xref: /openbmc/qemu/hw/ppc/spapr.c (revision 922a01a013d2270682a188258cbccacfecf8129c)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/block-backend.h"
39 #include "sysemu/cpus.h"
40 #include "sysemu/hw_accel.h"
41 #include "kvm_ppc.h"
42 #include "migration/misc.h"
43 #include "migration/global_state.h"
44 #include "migration/register.h"
45 #include "mmu-hash64.h"
46 #include "mmu-book3s-v3.h"
47 #include "cpu-models.h"
48 #include "qom/cpu.h"
49 
50 #include "hw/boards.h"
51 #include "hw/ppc/ppc.h"
52 #include "hw/loader.h"
53 
54 #include "hw/ppc/fdt.h"
55 #include "hw/ppc/spapr.h"
56 #include "hw/ppc/spapr_vio.h"
57 #include "hw/pci-host/spapr.h"
58 #include "hw/ppc/xics.h"
59 #include "hw/pci/msi.h"
60 
61 #include "hw/pci/pci.h"
62 #include "hw/scsi/scsi.h"
63 #include "hw/virtio/virtio-scsi.h"
64 #include "hw/virtio/vhost-scsi-common.h"
65 
66 #include "exec/address-spaces.h"
67 #include "hw/usb.h"
68 #include "qemu/config-file.h"
69 #include "qemu/error-report.h"
70 #include "trace.h"
71 #include "hw/nmi.h"
72 #include "hw/intc/intc.h"
73 
74 #include "hw/compat.h"
75 #include "qemu/cutils.h"
76 #include "hw/ppc/spapr_cpu_core.h"
77 
78 #include <libfdt.h>
79 
80 /* SLOF memory layout:
81  *
82  * SLOF raw image loaded at 0, copies its romfs right below the flat
83  * device-tree, then position SLOF itself 31M below that
84  *
85  * So we set FW_OVERHEAD to 40MB which should account for all of that
86  * and more
87  *
88  * We load our kernel at 4M, leaving space for SLOF initial image
89  */
90 #define FDT_MAX_SIZE            0x100000
91 #define RTAS_MAX_SIZE           0x10000
92 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
93 #define FW_MAX_SIZE             0x400000
94 #define FW_FILE_NAME            "slof.bin"
95 #define FW_OVERHEAD             0x2800000
96 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
97 
98 #define MIN_RMA_SLOF            128UL
99 
100 #define PHANDLE_XICP            0x00001111
101 
102 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
103                                   const char *type_ics,
104                                   int nr_irqs, Error **errp)
105 {
106     Error *local_err = NULL;
107     Object *obj;
108 
109     obj = object_new(type_ics);
110     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
111     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
112                                    &error_abort);
113     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
114     if (local_err) {
115         goto error;
116     }
117     object_property_set_bool(obj, true, "realized", &local_err);
118     if (local_err) {
119         goto error;
120     }
121 
122     return ICS_SIMPLE(obj);
123 
124 error:
125     error_propagate(errp, local_err);
126     return NULL;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 static inline int xics_max_server_number(void)
164 {
165     return DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(), smp_threads);
166 }
167 
168 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
169 {
170     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
171     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
172 
173     if (kvm_enabled()) {
174         if (machine_kernel_irqchip_allowed(machine) &&
175             !xics_kvm_init(spapr, errp)) {
176             spapr->icp_type = TYPE_KVM_ICP;
177             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
178         }
179         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
180             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
181             return;
182         }
183     }
184 
185     if (!spapr->ics) {
186         xics_spapr_init(spapr);
187         spapr->icp_type = TYPE_ICP;
188         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
189         if (!spapr->ics) {
190             return;
191         }
192     }
193 
194     if (smc->pre_2_10_has_unused_icps) {
195         int i;
196 
197         for (i = 0; i < xics_max_server_number(); i++) {
198             /* Dummy entries get deregistered when real ICPState objects
199              * are registered during CPU core hotplug.
200              */
201             pre_2_10_vmstate_register_dummy_icp(i);
202         }
203     }
204 }
205 
206 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
207                                   int smt_threads)
208 {
209     int i, ret = 0;
210     uint32_t servers_prop[smt_threads];
211     uint32_t gservers_prop[smt_threads * 2];
212     int index = spapr_vcpu_id(cpu);
213 
214     if (cpu->compat_pvr) {
215         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
216         if (ret < 0) {
217             return ret;
218         }
219     }
220 
221     /* Build interrupt servers and gservers properties */
222     for (i = 0; i < smt_threads; i++) {
223         servers_prop[i] = cpu_to_be32(index + i);
224         /* Hack, direct the group queues back to cpu 0 */
225         gservers_prop[i*2] = cpu_to_be32(index + i);
226         gservers_prop[i*2 + 1] = 0;
227     }
228     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
229                       servers_prop, sizeof(servers_prop));
230     if (ret < 0) {
231         return ret;
232     }
233     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
234                       gservers_prop, sizeof(gservers_prop));
235 
236     return ret;
237 }
238 
239 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
240 {
241     int index = spapr_vcpu_id(cpu);
242     uint32_t associativity[] = {cpu_to_be32(0x5),
243                                 cpu_to_be32(0x0),
244                                 cpu_to_be32(0x0),
245                                 cpu_to_be32(0x0),
246                                 cpu_to_be32(cpu->node_id),
247                                 cpu_to_be32(index)};
248 
249     /* Advertise NUMA via ibm,associativity */
250     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
251                           sizeof(associativity));
252 }
253 
254 /* Populate the "ibm,pa-features" property */
255 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
256                                        PowerPCCPU *cpu,
257                                        void *fdt, int offset,
258                                        bool legacy_guest)
259 {
260     CPUPPCState *env = &cpu->env;
261     uint8_t pa_features_206[] = { 6, 0,
262         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
263     uint8_t pa_features_207[] = { 24, 0,
264         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
265         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
266         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
267         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
268     uint8_t pa_features_300[] = { 66, 0,
269         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
270         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
271         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
272         /* 6: DS207 */
273         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
274         /* 16: Vector */
275         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
276         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
277         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
278         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
279         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
280         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
281         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
282         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
283         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
284         /* 42: PM, 44: PC RA, 46: SC vec'd */
285         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
286         /* 48: SIMD, 50: QP BFP, 52: String */
287         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
288         /* 54: DecFP, 56: DecI, 58: SHA */
289         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
290         /* 60: NM atomic, 62: RNG */
291         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
292     };
293     uint8_t *pa_features = NULL;
294     size_t pa_size;
295 
296     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
297         pa_features = pa_features_206;
298         pa_size = sizeof(pa_features_206);
299     }
300     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
301         pa_features = pa_features_207;
302         pa_size = sizeof(pa_features_207);
303     }
304     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
305         pa_features = pa_features_300;
306         pa_size = sizeof(pa_features_300);
307     }
308     if (!pa_features) {
309         return;
310     }
311 
312     if (env->ci_large_pages) {
313         /*
314          * Note: we keep CI large pages off by default because a 64K capable
315          * guest provisioned with large pages might otherwise try to map a qemu
316          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
317          * even if that qemu runs on a 4k host.
318          * We dd this bit back here if we are confident this is not an issue
319          */
320         pa_features[3] |= 0x20;
321     }
322     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
323         pa_features[24] |= 0x80;    /* Transactional memory support */
324     }
325     if (legacy_guest && pa_size > 40) {
326         /* Workaround for broken kernels that attempt (guest) radix
327          * mode when they can't handle it, if they see the radix bit set
328          * in pa-features. So hide it from them. */
329         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
330     }
331 
332     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
333 }
334 
335 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
336 {
337     int ret = 0, offset, cpus_offset;
338     CPUState *cs;
339     char cpu_model[32];
340     int smt = kvmppc_smt_threads();
341     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
342 
343     CPU_FOREACH(cs) {
344         PowerPCCPU *cpu = POWERPC_CPU(cs);
345         DeviceClass *dc = DEVICE_GET_CLASS(cs);
346         int index = spapr_vcpu_id(cpu);
347         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
348 
349         if ((index % smt) != 0) {
350             continue;
351         }
352 
353         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
354 
355         cpus_offset = fdt_path_offset(fdt, "/cpus");
356         if (cpus_offset < 0) {
357             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
358             if (cpus_offset < 0) {
359                 return cpus_offset;
360             }
361         }
362         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
363         if (offset < 0) {
364             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
365             if (offset < 0) {
366                 return offset;
367             }
368         }
369 
370         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
371                           pft_size_prop, sizeof(pft_size_prop));
372         if (ret < 0) {
373             return ret;
374         }
375 
376         if (nb_numa_nodes > 1) {
377             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
378             if (ret < 0) {
379                 return ret;
380             }
381         }
382 
383         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
384         if (ret < 0) {
385             return ret;
386         }
387 
388         spapr_populate_pa_features(spapr, cpu, fdt, offset,
389                                    spapr->cas_legacy_guest_workaround);
390     }
391     return ret;
392 }
393 
394 static hwaddr spapr_node0_size(MachineState *machine)
395 {
396     if (nb_numa_nodes) {
397         int i;
398         for (i = 0; i < nb_numa_nodes; ++i) {
399             if (numa_info[i].node_mem) {
400                 return MIN(pow2floor(numa_info[i].node_mem),
401                            machine->ram_size);
402             }
403         }
404     }
405     return machine->ram_size;
406 }
407 
408 static void add_str(GString *s, const gchar *s1)
409 {
410     g_string_append_len(s, s1, strlen(s1) + 1);
411 }
412 
413 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
414                                        hwaddr size)
415 {
416     uint32_t associativity[] = {
417         cpu_to_be32(0x4), /* length */
418         cpu_to_be32(0x0), cpu_to_be32(0x0),
419         cpu_to_be32(0x0), cpu_to_be32(nodeid)
420     };
421     char mem_name[32];
422     uint64_t mem_reg_property[2];
423     int off;
424 
425     mem_reg_property[0] = cpu_to_be64(start);
426     mem_reg_property[1] = cpu_to_be64(size);
427 
428     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
429     off = fdt_add_subnode(fdt, 0, mem_name);
430     _FDT(off);
431     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
432     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
433                       sizeof(mem_reg_property))));
434     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
435                       sizeof(associativity))));
436     return off;
437 }
438 
439 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
440 {
441     MachineState *machine = MACHINE(spapr);
442     hwaddr mem_start, node_size;
443     int i, nb_nodes = nb_numa_nodes;
444     NodeInfo *nodes = numa_info;
445     NodeInfo ramnode;
446 
447     /* No NUMA nodes, assume there is just one node with whole RAM */
448     if (!nb_numa_nodes) {
449         nb_nodes = 1;
450         ramnode.node_mem = machine->ram_size;
451         nodes = &ramnode;
452     }
453 
454     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
455         if (!nodes[i].node_mem) {
456             continue;
457         }
458         if (mem_start >= machine->ram_size) {
459             node_size = 0;
460         } else {
461             node_size = nodes[i].node_mem;
462             if (node_size > machine->ram_size - mem_start) {
463                 node_size = machine->ram_size - mem_start;
464             }
465         }
466         if (!mem_start) {
467             /* ppc_spapr_init() checks for rma_size <= node0_size already */
468             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
469             mem_start += spapr->rma_size;
470             node_size -= spapr->rma_size;
471         }
472         for ( ; node_size; ) {
473             hwaddr sizetmp = pow2floor(node_size);
474 
475             /* mem_start != 0 here */
476             if (ctzl(mem_start) < ctzl(sizetmp)) {
477                 sizetmp = 1ULL << ctzl(mem_start);
478             }
479 
480             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
481             node_size -= sizetmp;
482             mem_start += sizetmp;
483         }
484     }
485 
486     return 0;
487 }
488 
489 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
490                                   sPAPRMachineState *spapr)
491 {
492     PowerPCCPU *cpu = POWERPC_CPU(cs);
493     CPUPPCState *env = &cpu->env;
494     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
495     int index = spapr_vcpu_id(cpu);
496     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
497                        0xffffffff, 0xffffffff};
498     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
499         : SPAPR_TIMEBASE_FREQ;
500     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
501     uint32_t page_sizes_prop[64];
502     size_t page_sizes_prop_size;
503     uint32_t vcpus_per_socket = smp_threads * smp_cores;
504     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
505     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
506     sPAPRDRConnector *drc;
507     int drc_index;
508     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
509     int i;
510 
511     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
512     if (drc) {
513         drc_index = spapr_drc_index(drc);
514         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
515     }
516 
517     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
518     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
519 
520     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
521     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
522                            env->dcache_line_size)));
523     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
524                            env->dcache_line_size)));
525     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
526                            env->icache_line_size)));
527     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
528                            env->icache_line_size)));
529 
530     if (pcc->l1_dcache_size) {
531         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
532                                pcc->l1_dcache_size)));
533     } else {
534         warn_report("Unknown L1 dcache size for cpu");
535     }
536     if (pcc->l1_icache_size) {
537         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
538                                pcc->l1_icache_size)));
539     } else {
540         warn_report("Unknown L1 icache size for cpu");
541     }
542 
543     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
544     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
545     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
546     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
547     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
548     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
549 
550     if (env->spr_cb[SPR_PURR].oea_read) {
551         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
552     }
553 
554     if (env->mmu_model & POWERPC_MMU_1TSEG) {
555         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
556                           segs, sizeof(segs))));
557     }
558 
559     /* Advertise VSX (vector extensions) if available
560      *   1               == VMX / Altivec available
561      *   2               == VSX available
562      *
563      * Only CPUs for which we create core types in spapr_cpu_core.c
564      * are possible, and all of those have VMX */
565     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
566         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
567     } else {
568         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
569     }
570 
571     /* Advertise DFP (Decimal Floating Point) if available
572      *   0 / no property == no DFP
573      *   1               == DFP available */
574     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
575         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
576     }
577 
578     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
579                                                   sizeof(page_sizes_prop));
580     if (page_sizes_prop_size) {
581         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
582                           page_sizes_prop, page_sizes_prop_size)));
583     }
584 
585     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
586 
587     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
588                            cs->cpu_index / vcpus_per_socket)));
589 
590     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
591                       pft_size_prop, sizeof(pft_size_prop))));
592 
593     if (nb_numa_nodes > 1) {
594         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
595     }
596 
597     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
598 
599     if (pcc->radix_page_info) {
600         for (i = 0; i < pcc->radix_page_info->count; i++) {
601             radix_AP_encodings[i] =
602                 cpu_to_be32(pcc->radix_page_info->entries[i]);
603         }
604         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
605                           radix_AP_encodings,
606                           pcc->radix_page_info->count *
607                           sizeof(radix_AP_encodings[0]))));
608     }
609 }
610 
611 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
612 {
613     CPUState *cs;
614     int cpus_offset;
615     char *nodename;
616     int smt = kvmppc_smt_threads();
617 
618     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
619     _FDT(cpus_offset);
620     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
621     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
622 
623     /*
624      * We walk the CPUs in reverse order to ensure that CPU DT nodes
625      * created by fdt_add_subnode() end up in the right order in FDT
626      * for the guest kernel the enumerate the CPUs correctly.
