xref: /openbmc/qemu/hw/ppc/spapr.c (revision 086df4f30a8f33e7b696c79b6f5cd2c94afbf98f)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/block-backend.h"
39 #include "sysemu/cpus.h"
40 #include "sysemu/hw_accel.h"
41 #include "kvm_ppc.h"
42 #include "migration/misc.h"
43 #include "migration/global_state.h"
44 #include "migration/register.h"
45 #include "mmu-hash64.h"
46 #include "mmu-book3s-v3.h"
47 #include "cpu-models.h"
48 #include "qom/cpu.h"
49 
50 #include "hw/boards.h"
51 #include "hw/ppc/ppc.h"
52 #include "hw/loader.h"
53 
54 #include "hw/ppc/fdt.h"
55 #include "hw/ppc/spapr.h"
56 #include "hw/ppc/spapr_vio.h"
57 #include "hw/pci-host/spapr.h"
58 #include "hw/ppc/xics.h"
59 #include "hw/pci/msi.h"
60 
61 #include "hw/pci/pci.h"
62 #include "hw/scsi/scsi.h"
63 #include "hw/virtio/virtio-scsi.h"
64 #include "hw/virtio/vhost-scsi-common.h"
65 
66 #include "exec/address-spaces.h"
67 #include "hw/usb.h"
68 #include "qemu/config-file.h"
69 #include "qemu/error-report.h"
70 #include "trace.h"
71 #include "hw/nmi.h"
72 #include "hw/intc/intc.h"
73 
74 #include "hw/compat.h"
75 #include "qemu/cutils.h"
76 #include "hw/ppc/spapr_cpu_core.h"
77 
78 #include <libfdt.h>
79 
80 /* SLOF memory layout:
81  *
82  * SLOF raw image loaded at 0, copies its romfs right below the flat
83  * device-tree, then position SLOF itself 31M below that
84  *
85  * So we set FW_OVERHEAD to 40MB which should account for all of that
86  * and more
87  *
88  * We load our kernel at 4M, leaving space for SLOF initial image
89  */
90 #define FDT_MAX_SIZE            0x100000
91 #define RTAS_MAX_SIZE           0x10000
92 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
93 #define FW_MAX_SIZE             0x400000
94 #define FW_FILE_NAME            "slof.bin"
95 #define FW_OVERHEAD             0x2800000
96 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
97 
98 #define MIN_RMA_SLOF            128UL
99 
100 #define PHANDLE_XICP            0x00001111
101 
102 /* These two functions implement the VCPU id numbering: one to compute them
103  * all and one to identify thread 0 of a VCORE. Any change to the first one
104  * is likely to have an impact on the second one, so let's keep them close.
105  */
106 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
107 {
108     assert(spapr->vsmt);
109     return
110         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
111 }
112 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
113                                       PowerPCCPU *cpu)
114 {
115     assert(spapr->vsmt);
116     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
117 }
118 
119 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
120                                   const char *type_ics,
121                                   int nr_irqs, Error **errp)
122 {
123     Error *local_err = NULL;
124     Object *obj;
125 
126     obj = object_new(type_ics);
127     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
128     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
129                                    &error_abort);
130     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
131     if (local_err) {
132         goto error;
133     }
134     object_property_set_bool(obj, true, "realized", &local_err);
135     if (local_err) {
136         goto error;
137     }
138 
139     return ICS_SIMPLE(obj);
140 
141 error:
142     error_propagate(errp, local_err);
143     return NULL;
144 }
145 
146 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
147 {
148     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
149      * and newer QEMUs don't even have them. In both cases, we don't want
150      * to send anything on the wire.
