xref: /openbmc/qemu/hw/ppc/spapr.c (revision 88c725c7)
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 "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
31 #include "hw/hw.h"
32 #include "qemu/log.h"
33 #include "hw/fw-path-provider.h"
34 #include "elf.h"
35 #include "net/net.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.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 "qom/cpu.h"
47 
48 #include "hw/boards.h"
49 #include "hw/ppc/ppc.h"
50 #include "hw/loader.h"
51 
52 #include "hw/ppc/fdt.h"
53 #include "hw/ppc/spapr.h"
54 #include "hw/ppc/spapr_vio.h"
55 #include "hw/pci-host/spapr.h"
56 #include "hw/ppc/xics.h"
57 #include "hw/pci/msi.h"
58 
59 #include "hw/pci/pci.h"
60 #include "hw/scsi/scsi.h"
61 #include "hw/virtio/virtio-scsi.h"
62 #include "hw/virtio/vhost-scsi-common.h"
63 
64 #include "exec/address-spaces.h"
65 #include "hw/usb.h"
66 #include "qemu/config-file.h"
67 #include "qemu/error-report.h"
68 #include "trace.h"
69 #include "hw/nmi.h"
70 #include "hw/intc/intc.h"
71 
72 #include "hw/compat.h"
73 #include "qemu/cutils.h"
74 #include "hw/ppc/spapr_cpu_core.h"
75 #include "qmp-commands.h"
76 
77 #include <libfdt.h>
78 
79 /* SLOF memory layout:
80  *
81  * SLOF raw image loaded at 0, copies its romfs right below the flat
82  * device-tree, then position SLOF itself 31M below that
83  *
84  * So we set FW_OVERHEAD to 40MB which should account for all of that
85  * and more
86  *
87  * We load our kernel at 4M, leaving space for SLOF initial image
88  */
89 #define FDT_MAX_SIZE            0x100000
90 #define RTAS_MAX_SIZE           0x10000
91 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
92 #define FW_MAX_SIZE             0x400000
93 #define FW_FILE_NAME            "slof.bin"
94 #define FW_OVERHEAD             0x2800000
95 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
96 
97 #define MIN_RMA_SLOF            128UL
98 
99 #define PHANDLE_XICP            0x00001111
100 
101 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
102                                   const char *type_ics,
103                                   int nr_irqs, Error **errp)
104 {
105     Error *local_err = NULL;
106     Object *obj;
107 
108     obj = object_new(type_ics);
109     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
110     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
111                                    &error_abort);
112     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
113     if (local_err) {
114         goto error;
115     }
116     object_property_set_bool(obj, true, "realized", &local_err);
117     if (local_err) {
118         goto error;
119     }
120 
121     return ICS_SIMPLE(obj);
122 
123 error:
124     error_propagate(errp, local_err);
125     return NULL;
126 }
127 
128 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
129 {
130     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
131      * and newer QEMUs don't even have them. In both cases, we don't want
132      * to send anything on the wire.
133      */
134     return false;
135 }
136 
137 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
138     .name = "icp/server",
139     .version_id = 1,
140     .minimum_version_id = 1,
141     .needed = pre_2_10_vmstate_dummy_icp_needed,
142     .fields = (VMStateField[]) {
143         VMSTATE_UNUSED(4), /* uint32_t xirr */
144         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
145         VMSTATE_UNUSED(1), /* uint8_t mfrr */
146         VMSTATE_END_OF_LIST()
147     },
148 };
149 
150 static void pre_2_10_vmstate_register_dummy_icp(int i)
151 {
152     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
153                      (void *)(uintptr_t) i);
154 }
155 
156 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
157 {
158     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
159                        (void *)(uintptr_t) i);
160 }
161 
162 static inline int xics_max_server_number(void)
163 {
164     return DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(), smp_threads);
165 }
166 
167 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
168 {
169     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
170     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
171 
172     if (kvm_enabled()) {
173         if (machine_kernel_irqchip_allowed(machine) &&
174             !xics_kvm_init(spapr, errp)) {
175             spapr->icp_type = TYPE_KVM_ICP;
176             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
177         }
178         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
179             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
180             return;
181         }
182     }
183 
184     if (!spapr->ics) {
185         xics_spapr_init(spapr);
186         spapr->icp_type = TYPE_ICP;
187         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
188         if (!spapr->ics) {
189             return;
190         }
191     }
192 
193     if (smc->pre_2_10_has_unused_icps) {
194         int i;
195 
196         for (i = 0; i < xics_max_server_number(); i++) {
197             /* Dummy entries get deregistered when real ICPState objects
198              * are registered during CPU core hotplug.
199              */
200             pre_2_10_vmstate_register_dummy_icp(i);
201         }
202     }
203 }
204 
205 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
206                                   int smt_threads)
207 {
208     int i, ret = 0;
209     uint32_t servers_prop[smt_threads];
210     uint32_t gservers_prop[smt_threads * 2];
211     int index = ppc_get_vcpu_dt_id(cpu);
212 
213     if (cpu->compat_pvr) {
214         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
215         if (ret < 0) {
216             return ret;
217         }
218     }
219 
220     /* Build interrupt servers and gservers properties */
221     for (i = 0; i < smt_threads; i++) {
222         servers_prop[i] = cpu_to_be32(index + i);
223         /* Hack, direct the group queues back to cpu 0 */
224         gservers_prop[i*2] = cpu_to_be32(index + i);
225         gservers_prop[i*2 + 1] = 0;
226     }
227     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
228                       servers_prop, sizeof(servers_prop));
229     if (ret < 0) {
230         return ret;
231     }
232     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
233                       gservers_prop, sizeof(gservers_prop));
234 
235     return ret;
236 }
237 
238 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
239 {
240     int index = ppc_get_vcpu_dt_id(cpu);
241     uint32_t associativity[] = {cpu_to_be32(0x5),
242                                 cpu_to_be32(0x0),
243                                 cpu_to_be32(0x0),
244                                 cpu_to_be32(0x0),
245                                 cpu_to_be32(cpu->node_id),
246                                 cpu_to_be32(index)};
247 
248     /* Advertise NUMA via ibm,associativity */
249     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
250                           sizeof(associativity));
251 }
252 
253 /* Populate the "ibm,pa-features" property */
254 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset,
255                                       bool legacy_guest)
256 {
257     uint8_t pa_features_206[] = { 6, 0,
258         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
259     uint8_t pa_features_207[] = { 24, 0,
260         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
261         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
262         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
263         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
264     uint8_t pa_features_300[] = { 66, 0,
265         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
266         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
267         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
268         /* 6: DS207 */
269         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
270         /* 16: Vector */
271         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
272         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
273         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
274         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
275         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
276         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
277         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
278         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
279         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
280         /* 42: PM, 44: PC RA, 46: SC vec'd */
281         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
282         /* 48: SIMD, 50: QP BFP, 52: String */
283         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
284         /* 54: DecFP, 56: DecI, 58: SHA */
285         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
286         /* 60: NM atomic, 62: RNG */
287         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
288     };
289     uint8_t *pa_features;
290     size_t pa_size;
291 
292     switch (POWERPC_MMU_VER(env->mmu_model)) {
293     case POWERPC_MMU_VER_2_06:
294         pa_features = pa_features_206;
295         pa_size = sizeof(pa_features_206);
296         break;
297     case POWERPC_MMU_VER_2_07:
298         pa_features = pa_features_207;
299         pa_size = sizeof(pa_features_207);
300         break;
301     case POWERPC_MMU_VER_3_00:
302         pa_features = pa_features_300;
303         pa_size = sizeof(pa_features_300);
304         break;
305     default:
306         return;
307     }
308 
309     if (env->ci_large_pages) {
310         /*
311          * Note: we keep CI large pages off by default because a 64K capable
312          * guest provisioned with large pages might otherwise try to map a qemu
313          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
314          * even if that qemu runs on a 4k host.
315          * We dd this bit back here if we are confident this is not an issue
316          */
317         pa_features[3] |= 0x20;
318     }
319     if (kvmppc_has_cap_htm() && pa_size > 24) {
320         pa_features[24] |= 0x80;    /* Transactional memory support */
321     }
322     if (legacy_guest && pa_size > 40) {
323         /* Workaround for broken kernels that attempt (guest) radix
324          * mode when they can't handle it, if they see the radix bit set
325          * in pa-features. So hide it from them. */
326         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
327     }
328 
329     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
330 }
331 
332 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
333 {
334     int ret = 0, offset, cpus_offset;
335     CPUState *cs;
336     char cpu_model[32];
337     int smt = kvmppc_smt_threads();
338     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
339 
340     CPU_FOREACH(cs) {
341         PowerPCCPU *cpu = POWERPC_CPU(cs);
342         CPUPPCState *env = &cpu->env;
343         DeviceClass *dc = DEVICE_GET_CLASS(cs);
344         int index = ppc_get_vcpu_dt_id(cpu);
345         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
346 
347         if ((index % smt) != 0) {
348             continue;
349         }
350 
351         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
352 
353         cpus_offset = fdt_path_offset(fdt, "/cpus");
354         if (cpus_offset < 0) {
355             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
356                                           "cpus");
357             if (cpus_offset < 0) {
358                 return cpus_offset;
359             }
360         }
361         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
362         if (offset < 0) {
363             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
364             if (offset < 0) {
365                 return offset;
366             }
367         }
368 
369         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
370                           pft_size_prop, sizeof(pft_size_prop));
371         if (ret < 0) {
372             return ret;
373         }
374 
375         if (nb_numa_nodes > 1) {
376             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
377             if (ret < 0) {
378                 return ret;
379             }
380         }
381 
382         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
383         if (ret < 0) {
384             return ret;
385         }
386 
387         spapr_populate_pa_features(env, fdt, offset,
388                                          spapr->cas_legacy_guest_workaround);
389     }
390     return ret;
391 }
392 
393 static hwaddr spapr_node0_size(void)
394 {
395     MachineState *machine = MACHINE(qdev_get_machine());
396 
397     if (nb_numa_nodes) {
398         int i;
399         for (i = 0; i < nb_numa_nodes; ++i) {
400             if (numa_info[i].node_mem) {
401                 return MIN(pow2floor(numa_info[i].node_mem),
402                            machine->ram_size);
403             }
404         }
405     }
406     return machine->ram_size;
407 }
408 
409 static void add_str(GString *s, const gchar *s1)
410 {
411     g_string_append_len(s, s1, strlen(s1) + 1);
412 }
413 
414 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
415                                        hwaddr size)
416 {
417     uint32_t associativity[] = {
418         cpu_to_be32(0x4), /* length */
419         cpu_to_be32(0x0), cpu_to_be32(0x0),
420         cpu_to_be32(0x0), cpu_to_be32(nodeid)
421     };
422     char mem_name[32];
423     uint64_t mem_reg_property[2];
424     int off;
425 
426     mem_reg_property[0] = cpu_to_be64(start);
427     mem_reg_property[1] = cpu_to_be64(size);
428 
429     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
430     off = fdt_add_subnode(fdt, 0, mem_name);
431     _FDT(off);
432     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
433     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
434                       sizeof(mem_reg_property))));
435     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
436                       sizeof(associativity))));
437     return off;
438 }
439 
440 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
441 {
442     MachineState *machine = MACHINE(spapr);
443     hwaddr mem_start, node_size;
444     int i, nb_nodes = nb_numa_nodes;
445     NodeInfo *nodes = numa_info;
446     NodeInfo ramnode;
447 
448     /* No NUMA nodes, assume there is just one node with whole RAM */
449     if (!nb_numa_nodes) {
450         nb_nodes = 1;
451         ramnode.node_mem = machine->ram_size;
452         nodes = &ramnode;
453     }
454 
455     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
456         if (!nodes[i].node_mem) {
457             continue;
458         }
459         if (mem_start >= machine->ram_size) {
460             node_size = 0;
461         } else {
462             node_size = nodes[i].node_mem;
463             if (node_size > machine->ram_size - mem_start) {
464                 node_size = machine->ram_size - mem_start;
465             }
466         }
467         if (!mem_start) {
468             /* ppc_spapr_init() checks for rma_size <= node0_size already */
469             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
470             mem_start += spapr->rma_size;
471             node_size -= spapr->rma_size;
472         }
473         for ( ; node_size; ) {
474             hwaddr sizetmp = pow2floor(node_size);
475 
476             /* mem_start != 0 here */
477             if (ctzl(mem_start) < ctzl(sizetmp)) {
478                 sizetmp = 1ULL << ctzl(mem_start);
479             }
480 
481             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
482             node_size -= sizetmp;
483             mem_start += sizetmp;
484         }
485     }
486 
487     return 0;
488 }
489 
490 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
491                                   sPAPRMachineState *spapr)
492 {
493     PowerPCCPU *cpu = POWERPC_CPU(cs);
494     CPUPPCState *env = &cpu->env;
495     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
496     int index = ppc_get_vcpu_dt_id(cpu);
497     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
498                        0xffffffff, 0xffffffff};
499     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
500         : SPAPR_TIMEBASE_FREQ;
501     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
502     uint32_t page_sizes_prop[64];
503     size_t page_sizes_prop_size;
504     uint32_t vcpus_per_socket = smp_threads * smp_cores;
505     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
506     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
507     sPAPRDRConnector *drc;
508     int drc_index;
509     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
510     int i;
511 
512     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
513     if (drc) {
514         drc_index = spapr_drc_index(drc);
515         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
516     }
517 
518     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
519     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
520 
521     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
522     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
523                            env->dcache_line_size)));
524     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
525                            env->dcache_line_size)));
526     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
527                            env->icache_line_size)));
528     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
529                            env->icache_line_size)));
530 
531     if (pcc->l1_dcache_size) {
532         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
533                                pcc->l1_dcache_size)));
534     } else {
535         warn_report("Unknown L1 dcache size for cpu");
536     }
537     if (pcc->l1_icache_size) {
538         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
539                                pcc->l1_icache_size)));
540     } else {
541         warn_report("Unknown L1 icache size for cpu");
542     }
543 
544     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
545     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
546     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
547     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
548     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
549     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
550 
551     if (env->spr_cb[SPR_PURR].oea_read) {
552         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
553     }
554 
555     if (env->mmu_model & POWERPC_MMU_1TSEG) {
556         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
557                           segs, sizeof(segs))));
558     }
559 
560     /* Advertise VMX/VSX (vector extensions) if available
561      *   0 / no property == no vector extensions
562      *   1               == VMX / Altivec available
563      *   2               == VSX available */
564     if (env->insns_flags & PPC_ALTIVEC) {
565         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
566 
567         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
568     }
569 
570     /* Advertise DFP (Decimal Floating Point) if available
571      *   0 / no property == no DFP
572      *   1               == DFP available */
573     if (env->insns_flags2 & PPC2_DFP) {
574         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
575     }
576 
577     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
578                                                   sizeof(page_sizes_prop));
579     if (page_sizes_prop_size) {
580         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
581                           page_sizes_prop, page_sizes_prop_size)));
582     }
583 
584     spapr_populate_pa_features(env, fdt, offset, false);
585 
586     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
587                            cs->cpu_index / vcpus_per_socket)));
588 
589     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
590                       pft_size_prop, sizeof(pft_size_prop))));
591 
592     if (nb_numa_nodes > 1) {
593         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
594     }
595 
596     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
597 
598     if (pcc->radix_page_info) {
599         for (i = 0; i < pcc->radix_page_info->count; i++) {
600             radix_AP_encodings[i] =
601                 cpu_to_be32(pcc->radix_page_info->entries[i]);
602         }
603         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
604                           radix_AP_encodings,
605                           pcc->radix_page_info->count *
606                           sizeof(radix_AP_encodings[0]))));
607     }
608 }
609 
610 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
611 {
612     CPUState *cs;
613     int cpus_offset;
614     char *nodename;
615     int smt = kvmppc_smt_threads();
616 
617     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
618     _FDT(cpus_offset);
619     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
620     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
621 
622     /*
623      * We walk the CPUs in reverse order to ensure that CPU DT nodes
624      * created by fdt_add_subnode() end up in the right order in FDT
625      * for the guest kernel the enumerate the CPUs correctly.