627      */
628     CPU_FOREACH_REVERSE(cs) {
629         PowerPCCPU *cpu = POWERPC_CPU(cs);
630         int index = spapr_vcpu_id(cpu);
631         DeviceClass *dc = DEVICE_GET_CLASS(cs);
632         int offset;
633 
634         if ((index % smt) != 0) {
635             continue;
636         }
637 
638         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
639         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
640         g_free(nodename);
641         _FDT(offset);
642         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
643     }
644 
645 }
646 
647 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
648 {
649     MemoryDeviceInfoList *info;
650 
651     for (info = list; info; info = info->next) {
652         MemoryDeviceInfo *value = info->value;
653 
654         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
655             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
656 
657             if (pcdimm_info->addr >= addr &&
658                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
659                 return pcdimm_info->node;
660             }
661         }
662     }
663 
664     return -1;
665 }
666 
667 /*
668  * Adds ibm,dynamic-reconfiguration-memory node.
669  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
670  * of this device tree node.
671  */
672 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
673 {
674     MachineState *machine = MACHINE(spapr);
675     int ret, i, offset;
676     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
677     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
678     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
679     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
680                        memory_region_size(&spapr->hotplug_memory.mr)) /
681                        lmb_size;
682     uint32_t *int_buf, *cur_index, buf_len;
683     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
684     MemoryDeviceInfoList *dimms = NULL;
685 
686     /*
687      * Don't create the node if there is no hotpluggable memory
688      */
689     if (machine->ram_size == machine->maxram_size) {
690         return 0;
691     }
692 
693     /*
694      * Allocate enough buffer size to fit in ibm,dynamic-memory
695      * or ibm,associativity-lookup-arrays
696      */
697     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
698               * sizeof(uint32_t);
699     cur_index = int_buf = g_malloc0(buf_len);
700 
701     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
702 
703     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
704                     sizeof(prop_lmb_size));
705     if (ret < 0) {
706         goto out;
707     }
708 
709     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
710     if (ret < 0) {
711         goto out;
712     }
713 
714     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
715     if (ret < 0) {
716         goto out;
717     }
718 
719     if (hotplug_lmb_start) {
720         MemoryDeviceInfoList **prev = &dimms;
721         qmp_pc_dimm_device_list(qdev_get_machine(), &prev);
722     }
723 
724     /* ibm,dynamic-memory */
725     int_buf[0] = cpu_to_be32(nr_lmbs);
726     cur_index++;
727     for (i = 0; i < nr_lmbs; i++) {
728         uint64_t addr = i * lmb_size;
729         uint32_t *dynamic_memory = cur_index;
730 
731         if (i >= hotplug_lmb_start) {
732             sPAPRDRConnector *drc;
733 
734             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
735             g_assert(drc);
736 
737             dynamic_memory[0] = cpu_to_be32(addr >> 32);
738             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
739             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
740             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
741             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
742             if (memory_region_present(get_system_memory(), addr)) {
743                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
744             } else {
745                 dynamic_memory[5] = cpu_to_be32(0);
746             }
747         } else {
748             /*
749              * LMB information for RMA, boot time RAM and gap b/n RAM and
750              * hotplug memory region -- all these are marked as reserved
751              * and as having no valid DRC.
752              */
753             dynamic_memory[0] = cpu_to_be32(addr >> 32);
754             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
755             dynamic_memory[2] = cpu_to_be32(0);
756             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
757             dynamic_memory[4] = cpu_to_be32(-1);
758             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
759                                             SPAPR_LMB_FLAGS_DRC_INVALID);
760         }
761 
762         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
763     }
764     qapi_free_MemoryDeviceInfoList(dimms);
765     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
766     if (ret < 0) {
767         goto out;
768     }
769 
770     /* ibm,associativity-lookup-arrays */
771     cur_index = int_buf;
772     int_buf[0] = cpu_to_be32(nr_nodes);
773     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
774     cur_index += 2;
775     for (i = 0; i < nr_nodes; i++) {
776         uint32_t associativity[] = {
777             cpu_to_be32(0x0),
778             cpu_to_be32(0x0),
779             cpu_to_be32(0x0),
780             cpu_to_be32(i)
781         };
782         memcpy(cur_index, associativity, sizeof(associativity));
783         cur_index += 4;
784     }
785     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
786             (cur_index - int_buf) * sizeof(uint32_t));
787 out:
788     g_free(int_buf);
789     return ret;
790 }
791 
792 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
793                                 sPAPROptionVector *ov5_updates)
794 {
795     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
796     int ret = 0, offset;
797 
798     /* Generate ibm,dynamic-reconfiguration-memory node if required */
799     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
800         g_assert(smc->dr_lmb_enabled);
801         ret = spapr_populate_drconf_memory(spapr, fdt);
802         if (ret) {
803             goto out;
804         }
805     }
806 
807     offset = fdt_path_offset(fdt, "/chosen");
808     if (offset < 0) {
809         offset = fdt_add_subnode(fdt, 0, "chosen");
810         if (offset < 0) {
811             return offset;
812         }
813     }
814     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
815                                  "ibm,architecture-vec-5");
816 
817 out:
818     return ret;
819 }
820 
821 static bool spapr_hotplugged_dev_before_cas(void)
822 {
823     Object *drc_container, *obj;
824     ObjectProperty *prop;
825     ObjectPropertyIterator iter;
826 
827     drc_container = container_get(object_get_root(), "/dr-connector");
828     object_property_iter_init(&iter, drc_container);
829     while ((prop = object_property_iter_next(&iter))) {
830         if (!strstart(prop->type, "link<", NULL)) {
831             continue;
832         }
833         obj = object_property_get_link(drc_container, prop->name, NULL);
834         if (spapr_drc_needed(obj)) {
835             return true;
836         }
837     }
838     return false;
839 }
840 
841 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
842                                  target_ulong addr, target_ulong size,
843                                  sPAPROptionVector *ov5_updates)
844 {
845     void *fdt, *fdt_skel;
846     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
847 
848     if (spapr_hotplugged_dev_before_cas()) {
849         return 1;
850     }
851 
852     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
853         error_report("SLOF provided an unexpected CAS buffer size "
854                      TARGET_FMT_lu " (min: %zu, max: %u)",
855                      size, sizeof(hdr), FW_MAX_SIZE);
856         exit(EXIT_FAILURE);
857     }
858 
859     size -= sizeof(hdr);
860 
861     /* Create skeleton */
862     fdt_skel = g_malloc0(size);
863     _FDT((fdt_create(fdt_skel, size)));
864     _FDT((fdt_begin_node(fdt_skel, "")));
865     _FDT((fdt_end_node(fdt_skel)));
866     _FDT((fdt_finish(fdt_skel)));
867     fdt = g_malloc0(size);
868     _FDT((fdt_open_into(fdt_skel, fdt, size)));
869     g_free(fdt_skel);
870 
871     /* Fixup cpu nodes */
872     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
873 
874     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
875         return -1;
876     }
877 
878     /* Pack resulting tree */
879     _FDT((fdt_pack(fdt)));
880 
881     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
882         trace_spapr_cas_failed(size);
883         return -1;
884     }
885 
886     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
887     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
888     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
889     g_free(fdt);
890 
891     return 0;
892 }
893 
894 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
895 {
896     int rtas;
897     GString *hypertas = g_string_sized_new(256);
898     GString *qemu_hypertas = g_string_sized_new(256);
899     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
900     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
901         memory_region_size(&spapr->hotplug_memory.mr);
902     uint32_t lrdr_capacity[] = {
903         cpu_to_be32(max_hotplug_addr >> 32),
904         cpu_to_be32(max_hotplug_addr & 0xffffffff),
905         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
906         cpu_to_be32(max_cpus / smp_threads),
907     };
908 
909     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
910 
911     /* hypertas */
912     add_str(hypertas, "hcall-pft");
913     add_str(hypertas, "hcall-term");
914     add_str(hypertas, "hcall-dabr");
915     add_str(hypertas, "hcall-interrupt");
916     add_str(hypertas, "hcall-tce");
917     add_str(hypertas, "hcall-vio");
918     add_str(hypertas, "hcall-splpar");
919     add_str(hypertas, "hcall-bulk");
920     add_str(hypertas, "hcall-set-mode");
921     add_str(hypertas, "hcall-sprg0");
922     add_str(hypertas, "hcall-copy");
923     add_str(hypertas, "hcall-debug");
924     add_str(qemu_hypertas, "hcall-memop1");
925 
926     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
927         add_str(hypertas, "hcall-multi-tce");
928     }
929 
930     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
931         add_str(hypertas, "hcall-hpt-resize");
932     }
933 
934     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
935                      hypertas->str, hypertas->len));
936     g_string_free(hypertas, TRUE);
937     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
938                      qemu_hypertas->str, qemu_hypertas->len));
939     g_string_free(qemu_hypertas, TRUE);
940 
941     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
942                      refpoints, sizeof(refpoints)));
943 
944     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
945                           RTAS_ERROR_LOG_MAX));
946     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
947                           RTAS_EVENT_SCAN_RATE));
948 
949     g_assert(msi_nonbroken);
950     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
951 
952     /*
953      * According to PAPR, rtas ibm,os-term does not guarantee a return
954      * back to the guest cpu.
955      *
956      * While an additional ibm,extended-os-term property indicates
957      * that rtas call return will always occur. Set this property.
958      */
959     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
960 
961     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
962                      lrdr_capacity, sizeof(lrdr_capacity)));
963 
964     spapr_dt_rtas_tokens(fdt, rtas);
965 }
966 
967 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
968  * that the guest may request and thus the valid values for bytes 24..26 of
969  * option vector 5: */
970 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
971 {
972     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
973 
974     char val[2 * 4] = {
975         23, 0x00, /* Xive mode, filled in below. */
976         24, 0x00, /* Hash/Radix, filled in below. */
977         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
978         26, 0x40, /* Radix options: GTSE == yes. */
979     };
980 
981     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
982                           first_ppc_cpu->compat_pvr)) {
983         /* If we're in a pre POWER9 compat mode then the guest should do hash */
984         val[3] = 0x00; /* Hash */
985     } else if (kvm_enabled()) {
986         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
987             val[3] = 0x80; /* OV5_MMU_BOTH */
988         } else if (kvmppc_has_cap_mmu_radix()) {
989             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
990         } else {
991             val[3] = 0x00; /* Hash */
992         }
993     } else {
994         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
995         val[3] = 0xC0;
996     }
997     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
998                      val, sizeof(val)));
999 }
1000 
1001 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1002 {
1003     MachineState *machine = MACHINE(spapr);
1004     int chosen;
1005     const char *boot_device = machine->boot_order;
1006     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1007     size_t cb = 0;
1008     char *bootlist = get_boot_devices_list(&cb, true);
1009 
1010     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1011 
1012     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1013     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1014                           spapr->initrd_base));
1015     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1016                           spapr->initrd_base + spapr->initrd_size));
1017 
1018     if (spapr->kernel_size) {
1019         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1020                               cpu_to_be64(spapr->kernel_size) };
1021 
1022         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1023                          &kprop, sizeof(kprop)));
1024         if (spapr->kernel_le) {
1025             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1026         }
1027     }
1028     if (boot_menu) {
1029         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1030     }
1031     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1032     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1033     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1034 
1035     if (cb && bootlist) {
1036         int i;
1037 
1038         for (i = 0; i < cb; i++) {
1039             if (bootlist[i] == '\n') {
1040                 bootlist[i] = ' ';
1041             }
1042         }
1043         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1044     }
1045 
1046     if (boot_device && strlen(boot_device)) {
1047         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1048     }
1049 
1050     if (!spapr->has_graphics && stdout_path) {
1051         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1052     }
1053 
1054     spapr_dt_ov5_platform_support(fdt, chosen);
1055 
1056     g_free(stdout_path);
1057     g_free(bootlist);
1058 }
1059 
1060 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1061 {
1062     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1063      * KVM to work under pHyp with some guest co-operation */
1064     int hypervisor;
1065     uint8_t hypercall[16];
1066 
1067     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1068     /* indicate KVM hypercall interface */
1069     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1070     if (kvmppc_has_cap_fixup_hcalls()) {
1071         /*
1072          * Older KVM versions with older guest kernels were broken
1073          * with the magic page, don't allow the guest to map it.
