
508     uint32_t vcpus_per_socket = smp_threads * smp_cores;
509     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
510     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
511     sPAPRDRConnector *drc;
512     int drc_index;
513     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
514     int i;
515 
516     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
517     if (drc) {
518         drc_index = spapr_drc_index(drc);
519         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
520     }
521 
522     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
523     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
524 
525     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
526     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
527                            env->dcache_line_size)));
528     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
529                            env->dcache_line_size)));
530     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
531                            env->icache_line_size)));
532     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
533                            env->icache_line_size)));
534 
535     if (pcc->l1_dcache_size) {
536         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
537                                pcc->l1_dcache_size)));
538     } else {
539         warn_report("Unknown L1 dcache size for cpu");
540     }
541     if (pcc->l1_icache_size) {
542         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
543                                pcc->l1_icache_size)));
544     } else {
545         warn_report("Unknown L1 icache size for cpu");
546     }
547 
548     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
549     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
550     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
551     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
552     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
553     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
554 
555     if (env->spr_cb[SPR_PURR].oea_read) {
556         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
557     }
558 
559     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
560         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
561                           segs, sizeof(segs))));
562     }
563 
564     /* Advertise VSX (vector extensions) if available
565      *   1               == VMX / Altivec available
566      *   2               == VSX available
567      *
568      * Only CPUs for which we create core types in spapr_cpu_core.c
569      * are possible, and all of those have VMX */
570     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
571         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
572     } else {
573         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
574     }
575 
576     /* Advertise DFP (Decimal Floating Point) if available
577      *   0 / no property == no DFP
578      *   1               == DFP available */
579     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
580         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
581     }
582 
583     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
584                                                       sizeof(page_sizes_prop));
585     if (page_sizes_prop_size) {
586         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
587                           page_sizes_prop, page_sizes_prop_size)));
588     }
589 
590     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
591 
592     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
593                            cs->cpu_index / vcpus_per_socket)));
594 
595     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
596                       pft_size_prop, sizeof(pft_size_prop))));
597 
598     if (nb_numa_nodes > 1) {
599         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
600     }
601 
602     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
603 
604     if (pcc->radix_page_info) {
605         for (i = 0; i < pcc->radix_page_info->count; i++) {
606             radix_AP_encodings[i] =
607                 cpu_to_be32(pcc->radix_page_info->entries[i]);
608         }
609         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
610                           radix_AP_encodings,
611                           pcc->radix_page_info->count *
612                           sizeof(radix_AP_encodings[0]))));
613     }
614 }
615 
616 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
617 {
618     CPUState *cs;
619     int cpus_offset;
620     char *nodename;
621 
622     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
623     _FDT(cpus_offset);
624     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
625     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
626 
627     /*
628      * We walk the CPUs in reverse order to ensure that CPU DT nodes
629      * created by fdt_add_subnode() end up in the right order in FDT
630      * for the guest kernel the enumerate the CPUs correctly.
631      */
632     CPU_FOREACH_REVERSE(cs) {
633         PowerPCCPU *cpu = POWERPC_CPU(cs);
634         int index = spapr_get_vcpu_id(cpu);
635         DeviceClass *dc = DEVICE_GET_CLASS(cs);
636         int offset;
637 
638         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
639             continue;
640         }
641 
642         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
643         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
644         g_free(nodename);
645         _FDT(offset);
646         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
647     }
648 
649 }
650 
651 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
652 {
653     MemoryDeviceInfoList *info;
654 
655     for (info = list; info; info = info->next) {
656         MemoryDeviceInfo *value = info->value;
657 
658         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
659             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
660 
661             if (pcdimm_info->addr >= addr &&
662                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
663                 return pcdimm_info->node;
664             }
665         }
666     }
667 
668     return -1;
669 }
670 
671 struct sPAPRDrconfCellV2 {
672      uint32_t seq_lmbs;
673      uint64_t base_addr;
674      uint32_t drc_index;
675      uint32_t aa_index;
676      uint32_t flags;
677 } QEMU_PACKED;
678 
679 typedef struct DrconfCellQueue {
680     struct sPAPRDrconfCellV2 cell;
681     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
682 } DrconfCellQueue;
683 
684 static DrconfCellQueue *
685 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
686                       uint32_t drc_index, uint32_t aa_index,
687                       uint32_t flags)
688 {
689     DrconfCellQueue *elem;
690 
691     elem = g_malloc0(sizeof(*elem));
692     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
693     elem->cell.base_addr = cpu_to_be64(base_addr);
694     elem->cell.drc_index = cpu_to_be32(drc_index);
695     elem->cell.aa_index = cpu_to_be32(aa_index);
696     elem->cell.flags = cpu_to_be32(flags);
697 
698     return elem;
699 }
700 
701 /* ibm,dynamic-memory-v2 */
702 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
703                                    int offset, MemoryDeviceInfoList *dimms)
704 {
705     uint8_t *int_buf, *cur_index, buf_len;
706     int ret;
707     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
708     uint64_t addr, cur_addr, size;
709     uint32_t nr_boot_lmbs = (spapr->hotplug_memory.base / lmb_size);
710     uint64_t mem_end = spapr->hotplug_memory.base +
711                        memory_region_size(&spapr->hotplug_memory.mr);
712     uint32_t node, nr_entries = 0;
713     sPAPRDRConnector *drc;
714     DrconfCellQueue *elem, *next;
715     MemoryDeviceInfoList *info;
716     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
717         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
718 
719     /* Entry to cover RAM and the gap area */
720     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
721                                  SPAPR_LMB_FLAGS_RESERVED |
722                                  SPAPR_LMB_FLAGS_DRC_INVALID);
723     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
724     nr_entries++;
725 
726     cur_addr = spapr->hotplug_memory.base;
727     for (info = dimms; info; info = info->next) {
728         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
729 
730         addr = di->addr;
731         size = di->size;
732         node = di->node;
733 
734         /* Entry for hot-pluggable area */
735         if (cur_addr < addr) {
736             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
737             g_assert(drc);
738             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
739                                          cur_addr, spapr_drc_index(drc), -1, 0);
740             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
741             nr_entries++;
742         }
743 
744         /* Entry for DIMM */
745         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
746         g_assert(drc);
747         elem = spapr_get_drconf_cell(size / lmb_size, addr,
748                                      spapr_drc_index(drc), node,
749                                      SPAPR_LMB_FLAGS_ASSIGNED);
750         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
751         nr_entries++;
752         cur_addr = addr + size;
753     }
754 
755     /* Entry for remaining hotpluggable area */
756     if (cur_addr < mem_end) {
757         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
758         g_assert(drc);
759         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
760                                      cur_addr, spapr_drc_index(drc), -1, 0);
761         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
762         nr_entries++;
763     }
764 
765     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
766     int_buf = cur_index = g_malloc0(buf_len);
767     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
768     cur_index += sizeof(nr_entries);
769 
770     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
771         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
772         cur_index += sizeof(elem->cell);
773         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
774         g_free(elem);
775     }
776 
777     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
778     g_free(int_buf);
779     if (ret < 0) {
780         return -1;
781     }
782     return 0;
783 }
784 
785 /* ibm,dynamic-memory */
786 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
787                                    int offset, MemoryDeviceInfoList *dimms)
788 {
789     int i, ret;
790     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
791     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
792     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
793                        memory_region_size(&spapr->hotplug_memory.mr)) /
794                        lmb_size;
795     uint32_t *int_buf, *cur_index, buf_len;
796 
797     /*
798      * Allocate enough buffer size to fit in ibm,dynamic-memory
799      */
800     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
801     cur_index = int_buf = g_malloc0(buf_len);
802     int_buf[0] = cpu_to_be32(nr_lmbs);
803     cur_index++;
804     for (i = 0; i < nr_lmbs; i++) {
805         uint64_t addr = i * lmb_size;
806         uint32_t *dynamic_memory = cur_index;
807 
808         if (i >= hotplug_lmb_start) {
809             sPAPRDRConnector *drc;
810 
811             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
812             g_assert(drc);
813 
814             dynamic_memory[0] = cpu_to_be32(addr >> 32);
815             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
816             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
817             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
818             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
819             if (memory_region_present(get_system_memory(), addr)) {
820                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
821             } else {
822                 dynamic_memory[5] = cpu_to_be32(0);
823             }
824         } else {
825             /*
826              * LMB information for RMA, boot time RAM and gap b/n RAM and
827              * hotplug memory region -- all these are marked as reserved
828              * and as having no valid DRC.
829              */
830             dynamic_memory[0] = cpu_to_be32(addr >> 32);
831             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
832             dynamic_memory[2] = cpu_to_be32(0);
833             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
834             dynamic_memory[4] = cpu_to_be32(-1);
835             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
836                                             SPAPR_LMB_FLAGS_DRC_INVALID);
837         }
838 
839         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
840     }
841     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
842     g_free(int_buf);
843     if (ret < 0) {
844         return -1;
845     }
846     return 0;
847 }
848 
849 /*
850  * Adds ibm,dynamic-reconfiguration-memory node.
851  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
852  * of this device tree node.
853  */
854 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
855 {
856     MachineState *machine = MACHINE(spapr);
857     int ret, i, offset;
858     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
859     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
860     uint32_t *int_buf, *cur_index, buf_len;
861     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
862     MemoryDeviceInfoList *dimms = NULL;
863 
864     /*
865      * Don't create the node if there is no hotpluggable memory
866      */
867     if (machine->ram_size == machine->maxram_size) {
868         return 0;
869     }
870 
871     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
872 
873     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
874                     sizeof(prop_lmb_size));
875     if (ret < 0) {
876         return ret;
877     }
878 
879     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
880     if (ret < 0) {
881         return ret;
882     }
883 
884     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
885     if (ret < 0) {
886         return ret;
887     }
888 
889     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
890     dimms = qmp_pc_dimm_device_list();
891     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
892         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
893     } else {
894         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
895     }
896     qapi_free_MemoryDeviceInfoList(dimms);
897 
898     if (ret < 0) {
899         return ret;
900     }
901 
902     /* ibm,associativity-lookup-arrays */
903     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
904     cur_index = int_buf = g_malloc0(buf_len);
905 
906     cur_index = int_buf;
907     int_buf[0] = cpu_to_be32(nr_nodes);
908     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
909     cur_index += 2;
910     for (i = 0; i < nr_nodes; i++) {
911         uint32_t associativity[] = {
912             cpu_to_be32(0x0),
913             cpu_to_be32(0x0),
914             cpu_to_be32(0x0),
915             cpu_to_be32(i)
916         };
917         memcpy(cur_index, associativity, sizeof(associativity));
918         cur_index += 4;
919     }
920     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
921             (cur_index - int_buf) * sizeof(uint32_t));
922     g_free(int_buf);
923 
924     return ret;
925 }
926 
927 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
928                                 sPAPROptionVector *ov5_updates)
929 {
930     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
931     int ret = 0, offset;
932 
933     /* Generate ibm,dynamic-reconfiguration-memory node if required */
934     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
935         g_assert(smc->dr_lmb_enabled);
936         ret = spapr_populate_drconf_memory(spapr, fdt);
937         if (ret) {
938             goto out;
939         }
940     }
941 
942     offset = fdt_path_offset(fdt, "/chosen");
943     if (offset < 0) {
944         offset = fdt_add_subnode(fdt, 0, "chosen");
945         if (offset < 0) {
946             return offset;
947         }
948     }
949     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
950                                  "ibm,architecture-vec-5");
951 
952 out:
953     return ret;
954 }
955 
956 static bool spapr_hotplugged_dev_before_cas(void)
957 {
958     Object *drc_container, *obj;
959     ObjectProperty *prop;
960     ObjectPropertyIterator iter;
961 
962     drc_container = container_get(object_get_root(), "/dr-connector");
963     object_property_iter_init(&iter, drc_container);
964     while ((prop = object_property_iter_next(&iter))) {
965         if (!strstart(prop->type, "link<", NULL)) {
966             continue;
967         }
968         obj = object_property_get_link(drc_container, prop->name, NULL);
969         if (spapr_drc_needed(obj)) {
970             return true;
971         }
972     }
973     return false;
974 }
975 
976 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
977                                  target_ulong addr, target_ulong size,
978                                  sPAPROptionVector *ov5_updates)
979 {
980     void *fdt, *fdt_skel;
981     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
982 
983     if (spapr_hotplugged_dev_before_cas()) {
984         return 1;
985     }
986 
987     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
988         error_report("SLOF provided an unexpected CAS buffer size "
989                      TARGET_FMT_lu " (min: %zu, max: %u)",
990                      size, sizeof(hdr), FW_MAX_SIZE);
991         exit(EXIT_FAILURE);
992     }
993 
994     size -= sizeof(hdr);
995 
996     /* Create skeleton */
997     fdt_skel = g_malloc0(size);
998     _FDT((fdt_create(fdt_skel, size)));
999     _FDT((fdt_finish_reservemap(fdt_skel)));
1000     _FDT((fdt_begin_node(fdt_skel, "")));
1001     _FDT((fdt_end_node(fdt_skel)));
1002     _FDT((fdt_finish(fdt_skel)));
1003     fdt = g_malloc0(size);
1004     _FDT((fdt_open_into(fdt_skel, fdt, size)));
1005     g_free(fdt_skel);
1006 
1007     /* Fixup cpu nodes */
1008     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
1009 
1010     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
1011         return -1;
1012     }
1013 
1014     /* Pack resulting tree */
1015     _FDT((fdt_pack(fdt)));
1016 
1017     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1018         trace_spapr_cas_failed(size);
1019         return -1;
1020     }
1021 
1022     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1023     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1024     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1025     g_free(fdt);
1026 
1027     return 0;
1028 }
1029 
1030 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1031 {
1032     int rtas;
1033     GString *hypertas = g_string_sized_new(256);
1034     GString *qemu_hypertas = g_string_sized_new(256);
1035     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1036     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
1037         memory_region_size(&spapr->hotplug_memory.mr);
1038     uint32_t lrdr_capacity[] = {
1039         cpu_to_be32(max_hotplug_addr >> 32),
1040         cpu_to_be32(max_hotplug_addr & 0xffffffff),
1041         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1042         cpu_to_be32(max_cpus / smp_threads),
1043     };
1044     uint32_t maxdomains[] = {
1045         cpu_to_be32(4),
1046         cpu_to_be32(0),
1047         cpu_to_be32(0),
1048         cpu_to_be32(0),
1049         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
1050     };
1051 
1052     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1053 
1054     /* hypertas */
1055     add_str(hypertas, "hcall-pft");
1056     add_str(hypertas, "hcall-term");
1057     add_str(hypertas, "hcall-dabr");
1058     add_str(hypertas, "hcall-interrupt");
1059     add_str(hypertas, "hcall-tce");
1060     add_str(hypertas, "hcall-vio");
1061     add_str(hypertas, "hcall-splpar");
1062     add_str(hypertas, "hcall-bulk");
1063     add_str(hypertas, "hcall-set-mode");
1064     add_str(hypertas, "hcall-sprg0");
1065     add_str(hypertas, "hcall-copy");
1066     add_str(hypertas, "hcall-debug");
1067     add_str(qemu_hypertas, "hcall-memop1");
1068 
1069     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1070         add_str(hypertas, "hcall-multi-tce");
1071     }
1072 
1073     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1074         add_str(hypertas, "hcall-hpt-resize");
1075     }
1076 
1077     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1078                      hypertas->str, hypertas->len));
1079     g_string_free(hypertas, TRUE);
1080     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1081                      qemu_hypertas->str, qemu_hypertas->len));
1082     g_string_free(qemu_hypertas, TRUE);
1083 
1084     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1085                      refpoints, sizeof(refpoints)));
1086 
1087     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1088                      maxdomains, sizeof(maxdomains)));
1089 
1090     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1091                           RTAS_ERROR_LOG_MAX));
1092     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1093                           RTAS_EVENT_SCAN_RATE));
1094 
1095     g_assert(msi_nonbroken);
1096     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1097 
1098     /*
1099      * According to PAPR, rtas ibm,os-term does not guarantee a return
1100      * back to the guest cpu.
