xref: /openbmc/qemu/hw/ppc/spapr.c (revision cb764d06)
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/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
46 #include "cpu-models.h"
47 #include "qom/cpu.h"
48 
49 #include "hw/boards.h"
50 #include "hw/ppc/ppc.h"
51 #include "hw/loader.h"
52 
53 #include "hw/ppc/fdt.h"
54 #include "hw/ppc/spapr.h"
55 #include "hw/ppc/spapr_vio.h"
56 #include "hw/pci-host/spapr.h"
57 #include "hw/ppc/xics.h"
58 #include "hw/pci/msi.h"
59 
60 #include "hw/pci/pci.h"
61 #include "hw/scsi/scsi.h"
62 #include "hw/virtio/virtio-scsi.h"
63 #include "hw/virtio/vhost-scsi-common.h"
64 
65 #include "exec/address-spaces.h"
66 #include "exec/ram_addr.h"
67 #include "hw/usb.h"
68 #include "qemu/config-file.h"
69 #include "qemu/error-report.h"
70 #include "trace.h"
71 #include "hw/nmi.h"
72 #include "hw/intc/intc.h"
73 
74 #include "hw/compat.h"
75 #include "qemu/cutils.h"
76 #include "hw/ppc/spapr_cpu_core.h"
77 #include "hw/mem/memory-device.h"
78 
79 #include <libfdt.h>
80 
81 /* SLOF memory layout:
82  *
83  * SLOF raw image loaded at 0, copies its romfs right below the flat
84  * device-tree, then position SLOF itself 31M below that
85  *
86  * So we set FW_OVERHEAD to 40MB which should account for all of that
87  * and more
88  *
89  * We load our kernel at 4M, leaving space for SLOF initial image
90  */
91 #define FDT_MAX_SIZE            0x100000
92 #define RTAS_MAX_SIZE           0x10000
93 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
94 #define FW_MAX_SIZE             0x400000
95 #define FW_FILE_NAME            "slof.bin"
96 #define FW_OVERHEAD             0x2800000
97 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
98 
99 #define MIN_RMA_SLOF            128UL
100 
101 #define PHANDLE_XICP            0x00001111
102 
103 /* These two functions implement the VCPU id numbering: one to compute them
104  * all and one to identify thread 0 of a VCORE. Any change to the first one
105  * is likely to have an impact on the second one, so let's keep them close.
106  */
107 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
108 {
109     assert(spapr->vsmt);
110     return
111         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
112 }
113 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
114                                       PowerPCCPU *cpu)
115 {
116     assert(spapr->vsmt);
117     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
118 }
119 
120 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
121                                   const char *type_ics,
122                                   int nr_irqs, Error **errp)
123 {
124     Error *local_err = NULL;
125     Object *obj;
126 
127     obj = object_new(type_ics);
128     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
129     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
130                                    &error_abort);
131     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
132     if (local_err) {
133         goto error;
134     }
135     object_property_set_bool(obj, true, "realized", &local_err);
136     if (local_err) {
137         goto error;
138     }
139 
140     return ICS_BASE(obj);
141 
142 error:
143     error_propagate(errp, local_err);
144     return NULL;
145 }
146 
147 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
148 {
149     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
150      * and newer QEMUs don't even have them. In both cases, we don't want
151      * to send anything on the wire.
152      */
153     return false;
154 }
155 
156 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
157     .name = "icp/server",
158     .version_id = 1,
159     .minimum_version_id = 1,
160     .needed = pre_2_10_vmstate_dummy_icp_needed,
161     .fields = (VMStateField[]) {
162         VMSTATE_UNUSED(4), /* uint32_t xirr */
163         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
164         VMSTATE_UNUSED(1), /* uint8_t mfrr */
165         VMSTATE_END_OF_LIST()
166     },
167 };
168 
169 static void pre_2_10_vmstate_register_dummy_icp(int i)
170 {
171     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
172                      (void *)(uintptr_t) i);
173 }
174 
175 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
176 {
177     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
178                        (void *)(uintptr_t) i);
179 }
180 
181 static int xics_max_server_number(sPAPRMachineState *spapr)
182 {
183     assert(spapr->vsmt);
184     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
185 }
186 
187 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
188 {
189     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
190     Error *local_err = NULL;
191 
192     if (kvm_enabled()) {
193         if (machine_kernel_irqchip_allowed(machine) &&
194             !xics_kvm_init(spapr, &local_err)) {
195             spapr->icp_type = TYPE_KVM_ICP;
196             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs,
197                                           &local_err);
198         }
199         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
200             error_prepend(&local_err,
201                           "kernel_irqchip requested but unavailable: ");
202             goto error;
203         }
204         error_free(local_err);
205         local_err = NULL;
206     }
207 
208     if (!spapr->ics) {
209         xics_spapr_init(spapr);
210         spapr->icp_type = TYPE_ICP;
211         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs,
212                                       &local_err);
213     }
214 
215 error:
216     error_propagate(errp, local_err);
217 }
218 
219 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
220                                   int smt_threads)
221 {
222     int i, ret = 0;
223     uint32_t servers_prop[smt_threads];
224     uint32_t gservers_prop[smt_threads * 2];
225     int index = spapr_get_vcpu_id(cpu);
226 
227     if (cpu->compat_pvr) {
228         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
229         if (ret < 0) {
230             return ret;
231         }
232     }
233 
234     /* Build interrupt servers and gservers properties */
235     for (i = 0; i < smt_threads; i++) {
236         servers_prop[i] = cpu_to_be32(index + i);
237         /* Hack, direct the group queues back to cpu 0 */
238         gservers_prop[i*2] = cpu_to_be32(index + i);
239         gservers_prop[i*2 + 1] = 0;
240     }
241     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
242                       servers_prop, sizeof(servers_prop));
243     if (ret < 0) {
244         return ret;
245     }
246     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
247                       gservers_prop, sizeof(gservers_prop));
248 
249     return ret;
250 }
251 
252 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
253 {
254     int index = spapr_get_vcpu_id(cpu);
255     uint32_t associativity[] = {cpu_to_be32(0x5),
256                                 cpu_to_be32(0x0),
257                                 cpu_to_be32(0x0),
258                                 cpu_to_be32(0x0),
259                                 cpu_to_be32(cpu->node_id),
260                                 cpu_to_be32(index)};
261 
262     /* Advertise NUMA via ibm,associativity */
263     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
264                           sizeof(associativity));
265 }
266 
267 /* Populate the "ibm,pa-features" property */
268 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
269                                        PowerPCCPU *cpu,
270                                        void *fdt, int offset,
271                                        bool legacy_guest)
272 {
273     uint8_t pa_features_206[] = { 6, 0,
274         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
275     uint8_t pa_features_207[] = { 24, 0,
276         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
277         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
278         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
279         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
280     uint8_t pa_features_300[] = { 66, 0,
281         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
282         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
283         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
284         /* 6: DS207 */
285         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
286         /* 16: Vector */
287         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
288         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
289         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
290         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
291         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
292         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
293         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
294         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
295         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
296         /* 42: PM, 44: PC RA, 46: SC vec'd */
297         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
298         /* 48: SIMD, 50: QP BFP, 52: String */
299         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
300         /* 54: DecFP, 56: DecI, 58: SHA */
301         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
302         /* 60: NM atomic, 62: RNG */
303         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
304     };
305     uint8_t *pa_features = NULL;
306     size_t pa_size;
307 
308     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
309         pa_features = pa_features_206;
310         pa_size = sizeof(pa_features_206);
311     }
312     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
313         pa_features = pa_features_207;
314         pa_size = sizeof(pa_features_207);
315     }
316     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
317         pa_features = pa_features_300;
318         pa_size = sizeof(pa_features_300);
319     }
320     if (!pa_features) {
321         return;
322     }
323 
324     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
325         /*
326          * Note: we keep CI large pages off by default because a 64K capable
327          * guest provisioned with large pages might otherwise try to map a qemu
328          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
329          * even if that qemu runs on a 4k host.
330          * We dd this bit back here if we are confident this is not an issue
331          */
332         pa_features[3] |= 0x20;
333     }
334     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
335         pa_features[24] |= 0x80;    /* Transactional memory support */
336     }
337     if (legacy_guest && pa_size > 40) {
338         /* Workaround for broken kernels that attempt (guest) radix
339          * mode when they can't handle it, if they see the radix bit set
340          * in pa-features. So hide it from them. */
341         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
342     }
343 
344     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
345 }
346 
347 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
348 {
349     int ret = 0, offset, cpus_offset;
350     CPUState *cs;
351     char cpu_model[32];
352     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
353 
354     CPU_FOREACH(cs) {
355         PowerPCCPU *cpu = POWERPC_CPU(cs);
356         DeviceClass *dc = DEVICE_GET_CLASS(cs);
357         int index = spapr_get_vcpu_id(cpu);
358         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
359 
360         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
361             continue;
362         }
363 
364         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
365 
366         cpus_offset = fdt_path_offset(fdt, "/cpus");
367         if (cpus_offset < 0) {
368             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
369             if (cpus_offset < 0) {
370                 return cpus_offset;
371             }
372         }
373         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
374         if (offset < 0) {
375             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
376             if (offset < 0) {
377                 return offset;
378             }
379         }
380 
381         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
382                           pft_size_prop, sizeof(pft_size_prop));
383         if (ret < 0) {
384             return ret;
385         }
386 
387         if (nb_numa_nodes > 1) {
388             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
389             if (ret < 0) {
390                 return ret;
391             }
392         }
393 
394         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
395         if (ret < 0) {
396             return ret;
397         }
398 
399         spapr_populate_pa_features(spapr, cpu, fdt, offset,
400                                    spapr->cas_legacy_guest_workaround);
401     }
402     return ret;
403 }
404 
405 static hwaddr spapr_node0_size(MachineState *machine)
406 {
407     if (nb_numa_nodes) {
408         int i;
409         for (i = 0; i < nb_numa_nodes; ++i) {
410             if (numa_info[i].node_mem) {
411                 return MIN(pow2floor(numa_info[i].node_mem),
412                            machine->ram_size);
413             }
414         }
415     }
416     return machine->ram_size;
417 }
418 
419 static void add_str(GString *s, const gchar *s1)
420 {
421     g_string_append_len(s, s1, strlen(s1) + 1);
422 }
423 
424 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
425                                        hwaddr size)
426 {
427     uint32_t associativity[] = {
428         cpu_to_be32(0x4), /* length */
429         cpu_to_be32(0x0), cpu_to_be32(0x0),
430         cpu_to_be32(0x0), cpu_to_be32(nodeid)
431     };
432     char mem_name[32];
433     uint64_t mem_reg_property[2];
434     int off;
435 
436     mem_reg_property[0] = cpu_to_be64(start);
437     mem_reg_property[1] = cpu_to_be64(size);
438 
439     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
440     off = fdt_add_subnode(fdt, 0, mem_name);
441     _FDT(off);
442     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
443     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
444                       sizeof(mem_reg_property))));
445     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
446                       sizeof(associativity))));
447     return off;
448 }
449 
450 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
451 {
452     MachineState *machine = MACHINE(spapr);
453     hwaddr mem_start, node_size;
454     int i, nb_nodes = nb_numa_nodes;
455     NodeInfo *nodes = numa_info;
456     NodeInfo ramnode;
457 
458     /* No NUMA nodes, assume there is just one node with whole RAM */
459     if (!nb_numa_nodes) {
460         nb_nodes = 1;
461         ramnode.node_mem = machine->ram_size;
462         nodes = &ramnode;
463     }
464 
465     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
466         if (!nodes[i].node_mem) {
467             continue;
468         }
469         if (mem_start >= machine->ram_size) {
470             node_size = 0;
471         } else {
472             node_size = nodes[i].node_mem;
473             if (node_size > machine->ram_size - mem_start) {
474                 node_size = machine->ram_size - mem_start;
475             }
476         }
477         if (!mem_start) {
478             /* spapr_machine_init() checks for rma_size <= node0_size
479              * already */
480             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
481             mem_start += spapr->rma_size;
482             node_size -= spapr->rma_size;
483         }
484         for ( ; node_size; ) {
485             hwaddr sizetmp = pow2floor(node_size);
486 
487             /* mem_start != 0 here */
488             if (ctzl(mem_start) < ctzl(sizetmp)) {
489                 sizetmp = 1ULL << ctzl(mem_start);
490             }
491 
492             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
493             node_size -= sizetmp;
494             mem_start += sizetmp;
495         }
496     }
497 
498     return 0;
499 }
500 
501 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
502                                   sPAPRMachineState *spapr)
503 {
504     PowerPCCPU *cpu = POWERPC_CPU(cs);
505     CPUPPCState *env = &cpu->env;
506     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
507     int index = spapr_get_vcpu_id(cpu);
508     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
509                        0xffffffff, 0xffffffff};
510     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
511         : SPAPR_TIMEBASE_FREQ;
512     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
513     uint32_t page_sizes_prop[64];
514     size_t page_sizes_prop_size;
515     uint32_t vcpus_per_socket = smp_threads * smp_cores;
516     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
517     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
518     sPAPRDRConnector *drc;
519     int drc_index;
520     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
521     int i;
522 
523     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
524     if (drc) {
525         drc_index = spapr_drc_index(drc);
526         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
527     }
528 
529     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
530     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
531 
532     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
533     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
534                            env->dcache_line_size)));
535     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
536                            env->dcache_line_size)));
537     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
538                            env->icache_line_size)));
539     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
540                            env->icache_line_size)));
541 
542     if (pcc->l1_dcache_size) {
543         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
544                                pcc->l1_dcache_size)));
545     } else {
546         warn_report("Unknown L1 dcache size for cpu");
547     }
548     if (pcc->l1_icache_size) {
549         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
550                                pcc->l1_icache_size)));
551     } else {
552         warn_report("Unknown L1 icache size for cpu");
553     }
554 
555     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
556     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
557     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
558     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
559     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
560     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
561 
562     if (env->spr_cb[SPR_PURR].oea_read) {
563         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
564     }
565 
566     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
567         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
568                           segs, sizeof(segs))));
569     }
570 
571     /* Advertise VSX (vector extensions) if available
572      *   1               == VMX / Altivec available
573      *   2               == VSX available
574      *
575      * Only CPUs for which we create core types in spapr_cpu_core.c
576      * are possible, and all of those have VMX */
577     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
578         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
579     } else {
580         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
581     }
582 
583     /* Advertise DFP (Decimal Floating Point) if available
584      *   0 / no property == no DFP
585      *   1               == DFP available */
586     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
587         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
588     }
589 
590     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
591                                                       sizeof(page_sizes_prop));
592     if (page_sizes_prop_size) {
593         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
594                           page_sizes_prop, page_sizes_prop_size)));
595     }
596 
597     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
598 
599     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
600                            cs->cpu_index / vcpus_per_socket)));
601 
602     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
603                       pft_size_prop, sizeof(pft_size_prop))));
604 
605     if (nb_numa_nodes > 1) {
606         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
607     }
608 
609     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
610 
611     if (pcc->radix_page_info) {
612         for (i = 0; i < pcc->radix_page_info->count; i++) {
613             radix_AP_encodings[i] =
614                 cpu_to_be32(pcc->radix_page_info->entries[i]);
615         }
616         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
617                           radix_AP_encodings,
618                           pcc->radix_page_info->count *
619                           sizeof(radix_AP_encodings[0]))));
620     }
621 }
622 
623 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
624 {
625     CPUState *cs;
626     int cpus_offset;
627     char *nodename;
628 
629     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
630     _FDT(cpus_offset);
631     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
632     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
633 
634     /*
635      * We walk the CPUs in reverse order to ensure that CPU DT nodes
636      * created by fdt_add_subnode() end up in the right order in FDT
637      * for the guest kernel the enumerate the CPUs correctly.