1074          */
1075         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1076                                   sizeof(hypercall))) {
1077             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1078                              hypercall, sizeof(hypercall)));
1079         }
1080     }
1081 }
1082 
1083 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1084                              hwaddr rtas_addr,
1085                              hwaddr rtas_size)
1086 {
1087     MachineState *machine = MACHINE(spapr);
1088     MachineClass *mc = MACHINE_GET_CLASS(machine);
1089     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1090     int ret;
1091     void *fdt;
1092     sPAPRPHBState *phb;
1093     char *buf;
1094 
1095     fdt = g_malloc0(FDT_MAX_SIZE);
1096     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1097 
1098     /* Root node */
1099     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1100     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1101     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1102 
1103     /*
1104      * Add info to guest to indentify which host is it being run on
1105      * and what is the uuid of the guest
1106      */
1107     if (kvmppc_get_host_model(&buf)) {
1108         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1109         g_free(buf);
1110     }
1111     if (kvmppc_get_host_serial(&buf)) {
1112         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1113         g_free(buf);
1114     }
1115 
1116     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1117 
1118     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1119     if (qemu_uuid_set) {
1120         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1121     }
1122     g_free(buf);
1123 
1124     if (qemu_get_vm_name()) {
1125         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1126                                 qemu_get_vm_name()));
1127     }
1128 
1129     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1130     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1131 
1132     /* /interrupt controller */
1133     spapr_dt_xics(xics_max_server_number(), fdt, PHANDLE_XICP);
1134 
1135     ret = spapr_populate_memory(spapr, fdt);
1136     if (ret < 0) {
1137         error_report("couldn't setup memory nodes in fdt");
1138         exit(1);
1139     }
1140 
1141     /* /vdevice */
1142     spapr_dt_vdevice(spapr->vio_bus, fdt);
1143 
1144     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1145         ret = spapr_rng_populate_dt(fdt);
1146         if (ret < 0) {
1147             error_report("could not set up rng device in the fdt");
1148             exit(1);
1149         }
1150     }
1151 
1152     QLIST_FOREACH(phb, &spapr->phbs, list) {
1153         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1154         if (ret < 0) {
1155             error_report("couldn't setup PCI devices in fdt");
1156             exit(1);
1157         }
1158     }
1159 
1160     /* cpus */
1161     spapr_populate_cpus_dt_node(fdt, spapr);
1162 
1163     if (smc->dr_lmb_enabled) {
1164         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1165     }
1166 
1167     if (mc->has_hotpluggable_cpus) {
1168         int offset = fdt_path_offset(fdt, "/cpus");
1169         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1170                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1171         if (ret < 0) {
1172             error_report("Couldn't set up CPU DR device tree properties");
1173             exit(1);
1174         }
1175     }
1176 
1177     /* /event-sources */
1178     spapr_dt_events(spapr, fdt);
1179 
1180     /* /rtas */
1181     spapr_dt_rtas(spapr, fdt);
1182 
1183     /* /chosen */
1184     spapr_dt_chosen(spapr, fdt);
1185 
1186     /* /hypervisor */
1187     if (kvm_enabled()) {
1188         spapr_dt_hypervisor(spapr, fdt);
1189     }
1190 
1191     /* Build memory reserve map */
1192     if (spapr->kernel_size) {
1193         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1194     }
1195     if (spapr->initrd_size) {
1196         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1197     }
1198 
1199     /* ibm,client-architecture-support updates */
1200     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1201     if (ret < 0) {
1202         error_report("couldn't setup CAS properties fdt");
1203         exit(1);
1204     }
1205 
1206     return fdt;
1207 }
1208 
1209 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1210 {
1211     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1212 }
1213 
1214 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1215                                     PowerPCCPU *cpu)
1216 {
1217     CPUPPCState *env = &cpu->env;
1218 
1219     /* The TCG path should also be holding the BQL at this point */
1220     g_assert(qemu_mutex_iothread_locked());
1221 
1222     if (msr_pr) {
1223         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1224         env->gpr[3] = H_PRIVILEGE;
1225     } else {
1226         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1227     }
1228 }
1229 
1230 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1231 {
1232     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1233 
1234     return spapr->patb_entry;
1235 }
1236 
1237 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1238 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1239 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1240 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1241 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1242 
1243 /*
1244  * Get the fd to access the kernel htab, re-opening it if necessary
1245  */
1246 static int get_htab_fd(sPAPRMachineState *spapr)
1247 {
1248     Error *local_err = NULL;
1249 
1250     if (spapr->htab_fd >= 0) {
1251         return spapr->htab_fd;
1252     }
1253 
1254     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1255     if (spapr->htab_fd < 0) {
1256         error_report_err(local_err);
1257     }
1258 
1259     return spapr->htab_fd;
1260 }
1261 
1262 void close_htab_fd(sPAPRMachineState *spapr)
1263 {
1264     if (spapr->htab_fd >= 0) {
1265         close(spapr->htab_fd);
1266     }
1267     spapr->htab_fd = -1;
1268 }
1269 
1270 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1271 {
1272     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1273 
1274     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1275 }
1276 
1277 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1278 {
1279     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1280 
1281     assert(kvm_enabled());
1282 
1283     if (!spapr->htab) {
1284         return 0;
1285     }
1286 
1287     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1288 }
1289 
1290 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1291                                                 hwaddr ptex, int n)
1292 {
1293     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1294     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1295 
1296     if (!spapr->htab) {
1297         /*
1298          * HTAB is controlled by KVM. Fetch into temporary buffer
1299          */
1300         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1301         kvmppc_read_hptes(hptes, ptex, n);
1302         return hptes;
1303     }
1304 
1305     /*
1306      * HTAB is controlled by QEMU. Just point to the internally
1307      * accessible PTEG.
1308      */
1309     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1310 }
1311 
1312 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1313                               const ppc_hash_pte64_t *hptes,
1314                               hwaddr ptex, int n)
1315 {
1316     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1317 
1318     if (!spapr->htab) {
1319         g_free((void *)hptes);
1320     }
1321 
1322     /* Nothing to do for qemu managed HPT */
1323 }
1324 
1325 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1326                              uint64_t pte0, uint64_t pte1)
1327 {
1328     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1329     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1330 
1331     if (!spapr->htab) {
1332         kvmppc_write_hpte(ptex, pte0, pte1);
1333     } else {
1334         stq_p(spapr->htab + offset, pte0);
1335         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1336     }
1337 }
1338 
1339 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1340 {
1341     int shift;
1342 
1343     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1344      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1345      * that's much more than is needed for Linux guests */
1346     shift = ctz64(pow2ceil(ramsize)) - 7;
1347     shift = MAX(shift, 18); /* Minimum architected size */
1348     shift = MIN(shift, 46); /* Maximum architected size */
1349     return shift;
1350 }
1351 
1352 void spapr_free_hpt(sPAPRMachineState *spapr)
1353 {
1354     g_free(spapr->htab);
1355     spapr->htab = NULL;
1356     spapr->htab_shift = 0;
1357     close_htab_fd(spapr);
1358 }
1359 
1360 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1361                           Error **errp)
1362 {
1363     long rc;
1364 
1365     /* Clean up any HPT info from a previous boot */
1366     spapr_free_hpt(spapr);
1367 
1368     rc = kvmppc_reset_htab(shift);
1369     if (rc < 0) {
1370         /* kernel-side HPT needed, but couldn't allocate one */
1371         error_setg_errno(errp, errno,
1372                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1373                          shift);
1374         /* This is almost certainly fatal, but if the caller really
1375          * wants to carry on with shift == 0, it's welcome to try */
1376     } else if (rc > 0) {
1377         /* kernel-side HPT allocated */
1378         if (rc != shift) {
1379             error_setg(errp,
1380                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1381                        shift, rc);
1382         }
1383 
1384         spapr->htab_shift = shift;
1385         spapr->htab = NULL;
1386     } else {
1387         /* kernel-side HPT not needed, allocate in userspace instead */
1388         size_t size = 1ULL << shift;
1389         int i;
1390 
1391         spapr->htab = qemu_memalign(size, size);
1392         if (!spapr->htab) {
1393             error_setg_errno(errp, errno,
1394                              "Could not allocate HPT of order %d", shift);
1395             return;
1396         }
1397 
1398         memset(spapr->htab, 0, size);
1399         spapr->htab_shift = shift;
1400 
1401         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1402             DIRTY_HPTE(HPTE(spapr->htab, i));
1403         }
1404     }
1405     /* We're setting up a hash table, so that means we're not radix */
1406     spapr->patb_entry = 0;
1407 }
1408 
1409 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1410 {
1411     int hpt_shift;
1412 
1413     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1414         || (spapr->cas_reboot
1415             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1416         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1417     } else {
1418         uint64_t current_ram_size;
1419 
1420         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1421         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1422     }
1423     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1424 
1425     if (spapr->vrma_adjust) {
1426         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1427                                           spapr->htab_shift);
1428     }
1429 }
1430 
1431 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1432 {
1433     bool matched = false;
1434 
1435     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1436         matched = true;
1437     }
1438 
1439     if (!matched) {
1440         error_report("Device %s is not supported by this machine yet.",
1441                      qdev_fw_name(DEVICE(sbdev)));
1442         exit(1);
1443     }
1444 }
1445 
1446 static int spapr_reset_drcs(Object *child, void *opaque)
1447 {
1448     sPAPRDRConnector *drc =
1449         (sPAPRDRConnector *) object_dynamic_cast(child,
1450                                                  TYPE_SPAPR_DR_CONNECTOR);
1451 
1452     if (drc) {
1453         spapr_drc_reset(drc);
1454     }
1455 
1456     return 0;
1457 }
1458 
1459 static void spapr_machine_reset(void)
1460 {
1461     MachineState *machine = MACHINE(qdev_get_machine());
1462     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1463     PowerPCCPU *first_ppc_cpu;
1464     uint32_t rtas_limit;
1465     hwaddr rtas_addr, fdt_addr;
1466     void *fdt;
1467     int rc;
1468 
1469     /* Check for unknown sysbus devices */
1470     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1471 
1472     spapr_caps_reset(spapr);
1473 
1474     first_ppc_cpu = POWERPC_CPU(first_cpu);
1475     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1476         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1477                          spapr->max_compat_pvr)) {
1478         /* If using KVM with radix mode available, VCPUs can be started
1479          * without a HPT because KVM will start them in radix mode.
1480          * Set the GR bit in PATB so that we know there is no HPT. */
1481         spapr->patb_entry = PATBE1_GR;
1482     } else {
1483         spapr_setup_hpt_and_vrma(spapr);
1484     }
1485 
1486     /* if this reset wasn't generated by CAS, we should reset our
1487      * negotiated options and start from scratch */
1488     if (!spapr->cas_reboot) {
1489         spapr_ovec_cleanup(spapr->ov5_cas);
1490         spapr->ov5_cas = spapr_ovec_new();
1491 
1492         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1493     }
1494 
1495     qemu_devices_reset();
1496 
1497     /* DRC reset may cause a device to be unplugged. This will cause troubles
1498      * if this device is used by another device (eg, a running vhost backend
1499      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1500      * situations, we reset DRCs after all devices have been reset.
1501      */
1502     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1503 
1504     spapr_clear_pending_events(spapr);
1505 
1506     /*
1507      * We place the device tree and RTAS just below either the top of the RMA,
1508      * or just below 2GB, whichever is lowere, so that it can be
1509      * processed with 32-bit real mode code if necessary
1510      */
1511     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1512     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1513     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1514 
1515     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1516 
1517     spapr_load_rtas(spapr, fdt, rtas_addr);
1518 
1519     rc = fdt_pack(fdt);
1520 
1521     /* Should only fail if we've built a corrupted tree */
1522     assert(rc == 0);
1523 
1524     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1525         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1526                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1527         exit(1);
1528     }
1529 
1530     /* Load the fdt */
1531     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1532     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1533     g_free(fdt);
1534 
1535     /* Set up the entry state */
1536     first_ppc_cpu->env.gpr[3] = fdt_addr;
1537     first_ppc_cpu->env.gpr[5] = 0;
1538     first_cpu->halted = 0;
1539     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1540 
1541     spapr->cas_reboot = false;
1542 }
1543 
1544 static void spapr_create_nvram(sPAPRMachineState *spapr)
1545 {
1546     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1547     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1548 
1549     if (dinfo) {
1550         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1551                             &error_fatal);
1552     }
1553 
1554     qdev_init_nofail(dev);
1555 
1556     spapr->nvram = (struct sPAPRNVRAM *)dev;
1557 }
1558 
1559 static void spapr_rtc_create(sPAPRMachineState *spapr)
1560 {
1561     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1562     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1563                               &error_fatal);
1564     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1565                               &error_fatal);
1566     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1567                               "date", &error_fatal);
1568 }
1569 
1570 /* Returns whether we want to use VGA or not */
1571 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1572 {
1573     switch (vga_interface_type) {
1574     case VGA_NONE:
1575         return false;
1576     case VGA_DEVICE:
1577         return true;
1578     case VGA_STD:
1579     case VGA_VIRTIO:
1580         return pci_vga_init(pci_bus) != NULL;
1581     default:
1582         error_setg(errp,
1583                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1584         return false;
1585     }
1586 }
1587 
1588 static int spapr_pre_load(void *opaque)
1589 {
1590     int rc;
1591 
1592     rc = spapr_caps_pre_load(opaque);
1593     if (rc) {
1594         return rc;
1595     }
1596 
1597     return 0;
1598 }
1599 
1600 static int spapr_post_load(void *opaque, int version_id)
1601 {
1602     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1603     int err = 0;
1604 
1605     err = spapr_caps_post_migration(spapr);
1606     if (err) {
1607         return err;
1608     }
1609 
1610     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1611         CPUState *cs;
1612         CPU_FOREACH(cs) {
1613             PowerPCCPU *cpu = POWERPC_CPU(cs);
1614             icp_resend(ICP(cpu->intc));
1615         }
1616     }
1617 
1618     /* In earlier versions, there was no separate qdev for the PAPR
1619      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1620      * So when migrating from those versions, poke the incoming offset
1621      * value into the RTC device */
1622     if (version_id < 3) {
1623         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1624     }
1625 
1626     if (kvm_enabled() && spapr->patb_entry) {
1627         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1628         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1629         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1630 
1631         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1632         if (err) {
1633             error_report("Process table config unsupported by the host");
1634             return -EINVAL;
1635         }
1636     }
1637 
1638     return err;
1639 }
1640 
1641 static int spapr_pre_save(void *opaque)
1642 {
1643     int rc;
1644 
1645     rc = spapr_caps_pre_save(opaque);
1646     if (rc) {
1647         return rc;
1648     }
1649 
1650     return 0;
1651 }
1652 
1653 static bool version_before_3(void *opaque, int version_id)
1654 {
1655     return version_id < 3;
1656 }
1657 
1658 static bool spapr_pending_events_needed(void *opaque)
1659 {
1660     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1661     return !QTAILQ_EMPTY(&spapr->pending_events);
1662 }
1663 
1664 static const VMStateDescription vmstate_spapr_event_entry = {
1665     .name = "spapr_event_log_entry",
1666     .version_id = 1,
1667     .minimum_version_id = 1,
1668     .fields = (VMStateField[]) {
1669         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1670         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1671         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1672                                      NULL, extended_length),
1673         VMSTATE_END_OF_LIST()
1674     },
1675 };
1676 
1677 static const VMStateDescription vmstate_spapr_pending_events = {
1678     .name = "spapr_pending_events",
1679     .version_id = 1,
1680     .minimum_version_id = 1,
1681     .needed = spapr_pending_events_needed,
1682     .fields = (VMStateField[]) {
1683         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1684                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1685         VMSTATE_END_OF_LIST()
1686     },
1687 };
1688 
1689 static bool spapr_ov5_cas_needed(void *opaque)
1690 {
1691     sPAPRMachineState *spapr = opaque;
1692     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1693     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1694     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1695     bool cas_needed;
1696 
1697     /* Prior to the introduction of sPAPROptionVector, we had two option
1698      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1699      * Both of these options encode machine topology into the device-tree
1700      * in such a way that the now-booted OS should still be able to interact
1701      * appropriately with QEMU regardless of what options were actually
1702      * negotiatied on the source side.
1703      *
1704      * As such, we can avoid migrating the CAS-negotiated options if these
1705      * are the only options available on the current machine/platform.
1706      * Since these are the only options available for pseries-2.7 and
1707      * earlier, this allows us to maintain old->new/new->old migration
1708      * compatibility.
1709      *
1710      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1711      * via default pseries-2.8 machines and explicit command-line parameters.
1712      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1713      * of the actual CAS-negotiated values to continue working properly. For
1714      * example, availability of memory unplug depends on knowing whether
1715      * OV5_HP_EVT was negotiated via CAS.
1716      *
1717      * Thus, for any cases where the set of available CAS-negotiatable
1718      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1719      * include the CAS-negotiated options in the migration stream.