1101      *
1102      * While an additional ibm,extended-os-term property indicates
1103      * that rtas call return will always occur. Set this property.
1104      */
1105     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1106 
1107     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1108                      lrdr_capacity, sizeof(lrdr_capacity)));
1109 
1110     spapr_dt_rtas_tokens(fdt, rtas);
1111 }
1112 
1113 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1114  * that the guest may request and thus the valid values for bytes 24..26 of
1115  * option vector 5: */
1116 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1117 {
1118     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1119 
1120     char val[2 * 4] = {
1121         23, 0x00, /* Xive mode, filled in below. */
1122         24, 0x00, /* Hash/Radix, filled in below. */
1123         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1124         26, 0x40, /* Radix options: GTSE == yes. */
1125     };
1126 
1127     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1128                           first_ppc_cpu->compat_pvr)) {
1129         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1130         val[3] = 0x00; /* Hash */
1131     } else if (kvm_enabled()) {
1132         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1133             val[3] = 0x80; /* OV5_MMU_BOTH */
1134         } else if (kvmppc_has_cap_mmu_radix()) {
1135             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1136         } else {
1137             val[3] = 0x00; /* Hash */
1138         }
1139     } else {
1140         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1141         val[3] = 0xC0;
1142     }
1143     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1144                      val, sizeof(val)));
1145 }
1146 
1147 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1148 {
1149     MachineState *machine = MACHINE(spapr);
1150     int chosen;
1151     const char *boot_device = machine->boot_order;
1152     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1153     size_t cb = 0;
1154     char *bootlist = get_boot_devices_list(&cb, true);
1155 
1156     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1157 
1158     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1159     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1160                           spapr->initrd_base));
1161     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1162                           spapr->initrd_base + spapr->initrd_size));
1163 
1164     if (spapr->kernel_size) {
1165         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1166                               cpu_to_be64(spapr->kernel_size) };
1167 
1168         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1169                          &kprop, sizeof(kprop)));
1170         if (spapr->kernel_le) {
1171             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1172         }
1173     }
1174     if (boot_menu) {
1175         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1176     }
1177     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1178     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1179     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1180 
1181     if (cb && bootlist) {
1182         int i;
1183 
1184         for (i = 0; i < cb; i++) {
1185             if (bootlist[i] == '\n') {
1186                 bootlist[i] = ' ';
1187             }
1188         }
1189         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1190     }
1191 
1192     if (boot_device && strlen(boot_device)) {
1193         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1194     }
1195 
1196     if (!spapr->has_graphics && stdout_path) {
1197         /*
1198          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1199          * kernel. New platforms should only use the "stdout-path" property. Set
1200          * the new property and continue using older property to remain
1201          * compatible with the existing firmware.
1202          */
1203         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1204         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1205     }
1206 
1207     spapr_dt_ov5_platform_support(fdt, chosen);
1208 
1209     g_free(stdout_path);
1210     g_free(bootlist);
1211 }
1212 
1213 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1214 {
1215     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1216      * KVM to work under pHyp with some guest co-operation */
1217     int hypervisor;
1218     uint8_t hypercall[16];
1219 
1220     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1221     /* indicate KVM hypercall interface */
1222     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1223     if (kvmppc_has_cap_fixup_hcalls()) {
1224         /*
1225          * Older KVM versions with older guest kernels were broken
1226          * with the magic page, don't allow the guest to map it.
1227          */
1228         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1229                                   sizeof(hypercall))) {
1230             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1231                              hypercall, sizeof(hypercall)));
1232         }
1233     }
1234 }
1235 
1236 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1237                              hwaddr rtas_addr,
1238                              hwaddr rtas_size)
1239 {
1240     MachineState *machine = MACHINE(spapr);
1241     MachineClass *mc = MACHINE_GET_CLASS(machine);
1242     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1243     int ret;
1244     void *fdt;
1245     sPAPRPHBState *phb;
1246     char *buf;
1247 
1248     fdt = g_malloc0(FDT_MAX_SIZE);
1249     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1250 
1251     /* Root node */
1252     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1253     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1254     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1255 
1256     /*
1257      * Add info to guest to indentify which host is it being run on
1258      * and what is the uuid of the guest
1259      */
1260     if (kvmppc_get_host_model(&buf)) {
1261         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1262         g_free(buf);
1263     }
1264     if (kvmppc_get_host_serial(&buf)) {
1265         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1266         g_free(buf);
1267     }
1268 
1269     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1270 
1271     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1272     if (qemu_uuid_set) {
1273         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1274     }
1275     g_free(buf);
1276 
1277     if (qemu_get_vm_name()) {
1278         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1279                                 qemu_get_vm_name()));
1280     }
1281 
1282     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1283     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1284 
1285     /* /interrupt controller */
1286     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1287 
1288     ret = spapr_populate_memory(spapr, fdt);
1289     if (ret < 0) {
1290         error_report("couldn't setup memory nodes in fdt");
1291         exit(1);
1292     }
1293 
1294     /* /vdevice */
1295     spapr_dt_vdevice(spapr->vio_bus, fdt);
1296 
1297     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1298         ret = spapr_rng_populate_dt(fdt);
1299         if (ret < 0) {
1300             error_report("could not set up rng device in the fdt");
1301             exit(1);
1302         }
1303     }
1304 
1305     QLIST_FOREACH(phb, &spapr->phbs, list) {
1306         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1307         if (ret < 0) {
1308             error_report("couldn't setup PCI devices in fdt");
1309             exit(1);
1310         }
1311     }
1312 
1313     /* cpus */
1314     spapr_populate_cpus_dt_node(fdt, spapr);
1315 
1316     if (smc->dr_lmb_enabled) {
1317         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1318     }
1319 
1320     if (mc->has_hotpluggable_cpus) {
1321         int offset = fdt_path_offset(fdt, "/cpus");
1322         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1323                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1324         if (ret < 0) {
1325             error_report("Couldn't set up CPU DR device tree properties");
1326             exit(1);
1327         }
1328     }
1329 
1330     /* /event-sources */
1331     spapr_dt_events(spapr, fdt);
1332 
1333     /* /rtas */
1334     spapr_dt_rtas(spapr, fdt);
1335 
1336     /* /chosen */
1337     spapr_dt_chosen(spapr, fdt);
1338 
1339     /* /hypervisor */
1340     if (kvm_enabled()) {
1341         spapr_dt_hypervisor(spapr, fdt);
1342     }
1343 
1344     /* Build memory reserve map */
1345     if (spapr->kernel_size) {
1346         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1347     }
1348     if (spapr->initrd_size) {
1349         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1350     }
1351 
1352     /* ibm,client-architecture-support updates */
1353     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1354     if (ret < 0) {
1355         error_report("couldn't setup CAS properties fdt");
1356         exit(1);
1357     }
1358 
1359     return fdt;
1360 }
1361 
1362 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1363 {
1364     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1365 }
1366 
1367 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1368                                     PowerPCCPU *cpu)
1369 {
1370     CPUPPCState *env = &cpu->env;
1371 
1372     /* The TCG path should also be holding the BQL at this point */
1373     g_assert(qemu_mutex_iothread_locked());
1374 
1375     if (msr_pr) {
1376         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1377         env->gpr[3] = H_PRIVILEGE;
1378     } else {
1379         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1380     }
1381 }
1382 
1383 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1384 {
1385     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1386 
1387     return spapr->patb_entry;
1388 }
1389 
1390 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1391 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1392 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1393 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1394 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1395 
1396 /*
1397  * Get the fd to access the kernel htab, re-opening it if necessary
1398  */
1399 static int get_htab_fd(sPAPRMachineState *spapr)
1400 {
1401     Error *local_err = NULL;
1402 
1403     if (spapr->htab_fd >= 0) {
1404         return spapr->htab_fd;
1405     }
1406 
1407     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1408     if (spapr->htab_fd < 0) {
1409         error_report_err(local_err);
1410     }
1411 
1412     return spapr->htab_fd;
1413 }
1414 
1415 void close_htab_fd(sPAPRMachineState *spapr)
1416 {
1417     if (spapr->htab_fd >= 0) {
1418         close(spapr->htab_fd);
1419     }
1420     spapr->htab_fd = -1;
1421 }
1422 
1423 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1424 {
1425     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1426 
1427     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1428 }
1429 
1430 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1431 {
1432     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1433 
1434     assert(kvm_enabled());
1435 
1436     if (!spapr->htab) {
1437         return 0;
1438     }
1439 
1440     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1441 }
1442 
1443 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1444                                                 hwaddr ptex, int n)
1445 {
1446     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1447     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1448 
1449     if (!spapr->htab) {
1450         /*
1451          * HTAB is controlled by KVM. Fetch into temporary buffer
1452          */
1453         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1454         kvmppc_read_hptes(hptes, ptex, n);
1455         return hptes;
1456     }
1457 
1458     /*
1459      * HTAB is controlled by QEMU. Just point to the internally
1460      * accessible PTEG.
1461      */
1462     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1463 }
1464 
1465 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1466                               const ppc_hash_pte64_t *hptes,
1467                               hwaddr ptex, int n)
1468 {
1469     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1470 
1471     if (!spapr->htab) {
1472         g_free((void *)hptes);
1473     }
1474 
1475     /* Nothing to do for qemu managed HPT */
1476 }
1477 
1478 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1479                              uint64_t pte0, uint64_t pte1)
1480 {
1481     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1482     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1483 
1484     if (!spapr->htab) {
1485         kvmppc_write_hpte(ptex, pte0, pte1);
1486     } else {
1487         stq_p(spapr->htab + offset, pte0);
1488         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1489     }
1490 }
1491 
1492 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1493 {
1494     int shift;
1495 
1496     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1497      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1498      * that's much more than is needed for Linux guests */
1499     shift = ctz64(pow2ceil(ramsize)) - 7;
1500     shift = MAX(shift, 18); /* Minimum architected size */
1501     shift = MIN(shift, 46); /* Maximum architected size */
1502     return shift;
1503 }
1504 
1505 void spapr_free_hpt(sPAPRMachineState *spapr)
1506 {
1507     g_free(spapr->htab);
1508     spapr->htab = NULL;
1509     spapr->htab_shift = 0;
1510     close_htab_fd(spapr);
1511 }
1512 
1513 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1514                           Error **errp)
1515 {
1516     long rc;
1517 
1518     /* Clean up any HPT info from a previous boot */
1519     spapr_free_hpt(spapr);
1520 
1521     rc = kvmppc_reset_htab(shift);
1522     if (rc < 0) {
1523         /* kernel-side HPT needed, but couldn't allocate one */
1524         error_setg_errno(errp, errno,
1525                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1526                          shift);
1527         /* This is almost certainly fatal, but if the caller really
1528          * wants to carry on with shift == 0, it's welcome to try */
1529     } else if (rc > 0) {
1530         /* kernel-side HPT allocated */
1531         if (rc != shift) {
1532             error_setg(errp,
1533                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1534                        shift, rc);
1535         }
1536 
1537         spapr->htab_shift = shift;
1538         spapr->htab = NULL;
1539     } else {
1540         /* kernel-side HPT not needed, allocate in userspace instead */
1541         size_t size = 1ULL << shift;
1542         int i;
1543 
1544         spapr->htab = qemu_memalign(size, size);
1545         if (!spapr->htab) {
1546             error_setg_errno(errp, errno,
1547                              "Could not allocate HPT of order %d", shift);
1548             return;
1549         }
1550 
1551         memset(spapr->htab, 0, size);
1552         spapr->htab_shift = shift;
1553 
1554         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1555             DIRTY_HPTE(HPTE(spapr->htab, i));
1556         }
1557     }
1558     /* We're setting up a hash table, so that means we're not radix */
1559     spapr->patb_entry = 0;
1560 }
1561 
1562 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1563 {
1564     int hpt_shift;
1565 
1566     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1567         || (spapr->cas_reboot
1568             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1569         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1570     } else {
1571         uint64_t current_ram_size;
1572 
1573         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1574         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1575     }
1576     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1577 
1578     if (spapr->vrma_adjust) {
1579         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1580                                           spapr->htab_shift);
1581     }
1582 }
1583 
1584 static int spapr_reset_drcs(Object *child, void *opaque)
1585 {
1586     sPAPRDRConnector *drc =
1587         (sPAPRDRConnector *) object_dynamic_cast(child,
1588                                                  TYPE_SPAPR_DR_CONNECTOR);
1589 
1590     if (drc) {
1591         spapr_drc_reset(drc);
1592     }
1593 
1594     return 0;
1595 }
1596 
1597 static void spapr_machine_reset(void)
1598 {
1599     MachineState *machine = MACHINE(qdev_get_machine());
1600     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1601     PowerPCCPU *first_ppc_cpu;
1602     uint32_t rtas_limit;
1603     hwaddr rtas_addr, fdt_addr;
1604     void *fdt;
1605     int rc;
1606 
1607     spapr_caps_reset(spapr);
1608 
1609     first_ppc_cpu = POWERPC_CPU(first_cpu);
1610     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1611         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1612                          spapr->max_compat_pvr)) {
1613         /* If using KVM with radix mode available, VCPUs can be started
1614          * without a HPT because KVM will start them in radix mode.