638      */
639     CPU_FOREACH_REVERSE(cs) {
640         PowerPCCPU *cpu = POWERPC_CPU(cs);
641         int index = spapr_get_vcpu_id(cpu);
642         DeviceClass *dc = DEVICE_GET_CLASS(cs);
643         int offset;
644 
645         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
646             continue;
647         }
648 
649         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
650         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
651         g_free(nodename);
652         _FDT(offset);
653         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
654     }
655 
656 }
657 
658 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
659 {
660     MemoryDeviceInfoList *info;
661 
662     for (info = list; info; info = info->next) {
663         MemoryDeviceInfo *value = info->value;
664 
665         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
666             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
667 
668             if (addr >= pcdimm_info->addr &&
669                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
670                 return pcdimm_info->node;
671             }
672         }
673     }
674 
675     return -1;
676 }
677 
678 struct sPAPRDrconfCellV2 {
679      uint32_t seq_lmbs;
680      uint64_t base_addr;
681      uint32_t drc_index;
682      uint32_t aa_index;
683      uint32_t flags;
684 } QEMU_PACKED;
685 
686 typedef struct DrconfCellQueue {
687     struct sPAPRDrconfCellV2 cell;
688     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
689 } DrconfCellQueue;
690 
691 static DrconfCellQueue *
692 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
693                       uint32_t drc_index, uint32_t aa_index,
694                       uint32_t flags)
695 {
696     DrconfCellQueue *elem;
697 
698     elem = g_malloc0(sizeof(*elem));
699     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
700     elem->cell.base_addr = cpu_to_be64(base_addr);
701     elem->cell.drc_index = cpu_to_be32(drc_index);
702     elem->cell.aa_index = cpu_to_be32(aa_index);
703     elem->cell.flags = cpu_to_be32(flags);
704 
705     return elem;
706 }
707 
708 /* ibm,dynamic-memory-v2 */
709 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
710                                    int offset, MemoryDeviceInfoList *dimms)
711 {
712     MachineState *machine = MACHINE(spapr);
713     uint8_t *int_buf, *cur_index, buf_len;
714     int ret;
715     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
716     uint64_t addr, cur_addr, size;
717     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
718     uint64_t mem_end = machine->device_memory->base +
719                        memory_region_size(&machine->device_memory->mr);
720     uint32_t node, nr_entries = 0;
721     sPAPRDRConnector *drc;
722     DrconfCellQueue *elem, *next;
723     MemoryDeviceInfoList *info;
724     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
725         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
726 
727     /* Entry to cover RAM and the gap area */
728     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
729                                  SPAPR_LMB_FLAGS_RESERVED |
730                                  SPAPR_LMB_FLAGS_DRC_INVALID);
731     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
732     nr_entries++;
733 
734     cur_addr = machine->device_memory->base;
735     for (info = dimms; info; info = info->next) {
736         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
737 
738         addr = di->addr;
739         size = di->size;
740         node = di->node;
741 
742         /* Entry for hot-pluggable area */
743         if (cur_addr < addr) {
744             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
745             g_assert(drc);
746             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
747                                          cur_addr, spapr_drc_index(drc), -1, 0);
748             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
749             nr_entries++;
750         }
751 
752         /* Entry for DIMM */
753         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
754         g_assert(drc);
755         elem = spapr_get_drconf_cell(size / lmb_size, addr,
756                                      spapr_drc_index(drc), node,
757                                      SPAPR_LMB_FLAGS_ASSIGNED);
758         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
759         nr_entries++;
760         cur_addr = addr + size;
761     }
762 
763     /* Entry for remaining hotpluggable area */
764     if (cur_addr < mem_end) {
765         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
766         g_assert(drc);
767         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
768                                      cur_addr, spapr_drc_index(drc), -1, 0);
769         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
770         nr_entries++;
771     }
772 
773     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
774     int_buf = cur_index = g_malloc0(buf_len);
775     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
776     cur_index += sizeof(nr_entries);
777 
778     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
779         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
780         cur_index += sizeof(elem->cell);
781         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
782         g_free(elem);
783     }
784 
785     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
786     g_free(int_buf);
787     if (ret < 0) {
788         return -1;
789     }
790     return 0;
791 }
792 
793 /* ibm,dynamic-memory */
794 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
795                                    int offset, MemoryDeviceInfoList *dimms)
796 {
797     MachineState *machine = MACHINE(spapr);
798     int i, ret;
799     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
800     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
801     uint32_t nr_lmbs = (machine->device_memory->base +
802                        memory_region_size(&machine->device_memory->mr)) /
803                        lmb_size;
804     uint32_t *int_buf, *cur_index, buf_len;
805 
806     /*
807      * Allocate enough buffer size to fit in ibm,dynamic-memory
808      */
809     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
810     cur_index = int_buf = g_malloc0(buf_len);
811     int_buf[0] = cpu_to_be32(nr_lmbs);
812     cur_index++;
813     for (i = 0; i < nr_lmbs; i++) {
814         uint64_t addr = i * lmb_size;
815         uint32_t *dynamic_memory = cur_index;
816 
817         if (i >= device_lmb_start) {
818             sPAPRDRConnector *drc;
819 
820             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
821             g_assert(drc);
822 
823             dynamic_memory[0] = cpu_to_be32(addr >> 32);
824             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
825             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
826             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
827             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
828             if (memory_region_present(get_system_memory(), addr)) {
829                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
830             } else {
831                 dynamic_memory[5] = cpu_to_be32(0);
832             }
833         } else {
834             /*
835              * LMB information for RMA, boot time RAM and gap b/n RAM and
836              * device memory region -- all these are marked as reserved
837              * and as having no valid DRC.
838              */
839             dynamic_memory[0] = cpu_to_be32(addr >> 32);
840             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
841             dynamic_memory[2] = cpu_to_be32(0);
842             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
843             dynamic_memory[4] = cpu_to_be32(-1);
844             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
845                                             SPAPR_LMB_FLAGS_DRC_INVALID);
846         }
847 
848         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
849     }
850     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
851     g_free(int_buf);
852     if (ret < 0) {
853         return -1;
854     }
855     return 0;
856 }
857 
858 /*
859  * Adds ibm,dynamic-reconfiguration-memory node.
860  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
861  * of this device tree node.
862  */
863 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
864 {
865     MachineState *machine = MACHINE(spapr);
866     int ret, i, offset;
867     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
868     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
869     uint32_t *int_buf, *cur_index, buf_len;
870     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
871     MemoryDeviceInfoList *dimms = NULL;
872 
873     /*
874      * Don't create the node if there is no device memory
875      */
876     if (machine->ram_size == machine->maxram_size) {
877         return 0;
878     }
879 
880     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
881 
882     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
883                     sizeof(prop_lmb_size));
884     if (ret < 0) {
885         return ret;
886     }
887 
888     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
889     if (ret < 0) {
890         return ret;
891     }
892 
893     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
894     if (ret < 0) {
895         return ret;
896     }
897 
898     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
899     dimms = qmp_memory_device_list();
900     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
901         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
902     } else {
903         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
904     }
905     qapi_free_MemoryDeviceInfoList(dimms);
906 
907     if (ret < 0) {
908         return ret;
909     }
910 
911     /* ibm,associativity-lookup-arrays */
912     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
913     cur_index = int_buf = g_malloc0(buf_len);
914 
915     cur_index = int_buf;
916     int_buf[0] = cpu_to_be32(nr_nodes);
917     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
918     cur_index += 2;
919     for (i = 0; i < nr_nodes; i++) {
920         uint32_t associativity[] = {
921             cpu_to_be32(0x0),
922             cpu_to_be32(0x0),
923             cpu_to_be32(0x0),
924             cpu_to_be32(i)
925         };
926         memcpy(cur_index, associativity, sizeof(associativity));
927         cur_index += 4;
928     }
929     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
930             (cur_index - int_buf) * sizeof(uint32_t));
931     g_free(int_buf);
932 
933     return ret;
934 }
935 
936 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
937                                 sPAPROptionVector *ov5_updates)
938 {
939     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
940     int ret = 0, offset;
941 
942     /* Generate ibm,dynamic-reconfiguration-memory node if required */
943     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
944         g_assert(smc->dr_lmb_enabled);
945         ret = spapr_populate_drconf_memory(spapr, fdt);
946         if (ret) {
947             goto out;
948         }
949     }
950 
951     offset = fdt_path_offset(fdt, "/chosen");
952     if (offset < 0) {
953         offset = fdt_add_subnode(fdt, 0, "chosen");
954         if (offset < 0) {
955             return offset;
956         }
957     }
958     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
959                                  "ibm,architecture-vec-5");
960 
961 out:
962     return ret;
963 }
964 
965 static bool spapr_hotplugged_dev_before_cas(void)
966 {
967     Object *drc_container, *obj;
968     ObjectProperty *prop;
969     ObjectPropertyIterator iter;
970 
971     drc_container = container_get(object_get_root(), "/dr-connector");
972     object_property_iter_init(&iter, drc_container);
973     while ((prop = object_property_iter_next(&iter))) {
974         if (!strstart(prop->type, "link<", NULL)) {
975             continue;
976         }
977         obj = object_property_get_link(drc_container, prop->name, NULL);
978         if (spapr_drc_needed(obj)) {
979             return true;
980         }
981     }
982     return false;
983 }
984 
985 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
986                                  target_ulong addr, target_ulong size,
987                                  sPAPROptionVector *ov5_updates)
988 {
989     void *fdt, *fdt_skel;
990     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
991 
992     if (spapr_hotplugged_dev_before_cas()) {
993         return 1;
994     }
995 
996     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
997         error_report("SLOF provided an unexpected CAS buffer size "
998                      TARGET_FMT_lu " (min: %zu, max: %u)",
999                      size, sizeof(hdr), FW_MAX_SIZE);
1000         exit(EXIT_FAILURE);
1001     }
1002 
1003     size -= sizeof(hdr);
1004 
1005     /* Create skeleton */
1006     fdt_skel = g_malloc0(size);
1007     _FDT((fdt_create(fdt_skel, size)));
1008     _FDT((fdt_finish_reservemap(fdt_skel)));
1009     _FDT((fdt_begin_node(fdt_skel, "")));
1010     _FDT((fdt_end_node(fdt_skel)));
1011     _FDT((fdt_finish(fdt_skel)));
1012     fdt = g_malloc0(size);
1013     _FDT((fdt_open_into(fdt_skel, fdt, size)));
1014     g_free(fdt_skel);
1015 
1016     /* Fixup cpu nodes */
1017     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
1018 
1019     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
1020         return -1;
1021     }
1022 
1023     /* Pack resulting tree */
1024     _FDT((fdt_pack(fdt)));
1025 
1026     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1027         trace_spapr_cas_failed(size);
1028         return -1;
1029     }
1030 
1031     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1032     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1033     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1034     g_free(fdt);
1035 
1036     return 0;
1037 }
1038 
1039 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1040 {
1041     int rtas;
1042     GString *hypertas = g_string_sized_new(256);
1043     GString *qemu_hypertas = g_string_sized_new(256);
1044     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1045     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1046         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1047     uint32_t lrdr_capacity[] = {
1048         cpu_to_be32(max_device_addr >> 32),
1049         cpu_to_be32(max_device_addr & 0xffffffff),
1050         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1051         cpu_to_be32(max_cpus / smp_threads),
1052     };
1053     uint32_t maxdomains[] = {
1054         cpu_to_be32(4),
1055         cpu_to_be32(0),
1056         cpu_to_be32(0),
1057         cpu_to_be32(0),
1058         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
1059     };
1060 
1061     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1062 
1063     /* hypertas */
1064     add_str(hypertas, "hcall-pft");
1065     add_str(hypertas, "hcall-term");
1066     add_str(hypertas, "hcall-dabr");
1067     add_str(hypertas, "hcall-interrupt");
1068     add_str(hypertas, "hcall-tce");
1069     add_str(hypertas, "hcall-vio");
1070     add_str(hypertas, "hcall-splpar");
1071     add_str(hypertas, "hcall-bulk");
1072     add_str(hypertas, "hcall-set-mode");
1073     add_str(hypertas, "hcall-sprg0");
1074     add_str(hypertas, "hcall-copy");
1075     add_str(hypertas, "hcall-debug");
1076     add_str(qemu_hypertas, "hcall-memop1");
1077 
1078     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1079         add_str(hypertas, "hcall-multi-tce");
1080     }
1081 
1082     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1083         add_str(hypertas, "hcall-hpt-resize");
1084     }
1085 
1086     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1087                      hypertas->str, hypertas->len));
1088     g_string_free(hypertas, TRUE);
1089     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1090                      qemu_hypertas->str, qemu_hypertas->len));
1091     g_string_free(qemu_hypertas, TRUE);
1092 
1093     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1094                      refpoints, sizeof(refpoints)));
1095 
1096     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1097                      maxdomains, sizeof(maxdomains)));
1098 
1099     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1100                           RTAS_ERROR_LOG_MAX));
1101     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1102                           RTAS_EVENT_SCAN_RATE));
1103 
1104     g_assert(msi_nonbroken);
1105     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1106 
1107     /*
1108      * According to PAPR, rtas ibm,os-term does not guarantee a return
1109      * back to the guest cpu.
1110      *
1111      * While an additional ibm,extended-os-term property indicates
1112      * that rtas call return will always occur. Set this property.
1113      */
1114     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1115 
1116     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1117                      lrdr_capacity, sizeof(lrdr_capacity)));
1118 
1119     spapr_dt_rtas_tokens(fdt, rtas);
1120 }
1121 
1122 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1123  * that the guest may request and thus the valid values for bytes 24..26 of
1124  * option vector 5: */
1125 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1126 {
1127     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1128 
1129     char val[2 * 4] = {
1130         23, 0x00, /* Xive mode, filled in below. */
1131         24, 0x00, /* Hash/Radix, filled in below. */
1132         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1133         26, 0x40, /* Radix options: GTSE == yes. */
1134     };
1135 
1136     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1137                           first_ppc_cpu->compat_pvr)) {
1138         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1139         val[3] = 0x00; /* Hash */
1140     } else if (kvm_enabled()) {
1141         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1142             val[3] = 0x80; /* OV5_MMU_BOTH */
1143         } else if (kvmppc_has_cap_mmu_radix()) {
1144             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1145         } else {
1146             val[3] = 0x00; /* Hash */
1147         }
1148     } else {
1149         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1150         val[3] = 0xC0;
1151     }
1152     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1153                      val, sizeof(val)));
1154 }
1155 
1156 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1157 {
1158     MachineState *machine = MACHINE(spapr);
1159     int chosen;
1160     const char *boot_device = machine->boot_order;
1161     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1162     size_t cb = 0;
1163     char *bootlist = get_boot_devices_list(&cb);
1164 
1165     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1166 
1167     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1168     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1169                           spapr->initrd_base));
1170     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1171                           spapr->initrd_base + spapr->initrd_size));
1172 
1173     if (spapr->kernel_size) {
1174         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1175                               cpu_to_be64(spapr->kernel_size) };
1176 
1177         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1178                          &kprop, sizeof(kprop)));
1179         if (spapr->kernel_le) {
1180             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1181         }
1182     }
1183     if (boot_menu) {
1184         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1185     }
1186     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1187     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1188     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1189 
1190     if (cb && bootlist) {
1191         int i;
1192 
1193         for (i = 0; i < cb; i++) {
1194             if (bootlist[i] == '\n') {
1195                 bootlist[i] = ' ';
1196             }
1197         }
1198         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1199     }
1200 
1201     if (boot_device && strlen(boot_device)) {
1202         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1203     }
1204 
1205     if (!spapr->has_graphics && stdout_path) {
1206         /*
1207          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1208          * kernel. New platforms should only use the "stdout-path" property. Set
1209          * the new property and continue using older property to remain
1210          * compatible with the existing firmware.