1720      */
1721     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1722     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1723 
1724     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1725      * the mask itself since in the future it's possible "legacy" bits may be
1726      * removed via machine options, which could generate a false positive
1727      * that breaks migration.
1728      */
1729     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1730     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1731 
1732     spapr_ovec_cleanup(ov5_mask);
1733     spapr_ovec_cleanup(ov5_legacy);
1734     spapr_ovec_cleanup(ov5_removed);
1735 
1736     return cas_needed;
1737 }
1738 
1739 static const VMStateDescription vmstate_spapr_ov5_cas = {
1740     .name = "spapr_option_vector_ov5_cas",
1741     .version_id = 1,
1742     .minimum_version_id = 1,
1743     .needed = spapr_ov5_cas_needed,
1744     .fields = (VMStateField[]) {
1745         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1746                                  vmstate_spapr_ovec, sPAPROptionVector),
1747         VMSTATE_END_OF_LIST()
1748     },
1749 };
1750 
1751 static bool spapr_patb_entry_needed(void *opaque)
1752 {
1753     sPAPRMachineState *spapr = opaque;
1754 
1755     return !!spapr->patb_entry;
1756 }
1757 
1758 static const VMStateDescription vmstate_spapr_patb_entry = {
1759     .name = "spapr_patb_entry",
1760     .version_id = 1,
1761     .minimum_version_id = 1,
1762     .needed = spapr_patb_entry_needed,
1763     .fields = (VMStateField[]) {
1764         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1765         VMSTATE_END_OF_LIST()
1766     },
1767 };
1768 
1769 static const VMStateDescription vmstate_spapr = {
1770     .name = "spapr",
1771     .version_id = 3,
1772     .minimum_version_id = 1,
1773     .pre_load = spapr_pre_load,
1774     .post_load = spapr_post_load,
1775     .pre_save = spapr_pre_save,
1776     .fields = (VMStateField[]) {
1777         /* used to be @next_irq */
1778         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1779 
1780         /* RTC offset */
1781         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1782 
1783         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1784         VMSTATE_END_OF_LIST()
1785     },
1786     .subsections = (const VMStateDescription*[]) {
1787         &vmstate_spapr_ov5_cas,
1788         &vmstate_spapr_patb_entry,
1789         &vmstate_spapr_pending_events,
1790         &vmstate_spapr_cap_htm,
1791         &vmstate_spapr_cap_vsx,
1792         &vmstate_spapr_cap_dfp,
1793         &vmstate_spapr_cap_cfpc,
1794         &vmstate_spapr_cap_sbbc,
1795         &vmstate_spapr_cap_ibs,
1796         NULL
1797     }
1798 };
1799 
1800 static int htab_save_setup(QEMUFile *f, void *opaque)
1801 {
1802     sPAPRMachineState *spapr = opaque;
1803 
1804     /* "Iteration" header */
1805     if (!spapr->htab_shift) {
1806         qemu_put_be32(f, -1);
1807     } else {
1808         qemu_put_be32(f, spapr->htab_shift);
1809     }
1810 
1811     if (spapr->htab) {
1812         spapr->htab_save_index = 0;
1813         spapr->htab_first_pass = true;
1814     } else {
1815         if (spapr->htab_shift) {
1816             assert(kvm_enabled());
1817         }
1818     }
1819 
1820 
1821     return 0;
1822 }
1823 
1824 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1825                             int chunkstart, int n_valid, int n_invalid)
1826 {
1827     qemu_put_be32(f, chunkstart);
1828     qemu_put_be16(f, n_valid);
1829     qemu_put_be16(f, n_invalid);
1830     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1831                     HASH_PTE_SIZE_64 * n_valid);
1832 }
1833 
1834 static void htab_save_end_marker(QEMUFile *f)
1835 {
1836     qemu_put_be32(f, 0);
1837     qemu_put_be16(f, 0);
1838     qemu_put_be16(f, 0);
1839 }
1840 
1841 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1842                                  int64_t max_ns)
1843 {
1844     bool has_timeout = max_ns != -1;
1845     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1846     int index = spapr->htab_save_index;
1847     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1848 
1849     assert(spapr->htab_first_pass);
1850 
1851     do {
1852         int chunkstart;
1853 
1854         /* Consume invalid HPTEs */
1855         while ((index < htabslots)
1856                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1857             CLEAN_HPTE(HPTE(spapr->htab, index));
1858             index++;
1859         }
1860 
1861         /* Consume valid HPTEs */
1862         chunkstart = index;
1863         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1864                && HPTE_VALID(HPTE(spapr->htab, index))) {
1865             CLEAN_HPTE(HPTE(spapr->htab, index));
1866             index++;
1867         }
1868 
1869         if (index > chunkstart) {
1870             int n_valid = index - chunkstart;
1871 
1872             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
1873 
1874             if (has_timeout &&
1875                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1876                 break;
1877             }
1878         }
1879     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1880 
1881     if (index >= htabslots) {
1882         assert(index == htabslots);
1883         index = 0;
1884         spapr->htab_first_pass = false;
1885     }
1886     spapr->htab_save_index = index;
1887 }
1888 
1889 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1890                                 int64_t max_ns)
1891 {
1892     bool final = max_ns < 0;
1893     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1894     int examined = 0, sent = 0;
1895     int index = spapr->htab_save_index;
1896     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1897 
1898     assert(!spapr->htab_first_pass);
1899 
1900     do {
1901         int chunkstart, invalidstart;
1902 
1903         /* Consume non-dirty HPTEs */
1904         while ((index < htabslots)
1905                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1906             index++;
1907             examined++;
1908         }
1909 
1910         chunkstart = index;
1911         /* Consume valid dirty HPTEs */
1912         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1913                && HPTE_DIRTY(HPTE(spapr->htab, index))
1914                && HPTE_VALID(HPTE(spapr->htab, index))) {
1915             CLEAN_HPTE(HPTE(spapr->htab, index));
1916             index++;
1917             examined++;
1918         }
1919 
1920         invalidstart = index;
1921         /* Consume invalid dirty HPTEs */
1922         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1923                && HPTE_DIRTY(HPTE(spapr->htab, index))
1924                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1925             CLEAN_HPTE(HPTE(spapr->htab, index));
1926             index++;
1927             examined++;
1928         }
1929 
1930         if (index > chunkstart) {
1931             int n_valid = invalidstart - chunkstart;
1932             int n_invalid = index - invalidstart;
1933 
1934             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
1935             sent += index - chunkstart;
1936 
1937             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1938                 break;
1939             }
1940         }
1941 
1942         if (examined >= htabslots) {
1943             break;
1944         }
1945 
1946         if (index >= htabslots) {
1947             assert(index == htabslots);
1948             index = 0;
1949         }
1950     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1951 
1952     if (index >= htabslots) {
1953         assert(index == htabslots);
1954         index = 0;
1955     }
1956 
1957     spapr->htab_save_index = index;
1958 
1959     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1960 }
1961 
1962 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1963 #define MAX_KVM_BUF_SIZE    2048
1964 
1965 static int htab_save_iterate(QEMUFile *f, void *opaque)
1966 {
1967     sPAPRMachineState *spapr = opaque;
1968     int fd;
1969     int rc = 0;
1970 
1971     /* Iteration header */
1972     if (!spapr->htab_shift) {
1973         qemu_put_be32(f, -1);
1974         return 1;
1975     } else {
1976         qemu_put_be32(f, 0);
1977     }
1978 
1979     if (!spapr->htab) {
1980         assert(kvm_enabled());
1981 
1982         fd = get_htab_fd(spapr);
1983         if (fd < 0) {
1984             return fd;
1985         }
1986 
1987         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1988         if (rc < 0) {
1989             return rc;
1990         }
1991     } else  if (spapr->htab_first_pass) {
1992         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1993     } else {
1994         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1995     }
1996 
1997     htab_save_end_marker(f);
1998 
1999     return rc;
2000 }
2001 
2002 static int htab_save_complete(QEMUFile *f, void *opaque)
2003 {
2004     sPAPRMachineState *spapr = opaque;
2005     int fd;
2006 
2007     /* Iteration header */
2008     if (!spapr->htab_shift) {
2009         qemu_put_be32(f, -1);
2010         return 0;
2011     } else {
2012         qemu_put_be32(f, 0);
2013     }
2014 
2015     if (!spapr->htab) {
2016         int rc;
2017 
2018         assert(kvm_enabled());
2019 
2020         fd = get_htab_fd(spapr);
2021         if (fd < 0) {
2022             return fd;
2023         }
2024 
2025         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2026         if (rc < 0) {
2027             return rc;
2028         }
2029     } else {
2030         if (spapr->htab_first_pass) {
2031             htab_save_first_pass(f, spapr, -1);
2032         }
2033         htab_save_later_pass(f, spapr, -1);
2034     }
2035 
2036     /* End marker */
2037     htab_save_end_marker(f);
2038 
2039     return 0;
2040 }
2041 
2042 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2043 {
2044     sPAPRMachineState *spapr = opaque;
2045     uint32_t section_hdr;
2046     int fd = -1;
2047     Error *local_err = NULL;
2048 
2049     if (version_id < 1 || version_id > 1) {
2050         error_report("htab_load() bad version");
2051         return -EINVAL;
2052     }
2053 
2054     section_hdr = qemu_get_be32(f);
2055 
2056     if (section_hdr == -1) {
2057         spapr_free_hpt(spapr);
2058         return 0;
2059     }
2060 
2061     if (section_hdr) {
2062         /* First section gives the htab size */
2063         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2064         if (local_err) {
2065             error_report_err(local_err);
2066             return -EINVAL;
2067         }
2068         return 0;
2069     }
2070 
2071     if (!spapr->htab) {
2072         assert(kvm_enabled());
2073 
2074         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2075         if (fd < 0) {
2076             error_report_err(local_err);
2077             return fd;
2078         }
2079     }
2080 
2081     while (true) {
2082         uint32_t index;
2083         uint16_t n_valid, n_invalid;
2084 
2085         index = qemu_get_be32(f);
2086         n_valid = qemu_get_be16(f);
2087         n_invalid = qemu_get_be16(f);
2088 
2089         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2090             /* End of Stream */
2091             break;
2092         }
2093 
2094         if ((index + n_valid + n_invalid) >
2095             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2096             /* Bad index in stream */
2097             error_report(
2098                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2099                 index, n_valid, n_invalid, spapr->htab_shift);
2100             return -EINVAL;
2101         }
2102 
2103         if (spapr->htab) {
2104             if (n_valid) {
2105                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2106                                 HASH_PTE_SIZE_64 * n_valid);
2107             }
2108             if (n_invalid) {
2109                 memset(HPTE(spapr->htab, index + n_valid), 0,
2110                        HASH_PTE_SIZE_64 * n_invalid);
2111             }
2112         } else {
2113             int rc;
2114 
2115             assert(fd >= 0);
2116 
2117             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2118             if (rc < 0) {
2119                 return rc;
2120             }
2121         }
2122     }
2123 
2124     if (!spapr->htab) {
2125         assert(fd >= 0);
2126         close(fd);
2127     }
2128 
2129     return 0;
2130 }
2131 
2132 static void htab_save_cleanup(void *opaque)
2133 {
2134     sPAPRMachineState *spapr = opaque;
2135 
2136     close_htab_fd(spapr);
2137 }
2138 
2139 static SaveVMHandlers savevm_htab_handlers = {
2140     .save_setup = htab_save_setup,
2141     .save_live_iterate = htab_save_iterate,
2142     .save_live_complete_precopy = htab_save_complete,
2143     .save_cleanup = htab_save_cleanup,
2144     .load_state = htab_load,
2145 };
2146 
2147 static void spapr_boot_set(void *opaque, const char *boot_device,
2148                            Error **errp)
2149 {
2150     MachineState *machine = MACHINE(opaque);
2151     machine->boot_order = g_strdup(boot_device);
2152 }
2153 
2154 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2155 {
2156     MachineState *machine = MACHINE(spapr);
2157     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2158     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2159     int i;
2160 
2161     for (i = 0; i < nr_lmbs; i++) {
2162         uint64_t addr;
2163 
2164         addr = i * lmb_size + spapr->hotplug_memory.base;
2165         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2166                                addr / lmb_size);
2167     }
2168 }
2169 
2170 /*
2171  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2172  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2173  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2174  */
2175 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2176 {
2177     int i;
2178 
2179     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2180         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2181                    " is not aligned to %llu MiB",
2182                    machine->ram_size,
2183                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2184         return;
2185     }
2186 
2187     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2188         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2189                    " is not aligned to %llu MiB",
2190                    machine->ram_size,
2191                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2192         return;
2193     }
2194 
2195     for (i = 0; i < nb_numa_nodes; i++) {
2196         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2197             error_setg(errp,
2198                        "Node %d memory size 0x%" PRIx64
2199                        " is not aligned to %llu MiB",
2200                        i, numa_info[i].node_mem,
2201                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2202             return;
2203         }
2204     }
2205 }
2206 
2207 /* find cpu slot in machine->possible_cpus by core_id */
2208 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2209 {
2210     int index = id / smp_threads;
2211 
2212     if (index >= ms->possible_cpus->len) {
2213         return NULL;
2214     }
2215     if (idx) {
2216         *idx = index;
2217     }
2218     return &ms->possible_cpus->cpus[index];
2219 }
2220 
2221 static void spapr_init_cpus(sPAPRMachineState *spapr)
2222 {
2223     MachineState *machine = MACHINE(spapr);
2224     MachineClass *mc = MACHINE_GET_CLASS(machine);
2225     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2226     int smt = kvmppc_smt_threads();
2227     const CPUArchIdList *possible_cpus;
2228     int boot_cores_nr = smp_cpus / smp_threads;
2229     int i;
2230 
2231     possible_cpus = mc->possible_cpu_arch_ids(machine);
2232     if (mc->has_hotpluggable_cpus) {
2233         if (smp_cpus % smp_threads) {
2234             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2235                          smp_cpus, smp_threads);
2236             exit(1);
2237         }
2238         if (max_cpus % smp_threads) {
2239             error_report("max_cpus (%u) must be multiple of threads (%u)",
2240                          max_cpus, smp_threads);
2241             exit(1);
2242         }
2243     } else {
2244         if (max_cpus != smp_cpus) {
2245             error_report("This machine version does not support CPU hotplug");
2246             exit(1);
2247         }
2248         boot_cores_nr = possible_cpus->len;
2249     }
2250 
2251     for (i = 0; i < possible_cpus->len; i++) {
2252         int core_id = i * smp_threads;
2253 
2254         if (mc->has_hotpluggable_cpus) {
2255             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2256                                    (core_id / smp_threads) * smt);
2257         }
2258 
2259         if (i < boot_cores_nr) {
2260             Object *core  = object_new(type);
2261             int nr_threads = smp_threads;
2262 
2263             /* Handle the partially filled core for older machine types */
2264             if ((i + 1) * smp_threads >= smp_cpus) {
2265                 nr_threads = smp_cpus - i * smp_threads;
2266             }
2267 
2268             object_property_set_int(core, nr_threads, "nr-threads",
2269                                     &error_fatal);
2270             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2271                                     &error_fatal);
2272             object_property_set_bool(core, true, "realized", &error_fatal);
2273         }
2274     }
2275 }
2276 
2277 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2278 {
2279     Error *local_err = NULL;
2280     bool vsmt_user = !!spapr->vsmt;
2281     int kvm_smt = kvmppc_smt_threads();
2282     int ret;
2283 
2284     if (!kvm_enabled() && (smp_threads > 1)) {
2285         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2286                      "on a pseries machine");
2287         goto out;
2288     }
2289     if (!is_power_of_2(smp_threads)) {
2290         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2291                      "machine because it must be a power of 2", smp_threads);
2292         goto out;
2293     }
2294 
2295     /* Detemine the VSMT mode to use: */
2296     if (vsmt_user) {
2297         if (spapr->vsmt < smp_threads) {
2298             error_setg(&local_err, "Cannot support VSMT mode %d"
2299                          " because it must be >= threads/core (%d)",
2300                          spapr->vsmt, smp_threads);
2301             goto out;
2302         }
2303         /* In this case, spapr->vsmt has been set by the command line */
2304     } else {
2305         /*
2306          * Default VSMT value is tricky, because we need it to be as
2307          * consistent as possible (for migration), but this requires
2308          * changing it for at least some existing cases.  We pick 8 as
2309          * the value that we'd get with KVM on POWER8, the
2310          * overwhelmingly common case in production systems.