151      */
152     return false;
153 }
154 
155 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
156     .name = "icp/server",
157     .version_id = 1,
158     .minimum_version_id = 1,
159     .needed = pre_2_10_vmstate_dummy_icp_needed,
160     .fields = (VMStateField[]) {
161         VMSTATE_UNUSED(4), /* uint32_t xirr */
162         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
163         VMSTATE_UNUSED(1), /* uint8_t mfrr */
164         VMSTATE_END_OF_LIST()
165     },
166 };
167 
168 static void pre_2_10_vmstate_register_dummy_icp(int i)
169 {
170     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
171                      (void *)(uintptr_t) i);
172 }
173 
174 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
175 {
176     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
177                        (void *)(uintptr_t) i);
178 }
179 
180 static int xics_max_server_number(sPAPRMachineState *spapr)
181 {
182     assert(spapr->vsmt);
183     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
184 }
185 
186 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
187 {
188     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
189 
190     if (kvm_enabled()) {
191         if (machine_kernel_irqchip_allowed(machine) &&
192             !xics_kvm_init(spapr, errp)) {
193             spapr->icp_type = TYPE_KVM_ICP;
194             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
195         }
196         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
197             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
198             return;
199         }
200     }
201 
202     if (!spapr->ics) {
203         xics_spapr_init(spapr);
204         spapr->icp_type = TYPE_ICP;
205         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
206         if (!spapr->ics) {
207             return;
208         }
209     }
210 }
211 
212 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
213                                   int smt_threads)
214 {
215     int i, ret = 0;
216     uint32_t servers_prop[smt_threads];
217     uint32_t gservers_prop[smt_threads * 2];
218     int index = spapr_get_vcpu_id(cpu);
219 
220     if (cpu->compat_pvr) {
221         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
222         if (ret < 0) {
223             return ret;
224         }
225     }
226 
227     /* Build interrupt servers and gservers properties */
228     for (i = 0; i < smt_threads; i++) {
229         servers_prop[i] = cpu_to_be32(index + i);
230         /* Hack, direct the group queues back to cpu 0 */
231         gservers_prop[i*2] = cpu_to_be32(index + i);
232         gservers_prop[i*2 + 1] = 0;
233     }
234     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
235                       servers_prop, sizeof(servers_prop));
236     if (ret < 0) {
237         return ret;
238     }
239     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
240                       gservers_prop, sizeof(gservers_prop));
241 
242     return ret;
243 }
244 
245 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
246 {
247     int index = spapr_get_vcpu_id(cpu);
248     uint32_t associativity[] = {cpu_to_be32(0x5),
249                                 cpu_to_be32(0x0),
250                                 cpu_to_be32(0x0),
251                                 cpu_to_be32(0x0),
252                                 cpu_to_be32(cpu->node_id),
253                                 cpu_to_be32(index)};
254 
255     /* Advertise NUMA via ibm,associativity */
256     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
257                           sizeof(associativity));
258 }
259 
260 /* Populate the "ibm,pa-features" property */
261 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
262                                        PowerPCCPU *cpu,
263                                        void *fdt, int offset,
264                                        bool legacy_guest)
265 {
266     CPUPPCState *env = &cpu->env;
267     uint8_t pa_features_206[] = { 6, 0,
268         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
269     uint8_t pa_features_207[] = { 24, 0,
270         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
271         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
272         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
273         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
274     uint8_t pa_features_300[] = { 66, 0,
275         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
276         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
277         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
278         /* 6: DS207 */
279         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
280         /* 16: Vector */
281         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
282         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
283         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
284         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
285         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
286         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
287         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
288         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
289         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
290         /* 42: PM, 44: PC RA, 46: SC vec'd */
291         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
292         /* 48: SIMD, 50: QP BFP, 52: String */
293         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
294         /* 54: DecFP, 56: DecI, 58: SHA */
295         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
296         /* 60: NM atomic, 62: RNG */
297         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
298     };
299     uint8_t *pa_features = NULL;
300     size_t pa_size;
301 
302     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
303         pa_features = pa_features_206;
304         pa_size = sizeof(pa_features_206);
305     }
306     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
307         pa_features = pa_features_207;
308         pa_size = sizeof(pa_features_207);
309     }
310     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
311         pa_features = pa_features_300;
312         pa_size = sizeof(pa_features_300);
313     }
314     if (!pa_features) {
315         return;
316     }
317 
318     if (env->ci_large_pages) {
319         /*
320          * Note: we keep CI large pages off by default because a 64K capable
321          * guest provisioned with large pages might otherwise try to map a qemu
322          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
323          * even if that qemu runs on a 4k host.