1211          */
1212         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1213         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1214     }
1215 
1216     spapr_dt_ov5_platform_support(fdt, chosen);
1217 
1218     g_free(stdout_path);
1219     g_free(bootlist);
1220 }
1221 
1222 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1223 {
1224     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1225      * KVM to work under pHyp with some guest co-operation */
1226     int hypervisor;
1227     uint8_t hypercall[16];
1228 
1229     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1230     /* indicate KVM hypercall interface */
1231     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1232     if (kvmppc_has_cap_fixup_hcalls()) {
1233         /*
1234          * Older KVM versions with older guest kernels were broken
1235          * with the magic page, don't allow the guest to map it.
1236          */
1237         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1238                                   sizeof(hypercall))) {
1239             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1240                              hypercall, sizeof(hypercall)));
1241         }
1242     }
1243 }
1244 
1245 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1246                              hwaddr rtas_addr,
1247                              hwaddr rtas_size)
1248 {
1249     MachineState *machine = MACHINE(spapr);
1250     MachineClass *mc = MACHINE_GET_CLASS(machine);
1251     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1252     int ret;
1253     void *fdt;
1254     sPAPRPHBState *phb;
1255     char *buf;
1256 
1257     fdt = g_malloc0(FDT_MAX_SIZE);
1258     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1259 
1260     /* Root node */
1261     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1262     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1263     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1264 
1265     /*
1266      * Add info to guest to indentify which host is it being run on
1267      * and what is the uuid of the guest
1268      */
1269     if (kvmppc_get_host_model(&buf)) {
1270         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1271         g_free(buf);
1272     }
1273     if (kvmppc_get_host_serial(&buf)) {
1274         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1275         g_free(buf);
1276     }
1277 
1278     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1279 
1280     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1281     if (qemu_uuid_set) {
1282         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1283     }
1284     g_free(buf);
1285 
1286     if (qemu_get_vm_name()) {
1287         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1288                                 qemu_get_vm_name()));
1289     }
1290 
1291     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1292     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1293 
1294     /* /interrupt controller */
1295     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1296 
1297     ret = spapr_populate_memory(spapr, fdt);
1298     if (ret < 0) {
1299         error_report("couldn't setup memory nodes in fdt");
1300         exit(1);
1301     }
1302 
1303     /* /vdevice */
1304     spapr_dt_vdevice(spapr->vio_bus, fdt);
1305 
1306     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1307         ret = spapr_rng_populate_dt(fdt);
1308         if (ret < 0) {
1309             error_report("could not set up rng device in the fdt");
1310             exit(1);
1311         }
1312     }
1313 
1314     QLIST_FOREACH(phb, &spapr->phbs, list) {
1315         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1316         if (ret < 0) {
1317             error_report("couldn't setup PCI devices in fdt");
1318             exit(1);
1319         }
1320     }
1321 
1322     /* cpus */
1323     spapr_populate_cpus_dt_node(fdt, spapr);
1324 
1325     if (smc->dr_lmb_enabled) {
1326         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1327     }
1328 
1329     if (mc->has_hotpluggable_cpus) {
1330         int offset = fdt_path_offset(fdt, "/cpus");
1331         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1332                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1333         if (ret < 0) {
1334             error_report("Couldn't set up CPU DR device tree properties");
1335             exit(1);
1336         }
1337     }
1338 
1339     /* /event-sources */
1340     spapr_dt_events(spapr, fdt);
1341 
1342     /* /rtas */
1343     spapr_dt_rtas(spapr, fdt);
1344 
1345     /* /chosen */
1346     spapr_dt_chosen(spapr, fdt);
1347 
1348     /* /hypervisor */
1349     if (kvm_enabled()) {
1350         spapr_dt_hypervisor(spapr, fdt);
1351     }
1352 
1353     /* Build memory reserve map */
1354     if (spapr->kernel_size) {
1355         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1356     }
1357     if (spapr->initrd_size) {
1358         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1359     }
1360 
1361     /* ibm,client-architecture-support updates */
1362     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1363     if (ret < 0) {
1364         error_report("couldn't setup CAS properties fdt");
1365         exit(1);
1366     }
1367 
1368     return fdt;
1369 }
1370 
1371 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1372 {
1373     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1374 }
1375 
1376 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1377                                     PowerPCCPU *cpu)
1378 {
1379     CPUPPCState *env = &cpu->env;
1380 
1381     /* The TCG path should also be holding the BQL at this point */
1382     g_assert(qemu_mutex_iothread_locked());
1383 
1384     if (msr_pr) {
1385         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1386         env->gpr[3] = H_PRIVILEGE;
1387     } else {
1388         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1389     }
1390 }
1391 
1392 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1393 {
1394     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1395 
1396     return spapr->patb_entry;
1397 }
1398 
1399 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1400 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1401 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1402 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1403 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1404 
1405 /*
1406  * Get the fd to access the kernel htab, re-opening it if necessary
1407  */
1408 static int get_htab_fd(sPAPRMachineState *spapr)
1409 {
1410     Error *local_err = NULL;
1411 
1412     if (spapr->htab_fd >= 0) {
1413         return spapr->htab_fd;
1414     }
1415 
1416     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1417     if (spapr->htab_fd < 0) {
1418         error_report_err(local_err);
1419     }
1420 
1421     return spapr->htab_fd;
1422 }
1423 
1424 void close_htab_fd(sPAPRMachineState *spapr)
1425 {
1426     if (spapr->htab_fd >= 0) {
1427         close(spapr->htab_fd);
1428     }
1429     spapr->htab_fd = -1;
1430 }
1431 
1432 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1433 {
1434     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1435 
1436     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1437 }
1438 
1439 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1440 {
1441     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1442 
1443     assert(kvm_enabled());
1444 
1445     if (!spapr->htab) {
1446         return 0;
1447     }
1448 
1449     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1450 }
1451 
1452 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1453                                                 hwaddr ptex, int n)
1454 {
1455     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1456     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1457 
1458     if (!spapr->htab) {
1459         /*
1460          * HTAB is controlled by KVM. Fetch into temporary buffer
1461          */
1462         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1463         kvmppc_read_hptes(hptes, ptex, n);
1464         return hptes;
1465     }
1466 
1467     /*
1468      * HTAB is controlled by QEMU. Just point to the internally
1469      * accessible PTEG.
1470      */
1471     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1472 }
1473 
1474 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1475                               const ppc_hash_pte64_t *hptes,
1476                               hwaddr ptex, int n)
1477 {
1478     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1479 
1480     if (!spapr->htab) {
1481         g_free((void *)hptes);
1482     }
1483 
1484     /* Nothing to do for qemu managed HPT */
1485 }
1486 
1487 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1488                              uint64_t pte0, uint64_t pte1)
1489 {
1490     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1491     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1492 
1493     if (!spapr->htab) {
1494         kvmppc_write_hpte(ptex, pte0, pte1);
1495     } else {
1496         stq_p(spapr->htab + offset, pte0);
1497         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1498     }
1499 }
1500 
1501 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1502 {
1503     int shift;
1504 
1505     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1506      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1507      * that's much more than is needed for Linux guests */
1508     shift = ctz64(pow2ceil(ramsize)) - 7;
1509     shift = MAX(shift, 18); /* Minimum architected size */
1510     shift = MIN(shift, 46); /* Maximum architected size */
1511     return shift;
1512 }
1513 
1514 void spapr_free_hpt(sPAPRMachineState *spapr)
1515 {
1516     g_free(spapr->htab);
1517     spapr->htab = NULL;
1518     spapr->htab_shift = 0;
1519     close_htab_fd(spapr);
1520 }
1521 
1522 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1523                           Error **errp)
1524 {
1525     long rc;
1526 
1527     /* Clean up any HPT info from a previous boot */
1528     spapr_free_hpt(spapr);
1529 
1530     rc = kvmppc_reset_htab(shift);
1531     if (rc < 0) {
1532         /* kernel-side HPT needed, but couldn't allocate one */
1533         error_setg_errno(errp, errno,
1534                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1535                          shift);
1536         /* This is almost certainly fatal, but if the caller really
1537          * wants to carry on with shift == 0, it's welcome to try */
1538     } else if (rc > 0) {
1539         /* kernel-side HPT allocated */
1540         if (rc != shift) {
1541             error_setg(errp,
1542                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1543                        shift, rc);
1544         }
1545 
1546         spapr->htab_shift = shift;
1547         spapr->htab = NULL;
1548     } else {
1549         /* kernel-side HPT not needed, allocate in userspace instead */
1550         size_t size = 1ULL << shift;
1551         int i;
1552 
1553         spapr->htab = qemu_memalign(size, size);
1554         if (!spapr->htab) {
1555             error_setg_errno(errp, errno,
1556                              "Could not allocate HPT of order %d", shift);
1557             return;
1558         }
1559 
1560         memset(spapr->htab, 0, size);
1561         spapr->htab_shift = shift;
1562 
1563         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1564             DIRTY_HPTE(HPTE(spapr->htab, i));
1565         }
1566     }
1567     /* We're setting up a hash table, so that means we're not radix */
1568     spapr->patb_entry = 0;
1569 }
1570 
1571 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1572 {
1573     int hpt_shift;
1574 
1575     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1576         || (spapr->cas_reboot
1577             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1578         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1579     } else {
1580         uint64_t current_ram_size;
1581 
1582         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1583         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1584     }
1585     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1586 
1587     if (spapr->vrma_adjust) {
1588         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1589                                           spapr->htab_shift);
1590     }
1591 }
1592 
1593 static int spapr_reset_drcs(Object *child, void *opaque)
1594 {
1595     sPAPRDRConnector *drc =
1596         (sPAPRDRConnector *) object_dynamic_cast(child,
1597                                                  TYPE_SPAPR_DR_CONNECTOR);
1598 
1599     if (drc) {
1600         spapr_drc_reset(drc);
1601     }
1602 
1603     return 0;
1604 }
1605 
1606 static void spapr_machine_reset(void)
1607 {
1608     MachineState *machine = MACHINE(qdev_get_machine());
1609     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1610     PowerPCCPU *first_ppc_cpu;
1611     uint32_t rtas_limit;
1612     hwaddr rtas_addr, fdt_addr;
1613     void *fdt;
1614     int rc;
1615 
1616     spapr_caps_apply(spapr);
1617 
1618     first_ppc_cpu = POWERPC_CPU(first_cpu);
1619     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1620         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1621                               spapr->max_compat_pvr)) {
1622         /* If using KVM with radix mode available, VCPUs can be started
1623          * without a HPT because KVM will start them in radix mode.
1624          * Set the GR bit in PATB so that we know there is no HPT. */
1625         spapr->patb_entry = PATBE1_GR;
1626     } else {
1627         spapr_setup_hpt_and_vrma(spapr);
1628     }
1629 
1630     /* if this reset wasn't generated by CAS, we should reset our
1631      * negotiated options and start from scratch */
1632     if (!spapr->cas_reboot) {
1633         spapr_ovec_cleanup(spapr->ov5_cas);
1634         spapr->ov5_cas = spapr_ovec_new();
1635 
1636         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1637     }
1638 
1639     qemu_devices_reset();
1640 
1641     /* DRC reset may cause a device to be unplugged. This will cause troubles
1642      * if this device is used by another device (eg, a running vhost backend
1643      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1644      * situations, we reset DRCs after all devices have been reset.
1645      */
1646     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1647 
1648     spapr_clear_pending_events(spapr);
1649 
1650     /*
1651      * We place the device tree and RTAS just below either the top of the RMA,
1652      * or just below 2GB, whichever is lowere, so that it can be
1653      * processed with 32-bit real mode code if necessary
1654      */
1655     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1656     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1657     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1658 
1659     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1660 
1661     spapr_load_rtas(spapr, fdt, rtas_addr);
1662 
1663     rc = fdt_pack(fdt);
1664 
1665     /* Should only fail if we've built a corrupted tree */
1666     assert(rc == 0);
1667 
1668     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1669         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1670                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1671         exit(1);
1672     }
1673 
1674     /* Load the fdt */
1675     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1676     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1677     g_free(fdt);
1678 
1679     /* Set up the entry state */
1680     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1681     first_ppc_cpu->env.gpr[5] = 0;
1682 
1683     spapr->cas_reboot = false;
1684 }
1685 
1686 static void spapr_create_nvram(sPAPRMachineState *spapr)
1687 {
1688     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1689     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1690 
1691     if (dinfo) {
1692         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1693                             &error_fatal);
1694     }
1695 
1696     qdev_init_nofail(dev);
1697 
1698     spapr->nvram = (struct sPAPRNVRAM *)dev;
1699 }
1700 
1701 static void spapr_rtc_create(sPAPRMachineState *spapr)
1702 {
1703     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1704     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1705                               &error_fatal);
1706     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1707                               &error_fatal);
1708     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1709                               "date", &error_fatal);
1710 }
1711 
1712 /* Returns whether we want to use VGA or not */
1713 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1714 {
1715     switch (vga_interface_type) {
1716     case VGA_NONE:
1717         return false;
1718     case VGA_DEVICE:
1719         return true;
1720     case VGA_STD:
1721     case VGA_VIRTIO:
1722         return pci_vga_init(pci_bus) != NULL;
1723     default:
1724         error_setg(errp,
1725                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1726         return false;
1727     }
1728 }
1729 
1730 static int spapr_pre_load(void *opaque)
1731 {
1732     int rc;
1733 
1734     rc = spapr_caps_pre_load(opaque);
1735     if (rc) {
1736         return rc;
1737     }
1738 
1739     return 0;
1740 }
1741 
1742 static int spapr_post_load(void *opaque, int version_id)
1743 {
1744     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1745     int err = 0;
1746 
1747     err = spapr_caps_post_migration(spapr);
1748     if (err) {
1749         return err;
1750     }
1751 
1752     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1753         CPUState *cs;
1754         CPU_FOREACH(cs) {
1755             PowerPCCPU *cpu = POWERPC_CPU(cs);
1756             icp_resend(ICP(cpu->intc));
1757         }
1758     }
1759 
1760     /* In earlier versions, there was no separate qdev for the PAPR
1761      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1762      * So when migrating from those versions, poke the incoming offset
1763      * value into the RTC device */
1764     if (version_id < 3) {
1765         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1766     }
1767 
1768     if (kvm_enabled() && spapr->patb_entry) {
1769         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1770         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1771         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1772 
1773         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1774         if (err) {
1775             error_report("Process table config unsupported by the host");
1776             return -EINVAL;
1777         }
1778     }
1779 
1780     return err;
1781 }
1782 
1783 static int spapr_pre_save(void *opaque)
1784 {
1785     int rc;
1786 
1787     rc = spapr_caps_pre_save(opaque);
1788     if (rc) {
1789         return rc;
1790     }
1791 
1792     return 0;
1793 }
1794 
1795 static bool version_before_3(void *opaque, int version_id)
1796 {
1797     return version_id < 3;
1798 }
1799 
1800 static bool spapr_pending_events_needed(void *opaque)
1801 {
1802     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1803     return !QTAILQ_EMPTY(&spapr->pending_events);
1804 }
1805 
1806 static const VMStateDescription vmstate_spapr_event_entry = {
1807     .name = "spapr_event_log_entry",
1808     .version_id = 1,
1809     .minimum_version_id = 1,
1810     .fields = (VMStateField[]) {
1811         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1812         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1813         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1814                                      NULL, extended_length),
1815         VMSTATE_END_OF_LIST()
1816     },
1817 };
1818 
1819 static const VMStateDescription vmstate_spapr_pending_events = {
1820     .name = "spapr_pending_events",
1821     .version_id = 1,
1822     .minimum_version_id = 1,
1823     .needed = spapr_pending_events_needed,
1824     .fields = (VMStateField[]) {
1825         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1826                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1827         VMSTATE_END_OF_LIST()
1828     },
1829 };
1830 
1831 static bool spapr_ov5_cas_needed(void *opaque)
1832 {
1833     sPAPRMachineState *spapr = opaque;
1834     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1835     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1836     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1837     bool cas_needed;
1838 
1839     /* Prior to the introduction of sPAPROptionVector, we had two option
1840      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1841      * Both of these options encode machine topology into the device-tree
1842      * in such a way that the now-booted OS should still be able to interact
1843      * appropriately with QEMU regardless of what options were actually
1844      * negotiatied on the source side.