626      */
627     CPU_FOREACH_REVERSE(cs) {
628         PowerPCCPU *cpu = POWERPC_CPU(cs);
629         int index = ppc_get_vcpu_dt_id(cpu);
630         DeviceClass *dc = DEVICE_GET_CLASS(cs);
631         int offset;
632 
633         if ((index % smt) != 0) {
634             continue;
635         }
636 
637         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
638         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
639         g_free(nodename);
640         _FDT(offset);
641         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
642     }
643 
644 }
645 
646 /*
647  * Adds ibm,dynamic-reconfiguration-memory node.
648  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
649  * of this device tree node.
650  */
651 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
652 {
653     MachineState *machine = MACHINE(spapr);
654     int ret, i, offset;
655     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
656     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
657     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
658     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
659                        memory_region_size(&spapr->hotplug_memory.mr)) /
660                        lmb_size;
661     uint32_t *int_buf, *cur_index, buf_len;
662     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
663 
664     /*
665      * Don't create the node if there is no hotpluggable memory
666      */
667     if (machine->ram_size == machine->maxram_size) {
668         return 0;
669     }
670 
671     /*
672      * Allocate enough buffer size to fit in ibm,dynamic-memory
673      * or ibm,associativity-lookup-arrays
674      */
675     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
676               * sizeof(uint32_t);
677     cur_index = int_buf = g_malloc0(buf_len);
678 
679     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
680 
681     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
682                     sizeof(prop_lmb_size));
683     if (ret < 0) {
684         goto out;
685     }
686 
687     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
688     if (ret < 0) {
689         goto out;
690     }
691 
692     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
693     if (ret < 0) {
694         goto out;
695     }
696 
697     /* ibm,dynamic-memory */
698     int_buf[0] = cpu_to_be32(nr_lmbs);
699     cur_index++;
700     for (i = 0; i < nr_lmbs; i++) {
701         uint64_t addr = i * lmb_size;
702         uint32_t *dynamic_memory = cur_index;
703 
704         if (i >= hotplug_lmb_start) {
705             sPAPRDRConnector *drc;
706 
707             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
708             g_assert(drc);
709 
710             dynamic_memory[0] = cpu_to_be32(addr >> 32);
711             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
712             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
713             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
714             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
715             if (memory_region_present(get_system_memory(), addr)) {
716                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
717             } else {
718                 dynamic_memory[5] = cpu_to_be32(0);
719             }
720         } else {
721             /*
722              * LMB information for RMA, boot time RAM and gap b/n RAM and
723              * hotplug memory region -- all these are marked as reserved
724              * and as having no valid DRC.
725              */
726             dynamic_memory[0] = cpu_to_be32(addr >> 32);
727             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
728             dynamic_memory[2] = cpu_to_be32(0);
729             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
730             dynamic_memory[4] = cpu_to_be32(-1);
731             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
732                                             SPAPR_LMB_FLAGS_DRC_INVALID);
733         }
734 
735         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
736     }
737     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
738     if (ret < 0) {
739         goto out;
740     }
741 
742     /* ibm,associativity-lookup-arrays */
743     cur_index = int_buf;
744     int_buf[0] = cpu_to_be32(nr_nodes);
745     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
746     cur_index += 2;
747     for (i = 0; i < nr_nodes; i++) {
748         uint32_t associativity[] = {
749             cpu_to_be32(0x0),
750             cpu_to_be32(0x0),
751             cpu_to_be32(0x0),
752             cpu_to_be32(i)
753         };
754         memcpy(cur_index, associativity, sizeof(associativity));
755         cur_index += 4;
756     }
757     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
758             (cur_index - int_buf) * sizeof(uint32_t));
759 out:
760     g_free(int_buf);
761     return ret;
762 }
763 
764 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
765                                 sPAPROptionVector *ov5_updates)
766 {
767     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
768     int ret = 0, offset;
769 
770     /* Generate ibm,dynamic-reconfiguration-memory node if required */
771     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
772         g_assert(smc->dr_lmb_enabled);
773         ret = spapr_populate_drconf_memory(spapr, fdt);
774         if (ret) {
775             goto out;
776         }
777     }
778 
779     offset = fdt_path_offset(fdt, "/chosen");
780     if (offset < 0) {
781         offset = fdt_add_subnode(fdt, 0, "chosen");
782         if (offset < 0) {
783             return offset;
784         }
785     }
786     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
787                                  "ibm,architecture-vec-5");
788 
789 out:
790     return ret;
791 }
792 
793 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
794                                  target_ulong addr, target_ulong size,
795                                  sPAPROptionVector *ov5_updates)
796 {
797     void *fdt, *fdt_skel;
798     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
799 
800     size -= sizeof(hdr);
801 
802     /* Create sceleton */
803     fdt_skel = g_malloc0(size);
804     _FDT((fdt_create(fdt_skel, size)));
805     _FDT((fdt_begin_node(fdt_skel, "")));
806     _FDT((fdt_end_node(fdt_skel)));
807     _FDT((fdt_finish(fdt_skel)));
808     fdt = g_malloc0(size);
809     _FDT((fdt_open_into(fdt_skel, fdt, size)));
810     g_free(fdt_skel);
811 
812     /* Fixup cpu nodes */
813     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
814 
815     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
816         return -1;
817     }
818 
819     /* Pack resulting tree */
820     _FDT((fdt_pack(fdt)));
821 
822     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
823         trace_spapr_cas_failed(size);
824         return -1;
825     }
826 
827     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
828     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
829     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
830     g_free(fdt);
831 
832     return 0;
833 }
834 
835 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
836 {
837     int rtas;
838     GString *hypertas = g_string_sized_new(256);
839     GString *qemu_hypertas = g_string_sized_new(256);
840     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
841     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
842         memory_region_size(&spapr->hotplug_memory.mr);
843     uint32_t lrdr_capacity[] = {
844         cpu_to_be32(max_hotplug_addr >> 32),
845         cpu_to_be32(max_hotplug_addr & 0xffffffff),
846         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
847         cpu_to_be32(max_cpus / smp_threads),
848     };
849 
850     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
851 
852     /* hypertas */
853     add_str(hypertas, "hcall-pft");
854     add_str(hypertas, "hcall-term");
855     add_str(hypertas, "hcall-dabr");
856     add_str(hypertas, "hcall-interrupt");
857     add_str(hypertas, "hcall-tce");
858     add_str(hypertas, "hcall-vio");
859     add_str(hypertas, "hcall-splpar");
860     add_str(hypertas, "hcall-bulk");
861     add_str(hypertas, "hcall-set-mode");
862     add_str(hypertas, "hcall-sprg0");
863     add_str(hypertas, "hcall-copy");
864     add_str(hypertas, "hcall-debug");
865     add_str(qemu_hypertas, "hcall-memop1");
866 
867     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
868         add_str(hypertas, "hcall-multi-tce");
869     }
870 
871     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
872         add_str(hypertas, "hcall-hpt-resize");
873     }
874 
875     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
876                      hypertas->str, hypertas->len));
877     g_string_free(hypertas, TRUE);
878     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
879                      qemu_hypertas->str, qemu_hypertas->len));
880     g_string_free(qemu_hypertas, TRUE);
881 
882     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
883                      refpoints, sizeof(refpoints)));
884 
885     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
886                           RTAS_ERROR_LOG_MAX));
887     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
888                           RTAS_EVENT_SCAN_RATE));
889 
890     if (msi_nonbroken) {
891         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
892     }
893 
894     /*
895      * According to PAPR, rtas ibm,os-term does not guarantee a return
896      * back to the guest cpu.
897      *
898      * While an additional ibm,extended-os-term property indicates
899      * that rtas call return will always occur. Set this property.
900      */
901     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
902 
903     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
904                      lrdr_capacity, sizeof(lrdr_capacity)));
905 
906     spapr_dt_rtas_tokens(fdt, rtas);
907 }
908 
909 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
910  * that the guest may request and thus the valid values for bytes 24..26 of
911  * option vector 5: */
912 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
913 {
914     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
915 
916     char val[2 * 4] = {
917         23, 0x00, /* Xive mode: 0 = legacy (as in ISA 2.7), 1 = Exploitation */
918         24, 0x00, /* Hash/Radix, filled in below. */
919         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
920         26, 0x40, /* Radix options: GTSE == yes. */
921     };
922 
923     if (kvm_enabled()) {
924         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
925             val[3] = 0x80; /* OV5_MMU_BOTH */
926         } else if (kvmppc_has_cap_mmu_radix()) {
927             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
928         } else {
929             val[3] = 0x00; /* Hash */
930         }
931     } else {
932         if (first_ppc_cpu->env.mmu_model & POWERPC_MMU_V3) {
933             /* V3 MMU supports both hash and radix (with dynamic switching) */
934             val[3] = 0xC0;
935         } else {
936             /* Otherwise we can only do hash */
937             val[3] = 0x00;
938         }
939     }
940     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
941                      val, sizeof(val)));
942 }
943 
944 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
945 {
946     MachineState *machine = MACHINE(spapr);
947     int chosen;
948     const char *boot_device = machine->boot_order;
949     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
950     size_t cb = 0;
951     char *bootlist = get_boot_devices_list(&cb, true);
952 
953     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
954 
955     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
956     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
957                           spapr->initrd_base));
958     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
959                           spapr->initrd_base + spapr->initrd_size));
960 
961     if (spapr->kernel_size) {
962         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
963                               cpu_to_be64(spapr->kernel_size) };
964 
965         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
966                          &kprop, sizeof(kprop)));
967         if (spapr->kernel_le) {
968             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
969         }
970     }
971     if (boot_menu) {
972         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
973     }
974     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
975     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
976     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
977 
978     if (cb && bootlist) {
979         int i;
980 
981         for (i = 0; i < cb; i++) {
982             if (bootlist[i] == '\n') {
983                 bootlist[i] = ' ';
984             }
985         }
986         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
987     }
988 
989     if (boot_device && strlen(boot_device)) {
990         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
991     }
992 
993     if (!spapr->has_graphics && stdout_path) {
994         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
995     }
996 
997     spapr_dt_ov5_platform_support(fdt, chosen);
998 
999     g_free(stdout_path);
1000     g_free(bootlist);
1001 }
1002 
1003 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1004 {
1005     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1006      * KVM to work under pHyp with some guest co-operation */
1007     int hypervisor;
1008     uint8_t hypercall[16];
1009 
1010     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1011     /* indicate KVM hypercall interface */
1012     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1013     if (kvmppc_has_cap_fixup_hcalls()) {
1014         /*
1015          * Older KVM versions with older guest kernels were broken
1016          * with the magic page, don't allow the guest to map it.