2311          */
2312         spapr->vsmt = 8;
2313     }
2314 
2315     /* KVM: If necessary, set the SMT mode: */
2316     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2317         ret = kvmppc_set_smt_threads(spapr->vsmt);
2318         if (ret) {
2319             /* Looks like KVM isn't able to change VSMT mode */
2320             error_setg(&local_err,
2321                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2322                        spapr->vsmt, ret);
2323             /* We can live with that if the default one is big enough
2324              * for the number of threads, and a submultiple of the one
2325              * we want.  In this case we'll waste some vcpu ids, but
2326              * behaviour will be correct */
2327             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2328                 warn_report_err(local_err);
2329                 local_err = NULL;
2330                 goto out;
2331             } else {
2332                 if (!vsmt_user) {
2333                     error_append_hint(&local_err,
2334                                       "On PPC, a VM with %d threads/core"
2335                                       " on a host with %d threads/core"
2336                                       " requires the use of VSMT mode %d.\n",
2337                                       smp_threads, kvm_smt, spapr->vsmt);
2338                 }
2339                 kvmppc_hint_smt_possible(&local_err);
2340                 goto out;
2341             }
2342         }
2343     }
2344     /* else TCG: nothing to do currently */
2345 out:
2346     error_propagate(errp, local_err);
2347 }
2348 
2349 /* pSeries LPAR / sPAPR hardware init */
2350 static void spapr_machine_init(MachineState *machine)
2351 {
2352     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2353     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2354     const char *kernel_filename = machine->kernel_filename;
2355     const char *initrd_filename = machine->initrd_filename;
2356     PCIHostState *phb;
2357     int i;
2358     MemoryRegion *sysmem = get_system_memory();
2359     MemoryRegion *ram = g_new(MemoryRegion, 1);
2360     MemoryRegion *rma_region;
2361     void *rma = NULL;
2362     hwaddr rma_alloc_size;
2363     hwaddr node0_size = spapr_node0_size(machine);
2364     long load_limit, fw_size;
2365     char *filename;
2366     Error *resize_hpt_err = NULL;
2367 
2368     msi_nonbroken = true;
2369 
2370     QLIST_INIT(&spapr->phbs);
2371     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2372 
2373     /* Check HPT resizing availability */
2374     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2375     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2376         /*
2377          * If the user explicitly requested a mode we should either
2378          * supply it, or fail completely (which we do below).  But if
2379          * it's not set explicitly, we reset our mode to something
2380          * that works
2381          */
2382         if (resize_hpt_err) {
2383             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2384             error_free(resize_hpt_err);
2385             resize_hpt_err = NULL;
2386         } else {
2387             spapr->resize_hpt = smc->resize_hpt_default;
2388         }
2389     }
2390 
2391     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2392 
2393     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2394         /*
2395          * User requested HPT resize, but this host can't supply it.  Bail out
2396          */
2397         error_report_err(resize_hpt_err);
2398         exit(1);
2399     }
2400 
2401     /* Allocate RMA if necessary */
2402     rma_alloc_size = kvmppc_alloc_rma(&rma);
2403 
2404     if (rma_alloc_size == -1) {
2405         error_report("Unable to create RMA");
2406         exit(1);
2407     }
2408 
2409     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2410         spapr->rma_size = rma_alloc_size;
2411     } else {
2412         spapr->rma_size = node0_size;
2413 
2414         /* With KVM, we don't actually know whether KVM supports an
2415          * unbounded RMA (PR KVM) or is limited by the hash table size
2416          * (HV KVM using VRMA), so we always assume the latter
2417          *
2418          * In that case, we also limit the initial allocations for RTAS
2419          * etc... to 256M since we have no way to know what the VRMA size
2420          * is going to be as it depends on the size of the hash table
2421          * isn't determined yet.
2422          */
2423         if (kvm_enabled()) {
2424             spapr->vrma_adjust = 1;
2425             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2426         }
2427 
2428         /* Actually we don't support unbounded RMA anymore since we
2429          * added proper emulation of HV mode. The max we can get is
2430          * 16G which also happens to be what we configure for PAPR
2431          * mode so make sure we don't do anything bigger than that
2432          */
2433         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2434     }
2435 
2436     if (spapr->rma_size > node0_size) {
2437         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2438                      spapr->rma_size);
2439         exit(1);
2440     }
2441 
2442     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2443     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2444 
2445     /* Set up Interrupt Controller before we create the VCPUs */
2446     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2447 
2448     /* Set up containers for ibm,client-architecture-support negotiated options
2449      */
2450     spapr->ov5 = spapr_ovec_new();
2451     spapr->ov5_cas = spapr_ovec_new();
2452 
2453     if (smc->dr_lmb_enabled) {
2454         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2455         spapr_validate_node_memory(machine, &error_fatal);
2456     }
2457 
2458     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2459     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2460         /* KVM and TCG always allow GTSE with radix... */
2461         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2462     }
2463     /* ... but not with hash (currently). */
2464 
2465     /* advertise support for dedicated HP event source to guests */
2466     if (spapr->use_hotplug_event_source) {
2467         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2468     }
2469 
2470     /* advertise support for HPT resizing */
2471     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2472         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2473     }
2474 
2475     /* init CPUs */
2476     spapr_set_vsmt_mode(spapr, &error_fatal);
2477 
2478     spapr_init_cpus(spapr);
2479 
2480     if (kvm_enabled()) {
2481         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2482         kvmppc_enable_logical_ci_hcalls();
2483         kvmppc_enable_set_mode_hcall();
2484 
2485         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2486         kvmppc_enable_clear_ref_mod_hcalls();
2487     }
2488 
2489     /* allocate RAM */
2490     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2491                                          machine->ram_size);
2492     memory_region_add_subregion(sysmem, 0, ram);
2493 
2494     if (rma_alloc_size && rma) {
2495         rma_region = g_new(MemoryRegion, 1);
2496         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2497                                    rma_alloc_size, rma);
2498         vmstate_register_ram_global(rma_region);
2499         memory_region_add_subregion(sysmem, 0, rma_region);
2500     }
2501 
2502     /* initialize hotplug memory address space */
2503     if (machine->ram_size < machine->maxram_size) {
2504         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2505         /*
2506          * Limit the number of hotpluggable memory slots to half the number
2507          * slots that KVM supports, leaving the other half for PCI and other
2508          * devices. However ensure that number of slots doesn't drop below 32.
2509          */
2510         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2511                            SPAPR_MAX_RAM_SLOTS;
2512 
2513         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2514             max_memslots = SPAPR_MAX_RAM_SLOTS;
2515         }
2516         if (machine->ram_slots > max_memslots) {
2517             error_report("Specified number of memory slots %"
2518                          PRIu64" exceeds max supported %d",
2519                          machine->ram_slots, max_memslots);
2520             exit(1);
2521         }
2522 
2523         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2524                                               SPAPR_HOTPLUG_MEM_ALIGN);
2525         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2526                            "hotplug-memory", hotplug_mem_size);
2527         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2528                                     &spapr->hotplug_memory.mr);
2529     }
2530 
2531     if (smc->dr_lmb_enabled) {
2532         spapr_create_lmb_dr_connectors(spapr);
2533     }
2534 
2535     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2536     if (!filename) {
2537         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2538         exit(1);
2539     }
2540     spapr->rtas_size = get_image_size(filename);
2541     if (spapr->rtas_size < 0) {
2542         error_report("Could not get size of LPAR rtas '%s'", filename);
2543         exit(1);
2544     }
2545     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2546     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2547         error_report("Could not load LPAR rtas '%s'", filename);
2548         exit(1);
2549     }
2550     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2551         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2552                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2553         exit(1);
2554     }
2555     g_free(filename);
2556 
2557     /* Set up RTAS event infrastructure */
2558     spapr_events_init(spapr);
2559 
2560     /* Set up the RTC RTAS interfaces */
2561     spapr_rtc_create(spapr);
2562 
2563     /* Set up VIO bus */
2564     spapr->vio_bus = spapr_vio_bus_init();
2565 
2566     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2567         if (serial_hds[i]) {
2568             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2569         }
2570     }
2571 
2572     /* We always have at least the nvram device on VIO */
2573     spapr_create_nvram(spapr);
2574 
2575     /* Set up PCI */
2576     spapr_pci_rtas_init();
2577 
2578     phb = spapr_create_phb(spapr, 0);
2579 
2580     for (i = 0; i < nb_nics; i++) {
2581         NICInfo *nd = &nd_table[i];
2582 
2583         if (!nd->model) {
2584             nd->model = g_strdup("ibmveth");
2585         }
2586 
2587         if (strcmp(nd->model, "ibmveth") == 0) {
2588             spapr_vlan_create(spapr->vio_bus, nd);
2589         } else {
2590             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2591         }
2592     }
2593 
2594     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2595         spapr_vscsi_create(spapr->vio_bus);
2596     }
2597 
2598     /* Graphics */
2599     if (spapr_vga_init(phb->bus, &error_fatal)) {
2600         spapr->has_graphics = true;
2601         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2602     }
2603 
2604     if (machine->usb) {
2605         if (smc->use_ohci_by_default) {
2606             pci_create_simple(phb->bus, -1, "pci-ohci");
2607         } else {
2608             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2609         }
2610 
2611         if (spapr->has_graphics) {
2612             USBBus *usb_bus = usb_bus_find(-1);
2613 
2614             usb_create_simple(usb_bus, "usb-kbd");
2615             usb_create_simple(usb_bus, "usb-mouse");
2616         }
2617     }
2618 
2619     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2620         error_report(
2621             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2622             MIN_RMA_SLOF);
2623         exit(1);
2624     }
2625 
2626     if (kernel_filename) {
2627         uint64_t lowaddr = 0;
2628 
2629         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2630                                       NULL, NULL, &lowaddr, NULL, 1,
2631                                       PPC_ELF_MACHINE, 0, 0);
2632         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2633             spapr->kernel_size = load_elf(kernel_filename,
2634                                           translate_kernel_address, NULL, NULL,
2635                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2636                                           0, 0);
2637             spapr->kernel_le = spapr->kernel_size > 0;
2638         }
2639         if (spapr->kernel_size < 0) {
2640             error_report("error loading %s: %s", kernel_filename,
2641                          load_elf_strerror(spapr->kernel_size));
2642             exit(1);
2643         }
2644 
2645         /* load initrd */
2646         if (initrd_filename) {
2647             /* Try to locate the initrd in the gap between the kernel
2648              * and the firmware. Add a bit of space just in case
2649              */
2650             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2651                                   + 0x1ffff) & ~0xffff;
2652             spapr->initrd_size = load_image_targphys(initrd_filename,
2653                                                      spapr->initrd_base,
2654                                                      load_limit
2655                                                      - spapr->initrd_base);
2656             if (spapr->initrd_size < 0) {
2657                 error_report("could not load initial ram disk '%s'",
2658                              initrd_filename);
2659                 exit(1);
2660             }
2661         }
2662     }
2663 
2664     if (bios_name == NULL) {
2665         bios_name = FW_FILE_NAME;
2666     }
2667     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2668     if (!filename) {
2669         error_report("Could not find LPAR firmware '%s'", bios_name);
2670         exit(1);
2671     }
2672     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2673     if (fw_size <= 0) {
2674         error_report("Could not load LPAR firmware '%s'", filename);
2675         exit(1);
2676     }
2677     g_free(filename);
2678 
2679     /* FIXME: Should register things through the MachineState's qdev
2680      * interface, this is a legacy from the sPAPREnvironment structure
2681      * which predated MachineState but had a similar function */
2682     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2683     register_savevm_live(NULL, "spapr/htab", -1, 1,
2684                          &savevm_htab_handlers, spapr);
2685 
2686     qemu_register_boot_set(spapr_boot_set, spapr);
2687 
2688     if (kvm_enabled()) {
2689         /* to stop and start vmclock */
2690         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2691                                          &spapr->tb);
2692 
2693         kvmppc_spapr_enable_inkernel_multitce();
2694     }
2695 }
2696 
2697 static int spapr_kvm_type(const char *vm_type)
2698 {
2699     if (!vm_type) {
2700         return 0;
2701     }
2702 
2703     if (!strcmp(vm_type, "HV")) {
2704         return 1;
2705     }
2706 
2707     if (!strcmp(vm_type, "PR")) {
2708         return 2;
2709     }
2710 
2711     error_report("Unknown kvm-type specified '%s'", vm_type);
2712     exit(1);
2713 }
2714 
2715 /*
2716  * Implementation of an interface to adjust firmware path
2717  * for the bootindex property handling.