1615          * Set the GR bit in PATB so that we know there is no HPT. */
1616         spapr->patb_entry = PATBE1_GR;
1617     } else {
1618         spapr_setup_hpt_and_vrma(spapr);
1619     }
1620 
1621     /* if this reset wasn't generated by CAS, we should reset our
1622      * negotiated options and start from scratch */
1623     if (!spapr->cas_reboot) {
1624         spapr_ovec_cleanup(spapr->ov5_cas);
1625         spapr->ov5_cas = spapr_ovec_new();
1626 
1627         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1628     }
1629 
1630     qemu_devices_reset();
1631 
1632     /* DRC reset may cause a device to be unplugged. This will cause troubles
1633      * if this device is used by another device (eg, a running vhost backend
1634      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1635      * situations, we reset DRCs after all devices have been reset.
1636      */
1637     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1638 
1639     spapr_clear_pending_events(spapr);
1640 
1641     /*
1642      * We place the device tree and RTAS just below either the top of the RMA,
1643      * or just below 2GB, whichever is lowere, so that it can be
1644      * processed with 32-bit real mode code if necessary
1645      */
1646     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1647     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1648     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1649 
1650     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1651 
1652     spapr_load_rtas(spapr, fdt, rtas_addr);
1653 
1654     rc = fdt_pack(fdt);
1655 
1656     /* Should only fail if we've built a corrupted tree */
1657     assert(rc == 0);
1658 
1659     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1660         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1661                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1662         exit(1);
1663     }
1664 
1665     /* Load the fdt */
1666     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1667     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1668     g_free(fdt);
1669 
1670     /* Set up the entry state */
1671     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1672     first_ppc_cpu->env.gpr[5] = 0;
1673 
1674     spapr->cas_reboot = false;
1675 }
1676 
1677 static void spapr_create_nvram(sPAPRMachineState *spapr)
1678 {
1679     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1680     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1681 
1682     if (dinfo) {
1683         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1684                             &error_fatal);
1685     }
1686 
1687     qdev_init_nofail(dev);
1688 
1689     spapr->nvram = (struct sPAPRNVRAM *)dev;
1690 }
1691 
1692 static void spapr_rtc_create(sPAPRMachineState *spapr)
1693 {
1694     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1695     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1696                               &error_fatal);
1697     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1698                               &error_fatal);
1699     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1700                               "date", &error_fatal);
1701 }
1702 
1703 /* Returns whether we want to use VGA or not */
1704 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1705 {
1706     switch (vga_interface_type) {
1707     case VGA_NONE:
1708         return false;
1709     case VGA_DEVICE:
1710         return true;
1711     case VGA_STD:
1712     case VGA_VIRTIO:
1713         return pci_vga_init(pci_bus) != NULL;
1714     default:
1715         error_setg(errp,
1716                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1717         return false;
1718     }
1719 }
1720 
1721 static int spapr_pre_load(void *opaque)
1722 {
1723     int rc;
1724 
1725     rc = spapr_caps_pre_load(opaque);
1726     if (rc) {
1727         return rc;
1728     }
1729 
1730     return 0;
1731 }
1732 
1733 static int spapr_post_load(void *opaque, int version_id)
1734 {
1735     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1736     int err = 0;
1737 
1738     err = spapr_caps_post_migration(spapr);
1739     if (err) {
1740         return err;
1741     }
1742 
1743     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1744         CPUState *cs;
1745         CPU_FOREACH(cs) {
1746             PowerPCCPU *cpu = POWERPC_CPU(cs);
1747             icp_resend(ICP(cpu->intc));
1748         }
1749     }
1750 
1751     /* In earlier versions, there was no separate qdev for the PAPR
1752      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1753      * So when migrating from those versions, poke the incoming offset
1754      * value into the RTC device */
1755     if (version_id < 3) {
1756         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1757     }
1758 
1759     if (kvm_enabled() && spapr->patb_entry) {
1760         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1761         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1762         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1763 
1764         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1765         if (err) {
1766             error_report("Process table config unsupported by the host");
1767             return -EINVAL;
1768         }
1769     }
1770 
1771     return err;
1772 }
1773 
1774 static int spapr_pre_save(void *opaque)
1775 {
1776     int rc;
1777 
1778     rc = spapr_caps_pre_save(opaque);
1779     if (rc) {
1780         return rc;
1781     }
1782 
1783     return 0;
1784 }
1785 
1786 static bool version_before_3(void *opaque, int version_id)
1787 {
1788     return version_id < 3;
1789 }
1790 
1791 static bool spapr_pending_events_needed(void *opaque)
1792 {
1793     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1794     return !QTAILQ_EMPTY(&spapr->pending_events);
1795 }
1796 
1797 static const VMStateDescription vmstate_spapr_event_entry = {
1798     .name = "spapr_event_log_entry",
1799     .version_id = 1,
1800     .minimum_version_id = 1,
1801     .fields = (VMStateField[]) {
1802         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1803         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1804         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1805                                      NULL, extended_length),
1806         VMSTATE_END_OF_LIST()
1807     },
1808 };
1809 
1810 static const VMStateDescription vmstate_spapr_pending_events = {
1811     .name = "spapr_pending_events",
1812     .version_id = 1,
1813     .minimum_version_id = 1,
1814     .needed = spapr_pending_events_needed,
1815     .fields = (VMStateField[]) {
1816         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1817                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1818         VMSTATE_END_OF_LIST()
1819     },
1820 };
1821 
1822 static bool spapr_ov5_cas_needed(void *opaque)
1823 {
1824     sPAPRMachineState *spapr = opaque;
1825     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1826     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1827     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1828     bool cas_needed;
1829 
1830     /* Prior to the introduction of sPAPROptionVector, we had two option
1831      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1832      * Both of these options encode machine topology into the device-tree
1833      * in such a way that the now-booted OS should still be able to interact
1834      * appropriately with QEMU regardless of what options were actually
1835      * negotiatied on the source side.
1836      *
1837      * As such, we can avoid migrating the CAS-negotiated options if these
1838      * are the only options available on the current machine/platform.
1839      * Since these are the only options available for pseries-2.7 and
1840      * earlier, this allows us to maintain old->new/new->old migration
1841      * compatibility.
1842      *
1843      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1844      * via default pseries-2.8 machines and explicit command-line parameters.
1845      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1846      * of the actual CAS-negotiated values to continue working properly. For
1847      * example, availability of memory unplug depends on knowing whether
1848      * OV5_HP_EVT was negotiated via CAS.
1849      *
1850      * Thus, for any cases where the set of available CAS-negotiatable
1851      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1852      * include the CAS-negotiated options in the migration stream, unless
1853      * if they affect boot time behaviour only.
1854      */
1855     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1856     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1857     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1858 
1859     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1860      * the mask itself since in the future it's possible "legacy" bits may be
1861      * removed via machine options, which could generate a false positive
1862      * that breaks migration.
1863      */
1864     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1865     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1866 
1867     spapr_ovec_cleanup(ov5_mask);
1868     spapr_ovec_cleanup(ov5_legacy);
1869     spapr_ovec_cleanup(ov5_removed);
1870 
1871     return cas_needed;
1872 }
1873 
1874 static const VMStateDescription vmstate_spapr_ov5_cas = {
1875     .name = "spapr_option_vector_ov5_cas",
1876     .version_id = 1,
1877     .minimum_version_id = 1,
1878     .needed = spapr_ov5_cas_needed,
1879     .fields = (VMStateField[]) {
1880         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1881                                  vmstate_spapr_ovec, sPAPROptionVector),
1882         VMSTATE_END_OF_LIST()
1883     },
1884 };
1885 
1886 static bool spapr_patb_entry_needed(void *opaque)
1887 {
1888     sPAPRMachineState *spapr = opaque;
1889 
1890     return !!spapr->patb_entry;
1891 }
1892 
1893 static const VMStateDescription vmstate_spapr_patb_entry = {
1894     .name = "spapr_patb_entry",
1895     .version_id = 1,
1896     .minimum_version_id = 1,
1897     .needed = spapr_patb_entry_needed,
1898     .fields = (VMStateField[]) {
1899         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1900         VMSTATE_END_OF_LIST()
1901     },
1902 };
1903 
1904 static const VMStateDescription vmstate_spapr = {
1905     .name = "spapr",
1906     .version_id = 3,
1907     .minimum_version_id = 1,
1908     .pre_load = spapr_pre_load,
1909     .post_load = spapr_post_load,
1910     .pre_save = spapr_pre_save,
1911     .fields = (VMStateField[]) {
1912         /* used to be @next_irq */
1913         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1914 
1915         /* RTC offset */
1916         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1917 
1918         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1919         VMSTATE_END_OF_LIST()
1920     },
1921     .subsections = (const VMStateDescription*[]) {
1922         &vmstate_spapr_ov5_cas,
1923         &vmstate_spapr_patb_entry,
1924         &vmstate_spapr_pending_events,
1925         &vmstate_spapr_cap_htm,
1926         &vmstate_spapr_cap_vsx,
1927         &vmstate_spapr_cap_dfp,
1928         &vmstate_spapr_cap_cfpc,
1929         &vmstate_spapr_cap_sbbc,
1930         &vmstate_spapr_cap_ibs,
1931         NULL
1932     }
1933 };
1934 
1935 static int htab_save_setup(QEMUFile *f, void *opaque)
1936 {
1937     sPAPRMachineState *spapr = opaque;
1938 
1939     /* "Iteration" header */
1940     if (!spapr->htab_shift) {
1941         qemu_put_be32(f, -1);
1942     } else {
1943         qemu_put_be32(f, spapr->htab_shift);
1944     }
1945 
1946     if (spapr->htab) {
1947         spapr->htab_save_index = 0;
1948         spapr->htab_first_pass = true;
1949     } else {
1950         if (spapr->htab_shift) {
1951             assert(kvm_enabled());
1952         }
1953     }
1954 
1955 
1956     return 0;
1957 }
1958 
1959 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1960                             int chunkstart, int n_valid, int n_invalid)
1961 {
1962     qemu_put_be32(f, chunkstart);
1963     qemu_put_be16(f, n_valid);
1964     qemu_put_be16(f, n_invalid);
1965     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1966                     HASH_PTE_SIZE_64 * n_valid);
1967 }
1968 
1969 static void htab_save_end_marker(QEMUFile *f)
1970 {
1971     qemu_put_be32(f, 0);
1972     qemu_put_be16(f, 0);
1973     qemu_put_be16(f, 0);
1974 }
1975 
1976 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1977                                  int64_t max_ns)
1978 {
1979     bool has_timeout = max_ns != -1;
1980     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1981     int index = spapr->htab_save_index;
1982     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1983 
1984     assert(spapr->htab_first_pass);
1985 
1986     do {
1987         int chunkstart;
1988 
1989         /* Consume invalid HPTEs */
1990         while ((index < htabslots)
1991                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1992             CLEAN_HPTE(HPTE(spapr->htab, index));
1993             index++;
1994         }
1995 
1996         /* Consume valid HPTEs */
1997         chunkstart = index;
1998         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1999                && HPTE_VALID(HPTE(spapr->htab, index))) {
2000             CLEAN_HPTE(HPTE(spapr->htab, index));
2001             index++;
2002         }
2003 
2004         if (index > chunkstart) {
2005             int n_valid = index - chunkstart;
2006 
2007             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2008 
2009             if (has_timeout &&
2010                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2011                 break;
2012             }
2013         }
2014     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2015 
2016     if (index >= htabslots) {
2017         assert(index == htabslots);
2018         index = 0;
2019         spapr->htab_first_pass = false;
2020     }
2021     spapr->htab_save_index = index;
2022 }
2023 
2024 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2025                                 int64_t max_ns)
2026 {
2027     bool final = max_ns < 0;
2028     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2029     int examined = 0, sent = 0;
2030     int index = spapr->htab_save_index;
2031     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2032 
2033     assert(!spapr->htab_first_pass);
2034 
2035     do {
2036         int chunkstart, invalidstart;
2037 
2038         /* Consume non-dirty HPTEs */
2039         while ((index < htabslots)
2040                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2041             index++;
2042             examined++;
2043         }
2044 
2045         chunkstart = index;
2046         /* Consume valid dirty HPTEs */
2047         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2048                && HPTE_DIRTY(HPTE(spapr->htab, index))
2049                && HPTE_VALID(HPTE(spapr->htab, index))) {
2050             CLEAN_HPTE(HPTE(spapr->htab, index));
2051             index++;
2052             examined++;
2053         }
2054 
2055         invalidstart = index;
2056         /* Consume invalid dirty HPTEs */
2057         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2058                && HPTE_DIRTY(HPTE(spapr->htab, index))
2059                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2060             CLEAN_HPTE(HPTE(spapr->htab, index));
2061             index++;
2062             examined++;
2063         }
2064 
2065         if (index > chunkstart) {
2066             int n_valid = invalidstart - chunkstart;
2067             int n_invalid = index - invalidstart;
2068 
2069             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2070             sent += index - chunkstart;
2071 
2072             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2073                 break;
2074             }
2075         }
2076 
2077         if (examined >= htabslots) {
2078             break;
2079         }
2080 
2081         if (index >= htabslots) {
2082             assert(index == htabslots);
2083             index = 0;
2084         }