1017          */
1018         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1019                                   sizeof(hypercall))) {
1020             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1021                              hypercall, sizeof(hypercall)));
1022         }
1023     }
1024 }
1025 
1026 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1027                              hwaddr rtas_addr,
1028                              hwaddr rtas_size)
1029 {
1030     MachineState *machine = MACHINE(qdev_get_machine());
1031     MachineClass *mc = MACHINE_GET_CLASS(machine);
1032     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1033     int ret;
1034     void *fdt;
1035     sPAPRPHBState *phb;
1036     char *buf;
1037 
1038     fdt = g_malloc0(FDT_MAX_SIZE);
1039     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1040 
1041     /* Root node */
1042     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1043     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1044     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1045 
1046     /*
1047      * Add info to guest to indentify which host is it being run on
1048      * and what is the uuid of the guest
1049      */
1050     if (kvmppc_get_host_model(&buf)) {
1051         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1052         g_free(buf);
1053     }
1054     if (kvmppc_get_host_serial(&buf)) {
1055         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1056         g_free(buf);
1057     }
1058 
1059     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1060 
1061     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1062     if (qemu_uuid_set) {
1063         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1064     }
1065     g_free(buf);
1066 
1067     if (qemu_get_vm_name()) {
1068         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1069                                 qemu_get_vm_name()));
1070     }
1071 
1072     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1073     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1074 
1075     /* /interrupt controller */
1076     spapr_dt_xics(xics_max_server_number(), fdt, PHANDLE_XICP);
1077 
1078     ret = spapr_populate_memory(spapr, fdt);
1079     if (ret < 0) {
1080         error_report("couldn't setup memory nodes in fdt");
1081         exit(1);
1082     }
1083 
1084     /* /vdevice */
1085     spapr_dt_vdevice(spapr->vio_bus, fdt);
1086 
1087     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1088         ret = spapr_rng_populate_dt(fdt);
1089         if (ret < 0) {
1090             error_report("could not set up rng device in the fdt");
1091             exit(1);
1092         }
1093     }
1094 
1095     QLIST_FOREACH(phb, &spapr->phbs, list) {
1096         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1097         if (ret < 0) {
1098             error_report("couldn't setup PCI devices in fdt");
1099             exit(1);
1100         }
1101     }
1102 
1103     /* cpus */
1104     spapr_populate_cpus_dt_node(fdt, spapr);
1105 
1106     if (smc->dr_lmb_enabled) {
1107         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1108     }
1109 
1110     if (mc->has_hotpluggable_cpus) {
1111         int offset = fdt_path_offset(fdt, "/cpus");
1112         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1113                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1114         if (ret < 0) {
1115             error_report("Couldn't set up CPU DR device tree properties");
1116             exit(1);
1117         }
1118     }
1119 
1120     /* /event-sources */
1121     spapr_dt_events(spapr, fdt);
1122 
1123     /* /rtas */
1124     spapr_dt_rtas(spapr, fdt);
1125 
1126     /* /chosen */
1127     spapr_dt_chosen(spapr, fdt);
1128 
1129     /* /hypervisor */
1130     if (kvm_enabled()) {
1131         spapr_dt_hypervisor(spapr, fdt);
1132     }
1133 
1134     /* Build memory reserve map */
1135     if (spapr->kernel_size) {
1136         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1137     }
1138     if (spapr->initrd_size) {
1139         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1140     }
1141 
1142     /* ibm,client-architecture-support updates */
1143     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1144     if (ret < 0) {
1145         error_report("couldn't setup CAS properties fdt");
1146         exit(1);
1147     }
1148 
1149     return fdt;
1150 }
1151 
1152 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1153 {
1154     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1155 }
1156 
1157 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1158                                     PowerPCCPU *cpu)
1159 {
1160     CPUPPCState *env = &cpu->env;
1161 
1162     /* The TCG path should also be holding the BQL at this point */
1163     g_assert(qemu_mutex_iothread_locked());
1164 
1165     if (msr_pr) {
1166         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1167         env->gpr[3] = H_PRIVILEGE;
1168     } else {
1169         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1170     }
1171 }
1172 
1173 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1174 {
1175     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1176 
1177     return spapr->patb_entry;
1178 }
1179 
1180 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1181 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1182 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1183 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1184 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1185 
1186 /*
1187  * Get the fd to access the kernel htab, re-opening it if necessary
1188  */
1189 static int get_htab_fd(sPAPRMachineState *spapr)
1190 {
1191     if (spapr->htab_fd >= 0) {
1192         return spapr->htab_fd;
1193     }
1194 
1195     spapr->htab_fd = kvmppc_get_htab_fd(false);
1196     if (spapr->htab_fd < 0) {
1197         error_report("Unable to open fd for reading hash table from KVM: %s",
1198                      strerror(errno));
1199     }
1200 
1201     return spapr->htab_fd;
1202 }
1203 
1204 void close_htab_fd(sPAPRMachineState *spapr)
1205 {
1206     if (spapr->htab_fd >= 0) {
1207         close(spapr->htab_fd);
1208     }
1209     spapr->htab_fd = -1;
1210 }
1211 
1212 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1213 {
1214     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1215 
1216     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1217 }
1218 
1219 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1220                                                 hwaddr ptex, int n)
1221 {
1222     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1223     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1224 
1225     if (!spapr->htab) {
1226         /*
1227          * HTAB is controlled by KVM. Fetch into temporary buffer
1228          */
1229         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1230         kvmppc_read_hptes(hptes, ptex, n);
1231         return hptes;
1232     }
1233 
1234     /*
1235      * HTAB is controlled by QEMU. Just point to the internally
1236      * accessible PTEG.
1237      */
1238     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1239 }
1240 
1241 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1242                               const ppc_hash_pte64_t *hptes,
1243                               hwaddr ptex, int n)
1244 {
1245     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1246 
1247     if (!spapr->htab) {
1248         g_free((void *)hptes);
1249     }
1250 
1251     /* Nothing to do for qemu managed HPT */
1252 }
1253 
1254 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1255                              uint64_t pte0, uint64_t pte1)
1256 {
1257     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1258     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1259 
1260     if (!spapr->htab) {
1261         kvmppc_write_hpte(ptex, pte0, pte1);
1262     } else {
1263         stq_p(spapr->htab + offset, pte0);
1264         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1265     }
1266 }
1267 
1268 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1269 {
1270     int shift;
1271 
1272     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1273      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1274      * that's much more than is needed for Linux guests */
1275     shift = ctz64(pow2ceil(ramsize)) - 7;
1276     shift = MAX(shift, 18); /* Minimum architected size */
1277     shift = MIN(shift, 46); /* Maximum architected size */
1278     return shift;
1279 }
1280 
1281 void spapr_free_hpt(sPAPRMachineState *spapr)
1282 {
1283     g_free(spapr->htab);
1284     spapr->htab = NULL;
1285     spapr->htab_shift = 0;
1286     close_htab_fd(spapr);
1287 }
1288 
1289 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1290                           Error **errp)
1291 {
1292     long rc;
1293 
1294     /* Clean up any HPT info from a previous boot */
1295     spapr_free_hpt(spapr);
1296 
1297     rc = kvmppc_reset_htab(shift);
1298     if (rc < 0) {
1299         /* kernel-side HPT needed, but couldn't allocate one */
1300         error_setg_errno(errp, errno,
1301                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1302                          shift);
1303         /* This is almost certainly fatal, but if the caller really
1304          * wants to carry on with shift == 0, it's welcome to try */
1305     } else if (rc > 0) {
1306         /* kernel-side HPT allocated */
1307         if (rc != shift) {
1308             error_setg(errp,
1309                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1310                        shift, rc);
1311         }
1312 
1313         spapr->htab_shift = shift;
1314         spapr->htab = NULL;
1315     } else {
1316         /* kernel-side HPT not needed, allocate in userspace instead */
1317         size_t size = 1ULL << shift;
1318         int i;
1319 
1320         spapr->htab = qemu_memalign(size, size);
1321         if (!spapr->htab) {
1322             error_setg_errno(errp, errno,
1323                              "Could not allocate HPT of order %d", shift);
1324             return;
1325         }
1326 
1327         memset(spapr->htab, 0, size);
1328         spapr->htab_shift = shift;
1329 
1330         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1331             DIRTY_HPTE(HPTE(spapr->htab, i));
1332         }
1333     }
1334 }
1335 
1336 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1337 {
1338     int hpt_shift;
1339 
1340     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1341         || (spapr->cas_reboot
1342             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1343         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1344     } else {
1345         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->ram_size);
1346     }
1347     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1348 
1349     if (spapr->vrma_adjust) {
1350         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1351                                           spapr->htab_shift);
1352     }
1353     /* We're setting up a hash table, so that means we're not radix */
1354     spapr->patb_entry = 0;
1355 }
1356 
1357 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1358 {
1359     bool matched = false;
1360 
1361     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1362         matched = true;
1363     }
1364 
1365     if (!matched) {
1366         error_report("Device %s is not supported by this machine yet.",
1367                      qdev_fw_name(DEVICE(sbdev)));
1368         exit(1);
1369     }
1370 }
1371 
1372 static void ppc_spapr_reset(void)
1373 {
1374     MachineState *machine = MACHINE(qdev_get_machine());
1375     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1376     PowerPCCPU *first_ppc_cpu;
1377     uint32_t rtas_limit;
1378     hwaddr rtas_addr, fdt_addr;
1379     void *fdt;
1380     int rc;
1381 
1382     /* Check for unknown sysbus devices */
1383     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1384 
1385     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1386         /* If using KVM with radix mode available, VCPUs can be started
1387          * without a HPT because KVM will start them in radix mode.
1388          * Set the GR bit in PATB so that we know there is no HPT. */
1389         spapr->patb_entry = PATBE1_GR;
1390     } else {
1391         spapr_setup_hpt_and_vrma(spapr);
1392     }
1393 
1394     qemu_devices_reset();
1395 
1396     /*
1397      * We place the device tree and RTAS just below either the top of the RMA,
1398      * or just below 2GB, whichever is lowere, so that it can be
1399      * processed with 32-bit real mode code if necessary
1400      */
1401     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1402     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1403     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1404 
1405     /* if this reset wasn't generated by CAS, we should reset our
1406      * negotiated options and start from scratch */
1407     if (!spapr->cas_reboot) {
1408         spapr_ovec_cleanup(spapr->ov5_cas);
1409         spapr->ov5_cas = spapr_ovec_new();
1410 
1411         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1412     }
1413 
1414     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1415 
1416     spapr_load_rtas(spapr, fdt, rtas_addr);
1417 
1418     rc = fdt_pack(fdt);
1419 
1420     /* Should only fail if we've built a corrupted tree */
1421     assert(rc == 0);
1422 
1423     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1424         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1425                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1426         exit(1);
1427     }
1428 
1429     /* Load the fdt */
1430     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1431     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1432     g_free(fdt);
1433 
1434     /* Set up the entry state */
1435     first_ppc_cpu = POWERPC_CPU(first_cpu);
1436     first_ppc_cpu->env.gpr[3] = fdt_addr;
1437     first_ppc_cpu->env.gpr[5] = 0;
1438     first_cpu->halted = 0;
1439     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1440 
1441     spapr->cas_reboot = false;
1442 }
1443 
1444 static void spapr_create_nvram(sPAPRMachineState *spapr)
1445 {
1446     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1447     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1448 
1449     if (dinfo) {
1450         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1451                             &error_fatal);
1452     }
1453 
1454     qdev_init_nofail(dev);
1455 
1456     spapr->nvram = (struct sPAPRNVRAM *)dev;
1457 }
1458 
1459 static void spapr_rtc_create(sPAPRMachineState *spapr)
1460 {
1461     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1462     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1463                               &error_fatal);
1464     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1465                               &error_fatal);
1466     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1467                               "date", &error_fatal);
1468 }
1469 
1470 /* Returns whether we want to use VGA or not */
1471 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1472 {
1473     switch (vga_interface_type) {
1474     case VGA_NONE:
1475         return false;
1476     case VGA_DEVICE:
1477         return true;
1478     case VGA_STD:
1479     case VGA_VIRTIO:
1480         return pci_vga_init(pci_bus) != NULL;
1481     default:
1482         error_setg(errp,
1483                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1484         return false;
1485     }
1486 }
1487 
1488 static int spapr_post_load(void *opaque, int version_id)
1489 {
1490     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1491     int err = 0;
1492 
1493     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1494         CPUState *cs;
1495         CPU_FOREACH(cs) {
1496             PowerPCCPU *cpu = POWERPC_CPU(cs);
1497             icp_resend(ICP(cpu->intc));
1498         }
1499     }
1500 
1501     /* In earlier versions, there was no separate qdev for the PAPR
1502      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1503      * So when migrating from those versions, poke the incoming offset
1504      * value into the RTC device */
1505     if (version_id < 3) {
1506         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1507     }
1508 
1509     if (spapr->patb_entry) {
1510         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1511         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1512         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1513 
1514         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1515         if (err) {
1516             error_report("Process table config unsupported by the host");
1517             return -EINVAL;
1518         }
1519     }
1520 
1521     return err;
1522 }
1523 
1524 static bool version_before_3(void *opaque, int version_id)
1525 {
1526     return version_id < 3;
1527 }
1528 
1529 static bool spapr_pending_events_needed(void *opaque)
1530 {
1531     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1532     return !QTAILQ_EMPTY(&spapr->pending_events);
1533 }
1534 
1535 static const VMStateDescription vmstate_spapr_event_entry = {
1536     .name = "spapr_event_log_entry",
1537     .version_id = 1,
1538     .minimum_version_id = 1,
1539     .fields = (VMStateField[]) {
1540         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1541         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1542         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1543                                      NULL, extended_length),
1544         VMSTATE_END_OF_LIST()
1545     },
1546 };
1547 
1548 static const VMStateDescription vmstate_spapr_pending_events = {
1549     .name = "spapr_pending_events",
1550     .version_id = 1,
1551     .minimum_version_id = 1,
1552     .needed = spapr_pending_events_needed,
1553     .fields = (VMStateField[]) {
1554         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1555                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1556         VMSTATE_END_OF_LIST()
1557     },
1558 };
1559 
1560 static bool spapr_ov5_cas_needed(void *opaque)
1561 {
1562     sPAPRMachineState *spapr = opaque;
1563     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1564     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1565     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1566     bool cas_needed;
1567 
1568     /* Prior to the introduction of sPAPROptionVector, we had two option
1569      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1570      * Both of these options encode machine topology into the device-tree
1571      * in such a way that the now-booted OS should still be able to interact
1572      * appropriately with QEMU regardless of what options were actually
1573      * negotiatied on the source side.