1845      *
1846      * As such, we can avoid migrating the CAS-negotiated options if these
1847      * are the only options available on the current machine/platform.
1848      * Since these are the only options available for pseries-2.7 and
1849      * earlier, this allows us to maintain old->new/new->old migration
1850      * compatibility.
1851      *
1852      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1853      * via default pseries-2.8 machines and explicit command-line parameters.
1854      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1855      * of the actual CAS-negotiated values to continue working properly. For
1856      * example, availability of memory unplug depends on knowing whether
1857      * OV5_HP_EVT was negotiated via CAS.
1858      *
1859      * Thus, for any cases where the set of available CAS-negotiatable
1860      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1861      * include the CAS-negotiated options in the migration stream, unless
1862      * if they affect boot time behaviour only.
1863      */
1864     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1865     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1866     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1867 
1868     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1869      * the mask itself since in the future it's possible "legacy" bits may be
1870      * removed via machine options, which could generate a false positive
1871      * that breaks migration.
1872      */
1873     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1874     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1875 
1876     spapr_ovec_cleanup(ov5_mask);
1877     spapr_ovec_cleanup(ov5_legacy);
1878     spapr_ovec_cleanup(ov5_removed);
1879 
1880     return cas_needed;
1881 }
1882 
1883 static const VMStateDescription vmstate_spapr_ov5_cas = {
1884     .name = "spapr_option_vector_ov5_cas",
1885     .version_id = 1,
1886     .minimum_version_id = 1,
1887     .needed = spapr_ov5_cas_needed,
1888     .fields = (VMStateField[]) {
1889         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1890                                  vmstate_spapr_ovec, sPAPROptionVector),
1891         VMSTATE_END_OF_LIST()
1892     },
1893 };
1894 
1895 static bool spapr_patb_entry_needed(void *opaque)
1896 {
1897     sPAPRMachineState *spapr = opaque;
1898 
1899     return !!spapr->patb_entry;
1900 }
1901 
1902 static const VMStateDescription vmstate_spapr_patb_entry = {
1903     .name = "spapr_patb_entry",
1904     .version_id = 1,
1905     .minimum_version_id = 1,
1906     .needed = spapr_patb_entry_needed,
1907     .fields = (VMStateField[]) {
1908         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1909         VMSTATE_END_OF_LIST()
1910     },
1911 };
1912 
1913 static const VMStateDescription vmstate_spapr = {
1914     .name = "spapr",
1915     .version_id = 3,
1916     .minimum_version_id = 1,
1917     .pre_load = spapr_pre_load,
1918     .post_load = spapr_post_load,
1919     .pre_save = spapr_pre_save,
1920     .fields = (VMStateField[]) {
1921         /* used to be @next_irq */
1922         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1923 
1924         /* RTC offset */
1925         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1926 
1927         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1928         VMSTATE_END_OF_LIST()
1929     },
1930     .subsections = (const VMStateDescription*[]) {
1931         &vmstate_spapr_ov5_cas,
1932         &vmstate_spapr_patb_entry,
1933         &vmstate_spapr_pending_events,
1934         &vmstate_spapr_cap_htm,
1935         &vmstate_spapr_cap_vsx,
1936         &vmstate_spapr_cap_dfp,
1937         &vmstate_spapr_cap_cfpc,
1938         &vmstate_spapr_cap_sbbc,
1939         &vmstate_spapr_cap_ibs,
1940         NULL
1941     }
1942 };
1943 
1944 static int htab_save_setup(QEMUFile *f, void *opaque)
1945 {
1946     sPAPRMachineState *spapr = opaque;
1947 
1948     /* "Iteration" header */
1949     if (!spapr->htab_shift) {
1950         qemu_put_be32(f, -1);
1951     } else {
1952         qemu_put_be32(f, spapr->htab_shift);
1953     }
1954 
1955     if (spapr->htab) {
1956         spapr->htab_save_index = 0;
1957         spapr->htab_first_pass = true;
1958     } else {
1959         if (spapr->htab_shift) {
1960             assert(kvm_enabled());
1961         }
1962     }
1963 
1964 
1965     return 0;
1966 }
1967 
1968 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1969                             int chunkstart, int n_valid, int n_invalid)
1970 {
1971     qemu_put_be32(f, chunkstart);
1972     qemu_put_be16(f, n_valid);
1973     qemu_put_be16(f, n_invalid);
1974     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1975                     HASH_PTE_SIZE_64 * n_valid);
1976 }
1977 
1978 static void htab_save_end_marker(QEMUFile *f)
1979 {
1980     qemu_put_be32(f, 0);
1981     qemu_put_be16(f, 0);
1982     qemu_put_be16(f, 0);
1983 }
1984 
1985 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1986                                  int64_t max_ns)
1987 {
1988     bool has_timeout = max_ns != -1;
1989     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1990     int index = spapr->htab_save_index;
1991     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1992 
1993     assert(spapr->htab_first_pass);
1994 
1995     do {
1996         int chunkstart;
1997 
1998         /* Consume invalid HPTEs */
1999         while ((index < htabslots)
2000                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2001             CLEAN_HPTE(HPTE(spapr->htab, index));
2002             index++;
2003         }
2004 
2005         /* Consume valid HPTEs */
2006         chunkstart = index;
2007         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2008                && HPTE_VALID(HPTE(spapr->htab, index))) {
2009             CLEAN_HPTE(HPTE(spapr->htab, index));
2010             index++;
2011         }
2012 
2013         if (index > chunkstart) {
2014             int n_valid = index - chunkstart;
2015 
2016             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2017 
2018             if (has_timeout &&
2019                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2020                 break;
2021             }
2022         }
2023     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2024 
2025     if (index >= htabslots) {
2026         assert(index == htabslots);
2027         index = 0;
2028         spapr->htab_first_pass = false;
2029     }
2030     spapr->htab_save_index = index;
2031 }
2032 
2033 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2034                                 int64_t max_ns)
2035 {
2036     bool final = max_ns < 0;
2037     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2038     int examined = 0, sent = 0;
2039     int index = spapr->htab_save_index;
2040     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2041 
2042     assert(!spapr->htab_first_pass);
2043 
2044     do {
2045         int chunkstart, invalidstart;
2046 
2047         /* Consume non-dirty HPTEs */
2048         while ((index < htabslots)
2049                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2050             index++;
2051             examined++;
2052         }
2053 
2054         chunkstart = index;
2055         /* Consume valid dirty HPTEs */
2056         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2057                && HPTE_DIRTY(HPTE(spapr->htab, index))
2058                && HPTE_VALID(HPTE(spapr->htab, index))) {
2059             CLEAN_HPTE(HPTE(spapr->htab, index));
2060             index++;
2061             examined++;
2062         }
2063 
2064         invalidstart = index;
2065         /* Consume invalid dirty HPTEs */
2066         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2067                && HPTE_DIRTY(HPTE(spapr->htab, index))
2068                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2069             CLEAN_HPTE(HPTE(spapr->htab, index));
2070             index++;
2071             examined++;
2072         }
2073 
2074         if (index > chunkstart) {
2075             int n_valid = invalidstart - chunkstart;
2076             int n_invalid = index - invalidstart;
2077 
2078             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2079             sent += index - chunkstart;
2080 
2081             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2082                 break;
2083             }
2084         }
2085 
2086         if (examined >= htabslots) {
2087             break;
2088         }
2089 
2090         if (index >= htabslots) {
2091             assert(index == htabslots);
2092             index = 0;
2093         }
2094     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2095 
2096     if (index >= htabslots) {
2097         assert(index == htabslots);
2098         index = 0;
2099     }
2100 
2101     spapr->htab_save_index = index;
2102 
2103     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2104 }
2105 
2106 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2107 #define MAX_KVM_BUF_SIZE    2048
2108 
2109 static int htab_save_iterate(QEMUFile *f, void *opaque)
2110 {
2111     sPAPRMachineState *spapr = opaque;
2112     int fd;
2113     int rc = 0;
2114 
2115     /* Iteration header */
2116     if (!spapr->htab_shift) {
2117         qemu_put_be32(f, -1);
2118         return 1;
2119     } else {
2120         qemu_put_be32(f, 0);
2121     }
2122 
2123     if (!spapr->htab) {
2124         assert(kvm_enabled());
2125 
2126         fd = get_htab_fd(spapr);
2127         if (fd < 0) {
2128             return fd;
2129         }
2130 
2131         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2132         if (rc < 0) {
2133             return rc;
2134         }
2135     } else  if (spapr->htab_first_pass) {
2136         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2137     } else {
2138         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2139     }
2140 
2141     htab_save_end_marker(f);
2142 
2143     return rc;
2144 }
2145 
2146 static int htab_save_complete(QEMUFile *f, void *opaque)
2147 {
2148     sPAPRMachineState *spapr = opaque;
2149     int fd;
2150 
2151     /* Iteration header */
2152     if (!spapr->htab_shift) {
2153         qemu_put_be32(f, -1);
2154         return 0;
2155     } else {
2156         qemu_put_be32(f, 0);
2157     }
2158 
2159     if (!spapr->htab) {
2160         int rc;
2161 
2162         assert(kvm_enabled());
2163 
2164         fd = get_htab_fd(spapr);
2165         if (fd < 0) {
2166             return fd;
2167         }
2168 
2169         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2170         if (rc < 0) {
2171             return rc;
2172         }
2173     } else {
2174         if (spapr->htab_first_pass) {
2175             htab_save_first_pass(f, spapr, -1);
2176         }
2177         htab_save_later_pass(f, spapr, -1);
2178     }
2179 
2180     /* End marker */
2181     htab_save_end_marker(f);
2182 
2183     return 0;
2184 }
2185 
2186 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2187 {
2188     sPAPRMachineState *spapr = opaque;
2189     uint32_t section_hdr;
2190     int fd = -1;
2191     Error *local_err = NULL;
2192 
2193     if (version_id < 1 || version_id > 1) {
2194         error_report("htab_load() bad version");
2195         return -EINVAL;
2196     }
2197 
2198     section_hdr = qemu_get_be32(f);
2199 
2200     if (section_hdr == -1) {
2201         spapr_free_hpt(spapr);
2202         return 0;
2203     }
2204 
2205     if (section_hdr) {
2206         /* First section gives the htab size */
2207         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2208         if (local_err) {
2209             error_report_err(local_err);
2210             return -EINVAL;
2211         }
2212         return 0;
2213     }
2214 
2215     if (!spapr->htab) {
2216         assert(kvm_enabled());
2217 
2218         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2219         if (fd < 0) {
2220             error_report_err(local_err);
2221             return fd;
2222         }
2223     }
2224 
2225     while (true) {
2226         uint32_t index;
2227         uint16_t n_valid, n_invalid;
2228 
2229         index = qemu_get_be32(f);
2230         n_valid = qemu_get_be16(f);
2231         n_invalid = qemu_get_be16(f);
2232 
2233         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2234             /* End of Stream */
2235             break;
2236         }
2237 
2238         if ((index + n_valid + n_invalid) >
2239             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2240             /* Bad index in stream */
2241             error_report(
2242                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2243                 index, n_valid, n_invalid, spapr->htab_shift);
2244             return -EINVAL;
2245         }
2246 
2247         if (spapr->htab) {
2248             if (n_valid) {
2249                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2250                                 HASH_PTE_SIZE_64 * n_valid);
2251             }
2252             if (n_invalid) {
2253                 memset(HPTE(spapr->htab, index + n_valid), 0,
2254                        HASH_PTE_SIZE_64 * n_invalid);
2255             }
2256         } else {
2257             int rc;
2258 
2259             assert(fd >= 0);
2260 
2261             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2262             if (rc < 0) {
2263                 return rc;
2264             }
2265         }
2266     }
2267 
2268     if (!spapr->htab) {
2269         assert(fd >= 0);
2270         close(fd);
2271     }
2272 
2273     return 0;
2274 }
2275 
2276 static void htab_save_cleanup(void *opaque)
2277 {
2278     sPAPRMachineState *spapr = opaque;
2279 
2280     close_htab_fd(spapr);
2281 }
2282 
2283 static SaveVMHandlers savevm_htab_handlers = {
2284     .save_setup = htab_save_setup,
2285     .save_live_iterate = htab_save_iterate,
2286     .save_live_complete_precopy = htab_save_complete,
2287     .save_cleanup = htab_save_cleanup,
2288     .load_state = htab_load,
2289 };
2290 
2291 static void spapr_boot_set(void *opaque, const char *boot_device,
2292                            Error **errp)
2293 {
2294     MachineState *machine = MACHINE(opaque);
2295     machine->boot_order = g_strdup(boot_device);
2296 }
2297 
2298 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2299 {
2300     MachineState *machine = MACHINE(spapr);
2301     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2302     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2303     int i;
2304 
2305     for (i = 0; i < nr_lmbs; i++) {
2306         uint64_t addr;
2307 
2308         addr = i * lmb_size + machine->device_memory->base;
2309         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2310                                addr / lmb_size);
2311     }
2312 }
2313 
2314 /*
2315  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2316  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2317  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2318  */
2319 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2320 {
2321     int i;
2322 
2323     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2324         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2325                    " is not aligned to %" PRIu64 " MiB",
2326                    machine->ram_size,
2327                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2328         return;
2329     }
2330 
2331     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2332         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2333                    " is not aligned to %" PRIu64 " MiB",
2334                    machine->ram_size,
2335                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2336         return;
2337     }
2338 
2339     for (i = 0; i < nb_numa_nodes; i++) {
2340         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2341             error_setg(errp,
2342                        "Node %d memory size 0x%" PRIx64
2343                        " is not aligned to %" PRIu64 " MiB",
2344                        i, numa_info[i].node_mem,
2345                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2346             return;
2347         }
2348     }
2349 }
2350 
2351 /* find cpu slot in machine->possible_cpus by core_id */
2352 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2353 {
2354     int index = id / smp_threads;
2355 
2356     if (index >= ms->possible_cpus->len) {
2357         return NULL;
2358     }
2359     if (idx) {
2360         *idx = index;
2361     }
2362     return &ms->possible_cpus->cpus[index];
2363 }
2364 
2365 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2366 {
2367     Error *local_err = NULL;
2368     bool vsmt_user = !!spapr->vsmt;
2369     int kvm_smt = kvmppc_smt_threads();
2370     int ret;
2371 
2372     if (!kvm_enabled() && (smp_threads > 1)) {
2373         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2374                      "on a pseries machine");
2375         goto out;
2376     }
2377     if (!is_power_of_2(smp_threads)) {
2378         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2379                      "machine because it must be a power of 2", smp_threads);
2380         goto out;
2381     }
2382 
2383     /* Detemine the VSMT mode to use: */
2384     if (vsmt_user) {
2385         if (spapr->vsmt < smp_threads) {
2386             error_setg(&local_err, "Cannot support VSMT mode %d"
2387                          " because it must be >= threads/core (%d)",
2388                          spapr->vsmt, smp_threads);
2389             goto out;
2390         }
2391         /* In this case, spapr->vsmt has been set by the command line */
2392     } else {
2393         /*
2394          * Default VSMT value is tricky, because we need it to be as
2395          * consistent as possible (for migration), but this requires
2396          * changing it for at least some existing cases.  We pick 8 as
2397          * the value that we'd get with KVM on POWER8, the
2398          * overwhelmingly common case in production systems.