2718  */
2719 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2720                                    DeviceState *dev)
2721 {
2722 #define CAST(type, obj, name) \
2723     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2724     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2725     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2726     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2727 
2728     if (d) {
2729         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2730         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2731         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2732 
2733         if (spapr) {
2734             /*
2735              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2736              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2737              * in the top 16 bits of the 64-bit LUN
2738              */
2739             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2740             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2741                                    (uint64_t)id << 48);
2742         } else if (virtio) {
2743             /*
2744              * We use SRP luns of the form 01000000 | (target << 8) | lun
2745              * in the top 32 bits of the 64-bit LUN
2746              * Note: the quote above is from SLOF and it is wrong,
2747              * the actual binding is:
2748              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2749              */
2750             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2751             if (d->lun >= 256) {
2752                 /* Use the LUN "flat space addressing method" */
2753                 id |= 0x4000;
2754             }
2755             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2756                                    (uint64_t)id << 32);
2757         } else if (usb) {
2758             /*
2759              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2760              * in the top 32 bits of the 64-bit LUN
2761              */
2762             unsigned usb_port = atoi(usb->port->path);
2763             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2764             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2765                                    (uint64_t)id << 32);
2766         }
2767     }
2768 
2769     /*
2770      * SLOF probes the USB devices, and if it recognizes that the device is a
2771      * storage device, it changes its name to "storage" instead of "usb-host",
2772      * and additionally adds a child node for the SCSI LUN, so the correct
2773      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2774      */
2775     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2776         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2777         if (usb_host_dev_is_scsi_storage(usbdev)) {
2778             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2779         }
2780     }
2781 
2782     if (phb) {
2783         /* Replace "pci" with "pci@800000020000000" */
2784         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2785     }
2786 
2787     if (vsc) {
2788         /* Same logic as virtio above */
2789         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2790         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2791     }
2792 
2793     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2794         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2795         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2796         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2797     }
2798 
2799     return NULL;
2800 }
2801 
2802 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2803 {
2804     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2805 
2806     return g_strdup(spapr->kvm_type);
2807 }
2808 
2809 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2810 {
2811     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2812 
2813     g_free(spapr->kvm_type);
2814     spapr->kvm_type = g_strdup(value);
2815 }
2816 
2817 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2818 {
2819     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2820 
2821     return spapr->use_hotplug_event_source;
2822 }
2823 
2824 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2825                                             Error **errp)
2826 {
2827     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2828 
2829     spapr->use_hotplug_event_source = value;
2830 }
2831 
2832 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2833 {
2834     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2835 
2836     switch (spapr->resize_hpt) {
2837     case SPAPR_RESIZE_HPT_DEFAULT:
2838         return g_strdup("default");
2839     case SPAPR_RESIZE_HPT_DISABLED:
2840         return g_strdup("disabled");
2841     case SPAPR_RESIZE_HPT_ENABLED:
2842         return g_strdup("enabled");
2843     case SPAPR_RESIZE_HPT_REQUIRED:
2844         return g_strdup("required");
2845     }
2846     g_assert_not_reached();
2847 }
2848 
2849 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2850 {
2851     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2852 
2853     if (strcmp(value, "default") == 0) {
2854         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2855     } else if (strcmp(value, "disabled") == 0) {
2856         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2857     } else if (strcmp(value, "enabled") == 0) {
2858         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2859     } else if (strcmp(value, "required") == 0) {
2860         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2861     } else {
2862         error_setg(errp, "Bad value for \"resize-hpt\" property");
2863     }
2864 }
2865 
2866 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
2867                                    void *opaque, Error **errp)
2868 {
2869     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2870 }
2871 
2872 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
2873                                    void *opaque, Error **errp)
2874 {
2875     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2876 }
2877 
2878 static void spapr_instance_init(Object *obj)
2879 {
2880     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2881 
2882     spapr->htab_fd = -1;
2883     spapr->use_hotplug_event_source = true;
2884     object_property_add_str(obj, "kvm-type",
2885                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2886     object_property_set_description(obj, "kvm-type",
2887                                     "Specifies the KVM virtualization mode (HV, PR)",
2888                                     NULL);
2889     object_property_add_bool(obj, "modern-hotplug-events",
2890                             spapr_get_modern_hotplug_events,
2891                             spapr_set_modern_hotplug_events,
2892                             NULL);
2893     object_property_set_description(obj, "modern-hotplug-events",
2894                                     "Use dedicated hotplug event mechanism in"
2895                                     " place of standard EPOW events when possible"
2896                                     " (required for memory hot-unplug support)",
2897                                     NULL);
2898 
2899     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
2900                             "Maximum permitted CPU compatibility mode",
2901                             &error_fatal);
2902 
2903     object_property_add_str(obj, "resize-hpt",
2904                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
2905     object_property_set_description(obj, "resize-hpt",
2906                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
2907                                     NULL);
2908     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
2909                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
2910     object_property_set_description(obj, "vsmt",
2911                                     "Virtual SMT: KVM behaves as if this were"
2912                                     " the host's SMT mode", &error_abort);
2913 }
2914 
2915 static void spapr_machine_finalizefn(Object *obj)
2916 {
2917     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2918 
2919     g_free(spapr->kvm_type);
2920 }
2921 
2922 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2923 {
2924     cpu_synchronize_state(cs);
2925     ppc_cpu_do_system_reset(cs);
2926 }
2927 
2928 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2929 {
2930     CPUState *cs;
2931 
2932     CPU_FOREACH(cs) {
2933         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2934     }
2935 }
2936 
2937 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2938                            uint32_t node, bool dedicated_hp_event_source,
2939                            Error **errp)
2940 {
2941     sPAPRDRConnector *drc;
2942     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2943     int i, fdt_offset, fdt_size;
2944     void *fdt;
2945     uint64_t addr = addr_start;
2946     bool hotplugged = spapr_drc_hotplugged(dev);
2947     Error *local_err = NULL;
2948 
2949     for (i = 0; i < nr_lmbs; i++) {
2950         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2951                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2952         g_assert(drc);
2953 
2954         fdt = create_device_tree(&fdt_size);
2955         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2956                                                 SPAPR_MEMORY_BLOCK_SIZE);
2957 
2958         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
2959         if (local_err) {
2960             while (addr > addr_start) {
2961                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
2962                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2963                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
2964                 spapr_drc_detach(drc);
2965             }
2966             g_free(fdt);
2967             error_propagate(errp, local_err);
2968             return;
2969         }
2970         if (!hotplugged) {
2971             spapr_drc_reset(drc);
2972         }
2973         addr += SPAPR_MEMORY_BLOCK_SIZE;
2974     }
2975     /* send hotplug notification to the
2976      * guest only in case of hotplugged memory
2977      */
2978     if (hotplugged) {
2979         if (dedicated_hp_event_source) {
2980             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2981                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2982             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2983                                                    nr_lmbs,
2984                                                    spapr_drc_index(drc));
2985         } else {
2986             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2987                                            nr_lmbs);
2988         }
2989     }
2990 }
2991 
2992 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2993                               uint32_t node, Error **errp)
2994 {
2995     Error *local_err = NULL;
2996     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2997     PCDIMMDevice *dimm = PC_DIMM(dev);
2998     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2999     MemoryRegion *mr;
3000     uint64_t align, size, addr;
3001 
3002     mr = ddc->get_memory_region(dimm, &local_err);
3003     if (local_err) {
3004         goto out;
3005     }
3006     align = memory_region_get_alignment(mr);
3007     size = memory_region_size(mr);
3008 
3009     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
3010     if (local_err) {
3011         goto out;
3012     }
3013 
3014     addr = object_property_get_uint(OBJECT(dimm),
3015                                     PC_DIMM_ADDR_PROP, &local_err);
3016     if (local_err) {
3017         goto out_unplug;
3018     }
3019 
3020     spapr_add_lmbs(dev, addr, size, node,
3021                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3022                    &local_err);
3023     if (local_err) {
3024         goto out_unplug;
3025     }
3026 
3027     return;
3028 
3029 out_unplug:
3030     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
3031 out:
3032     error_propagate(errp, local_err);
3033 }
3034 
3035 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3036                                   Error **errp)
3037 {
3038     PCDIMMDevice *dimm = PC_DIMM(dev);
3039     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3040     MemoryRegion *mr;
3041     uint64_t size;
3042     char *mem_dev;
3043 
3044     mr = ddc->get_memory_region(dimm, errp);
3045     if (!mr) {
3046         return;
3047     }
3048     size = memory_region_size(mr);
3049 
3050     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3051         error_setg(errp, "Hotplugged memory size must be a multiple of "
3052                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
3053         return;
3054     }
3055 
3056     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
3057     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
3058         error_setg(errp, "Memory backend has bad page size. "
3059                    "Use 'memory-backend-file' with correct mem-path.");
3060         goto out;
3061     }
3062 
3063 out:
3064     g_free(mem_dev);
3065 }
3066 
3067 struct sPAPRDIMMState {
3068     PCDIMMDevice *dimm;
3069     uint32_t nr_lmbs;
3070     QTAILQ_ENTRY(sPAPRDIMMState) next;
3071 };
3072 
3073 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3074                                                        PCDIMMDevice *dimm)
3075 {
3076     sPAPRDIMMState *dimm_state = NULL;
3077 
3078     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3079         if (dimm_state->dimm == dimm) {
3080             break;
3081         }
3082     }
3083     return dimm_state;
3084 }
3085 
3086 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3087                                                       uint32_t nr_lmbs,
3088                                                       PCDIMMDevice *dimm)
3089 {
3090     sPAPRDIMMState *ds = NULL;
3091 
3092     /*
3093      * If this request is for a DIMM whose removal had failed earlier
3094      * (due to guest's refusal to remove the LMBs), we would have this
3095      * dimm already in the pending_dimm_unplugs list. In that
3096      * case don't add again.
3097      */
3098     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3099     if (!ds) {
3100         ds = g_malloc0(sizeof(sPAPRDIMMState));
3101         ds->nr_lmbs = nr_lmbs;
3102         ds->dimm = dimm;
3103         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3104     }
3105     return ds;
3106 }
3107 
3108 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3109                                               sPAPRDIMMState *dimm_state)
3110 {
3111     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3112     g_free(dimm_state);
3113 }
3114 
3115 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3116                                                         PCDIMMDevice *dimm)
3117 {
3118     sPAPRDRConnector *drc;
3119     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3120     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3121     uint64_t size = memory_region_size(mr);
3122     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3123     uint32_t avail_lmbs = 0;
3124     uint64_t addr_start, addr;
3125     int i;
3126 
3127     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3128                                          &error_abort);
3129 
3130     addr = addr_start;
3131     for (i = 0; i < nr_lmbs; i++) {
3132         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3133                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3134         g_assert(drc);
3135         if (drc->dev) {
3136             avail_lmbs++;
3137         }
3138         addr += SPAPR_MEMORY_BLOCK_SIZE;
3139     }
3140 
3141     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3142 }
3143 
3144 /* Callback to be called during DRC release. */
3145 void spapr_lmb_release(DeviceState *dev)
3146 {
3147     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_hotplug_handler(dev));
3148     PCDIMMDevice *dimm = PC_DIMM(dev);
3149     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3150     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3151     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3152 
3153     /* This information will get lost if a migration occurs
3154      * during the unplug process. In this case recover it. */
3155     if (ds == NULL) {
3156         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3157         g_assert(ds);
3158         /* The DRC being examined by the caller at least must be counted */
3159         g_assert(ds->nr_lmbs);
3160     }
3161 
3162     if (--ds->nr_lmbs) {
3163         return;
3164     }
3165 
3166     /*
3167      * Now that all the LMBs have been removed by the guest, call the
3168      * pc-dimm unplug handler to cleanup up the pc-dimm device.
3169      */
3170     pc_dimm_memory_unplug(dev, &spapr->hotplug_memory, mr);
3171     object_unparent(OBJECT(dev));
3172     spapr_pending_dimm_unplugs_remove(spapr, ds);
3173 }
3174 
3175 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3176                                         DeviceState *dev, Error **errp)
3177 {
3178     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3179     Error *local_err = NULL;
3180     PCDIMMDevice *dimm = PC_DIMM(dev);
3181     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3182     MemoryRegion *mr;
3183     uint32_t nr_lmbs;
3184     uint64_t size, addr_start, addr;
3185     int i;
3186     sPAPRDRConnector *drc;
3187 
3188     mr = ddc->get_memory_region(dimm, &local_err);
3189     if (local_err) {
3190         goto out;
3191     }
3192     size = memory_region_size(mr);
3193     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3194 
3195     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3196                                          &local_err);
3197     if (local_err) {
3198         goto out;
3199     }
3200 
3201     /*
3202      * An existing pending dimm state for this DIMM means that there is an
3203      * unplug operation in progress, waiting for the spapr_lmb_release
3204      * callback to complete the job (BQL can't cover that far). In this case,
3205      * bail out to avoid detaching DRCs that were already released.
3206      */
3207     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3208         error_setg(&local_err,
3209                    "Memory unplug already in progress for device %s",
3210                    dev->id);
3211         goto out;
3212     }
3213 
3214     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3215 
3216     addr = addr_start;
3217     for (i = 0; i < nr_lmbs; i++) {
3218         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3219                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3220         g_assert(drc);
3221 
3222         spapr_drc_detach(drc);
3223         addr += SPAPR_MEMORY_BLOCK_SIZE;
3224     }
3225 
3226     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3227                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3228     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3229                                               nr_lmbs, spapr_drc_index(drc));
3230 out:
3231     error_propagate(errp, local_err);
3232 }
3233 
3234 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3235                                            sPAPRMachineState *spapr)
3236 {
3237     PowerPCCPU *cpu = POWERPC_CPU(cs);
3238     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3239     int id = spapr_vcpu_id(cpu);
3240     void *fdt;
3241     int offset, fdt_size;
3242     char *nodename;
3243 
3244     fdt = create_device_tree(&fdt_size);
3245     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3246     offset = fdt_add_subnode(fdt, 0, nodename);
3247 
3248     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3249     g_free(nodename);
3250 
3251     *fdt_offset = offset;
3252     return fdt;
3253 }
3254 
3255 /* Callback to be called during DRC release. */
3256 void spapr_core_release(DeviceState *dev)
3257 {
3258     MachineState *ms = MACHINE(qdev_get_hotplug_handler(dev));
3259     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3260     CPUCore *cc = CPU_CORE(dev);
3261     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3262 
3263     if (smc->pre_2_10_has_unused_icps) {
3264         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3265         int i;
3266 
3267         for (i = 0; i < cc->nr_threads; i++) {
3268             CPUState *cs = CPU(sc->threads[i]);
3269 
3270             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3271         }
3272     }
3273 
3274     assert(core_slot);
3275     core_slot->cpu = NULL;
3276     object_unparent(OBJECT(dev));
3277 }
3278 
3279 static
3280 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3281                                Error **errp)
3282 {
3283     int index;
3284     sPAPRDRConnector *drc;
3285     CPUCore *cc = CPU_CORE(dev);
3286     int smt = kvmppc_smt_threads();
3287 
3288     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3289         error_setg(errp, "Unable to find CPU core with core-id: %d",
3290                    cc->core_id);
3291         return;
3292     }
3293     if (index == 0) {
3294         error_setg(errp, "Boot CPU core may not be unplugged");
3295         return;
3296     }
3297 
3298     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3299     g_assert(drc);
3300 
3301     spapr_drc_detach(drc);
3302 
3303     spapr_hotplug_req_remove_by_index(drc);
3304 }
3305 
3306 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3307                             Error **errp)
3308 {
3309     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3310     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3311     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3312     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3313     CPUCore *cc = CPU_CORE(dev);
3314     CPUState *cs = CPU(core->threads[0]);
3315     sPAPRDRConnector *drc;
3316     Error *local_err = NULL;
3317     int smt = kvmppc_smt_threads();
3318     CPUArchId *core_slot;
3319     int index;
3320     bool hotplugged = spapr_drc_hotplugged(dev);
3321 
3322     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3323     if (!core_slot) {
3324         error_setg(errp, "Unable to find CPU core with core-id: %d",
3325                    cc->core_id);
3326         return;
3327     }
3328     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3329 
3330     g_assert(drc || !mc->has_hotpluggable_cpus);
3331 
3332     if (drc) {
3333         void *fdt;
3334         int fdt_offset;
3335 
3336         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3337 
3338         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3339         if (local_err) {
3340             g_free(fdt);
3341             error_propagate(errp, local_err);
3342             return;
3343         }
3344 
3345         if (hotplugged) {
3346             /*
3347              * Send hotplug notification interrupt to the guest only
3348              * in case of hotplugged CPUs.