1574      *
1575      * As such, we can avoid migrating the CAS-negotiated options if these
1576      * are the only options available on the current machine/platform.
1577      * Since these are the only options available for pseries-2.7 and
1578      * earlier, this allows us to maintain old->new/new->old migration
1579      * compatibility.
1580      *
1581      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1582      * via default pseries-2.8 machines and explicit command-line parameters.
1583      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1584      * of the actual CAS-negotiated values to continue working properly. For
1585      * example, availability of memory unplug depends on knowing whether
1586      * OV5_HP_EVT was negotiated via CAS.
1587      *
1588      * Thus, for any cases where the set of available CAS-negotiatable
1589      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1590      * include the CAS-negotiated options in the migration stream.
1591      */
1592     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1593     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1594 
1595     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1596      * the mask itself since in the future it's possible "legacy" bits may be
1597      * removed via machine options, which could generate a false positive
1598      * that breaks migration.
1599      */
1600     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1601     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1602 
1603     spapr_ovec_cleanup(ov5_mask);
1604     spapr_ovec_cleanup(ov5_legacy);
1605     spapr_ovec_cleanup(ov5_removed);
1606 
1607     return cas_needed;
1608 }
1609 
1610 static const VMStateDescription vmstate_spapr_ov5_cas = {
1611     .name = "spapr_option_vector_ov5_cas",
1612     .version_id = 1,
1613     .minimum_version_id = 1,
1614     .needed = spapr_ov5_cas_needed,
1615     .fields = (VMStateField[]) {
1616         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1617                                  vmstate_spapr_ovec, sPAPROptionVector),
1618         VMSTATE_END_OF_LIST()
1619     },
1620 };
1621 
1622 static bool spapr_patb_entry_needed(void *opaque)
1623 {
1624     sPAPRMachineState *spapr = opaque;
1625 
1626     return !!spapr->patb_entry;
1627 }
1628 
1629 static const VMStateDescription vmstate_spapr_patb_entry = {
1630     .name = "spapr_patb_entry",
1631     .version_id = 1,
1632     .minimum_version_id = 1,
1633     .needed = spapr_patb_entry_needed,
1634     .fields = (VMStateField[]) {
1635         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1636         VMSTATE_END_OF_LIST()
1637     },
1638 };
1639 
1640 static const VMStateDescription vmstate_spapr = {
1641     .name = "spapr",
1642     .version_id = 3,
1643     .minimum_version_id = 1,
1644     .post_load = spapr_post_load,
1645     .fields = (VMStateField[]) {
1646         /* used to be @next_irq */
1647         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1648 
1649         /* RTC offset */
1650         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1651 
1652         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1653         VMSTATE_END_OF_LIST()
1654     },
1655     .subsections = (const VMStateDescription*[]) {
1656         &vmstate_spapr_ov5_cas,
1657         &vmstate_spapr_patb_entry,
1658         &vmstate_spapr_pending_events,
1659         NULL
1660     }
1661 };
1662 
1663 static int htab_save_setup(QEMUFile *f, void *opaque)
1664 {
1665     sPAPRMachineState *spapr = opaque;
1666 
1667     /* "Iteration" header */
1668     if (!spapr->htab_shift) {
1669         qemu_put_be32(f, -1);
1670     } else {
1671         qemu_put_be32(f, spapr->htab_shift);
1672     }
1673 
1674     if (spapr->htab) {
1675         spapr->htab_save_index = 0;
1676         spapr->htab_first_pass = true;
1677     } else {
1678         if (spapr->htab_shift) {
1679             assert(kvm_enabled());
1680         }
1681     }
1682 
1683 
1684     return 0;
1685 }
1686 
1687 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1688                                  int64_t max_ns)
1689 {
1690     bool has_timeout = max_ns != -1;
1691     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1692     int index = spapr->htab_save_index;
1693     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1694 
1695     assert(spapr->htab_first_pass);
1696 
1697     do {
1698         int chunkstart;
1699 
1700         /* Consume invalid HPTEs */
1701         while ((index < htabslots)
1702                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1703             CLEAN_HPTE(HPTE(spapr->htab, index));
1704             index++;
1705         }
1706 
1707         /* Consume valid HPTEs */
1708         chunkstart = index;
1709         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1710                && HPTE_VALID(HPTE(spapr->htab, index))) {
1711             CLEAN_HPTE(HPTE(spapr->htab, index));
1712             index++;
1713         }
1714 
1715         if (index > chunkstart) {
1716             int n_valid = index - chunkstart;
1717 
1718             qemu_put_be32(f, chunkstart);
1719             qemu_put_be16(f, n_valid);
1720             qemu_put_be16(f, 0);
1721             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1722                             HASH_PTE_SIZE_64 * n_valid);
1723 
1724             if (has_timeout &&
1725                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1726                 break;
1727             }
1728         }
1729     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1730 
1731     if (index >= htabslots) {
1732         assert(index == htabslots);
1733         index = 0;
1734         spapr->htab_first_pass = false;
1735     }
1736     spapr->htab_save_index = index;
1737 }
1738 
1739 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1740                                 int64_t max_ns)
1741 {
1742     bool final = max_ns < 0;
1743     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1744     int examined = 0, sent = 0;
1745     int index = spapr->htab_save_index;
1746     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1747 
1748     assert(!spapr->htab_first_pass);
1749 
1750     do {
1751         int chunkstart, invalidstart;
1752 
1753         /* Consume non-dirty HPTEs */
1754         while ((index < htabslots)
1755                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1756             index++;
1757             examined++;
1758         }
1759 
1760         chunkstart = index;
1761         /* Consume valid dirty HPTEs */
1762         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1763                && HPTE_DIRTY(HPTE(spapr->htab, index))
1764                && HPTE_VALID(HPTE(spapr->htab, index))) {
1765             CLEAN_HPTE(HPTE(spapr->htab, index));
1766             index++;
1767             examined++;
1768         }
1769 
1770         invalidstart = index;
1771         /* Consume invalid dirty HPTEs */
1772         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1773                && HPTE_DIRTY(HPTE(spapr->htab, index))
1774                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1775             CLEAN_HPTE(HPTE(spapr->htab, index));
1776             index++;
1777             examined++;
1778         }
1779 
1780         if (index > chunkstart) {
1781             int n_valid = invalidstart - chunkstart;
1782             int n_invalid = index - invalidstart;
1783 
1784             qemu_put_be32(f, chunkstart);
1785             qemu_put_be16(f, n_valid);
1786             qemu_put_be16(f, n_invalid);
1787             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1788                             HASH_PTE_SIZE_64 * n_valid);
1789             sent += index - chunkstart;
1790 
1791             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1792                 break;
1793             }
1794         }
1795 
1796         if (examined >= htabslots) {
1797             break;
1798         }
1799 
1800         if (index >= htabslots) {
1801             assert(index == htabslots);
1802             index = 0;
1803         }
1804     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1805 
1806     if (index >= htabslots) {
1807         assert(index == htabslots);
1808         index = 0;
1809     }
1810 
1811     spapr->htab_save_index = index;
1812 
1813     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1814 }
1815 
1816 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1817 #define MAX_KVM_BUF_SIZE    2048
1818 
1819 static int htab_save_iterate(QEMUFile *f, void *opaque)
1820 {
1821     sPAPRMachineState *spapr = opaque;
1822     int fd;
1823     int rc = 0;
1824 
1825     /* Iteration header */
1826     if (!spapr->htab_shift) {
1827         qemu_put_be32(f, -1);
1828         return 1;
1829     } else {
1830         qemu_put_be32(f, 0);
1831     }
1832 
1833     if (!spapr->htab) {
1834         assert(kvm_enabled());
1835 
1836         fd = get_htab_fd(spapr);
1837         if (fd < 0) {
1838             return fd;
1839         }
1840 
1841         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1842         if (rc < 0) {
1843             return rc;
1844         }
1845     } else  if (spapr->htab_first_pass) {
1846         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1847     } else {
1848         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1849     }
1850 
1851     /* End marker */
1852     qemu_put_be32(f, 0);
1853     qemu_put_be16(f, 0);
1854     qemu_put_be16(f, 0);
1855 
1856     return rc;
1857 }
1858 
1859 static int htab_save_complete(QEMUFile *f, void *opaque)
1860 {
1861     sPAPRMachineState *spapr = opaque;
1862     int fd;
1863 
1864     /* Iteration header */
1865     if (!spapr->htab_shift) {
1866         qemu_put_be32(f, -1);
1867         return 0;
1868     } else {
1869         qemu_put_be32(f, 0);
1870     }
1871 
1872     if (!spapr->htab) {
1873         int rc;
1874 
1875         assert(kvm_enabled());
1876 
1877         fd = get_htab_fd(spapr);
1878         if (fd < 0) {
1879             return fd;
1880         }
1881 
1882         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1883         if (rc < 0) {
1884             return rc;
1885         }
1886     } else {
1887         if (spapr->htab_first_pass) {
1888             htab_save_first_pass(f, spapr, -1);
1889         }
1890         htab_save_later_pass(f, spapr, -1);
1891     }
1892 
1893     /* End marker */
1894     qemu_put_be32(f, 0);
1895     qemu_put_be16(f, 0);
1896     qemu_put_be16(f, 0);
1897 
1898     return 0;
1899 }
1900 
1901 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1902 {
1903     sPAPRMachineState *spapr = opaque;
1904     uint32_t section_hdr;
1905     int fd = -1;
1906 
1907     if (version_id < 1 || version_id > 1) {
1908         error_report("htab_load() bad version");
1909         return -EINVAL;
1910     }
1911 
1912     section_hdr = qemu_get_be32(f);
1913 
1914     if (section_hdr == -1) {
1915         spapr_free_hpt(spapr);
1916         return 0;
1917     }
1918 
1919     if (section_hdr) {
1920         Error *local_err = NULL;
1921 
1922         /* First section gives the htab size */
1923         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1924         if (local_err) {
1925             error_report_err(local_err);
1926             return -EINVAL;
1927         }
1928         return 0;
1929     }
1930 
1931     if (!spapr->htab) {
1932         assert(kvm_enabled());
1933 
1934         fd = kvmppc_get_htab_fd(true);
1935         if (fd < 0) {
1936             error_report("Unable to open fd to restore KVM hash table: %s",
1937                          strerror(errno));
1938         }
1939     }
1940 
1941     while (true) {
1942         uint32_t index;
1943         uint16_t n_valid, n_invalid;
1944 
1945         index = qemu_get_be32(f);
1946         n_valid = qemu_get_be16(f);
1947         n_invalid = qemu_get_be16(f);
1948 
1949         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1950             /* End of Stream */
1951             break;
1952         }
1953 
1954         if ((index + n_valid + n_invalid) >
1955             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1956             /* Bad index in stream */
1957             error_report(
1958                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1959                 index, n_valid, n_invalid, spapr->htab_shift);
1960             return -EINVAL;
1961         }
1962 
1963         if (spapr->htab) {
1964             if (n_valid) {
1965                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1966                                 HASH_PTE_SIZE_64 * n_valid);
1967             }
1968             if (n_invalid) {
1969                 memset(HPTE(spapr->htab, index + n_valid), 0,
1970                        HASH_PTE_SIZE_64 * n_invalid);
1971             }
1972         } else {
1973             int rc;
1974 
1975             assert(fd >= 0);
1976 
1977             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1978             if (rc < 0) {
1979                 return rc;
1980             }
1981         }
1982     }
1983 
1984     if (!spapr->htab) {
1985         assert(fd >= 0);
1986         close(fd);
1987     }
1988 
1989     return 0;
1990 }
1991 
1992 static void htab_save_cleanup(void *opaque)
1993 {
1994     sPAPRMachineState *spapr = opaque;
1995 
1996     close_htab_fd(spapr);
1997 }
1998 
1999 static SaveVMHandlers savevm_htab_handlers = {
2000     .save_setup = htab_save_setup,
2001     .save_live_iterate = htab_save_iterate,
2002     .save_live_complete_precopy = htab_save_complete,
2003     .save_cleanup = htab_save_cleanup,
2004     .load_state = htab_load,
2005 };
2006 
2007 static void spapr_boot_set(void *opaque, const char *boot_device,
2008                            Error **errp)
2009 {
2010     MachineState *machine = MACHINE(qdev_get_machine());
2011     machine->boot_order = g_strdup(boot_device);
2012 }
2013 
2014 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2015 {
2016     MachineState *machine = MACHINE(spapr);
2017     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2018     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2019     int i;
2020 
2021     for (i = 0; i < nr_lmbs; i++) {
2022         uint64_t addr;
2023 
2024         addr = i * lmb_size + spapr->hotplug_memory.base;
2025         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2026                                addr / lmb_size);
2027     }
2028 }
2029 
2030 /*
2031  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2032  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2033  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2034  */
2035 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2036 {
2037     int i;
2038 
2039     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2040         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2041                    " is not aligned to %llu MiB",
2042                    machine->ram_size,
2043                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2044         return;
2045     }
2046 
2047     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2048         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2049                    " is not aligned to %llu MiB",
2050                    machine->ram_size,
2051                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2052         return;
2053     }
2054 
2055     for (i = 0; i < nb_numa_nodes; i++) {
2056         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2057             error_setg(errp,
2058                        "Node %d memory size 0x%" PRIx64
2059                        " is not aligned to %llu MiB",
2060                        i, numa_info[i].node_mem,
2061                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2062             return;
2063         }
2064     }
2065 }
2066 
2067 /* find cpu slot in machine->possible_cpus by core_id */
2068 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2069 {
2070     int index = id / smp_threads;
2071 
2072     if (index >= ms->possible_cpus->len) {
2073         return NULL;
2074     }
2075     if (idx) {
2076         *idx = index;
2077     }
2078     return &ms->possible_cpus->cpus[index];
2079 }
2080 
2081 static void spapr_init_cpus(sPAPRMachineState *spapr)
2082 {
2083     MachineState *machine = MACHINE(spapr);
2084     MachineClass *mc = MACHINE_GET_CLASS(machine);