2399          */
2400         spapr->vsmt = MAX(8, smp_threads);
2401     }
2402 
2403     /* KVM: If necessary, set the SMT mode: */
2404     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2405         ret = kvmppc_set_smt_threads(spapr->vsmt);
2406         if (ret) {
2407             /* Looks like KVM isn't able to change VSMT mode */
2408             error_setg(&local_err,
2409                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2410                        spapr->vsmt, ret);
2411             /* We can live with that if the default one is big enough
2412              * for the number of threads, and a submultiple of the one
2413              * we want.  In this case we'll waste some vcpu ids, but
2414              * behaviour will be correct */
2415             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2416                 warn_report_err(local_err);
2417                 local_err = NULL;
2418                 goto out;
2419             } else {
2420                 if (!vsmt_user) {
2421                     error_append_hint(&local_err,
2422                                       "On PPC, a VM with %d threads/core"
2423                                       " on a host with %d threads/core"
2424                                       " requires the use of VSMT mode %d.\n",
2425                                       smp_threads, kvm_smt, spapr->vsmt);
2426                 }
2427                 kvmppc_hint_smt_possible(&local_err);
2428                 goto out;
2429             }
2430         }
2431     }
2432     /* else TCG: nothing to do currently */
2433 out:
2434     error_propagate(errp, local_err);
2435 }
2436 
2437 static void spapr_init_cpus(sPAPRMachineState *spapr)
2438 {
2439     MachineState *machine = MACHINE(spapr);
2440     MachineClass *mc = MACHINE_GET_CLASS(machine);
2441     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2442     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2443     const CPUArchIdList *possible_cpus;
2444     int boot_cores_nr = smp_cpus / smp_threads;
2445     int i;
2446 
2447     possible_cpus = mc->possible_cpu_arch_ids(machine);
2448     if (mc->has_hotpluggable_cpus) {
2449         if (smp_cpus % smp_threads) {
2450             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2451                          smp_cpus, smp_threads);
2452             exit(1);
2453         }
2454         if (max_cpus % smp_threads) {
2455             error_report("max_cpus (%u) must be multiple of threads (%u)",
2456                          max_cpus, smp_threads);
2457             exit(1);
2458         }
2459     } else {
2460         if (max_cpus != smp_cpus) {
2461             error_report("This machine version does not support CPU hotplug");
2462             exit(1);
2463         }
2464         boot_cores_nr = possible_cpus->len;
2465     }
2466 
2467     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2468      * call xics_max_server_number() or spapr_vcpu_id().
2469      */
2470     spapr_set_vsmt_mode(spapr, &error_fatal);
2471 
2472     if (smc->pre_2_10_has_unused_icps) {
2473         int i;
2474 
2475         for (i = 0; i < xics_max_server_number(spapr); i++) {
2476             /* Dummy entries get deregistered when real ICPState objects
2477              * are registered during CPU core hotplug.
2478              */
2479             pre_2_10_vmstate_register_dummy_icp(i);
2480         }
2481     }
2482 
2483     for (i = 0; i < possible_cpus->len; i++) {
2484         int core_id = i * smp_threads;
2485 
2486         if (mc->has_hotpluggable_cpus) {
2487             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2488                                    spapr_vcpu_id(spapr, core_id));
2489         }
2490 
2491         if (i < boot_cores_nr) {
2492             Object *core  = object_new(type);
2493             int nr_threads = smp_threads;
2494 
2495             /* Handle the partially filled core for older machine types */
2496             if ((i + 1) * smp_threads >= smp_cpus) {
2497                 nr_threads = smp_cpus - i * smp_threads;
2498             }
2499 
2500             object_property_set_int(core, nr_threads, "nr-threads",
2501                                     &error_fatal);
2502             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2503                                     &error_fatal);
2504             object_property_set_bool(core, true, "realized", &error_fatal);
2505         }
2506     }
2507 }
2508 
2509 /* pSeries LPAR / sPAPR hardware init */
2510 static void spapr_machine_init(MachineState *machine)
2511 {
2512     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2513     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2514     const char *kernel_filename = machine->kernel_filename;
2515     const char *initrd_filename = machine->initrd_filename;
2516     PCIHostState *phb;
2517     int i;
2518     MemoryRegion *sysmem = get_system_memory();
2519     MemoryRegion *ram = g_new(MemoryRegion, 1);
2520     hwaddr node0_size = spapr_node0_size(machine);
2521     long load_limit, fw_size;
2522     char *filename;
2523     Error *resize_hpt_err = NULL;
2524 
2525     msi_nonbroken = true;
2526 
2527     QLIST_INIT(&spapr->phbs);
2528     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2529 
2530     /* Determine capabilities to run with */
2531     spapr_caps_init(spapr);
2532 
2533     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2534     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2535         /*
2536          * If the user explicitly requested a mode we should either
2537          * supply it, or fail completely (which we do below).  But if
2538          * it's not set explicitly, we reset our mode to something
2539          * that works
2540          */
2541         if (resize_hpt_err) {
2542             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2543             error_free(resize_hpt_err);
2544             resize_hpt_err = NULL;
2545         } else {
2546             spapr->resize_hpt = smc->resize_hpt_default;
2547         }
2548     }
2549 
2550     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2551 
2552     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2553         /*
2554          * User requested HPT resize, but this host can't supply it.  Bail out
2555          */
2556         error_report_err(resize_hpt_err);
2557         exit(1);
2558     }
2559 
2560     spapr->rma_size = node0_size;
2561 
2562     /* With KVM, we don't actually know whether KVM supports an
2563      * unbounded RMA (PR KVM) or is limited by the hash table size
2564      * (HV KVM using VRMA), so we always assume the latter
2565      *
2566      * In that case, we also limit the initial allocations for RTAS
2567      * etc... to 256M since we have no way to know what the VRMA size
2568      * is going to be as it depends on the size of the hash table
2569      * which isn't determined yet.
2570      */
2571     if (kvm_enabled()) {
2572         spapr->vrma_adjust = 1;
2573         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2574     }
2575 
2576     /* Actually we don't support unbounded RMA anymore since we added
2577      * proper emulation of HV mode. The max we can get is 16G which
2578      * also happens to be what we configure for PAPR mode so make sure
2579      * we don't do anything bigger than that
2580      */
2581     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2582 
2583     if (spapr->rma_size > node0_size) {
2584         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2585                      spapr->rma_size);
2586         exit(1);
2587     }
2588 
2589     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2590     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2591 
2592     /* Set up Interrupt Controller before we create the VCPUs */
2593     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2594 
2595     /* Set up containers for ibm,client-architecture-support negotiated options
2596      */
2597     spapr->ov5 = spapr_ovec_new();
2598     spapr->ov5_cas = spapr_ovec_new();
2599 
2600     if (smc->dr_lmb_enabled) {
2601         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2602         spapr_validate_node_memory(machine, &error_fatal);
2603     }
2604 
2605     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2606 
2607     /* advertise support for dedicated HP event source to guests */
2608     if (spapr->use_hotplug_event_source) {
2609         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2610     }
2611 
2612     /* advertise support for HPT resizing */
2613     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2614         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2615     }
2616 
2617     /* advertise support for ibm,dyamic-memory-v2 */
2618     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2619 
2620     /* init CPUs */
2621     spapr_init_cpus(spapr);
2622 
2623     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2624         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2625                               spapr->max_compat_pvr)) {
2626         /* KVM and TCG always allow GTSE with radix... */
2627         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2628     }
2629     /* ... but not with hash (currently). */
2630 
2631     if (kvm_enabled()) {
2632         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2633         kvmppc_enable_logical_ci_hcalls();
2634         kvmppc_enable_set_mode_hcall();
2635 
2636         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2637         kvmppc_enable_clear_ref_mod_hcalls();
2638     }
2639 
2640     /* allocate RAM */
2641     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2642                                          machine->ram_size);
2643     memory_region_add_subregion(sysmem, 0, ram);
2644 
2645     /* always allocate the device memory information */
2646     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2647 
2648     /* initialize hotplug memory address space */
2649     if (machine->ram_size < machine->maxram_size) {
2650         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2651         /*
2652          * Limit the number of hotpluggable memory slots to half the number
2653          * slots that KVM supports, leaving the other half for PCI and other
2654          * devices. However ensure that number of slots doesn't drop below 32.
2655          */
2656         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2657                            SPAPR_MAX_RAM_SLOTS;
2658 
2659         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2660             max_memslots = SPAPR_MAX_RAM_SLOTS;
2661         }
2662         if (machine->ram_slots > max_memslots) {
2663             error_report("Specified number of memory slots %"
2664                          PRIu64" exceeds max supported %d",
2665                          machine->ram_slots, max_memslots);
2666             exit(1);
2667         }
2668 
2669         machine->device_memory->base = ROUND_UP(machine->ram_size,
2670                                                 SPAPR_DEVICE_MEM_ALIGN);
2671         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2672                            "device-memory", device_mem_size);
2673         memory_region_add_subregion(sysmem, machine->device_memory->base,
2674                                     &machine->device_memory->mr);
2675     }
2676 
2677     if (smc->dr_lmb_enabled) {
2678         spapr_create_lmb_dr_connectors(spapr);
2679     }
2680 
2681     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2682     if (!filename) {
2683         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2684         exit(1);
2685     }
2686     spapr->rtas_size = get_image_size(filename);
2687     if (spapr->rtas_size < 0) {
2688         error_report("Could not get size of LPAR rtas '%s'", filename);
2689         exit(1);
2690     }
2691     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2692     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2693         error_report("Could not load LPAR rtas '%s'", filename);
2694         exit(1);
2695     }
2696     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2697         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2698                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2699         exit(1);
2700     }
2701     g_free(filename);
2702 
2703     /* Set up RTAS event infrastructure */
2704     spapr_events_init(spapr);
2705 
2706     /* Set up the RTC RTAS interfaces */
2707     spapr_rtc_create(spapr);
2708 
2709     /* Set up VIO bus */
2710     spapr->vio_bus = spapr_vio_bus_init();
2711 
2712     for (i = 0; i < serial_max_hds(); i++) {
2713         if (serial_hd(i)) {
2714             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2715         }
2716     }
2717 
2718     /* We always have at least the nvram device on VIO */
2719     spapr_create_nvram(spapr);
2720 
2721     /* Set up PCI */
2722     spapr_pci_rtas_init();
2723 
2724     phb = spapr_create_phb(spapr, 0);
2725 
2726     for (i = 0; i < nb_nics; i++) {
2727         NICInfo *nd = &nd_table[i];
2728 
2729         if (!nd->model) {
2730             nd->model = g_strdup("spapr-vlan");
2731         }
2732 
2733         if (g_str_equal(nd->model, "spapr-vlan") ||
2734             g_str_equal(nd->model, "ibmveth")) {
2735             spapr_vlan_create(spapr->vio_bus, nd);
2736         } else {
2737             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2738         }
2739     }
2740 
2741     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2742         spapr_vscsi_create(spapr->vio_bus);
2743     }
2744 
2745     /* Graphics */
2746     if (spapr_vga_init(phb->bus, &error_fatal)) {
2747         spapr->has_graphics = true;
2748         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2749     }
2750 
2751     if (machine->usb) {
2752         if (smc->use_ohci_by_default) {
2753             pci_create_simple(phb->bus, -1, "pci-ohci");
2754         } else {
2755             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2756         }
2757 
2758         if (spapr->has_graphics) {
2759             USBBus *usb_bus = usb_bus_find(-1);
2760 
2761             usb_create_simple(usb_bus, "usb-kbd");
2762             usb_create_simple(usb_bus, "usb-mouse");
2763         }
2764     }
2765 
2766     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2767         error_report(
2768             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2769             MIN_RMA_SLOF);
2770         exit(1);
2771     }
2772 
2773     if (kernel_filename) {
2774         uint64_t lowaddr = 0;
2775 
2776         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2777                                       NULL, NULL, &lowaddr, NULL, 1,
2778                                       PPC_ELF_MACHINE, 0, 0);
2779         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2780             spapr->kernel_size = load_elf(kernel_filename,
2781                                           translate_kernel_address, NULL, NULL,
2782                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2783                                           0, 0);
2784             spapr->kernel_le = spapr->kernel_size > 0;
2785         }
2786         if (spapr->kernel_size < 0) {
2787             error_report("error loading %s: %s", kernel_filename,
2788                          load_elf_strerror(spapr->kernel_size));
2789             exit(1);
2790         }
2791 
2792         /* load initrd */
2793         if (initrd_filename) {
2794             /* Try to locate the initrd in the gap between the kernel
2795              * and the firmware. Add a bit of space just in case
2796              */
2797             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2798                                   + 0x1ffff) & ~0xffff;
2799             spapr->initrd_size = load_image_targphys(initrd_filename,
2800                                                      spapr->initrd_base,
2801                                                      load_limit
2802                                                      - spapr->initrd_base);
2803             if (spapr->initrd_size < 0) {
2804                 error_report("could not load initial ram disk '%s'",
2805                              initrd_filename);
2806                 exit(1);
2807             }
2808         }
2809     }
2810 
2811     if (bios_name == NULL) {
2812         bios_name = FW_FILE_NAME;
2813     }
2814     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2815     if (!filename) {
2816         error_report("Could not find LPAR firmware '%s'", bios_name);
2817         exit(1);
2818     }
2819     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2820     if (fw_size <= 0) {
2821         error_report("Could not load LPAR firmware '%s'", filename);
2822         exit(1);
2823     }
2824     g_free(filename);
2825 
2826     /* FIXME: Should register things through the MachineState's qdev
2827      * interface, this is a legacy from the sPAPREnvironment structure
2828      * which predated MachineState but had a similar function */
2829     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2830     register_savevm_live(NULL, "spapr/htab", -1, 1,
2831                          &savevm_htab_handlers, spapr);
2832 
2833     qemu_register_boot_set(spapr_boot_set, spapr);
2834 
2835     if (kvm_enabled()) {
2836         /* to stop and start vmclock */
2837         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2838                                          &spapr->tb);
2839 
2840         kvmppc_spapr_enable_inkernel_multitce();
2841     }
2842 }
2843 
2844 static int spapr_kvm_type(const char *vm_type)
2845 {
2846     if (!vm_type) {
2847         return 0;
2848     }
2849 
2850     if (!strcmp(vm_type, "HV")) {
2851         return 1;
2852     }
2853 
2854     if (!strcmp(vm_type, "PR")) {
2855         return 2;
2856     }
2857 
2858     error_report("Unknown kvm-type specified '%s'", vm_type);
2859     exit(1);
2860 }
2861 
2862 /*
2863  * Implementation of an interface to adjust firmware path
2864  * for the bootindex property handling.