324          * We dd this bit back here if we are confident this is not an issue
325          */
326         pa_features[3] |= 0x20;
327     }
328     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
329         pa_features[24] |= 0x80;    /* Transactional memory support */
330     }
331     if (legacy_guest && pa_size > 40) {
332         /* Workaround for broken kernels that attempt (guest) radix
333          * mode when they can't handle it, if they see the radix bit set
334          * in pa-features. So hide it from them. */
335         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
336     }
337 
338     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
339 }
340 
341 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
342 {
343     int ret = 0, offset, cpus_offset;
344     CPUState *cs;
345     char cpu_model[32];
346     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
347 
348     CPU_FOREACH(cs) {
349         PowerPCCPU *cpu = POWERPC_CPU(cs);
350         DeviceClass *dc = DEVICE_GET_CLASS(cs);
351         int index = spapr_get_vcpu_id(cpu);
352         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
353 
354         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
355             continue;
356         }
357 
358         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
359 
360         cpus_offset = fdt_path_offset(fdt, "/cpus");
361         if (cpus_offset < 0) {
362             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
363             if (cpus_offset < 0) {
364                 return cpus_offset;
365             }
366         }
367         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
368         if (offset < 0) {
369             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
370             if (offset < 0) {
371                 return offset;
372             }
373         }
374 
375         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
376                           pft_size_prop, sizeof(pft_size_prop));
377         if (ret < 0) {
378             return ret;
379         }
380 
381         if (nb_numa_nodes > 1) {
382             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
383             if (ret < 0) {
384                 return ret;
385             }
386         }
387 
388         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
389         if (ret < 0) {
390             return ret;
391         }
392 
393         spapr_populate_pa_features(spapr, cpu, fdt, offset,
394                                    spapr->cas_legacy_guest_workaround);
395     }
396     return ret;
397 }
398 
399 static hwaddr spapr_node0_size(MachineState *machine)
400 {
401     if (nb_numa_nodes) {
402         int i;
403         for (i = 0; i < nb_numa_nodes; ++i) {
404             if (numa_info[i].node_mem) {
405                 return MIN(pow2floor(numa_info[i].node_mem),
406                            machine->ram_size);
407             }
408         }
409     }
410     return machine->ram_size;
411 }
412 
413 static void add_str(GString *s, const gchar *s1)
414 {
415     g_string_append_len(s, s1, strlen(s1) + 1);
416 }
417 
418 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
419                                        hwaddr size)
420 {
421     uint32_t associativity[] = {
422         cpu_to_be32(0x4), /* length */
423         cpu_to_be32(0x0), cpu_to_be32(0x0),
424         cpu_to_be32(0x0), cpu_to_be32(nodeid)
425     };
426     char mem_name[32];
427     uint64_t mem_reg_property[2];
428     int off;
429 
430     mem_reg_property[0] = cpu_to_be64(start);
431     mem_reg_property[1] = cpu_to_be64(size);
432 
433     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
434     off = fdt_add_subnode(fdt, 0, mem_name);
435     _FDT(off);
436     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
437     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
438                       sizeof(mem_reg_property))));
439     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
440                       sizeof(associativity))));
441     return off;
442 }
443 
444 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
445 {
446     MachineState *machine = MACHINE(spapr);
447     hwaddr mem_start, node_size;
448     int i, nb_nodes = nb_numa_nodes;
449     NodeInfo *nodes = numa_info;
450     NodeInfo ramnode;
451 
452     /* No NUMA nodes, assume there is just one node with whole RAM */
453     if (!nb_numa_nodes) {
454         nb_nodes = 1;
455         ramnode.node_mem = machine->ram_size;
456         nodes = &ramnode;
457     }
458 
459     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
460         if (!nodes[i].node_mem) {
461             continue;
462         }
463         if (mem_start >= machine->ram_size) {
464             node_size = 0;
465         } else {
466             node_size = nodes[i].node_mem;
467             if (node_size > machine->ram_size - mem_start) {
468                 node_size = machine->ram_size - mem_start;
469             }
470         }
471         if (!mem_start) {
472             /* spapr_machine_init() checks for rma_size <= node0_size
473              * already */
474             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
475             mem_start += spapr->rma_size;
476             node_size -= spapr->rma_size;
477         }
478         for ( ; node_size; ) {
479             hwaddr sizetmp = pow2floor(node_size);
480 
481             /* mem_start != 0 here */
482             if (ctzl(mem_start) < ctzl(sizetmp)) {
483                 sizetmp = 1ULL << ctzl(mem_start);
484             }
485 
486             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
487             node_size -= sizetmp;
488             mem_start += sizetmp;
489         }
490     }
491 