2085     char *type = spapr_get_cpu_core_type(machine->cpu_model);
2086     int smt = kvmppc_smt_threads();
2087     const CPUArchIdList *possible_cpus;
2088     int boot_cores_nr = smp_cpus / smp_threads;
2089     int i;
2090 
2091     if (!type) {
2092         error_report("Unable to find sPAPR CPU Core definition");
2093         exit(1);
2094     }
2095 
2096     possible_cpus = mc->possible_cpu_arch_ids(machine);
2097     if (mc->has_hotpluggable_cpus) {
2098         if (smp_cpus % smp_threads) {
2099             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2100                          smp_cpus, smp_threads);
2101             exit(1);
2102         }
2103         if (max_cpus % smp_threads) {
2104             error_report("max_cpus (%u) must be multiple of threads (%u)",
2105                          max_cpus, smp_threads);
2106             exit(1);
2107         }
2108     } else {
2109         if (max_cpus != smp_cpus) {
2110             error_report("This machine version does not support CPU hotplug");
2111             exit(1);
2112         }
2113         boot_cores_nr = possible_cpus->len;
2114     }
2115 
2116     for (i = 0; i < possible_cpus->len; i++) {
2117         int core_id = i * smp_threads;
2118 
2119         if (mc->has_hotpluggable_cpus) {
2120             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2121                                    (core_id / smp_threads) * smt);
2122         }
2123 
2124         if (i < boot_cores_nr) {
2125             Object *core  = object_new(type);
2126             int nr_threads = smp_threads;
2127 
2128             /* Handle the partially filled core for older machine types */
2129             if ((i + 1) * smp_threads >= smp_cpus) {
2130                 nr_threads = smp_cpus - i * smp_threads;
2131             }
2132 
2133             object_property_set_int(core, nr_threads, "nr-threads",
2134                                     &error_fatal);
2135             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2136                                     &error_fatal);
2137             object_property_set_bool(core, true, "realized", &error_fatal);
2138         }
2139     }
2140     g_free(type);
2141 }
2142 
2143 /* pSeries LPAR / sPAPR hardware init */
2144 static void ppc_spapr_init(MachineState *machine)
2145 {
2146     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2147     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2148     const char *kernel_filename = machine->kernel_filename;
2149     const char *initrd_filename = machine->initrd_filename;
2150     PCIHostState *phb;
2151     int i;
2152     MemoryRegion *sysmem = get_system_memory();
2153     MemoryRegion *ram = g_new(MemoryRegion, 1);
2154     MemoryRegion *rma_region;
2155     void *rma = NULL;
2156     hwaddr rma_alloc_size;
2157     hwaddr node0_size = spapr_node0_size();
2158     long load_limit, fw_size;
2159     char *filename;
2160     Error *resize_hpt_err = NULL;
2161 
2162     msi_nonbroken = true;
2163 
2164     QLIST_INIT(&spapr->phbs);
2165     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2166 
2167     /* Check HPT resizing availability */
2168     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2169     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2170         /*
2171          * If the user explicitly requested a mode we should either
2172          * supply it, or fail completely (which we do below).  But if
2173          * it's not set explicitly, we reset our mode to something
2174          * that works
2175          */
2176         if (resize_hpt_err) {
2177             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2178             error_free(resize_hpt_err);
2179             resize_hpt_err = NULL;
2180         } else {
2181             spapr->resize_hpt = smc->resize_hpt_default;
2182         }
2183     }
2184 
2185     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2186 
2187     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2188         /*
2189          * User requested HPT resize, but this host can't supply it.  Bail out
2190          */
2191         error_report_err(resize_hpt_err);
2192         exit(1);
2193     }
2194 
2195     /* Allocate RMA if necessary */
2196     rma_alloc_size = kvmppc_alloc_rma(&rma);
2197 
2198     if (rma_alloc_size == -1) {
2199         error_report("Unable to create RMA");
2200         exit(1);
2201     }
2202 
2203     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2204         spapr->rma_size = rma_alloc_size;
2205     } else {
2206         spapr->rma_size = node0_size;
2207 
2208         /* With KVM, we don't actually know whether KVM supports an
2209          * unbounded RMA (PR KVM) or is limited by the hash table size
2210          * (HV KVM using VRMA), so we always assume the latter
2211          *
2212          * In that case, we also limit the initial allocations for RTAS
2213          * etc... to 256M since we have no way to know what the VRMA size
2214          * is going to be as it depends on the size of the hash table
2215          * isn't determined yet.
2216          */
2217         if (kvm_enabled()) {
2218             spapr->vrma_adjust = 1;
2219             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2220         }
2221 
2222         /* Actually we don't support unbounded RMA anymore since we
2223          * added proper emulation of HV mode. The max we can get is
2224          * 16G which also happens to be what we configure for PAPR
2225          * mode so make sure we don't do anything bigger than that
2226          */
2227         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2228     }
2229 
2230     if (spapr->rma_size > node0_size) {
2231         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2232                      spapr->rma_size);
2233         exit(1);
2234     }
2235 
2236     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2237     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2238 
2239     /* Set up Interrupt Controller before we create the VCPUs */
2240     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2241 
2242     /* Set up containers for ibm,client-set-architecture negotiated options */
2243     spapr->ov5 = spapr_ovec_new();
2244     spapr->ov5_cas = spapr_ovec_new();
2245 
2246     if (smc->dr_lmb_enabled) {
2247         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2248         spapr_validate_node_memory(machine, &error_fatal);
2249     }
2250 
2251     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2252     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2253         /* KVM and TCG always allow GTSE with radix... */
2254         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2255     }
2256     /* ... but not with hash (currently). */
2257 
2258     /* advertise support for dedicated HP event source to guests */
2259     if (spapr->use_hotplug_event_source) {
2260         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2261     }
2262 
2263     /* advertise support for HPT resizing */
2264     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2265         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2266     }
2267 
2268     /* init CPUs */
2269     if (machine->cpu_model == NULL) {
2270         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2271     }
2272 
2273     spapr_cpu_parse_features(spapr);
2274 
2275     spapr_init_cpus(spapr);
2276 
2277     if (kvm_enabled()) {
2278         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2279         kvmppc_enable_logical_ci_hcalls();
2280         kvmppc_enable_set_mode_hcall();
2281 
2282         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2283         kvmppc_enable_clear_ref_mod_hcalls();
2284     }
2285 
2286     /* allocate RAM */
2287     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2288                                          machine->ram_size);
2289     memory_region_add_subregion(sysmem, 0, ram);
2290 
2291     if (rma_alloc_size && rma) {
2292         rma_region = g_new(MemoryRegion, 1);
2293         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2294                                    rma_alloc_size, rma);
2295         vmstate_register_ram_global(rma_region);
2296         memory_region_add_subregion(sysmem, 0, rma_region);
2297     }
2298 
2299     /* initialize hotplug memory address space */
2300     if (machine->ram_size < machine->maxram_size) {
2301         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2302         /*
2303          * Limit the number of hotpluggable memory slots to half the number
2304          * slots that KVM supports, leaving the other half for PCI and other
2305          * devices. However ensure that number of slots doesn't drop below 32.
2306          */
2307         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2308                            SPAPR_MAX_RAM_SLOTS;
2309 
2310         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2311             max_memslots = SPAPR_MAX_RAM_SLOTS;
2312         }
2313         if (machine->ram_slots > max_memslots) {
2314             error_report("Specified number of memory slots %"
2315                          PRIu64" exceeds max supported %d",
2316                          machine->ram_slots, max_memslots);
2317             exit(1);
2318         }
2319 
2320         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2321                                               SPAPR_HOTPLUG_MEM_ALIGN);
2322         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2323                            "hotplug-memory", hotplug_mem_size);
2324         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2325                                     &spapr->hotplug_memory.mr);
2326     }
2327 
2328     if (smc->dr_lmb_enabled) {
2329         spapr_create_lmb_dr_connectors(spapr);
2330     }
2331 
2332     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2333     if (!filename) {
2334         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2335         exit(1);
2336     }
2337     spapr->rtas_size = get_image_size(filename);
2338     if (spapr->rtas_size < 0) {
2339         error_report("Could not get size of LPAR rtas '%s'", filename);
2340         exit(1);
2341     }
2342     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2343     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2344         error_report("Could not load LPAR rtas '%s'", filename);
2345         exit(1);
2346     }
2347     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2348         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2349                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2350         exit(1);
2351     }
2352     g_free(filename);
2353 
2354     /* Set up RTAS event infrastructure */
2355     spapr_events_init(spapr);
2356 
2357     /* Set up the RTC RTAS interfaces */
2358     spapr_rtc_create(spapr);
2359 
2360     /* Set up VIO bus */
2361     spapr->vio_bus = spapr_vio_bus_init();
2362 
2363     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2364         if (serial_hds[i]) {
2365             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2366         }
2367     }
2368 
2369     /* We always have at least the nvram device on VIO */
2370     spapr_create_nvram(spapr);
2371 
2372     /* Set up PCI */
2373     spapr_pci_rtas_init();
2374 
2375     phb = spapr_create_phb(spapr, 0);
2376 
2377     for (i = 0; i < nb_nics; i++) {
2378         NICInfo *nd = &nd_table[i];
2379 
2380         if (!nd->model) {
2381             nd->model = g_strdup("ibmveth");
2382         }
2383 
2384         if (strcmp(nd->model, "ibmveth") == 0) {
2385             spapr_vlan_create(spapr->vio_bus, nd);
2386         } else {
2387             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2388         }
2389     }
2390 
2391     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2392         spapr_vscsi_create(spapr->vio_bus);
2393     }
2394 
2395     /* Graphics */
2396     if (spapr_vga_init(phb->bus, &error_fatal)) {
2397         spapr->has_graphics = true;
2398         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2399     }
2400 
2401     if (machine->usb) {
2402         if (smc->use_ohci_by_default) {
2403             pci_create_simple(phb->bus, -1, "pci-ohci");
2404         } else {
2405             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2406         }
2407 
2408         if (spapr->has_graphics) {
2409             USBBus *usb_bus = usb_bus_find(-1);
2410 
2411             usb_create_simple(usb_bus, "usb-kbd");
2412             usb_create_simple(usb_bus, "usb-mouse");
2413         }
2414     }
2415 
2416     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2417         error_report(
2418             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2419             MIN_RMA_SLOF);
2420         exit(1);
2421     }
2422 
2423     if (kernel_filename) {
2424         uint64_t lowaddr = 0;
2425 
2426         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2427                                       NULL, NULL, &lowaddr, NULL, 1,
2428                                       PPC_ELF_MACHINE, 0, 0);
2429         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2430             spapr->kernel_size = load_elf(kernel_filename,
2431                                           translate_kernel_address, NULL, NULL,
2432                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2433                                           0, 0);
2434             spapr->kernel_le = spapr->kernel_size > 0;
2435         }
2436         if (spapr->kernel_size < 0) {
2437             error_report("error loading %s: %s", kernel_filename,
2438                          load_elf_strerror(spapr->kernel_size));
2439             exit(1);
2440         }
2441 
2442         /* load initrd */
2443         if (initrd_filename) {
2444             /* Try to locate the initrd in the gap between the kernel
2445              * and the firmware. Add a bit of space just in case
2446              */
2447             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2448                                   + 0x1ffff) & ~0xffff;
2449             spapr->initrd_size = load_image_targphys(initrd_filename,
2450                                                      spapr->initrd_base,
2451                                                      load_limit
2452                                                      - spapr->initrd_base);
2453             if (spapr->initrd_size < 0) {
2454                 error_report("could not load initial ram disk '%s'",
2455                              initrd_filename);
2456                 exit(1);
2457             }
2458         }
2459     }
2460 
2461     if (bios_name == NULL) {
2462         bios_name = FW_FILE_NAME;
2463     }
2464     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2465     if (!filename) {
2466         error_report("Could not find LPAR firmware '%s'", bios_name);
2467         exit(1);
2468     }
2469     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2470     if (fw_size <= 0) {
2471         error_report("Could not load LPAR firmware '%s'", filename);
2472         exit(1);
2473     }
2474     g_free(filename);
2475 
2476     /* FIXME: Should register things through the MachineState's qdev
2477      * interface, this is a legacy from the sPAPREnvironment structure
2478      * which predated MachineState but had a similar function */
2479     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2480     register_savevm_live(NULL, "spapr/htab", -1, 1,
2481                          &savevm_htab_handlers, spapr);
2482 
2483     qemu_register_boot_set(spapr_boot_set, spapr);
2484 
2485     if (kvm_enabled()) {
2486         /* to stop and start vmclock */
2487         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2488                                          &spapr->tb);
2489 
2490         kvmppc_spapr_enable_inkernel_multitce();
2491     }
2492 }
2493 
2494 static int spapr_kvm_type(const char *vm_type)
2495 {
2496     if (!vm_type) {
2497         return 0;
2498     }
2499 
2500     if (!strcmp(vm_type, "HV")) {
2501         return 1;
2502     }
2503 
2504     if (!strcmp(vm_type, "PR")) {
2505         return 2;
2506     }
2507 
2508     error_report("Unknown kvm-type specified '%s'", vm_type);
2509     exit(1);
2510 }
2511 
2512 /*
2513  * Implementation of an interface to adjust firmware path
2514  * for the bootindex property handling.