3349              */
3350             spapr_hotplug_req_add_by_index(drc);
3351         } else {
3352             spapr_drc_reset(drc);
3353         }
3354     }
3355 
3356     core_slot->cpu = OBJECT(dev);
3357 
3358     if (smc->pre_2_10_has_unused_icps) {
3359         int i;
3360 
3361         for (i = 0; i < cc->nr_threads; i++) {
3362             cs = CPU(core->threads[i]);
3363             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3364         }
3365     }
3366 }
3367 
3368 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3369                                 Error **errp)
3370 {
3371     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3372     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3373     Error *local_err = NULL;
3374     CPUCore *cc = CPU_CORE(dev);
3375     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3376     const char *type = object_get_typename(OBJECT(dev));
3377     CPUArchId *core_slot;
3378     int index;
3379 
3380     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3381         error_setg(&local_err, "CPU hotplug not supported for this machine");
3382         goto out;
3383     }
3384 
3385     if (strcmp(base_core_type, type)) {
3386         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3387         goto out;
3388     }
3389 
3390     if (cc->core_id % smp_threads) {
3391         error_setg(&local_err, "invalid core id %d", cc->core_id);
3392         goto out;
3393     }
3394 
3395     /*
3396      * In general we should have homogeneous threads-per-core, but old
3397      * (pre hotplug support) machine types allow the last core to have
3398      * reduced threads as a compatibility hack for when we allowed
3399      * total vcpus not a multiple of threads-per-core.
3400      */
3401     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3402         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3403                    cc->nr_threads, smp_threads);
3404         goto out;
3405     }
3406 
3407     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3408     if (!core_slot) {
3409         error_setg(&local_err, "core id %d out of range", cc->core_id);
3410         goto out;
3411     }
3412 
3413     if (core_slot->cpu) {
3414         error_setg(&local_err, "core %d already populated", cc->core_id);
3415         goto out;
3416     }
3417 
3418     numa_cpu_pre_plug(core_slot, dev, &local_err);
3419 
3420 out:
3421     error_propagate(errp, local_err);
3422 }
3423 
3424 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3425                                       DeviceState *dev, Error **errp)
3426 {
3427     MachineState *ms = MACHINE(hotplug_dev);
3428     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3429 
3430     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3431         int node;
3432 
3433         if (!smc->dr_lmb_enabled) {
3434             error_setg(errp, "Memory hotplug not supported for this machine");
3435             return;
3436         }
3437         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3438         if (*errp) {
3439             return;
3440         }
3441         if (node < 0 || node >= MAX_NODES) {
3442             error_setg(errp, "Invaild node %d", node);
3443             return;
3444         }
3445 
3446         /*
3447          * Currently PowerPC kernel doesn't allow hot-adding memory to
3448          * memory-less node, but instead will silently add the memory
3449          * to the first node that has some memory. This causes two
3450          * unexpected behaviours for the user.
3451          *
3452          * - Memory gets hotplugged to a different node than what the user
3453          *   specified.
3454          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3455          *   to memory-less node, a reboot will set things accordingly
3456          *   and the previously hotplugged memory now ends in the right node.
3457          *   This appears as if some memory moved from one node to another.
3458          *
3459          * So until kernel starts supporting memory hotplug to memory-less
3460          * nodes, just prevent such attempts upfront in QEMU.
3461          */
3462         if (nb_numa_nodes && !numa_info[node].node_mem) {
3463             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3464                        node);
3465             return;
3466         }
3467 
3468         spapr_memory_plug(hotplug_dev, dev, node, errp);
3469     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3470         spapr_core_plug(hotplug_dev, dev, errp);
3471     }
3472 }
3473 
3474 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3475                                                 DeviceState *dev, Error **errp)
3476 {
3477     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3478     MachineClass *mc = MACHINE_GET_CLASS(sms);
3479 
3480     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3481         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3482             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3483         } else {
3484             /* NOTE: this means there is a window after guest reset, prior to
3485              * CAS negotiation, where unplug requests will fail due to the
3486              * capability not being detected yet. This is a bit different than
3487              * the case with PCI unplug, where the events will be queued and
3488              * eventually handled by the guest after boot
3489              */
3490             error_setg(errp, "Memory hot unplug not supported for this guest");
3491         }
3492     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3493         if (!mc->has_hotpluggable_cpus) {
3494             error_setg(errp, "CPU hot unplug not supported on this machine");
3495             return;
3496         }
3497         spapr_core_unplug_request(hotplug_dev, dev, errp);
3498     }
3499 }
3500 
3501 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3502                                           DeviceState *dev, Error **errp)
3503 {
3504     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3505         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3506     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3507         spapr_core_pre_plug(hotplug_dev, dev, errp);
3508     }
3509 }
3510 
3511 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3512                                                  DeviceState *dev)
3513 {
3514     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3515         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3516         return HOTPLUG_HANDLER(machine);
3517     }
3518     return NULL;
3519 }
3520 
3521 static CpuInstanceProperties
3522 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3523 {
3524     CPUArchId *core_slot;
3525     MachineClass *mc = MACHINE_GET_CLASS(machine);
3526 
3527     /* make sure possible_cpu are intialized */
3528     mc->possible_cpu_arch_ids(machine);
3529     /* get CPU core slot containing thread that matches cpu_index */
3530     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3531     assert(core_slot);
3532     return core_slot->props;
3533 }
3534 
3535 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3536 {
3537     return idx / smp_cores % nb_numa_nodes;
3538 }
3539 
3540 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3541 {
3542     int i;
3543     const char *core_type;
3544     int spapr_max_cores = max_cpus / smp_threads;
3545     MachineClass *mc = MACHINE_GET_CLASS(machine);
3546 
3547     if (!mc->has_hotpluggable_cpus) {
3548         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3549     }
3550     if (machine->possible_cpus) {
3551         assert(machine->possible_cpus->len == spapr_max_cores);
3552         return machine->possible_cpus;
3553     }
3554 
3555     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3556     if (!core_type) {
3557         error_report("Unable to find sPAPR CPU Core definition");
3558         exit(1);
3559     }
3560 
3561     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3562                              sizeof(CPUArchId) * spapr_max_cores);
3563     machine->possible_cpus->len = spapr_max_cores;
3564     for (i = 0; i < machine->possible_cpus->len; i++) {
3565         int core_id = i * smp_threads;
3566 
3567         machine->possible_cpus->cpus[i].type = core_type;
3568         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3569         machine->possible_cpus->cpus[i].arch_id = core_id;
3570         machine->possible_cpus->cpus[i].props.has_core_id = true;
3571         machine->possible_cpus->cpus[i].props.core_id = core_id;
3572     }
3573     return machine->possible_cpus;
3574 }
3575 
3576 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3577                                 uint64_t *buid, hwaddr *pio,
3578                                 hwaddr *mmio32, hwaddr *mmio64,
3579                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3580 {
3581     /*
3582      * New-style PHB window placement.
3583      *
3584      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3585      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3586      * windows.
3587      *
3588      * Some guest kernels can't work with MMIO windows above 1<<46
3589      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3590      *
3591      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3592      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3593      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3594      * 1TiB 64-bit MMIO windows for each PHB.
3595      */
3596     const uint64_t base_buid = 0x800000020000000ULL;
3597 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3598                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3599     int i;
3600 
3601     /* Sanity check natural alignments */
3602     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3603     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3604     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3605     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3606     /* Sanity check bounds */
3607     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3608                       SPAPR_PCI_MEM32_WIN_SIZE);
3609     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3610                       SPAPR_PCI_MEM64_WIN_SIZE);
3611 
3612     if (index >= SPAPR_MAX_PHBS) {
3613         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3614                    SPAPR_MAX_PHBS - 1);
3615         return;
3616     }
3617 
3618     *buid = base_buid + index;
3619     for (i = 0; i < n_dma; ++i) {
3620         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3621     }
3622 
3623     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3624     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3625     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3626 }
3627 
3628 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3629 {
3630     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3631 
3632     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3633 }
3634 
3635 static void spapr_ics_resend(XICSFabric *dev)
3636 {
3637     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3638 
3639     ics_resend(spapr->ics);
3640 }
3641 
3642 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3643 {
3644     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3645 
3646     return cpu ? ICP(cpu->intc) : NULL;
3647 }
3648 
3649 #define ICS_IRQ_FREE(ics, srcno)   \
3650     (!((ics)->irqs[(srcno)].flags & (XICS_FLAGS_IRQ_MASK)))
3651 
3652 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
3653 {
3654     int first, i;
3655 
3656     for (first = 0; first < ics->nr_irqs; first += alignnum) {
3657         if (num > (ics->nr_irqs - first)) {
3658             return -1;
3659         }
3660         for (i = first; i < first + num; ++i) {
3661             if (!ICS_IRQ_FREE(ics, i)) {
3662                 break;
3663             }
3664         }
3665         if (i == (first + num)) {
3666             return first;
3667         }
3668     }
3669 
3670     return -1;
3671 }
3672 
3673 /*
3674  * Allocate the IRQ number and set the IRQ type, LSI or MSI
3675  */
3676 static void spapr_irq_set_lsi(sPAPRMachineState *spapr, int irq, bool lsi)
3677 {
3678     ics_set_irq_type(spapr->ics, irq - spapr->ics->offset, lsi);
3679 }
3680 
3681 int spapr_irq_alloc(sPAPRMachineState *spapr, int irq_hint, bool lsi,
3682                     Error **errp)
3683 {
3684     ICSState *ics = spapr->ics;
3685     int irq;
3686 
3687     if (!ics) {
3688         return -1;
3689     }
3690     if (irq_hint) {
3691         if (!ICS_IRQ_FREE(ics, irq_hint - ics->offset)) {
3692             error_setg(errp, "can't allocate IRQ %d: already in use", irq_hint);
3693             return -1;
3694         }
3695         irq = irq_hint;
3696     } else {
3697         irq = ics_find_free_block(ics, 1, 1);
3698         if (irq < 0) {
3699             error_setg(errp, "can't allocate IRQ: no IRQ left");
3700             return -1;
3701         }
3702         irq += ics->offset;
3703     }
3704 
3705     spapr_irq_set_lsi(spapr, irq, lsi);
3706     trace_spapr_irq_alloc(irq);
3707 
3708     return irq;
3709 }
3710 
3711 /*
3712  * Allocate block of consecutive IRQs, and return the number of the first IRQ in
3713  * the block. If align==true, aligns the first IRQ number to num.
3714  */
3715 int spapr_irq_alloc_block(sPAPRMachineState *spapr, int num, bool lsi,
3716                           bool align, Error **errp)
3717 {
3718     ICSState *ics = spapr->ics;
3719     int i, first = -1;
3720 
3721     if (!ics) {
3722         return -1;
3723     }
3724 
3725     /*
3726      * MSIMesage::data is used for storing VIRQ so
3727      * it has to be aligned to num to support multiple
3728      * MSI vectors. MSI-X is not affected by this.
3729      * The hint is used for the first IRQ, the rest should
3730      * be allocated continuously.