492     return 0;
493 }
494 
495 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
496                                   sPAPRMachineState *spapr)
497 {
498     PowerPCCPU *cpu = POWERPC_CPU(cs);
499     CPUPPCState *env = &cpu->env;
500     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
501     int index = spapr_get_vcpu_id(cpu);
502     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
503                        0xffffffff, 0xffffffff};
504     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
505         : SPAPR_TIMEBASE_FREQ;
506     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
507     uint32_t page_sizes_prop[64];
508     size_t page_sizes_prop_size;
509     uint32_t vcpus_per_socket = smp_threads * smp_cores;
510     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
511     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
512     sPAPRDRConnector *drc;
513     int drc_index;
514     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
515     int i;
516 
517     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
518     if (drc) {
519         drc_index = spapr_drc_index(drc);
520         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
521     }
522 
523     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
524     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
525 
526     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
527     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
528                            env->dcache_line_size)));
529     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
530                            env->dcache_line_size)));
531     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
532                            env->icache_line_size)));
533     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
534                            env->icache_line_size)));
535 
536     if (pcc->l1_dcache_size) {
537         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
538                                pcc->l1_dcache_size)));
539     } else {
540         warn_report("Unknown L1 dcache size for cpu");
541     }
542     if (pcc->l1_icache_size) {
543         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
544                                pcc->l1_icache_size)));
545     } else {
546         warn_report("Unknown L1 icache size for cpu");
547     }
548 
549     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
550     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
551     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
552     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
553     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
554     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
555 
556     if (env->spr_cb[SPR_PURR].oea_read) {
557         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
558     }
559 
560     if (env->mmu_model & POWERPC_MMU_1TSEG) {
561         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
562                           segs, sizeof(segs))));
563     }
564 
565     /* Advertise VSX (vector extensions) if available
566      *   1               == VMX / Altivec available
567      *   2               == VSX available
568      *
569      * Only CPUs for which we create core types in spapr_cpu_core.c
570      * are possible, and all of those have VMX */
571     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
572         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
573     } else {
574         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
575     }
576 
577     /* Advertise DFP (Decimal Floating Point) if available
578      *   0 / no property == no DFP
579      *   1               == DFP available */
580     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
581         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
582     }
583 
584     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
585                                                   sizeof(page_sizes_prop));
586     if (page_sizes_prop_size) {
587         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
588                           page_sizes_prop, page_sizes_prop_size)));
589     }
590 
591     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
592 
593     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
594                            cs->cpu_index / vcpus_per_socket)));
595 
596     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
597                       pft_size_prop, sizeof(pft_size_prop))));
598 
599     if (nb_numa_nodes > 1) {
600         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
601     }
602 
603     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
604 
605     if (pcc->radix_page_info) {
606         for (i = 0; i < pcc->radix_page_info->count; i++) {
607             radix_AP_encodings[i] =
608                 cpu_to_be32(pcc->radix_page_info->entries[i]);
609         }
610         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
611                           radix_AP_encodings,
612                           pcc->radix_page_info->count *
613                           sizeof(radix_AP_encodings[0]))));
614     }
615 }
616 
617 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
618 {
619     CPUState *cs;
620     int cpus_offset;
621     char *nodename;
622 
623     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
624     _FDT(cpus_offset);
625     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
626     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
627 
628     /*
629      * We walk the CPUs in reverse order to ensure that CPU DT nodes
630      * created by fdt_add_subnode() end up in the right order in FDT
631      * for the guest kernel the enumerate the CPUs correctly.