2515  */
2516 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2517                                    DeviceState *dev)
2518 {
2519 #define CAST(type, obj, name) \
2520     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2521     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2522     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2523     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2524 
2525     if (d) {
2526         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2527         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2528         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2529 
2530         if (spapr) {
2531             /*
2532              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2533              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2534              * in the top 16 bits of the 64-bit LUN
2535              */
2536             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2537             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2538                                    (uint64_t)id << 48);
2539         } else if (virtio) {
2540             /*
2541              * We use SRP luns of the form 01000000 | (target << 8) | lun
2542              * in the top 32 bits of the 64-bit LUN
2543              * Note: the quote above is from SLOF and it is wrong,
2544              * the actual binding is:
2545              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2546              */
2547             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2548             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2549                                    (uint64_t)id << 32);
2550         } else if (usb) {
2551             /*
2552              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2553              * in the top 32 bits of the 64-bit LUN
2554              */
2555             unsigned usb_port = atoi(usb->port->path);
2556             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2557             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2558                                    (uint64_t)id << 32);
2559         }
2560     }
2561 
2562     /*
2563      * SLOF probes the USB devices, and if it recognizes that the device is a
2564      * storage device, it changes its name to "storage" instead of "usb-host",
2565      * and additionally adds a child node for the SCSI LUN, so the correct
2566      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2567      */
2568     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2569         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2570         if (usb_host_dev_is_scsi_storage(usbdev)) {
2571             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2572         }
2573     }
2574 
2575     if (phb) {
2576         /* Replace "pci" with "pci@800000020000000" */
2577         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2578     }
2579 
2580     if (vsc) {
2581         /* Same logic as virtio above */
2582         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2583         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2584     }
2585 
2586     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2587         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2588         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2589         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2590     }
2591 
2592     return NULL;
2593 }
2594 
2595 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2596 {
2597     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2598 
2599     return g_strdup(spapr->kvm_type);
2600 }
2601 
2602 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2603 {
2604     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2605 
2606     g_free(spapr->kvm_type);
2607     spapr->kvm_type = g_strdup(value);
2608 }
2609 
2610 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2611 {
2612     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2613 
2614     return spapr->use_hotplug_event_source;
2615 }
2616 
2617 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2618                                             Error **errp)
2619 {
2620     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2621 
2622     spapr->use_hotplug_event_source = value;
2623 }
2624 
2625 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2626 {
2627     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2628 
2629     switch (spapr->resize_hpt) {
2630     case SPAPR_RESIZE_HPT_DEFAULT:
2631         return g_strdup("default");
2632     case SPAPR_RESIZE_HPT_DISABLED:
2633         return g_strdup("disabled");
2634     case SPAPR_RESIZE_HPT_ENABLED:
2635         return g_strdup("enabled");
2636     case SPAPR_RESIZE_HPT_REQUIRED:
2637         return g_strdup("required");
2638     }
2639     g_assert_not_reached();
2640 }
2641 
2642 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2643 {
2644     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2645 
2646     if (strcmp(value, "default") == 0) {
2647         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2648     } else if (strcmp(value, "disabled") == 0) {
2649         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2650     } else if (strcmp(value, "enabled") == 0) {
2651         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2652     } else if (strcmp(value, "required") == 0) {
2653         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2654     } else {
2655         error_setg(errp, "Bad value for \"resize-hpt\" property");
2656     }
2657 }
2658 
2659 static void spapr_machine_initfn(Object *obj)
2660 {
2661     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2662 
2663     spapr->htab_fd = -1;
2664     spapr->use_hotplug_event_source = true;
2665     object_property_add_str(obj, "kvm-type",
2666                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2667     object_property_set_description(obj, "kvm-type",
2668                                     "Specifies the KVM virtualization mode (HV, PR)",
2669                                     NULL);
2670     object_property_add_bool(obj, "modern-hotplug-events",
2671                             spapr_get_modern_hotplug_events,
2672                             spapr_set_modern_hotplug_events,
2673                             NULL);
2674     object_property_set_description(obj, "modern-hotplug-events",
2675                                     "Use dedicated hotplug event mechanism in"
2676                                     " place of standard EPOW events when possible"
2677                                     " (required for memory hot-unplug support)",
2678                                     NULL);
2679 
2680     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
2681                             "Maximum permitted CPU compatibility mode",
2682                             &error_fatal);
2683 
2684     object_property_add_str(obj, "resize-hpt",
2685                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
2686     object_property_set_description(obj, "resize-hpt",
2687                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
2688                                     NULL);
2689 }
2690 
2691 static void spapr_machine_finalizefn(Object *obj)
2692 {
2693     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2694 
2695     g_free(spapr->kvm_type);
2696 }
2697 
2698 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2699 {
2700     cpu_synchronize_state(cs);
2701     ppc_cpu_do_system_reset(cs);
2702 }
2703 
2704 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2705 {
2706     CPUState *cs;
2707 
2708     CPU_FOREACH(cs) {
2709         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2710     }
2711 }
2712 
2713 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2714                            uint32_t node, bool dedicated_hp_event_source,
2715                            Error **errp)
2716 {
2717     sPAPRDRConnector *drc;
2718     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2719     int i, fdt_offset, fdt_size;
2720     void *fdt;
2721     uint64_t addr = addr_start;
2722     bool hotplugged = spapr_drc_hotplugged(dev);
2723     Error *local_err = NULL;
2724 
2725     for (i = 0; i < nr_lmbs; i++) {
2726         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2727                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2728         g_assert(drc);
2729 
2730         fdt = create_device_tree(&fdt_size);
2731         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2732                                                 SPAPR_MEMORY_BLOCK_SIZE);
2733 
2734         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
2735         if (local_err) {
2736             while (addr > addr_start) {
2737                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
2738                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2739                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
2740                 spapr_drc_detach(drc);
2741             }
2742             g_free(fdt);
2743             error_propagate(errp, local_err);
2744             return;
2745         }
2746         if (!hotplugged) {
2747             spapr_drc_reset(drc);
2748         }
2749         addr += SPAPR_MEMORY_BLOCK_SIZE;
2750     }
2751     /* send hotplug notification to the
2752      * guest only in case of hotplugged memory
2753      */
2754     if (hotplugged) {
2755         if (dedicated_hp_event_source) {
2756             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2757                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2758             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2759                                                    nr_lmbs,
2760                                                    spapr_drc_index(drc));
2761         } else {
2762             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2763                                            nr_lmbs);
2764         }
2765     }
2766 }
2767 
2768 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2769                               uint32_t node, Error **errp)
2770 {
2771     Error *local_err = NULL;
2772     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2773     PCDIMMDevice *dimm = PC_DIMM(dev);
2774     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2775     MemoryRegion *mr;
2776     uint64_t align, size, addr;
2777 
2778     mr = ddc->get_memory_region(dimm, &local_err);
2779     if (local_err) {
2780         goto out;
2781     }
2782     align = memory_region_get_alignment(mr);
2783     size = memory_region_size(mr);
2784 
2785     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2786     if (local_err) {
2787         goto out;
2788     }
2789 
2790     addr = object_property_get_uint(OBJECT(dimm),
2791                                     PC_DIMM_ADDR_PROP, &local_err);
2792     if (local_err) {
2793         goto out_unplug;
2794     }
2795 
2796     spapr_add_lmbs(dev, addr, size, node,
2797                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2798                    &local_err);
2799     if (local_err) {
2800         goto out_unplug;
2801     }
2802 
2803     return;
2804 
2805 out_unplug:
2806     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2807 out:
2808     error_propagate(errp, local_err);
2809 }
2810 
2811 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2812                                   Error **errp)
2813 {
2814     PCDIMMDevice *dimm = PC_DIMM(dev);
2815     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2816     MemoryRegion *mr;
2817     uint64_t size;
2818     char *mem_dev;
2819 
2820     mr = ddc->get_memory_region(dimm, errp);
2821     if (!mr) {
2822         return;
2823     }
2824     size = memory_region_size(mr);
2825 
2826     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2827         error_setg(errp, "Hotplugged memory size must be a multiple of "
2828                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2829         return;
2830     }
2831 
2832     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2833     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2834         error_setg(errp, "Memory backend has bad page size. "
2835                    "Use 'memory-backend-file' with correct mem-path.");
2836         goto out;
2837     }
2838 
2839 out:
2840     g_free(mem_dev);
2841 }
2842 
2843 struct sPAPRDIMMState {
2844     PCDIMMDevice *dimm;
2845     uint32_t nr_lmbs;
2846     QTAILQ_ENTRY(sPAPRDIMMState) next;
2847 };
2848 
2849 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
2850                                                        PCDIMMDevice *dimm)
2851 {
2852     sPAPRDIMMState *dimm_state = NULL;
2853 
2854     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
2855         if (dimm_state->dimm == dimm) {
2856             break;
2857         }
2858     }
2859     return dimm_state;
2860 }
2861 
2862 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
2863                                                       uint32_t nr_lmbs,
2864                                                       PCDIMMDevice *dimm)
2865 {
2866     sPAPRDIMMState *ds = NULL;
2867 
2868     /*
2869      * If this request is for a DIMM whose removal had failed earlier
2870      * (due to guest's refusal to remove the LMBs), we would have this
2871      * dimm already in the pending_dimm_unplugs list. In that
2872      * case don't add again.