2085     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2086 
2087     if (index >= htabslots) {
2088         assert(index == htabslots);
2089         index = 0;
2090     }
2091 
2092     spapr->htab_save_index = index;
2093 
2094     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2095 }
2096 
2097 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2098 #define MAX_KVM_BUF_SIZE    2048
2099 
2100 static int htab_save_iterate(QEMUFile *f, void *opaque)
2101 {
2102     sPAPRMachineState *spapr = opaque;
2103     int fd;
2104     int rc = 0;
2105 
2106     /* Iteration header */
2107     if (!spapr->htab_shift) {
2108         qemu_put_be32(f, -1);
2109         return 1;
2110     } else {
2111         qemu_put_be32(f, 0);
2112     }
2113 
2114     if (!spapr->htab) {
2115         assert(kvm_enabled());
2116 
2117         fd = get_htab_fd(spapr);
2118         if (fd < 0) {
2119             return fd;
2120         }
2121 
2122         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2123         if (rc < 0) {
2124             return rc;
2125         }
2126     } else  if (spapr->htab_first_pass) {
2127         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2128     } else {
2129         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2130     }
2131 
2132     htab_save_end_marker(f);
2133 
2134     return rc;
2135 }
2136 
2137 static int htab_save_complete(QEMUFile *f, void *opaque)
2138 {
2139     sPAPRMachineState *spapr = opaque;
2140     int fd;
2141 
2142     /* Iteration header */
2143     if (!spapr->htab_shift) {
2144         qemu_put_be32(f, -1);
2145         return 0;
2146     } else {
2147         qemu_put_be32(f, 0);
2148     }
2149 
2150     if (!spapr->htab) {
2151         int rc;
2152 
2153         assert(kvm_enabled());
2154 
2155         fd = get_htab_fd(spapr);
2156         if (fd < 0) {
2157             return fd;
2158         }
2159 
2160         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2161         if (rc < 0) {
2162             return rc;
2163         }
2164     } else {
2165         if (spapr->htab_first_pass) {
2166             htab_save_first_pass(f, spapr, -1);
2167         }
2168         htab_save_later_pass(f, spapr, -1);
2169     }
2170 
2171     /* End marker */
2172     htab_save_end_marker(f);
2173 
2174     return 0;
2175 }
2176 
2177 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2178 {
2179     sPAPRMachineState *spapr = opaque;
2180     uint32_t section_hdr;
2181     int fd = -1;
2182     Error *local_err = NULL;
2183 
2184     if (version_id < 1 || version_id > 1) {
2185         error_report("htab_load() bad version");
2186         return -EINVAL;
2187     }
2188 
2189     section_hdr = qemu_get_be32(f);
2190 
2191     if (section_hdr == -1) {
2192         spapr_free_hpt(spapr);
2193         return 0;
2194     }
2195 
2196     if (section_hdr) {
2197         /* First section gives the htab size */
2198         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2199         if (local_err) {
2200             error_report_err(local_err);
2201             return -EINVAL;
2202         }
2203         return 0;
2204     }
2205 
2206     if (!spapr->htab) {
2207         assert(kvm_enabled());
2208 
2209         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2210         if (fd < 0) {
2211             error_report_err(local_err);
2212             return fd;
2213         }
2214     }
2215 
2216     while (true) {
2217         uint32_t index;
2218         uint16_t n_valid, n_invalid;
2219 
2220         index = qemu_get_be32(f);
2221         n_valid = qemu_get_be16(f);
2222         n_invalid = qemu_get_be16(f);
2223 
2224         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2225             /* End of Stream */
2226             break;
2227         }
2228 
2229         if ((index + n_valid + n_invalid) >
2230             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2231             /* Bad index in stream */
2232             error_report(
2233                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2234                 index, n_valid, n_invalid, spapr->htab_shift);
2235             return -EINVAL;
2236         }
2237 
2238         if (spapr->htab) {
2239             if (n_valid) {
2240                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2241                                 HASH_PTE_SIZE_64 * n_valid);
2242             }
2243             if (n_invalid) {
2244                 memset(HPTE(spapr->htab, index + n_valid), 0,
2245                        HASH_PTE_SIZE_64 * n_invalid);
2246             }
2247         } else {
2248             int rc;
2249 
2250             assert(fd >= 0);
2251 
2252             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2253             if (rc < 0) {
2254                 return rc;
2255             }
2256         }
2257     }
2258 
2259     if (!spapr->htab) {
2260         assert(fd >= 0);
2261         close(fd);
2262     }
2263 
2264     return 0;
2265 }
2266 
2267 static void htab_save_cleanup(void *opaque)
2268 {
2269     sPAPRMachineState *spapr = opaque;
2270 
2271     close_htab_fd(spapr);
2272 }
2273 
2274 static SaveVMHandlers savevm_htab_handlers = {
2275     .save_setup = htab_save_setup,
2276     .save_live_iterate = htab_save_iterate,
2277     .save_live_complete_precopy = htab_save_complete,
2278     .save_cleanup = htab_save_cleanup,
2279     .load_state = htab_load,
2280 };
2281 
2282 static void spapr_boot_set(void *opaque, const char *boot_device,
2283                            Error **errp)
2284 {
2285     MachineState *machine = MACHINE(opaque);
2286     machine->boot_order = g_strdup(boot_device);
2287 }
2288 
2289 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2290 {
2291     MachineState *machine = MACHINE(spapr);
2292     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2293     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2294     int i;
2295 
2296     for (i = 0; i < nr_lmbs; i++) {
2297         uint64_t addr;
2298 
2299         addr = i * lmb_size + spapr->hotplug_memory.base;
2300         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2301                                addr / lmb_size);
2302     }
2303 }
2304 
2305 /*
2306  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2307  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2308  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2309  */
2310 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2311 {
2312     int i;
2313 
2314     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2315         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2316                    " is not aligned to %llu MiB",
2317                    machine->ram_size,
2318                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2319         return;
2320     }
2321 
2322     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2323         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2324                    " is not aligned to %llu MiB",
2325                    machine->ram_size,
2326                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2327         return;
2328     }
2329 
2330     for (i = 0; i < nb_numa_nodes; i++) {
2331         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2332             error_setg(errp,
2333                        "Node %d memory size 0x%" PRIx64
2334                        " is not aligned to %llu MiB",
2335                        i, numa_info[i].node_mem,
2336                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2337             return;
2338         }
2339     }
2340 }
2341 
2342 /* find cpu slot in machine->possible_cpus by core_id */
2343 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2344 {
2345     int index = id / smp_threads;
2346 
2347     if (index >= ms->possible_cpus->len) {
2348         return NULL;
2349     }
2350     if (idx) {
2351         *idx = index;
2352     }
2353     return &ms->possible_cpus->cpus[index];
2354 }
2355 
2356 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2357 {
2358     Error *local_err = NULL;
2359     bool vsmt_user = !!spapr->vsmt;
2360     int kvm_smt = kvmppc_smt_threads();
2361     int ret;
2362 
2363     if (!kvm_enabled() && (smp_threads > 1)) {
2364         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2365                      "on a pseries machine");
2366         goto out;
2367     }
2368     if (!is_power_of_2(smp_threads)) {
2369         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2370                      "machine because it must be a power of 2", smp_threads);
2371         goto out;
2372     }
2373 
2374     /* Detemine the VSMT mode to use: */
2375     if (vsmt_user) {
2376         if (spapr->vsmt < smp_threads) {
2377             error_setg(&local_err, "Cannot support VSMT mode %d"
2378                          " because it must be >= threads/core (%d)",
2379                          spapr->vsmt, smp_threads);
2380             goto out;
2381         }
2382         /* In this case, spapr->vsmt has been set by the command line */
2383     } else {
2384         /*
2385          * Default VSMT value is tricky, because we need it to be as
2386          * consistent as possible (for migration), but this requires
2387          * changing it for at least some existing cases.  We pick 8 as
2388          * the value that we'd get with KVM on POWER8, the
2389          * overwhelmingly common case in production systems.
2390          */
2391         spapr->vsmt = MAX(8, smp_threads);
2392     }
2393 
2394     /* KVM: If necessary, set the SMT mode: */
2395     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2396         ret = kvmppc_set_smt_threads(spapr->vsmt);
2397         if (ret) {
2398             /* Looks like KVM isn't able to change VSMT mode */
2399             error_setg(&local_err,
2400                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2401                        spapr->vsmt, ret);
2402             /* We can live with that if the default one is big enough
2403              * for the number of threads, and a submultiple of the one
2404              * we want.  In this case we'll waste some vcpu ids, but
2405              * behaviour will be correct */
2406             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2407                 warn_report_err(local_err);
2408                 local_err = NULL;
2409                 goto out;
2410             } else {
2411                 if (!vsmt_user) {
2412                     error_append_hint(&local_err,
2413                                       "On PPC, a VM with %d threads/core"
2414                                       " on a host with %d threads/core"
2415                                       " requires the use of VSMT mode %d.\n",
2416                                       smp_threads, kvm_smt, spapr->vsmt);
2417                 }
2418                 kvmppc_hint_smt_possible(&local_err);
2419                 goto out;
2420             }
2421         }
2422     }
2423     /* else TCG: nothing to do currently */
2424 out:
2425     error_propagate(errp, local_err);
2426 }
2427 
2428 static void spapr_init_cpus(sPAPRMachineState *spapr)
2429 {
2430     MachineState *machine = MACHINE(spapr);
2431     MachineClass *mc = MACHINE_GET_CLASS(machine);
2432     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2433     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2434     const CPUArchIdList *possible_cpus;
2435     int boot_cores_nr = smp_cpus / smp_threads;
2436     int i;
2437 
2438     possible_cpus = mc->possible_cpu_arch_ids(machine);
2439     if (mc->has_hotpluggable_cpus) {
2440         if (smp_cpus % smp_threads) {
2441             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2442                          smp_cpus, smp_threads);
2443             exit(1);
2444         }
2445         if (max_cpus % smp_threads) {
2446             error_report("max_cpus (%u) must be multiple of threads (%u)",
2447                          max_cpus, smp_threads);
2448             exit(1);
2449         }
2450     } else {
2451         if (max_cpus != smp_cpus) {
2452             error_report("This machine version does not support CPU hotplug");
2453             exit(1);
2454         }
2455         boot_cores_nr = possible_cpus->len;
2456     }
2457 
2458     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2459      * call xics_max_server_number() or spapr_vcpu_id().
2460      */
2461     spapr_set_vsmt_mode(spapr, &error_fatal);
2462 
2463     if (smc->pre_2_10_has_unused_icps) {
2464         int i;
2465 
2466         for (i = 0; i < xics_max_server_number(spapr); i++) {
2467             /* Dummy entries get deregistered when real ICPState objects
2468              * are registered during CPU core hotplug.
2469              */
2470             pre_2_10_vmstate_register_dummy_icp(i);
2471         }
2472     }
2473 
2474     for (i = 0; i < possible_cpus->len; i++) {
2475         int core_id = i * smp_threads;
2476 
2477         if (mc->has_hotpluggable_cpus) {
2478             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2479                                    spapr_vcpu_id(spapr, core_id));
2480         }
2481 
2482         if (i < boot_cores_nr) {
2483             Object *core  = object_new(type);
2484             int nr_threads = smp_threads;
2485 
2486             /* Handle the partially filled core for older machine types */
2487             if ((i + 1) * smp_threads >= smp_cpus) {
2488                 nr_threads = smp_cpus - i * smp_threads;
2489             }
2490 
2491             object_property_set_int(core, nr_threads, "nr-threads",
2492                                     &error_fatal);
2493             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2494                                     &error_fatal);
2495             object_property_set_bool(core, true, "realized", &error_fatal);
2496         }
2497     }
2498 }
2499 
2500 /* pSeries LPAR / sPAPR hardware init */
2501 static void spapr_machine_init(MachineState *machine)
2502 {
2503     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2504     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2505     const char *kernel_filename = machine->kernel_filename;
2506     const char *initrd_filename = machine->initrd_filename;
2507     PCIHostState *phb;
2508     int i;
2509     MemoryRegion *sysmem = get_system_memory();
2510     MemoryRegion *ram = g_new(MemoryRegion, 1);
2511     hwaddr node0_size = spapr_node0_size(machine);
2512     long load_limit, fw_size;
2513     char *filename;
2514     Error *resize_hpt_err = NULL;
2515     PowerPCCPU *first_ppc_cpu;
2516 
2517     msi_nonbroken = true;
2518 
2519     QLIST_INIT(&spapr->phbs);
2520     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2521 
2522     /* Check HPT resizing availability */
2523     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2524     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2525         /*
2526          * If the user explicitly requested a mode we should either
2527          * supply it, or fail completely (which we do below).  But if
2528          * it's not set explicitly, we reset our mode to something
2529          * that works
2530          */
2531         if (resize_hpt_err) {
2532             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2533             error_free(resize_hpt_err);
2534             resize_hpt_err = NULL;
2535         } else {
2536             spapr->resize_hpt = smc->resize_hpt_default;
2537         }
2538     }
2539 
2540     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2541 
2542     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2543         /*
2544          * User requested HPT resize, but this host can't supply it.  Bail out
2545          */
2546         error_report_err(resize_hpt_err);
2547         exit(1);
2548     }
2549 
2550     spapr->rma_size = node0_size;
2551 
2552     /* With KVM, we don't actually know whether KVM supports an
2553      * unbounded RMA (PR KVM) or is limited by the hash table size
2554      * (HV KVM using VRMA), so we always assume the latter
2555      *
2556      * In that case, we also limit the initial allocations for RTAS
2557      * etc... to 256M since we have no way to know what the VRMA size
2558      * is going to be as it depends on the size of the hash table
2559      * which isn't determined yet.