632      */
633     CPU_FOREACH_REVERSE(cs) {
634         PowerPCCPU *cpu = POWERPC_CPU(cs);
635         int index = spapr_get_vcpu_id(cpu);
636         DeviceClass *dc = DEVICE_GET_CLASS(cs);
637         int offset;
638 
639         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
640             continue;
641         }
642 
643         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
644         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
645         g_free(nodename);
646         _FDT(offset);
647         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
648     }
649 
650 }
651 
652 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
653 {
654     MemoryDeviceInfoList *info;
655 
656     for (info = list; info; info = info->next) {
657         MemoryDeviceInfo *value = info->value;
658 
659         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
660             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
661 
662             if (pcdimm_info->addr >= addr &&
663                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
664                 return pcdimm_info->node;
665             }
666         }
667     }
668 
669     return -1;
670 }
671 
672 /*
673  * Adds ibm,dynamic-reconfiguration-memory node.
674  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
675  * of this device tree node.
676  */
677 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
678 {
679     MachineState *machine = MACHINE(spapr);
680     int ret, i, offset;
681     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
682     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
683     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
684     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
685                        memory_region_size(&spapr->hotplug_memory.mr)) /
686                        lmb_size;
687     uint32_t *int_buf, *cur_index, buf_len;
688     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
689     MemoryDeviceInfoList *dimms = NULL;
690 
691     /*
692      * Don't create the node if there is no hotpluggable memory
693      */
694     if (machine->ram_size == machine->maxram_size) {
695         return 0;
696     }
697 
698     /*
699      * Allocate enough buffer size to fit in ibm,dynamic-memory
700      * or ibm,associativity-lookup-arrays
701      */
702     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
703               * sizeof(uint32_t);
704     cur_index = int_buf = g_malloc0(buf_len);
705 
706     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
707 
708     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
709                     sizeof(prop_lmb_size));
710     if (ret < 0) {
711         goto out;
712     }
713 
714     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
715     if (ret < 0) {
716         goto out;
717     }
718 
719     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
720     if (ret < 0) {
721         goto out;
722     }
723 
724     if (hotplug_lmb_start) {
725         MemoryDeviceInfoList **prev = &dimms;
726         qmp_pc_dimm_device_list(qdev_get_machine(), &prev);
727     }
728 
729     /* ibm,dynamic-memory */
730     int_buf[0] = cpu_to_be32(nr_lmbs);
731     cur_index++;
732     for (i = 0; i < nr_lmbs; i++) {
733         uint64_t addr = i * lmb_size;
734         uint32_t *dynamic_memory = cur_index;
735 
736         if (i >= hotplug_lmb_start) {
737             sPAPRDRConnector *drc;
738 
739             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
740             g_assert(drc);
741 
742             dynamic_memory[0] = cpu_to_be32(addr >> 32);
743             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
744             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
745             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
746             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
747             if (memory_region_present(get_system_memory(), addr)) {
748                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
749             } else {
750                 dynamic_memory[5] = cpu_to_be32(0);
751             }
752         } else {
753             /*
754              * LMB information for RMA, boot time RAM and gap b/n RAM and
755              * hotplug memory region -- all these are marked as reserved
756              * and as having no valid DRC.
757              */
758             dynamic_memory[0] = cpu_to_be32(addr >> 32);
759             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
760             dynamic_memory[2] = cpu_to_be32(0);
761             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
762             dynamic_memory[4] = cpu_to_be32(-1);
763             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
764                                             SPAPR_LMB_FLAGS_DRC_INVALID);
765         }
766 
767         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
768     }
769     qapi_free_MemoryDeviceInfoList(dimms);
770     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
771     if (ret < 0) {
772         goto out;
773     }
774 
775     /* ibm,associativity-lookup-arrays */
776     cur_index = int_buf;
777     int_buf[0] = cpu_to_be32(nr_nodes);
778     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
779     cur_index += 2;
780     for (i = 0; i < nr_nodes; i++) {
781         uint32_t associativity[] = {
782             cpu_to_be32(0x0),
783             cpu_to_be32(0x0),
784             cpu_to_be32(0x0),
785             cpu_to_be32(i)
786         };
787         memcpy(cur_index, associativity, sizeof(associativity));
788         cur_index += 4;
789     }
790     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
791             (cur_index - int_buf) * sizeof(uint32_t));
792 out:
793     g_free(int_buf);
794     return ret;
795 }
796 
797 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
798                                 sPAPROptionVector *ov5_updates)
799 {
800     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
801     int ret = 0, offset;
802 
803     /* Generate ibm,dynamic-reconfiguration-memory node if required */
804     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
805         g_assert(smc->dr_lmb_enabled);
806         ret = spapr_populate_drconf_memory(spapr, fdt);
807         if (ret) {
808             goto out;
809         }
810     }
811 
812     offset = fdt_path_offset(fdt, "/chosen");
813     if (offset < 0) {
814         offset = fdt_add_subnode(fdt, 0, "chosen");
815         if (offset < 0) {
816             return offset;
817         }
818     }
819     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
820                                  "ibm,architecture-vec-5");
821 
822 out:
823     return ret;
824 }
825 
826 static bool spapr_hotplugged_dev_before_cas(void)
827 {
828     Object *drc_container, *obj;
829     ObjectProperty *prop;
830     ObjectPropertyIterator iter;
831 
832     drc_container = container_get(object_get_root(), "/dr-connector");
833     object_property_iter_init(&iter, drc_container);
834     while ((prop = object_property_iter_next(&iter))) {
835         if (!strstart(prop->type, "link<", NULL)) {
836             continue;
837         }
838         obj = object_property_get_link(drc_container, prop->name, NULL);
839         if (spapr_drc_needed(obj)) {
840             return true;
841         }
842     }
843     return false;
844 }
845 
846 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
847                                  target_ulong addr, target_ulong size,
848                                  sPAPROptionVector *ov5_updates)
849 {
850     void *fdt, *fdt_skel;
851     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
852 
853     if (spapr_hotplugged_dev_before_cas()) {
854         return 1;
855     }
856 
857     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
858         error_report("SLOF provided an unexpected CAS buffer size "
859                      TARGET_FMT_lu " (min: %zu, max: %u)",
860                      size, sizeof(hdr), FW_MAX_SIZE);
861         exit(EXIT_FAILURE);
862     }
863 
864     size -= sizeof(hdr);
865 
866     /* Create skeleton */
867     fdt_skel = g_malloc0(size);
868     _FDT((fdt_create(fdt_skel, size)));
869     _FDT((fdt_begin_node(fdt_skel, "")));
870     _FDT((fdt_end_node(fdt_skel)));
871     _FDT((fdt_finish(fdt_skel)));
872     fdt = g_malloc0(size);
873     _FDT((fdt_open_into(fdt_skel, fdt, size)));
874     g_free(fdt_skel);
875 
876     /* Fixup cpu nodes */
877     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
878 
879     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
880         return -1;
881     }
882 
883     /* Pack resulting tree */
884     _FDT((fdt_pack(fdt)));
885 
886     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
887         trace_spapr_cas_failed(size);
888         return -1;
889     }
890 
891     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
892     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
893     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
894     g_free(fdt);
895 
896     return 0;
897 }
898 
899 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
900 {
901     int rtas;
902     GString *hypertas = g_string_sized_new(256);
903     GString *qemu_hypertas = g_string_sized_new(256);
904     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
905     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
906         memory_region_size(&spapr->hotplug_memory.mr);
907     uint32_t lrdr_capacity[] = {
908         cpu_to_be32(max_hotplug_addr >> 32),
909         cpu_to_be32(max_hotplug_addr & 0xffffffff),
910         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
911         cpu_to_be32(max_cpus / smp_threads),
912     };
913 
914     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
915 
916     /* hypertas */
917     add_str(hypertas, "hcall-pft");
918     add_str(hypertas, "hcall-term");
919     add_str(hypertas, "hcall-dabr");
920     add_str(hypertas, "hcall-interrupt");
921     add_str(hypertas, "hcall-tce");
922     add_str(hypertas, "hcall-vio");
923     add_str(hypertas, "hcall-splpar");
924     add_str(hypertas, "hcall-bulk");
925     add_str(hypertas, "hcall-set-mode");
926     add_str(hypertas, "hcall-sprg0");
927     add_str(hypertas, "hcall-copy");
928     add_str(hypertas, "hcall-debug");
929     add_str(qemu_hypertas, "hcall-memop1");
930 
931     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
932         add_str(hypertas, "hcall-multi-tce");
933     }
934 
935     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
936         add_str(hypertas, "hcall-hpt-resize");
937     }
938 
939     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
940                      hypertas->str, hypertas->len));
941     g_string_free(hypertas, TRUE);
942     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
943                      qemu_hypertas->str, qemu_hypertas->len));
944     g_string_free(qemu_hypertas, TRUE);
945 
946     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
947                      refpoints, sizeof(refpoints)));
948 
949     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
950                           RTAS_ERROR_LOG_MAX));
951     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
952                           RTAS_EVENT_SCAN_RATE));
953 
954     g_assert(msi_nonbroken);
955     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
956 
957     /*
958      * According to PAPR, rtas ibm,os-term does not guarantee a return
959      * back to the guest cpu.