2865  */
2866 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2867                                    DeviceState *dev)
2868 {
2869 #define CAST(type, obj, name) \
2870     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2871     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2872     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2873     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2874 
2875     if (d) {
2876         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2877         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2878         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2879 
2880         if (spapr) {
2881             /*
2882              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2883              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2884              * in the top 16 bits of the 64-bit LUN
2885              */
2886             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2887             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2888                                    (uint64_t)id << 48);
2889         } else if (virtio) {
2890             /*
2891              * We use SRP luns of the form 01000000 | (target << 8) | lun
2892              * in the top 32 bits of the 64-bit LUN
2893              * Note: the quote above is from SLOF and it is wrong,
2894              * the actual binding is:
2895              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2896              */
2897             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2898             if (d->lun >= 256) {
2899                 /* Use the LUN "flat space addressing method" */
2900                 id |= 0x4000;
2901             }
2902             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2903                                    (uint64_t)id << 32);
2904         } else if (usb) {
2905             /*
2906              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2907              * in the top 32 bits of the 64-bit LUN
2908              */
2909             unsigned usb_port = atoi(usb->port->path);
2910             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2911             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2912                                    (uint64_t)id << 32);
2913         }
2914     }
2915 
2916     /*
2917      * SLOF probes the USB devices, and if it recognizes that the device is a
2918      * storage device, it changes its name to "storage" instead of "usb-host",
2919      * and additionally adds a child node for the SCSI LUN, so the correct
2920      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2921      */
2922     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2923         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2924         if (usb_host_dev_is_scsi_storage(usbdev)) {
2925             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2926         }
2927     }
2928 
2929     if (phb) {
2930         /* Replace "pci" with "pci@800000020000000" */
2931         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2932     }
2933 
2934     if (vsc) {
2935         /* Same logic as virtio above */
2936         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2937         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2938     }
2939 
2940     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2941         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2942         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2943         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2944     }
2945 
2946     return NULL;
2947 }
2948 
2949 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2950 {
2951     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2952 
2953     return g_strdup(spapr->kvm_type);
2954 }
2955 
2956 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2957 {
2958     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2959 
2960     g_free(spapr->kvm_type);
2961     spapr->kvm_type = g_strdup(value);
2962 }
2963 
2964 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2965 {
2966     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2967 
2968     return spapr->use_hotplug_event_source;
2969 }
2970 
2971 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2972                                             Error **errp)
2973 {
2974     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2975 
2976     spapr->use_hotplug_event_source = value;
2977 }
2978 
2979 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2980 {
2981     return true;
2982 }
2983 
2984 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2985 {
2986     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2987 
2988     switch (spapr->resize_hpt) {
2989     case SPAPR_RESIZE_HPT_DEFAULT:
2990         return g_strdup("default");
2991     case SPAPR_RESIZE_HPT_DISABLED:
2992         return g_strdup("disabled");
2993     case SPAPR_RESIZE_HPT_ENABLED:
2994         return g_strdup("enabled");
2995     case SPAPR_RESIZE_HPT_REQUIRED:
2996         return g_strdup("required");
2997     }
2998     g_assert_not_reached();
2999 }
3000 
3001 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3002 {
3003     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3004 
3005     if (strcmp(value, "default") == 0) {
3006         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3007     } else if (strcmp(value, "disabled") == 0) {
3008         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3009     } else if (strcmp(value, "enabled") == 0) {
3010         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3011     } else if (strcmp(value, "required") == 0) {
3012         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3013     } else {
3014         error_setg(errp, "Bad value for \"resize-hpt\" property");
3015     }
3016 }
3017 
3018 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3019                                    void *opaque, Error **errp)
3020 {
3021     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3022 }
3023 
3024 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3025                                    void *opaque, Error **errp)
3026 {
3027     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3028 }
3029 
3030 static void spapr_instance_init(Object *obj)
3031 {
3032     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3033 
3034     spapr->htab_fd = -1;
3035     spapr->use_hotplug_event_source = true;
3036     object_property_add_str(obj, "kvm-type",
3037                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3038     object_property_set_description(obj, "kvm-type",
3039                                     "Specifies the KVM virtualization mode (HV, PR)",
3040                                     NULL);
3041     object_property_add_bool(obj, "modern-hotplug-events",
3042                             spapr_get_modern_hotplug_events,
3043                             spapr_set_modern_hotplug_events,
3044                             NULL);
3045     object_property_set_description(obj, "modern-hotplug-events",
3046                                     "Use dedicated hotplug event mechanism in"
3047                                     " place of standard EPOW events when possible"
3048                                     " (required for memory hot-unplug support)",
3049                                     NULL);
3050     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3051                             "Maximum permitted CPU compatibility mode",
3052                             &error_fatal);
3053 
3054     object_property_add_str(obj, "resize-hpt",
3055                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3056     object_property_set_description(obj, "resize-hpt",
3057                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3058                                     NULL);
3059     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3060                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3061     object_property_set_description(obj, "vsmt",
3062                                     "Virtual SMT: KVM behaves as if this were"
3063                                     " the host's SMT mode", &error_abort);
3064     object_property_add_bool(obj, "vfio-no-msix-emulation",
3065                              spapr_get_msix_emulation, NULL, NULL);
3066 }
3067 
3068 static void spapr_machine_finalizefn(Object *obj)
3069 {
3070     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3071 
3072     g_free(spapr->kvm_type);
3073 }
3074 
3075 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3076 {
3077     cpu_synchronize_state(cs);
3078     ppc_cpu_do_system_reset(cs);
3079 }
3080 
3081 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3082 {
3083     CPUState *cs;
3084 
3085     CPU_FOREACH(cs) {
3086         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3087     }
3088 }
3089 
3090 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3091                            uint32_t node, bool dedicated_hp_event_source,
3092                            Error **errp)
3093 {
3094     sPAPRDRConnector *drc;
3095     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3096     int i, fdt_offset, fdt_size;
3097     void *fdt;
3098     uint64_t addr = addr_start;
3099     bool hotplugged = spapr_drc_hotplugged(dev);
3100     Error *local_err = NULL;
3101 
3102     for (i = 0; i < nr_lmbs; i++) {
3103         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3104                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3105         g_assert(drc);
3106 
3107         fdt = create_device_tree(&fdt_size);
3108         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3109                                                 SPAPR_MEMORY_BLOCK_SIZE);
3110 
3111         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3112         if (local_err) {
3113             while (addr > addr_start) {
3114                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3115                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3116                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3117                 spapr_drc_detach(drc);
3118             }
3119             g_free(fdt);
3120             error_propagate(errp, local_err);
3121             return;
3122         }
3123         if (!hotplugged) {
3124             spapr_drc_reset(drc);
3125         }
3126         addr += SPAPR_MEMORY_BLOCK_SIZE;
3127     }
3128     /* send hotplug notification to the
3129      * guest only in case of hotplugged memory
3130      */
3131     if (hotplugged) {
3132         if (dedicated_hp_event_source) {
3133             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3134                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3135             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3136                                                    nr_lmbs,
3137                                                    spapr_drc_index(drc));
3138         } else {
3139             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3140                                            nr_lmbs);
3141         }
3142     }
3143 }
3144 
3145 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3146                               Error **errp)
3147 {
3148     Error *local_err = NULL;
3149     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3150     PCDIMMDevice *dimm = PC_DIMM(dev);
3151     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3152     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3153     uint64_t align, size, addr;
3154     uint32_t node;
3155 
3156     align = memory_region_get_alignment(mr);
3157     size = memory_region_size(mr);
3158 
3159     pc_dimm_plug(dev, MACHINE(ms), align, &local_err);
3160     if (local_err) {
3161         goto out;
3162     }
3163 
3164     addr = object_property_get_uint(OBJECT(dimm),
3165                                     PC_DIMM_ADDR_PROP, &local_err);
3166     if (local_err) {
3167         goto out_unplug;
3168     }
3169 
3170     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3171                                     &error_abort);
3172     spapr_add_lmbs(dev, addr, size, node,
3173                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3174                    &local_err);
3175     if (local_err) {
3176         goto out_unplug;
3177     }
3178 
3179     return;
3180 
3181 out_unplug:
3182     pc_dimm_unplug(dev, MACHINE(ms));
3183 out:
3184     error_propagate(errp, local_err);
3185 }
3186 
3187 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3188                                   Error **errp)
3189 {
3190     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3191     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3192     PCDIMMDevice *dimm = PC_DIMM(dev);
3193     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3194     MemoryRegion *mr;
3195     uint64_t size;
3196     Object *memdev;
3197     hwaddr pagesize;
3198 
3199     if (!smc->dr_lmb_enabled) {
3200         error_setg(errp, "Memory hotplug not supported for this machine");
3201         return;
3202     }
3203 
3204     mr = ddc->get_memory_region(dimm, errp);
3205     if (!mr) {
3206         return;
3207     }
3208     size = memory_region_size(mr);
3209 
3210     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3211         error_setg(errp, "Hotplugged memory size must be a multiple of "
3212                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3213         return;
3214     }
3215 
3216     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3217                                       &error_abort);
3218     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3219     spapr_check_pagesize(spapr, pagesize, errp);
3220 }
3221 
3222 struct sPAPRDIMMState {
3223     PCDIMMDevice *dimm;
3224     uint32_t nr_lmbs;
3225     QTAILQ_ENTRY(sPAPRDIMMState) next;
3226 };
3227 
3228 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3229                                                        PCDIMMDevice *dimm)
3230 {
3231     sPAPRDIMMState *dimm_state = NULL;
3232 
3233     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3234         if (dimm_state->dimm == dimm) {
3235             break;
3236         }
3237     }
3238     return dimm_state;
3239 }
3240 
3241 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3242                                                       uint32_t nr_lmbs,
3243                                                       PCDIMMDevice *dimm)
3244 {
3245     sPAPRDIMMState *ds = NULL;
3246 
3247     /*
3248      * If this request is for a DIMM whose removal had failed earlier
3249      * (due to guest's refusal to remove the LMBs), we would have this
3250      * dimm already in the pending_dimm_unplugs list. In that
3251      * case don't add again.
3252      */
3253     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3254     if (!ds) {
3255         ds = g_malloc0(sizeof(sPAPRDIMMState));
3256         ds->nr_lmbs = nr_lmbs;
3257         ds->dimm = dimm;
3258         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3259     }
3260     return ds;
3261 }
3262 
3263 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3264                                               sPAPRDIMMState *dimm_state)
3265 {
3266     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3267     g_free(dimm_state);
3268 }
3269 
3270 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3271                                                         PCDIMMDevice *dimm)
3272 {
3273     sPAPRDRConnector *drc;
3274     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3275     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3276     uint64_t size = memory_region_size(mr);
3277     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3278     uint32_t avail_lmbs = 0;
3279     uint64_t addr_start, addr;
3280     int i;
3281 
3282     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3283                                          &error_abort);
3284 
3285     addr = addr_start;
3286     for (i = 0; i < nr_lmbs; i++) {
3287         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3288                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3289         g_assert(drc);
3290         if (drc->dev) {
3291             avail_lmbs++;
3292         }
3293         addr += SPAPR_MEMORY_BLOCK_SIZE;
3294     }
3295 
3296     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3297 }
3298 
3299 /* Callback to be called during DRC release. */
3300 void spapr_lmb_release(DeviceState *dev)
3301 {
3302     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3303     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3304     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3305 
3306     /* This information will get lost if a migration occurs
3307      * during the unplug process. In this case recover it. */
3308     if (ds == NULL) {
3309         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3310         g_assert(ds);
3311         /* The DRC being examined by the caller at least must be counted */
3312         g_assert(ds->nr_lmbs);
3313     }
3314 
3315     if (--ds->nr_lmbs) {
3316         return;
3317     }
3318 
3319     /*
3320      * Now that all the LMBs have been removed by the guest, call the
3321      * unplug handler chain. This can never fail.
3322      */
3323     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3324 }
3325 
3326 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3327 {
3328     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3329     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3330 
3331     pc_dimm_unplug(dev, MACHINE(hotplug_dev));
3332     object_unparent(OBJECT(dev));
3333     spapr_pending_dimm_unplugs_remove(spapr, ds);
3334 }
3335 
3336 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3337                                         DeviceState *dev, Error **errp)
3338 {
3339     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3340     Error *local_err = NULL;
3341     PCDIMMDevice *dimm = PC_DIMM(dev);
3342     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3343     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3344     uint32_t nr_lmbs;
3345     uint64_t size, addr_start, addr;
3346     int i;
3347     sPAPRDRConnector *drc;
3348 
3349     size = memory_region_size(mr);
3350     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3351 
3352     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3353                                          &local_err);
3354     if (local_err) {
3355         goto out;
3356     }
3357 
3358     /*
3359      * An existing pending dimm state for this DIMM means that there is an
3360      * unplug operation in progress, waiting for the spapr_lmb_release
3361      * callback to complete the job (BQL can't cover that far). In this case,
3362      * bail out to avoid detaching DRCs that were already released.