2560      */
2561     if (kvm_enabled()) {
2562         spapr->vrma_adjust = 1;
2563         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2564     }
2565 
2566     /* Actually we don't support unbounded RMA anymore since we added
2567      * proper emulation of HV mode. The max we can get is 16G which
2568      * also happens to be what we configure for PAPR mode so make sure
2569      * we don't do anything bigger than that
2570      */
2571     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2572 
2573     if (spapr->rma_size > node0_size) {
2574         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2575                      spapr->rma_size);
2576         exit(1);
2577     }
2578 
2579     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2580     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2581 
2582     /* Set up Interrupt Controller before we create the VCPUs */
2583     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2584 
2585     /* Set up containers for ibm,client-architecture-support negotiated options
2586      */
2587     spapr->ov5 = spapr_ovec_new();
2588     spapr->ov5_cas = spapr_ovec_new();
2589 
2590     if (smc->dr_lmb_enabled) {
2591         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2592         spapr_validate_node_memory(machine, &error_fatal);
2593     }
2594 
2595     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2596 
2597     /* advertise support for dedicated HP event source to guests */
2598     if (spapr->use_hotplug_event_source) {
2599         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2600     }
2601 
2602     /* advertise support for HPT resizing */
2603     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2604         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2605     }
2606 
2607     /* advertise support for ibm,dyamic-memory-v2 */
2608     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2609 
2610     /* init CPUs */
2611     spapr_init_cpus(spapr);
2612 
2613     first_ppc_cpu = POWERPC_CPU(first_cpu);
2614     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2615         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
2616                          spapr->max_compat_pvr)) {
2617         /* KVM and TCG always allow GTSE with radix... */
2618         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2619     }
2620     /* ... but not with hash (currently). */
2621 
2622     if (kvm_enabled()) {
2623         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2624         kvmppc_enable_logical_ci_hcalls();
2625         kvmppc_enable_set_mode_hcall();
2626 
2627         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2628         kvmppc_enable_clear_ref_mod_hcalls();
2629     }
2630 
2631     /* allocate RAM */
2632     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2633                                          machine->ram_size);
2634     memory_region_add_subregion(sysmem, 0, ram);
2635 
2636     /* initialize hotplug memory address space */
2637     if (machine->ram_size < machine->maxram_size) {
2638         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2639         /*
2640          * Limit the number of hotpluggable memory slots to half the number
2641          * slots that KVM supports, leaving the other half for PCI and other
2642          * devices. However ensure that number of slots doesn't drop below 32.
2643          */
2644         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2645                            SPAPR_MAX_RAM_SLOTS;
2646 
2647         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2648             max_memslots = SPAPR_MAX_RAM_SLOTS;
2649         }
2650         if (machine->ram_slots > max_memslots) {
2651             error_report("Specified number of memory slots %"
2652                          PRIu64" exceeds max supported %d",
2653                          machine->ram_slots, max_memslots);
2654             exit(1);
2655         }
2656 
2657         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2658                                               SPAPR_HOTPLUG_MEM_ALIGN);
2659         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2660                            "hotplug-memory", hotplug_mem_size);
2661         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2662                                     &spapr->hotplug_memory.mr);
2663     }
2664 
2665     if (smc->dr_lmb_enabled) {
2666         spapr_create_lmb_dr_connectors(spapr);
2667     }
2668 
2669     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2670     if (!filename) {
2671         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2672         exit(1);
2673     }
2674     spapr->rtas_size = get_image_size(filename);
2675     if (spapr->rtas_size < 0) {
2676         error_report("Could not get size of LPAR rtas '%s'", filename);
2677         exit(1);
2678     }
2679     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2680     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2681         error_report("Could not load LPAR rtas '%s'", filename);
2682         exit(1);
2683     }
2684     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2685         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2686                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2687         exit(1);
2688     }
2689     g_free(filename);
2690 
2691     /* Set up RTAS event infrastructure */
2692     spapr_events_init(spapr);
2693 
2694     /* Set up the RTC RTAS interfaces */
2695     spapr_rtc_create(spapr);
2696 
2697     /* Set up VIO bus */
2698     spapr->vio_bus = spapr_vio_bus_init();
2699 
2700     for (i = 0; i < serial_max_hds(); i++) {
2701         if (serial_hd(i)) {
2702             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2703         }
2704     }
2705 
2706     /* We always have at least the nvram device on VIO */
2707     spapr_create_nvram(spapr);
2708 
2709     /* Set up PCI */
2710     spapr_pci_rtas_init();
2711 
2712     phb = spapr_create_phb(spapr, 0);
2713 
2714     for (i = 0; i < nb_nics; i++) {
2715         NICInfo *nd = &nd_table[i];
2716 
2717         if (!nd->model) {
2718             nd->model = g_strdup("spapr-vlan");
2719         }
2720 
2721         if (g_str_equal(nd->model, "spapr-vlan") ||
2722             g_str_equal(nd->model, "ibmveth")) {
2723             spapr_vlan_create(spapr->vio_bus, nd);
2724         } else {
2725             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2726         }
2727     }
2728 
2729     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2730         spapr_vscsi_create(spapr->vio_bus);
2731     }
2732 
2733     /* Graphics */
2734     if (spapr_vga_init(phb->bus, &error_fatal)) {
2735         spapr->has_graphics = true;
2736         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2737     }
2738 
2739     if (machine->usb) {
2740         if (smc->use_ohci_by_default) {
2741             pci_create_simple(phb->bus, -1, "pci-ohci");
2742         } else {
2743             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2744         }
2745 
2746         if (spapr->has_graphics) {
2747             USBBus *usb_bus = usb_bus_find(-1);
2748 
2749             usb_create_simple(usb_bus, "usb-kbd");
2750             usb_create_simple(usb_bus, "usb-mouse");
2751         }
2752     }
2753 
2754     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2755         error_report(
2756             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2757             MIN_RMA_SLOF);
2758         exit(1);
2759     }
2760 
2761     if (kernel_filename) {
2762         uint64_t lowaddr = 0;
2763 
2764         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2765                                       NULL, NULL, &lowaddr, NULL, 1,
2766                                       PPC_ELF_MACHINE, 0, 0);
2767         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2768             spapr->kernel_size = load_elf(kernel_filename,
2769                                           translate_kernel_address, NULL, NULL,
2770                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2771                                           0, 0);
2772             spapr->kernel_le = spapr->kernel_size > 0;
2773         }
2774         if (spapr->kernel_size < 0) {
2775             error_report("error loading %s: %s", kernel_filename,
2776                          load_elf_strerror(spapr->kernel_size));
2777             exit(1);
2778         }
2779 
2780         /* load initrd */
2781         if (initrd_filename) {
2782             /* Try to locate the initrd in the gap between the kernel
2783              * and the firmware. Add a bit of space just in case
2784              */
2785             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2786                                   + 0x1ffff) & ~0xffff;
2787             spapr->initrd_size = load_image_targphys(initrd_filename,
2788                                                      spapr->initrd_base,
2789                                                      load_limit
2790                                                      - spapr->initrd_base);
2791             if (spapr->initrd_size < 0) {
2792                 error_report("could not load initial ram disk '%s'",
2793                              initrd_filename);
2794                 exit(1);
2795             }
2796         }
2797     }
2798 
2799     if (bios_name == NULL) {
2800         bios_name = FW_FILE_NAME;
2801     }
2802     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2803     if (!filename) {
2804         error_report("Could not find LPAR firmware '%s'", bios_name);
2805         exit(1);
2806     }
2807     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2808     if (fw_size <= 0) {
2809         error_report("Could not load LPAR firmware '%s'", filename);
2810         exit(1);
2811     }
2812     g_free(filename);
2813 
2814     /* FIXME: Should register things through the MachineState's qdev
2815      * interface, this is a legacy from the sPAPREnvironment structure
2816      * which predated MachineState but had a similar function */
2817     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2818     register_savevm_live(NULL, "spapr/htab", -1, 1,
2819                          &savevm_htab_handlers, spapr);
2820 
2821     qemu_register_boot_set(spapr_boot_set, spapr);
2822 
2823     if (kvm_enabled()) {
2824         /* to stop and start vmclock */
2825         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2826                                          &spapr->tb);
2827 
2828         kvmppc_spapr_enable_inkernel_multitce();
2829     }
2830 }
2831 
2832 static int spapr_kvm_type(const char *vm_type)
2833 {
2834     if (!vm_type) {
2835         return 0;
2836     }
2837 
2838     if (!strcmp(vm_type, "HV")) {
2839         return 1;
2840     }
2841 
2842     if (!strcmp(vm_type, "PR")) {
2843         return 2;
2844     }
2845 
2846     error_report("Unknown kvm-type specified '%s'", vm_type);
2847     exit(1);
2848 }
2849 
2850 /*
2851  * Implementation of an interface to adjust firmware path
2852  * for the bootindex property handling.