960      *
961      * While an additional ibm,extended-os-term property indicates
962      * that rtas call return will always occur. Set this property.
963      */
964     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
965 
966     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
967                      lrdr_capacity, sizeof(lrdr_capacity)));
968 
969     spapr_dt_rtas_tokens(fdt, rtas);
970 }
971 
972 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
973  * that the guest may request and thus the valid values for bytes 24..26 of
974  * option vector 5: */
975 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
976 {
977     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
978 
979     char val[2 * 4] = {
980         23, 0x00, /* Xive mode, filled in below. */
981         24, 0x00, /* Hash/Radix, filled in below. */
982         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
983         26, 0x40, /* Radix options: GTSE == yes. */
984     };
985 
986     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
987                           first_ppc_cpu->compat_pvr)) {
988         /* If we're in a pre POWER9 compat mode then the guest should do hash */
989         val[3] = 0x00; /* Hash */
990     } else if (kvm_enabled()) {
991         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
992             val[3] = 0x80; /* OV5_MMU_BOTH */
993         } else if (kvmppc_has_cap_mmu_radix()) {
994             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
995         } else {
996             val[3] = 0x00; /* Hash */
997         }
998     } else {
999         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1000         val[3] = 0xC0;
1001     }
1002     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1003                      val, sizeof(val)));
1004 }
1005 
1006 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1007 {
1008     MachineState *machine = MACHINE(spapr);
1009     int chosen;
1010     const char *boot_device = machine->boot_order;
1011     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1012     size_t cb = 0;
1013     char *bootlist = get_boot_devices_list(&cb, true);
1014 
1015     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1016 
1017     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1018     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1019                           spapr->initrd_base));
1020     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1021                           spapr->initrd_base + spapr->initrd_size));
1022 
1023     if (spapr->kernel_size) {
1024         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1025                               cpu_to_be64(spapr->kernel_size) };
1026 
1027         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1028                          &kprop, sizeof(kprop)));
1029         if (spapr->kernel_le) {
1030             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1031         }
1032     }
1033     if (boot_menu) {
1034         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1035     }
1036     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1037     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1038     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1039 
1040     if (cb && bootlist) {
1041         int i;
1042 
1043         for (i = 0; i < cb; i++) {
1044             if (bootlist[i] == '\n') {
1045                 bootlist[i] = ' ';
1046             }
1047         }
1048         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1049     }
1050 
1051     if (boot_device && strlen(boot_device)) {
1052         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1053     }
1054 
1055     if (!spapr->has_graphics && stdout_path) {
1056         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1057     }
1058 
1059     spapr_dt_ov5_platform_support(fdt, chosen);
1060 
1061     g_free(stdout_path);
1062     g_free(bootlist);
1063 }
1064 
1065 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1066 {
1067     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1068      * KVM to work under pHyp with some guest co-operation */
1069     int hypervisor;
1070     uint8_t hypercall[16];
1071 
1072     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1073     /* indicate KVM hypercall interface */
1074     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1075     if (kvmppc_has_cap_fixup_hcalls()) {
1076         /*
1077          * Older KVM versions with older guest kernels were broken
1078          * with the magic page, don't allow the guest to map it.