2873      */
2874     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
2875     if (!ds) {
2876         ds = g_malloc0(sizeof(sPAPRDIMMState));
2877         ds->nr_lmbs = nr_lmbs;
2878         ds->dimm = dimm;
2879         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
2880     }
2881     return ds;
2882 }
2883 
2884 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
2885                                               sPAPRDIMMState *dimm_state)
2886 {
2887     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
2888     g_free(dimm_state);
2889 }
2890 
2891 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
2892                                                         PCDIMMDevice *dimm)
2893 {
2894     sPAPRDRConnector *drc;
2895     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2896     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
2897     uint64_t size = memory_region_size(mr);
2898     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2899     uint32_t avail_lmbs = 0;
2900     uint64_t addr_start, addr;
2901     int i;
2902 
2903     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2904                                          &error_abort);
2905 
2906     addr = addr_start;
2907     for (i = 0; i < nr_lmbs; i++) {
2908         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2909                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2910         g_assert(drc);
2911         if (drc->dev) {
2912             avail_lmbs++;
2913         }
2914         addr += SPAPR_MEMORY_BLOCK_SIZE;
2915     }
2916 
2917     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
2918 }
2919 
2920 /* Callback to be called during DRC release. */
2921 void spapr_lmb_release(DeviceState *dev)
2922 {
2923     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_hotplug_handler(dev));
2924     PCDIMMDevice *dimm = PC_DIMM(dev);
2925     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2926     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
2927     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
2928 
2929     /* This information will get lost if a migration occurs
2930      * during the unplug process. In this case recover it. */
2931     if (ds == NULL) {
2932         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
2933         g_assert(ds);
2934         /* The DRC being examined by the caller at least must be counted */
2935         g_assert(ds->nr_lmbs);
2936     }
2937 
2938     if (--ds->nr_lmbs) {
2939         return;
2940     }
2941 
2942     spapr_pending_dimm_unplugs_remove(spapr, ds);
2943 
2944     /*
2945      * Now that all the LMBs have been removed by the guest, call the
2946      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2947      */
2948     pc_dimm_memory_unplug(dev, &spapr->hotplug_memory, mr);
2949     object_unparent(OBJECT(dev));
2950 }
2951 
2952 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2953                                         DeviceState *dev, Error **errp)
2954 {
2955     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
2956     Error *local_err = NULL;
2957     PCDIMMDevice *dimm = PC_DIMM(dev);
2958     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2959     MemoryRegion *mr;
2960     uint32_t nr_lmbs;
2961     uint64_t size, addr_start, addr;
2962     int i;
2963     sPAPRDRConnector *drc;
2964 
2965     mr = ddc->get_memory_region(dimm, &local_err);
2966     if (local_err) {
2967         goto out;
2968     }
2969     size = memory_region_size(mr);
2970     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2971 
2972     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2973                                          &local_err);
2974     if (local_err) {
2975         goto out;
2976     }
2977 
2978     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
2979 
2980     addr = addr_start;
2981     for (i = 0; i < nr_lmbs; i++) {
2982         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2983                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2984         g_assert(drc);
2985 
2986         spapr_drc_detach(drc);
2987         addr += SPAPR_MEMORY_BLOCK_SIZE;
2988     }
2989 
2990     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2991                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2992     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2993                                               nr_lmbs, spapr_drc_index(drc));
2994 out:
2995     error_propagate(errp, local_err);
2996 }
2997 
2998 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2999                                            sPAPRMachineState *spapr)
3000 {
3001     PowerPCCPU *cpu = POWERPC_CPU(cs);
3002     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3003     int id = ppc_get_vcpu_dt_id(cpu);
3004     void *fdt;
3005     int offset, fdt_size;
3006     char *nodename;
3007 
3008     fdt = create_device_tree(&fdt_size);
3009     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3010     offset = fdt_add_subnode(fdt, 0, nodename);
3011 
3012     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3013     g_free(nodename);
3014 
3015     *fdt_offset = offset;
3016     return fdt;
3017 }
3018 
3019 /* Callback to be called during DRC release. */
3020 void spapr_core_release(DeviceState *dev)
3021 {
3022     MachineState *ms = MACHINE(qdev_get_hotplug_handler(dev));
3023     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3024     CPUCore *cc = CPU_CORE(dev);
3025     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3026 
3027     if (smc->pre_2_10_has_unused_icps) {
3028         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3029         sPAPRCPUCoreClass *scc = SPAPR_CPU_CORE_GET_CLASS(OBJECT(cc));
3030         const char *typename = object_class_get_name(scc->cpu_class);
3031         size_t size = object_type_get_instance_size(typename);
3032         int i;
3033 
3034         for (i = 0; i < cc->nr_threads; i++) {
3035             CPUState *cs = CPU(sc->threads + i * size);
3036 
3037             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3038         }
3039     }
3040 
3041     assert(core_slot);
3042     core_slot->cpu = NULL;
3043     object_unparent(OBJECT(dev));
3044 }
3045 
3046 static
3047 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3048                                Error **errp)
3049 {
3050     int index;
3051     sPAPRDRConnector *drc;
3052     CPUCore *cc = CPU_CORE(dev);
3053     int smt = kvmppc_smt_threads();
3054 
3055     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3056         error_setg(errp, "Unable to find CPU core with core-id: %d",
3057                    cc->core_id);
3058         return;
3059     }
3060     if (index == 0) {
3061         error_setg(errp, "Boot CPU core may not be unplugged");
3062         return;
3063     }
3064 
3065     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3066     g_assert(drc);
3067 
3068     spapr_drc_detach(drc);
3069 
3070     spapr_hotplug_req_remove_by_index(drc);
3071 }
3072 
3073 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3074                             Error **errp)
3075 {
3076     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3077     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3078     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3079     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3080     CPUCore *cc = CPU_CORE(dev);
3081     CPUState *cs = CPU(core->threads);
3082     sPAPRDRConnector *drc;
3083     Error *local_err = NULL;
3084     int smt = kvmppc_smt_threads();
3085     CPUArchId *core_slot;
3086     int index;
3087     bool hotplugged = spapr_drc_hotplugged(dev);
3088 
3089     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3090     if (!core_slot) {
3091         error_setg(errp, "Unable to find CPU core with core-id: %d",
3092                    cc->core_id);
3093         return;
3094     }
3095     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3096 
3097     g_assert(drc || !mc->has_hotpluggable_cpus);
3098 
3099     if (drc) {
3100         void *fdt;
3101         int fdt_offset;
3102 
3103         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3104 
3105         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3106         if (local_err) {
3107             g_free(fdt);
3108             error_propagate(errp, local_err);
3109             return;
3110         }
3111 
3112         if (hotplugged) {
3113             /*
3114              * Send hotplug notification interrupt to the guest only
3115              * in case of hotplugged CPUs.
3116              */
3117             spapr_hotplug_req_add_by_index(drc);
3118         } else {
3119             spapr_drc_reset(drc);
3120         }
3121     }
3122 
3123     core_slot->cpu = OBJECT(dev);
3124 
3125     if (smc->pre_2_10_has_unused_icps) {
3126         sPAPRCPUCoreClass *scc = SPAPR_CPU_CORE_GET_CLASS(OBJECT(cc));
3127         const char *typename = object_class_get_name(scc->cpu_class);
3128         size_t size = object_type_get_instance_size(typename);
3129         int i;
3130 
3131         for (i = 0; i < cc->nr_threads; i++) {
3132             sPAPRCPUCore *sc = SPAPR_CPU_CORE(dev);
3133             void *obj = sc->threads + i * size;
3134 
3135             cs = CPU(obj);
3136             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3137         }
3138     }
3139 }
3140 
3141 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3142                                 Error **errp)
3143 {
3144     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3145     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3146     Error *local_err = NULL;
3147     CPUCore *cc = CPU_CORE(dev);
3148     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
3149     const char *type = object_get_typename(OBJECT(dev));
3150     CPUArchId *core_slot;
3151     int index;
3152 
3153     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3154         error_setg(&local_err, "CPU hotplug not supported for this machine");
3155         goto out;
3156     }
3157 
3158     if (strcmp(base_core_type, type)) {
3159         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3160         goto out;
3161     }
3162 
3163     if (cc->core_id % smp_threads) {
3164         error_setg(&local_err, "invalid core id %d", cc->core_id);
3165         goto out;
3166     }
3167 
3168     /*
3169      * In general we should have homogeneous threads-per-core, but old
3170      * (pre hotplug support) machine types allow the last core to have
3171      * reduced threads as a compatibility hack for when we allowed
3172      * total vcpus not a multiple of threads-per-core.
3173      */
3174     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3175         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3176                    cc->nr_threads, smp_threads);
3177         goto out;
3178     }
3179 
3180     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3181     if (!core_slot) {
3182         error_setg(&local_err, "core id %d out of range", cc->core_id);
3183         goto out;
3184     }
3185 
3186     if (core_slot->cpu) {
3187         error_setg(&local_err, "core %d already populated", cc->core_id);
3188         goto out;
3189     }
3190 
3191     numa_cpu_pre_plug(core_slot, dev, &local_err);
3192 
3193 out:
3194     g_free(base_core_type);
3195     error_propagate(errp, local_err);
3196 }
3197 
3198 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3199                                       DeviceState *dev, Error **errp)
3200 {
3201     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
3202 
3203     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3204         int node;
3205 
3206         if (!smc->dr_lmb_enabled) {
3207             error_setg(errp, "Memory hotplug not supported for this machine");
3208             return;
3209         }
3210         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3211         if (*errp) {
3212             return;
3213         }
3214         if (node < 0 || node >= MAX_NODES) {
3215             error_setg(errp, "Invaild node %d", node);
3216             return;
3217         }
3218 
3219         /*
3220          * Currently PowerPC kernel doesn't allow hot-adding memory to
3221          * memory-less node, but instead will silently add the memory
3222          * to the first node that has some memory. This causes two
3223          * unexpected behaviours for the user.
3224          *
3225          * - Memory gets hotplugged to a different node than what the user
3226          *   specified.
3227          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3228          *   to memory-less node, a reboot will set things accordingly
3229          *   and the previously hotplugged memory now ends in the right node.
3230          *   This appears as if some memory moved from one node to another.
3231          *
3232          * So until kernel starts supporting memory hotplug to memory-less
3233          * nodes, just prevent such attempts upfront in QEMU.
3234          */
3235         if (nb_numa_nodes && !numa_info[node].node_mem) {
3236             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3237                        node);
3238             return;
3239         }
3240 
3241         spapr_memory_plug(hotplug_dev, dev, node, errp);
3242     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3243         spapr_core_plug(hotplug_dev, dev, errp);
3244     }
3245 }
3246 
3247 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3248                                                 DeviceState *dev, Error **errp)
3249 {
3250     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
3251     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
3252 
3253     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3254         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3255             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3256         } else {
3257             /* NOTE: this means there is a window after guest reset, prior to
3258              * CAS negotiation, where unplug requests will fail due to the
3259              * capability not being detected yet. This is a bit different than
3260              * the case with PCI unplug, where the events will be queued and
3261              * eventually handled by the guest after boot
3262              */
3263             error_setg(errp, "Memory hot unplug not supported for this guest");
3264         }
3265     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3266         if (!mc->has_hotpluggable_cpus) {
3267             error_setg(errp, "CPU hot unplug not supported on this machine");
3268             return;
3269         }
3270         spapr_core_unplug_request(hotplug_dev, dev, errp);
3271     }
3272 }
3273 
3274 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3275                                           DeviceState *dev, Error **errp)
3276 {
3277     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3278         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3279     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3280         spapr_core_pre_plug(hotplug_dev, dev, errp);
3281     }
3282 }
3283 
3284 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3285                                                  DeviceState *dev)
3286 {
3287     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3288         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3289         return HOTPLUG_HANDLER(machine);
3290     }
3291     return NULL;
3292 }
3293 
3294 static CpuInstanceProperties
3295 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3296 {
3297     CPUArchId *core_slot;
3298     MachineClass *mc = MACHINE_GET_CLASS(machine);
3299 
3300     /* make sure possible_cpu are intialized */
3301     mc->possible_cpu_arch_ids(machine);
3302     /* get CPU core slot containing thread that matches cpu_index */
3303     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3304     assert(core_slot);
3305     return core_slot->props;
3306 }
3307 
3308 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3309 {
3310     int i;
3311     int spapr_max_cores = max_cpus / smp_threads;
3312     MachineClass *mc = MACHINE_GET_CLASS(machine);
3313 
3314     if (!mc->has_hotpluggable_cpus) {
3315         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3316     }
3317     if (machine->possible_cpus) {
3318         assert(machine->possible_cpus->len == spapr_max_cores);
3319         return machine->possible_cpus;
3320     }
3321 
3322     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3323                              sizeof(CPUArchId) * spapr_max_cores);
3324     machine->possible_cpus->len = spapr_max_cores;
3325     for (i = 0; i < machine->possible_cpus->len; i++) {
3326         int core_id = i * smp_threads;
3327 
3328         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3329         machine->possible_cpus->cpus[i].arch_id = core_id;
3330         machine->possible_cpus->cpus[i].props.has_core_id = true;
3331         machine->possible_cpus->cpus[i].props.core_id = core_id;
3332 
3333         /* default distribution of CPUs over NUMA nodes */
3334         if (nb_numa_nodes) {
3335             /* preset values but do not enable them i.e. 'has_node_id = false',
3336              * numa init code will enable them later if manual mapping wasn't
3337              * present on CLI */
3338             machine->possible_cpus->cpus[i].props.node_id =
3339                 core_id / smp_threads / smp_cores % nb_numa_nodes;
3340         }
3341     }
3342     return machine->possible_cpus;
3343 }
3344 
3345 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3346                                 uint64_t *buid, hwaddr *pio,
3347                                 hwaddr *mmio32, hwaddr *mmio64,
3348                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3349 {
3350     /*
3351      * New-style PHB window placement.
3352      *
3353      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3354      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3355      * windows.
3356      *
3357      * Some guest kernels can't work with MMIO windows above 1<<46
3358      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3359      *
3360      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3361      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3362      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3363      * 1TiB 64-bit MMIO windows for each PHB.