2853  */
2854 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2855                                    DeviceState *dev)
2856 {
2857 #define CAST(type, obj, name) \
2858     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2859     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2860     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2861     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2862 
2863     if (d) {
2864         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2865         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2866         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2867 
2868         if (spapr) {
2869             /*
2870              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2871              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2872              * in the top 16 bits of the 64-bit LUN
2873              */
2874             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2875             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2876                                    (uint64_t)id << 48);
2877         } else if (virtio) {
2878             /*
2879              * We use SRP luns of the form 01000000 | (target << 8) | lun
2880              * in the top 32 bits of the 64-bit LUN
2881              * Note: the quote above is from SLOF and it is wrong,
2882              * the actual binding is:
2883              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2884              */
2885             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2886             if (d->lun >= 256) {
2887                 /* Use the LUN "flat space addressing method" */
2888                 id |= 0x4000;
2889             }
2890             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2891                                    (uint64_t)id << 32);
2892         } else if (usb) {
2893             /*
2894              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2895              * in the top 32 bits of the 64-bit LUN
2896              */
2897             unsigned usb_port = atoi(usb->port->path);
2898             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2899             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2900                                    (uint64_t)id << 32);
2901         }
2902     }
2903 
2904     /*
2905      * SLOF probes the USB devices, and if it recognizes that the device is a
2906      * storage device, it changes its name to "storage" instead of "usb-host",
2907      * and additionally adds a child node for the SCSI LUN, so the correct
2908      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2909      */
2910     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2911         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2912         if (usb_host_dev_is_scsi_storage(usbdev)) {
2913             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2914         }
2915     }
2916 
2917     if (phb) {
2918         /* Replace "pci" with "pci@800000020000000" */
2919         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2920     }
2921 
2922     if (vsc) {
2923         /* Same logic as virtio above */
2924         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2925         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2926     }
2927 
2928     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2929         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2930         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2931         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2932     }
2933 
2934     return NULL;
2935 }
2936 
2937 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2938 {
2939     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2940 
2941     return g_strdup(spapr->kvm_type);
2942 }
2943 
2944 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2945 {
2946     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2947 
2948     g_free(spapr->kvm_type);
2949     spapr->kvm_type = g_strdup(value);
2950 }
2951 
2952 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2953 {
2954     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2955 
2956     return spapr->use_hotplug_event_source;
2957 }
2958 
2959 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2960                                             Error **errp)
2961 {
2962     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2963 
2964     spapr->use_hotplug_event_source = value;
2965 }
2966 
2967 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2968 {
2969     return true;
2970 }
2971 
2972 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2973 {
2974     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2975 
2976     switch (spapr->resize_hpt) {
2977     case SPAPR_RESIZE_HPT_DEFAULT:
2978         return g_strdup("default");
2979     case SPAPR_RESIZE_HPT_DISABLED:
2980         return g_strdup("disabled");
2981     case SPAPR_RESIZE_HPT_ENABLED:
2982         return g_strdup("enabled");
2983     case SPAPR_RESIZE_HPT_REQUIRED:
2984         return g_strdup("required");
2985     }
2986     g_assert_not_reached();
2987 }
2988 
2989 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2990 {
2991     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2992 
2993     if (strcmp(value, "default") == 0) {
2994         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2995     } else if (strcmp(value, "disabled") == 0) {
2996         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2997     } else if (strcmp(value, "enabled") == 0) {
2998         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2999     } else if (strcmp(value, "required") == 0) {
3000         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3001     } else {
3002         error_setg(errp, "Bad value for \"resize-hpt\" property");
3003     }
3004 }
3005 
3006 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3007                                    void *opaque, Error **errp)
3008 {
3009     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3010 }
3011 
3012 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3013                                    void *opaque, Error **errp)
3014 {
3015     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3016 }
3017 
3018 static void spapr_instance_init(Object *obj)
3019 {
3020     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3021 
3022     spapr->htab_fd = -1;
3023     spapr->use_hotplug_event_source = true;
3024     object_property_add_str(obj, "kvm-type",
3025                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3026     object_property_set_description(obj, "kvm-type",
3027                                     "Specifies the KVM virtualization mode (HV, PR)",
3028                                     NULL);
3029     object_property_add_bool(obj, "modern-hotplug-events",
3030                             spapr_get_modern_hotplug_events,
3031                             spapr_set_modern_hotplug_events,
3032                             NULL);
3033     object_property_set_description(obj, "modern-hotplug-events",
3034                                     "Use dedicated hotplug event mechanism in"
3035                                     " place of standard EPOW events when possible"
3036                                     " (required for memory hot-unplug support)",
3037                                     NULL);
3038     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3039                             "Maximum permitted CPU compatibility mode",
3040                             &error_fatal);
3041 
3042     object_property_add_str(obj, "resize-hpt",
3043                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3044     object_property_set_description(obj, "resize-hpt",
3045                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3046                                     NULL);
3047     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3048                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3049     object_property_set_description(obj, "vsmt",
3050                                     "Virtual SMT: KVM behaves as if this were"
3051                                     " the host's SMT mode", &error_abort);
3052     object_property_add_bool(obj, "vfio-no-msix-emulation",
3053                              spapr_get_msix_emulation, NULL, NULL);
3054 }
3055 
3056 static void spapr_machine_finalizefn(Object *obj)
3057 {
3058     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3059 
3060     g_free(spapr->kvm_type);
3061 }
3062 
3063 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3064 {
3065     cpu_synchronize_state(cs);
3066     ppc_cpu_do_system_reset(cs);
3067 }
3068 
3069 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3070 {
3071     CPUState *cs;
3072 
3073     CPU_FOREACH(cs) {
3074         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3075     }
3076 }
3077 
3078 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3079                            uint32_t node, bool dedicated_hp_event_source,
3080                            Error **errp)
3081 {
3082     sPAPRDRConnector *drc;
3083     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3084     int i, fdt_offset, fdt_size;
3085     void *fdt;
3086     uint64_t addr = addr_start;
3087     bool hotplugged = spapr_drc_hotplugged(dev);
3088     Error *local_err = NULL;
3089 
3090     for (i = 0; i < nr_lmbs; i++) {
3091         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3092                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3093         g_assert(drc);
3094 
3095         fdt = create_device_tree(&fdt_size);
3096         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3097                                                 SPAPR_MEMORY_BLOCK_SIZE);
3098 
3099         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3100         if (local_err) {
3101             while (addr > addr_start) {
3102                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3103                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3104                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3105                 spapr_drc_detach(drc);
3106             }
3107             g_free(fdt);
3108             error_propagate(errp, local_err);
3109             return;
3110         }
3111         if (!hotplugged) {
3112             spapr_drc_reset(drc);
3113         }
3114         addr += SPAPR_MEMORY_BLOCK_SIZE;
3115     }
3116     /* send hotplug notification to the
3117      * guest only in case of hotplugged memory
3118      */
3119     if (hotplugged) {
3120         if (dedicated_hp_event_source) {
3121             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3122                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3123             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3124                                                    nr_lmbs,
3125                                                    spapr_drc_index(drc));
3126         } else {
3127             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3128                                            nr_lmbs);
3129         }
3130     }
3131 }
3132 
3133 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3134                               uint32_t node, Error **errp)
3135 {
3136     Error *local_err = NULL;
3137     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3138     PCDIMMDevice *dimm = PC_DIMM(dev);
3139     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3140     MemoryRegion *mr;
3141     uint64_t align, size, addr;
3142 
3143     mr = ddc->get_memory_region(dimm, &local_err);
3144     if (local_err) {
3145         goto out;
3146     }
3147     align = memory_region_get_alignment(mr);
3148     size = memory_region_size(mr);
3149 
3150     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
3151     if (local_err) {
3152         goto out;
3153     }
3154 
3155     addr = object_property_get_uint(OBJECT(dimm),
3156                                     PC_DIMM_ADDR_PROP, &local_err);
3157     if (local_err) {
3158         goto out_unplug;
3159     }
3160 
3161     spapr_add_lmbs(dev, addr, size, node,
3162                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3163                    &local_err);
3164     if (local_err) {
3165         goto out_unplug;
3166     }
3167 
3168     return;
3169 
3170 out_unplug:
3171     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
3172 out:
3173     error_propagate(errp, local_err);
3174 }
3175 
3176 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3177                                   Error **errp)
3178 {
3179     PCDIMMDevice *dimm = PC_DIMM(dev);
3180     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3181     MemoryRegion *mr;
3182     uint64_t size;
3183     char *mem_dev;
3184 
3185     mr = ddc->get_memory_region(dimm, errp);
3186     if (!mr) {
3187         return;
3188     }
3189     size = memory_region_size(mr);
3190 
3191     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3192         error_setg(errp, "Hotplugged memory size must be a multiple of "
3193                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
3194         return;
3195     }
3196 
3197     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
3198     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
3199         error_setg(errp, "Memory backend has bad page size. "
3200                    "Use 'memory-backend-file' with correct mem-path.");
3201         goto out;
3202     }
3203 
3204 out:
3205     g_free(mem_dev);
3206 }
3207 
3208 struct sPAPRDIMMState {
3209     PCDIMMDevice *dimm;
3210     uint32_t nr_lmbs;
3211     QTAILQ_ENTRY(sPAPRDIMMState) next;
3212 };
3213 
3214 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3215                                                        PCDIMMDevice *dimm)
3216 {
3217     sPAPRDIMMState *dimm_state = NULL;
3218 
3219     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3220         if (dimm_state->dimm == dimm) {
3221             break;
3222         }
3223     }
3224     return dimm_state;
3225 }
3226 
3227 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3228                                                       uint32_t nr_lmbs,
3229                                                       PCDIMMDevice *dimm)
3230 {
3231     sPAPRDIMMState *ds = NULL;
3232 
3233     /*
3234      * If this request is for a DIMM whose removal had failed earlier
3235      * (due to guest's refusal to remove the LMBs), we would have this
3236      * dimm already in the pending_dimm_unplugs list. In that
3237      * case don't add again.
3238      */
3239     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3240     if (!ds) {
3241         ds = g_malloc0(sizeof(sPAPRDIMMState));
3242         ds->nr_lmbs = nr_lmbs;
3243         ds->dimm = dimm;
3244         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3245     }
3246     return ds;
3247 }
3248 
3249 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3250                                               sPAPRDIMMState *dimm_state)
3251 {
3252     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3253     g_free(dimm_state);
3254 }
3255 
3256 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3257                                                         PCDIMMDevice *dimm)
3258 {
3259     sPAPRDRConnector *drc;
3260     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3261     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3262     uint64_t size = memory_region_size(mr);
3263     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3264     uint32_t avail_lmbs = 0;
3265     uint64_t addr_start, addr;
3266     int i;
3267 
3268     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3269                                          &error_abort);
3270 
3271     addr = addr_start;
3272     for (i = 0; i < nr_lmbs; i++) {
3273         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3274                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3275         g_assert(drc);
3276         if (drc->dev) {
3277             avail_lmbs++;
3278         }
3279         addr += SPAPR_MEMORY_BLOCK_SIZE;
3280     }
3281 
3282     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3283 }
3284 
3285 /* Callback to be called during DRC release. */
3286 void spapr_lmb_release(DeviceState *dev)
3287 {
3288     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_hotplug_handler(dev));
3289     PCDIMMDevice *dimm = PC_DIMM(dev);
3290     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3291     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3292     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3293 
3294     /* This information will get lost if a migration occurs
3295      * during the unplug process. In this case recover it. */
3296     if (ds == NULL) {
3297         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3298         g_assert(ds);
3299         /* The DRC being examined by the caller at least must be counted */
3300         g_assert(ds->nr_lmbs);
3301     }
3302 
3303     if (--ds->nr_lmbs) {
3304         return;
3305     }
3306 
3307     /*
3308      * Now that all the LMBs have been removed by the guest, call the
3309      * pc-dimm unplug handler to cleanup up the pc-dimm device.
3310      */
3311     pc_dimm_memory_unplug(dev, &spapr->hotplug_memory, mr);
3312     object_unparent(OBJECT(dev));
3313     spapr_pending_dimm_unplugs_remove(spapr, ds);
3314 }
3315 
3316 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3317                                         DeviceState *dev, Error **errp)
3318 {
3319     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3320     Error *local_err = NULL;
3321     PCDIMMDevice *dimm = PC_DIMM(dev);
3322     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3323     MemoryRegion *mr;
3324     uint32_t nr_lmbs;
3325     uint64_t size, addr_start, addr;
3326     int i;
3327     sPAPRDRConnector *drc;
3328 
3329     mr = ddc->get_memory_region(dimm, &local_err);
3330     if (local_err) {
3331         goto out;
3332     }
3333     size = memory_region_size(mr);
3334     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3335 
3336     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3337                                          &local_err);
3338     if (local_err) {
3339         goto out;
3340     }
3341 
3342     /*
3343      * An existing pending dimm state for this DIMM means that there is an
3344      * unplug operation in progress, waiting for the spapr_lmb_release
3345      * callback to complete the job (BQL can't cover that far). In this case,
3346      * bail out to avoid detaching DRCs that were already released.
3347      */
3348     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3349         error_setg(&local_err,
3350                    "Memory unplug already in progress for device %s",
3351                    dev->id);
3352         goto out;
3353     }
3354 
3355     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3356 
3357     addr = addr_start;
3358     for (i = 0; i < nr_lmbs; i++) {
3359         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3360                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3361         g_assert(drc);
3362 
3363         spapr_drc_detach(drc);
3364         addr += SPAPR_MEMORY_BLOCK_SIZE;
3365     }
3366 
3367     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3368                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3369     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3370                                               nr_lmbs, spapr_drc_index(drc));
3371 out:
3372     error_propagate(errp, local_err);
3373 }
3374 
3375 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3376                                            sPAPRMachineState *spapr)
3377 {
3378     PowerPCCPU *cpu = POWERPC_CPU(cs);
3379     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3380     int id = spapr_get_vcpu_id(cpu);
3381     void *fdt;
3382     int offset, fdt_size;
3383     char *nodename;
3384 
3385     fdt = create_device_tree(&fdt_size);
3386     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3387     offset = fdt_add_subnode(fdt, 0, nodename);
3388 
3389     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3390     g_free(nodename);
3391 
3392     *fdt_offset = offset;
3393     return fdt;
3394 }
3395 
3396 /* Callback to be called during DRC release. */
3397 void spapr_core_release(DeviceState *dev)
3398 {
3399     MachineState *ms = MACHINE(qdev_get_hotplug_handler(dev));
3400     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3401     CPUCore *cc = CPU_CORE(dev);
3402     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3403 
3404     if (smc->pre_2_10_has_unused_icps) {
3405         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3406         int i;
3407 
3408         for (i = 0; i < cc->nr_threads; i++) {
3409             CPUState *cs = CPU(sc->threads[i]);
3410 
3411             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3412         }
3413     }
3414 
3415     assert(core_slot);
3416     core_slot->cpu = NULL;
3417     object_unparent(OBJECT(dev));
3418 }
3419 
3420 static
3421 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3422                                Error **errp)
3423 {
3424     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3425     int index;
3426     sPAPRDRConnector *drc;
3427     CPUCore *cc = CPU_CORE(dev);
3428 
3429     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3430         error_setg(errp, "Unable to find CPU core with core-id: %d",
3431                    cc->core_id);
3432         return;
3433     }
3434     if (index == 0) {
3435         error_setg(errp, "Boot CPU core may not be unplugged");
3436         return;
3437     }
3438 
3439     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3440                           spapr_vcpu_id(spapr, cc->core_id));
3441     g_assert(drc);
3442 
3443     spapr_drc_detach(drc);
3444 
3445     spapr_hotplug_req_remove_by_index(drc);
3446 }
3447 
3448 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3449                             Error **errp)
3450 {
3451     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3452     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3453     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3454     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3455     CPUCore *cc = CPU_CORE(dev);
3456     CPUState *cs = CPU(core->threads[0]);
3457     sPAPRDRConnector *drc;
3458     Error *local_err = NULL;
3459     CPUArchId *core_slot;
3460     int index;
3461     bool hotplugged = spapr_drc_hotplugged(dev);
3462 
3463     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3464     if (!core_slot) {
3465         error_setg(errp, "Unable to find CPU core with core-id: %d",
3466                    cc->core_id);
3467         return;
3468     }
3469     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3470                           spapr_vcpu_id(spapr, cc->core_id));
3471 
3472     g_assert(drc || !mc->has_hotpluggable_cpus);
3473 
3474     if (drc) {
3475         void *fdt;
3476         int fdt_offset;
3477 
3478         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3479 
3480         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3481         if (local_err) {
3482             g_free(fdt);
3483             error_propagate(errp, local_err);
3484             return;
3485         }
3486 
3487         if (hotplugged) {
3488             /*
3489              * Send hotplug notification interrupt to the guest only
3490              * in case of hotplugged CPUs.
3491              */
3492             spapr_hotplug_req_add_by_index(drc);
3493         } else {
3494             spapr_drc_reset(drc);
3495         }
3496     }
3497 
3498     core_slot->cpu = OBJECT(dev);
3499 
3500     if (smc->pre_2_10_has_unused_icps) {
3501         int i;
3502 
3503         for (i = 0; i < cc->nr_threads; i++) {
3504             cs = CPU(core->threads[i]);
3505             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3506         }
3507     }
3508 }
3509 
3510 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3511                                 Error **errp)
3512 {
3513     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3514     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3515     Error *local_err = NULL;
3516     CPUCore *cc = CPU_CORE(dev);
3517     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3518     const char *type = object_get_typename(OBJECT(dev));
3519     CPUArchId *core_slot;
3520     int index;
3521 
3522     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3523         error_setg(&local_err, "CPU hotplug not supported for this machine");
3524         goto out;
3525     }
3526 
3527     if (strcmp(base_core_type, type)) {
3528         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3529         goto out;
3530     }
3531 
3532     if (cc->core_id % smp_threads) {
3533         error_setg(&local_err, "invalid core id %d", cc->core_id);
3534         goto out;
3535     }
3536 
3537     /*
3538      * In general we should have homogeneous threads-per-core, but old
3539      * (pre hotplug support) machine types allow the last core to have
3540      * reduced threads as a compatibility hack for when we allowed
3541      * total vcpus not a multiple of threads-per-core.