3363      */
3364     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3365         error_setg(&local_err,
3366                    "Memory unplug already in progress for device %s",
3367                    dev->id);
3368         goto out;
3369     }
3370 
3371     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3372 
3373     addr = addr_start;
3374     for (i = 0; i < nr_lmbs; i++) {
3375         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3376                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3377         g_assert(drc);
3378 
3379         spapr_drc_detach(drc);
3380         addr += SPAPR_MEMORY_BLOCK_SIZE;
3381     }
3382 
3383     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3384                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3385     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3386                                               nr_lmbs, spapr_drc_index(drc));
3387 out:
3388     error_propagate(errp, local_err);
3389 }
3390 
3391 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3392                                            sPAPRMachineState *spapr)
3393 {
3394     PowerPCCPU *cpu = POWERPC_CPU(cs);
3395     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3396     int id = spapr_get_vcpu_id(cpu);
3397     void *fdt;
3398     int offset, fdt_size;
3399     char *nodename;
3400 
3401     fdt = create_device_tree(&fdt_size);
3402     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3403     offset = fdt_add_subnode(fdt, 0, nodename);
3404 
3405     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3406     g_free(nodename);
3407 
3408     *fdt_offset = offset;
3409     return fdt;
3410 }
3411 
3412 /* Callback to be called during DRC release. */
3413 void spapr_core_release(DeviceState *dev)
3414 {
3415     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3416 
3417     /* Call the unplug handler chain. This can never fail. */
3418     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3419 }
3420 
3421 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3422 {
3423     MachineState *ms = MACHINE(hotplug_dev);
3424     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3425     CPUCore *cc = CPU_CORE(dev);
3426     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3427 
3428     if (smc->pre_2_10_has_unused_icps) {
3429         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3430         int i;
3431 
3432         for (i = 0; i < cc->nr_threads; i++) {
3433             CPUState *cs = CPU(sc->threads[i]);
3434 
3435             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3436         }
3437     }
3438 
3439     assert(core_slot);
3440     core_slot->cpu = NULL;
3441     object_unparent(OBJECT(dev));
3442 }
3443 
3444 static
3445 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3446                                Error **errp)
3447 {
3448     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3449     int index;
3450     sPAPRDRConnector *drc;
3451     CPUCore *cc = CPU_CORE(dev);
3452 
3453     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3454         error_setg(errp, "Unable to find CPU core with core-id: %d",
3455                    cc->core_id);
3456         return;
3457     }
3458     if (index == 0) {
3459         error_setg(errp, "Boot CPU core may not be unplugged");
3460         return;
3461     }
3462 
3463     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3464                           spapr_vcpu_id(spapr, cc->core_id));
3465     g_assert(drc);
3466 
3467     spapr_drc_detach(drc);
3468 
3469     spapr_hotplug_req_remove_by_index(drc);
3470 }
3471 
3472 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3473                             Error **errp)
3474 {
3475     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3476     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3477     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3478     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3479     CPUCore *cc = CPU_CORE(dev);
3480     CPUState *cs = CPU(core->threads[0]);
3481     sPAPRDRConnector *drc;
3482     Error *local_err = NULL;
3483     CPUArchId *core_slot;
3484     int index;
3485     bool hotplugged = spapr_drc_hotplugged(dev);
3486 
3487     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3488     if (!core_slot) {
3489         error_setg(errp, "Unable to find CPU core with core-id: %d",
3490                    cc->core_id);
3491         return;
3492     }
3493     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3494                           spapr_vcpu_id(spapr, cc->core_id));
3495 
3496     g_assert(drc || !mc->has_hotpluggable_cpus);
3497 
3498     if (drc) {
3499         void *fdt;
3500         int fdt_offset;
3501 
3502         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3503 
3504         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3505         if (local_err) {
3506             g_free(fdt);
3507             error_propagate(errp, local_err);
3508             return;
3509         }
3510 
3511         if (hotplugged) {
3512             /*
3513              * Send hotplug notification interrupt to the guest only
3514              * in case of hotplugged CPUs.
3515              */
3516             spapr_hotplug_req_add_by_index(drc);
3517         } else {
3518             spapr_drc_reset(drc);
3519         }
3520     }
3521 
3522     core_slot->cpu = OBJECT(dev);
3523 
3524     if (smc->pre_2_10_has_unused_icps) {
3525         int i;
3526 
3527         for (i = 0; i < cc->nr_threads; i++) {
3528             cs = CPU(core->threads[i]);
3529             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3530         }
3531     }
3532 }
3533 
3534 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3535                                 Error **errp)
3536 {
3537     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3538     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3539     Error *local_err = NULL;
3540     CPUCore *cc = CPU_CORE(dev);
3541     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3542     const char *type = object_get_typename(OBJECT(dev));
3543     CPUArchId *core_slot;
3544     int index;
3545 
3546     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3547         error_setg(&local_err, "CPU hotplug not supported for this machine");
3548         goto out;
3549     }
3550 
3551     if (strcmp(base_core_type, type)) {
3552         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3553         goto out;
3554     }
3555 
3556     if (cc->core_id % smp_threads) {
3557         error_setg(&local_err, "invalid core id %d", cc->core_id);
3558         goto out;
3559     }
3560 
3561     /*
3562      * In general we should have homogeneous threads-per-core, but old
3563      * (pre hotplug support) machine types allow the last core to have
3564      * reduced threads as a compatibility hack for when we allowed
3565      * total vcpus not a multiple of threads-per-core.
3566      */
3567     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3568         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3569                    cc->nr_threads, smp_threads);
3570         goto out;
3571     }
3572 
3573     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3574     if (!core_slot) {
3575         error_setg(&local_err, "core id %d out of range", cc->core_id);
3576         goto out;
3577     }
3578 
3579     if (core_slot->cpu) {
3580         error_setg(&local_err, "core %d already populated", cc->core_id);
3581         goto out;
3582     }
3583 
3584     numa_cpu_pre_plug(core_slot, dev, &local_err);
3585 
3586 out:
3587     error_propagate(errp, local_err);
3588 }
3589 
3590 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3591                                       DeviceState *dev, Error **errp)
3592 {
3593     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3594         spapr_memory_plug(hotplug_dev, dev, errp);
3595     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3596         spapr_core_plug(hotplug_dev, dev, errp);
3597     }
3598 }
3599 
3600 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3601                                         DeviceState *dev, Error **errp)
3602 {
3603     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3604         spapr_memory_unplug(hotplug_dev, dev);
3605     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3606         spapr_core_unplug(hotplug_dev, dev);
3607     }
3608 }
3609 
3610 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3611                                                 DeviceState *dev, Error **errp)
3612 {
3613     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3614     MachineClass *mc = MACHINE_GET_CLASS(sms);
3615 
3616     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3617         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3618             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3619         } else {
3620             /* NOTE: this means there is a window after guest reset, prior to
3621              * CAS negotiation, where unplug requests will fail due to the
3622              * capability not being detected yet. This is a bit different than
3623              * the case with PCI unplug, where the events will be queued and
3624              * eventually handled by the guest after boot
3625              */
3626             error_setg(errp, "Memory hot unplug not supported for this guest");
3627         }
3628     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3629         if (!mc->has_hotpluggable_cpus) {
3630             error_setg(errp, "CPU hot unplug not supported on this machine");
3631             return;
3632         }
3633         spapr_core_unplug_request(hotplug_dev, dev, errp);
3634     }
3635 }
3636 
3637 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3638                                           DeviceState *dev, Error **errp)
3639 {
3640     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3641         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3642     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3643         spapr_core_pre_plug(hotplug_dev, dev, errp);
3644     }
3645 }
3646 
3647 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3648                                                  DeviceState *dev)
3649 {
3650     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3651         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3652         return HOTPLUG_HANDLER(machine);
3653     }
3654     return NULL;
3655 }
3656 
3657 static CpuInstanceProperties
3658 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3659 {
3660     CPUArchId *core_slot;
3661     MachineClass *mc = MACHINE_GET_CLASS(machine);
3662 
3663     /* make sure possible_cpu are intialized */
3664     mc->possible_cpu_arch_ids(machine);
3665     /* get CPU core slot containing thread that matches cpu_index */
3666     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3667     assert(core_slot);
3668     return core_slot->props;
3669 }
3670 
3671 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3672 {
3673     return idx / smp_cores % nb_numa_nodes;
3674 }
3675 
3676 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3677 {
3678     int i;
3679     const char *core_type;
3680     int spapr_max_cores = max_cpus / smp_threads;
3681     MachineClass *mc = MACHINE_GET_CLASS(machine);
3682 
3683     if (!mc->has_hotpluggable_cpus) {
3684         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3685     }
3686     if (machine->possible_cpus) {
3687         assert(machine->possible_cpus->len == spapr_max_cores);
3688         return machine->possible_cpus;
3689     }
3690 
3691     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3692     if (!core_type) {
3693         error_report("Unable to find sPAPR CPU Core definition");
3694         exit(1);
3695     }
3696 
3697     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3698                              sizeof(CPUArchId) * spapr_max_cores);
3699     machine->possible_cpus->len = spapr_max_cores;
3700     for (i = 0; i < machine->possible_cpus->len; i++) {
3701         int core_id = i * smp_threads;
3702 
3703         machine->possible_cpus->cpus[i].type = core_type;
3704         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3705         machine->possible_cpus->cpus[i].arch_id = core_id;
3706         machine->possible_cpus->cpus[i].props.has_core_id = true;
3707         machine->possible_cpus->cpus[i].props.core_id = core_id;
3708     }
3709     return machine->possible_cpus;
3710 }
3711 
3712 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3713                                 uint64_t *buid, hwaddr *pio,
3714                                 hwaddr *mmio32, hwaddr *mmio64,
3715                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3716 {
3717     /*
3718      * New-style PHB window placement.
3719      *
3720      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3721      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3722      * windows.
3723      *
3724      * Some guest kernels can't work with MMIO windows above 1<<46
3725      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3726      *
3727      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3728      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3729      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3730      * 1TiB 64-bit MMIO windows for each PHB.
3731      */
3732     const uint64_t base_buid = 0x800000020000000ULL;
3733 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3734                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3735     int i;
3736 
3737     /* Sanity check natural alignments */
3738     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3739     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3740     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3741     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3742     /* Sanity check bounds */
3743     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3744                       SPAPR_PCI_MEM32_WIN_SIZE);
3745     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3746                       SPAPR_PCI_MEM64_WIN_SIZE);
3747 
3748     if (index >= SPAPR_MAX_PHBS) {
3749         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3750                    SPAPR_MAX_PHBS - 1);
3751         return;
3752     }
3753 
3754     *buid = base_buid + index;
3755     for (i = 0; i < n_dma; ++i) {
3756         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3757     }
3758 
3759     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3760     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3761     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3762 }
3763 
3764 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3765 {
3766     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3767 
3768     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3769 }
3770 
3771 static void spapr_ics_resend(XICSFabric *dev)
3772 {
3773     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3774 
3775     ics_resend(spapr->ics);
3776 }
3777 
3778 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3779 {
3780     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3781 
3782     return cpu ? ICP(cpu->intc) : NULL;
3783 }
3784 
3785 #define ICS_IRQ_FREE(ics, srcno)   \
3786     (!((ics)->irqs[(srcno)].flags & (XICS_FLAGS_IRQ_MASK)))
3787 
3788 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
3789 {
3790     int first, i;
3791 
3792     for (first = 0; first < ics->nr_irqs; first += alignnum) {
3793         if (num > (ics->nr_irqs - first)) {
3794             return -1;
3795         }
3796         for (i = first; i < first + num; ++i) {
3797             if (!ICS_IRQ_FREE(ics, i)) {
3798                 break;
3799             }
3800         }
3801         if (i == (first + num)) {
3802             return first;
3803         }
3804     }
3805 
3806     return -1;
3807 }
3808 
3809 int spapr_irq_find(sPAPRMachineState *spapr, int num, bool align, Error **errp)
3810 {
3811     ICSState *ics = spapr->ics;
3812     int first = -1;
3813 
3814     assert(ics);
3815 
3816     /*
3817      * MSIMesage::data is used for storing VIRQ so
3818      * it has to be aligned to num to support multiple
3819      * MSI vectors. MSI-X is not affected by this.
3820      * The hint is used for the first IRQ, the rest should
3821      * be allocated continuously.