3731      */
3732     if (align) {
3733         assert((num == 1) || (num == 2) || (num == 4) ||
3734                (num == 8) || (num == 16) || (num == 32));
3735         first = ics_find_free_block(ics, num, num);
3736     } else {
3737         first = ics_find_free_block(ics, num, 1);
3738     }
3739     if (first < 0) {
3740         error_setg(errp, "can't find a free %d-IRQ block", num);
3741         return -1;
3742     }
3743 
3744     first += ics->offset;
3745     for (i = first; i < first + num; ++i) {
3746         spapr_irq_set_lsi(spapr, i, lsi);
3747     }
3748 
3749     trace_spapr_irq_alloc_block(first, num, lsi, align);
3750 
3751     return first;
3752 }
3753 
3754 void spapr_irq_free(sPAPRMachineState *spapr, int irq, int num)
3755 {
3756     ICSState *ics = spapr->ics;
3757     int srcno = irq - ics->offset;
3758     int i;
3759 
3760     if (ics_valid_irq(ics, irq)) {
3761         trace_spapr_irq_free(0, irq, num);
3762         for (i = srcno; i < srcno + num; ++i) {
3763             if (ICS_IRQ_FREE(ics, i)) {
3764                 trace_spapr_irq_free_warn(0, i + ics->offset);
3765             }
3766             memset(&ics->irqs[i], 0, sizeof(ICSIRQState));
3767         }
3768     }
3769 }
3770 
3771 qemu_irq spapr_qirq(sPAPRMachineState *spapr, int irq)
3772 {
3773     ICSState *ics = spapr->ics;
3774 
3775     if (ics_valid_irq(ics, irq)) {
3776         return ics->qirqs[irq - ics->offset];
3777     }
3778 
3779     return NULL;
3780 }
3781 
3782 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3783                                  Monitor *mon)
3784 {
3785     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3786     CPUState *cs;
3787 
3788     CPU_FOREACH(cs) {
3789         PowerPCCPU *cpu = POWERPC_CPU(cs);
3790 
3791         icp_pic_print_info(ICP(cpu->intc), mon);
3792     }
3793 
3794     ics_pic_print_info(spapr->ics, mon);
3795 }
3796 
3797 int spapr_vcpu_id(PowerPCCPU *cpu)
3798 {
3799     CPUState *cs = CPU(cpu);
3800 
3801     if (kvm_enabled()) {
3802         return kvm_arch_vcpu_id(cs);
3803     } else {
3804         return cs->cpu_index;
3805     }
3806 }
3807 
3808 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3809 {
3810     CPUState *cs;
3811 
3812     CPU_FOREACH(cs) {
3813         PowerPCCPU *cpu = POWERPC_CPU(cs);
3814 
3815         if (spapr_vcpu_id(cpu) == vcpu_id) {
3816             return cpu;
3817         }
3818     }
3819 
3820     return NULL;
3821 }
3822 
3823 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3824 {
3825     MachineClass *mc = MACHINE_CLASS(oc);
3826     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3827     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3828     NMIClass *nc = NMI_CLASS(oc);
3829     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3830     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3831     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3832     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3833 
3834     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3835 
3836     /*
3837      * We set up the default / latest behaviour here.  The class_init
3838      * functions for the specific versioned machine types can override
3839      * these details for backwards compatibility
3840      */
3841     mc->init = spapr_machine_init;
3842     mc->reset = spapr_machine_reset;
3843     mc->block_default_type = IF_SCSI;
3844     mc->max_cpus = 1024;
3845     mc->no_parallel = 1;
3846     mc->default_boot_order = "";
3847     mc->default_ram_size = 512 * M_BYTE;
3848     mc->kvm_type = spapr_kvm_type;
3849     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3850     mc->pci_allow_0_address = true;
3851     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3852     hc->pre_plug = spapr_machine_device_pre_plug;
3853     hc->plug = spapr_machine_device_plug;
3854     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3855     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3856     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3857     hc->unplug_request = spapr_machine_device_unplug_request;
3858 
3859     smc->dr_lmb_enabled = true;
3860     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3861     mc->has_hotpluggable_cpus = true;
3862     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3863     fwc->get_dev_path = spapr_get_fw_dev_path;
3864     nc->nmi_monitor_handler = spapr_nmi;
3865     smc->phb_placement = spapr_phb_placement;
3866     vhc->hypercall = emulate_spapr_hypercall;
3867     vhc->hpt_mask = spapr_hpt_mask;
3868     vhc->map_hptes = spapr_map_hptes;
3869     vhc->unmap_hptes = spapr_unmap_hptes;
3870     vhc->store_hpte = spapr_store_hpte;
3871     vhc->get_patbe = spapr_get_patbe;
3872     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3873     xic->ics_get = spapr_ics_get;
3874     xic->ics_resend = spapr_ics_resend;
3875     xic->icp_get = spapr_icp_get;
3876     ispc->print_info = spapr_pic_print_info;
3877     /* Force NUMA node memory size to be a multiple of
3878      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3879      * in which LMBs are represented and hot-added
3880      */
3881     mc->numa_mem_align_shift = 28;
3882 
3883     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
3884     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
3885     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
3886     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
3887     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
3888     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
3889     spapr_caps_add_properties(smc, &error_abort);
3890 }
3891 
3892 static const TypeInfo spapr_machine_info = {
3893     .name          = TYPE_SPAPR_MACHINE,
3894     .parent        = TYPE_MACHINE,
3895     .abstract      = true,
3896     .instance_size = sizeof(sPAPRMachineState),
3897     .instance_init = spapr_instance_init,
3898     .instance_finalize = spapr_machine_finalizefn,
3899     .class_size    = sizeof(sPAPRMachineClass),
3900     .class_init    = spapr_machine_class_init,
3901     .interfaces = (InterfaceInfo[]) {
3902         { TYPE_FW_PATH_PROVIDER },
3903         { TYPE_NMI },
3904         { TYPE_HOTPLUG_HANDLER },
3905         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3906         { TYPE_XICS_FABRIC },
3907         { TYPE_INTERRUPT_STATS_PROVIDER },
3908         { }
3909     },
3910 };
3911 
3912 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3913     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3914                                                     void *data)      \
3915     {                                                                \
3916         MachineClass *mc = MACHINE_CLASS(oc);                        \
3917         spapr_machine_##suffix##_class_options(mc);                  \
3918         if (latest) {                                                \
3919             mc->alias = "pseries";                                   \
3920             mc->is_default = 1;                                      \
3921         }                                                            \
3922     }                                                                \
3923     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3924     {                                                                \
3925         MachineState *machine = MACHINE(obj);                        \
3926         spapr_machine_##suffix##_instance_options(machine);          \
3927     }                                                                \
3928     static const TypeInfo spapr_machine_##suffix##_info = {          \
3929         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3930         .parent = TYPE_SPAPR_MACHINE,                                \
3931         .class_init = spapr_machine_##suffix##_class_init,           \
3932         .instance_init = spapr_machine_##suffix##_instance_init,     \
3933     };                                                               \
3934     static void spapr_machine_register_##suffix(void)                \
3935     {                                                                \
3936         type_register(&spapr_machine_##suffix##_info);               \
3937     }                                                                \
3938     type_init(spapr_machine_register_##suffix)
3939 
3940 /*
3941  * pseries-2.12
3942  */
3943 static void spapr_machine_2_12_instance_options(MachineState *machine)
3944 {
3945 }
3946 
3947 static void spapr_machine_2_12_class_options(MachineClass *mc)
3948 {
3949     /* Defaults for the latest behaviour inherited from the base class */
3950 }
3951 
3952 DEFINE_SPAPR_MACHINE(2_12, "2.12", true);
3953 
3954 /*
3955  * pseries-2.11
3956  */
3957 #define SPAPR_COMPAT_2_11                                              \
3958     HW_COMPAT_2_11
3959 
3960 static void spapr_machine_2_11_instance_options(MachineState *machine)
3961 {
3962     spapr_machine_2_12_instance_options(machine);
3963 }
3964 
3965 static void spapr_machine_2_11_class_options(MachineClass *mc)
3966 {
3967     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3968 
3969     spapr_machine_2_12_class_options(mc);
3970     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
3971     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
3972 }
3973 
3974 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
3975 
3976 /*
3977  * pseries-2.10
3978  */
3979 #define SPAPR_COMPAT_2_10                                              \
3980     HW_COMPAT_2_10
3981 
3982 static void spapr_machine_2_10_instance_options(MachineState *machine)
3983 {
3984     spapr_machine_2_11_instance_options(machine);
3985 }
3986 
3987 static void spapr_machine_2_10_class_options(MachineClass *mc)
3988 {
3989     spapr_machine_2_11_class_options(mc);
3990     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
3991 }
3992 
3993 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
3994 
3995 /*
3996  * pseries-2.9
3997  */
3998 #define SPAPR_COMPAT_2_9                                               \
3999     HW_COMPAT_2_9                                                      \
4000     {                                                                  \
4001         .driver = TYPE_POWERPC_CPU,                                    \
4002         .property = "pre-2.10-migration",                              \
4003         .value    = "on",                                              \
4004     },                                                                 \
4005 
4006 static void spapr_machine_2_9_instance_options(MachineState *machine)
4007 {
4008     spapr_machine_2_10_instance_options(machine);
4009 }
4010 
4011 static void spapr_machine_2_9_class_options(MachineClass *mc)
4012 {
4013     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4014 
4015     spapr_machine_2_10_class_options(mc);
4016     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4017     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4018     smc->pre_2_10_has_unused_icps = true;
4019     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4020 }
4021 
4022 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4023 
4024 /*
4025  * pseries-2.8
4026  */
4027 #define SPAPR_COMPAT_2_8                                        \
4028     HW_COMPAT_2_8                                               \
4029     {                                                           \
4030         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4031         .property = "pcie-extended-configuration-space",        \
4032         .value    = "off",                                      \
4033     },
4034 
4035 static void spapr_machine_2_8_instance_options(MachineState *machine)
4036 {
4037     spapr_machine_2_9_instance_options(machine);
4038 }
4039 
4040 static void spapr_machine_2_8_class_options(MachineClass *mc)
4041 {
4042     spapr_machine_2_9_class_options(mc);
4043     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4044     mc->numa_mem_align_shift = 23;
4045 }
4046 
4047 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4048 
4049 /*
4050  * pseries-2.7
4051  */
4052 #define SPAPR_COMPAT_2_7                            \
4053     HW_COMPAT_2_7                                   \
4054     {                                               \
4055         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4056         .property = "mem_win_size",                 \
4057         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4058     },                                              \
4059     {                                               \
4060         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4061         .property = "mem64_win_size",               \
4062         .value    = "0",                            \
4063     },                                              \
4064     {                                               \
4065         .driver = TYPE_POWERPC_CPU,                 \
4066         .property = "pre-2.8-migration",            \
4067         .value    = "on",                           \
4068     },                                              \
4069     {                                               \
4070         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4071         .property = "pre-2.8-migration",            \
4072         .value    = "on",                           \
4073     },
4074 
4075 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4076                               uint64_t *buid, hwaddr *pio,
4077                               hwaddr *mmio32, hwaddr *mmio64,
4078                               unsigned n_dma, uint32_t *liobns, Error **errp)
4079 {
4080     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4081     const uint64_t base_buid = 0x800000020000000ULL;
4082     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4083     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4084     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4085     const uint32_t max_index = 255;
4086     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4087 
4088     uint64_t ram_top = MACHINE(spapr)->ram_size;
4089     hwaddr phb0_base, phb_base;
4090     int i;
4091 
4092     /* Do we have hotpluggable memory? */
4093     if (MACHINE(spapr)->maxram_size > ram_top) {
4094         /* Can't just use maxram_size, because there may be an
4095          * alignment gap between normal and hotpluggable memory
4096          * regions */
4097         ram_top = spapr->hotplug_memory.base +
4098             memory_region_size(&spapr->hotplug_memory.mr);
4099     }
4100 
4101     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4102 
4103     if (index > max_index) {
4104         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4105                    max_index);
4106         return;
4107     }
4108 
4109     *buid = base_buid + index;
4110     for (i = 0; i < n_dma; ++i) {
4111         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4112     }
4113 
4114     phb_base = phb0_base + index * phb_spacing;
4115     *pio = phb_base + pio_offset;
4116     *mmio32 = phb_base + mmio_offset;
4117     /*
4118      * We don't set the 64-bit MMIO window, relying on the PHB's
4119      * fallback behaviour of automatically splitting a large "32-bit"
4120      * window into contiguous 32-bit and 64-bit windows
4121      */
4122 }
4123 
4124 static void spapr_machine_2_7_instance_options(MachineState *machine)
4125 {
4126     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4127 
4128     spapr_machine_2_8_instance_options(machine);
4129     spapr->use_hotplug_event_source = false;
4130 }
4131 
4132 static void spapr_machine_2_7_class_options(MachineClass *mc)
4133 {
4134     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4135 
4136     spapr_machine_2_8_class_options(mc);
4137     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4138     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4139     smc->phb_placement = phb_placement_2_7;
4140 }
4141 
4142 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4143 
4144 /*
4145  * pseries-2.6
4146  */
4147 #define SPAPR_COMPAT_2_6 \
4148     HW_COMPAT_2_6 \
4149     { \
4150         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4151         .property = "ddw",\
4152         .value    = stringify(off),\
4153     },
4154 
4155 static void spapr_machine_2_6_instance_options(MachineState *machine)
4156 {
4157     spapr_machine_2_7_instance_options(machine);
4158 }
4159 
4160 static void spapr_machine_2_6_class_options(MachineClass *mc)
4161 {
4162     spapr_machine_2_7_class_options(mc);
4163     mc->has_hotpluggable_cpus = false;
4164     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4165 }
4166 
4167 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4168 
4169 /*
4170  * pseries-2.5
4171  */
4172 #define SPAPR_COMPAT_2_5 \
4173     HW_COMPAT_2_5 \
4174     { \
4175         .driver   = "spapr-vlan", \
4176         .property = "use-rx-buffer-pools", \
4177         .value    = "off", \
4178     },
4179 
4180 static void spapr_machine_2_5_instance_options(MachineState *machine)
4181 {
4182     spapr_machine_2_6_instance_options(machine);
4183 }
4184 
4185 static void spapr_machine_2_5_class_options(MachineClass *mc)
4186 {
4187     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4188 
4189     spapr_machine_2_6_class_options(mc);
4190     smc->use_ohci_by_default = true;
4191     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4192 }
4193 
4194 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4195 
4196 /*
4197  * pseries-2.4
4198  */
4199 #define SPAPR_COMPAT_2_4 \
4200         HW_COMPAT_2_4
4201 
4202 static void spapr_machine_2_4_instance_options(MachineState *machine)
4203 {
4204     spapr_machine_2_5_instance_options(machine);
4205 }
4206 
4207 static void spapr_machine_2_4_class_options(MachineClass *mc)
4208 {
4209     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4210 
4211     spapr_machine_2_5_class_options(mc);
4212     smc->dr_lmb_enabled = false;
4213     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4214 }
4215 
4216 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4217 
4218 /*
4219  * pseries-2.3
4220  */
4221 #define SPAPR_COMPAT_2_3 \
4222         HW_COMPAT_2_3 \
4223         {\
4224             .driver   = "spapr-pci-host-bridge",\
4225             .property = "dynamic-reconfiguration",\
4226             .value    = "off",\
4227         },
4228 
4229 static void spapr_machine_2_3_instance_options(MachineState *machine)
4230 {
4231     spapr_machine_2_4_instance_options(machine);
4232 }
4233 
4234 static void spapr_machine_2_3_class_options(MachineClass *mc)
4235 {
4236     spapr_machine_2_4_class_options(mc);
4237     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4238 }
4239 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4240 
4241 /*
4242  * pseries-2.2
4243  */
4244 
4245 #define SPAPR_COMPAT_2_2 \
4246         HW_COMPAT_2_2 \
4247         {\
4248             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4249             .property = "mem_win_size",\
4250             .value    = "0x20000000",\
4251         },
4252 
4253 static void spapr_machine_2_2_instance_options(MachineState *machine)
4254 {
4255     spapr_machine_2_3_instance_options(machine);
4256     machine->suppress_vmdesc = true;
4257 }
4258 
4259 static void spapr_machine_2_2_class_options(MachineClass *mc)
4260 {
4261     spapr_machine_2_3_class_options(mc);
4262     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4263 }
4264 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4265 
4266 /*
4267  * pseries-2.1
4268  */
4269 #define SPAPR_COMPAT_2_1 \
4270         HW_COMPAT_2_1
4271 
4272 static void spapr_machine_2_1_instance_options(MachineState *machine)
4273 {
4274     spapr_machine_2_2_instance_options(machine);
4275 }
4276 
4277 static void spapr_machine_2_1_class_options(MachineClass *mc)
4278 {
4279     spapr_machine_2_2_class_options(mc);
4280     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4281 }
4282 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4283 
4284 static void spapr_machine_register_types(void)
4285 {
4286     type_register_static(&spapr_machine_info);
4287 }
4288 
4289 type_init(spapr_machine_register_types)
4290