3542      */
3543     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3544         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3545                    cc->nr_threads, smp_threads);
3546         goto out;
3547     }
3548 
3549     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3550     if (!core_slot) {
3551         error_setg(&local_err, "core id %d out of range", cc->core_id);
3552         goto out;
3553     }
3554 
3555     if (core_slot->cpu) {
3556         error_setg(&local_err, "core %d already populated", cc->core_id);
3557         goto out;
3558     }
3559 
3560     numa_cpu_pre_plug(core_slot, dev, &local_err);
3561 
3562 out:
3563     error_propagate(errp, local_err);
3564 }
3565 
3566 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3567                                       DeviceState *dev, Error **errp)
3568 {
3569     MachineState *ms = MACHINE(hotplug_dev);
3570     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3571 
3572     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3573         int node;
3574 
3575         if (!smc->dr_lmb_enabled) {
3576             error_setg(errp, "Memory hotplug not supported for this machine");
3577             return;
3578         }
3579         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3580         if (*errp) {
3581             return;
3582         }
3583         if (node < 0 || node >= MAX_NODES) {
3584             error_setg(errp, "Invaild node %d", node);
3585             return;
3586         }
3587 
3588         spapr_memory_plug(hotplug_dev, dev, node, errp);
3589     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3590         spapr_core_plug(hotplug_dev, dev, errp);
3591     }
3592 }
3593 
3594 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3595                                                 DeviceState *dev, Error **errp)
3596 {
3597     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3598     MachineClass *mc = MACHINE_GET_CLASS(sms);
3599 
3600     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3601         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3602             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3603         } else {
3604             /* NOTE: this means there is a window after guest reset, prior to
3605              * CAS negotiation, where unplug requests will fail due to the
3606              * capability not being detected yet. This is a bit different than
3607              * the case with PCI unplug, where the events will be queued and
3608              * eventually handled by the guest after boot
3609              */
3610             error_setg(errp, "Memory hot unplug not supported for this guest");
3611         }
3612     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3613         if (!mc->has_hotpluggable_cpus) {
3614             error_setg(errp, "CPU hot unplug not supported on this machine");
3615             return;
3616         }
3617         spapr_core_unplug_request(hotplug_dev, dev, errp);
3618     }
3619 }
3620 
3621 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3622                                           DeviceState *dev, Error **errp)
3623 {
3624     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3625         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3626     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3627         spapr_core_pre_plug(hotplug_dev, dev, errp);
3628     }
3629 }
3630 
3631 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3632                                                  DeviceState *dev)
3633 {
3634     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3635         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3636         return HOTPLUG_HANDLER(machine);
3637     }
3638     return NULL;
3639 }
3640 
3641 static CpuInstanceProperties
3642 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3643 {
3644     CPUArchId *core_slot;
3645     MachineClass *mc = MACHINE_GET_CLASS(machine);
3646 
3647     /* make sure possible_cpu are intialized */
3648     mc->possible_cpu_arch_ids(machine);
3649     /* get CPU core slot containing thread that matches cpu_index */
3650     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3651     assert(core_slot);
3652     return core_slot->props;
3653 }
3654 
3655 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3656 {
3657     return idx / smp_cores % nb_numa_nodes;
3658 }
3659 
3660 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3661 {
3662     int i;
3663     const char *core_type;
3664     int spapr_max_cores = max_cpus / smp_threads;
3665     MachineClass *mc = MACHINE_GET_CLASS(machine);
3666 
3667     if (!mc->has_hotpluggable_cpus) {
3668         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3669     }
3670     if (machine->possible_cpus) {
3671         assert(machine->possible_cpus->len == spapr_max_cores);
3672         return machine->possible_cpus;
3673     }
3674 
3675     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3676     if (!core_type) {
3677         error_report("Unable to find sPAPR CPU Core definition");
3678         exit(1);
3679     }
3680 
3681     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3682                              sizeof(CPUArchId) * spapr_max_cores);
3683     machine->possible_cpus->len = spapr_max_cores;
3684     for (i = 0; i < machine->possible_cpus->len; i++) {
3685         int core_id = i * smp_threads;
3686 
3687         machine->possible_cpus->cpus[i].type = core_type;
3688         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3689         machine->possible_cpus->cpus[i].arch_id = core_id;
3690         machine->possible_cpus->cpus[i].props.has_core_id = true;
3691         machine->possible_cpus->cpus[i].props.core_id = core_id;
3692     }
3693     return machine->possible_cpus;
3694 }
3695 
3696 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3697                                 uint64_t *buid, hwaddr *pio,
3698                                 hwaddr *mmio32, hwaddr *mmio64,
3699                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3700 {
3701     /*
3702      * New-style PHB window placement.
3703      *
3704      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3705      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3706      * windows.
3707      *
3708      * Some guest kernels can't work with MMIO windows above 1<<46
3709      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3710      *
3711      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3712      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3713      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3714      * 1TiB 64-bit MMIO windows for each PHB.
3715      */
3716     const uint64_t base_buid = 0x800000020000000ULL;
3717 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3718                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3719     int i;
3720 
3721     /* Sanity check natural alignments */
3722     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3723     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3724     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3725     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3726     /* Sanity check bounds */
3727     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3728                       SPAPR_PCI_MEM32_WIN_SIZE);
3729     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3730                       SPAPR_PCI_MEM64_WIN_SIZE);
3731 
3732     if (index >= SPAPR_MAX_PHBS) {
3733         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3734                    SPAPR_MAX_PHBS - 1);
3735         return;
3736     }
3737 
3738     *buid = base_buid + index;
3739     for (i = 0; i < n_dma; ++i) {
3740         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3741     }
3742 
3743     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3744     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3745     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3746 }
3747 
3748 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3749 {
3750     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3751 
3752     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3753 }
3754 
3755 static void spapr_ics_resend(XICSFabric *dev)
3756 {
3757     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3758 
3759     ics_resend(spapr->ics);
3760 }
3761 
3762 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3763 {
3764     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3765 
3766     return cpu ? ICP(cpu->intc) : NULL;
3767 }
3768 
3769 #define ICS_IRQ_FREE(ics, srcno)   \
3770     (!((ics)->irqs[(srcno)].flags & (XICS_FLAGS_IRQ_MASK)))
3771 
3772 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
3773 {
3774     int first, i;
3775 
3776     for (first = 0; first < ics->nr_irqs; first += alignnum) {
3777         if (num > (ics->nr_irqs - first)) {
3778             return -1;
3779         }
3780         for (i = first; i < first + num; ++i) {
3781             if (!ICS_IRQ_FREE(ics, i)) {
3782                 break;
3783             }
3784         }
3785         if (i == (first + num)) {
3786             return first;
3787         }
3788     }
3789 
3790     return -1;
3791 }
3792 
3793 /*
3794  * Allocate the IRQ number and set the IRQ type, LSI or MSI
3795  */
3796 static void spapr_irq_set_lsi(sPAPRMachineState *spapr, int irq, bool lsi)
3797 {
3798     ics_set_irq_type(spapr->ics, irq - spapr->ics->offset, lsi);
3799 }
3800 
3801 int spapr_irq_alloc(sPAPRMachineState *spapr, int irq_hint, bool lsi,
3802                     Error **errp)
3803 {
3804     ICSState *ics = spapr->ics;
3805     int irq;
3806 
3807     assert(ics);
3808 
3809     if (irq_hint) {
3810         if (!ICS_IRQ_FREE(ics, irq_hint - ics->offset)) {
3811             error_setg(errp, "can't allocate IRQ %d: already in use", irq_hint);
3812             return -1;
3813         }
3814         irq = irq_hint;
3815     } else {
3816         irq = ics_find_free_block(ics, 1, 1);
3817         if (irq < 0) {
3818             error_setg(errp, "can't allocate IRQ: no IRQ left");
3819             return -1;
3820         }
3821         irq += ics->offset;
3822     }
3823 
3824     spapr_irq_set_lsi(spapr, irq, lsi);
3825     trace_spapr_irq_alloc(irq);
3826 
3827     return irq;
3828 }
3829 
3830 /*
3831  * Allocate block of consecutive IRQs, and return the number of the first IRQ in
3832  * the block. If align==true, aligns the first IRQ number to num.
3833  */
3834 int spapr_irq_alloc_block(sPAPRMachineState *spapr, int num, bool lsi,
3835                           bool align, Error **errp)
3836 {
3837     ICSState *ics = spapr->ics;
3838     int i, first = -1;
3839 
3840     assert(ics);
3841 
3842     /*
3843      * MSIMesage::data is used for storing VIRQ so
3844      * it has to be aligned to num to support multiple
3845      * MSI vectors. MSI-X is not affected by this.
3846      * The hint is used for the first IRQ, the rest should
3847      * be allocated continuously.
3848      */
3849     if (align) {
3850         assert((num == 1) || (num == 2) || (num == 4) ||
3851                (num == 8) || (num == 16) || (num == 32));
3852         first = ics_find_free_block(ics, num, num);
3853     } else {
3854         first = ics_find_free_block(ics, num, 1);
3855     }
3856     if (first < 0) {
3857         error_setg(errp, "can't find a free %d-IRQ block", num);
3858         return -1;
3859     }
3860 
3861     first += ics->offset;
3862     for (i = first; i < first + num; ++i) {
3863         spapr_irq_set_lsi(spapr, i, lsi);
3864     }
3865 
3866     trace_spapr_irq_alloc_block(first, num, lsi, align);
3867 
3868     return first;
3869 }
3870 
3871 void spapr_irq_free(sPAPRMachineState *spapr, int irq, int num)
3872 {
3873     ICSState *ics = spapr->ics;
3874     int srcno = irq - ics->offset;
3875     int i;
3876 
3877     if (ics_valid_irq(ics, irq)) {
3878         trace_spapr_irq_free(0, irq, num);
3879         for (i = srcno; i < srcno + num; ++i) {
3880             if (ICS_IRQ_FREE(ics, i)) {
3881                 trace_spapr_irq_free_warn(0, i + ics->offset);
3882             }
3883             memset(&ics->irqs[i], 0, sizeof(ICSIRQState));
3884         }
3885     }
3886 }
3887 
3888 qemu_irq spapr_qirq(sPAPRMachineState *spapr, int irq)
3889 {
3890     ICSState *ics = spapr->ics;
3891 
3892     if (ics_valid_irq(ics, irq)) {
3893         return ics->qirqs[irq - ics->offset];
3894     }
3895 
3896     return NULL;
3897 }
3898 
3899 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3900                                  Monitor *mon)
3901 {
3902     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3903     CPUState *cs;
3904 
3905     CPU_FOREACH(cs) {
3906         PowerPCCPU *cpu = POWERPC_CPU(cs);
3907 
3908         icp_pic_print_info(ICP(cpu->intc), mon);
3909     }
3910 
3911     ics_pic_print_info(spapr->ics, mon);
3912 }
3913 
3914 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3915 {
3916     return cpu->vcpu_id;
3917 }
3918 
3919 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3920 {
3921     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3922     int vcpu_id;
3923 
3924     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3925 
3926     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3927         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3928         error_append_hint(errp, "Adjust the number of cpus to %d "
3929                           "or try to raise the number of threads per core\n",
3930                           vcpu_id * smp_threads / spapr->vsmt);
3931         return;
3932     }
3933 
3934     cpu->vcpu_id = vcpu_id;
3935 }
3936 
3937 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3938 {
3939     CPUState *cs;
3940 
3941     CPU_FOREACH(cs) {
3942         PowerPCCPU *cpu = POWERPC_CPU(cs);
3943 
3944         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3945             return cpu;
3946         }
3947     }
3948 
3949     return NULL;
3950 }
3951 
3952 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3953 {
3954     MachineClass *mc = MACHINE_CLASS(oc);
3955     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3956     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3957     NMIClass *nc = NMI_CLASS(oc);
3958     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3959     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3960     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3961     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3962 
3963     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3964 
3965     /*
3966      * We set up the default / latest behaviour here.  The class_init
3967      * functions for the specific versioned machine types can override
3968      * these details for backwards compatibility
3969      */
3970     mc->init = spapr_machine_init;
3971     mc->reset = spapr_machine_reset;
3972     mc->block_default_type = IF_SCSI;
3973     mc->max_cpus = 1024;
3974     mc->no_parallel = 1;
3975     mc->default_boot_order = "";
3976     mc->default_ram_size = 512 * M_BYTE;
3977     mc->kvm_type = spapr_kvm_type;
3978     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3979     mc->pci_allow_0_address = true;
3980     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3981     hc->pre_plug = spapr_machine_device_pre_plug;
3982     hc->plug = spapr_machine_device_plug;
3983     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3984     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3985     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3986     hc->unplug_request = spapr_machine_device_unplug_request;
3987 
3988     smc->dr_lmb_enabled = true;
3989     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3990     mc->has_hotpluggable_cpus = true;
3991     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3992     fwc->get_dev_path = spapr_get_fw_dev_path;
3993     nc->nmi_monitor_handler = spapr_nmi;
3994     smc->phb_placement = spapr_phb_placement;
3995     vhc->hypercall = emulate_spapr_hypercall;
3996     vhc->hpt_mask = spapr_hpt_mask;
3997     vhc->map_hptes = spapr_map_hptes;
3998     vhc->unmap_hptes = spapr_unmap_hptes;
3999     vhc->store_hpte = spapr_store_hpte;
4000     vhc->get_patbe = spapr_get_patbe;
4001     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4002     xic->ics_get = spapr_ics_get;
4003     xic->ics_resend = spapr_ics_resend;
4004     xic->icp_get = spapr_icp_get;
4005     ispc->print_info = spapr_pic_print_info;
4006     /* Force NUMA node memory size to be a multiple of
4007      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4008      * in which LMBs are represented and hot-added