3822      */
3823     if (align) {
3824         assert((num == 1) || (num == 2) || (num == 4) ||
3825                (num == 8) || (num == 16) || (num == 32));
3826         first = ics_find_free_block(ics, num, num);
3827     } else {
3828         first = ics_find_free_block(ics, num, 1);
3829     }
3830 
3831     if (first < 0) {
3832         error_setg(errp, "can't find a free %d-IRQ block", num);
3833         return -1;
3834     }
3835 
3836     return first + ics->offset;
3837 }
3838 
3839 int spapr_irq_claim(sPAPRMachineState *spapr, int irq, bool lsi, Error **errp)
3840 {
3841     ICSState *ics = spapr->ics;
3842 
3843     assert(ics);
3844 
3845     if (!ics_valid_irq(ics, irq)) {
3846         error_setg(errp, "IRQ %d is invalid", irq);
3847         return -1;
3848     }
3849 
3850     if (!ICS_IRQ_FREE(ics, irq - ics->offset)) {
3851         error_setg(errp, "IRQ %d is not free", irq);
3852         return -1;
3853     }
3854 
3855     ics_set_irq_type(ics, irq - ics->offset, lsi);
3856     return 0;
3857 }
3858 
3859 void spapr_irq_free(sPAPRMachineState *spapr, int irq, int num)
3860 {
3861     ICSState *ics = spapr->ics;
3862     int srcno = irq - ics->offset;
3863     int i;
3864 
3865     if (ics_valid_irq(ics, irq)) {
3866         trace_spapr_irq_free(0, irq, num);
3867         for (i = srcno; i < srcno + num; ++i) {
3868             if (ICS_IRQ_FREE(ics, i)) {
3869                 trace_spapr_irq_free_warn(0, i + ics->offset);
3870             }
3871             memset(&ics->irqs[i], 0, sizeof(ICSIRQState));
3872         }
3873     }
3874 }
3875 
3876 qemu_irq spapr_qirq(sPAPRMachineState *spapr, int irq)
3877 {
3878     ICSState *ics = spapr->ics;
3879 
3880     if (ics_valid_irq(ics, irq)) {
3881         return ics->qirqs[irq - ics->offset];
3882     }
3883 
3884     return NULL;
3885 }
3886 
3887 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3888                                  Monitor *mon)
3889 {
3890     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3891     CPUState *cs;
3892 
3893     CPU_FOREACH(cs) {
3894         PowerPCCPU *cpu = POWERPC_CPU(cs);
3895 
3896         icp_pic_print_info(ICP(cpu->intc), mon);
3897     }
3898 
3899     ics_pic_print_info(spapr->ics, mon);
3900 }
3901 
3902 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3903 {
3904     return cpu->vcpu_id;
3905 }
3906 
3907 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3908 {
3909     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3910     int vcpu_id;
3911 
3912     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3913 
3914     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3915         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3916         error_append_hint(errp, "Adjust the number of cpus to %d "
3917                           "or try to raise the number of threads per core\n",
3918                           vcpu_id * smp_threads / spapr->vsmt);
3919         return;
3920     }
3921 
3922     cpu->vcpu_id = vcpu_id;
3923 }
3924 
3925 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3926 {
3927     CPUState *cs;
3928 
3929     CPU_FOREACH(cs) {
3930         PowerPCCPU *cpu = POWERPC_CPU(cs);
3931 
3932         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3933             return cpu;
3934         }
3935     }
3936 
3937     return NULL;
3938 }
3939 
3940 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3941 {
3942     MachineClass *mc = MACHINE_CLASS(oc);
3943     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3944     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3945     NMIClass *nc = NMI_CLASS(oc);
3946     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3947     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3948     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3949     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3950 
3951     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3952     mc->ignore_boot_device_suffixes = true;
3953 
3954     /*
3955      * We set up the default / latest behaviour here.  The class_init
3956      * functions for the specific versioned machine types can override
3957      * these details for backwards compatibility
3958      */
3959     mc->init = spapr_machine_init;
3960     mc->reset = spapr_machine_reset;
3961     mc->block_default_type = IF_SCSI;
3962     mc->max_cpus = 1024;
3963     mc->no_parallel = 1;
3964     mc->default_boot_order = "";
3965     mc->default_ram_size = 512 * MiB;
3966     mc->default_display = "std";
3967     mc->kvm_type = spapr_kvm_type;
3968     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3969     mc->pci_allow_0_address = true;
3970     assert(!mc->get_hotplug_handler);
3971     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3972     hc->pre_plug = spapr_machine_device_pre_plug;
3973     hc->plug = spapr_machine_device_plug;
3974     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3975     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3976     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3977     hc->unplug_request = spapr_machine_device_unplug_request;
3978     hc->unplug = spapr_machine_device_unplug;
3979 
3980     smc->dr_lmb_enabled = true;
3981     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3982     mc->has_hotpluggable_cpus = true;
3983     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3984     fwc->get_dev_path = spapr_get_fw_dev_path;
3985     nc->nmi_monitor_handler = spapr_nmi;
3986     smc->phb_placement = spapr_phb_placement;
3987     vhc->hypercall = emulate_spapr_hypercall;
3988     vhc->hpt_mask = spapr_hpt_mask;
3989     vhc->map_hptes = spapr_map_hptes;
3990     vhc->unmap_hptes = spapr_unmap_hptes;
3991     vhc->store_hpte = spapr_store_hpte;
3992     vhc->get_patbe = spapr_get_patbe;
3993     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3994     xic->ics_get = spapr_ics_get;
3995     xic->ics_resend = spapr_ics_resend;
3996     xic->icp_get = spapr_icp_get;
3997     ispc->print_info = spapr_pic_print_info;
3998     /* Force NUMA node memory size to be a multiple of
3999      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4000      * in which LMBs are represented and hot-added
4001      */
4002     mc->numa_mem_align_shift = 28;
4003 
4004     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4005     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4006     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4007     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4008     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4009     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4010     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4011     spapr_caps_add_properties(smc, &error_abort);
4012 }
4013 
4014 static const TypeInfo spapr_machine_info = {
4015     .name          = TYPE_SPAPR_MACHINE,
4016     .parent        = TYPE_MACHINE,
4017     .abstract      = true,
4018     .instance_size = sizeof(sPAPRMachineState),
4019     .instance_init = spapr_instance_init,
4020     .instance_finalize = spapr_machine_finalizefn,
4021     .class_size    = sizeof(sPAPRMachineClass),
4022     .class_init    = spapr_machine_class_init,
4023     .interfaces = (InterfaceInfo[]) {
4024         { TYPE_FW_PATH_PROVIDER },
4025         { TYPE_NMI },
4026         { TYPE_HOTPLUG_HANDLER },
4027         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4028         { TYPE_XICS_FABRIC },
4029         { TYPE_INTERRUPT_STATS_PROVIDER },
4030         { }
4031     },
4032 };
4033 
4034 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4035     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4036                                                     void *data)      \
4037     {                                                                \
4038         MachineClass *mc = MACHINE_CLASS(oc);                        \
4039         spapr_machine_##suffix##_class_options(mc);                  \
4040         if (latest) {                                                \
4041             mc->alias = "pseries";                                   \
4042             mc->is_default = 1;                                      \
4043         }                                                            \
4044     }                                                                \
4045     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
4046     {                                                                \
4047         MachineState *machine = MACHINE(obj);                        \
4048         spapr_machine_##suffix##_instance_options(machine);          \
4049     }                                                                \
4050     static const TypeInfo spapr_machine_##suffix##_info = {          \
4051         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4052         .parent = TYPE_SPAPR_MACHINE,                                \
4053         .class_init = spapr_machine_##suffix##_class_init,           \
4054         .instance_init = spapr_machine_##suffix##_instance_init,     \
4055     };                                                               \
4056     static void spapr_machine_register_##suffix(void)                \
4057     {                                                                \
4058         type_register(&spapr_machine_##suffix##_info);               \
4059     }                                                                \
4060     type_init(spapr_machine_register_##suffix)
4061 
4062 /*
4063  * pseries-3.0
4064  */
4065 static void spapr_machine_3_0_instance_options(MachineState *machine)
4066 {
4067 }
4068 
4069 static void spapr_machine_3_0_class_options(MachineClass *mc)
4070 {
4071     /* Defaults for the latest behaviour inherited from the base class */
4072 }
4073 
4074 DEFINE_SPAPR_MACHINE(3_0, "3.0", true);
4075 
4076 /*
4077  * pseries-2.12
4078  */
4079 #define SPAPR_COMPAT_2_12                                              \
4080     HW_COMPAT_2_12                                                     \
4081     {                                                                  \
4082         .driver = TYPE_POWERPC_CPU,                                    \
4083         .property = "pre-3.0-migration",                               \
4084         .value    = "on",                                              \
4085     },                                                                 \
4086     {                                                                  \
4087         .driver = TYPE_SPAPR_CPU_CORE,                                 \
4088         .property = "pre-3.0-migration",                               \
4089         .value    = "on",                                              \
4090     },
4091 
4092 static void spapr_machine_2_12_instance_options(MachineState *machine)
4093 {
4094     spapr_machine_3_0_instance_options(machine);
4095 }
4096 
4097 static void spapr_machine_2_12_class_options(MachineClass *mc)
4098 {
4099     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4100 
4101     spapr_machine_3_0_class_options(mc);
4102     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4103 
4104     /* We depend on kvm_enabled() to choose a default value for the
4105      * hpt-max-page-size capability. Of course we can't do it here
4106      * because this is too early and the HW accelerator isn't initialzed
4107      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4108      */
4109     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4110 }
4111 
4112 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4113 
4114 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
4115 {
4116     spapr_machine_2_12_instance_options(machine);
4117 }
4118 
4119 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4120 {
4121     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4122 
4123     spapr_machine_2_12_class_options(mc);
4124     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4125     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4126     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4127 }
4128 
4129 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4130 
4131 /*
4132  * pseries-2.11
4133  */
4134 #define SPAPR_COMPAT_2_11                                              \
4135     HW_COMPAT_2_11
4136 
4137 static void spapr_machine_2_11_instance_options(MachineState *machine)
4138 {
4139     spapr_machine_2_12_instance_options(machine);
4140 }
4141 
4142 static void spapr_machine_2_11_class_options(MachineClass *mc)
4143 {
4144     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4145 
4146     spapr_machine_2_12_class_options(mc);
4147     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4148     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4149 }
4150 
4151 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4152 
4153 /*
4154  * pseries-2.10
4155  */
4156 #define SPAPR_COMPAT_2_10                                              \
4157     HW_COMPAT_2_10
4158 
4159 static void spapr_machine_2_10_instance_options(MachineState *machine)
4160 {
4161     spapr_machine_2_11_instance_options(machine);
4162 }
4163 
4164 static void spapr_machine_2_10_class_options(MachineClass *mc)
4165 {
4166     spapr_machine_2_11_class_options(mc);
4167     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4168 }
4169 
4170 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4171 
4172 /*
4173  * pseries-2.9
4174  */
4175 #define SPAPR_COMPAT_2_9                                               \
4176     HW_COMPAT_2_9                                                      \
4177     {                                                                  \
4178         .driver = TYPE_POWERPC_CPU,                                    \
4179         .property = "pre-2.10-migration",                              \
4180         .value    = "on",                                              \
4181     },                                                                 \
4182 
4183 static void spapr_machine_2_9_instance_options(MachineState *machine)
4184 {
4185     spapr_machine_2_10_instance_options(machine);
4186 }
4187 
4188 static void spapr_machine_2_9_class_options(MachineClass *mc)
4189 {
4190     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4191 
4192     spapr_machine_2_10_class_options(mc);
4193     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4194     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4195     smc->pre_2_10_has_unused_icps = true;
4196     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4197 }
4198 
4199 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4200 
4201 /*
4202  * pseries-2.8
4203  */
4204 #define SPAPR_COMPAT_2_8                                        \
4205     HW_COMPAT_2_8                                               \
4206     {                                                           \
4207         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4208         .property = "pcie-extended-configuration-space",        \
4209         .value    = "off",                                      \
4210     },
4211 
4212 static void spapr_machine_2_8_instance_options(MachineState *machine)
4213 {
4214     spapr_machine_2_9_instance_options(machine);
4215 }
4216 
4217 static void spapr_machine_2_8_class_options(MachineClass *mc)
4218 {
4219     spapr_machine_2_9_class_options(mc);
4220     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4221     mc->numa_mem_align_shift = 23;
4222 }
4223 
4224 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4225 
4226 /*
4227  * pseries-2.7
4228  */
4229 #define SPAPR_COMPAT_2_7                            \
4230     HW_COMPAT_2_7                                   \
4231     {                                               \
4232         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4233         .property = "mem_win_size",                 \
4234         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4235     },                                              \
4236     {                                               \
4237         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4238         .property = "mem64_win_size",               \
4239         .value    = "0",                            \
4240     },                                              \
4241     {                                               \
4242         .driver = TYPE_POWERPC_CPU,                 \
4243         .property = "pre-2.8-migration",            \
4244         .value    = "on",                           \
4245     },                                              \
4246     {                                               \
4247         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4248         .property = "pre-2.8-migration",            \
4249         .value    = "on",                           \
4250     },
4251 
4252 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4253                               uint64_t *buid, hwaddr *pio,
4254                               hwaddr *mmio32, hwaddr *mmio64,
4255                               unsigned n_dma, uint32_t *liobns, Error **errp)
4256 {
4257     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4258     const uint64_t base_buid = 0x800000020000000ULL;
4259     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4260     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4261     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4262     const uint32_t max_index = 255;
4263     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4264 
4265     uint64_t ram_top = MACHINE(spapr)->ram_size;
4266     hwaddr phb0_base, phb_base;
4267     int i;
4268 
4269     /* Do we have device memory? */
4270     if (MACHINE(spapr)->maxram_size > ram_top) {
4271         /* Can't just use maxram_size, because there may be an
4272          * alignment gap between normal and device memory regions
4273          */
4274         ram_top = MACHINE(spapr)->device_memory->base +
4275             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4276     }
4277 
4278     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4279 
4280     if (index > max_index) {
4281         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4282                    max_index);
4283         return;
4284     }
4285 
4286     *buid = base_buid + index;
4287     for (i = 0; i < n_dma; ++i) {
4288         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4289     }
4290 
4291     phb_base = phb0_base + index * phb_spacing;
4292     *pio = phb_base + pio_offset;
4293     *mmio32 = phb_base + mmio_offset;
4294     /*
4295      * We don't set the 64-bit MMIO window, relying on the PHB's
4296      * fallback behaviour of automatically splitting a large "32-bit"
4297      * window into contiguous 32-bit and 64-bit windows
4298      */
4299 }
4300 
4301 static void spapr_machine_2_7_instance_options(MachineState *machine)
4302 {
4303     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4304 
4305     spapr_machine_2_8_instance_options(machine);
4306     spapr->use_hotplug_event_source = false;
4307 }
4308 
4309 static void spapr_machine_2_7_class_options(MachineClass *mc)
4310 {
4311     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4312 
4313     spapr_machine_2_8_class_options(mc);
4314     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4315     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4316     smc->phb_placement = phb_placement_2_7;
4317 }
4318 
4319 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4320 
4321 /*
4322  * pseries-2.6
4323  */
4324 #define SPAPR_COMPAT_2_6 \
4325     HW_COMPAT_2_6 \
4326     { \
4327         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4328         .property = "ddw",\
4329         .value    = stringify(off),\
4330     },
4331 
4332 static void spapr_machine_2_6_instance_options(MachineState *machine)
4333 {
4334     spapr_machine_2_7_instance_options(machine);
4335 }
4336 
4337 static void spapr_machine_2_6_class_options(MachineClass *mc)
4338 {
4339     spapr_machine_2_7_class_options(mc);
4340     mc->has_hotpluggable_cpus = false;
4341     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4342 }
4343 
4344 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4345 
4346 /*
4347  * pseries-2.5
4348  */
4349 #define SPAPR_COMPAT_2_5 \
4350     HW_COMPAT_2_5 \
4351     { \
4352         .driver   = "spapr-vlan", \
4353         .property = "use-rx-buffer-pools", \
4354         .value    = "off", \
4355     },
4356 
4357 static void spapr_machine_2_5_instance_options(MachineState *machine)
4358 {
4359     spapr_machine_2_6_instance_options(machine);
4360 }
4361 
4362 static void spapr_machine_2_5_class_options(MachineClass *mc)
4363 {
4364     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4365 
4366     spapr_machine_2_6_class_options(mc);
4367     smc->use_ohci_by_default = true;
4368     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4369 }
4370 
4371 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4372 
4373 /*
4374  * pseries-2.4
4375  */
4376 #define SPAPR_COMPAT_2_4 \
4377         HW_COMPAT_2_4
4378 
4379 static void spapr_machine_2_4_instance_options(MachineState *machine)
4380 {
4381     spapr_machine_2_5_instance_options(machine);
4382 }
4383 
4384 static void spapr_machine_2_4_class_options(MachineClass *mc)
4385 {
4386     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4387 
4388     spapr_machine_2_5_class_options(mc);
4389     smc->dr_lmb_enabled = false;
4390     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4391 }
4392 
4393 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4394 
4395 /*
4396  * pseries-2.3
4397  */
4398 #define SPAPR_COMPAT_2_3 \
4399         HW_COMPAT_2_3 \
4400         {\
4401             .driver   = "spapr-pci-host-bridge",\
4402             .property = "dynamic-reconfiguration",\
4403             .value    = "off",\
4404         },
4405 
4406 static void spapr_machine_2_3_instance_options(MachineState *machine)
4407 {
4408     spapr_machine_2_4_instance_options(machine);
4409 }
4410 
4411 static void spapr_machine_2_3_class_options(MachineClass *mc)
4412 {
4413     spapr_machine_2_4_class_options(mc);
4414     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4415 }
4416 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4417 
4418 /*
4419  * pseries-2.2
4420  */
4421 
4422 #define SPAPR_COMPAT_2_2 \
4423         HW_COMPAT_2_2 \
4424         {\
4425             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4426             .property = "mem_win_size",\
4427             .value    = "0x20000000",\
4428         },
4429 
4430 static void spapr_machine_2_2_instance_options(MachineState *machine)
4431 {
4432     spapr_machine_2_3_instance_options(machine);
4433     machine->suppress_vmdesc = true;
4434 }
4435 
4436 static void spapr_machine_2_2_class_options(MachineClass *mc)
4437 {
4438     spapr_machine_2_3_class_options(mc);
4439     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4440 }
4441 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4442 
4443 /*
4444  * pseries-2.1
4445  */
4446 #define SPAPR_COMPAT_2_1 \
4447         HW_COMPAT_2_1
4448 
4449 static void spapr_machine_2_1_instance_options(MachineState *machine)
4450 {
4451     spapr_machine_2_2_instance_options(machine);
4452 }
4453 
4454 static void spapr_machine_2_1_class_options(MachineClass *mc)
4455 {
4456     spapr_machine_2_2_class_options(mc);
4457     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4458 }
4459 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4460 
4461 static void spapr_machine_register_types(void)
4462 {
4463     type_register_static(&spapr_machine_info);
4464 }
4465 
4466 type_init(spapr_machine_register_types)
4467