3364      */
3365     const uint64_t base_buid = 0x800000020000000ULL;
3366 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3367                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3368     int i;
3369 
3370     /* Sanity check natural alignments */
3371     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3372     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3373     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3374     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3375     /* Sanity check bounds */
3376     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3377                       SPAPR_PCI_MEM32_WIN_SIZE);
3378     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3379                       SPAPR_PCI_MEM64_WIN_SIZE);
3380 
3381     if (index >= SPAPR_MAX_PHBS) {
3382         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3383                    SPAPR_MAX_PHBS - 1);
3384         return;
3385     }
3386 
3387     *buid = base_buid + index;
3388     for (i = 0; i < n_dma; ++i) {
3389         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3390     }
3391 
3392     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3393     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3394     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3395 }
3396 
3397 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3398 {
3399     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3400 
3401     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3402 }
3403 
3404 static void spapr_ics_resend(XICSFabric *dev)
3405 {
3406     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3407 
3408     ics_resend(spapr->ics);
3409 }
3410 
3411 static ICPState *spapr_icp_get(XICSFabric *xi, int cpu_dt_id)
3412 {
3413     PowerPCCPU *cpu = ppc_get_vcpu_by_dt_id(cpu_dt_id);
3414 
3415     return cpu ? ICP(cpu->intc) : NULL;
3416 }
3417 
3418 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3419                                  Monitor *mon)
3420 {
3421     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3422     CPUState *cs;
3423 
3424     CPU_FOREACH(cs) {
3425         PowerPCCPU *cpu = POWERPC_CPU(cs);
3426 
3427         icp_pic_print_info(ICP(cpu->intc), mon);
3428     }
3429 
3430     ics_pic_print_info(spapr->ics, mon);
3431 }
3432 
3433 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3434 {
3435     MachineClass *mc = MACHINE_CLASS(oc);
3436     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3437     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3438     NMIClass *nc = NMI_CLASS(oc);
3439     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3440     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3441     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3442     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3443 
3444     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3445 
3446     /*
3447      * We set up the default / latest behaviour here.  The class_init
3448      * functions for the specific versioned machine types can override
3449      * these details for backwards compatibility
3450      */
3451     mc->init = ppc_spapr_init;
3452     mc->reset = ppc_spapr_reset;
3453     mc->block_default_type = IF_SCSI;
3454     mc->max_cpus = 1024;
3455     mc->no_parallel = 1;
3456     mc->default_boot_order = "";
3457     mc->default_ram_size = 512 * M_BYTE;
3458     mc->kvm_type = spapr_kvm_type;
3459     mc->has_dynamic_sysbus = true;
3460     mc->pci_allow_0_address = true;
3461     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3462     hc->pre_plug = spapr_machine_device_pre_plug;
3463     hc->plug = spapr_machine_device_plug;
3464     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3465     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3466     hc->unplug_request = spapr_machine_device_unplug_request;
3467 
3468     smc->dr_lmb_enabled = true;
3469     smc->tcg_default_cpu = "POWER8";
3470     mc->has_hotpluggable_cpus = true;
3471     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3472     fwc->get_dev_path = spapr_get_fw_dev_path;
3473     nc->nmi_monitor_handler = spapr_nmi;
3474     smc->phb_placement = spapr_phb_placement;
3475     vhc->hypercall = emulate_spapr_hypercall;
3476     vhc->hpt_mask = spapr_hpt_mask;
3477     vhc->map_hptes = spapr_map_hptes;
3478     vhc->unmap_hptes = spapr_unmap_hptes;
3479     vhc->store_hpte = spapr_store_hpte;
3480     vhc->get_patbe = spapr_get_patbe;
3481     xic->ics_get = spapr_ics_get;
3482     xic->ics_resend = spapr_ics_resend;
3483     xic->icp_get = spapr_icp_get;
3484     ispc->print_info = spapr_pic_print_info;
3485     /* Force NUMA node memory size to be a multiple of
3486      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3487      * in which LMBs are represented and hot-added
3488      */
3489     mc->numa_mem_align_shift = 28;
3490 }
3491 
3492 static const TypeInfo spapr_machine_info = {
3493     .name          = TYPE_SPAPR_MACHINE,
3494     .parent        = TYPE_MACHINE,
3495     .abstract      = true,
3496     .instance_size = sizeof(sPAPRMachineState),
3497     .instance_init = spapr_machine_initfn,
3498     .instance_finalize = spapr_machine_finalizefn,
3499     .class_size    = sizeof(sPAPRMachineClass),
3500     .class_init    = spapr_machine_class_init,
3501     .interfaces = (InterfaceInfo[]) {
3502         { TYPE_FW_PATH_PROVIDER },
3503         { TYPE_NMI },
3504         { TYPE_HOTPLUG_HANDLER },
3505         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3506         { TYPE_XICS_FABRIC },
3507         { TYPE_INTERRUPT_STATS_PROVIDER },
3508         { }
3509     },
3510 };
3511 
3512 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3513     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3514                                                     void *data)      \
3515     {                                                                \
3516         MachineClass *mc = MACHINE_CLASS(oc);                        \
3517         spapr_machine_##suffix##_class_options(mc);                  \
3518         if (latest) {                                                \
3519             mc->alias = "pseries";                                   \
3520             mc->is_default = 1;                                      \
3521         }                                                            \
3522     }                                                                \
3523     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3524     {                                                                \
3525         MachineState *machine = MACHINE(obj);                        \
3526         spapr_machine_##suffix##_instance_options(machine);          \
3527     }                                                                \
3528     static const TypeInfo spapr_machine_##suffix##_info = {          \
3529         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3530         .parent = TYPE_SPAPR_MACHINE,                                \
3531         .class_init = spapr_machine_##suffix##_class_init,           \
3532         .instance_init = spapr_machine_##suffix##_instance_init,     \
3533     };                                                               \
3534     static void spapr_machine_register_##suffix(void)                \
3535     {                                                                \
3536         type_register(&spapr_machine_##suffix##_info);               \
3537     }                                                                \
3538     type_init(spapr_machine_register_##suffix)
3539 
3540 /*
3541  * pseries-2.10
3542  */
3543 static void spapr_machine_2_10_instance_options(MachineState *machine)
3544 {
3545 }
3546 
3547 static void spapr_machine_2_10_class_options(MachineClass *mc)
3548 {
3549     /* Defaults for the latest behaviour inherited from the base class */
3550 }
3551 
3552 DEFINE_SPAPR_MACHINE(2_10, "2.10", true);
3553 
3554 /*
3555  * pseries-2.9
3556  */
3557 #define SPAPR_COMPAT_2_9                                               \
3558     HW_COMPAT_2_9                                                      \
3559     {                                                                  \
3560         .driver = TYPE_POWERPC_CPU,                                    \
3561         .property = "pre-2.10-migration",                              \
3562         .value    = "on",                                              \
3563     },                                                                 \
3564 
3565 static void spapr_machine_2_9_instance_options(MachineState *machine)
3566 {
3567     spapr_machine_2_10_instance_options(machine);
3568 }
3569 
3570 static void spapr_machine_2_9_class_options(MachineClass *mc)
3571 {
3572     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3573 
3574     spapr_machine_2_10_class_options(mc);
3575     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3576     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
3577     smc->pre_2_10_has_unused_icps = true;
3578     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
3579 }
3580 
3581 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3582 
3583 /*
3584  * pseries-2.8
3585  */
3586 #define SPAPR_COMPAT_2_8                                        \
3587     HW_COMPAT_2_8                                               \
3588     {                                                           \
3589         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3590         .property = "pcie-extended-configuration-space",        \
3591         .value    = "off",                                      \
3592     },
3593 
3594 static void spapr_machine_2_8_instance_options(MachineState *machine)
3595 {
3596     spapr_machine_2_9_instance_options(machine);
3597 }
3598 
3599 static void spapr_machine_2_8_class_options(MachineClass *mc)
3600 {
3601     spapr_machine_2_9_class_options(mc);
3602     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3603     mc->numa_mem_align_shift = 23;
3604 }
3605 
3606 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3607 
3608 /*
3609  * pseries-2.7
3610  */
3611 #define SPAPR_COMPAT_2_7                            \
3612     HW_COMPAT_2_7                                   \
3613     {                                               \
3614         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3615         .property = "mem_win_size",                 \
3616         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3617     },                                              \
3618     {                                               \
3619         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3620         .property = "mem64_win_size",               \
3621         .value    = "0",                            \
3622     },                                              \
3623     {                                               \
3624         .driver = TYPE_POWERPC_CPU,                 \
3625         .property = "pre-2.8-migration",            \
3626         .value    = "on",                           \
3627     },                                              \
3628     {                                               \
3629         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3630         .property = "pre-2.8-migration",            \
3631         .value    = "on",                           \
3632     },
3633 
3634 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3635                               uint64_t *buid, hwaddr *pio,
3636                               hwaddr *mmio32, hwaddr *mmio64,
3637                               unsigned n_dma, uint32_t *liobns, Error **errp)
3638 {
3639     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3640     const uint64_t base_buid = 0x800000020000000ULL;
3641     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3642     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3643     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3644     const uint32_t max_index = 255;
3645     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3646 
3647     uint64_t ram_top = MACHINE(spapr)->ram_size;
3648     hwaddr phb0_base, phb_base;
3649     int i;
3650 
3651     /* Do we have hotpluggable memory? */
3652     if (MACHINE(spapr)->maxram_size > ram_top) {
3653         /* Can't just use maxram_size, because there may be an
3654          * alignment gap between normal and hotpluggable memory
3655          * regions */
3656         ram_top = spapr->hotplug_memory.base +
3657             memory_region_size(&spapr->hotplug_memory.mr);
3658     }
3659 
3660     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3661 
3662     if (index > max_index) {
3663         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3664                    max_index);
3665         return;
3666     }
3667 
3668     *buid = base_buid + index;
3669     for (i = 0; i < n_dma; ++i) {
3670         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3671     }
3672 
3673     phb_base = phb0_base + index * phb_spacing;
3674     *pio = phb_base + pio_offset;
3675     *mmio32 = phb_base + mmio_offset;
3676     /*
3677      * We don't set the 64-bit MMIO window, relying on the PHB's
3678      * fallback behaviour of automatically splitting a large "32-bit"
3679      * window into contiguous 32-bit and 64-bit windows
3680      */
3681 }
3682 
3683 static void spapr_machine_2_7_instance_options(MachineState *machine)
3684 {
3685     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3686 
3687     spapr_machine_2_8_instance_options(machine);
3688     spapr->use_hotplug_event_source = false;
3689 }
3690 
3691 static void spapr_machine_2_7_class_options(MachineClass *mc)
3692 {
3693     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3694 
3695     spapr_machine_2_8_class_options(mc);
3696     smc->tcg_default_cpu = "POWER7";
3697     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3698     smc->phb_placement = phb_placement_2_7;
3699 }
3700 
3701 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3702 
3703 /*
3704  * pseries-2.6
3705  */
3706 #define SPAPR_COMPAT_2_6 \
3707     HW_COMPAT_2_6 \
3708     { \
3709         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3710         .property = "ddw",\
3711         .value    = stringify(off),\
3712     },
3713 
3714 static void spapr_machine_2_6_instance_options(MachineState *machine)
3715 {
3716     spapr_machine_2_7_instance_options(machine);
3717 }
3718 
3719 static void spapr_machine_2_6_class_options(MachineClass *mc)
3720 {
3721     spapr_machine_2_7_class_options(mc);
3722     mc->has_hotpluggable_cpus = false;
3723     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3724 }
3725 
3726 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3727 
3728 /*
3729  * pseries-2.5
3730  */
3731 #define SPAPR_COMPAT_2_5 \
3732     HW_COMPAT_2_5 \
3733     { \
3734         .driver   = "spapr-vlan", \
3735         .property = "use-rx-buffer-pools", \
3736         .value    = "off", \
3737     },
3738 
3739 static void spapr_machine_2_5_instance_options(MachineState *machine)
3740 {
3741     spapr_machine_2_6_instance_options(machine);
3742 }
3743 
3744 static void spapr_machine_2_5_class_options(MachineClass *mc)
3745 {
3746     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3747 
3748     spapr_machine_2_6_class_options(mc);
3749     smc->use_ohci_by_default = true;
3750     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3751 }
3752 
3753 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3754 
3755 /*
3756  * pseries-2.4
3757  */
3758 #define SPAPR_COMPAT_2_4 \
3759         HW_COMPAT_2_4
3760 
3761 static void spapr_machine_2_4_instance_options(MachineState *machine)
3762 {
3763     spapr_machine_2_5_instance_options(machine);
3764 }
3765 
3766 static void spapr_machine_2_4_class_options(MachineClass *mc)
3767 {
3768     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3769 
3770     spapr_machine_2_5_class_options(mc);
3771     smc->dr_lmb_enabled = false;
3772     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3773 }
3774 
3775 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3776 
3777 /*
3778  * pseries-2.3
3779  */
3780 #define SPAPR_COMPAT_2_3 \
3781         HW_COMPAT_2_3 \
3782         {\
3783             .driver   = "spapr-pci-host-bridge",\
3784             .property = "dynamic-reconfiguration",\
3785             .value    = "off",\
3786         },
3787 
3788 static void spapr_machine_2_3_instance_options(MachineState *machine)
3789 {
3790     spapr_machine_2_4_instance_options(machine);
3791 }
3792 
3793 static void spapr_machine_2_3_class_options(MachineClass *mc)
3794 {
3795     spapr_machine_2_4_class_options(mc);
3796     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3797 }
3798 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3799 
3800 /*
3801  * pseries-2.2
3802  */
3803 
3804 #define SPAPR_COMPAT_2_2 \
3805         HW_COMPAT_2_2 \
3806         {\
3807             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3808             .property = "mem_win_size",\
3809             .value    = "0x20000000",\
3810         },
3811 
3812 static void spapr_machine_2_2_instance_options(MachineState *machine)
3813 {
3814     spapr_machine_2_3_instance_options(machine);
3815     machine->suppress_vmdesc = true;
3816 }
3817 
3818 static void spapr_machine_2_2_class_options(MachineClass *mc)
3819 {
3820     spapr_machine_2_3_class_options(mc);
3821     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3822 }
3823 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3824 
3825 /*
3826  * pseries-2.1
3827  */
3828 #define SPAPR_COMPAT_2_1 \
3829         HW_COMPAT_2_1
3830 
3831 static void spapr_machine_2_1_instance_options(MachineState *machine)
3832 {
3833     spapr_machine_2_2_instance_options(machine);
3834 }
3835 
3836 static void spapr_machine_2_1_class_options(MachineClass *mc)
3837 {
3838     spapr_machine_2_2_class_options(mc);
3839     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3840 }
3841 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3842 
3843 static void spapr_machine_register_types(void)
3844 {
3845     type_register_static(&spapr_machine_info);
3846 }
3847 
3848 type_init(spapr_machine_register_types)
3849