4009      */
4010     mc->numa_mem_align_shift = 28;
4011 
4012     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4013     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4014     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4015     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4016     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4017     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4018     spapr_caps_add_properties(smc, &error_abort);
4019 }
4020 
4021 static const TypeInfo spapr_machine_info = {
4022     .name          = TYPE_SPAPR_MACHINE,
4023     .parent        = TYPE_MACHINE,
4024     .abstract      = true,
4025     .instance_size = sizeof(sPAPRMachineState),
4026     .instance_init = spapr_instance_init,
4027     .instance_finalize = spapr_machine_finalizefn,
4028     .class_size    = sizeof(sPAPRMachineClass),
4029     .class_init    = spapr_machine_class_init,
4030     .interfaces = (InterfaceInfo[]) {
4031         { TYPE_FW_PATH_PROVIDER },
4032         { TYPE_NMI },
4033         { TYPE_HOTPLUG_HANDLER },
4034         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4035         { TYPE_XICS_FABRIC },
4036         { TYPE_INTERRUPT_STATS_PROVIDER },
4037         { }
4038     },
4039 };
4040 
4041 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4042     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4043                                                     void *data)      \
4044     {                                                                \
4045         MachineClass *mc = MACHINE_CLASS(oc);                        \
4046         spapr_machine_##suffix##_class_options(mc);                  \
4047         if (latest) {                                                \
4048             mc->alias = "pseries";                                   \
4049             mc->is_default = 1;                                      \
4050         }                                                            \
4051     }                                                                \
4052     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
4053     {                                                                \
4054         MachineState *machine = MACHINE(obj);                        \
4055         spapr_machine_##suffix##_instance_options(machine);          \
4056     }                                                                \
4057     static const TypeInfo spapr_machine_##suffix##_info = {          \
4058         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4059         .parent = TYPE_SPAPR_MACHINE,                                \
4060         .class_init = spapr_machine_##suffix##_class_init,           \
4061         .instance_init = spapr_machine_##suffix##_instance_init,     \
4062     };                                                               \
4063     static void spapr_machine_register_##suffix(void)                \
4064     {                                                                \
4065         type_register(&spapr_machine_##suffix##_info);               \
4066     }                                                                \
4067     type_init(spapr_machine_register_##suffix)
4068 
4069 /*
4070  * pseries-2.13
4071  */
4072 static void spapr_machine_2_13_instance_options(MachineState *machine)
4073 {
4074 }
4075 
4076 static void spapr_machine_2_13_class_options(MachineClass *mc)
4077 {
4078     /* Defaults for the latest behaviour inherited from the base class */
4079 }
4080 
4081 DEFINE_SPAPR_MACHINE(2_13, "2.13", true);
4082 
4083 /*
4084  * pseries-2.12
4085  */
4086 #define SPAPR_COMPAT_2_12                                              \
4087     HW_COMPAT_2_12                                                     \
4088     {                                                                  \
4089         .driver = TYPE_POWERPC_CPU,                                    \
4090         .property = "pre-2.13-migration",                              \
4091         .value    = "on",                                              \
4092     },
4093 
4094 static void spapr_machine_2_12_instance_options(MachineState *machine)
4095 {
4096     spapr_machine_2_13_instance_options(machine);
4097 }
4098 
4099 static void spapr_machine_2_12_class_options(MachineClass *mc)
4100 {
4101     spapr_machine_2_13_class_options(mc);
4102     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4103 }
4104 
4105 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4106 
4107 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
4108 {
4109     spapr_machine_2_12_instance_options(machine);
4110 }
4111 
4112 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4113 {
4114     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4115 
4116     spapr_machine_2_12_class_options(mc);
4117     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4118     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4119     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4120 }
4121 
4122 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4123 
4124 /*
4125  * pseries-2.11
4126  */
4127 #define SPAPR_COMPAT_2_11                                              \
4128     HW_COMPAT_2_11
4129 
4130 static void spapr_machine_2_11_instance_options(MachineState *machine)
4131 {
4132     spapr_machine_2_12_instance_options(machine);
4133 }
4134 
4135 static void spapr_machine_2_11_class_options(MachineClass *mc)
4136 {
4137     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4138 
4139     spapr_machine_2_12_class_options(mc);
4140     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4141     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4142 }
4143 
4144 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4145 
4146 /*
4147  * pseries-2.10
4148  */
4149 #define SPAPR_COMPAT_2_10                                              \
4150     HW_COMPAT_2_10
4151 
4152 static void spapr_machine_2_10_instance_options(MachineState *machine)
4153 {
4154     spapr_machine_2_11_instance_options(machine);
4155 }
4156 
4157 static void spapr_machine_2_10_class_options(MachineClass *mc)
4158 {
4159     spapr_machine_2_11_class_options(mc);
4160     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4161 }
4162 
4163 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4164 
4165 /*
4166  * pseries-2.9
4167  */
4168 #define SPAPR_COMPAT_2_9                                               \
4169     HW_COMPAT_2_9                                                      \
4170     {                                                                  \
4171         .driver = TYPE_POWERPC_CPU,                                    \
4172         .property = "pre-2.10-migration",                              \
4173         .value    = "on",                                              \
4174     },                                                                 \
4175 
4176 static void spapr_machine_2_9_instance_options(MachineState *machine)
4177 {
4178     spapr_machine_2_10_instance_options(machine);
4179 }
4180 
4181 static void spapr_machine_2_9_class_options(MachineClass *mc)
4182 {
4183     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4184 
4185     spapr_machine_2_10_class_options(mc);
4186     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4187     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4188     smc->pre_2_10_has_unused_icps = true;
4189     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4190 }
4191 
4192 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4193 
4194 /*
4195  * pseries-2.8
4196  */
4197 #define SPAPR_COMPAT_2_8                                        \
4198     HW_COMPAT_2_8                                               \
4199     {                                                           \
4200         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4201         .property = "pcie-extended-configuration-space",        \
4202         .value    = "off",                                      \
4203     },
4204 
4205 static void spapr_machine_2_8_instance_options(MachineState *machine)
4206 {
4207     spapr_machine_2_9_instance_options(machine);
4208 }
4209 
4210 static void spapr_machine_2_8_class_options(MachineClass *mc)
4211 {
4212     spapr_machine_2_9_class_options(mc);
4213     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4214     mc->numa_mem_align_shift = 23;
4215 }
4216 
4217 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4218 
4219 /*
4220  * pseries-2.7
4221  */
4222 #define SPAPR_COMPAT_2_7                            \
4223     HW_COMPAT_2_7                                   \
4224     {                                               \
4225         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4226         .property = "mem_win_size",                 \
4227         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4228     },                                              \
4229     {                                               \
4230         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4231         .property = "mem64_win_size",               \
4232         .value    = "0",                            \
4233     },                                              \
4234     {                                               \
4235         .driver = TYPE_POWERPC_CPU,                 \
4236         .property = "pre-2.8-migration",            \
4237         .value    = "on",                           \
4238     },                                              \
4239     {                                               \
4240         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4241         .property = "pre-2.8-migration",            \
4242         .value    = "on",                           \
4243     },
4244 
4245 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4246                               uint64_t *buid, hwaddr *pio,
4247                               hwaddr *mmio32, hwaddr *mmio64,
4248                               unsigned n_dma, uint32_t *liobns, Error **errp)
4249 {
4250     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4251     const uint64_t base_buid = 0x800000020000000ULL;
4252     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4253     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4254     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4255     const uint32_t max_index = 255;
4256     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4257 
4258     uint64_t ram_top = MACHINE(spapr)->ram_size;
4259     hwaddr phb0_base, phb_base;
4260     int i;
4261 
4262     /* Do we have hotpluggable memory? */
4263     if (MACHINE(spapr)->maxram_size > ram_top) {
4264         /* Can't just use maxram_size, because there may be an
4265          * alignment gap between normal and hotpluggable memory
4266          * regions */
4267         ram_top = spapr->hotplug_memory.base +
4268             memory_region_size(&spapr->hotplug_memory.mr);
4269     }
4270 
4271     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4272 
4273     if (index > max_index) {
4274         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4275                    max_index);
4276         return;
4277     }
4278 
4279     *buid = base_buid + index;
4280     for (i = 0; i < n_dma; ++i) {
4281         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4282     }
4283 
4284     phb_base = phb0_base + index * phb_spacing;
4285     *pio = phb_base + pio_offset;
4286     *mmio32 = phb_base + mmio_offset;
4287     /*
4288      * We don't set the 64-bit MMIO window, relying on the PHB's
4289      * fallback behaviour of automatically splitting a large "32-bit"
4290      * window into contiguous 32-bit and 64-bit windows
4291      */
4292 }
4293 
4294 static void spapr_machine_2_7_instance_options(MachineState *machine)
4295 {
4296     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4297 
4298     spapr_machine_2_8_instance_options(machine);
4299     spapr->use_hotplug_event_source = false;
4300 }
4301 
4302 static void spapr_machine_2_7_class_options(MachineClass *mc)
4303 {
4304     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4305 
4306     spapr_machine_2_8_class_options(mc);
4307     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4308     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4309     smc->phb_placement = phb_placement_2_7;
4310 }
4311 
4312 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4313 
4314 /*
4315  * pseries-2.6
4316  */
4317 #define SPAPR_COMPAT_2_6 \
4318     HW_COMPAT_2_6 \
4319     { \
4320         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4321         .property = "ddw",\
4322         .value    = stringify(off),\
4323     },
4324 
4325 static void spapr_machine_2_6_instance_options(MachineState *machine)
4326 {
4327     spapr_machine_2_7_instance_options(machine);
4328 }
4329 
4330 static void spapr_machine_2_6_class_options(MachineClass *mc)
4331 {
4332     spapr_machine_2_7_class_options(mc);
4333     mc->has_hotpluggable_cpus = false;
4334     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4335 }
4336 
4337 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4338 
4339 /*
4340  * pseries-2.5
4341  */
4342 #define SPAPR_COMPAT_2_5 \
4343     HW_COMPAT_2_5 \
4344     { \
4345         .driver   = "spapr-vlan", \
4346         .property = "use-rx-buffer-pools", \
4347         .value    = "off", \
4348     },
4349 
4350 static void spapr_machine_2_5_instance_options(MachineState *machine)
4351 {
4352     spapr_machine_2_6_instance_options(machine);
4353 }
4354 
4355 static void spapr_machine_2_5_class_options(MachineClass *mc)
4356 {
4357     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4358 
4359     spapr_machine_2_6_class_options(mc);
4360     smc->use_ohci_by_default = true;
4361     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4362 }
4363 
4364 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4365 
4366 /*
4367  * pseries-2.4
4368  */
4369 #define SPAPR_COMPAT_2_4 \
4370         HW_COMPAT_2_4
4371 
4372 static void spapr_machine_2_4_instance_options(MachineState *machine)
4373 {
4374     spapr_machine_2_5_instance_options(machine);
4375 }
4376 
4377 static void spapr_machine_2_4_class_options(MachineClass *mc)
4378 {
4379     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4380 
4381     spapr_machine_2_5_class_options(mc);
4382     smc->dr_lmb_enabled = false;
4383     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4384 }
4385 
4386 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4387 
4388 /*
4389  * pseries-2.3
4390  */
4391 #define SPAPR_COMPAT_2_3 \
4392         HW_COMPAT_2_3 \
4393         {\
4394             .driver   = "spapr-pci-host-bridge",\
4395             .property = "dynamic-reconfiguration",\
4396             .value    = "off",\
4397         },
4398 
4399 static void spapr_machine_2_3_instance_options(MachineState *machine)
4400 {
4401     spapr_machine_2_4_instance_options(machine);
4402 }
4403 
4404 static void spapr_machine_2_3_class_options(MachineClass *mc)
4405 {
4406     spapr_machine_2_4_class_options(mc);
4407     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4408 }
4409 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4410 
4411 /*
4412  * pseries-2.2
4413  */
4414 
4415 #define SPAPR_COMPAT_2_2 \
4416         HW_COMPAT_2_2 \
4417         {\
4418             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4419             .property = "mem_win_size",\
4420             .value    = "0x20000000",\
4421         },
4422 
4423 static void spapr_machine_2_2_instance_options(MachineState *machine)
4424 {
4425     spapr_machine_2_3_instance_options(machine);
4426     machine->suppress_vmdesc = true;
4427 }
4428 
4429 static void spapr_machine_2_2_class_options(MachineClass *mc)
4430 {
4431     spapr_machine_2_3_class_options(mc);
4432     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4433 }
4434 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4435 
4436 /*
4437  * pseries-2.1
4438  */
4439 #define SPAPR_COMPAT_2_1 \
4440         HW_COMPAT_2_1
4441 
4442 static void spapr_machine_2_1_instance_options(MachineState *machine)
4443 {
4444     spapr_machine_2_2_instance_options(machine);
4445 }
4446 
4447 static void spapr_machine_2_1_class_options(MachineClass *mc)
4448 {
4449     spapr_machine_2_2_class_options(mc);
4450     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4451 }
4452 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4453 
4454 static void spapr_machine_register_types(void)
4455 {
4456     type_register_static(&spapr_machine_info);
4457 }
4458 
4459 type_init(spapr_machine_register_types)
4460