1079          */
1080         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1081                                   sizeof(hypercall))) {
1082             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1083                              hypercall, sizeof(hypercall)));
1084         }
1085     }
1086 }
1087 
1088 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1089                              hwaddr rtas_addr,
1090                              hwaddr rtas_size)
1091 {
1092     MachineState *machine = MACHINE(spapr);
1093     MachineClass *mc = MACHINE_GET_CLASS(machine);
1094     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1095     int ret;
1096     void *fdt;
1097     sPAPRPHBState *phb;
1098     char *buf;
1099 
1100     fdt = g_malloc0(FDT_MAX_SIZE);
1101     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1102 
1103     /* Root node */
1104     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1105     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1106     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1107 
1108     /*
1109      * Add info to guest to indentify which host is it being run on
1110      * and what is the uuid of the guest
1111      */
1112     if (kvmppc_get_host_model(&buf)) {
1113         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1114         g_free(buf);
1115     }
1116     if (kvmppc_get_host_serial(&buf)) {
1117         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1118         g_free(buf);
1119     }
1120 
1121     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1122 
1123     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1124     if (qemu_uuid_set) {
1125         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1126     }
1127     g_free(buf);
1128 
1129     if (qemu_get_vm_name()) {
1130         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1131                                 qemu_get_vm_name()));
1132     }
1133 
1134     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1135     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1136 
1137     /* /interrupt controller */
1138     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1139 
1140     ret = spapr_populate_memory(spapr, fdt);
1141     if (ret < 0) {
1142         error_report("couldn't setup memory nodes in fdt");
1143         exit(1);
1144     }
1145 
1146     /* /vdevice */
1147     spapr_dt_vdevice(spapr->vio_bus, fdt);
1148 
1149     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1150         ret = spapr_rng_populate_dt(fdt);
1151         if (ret < 0) {
1152             error_report("could not set up rng device in the fdt");
1153             exit(1);
1154         }
1155     }
1156 
1157     QLIST_FOREACH(phb, &spapr->phbs, list) {
1158         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1159         if (ret < 0) {
1160             error_report("couldn't setup PCI devices in fdt");
1161             exit(1);
1162         }
1163     }
1164 
1165     /* cpus */
1166     spapr_populate_cpus_dt_node(fdt, spapr);
1167 
1168     if (smc->dr_lmb_enabled) {
1169         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1170     }
1171 
1172     if (mc->has_hotpluggable_cpus) {
1173         int offset = fdt_path_offset(fdt, "/cpus");
1174         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1175                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1176         if (ret < 0) {
1177             error_report("Couldn't set up CPU DR device tree properties");
1178             exit(1);
1179         }
1180     }
1181 
1182     /* /event-sources */
1183     spapr_dt_events(spapr, fdt);
1184 
1185     /* /rtas */
1186     spapr_dt_rtas(spapr, fdt);
1187 
1188     /* /chosen */
1189     spapr_dt_chosen(spapr, fdt);
1190 
1191     /* /hypervisor */
1192     if (kvm_enabled()) {
1193         spapr_dt_hypervisor(spapr, fdt);
1194     }
1195 
1196     /* Build memory reserve map */
1197     if (spapr->kernel_size) {
1198         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1199     }
1200     if (spapr->initrd_size) {
1201         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1202     }
1203 
1204     /* ibm,client-architecture-support updates */
1205     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1206     if (ret < 0) {
1207         error_report("couldn't setup CAS properties fdt");
1208         exit(1);
1209     }
1210 
1211     return fdt;
1212 }
1213 
1214 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1215 {
1216     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1217 }
1218 
1219 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1220                                     PowerPCCPU *cpu)
1221 {
1222     CPUPPCState *env = &cpu->env;
1223 
1224     /* The TCG path should also be holding the BQL at this point */
1225     g_assert(qemu_mutex_iothread_locked());
1226 
1227     if (msr_pr) {
1228         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1229         env->gpr[3] = H_PRIVILEGE;
1230     } else {
1231         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1232     }
1233 }
1234 
1235 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1236 {
1237     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1238 
1239     return spapr->patb_entry;
1240 }
1241 
1242 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1243 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1244 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1245 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1246 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1247 
1248 /*
1249  * Get the fd to access the kernel htab, re-opening it if necessary
1250  */
1251 static int get_htab_fd(sPAPRMachineState *spapr)
1252 {
1253     Error *local_err = NULL;
1254 
1255     if (spapr->htab_fd >= 0) {
1256         return spapr->htab_fd;
1257     }
1258 
1259     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1260     if (spapr->htab_fd < 0) {
1261         error_report_err(local_err);
1262     }
1263 
1264     return spapr->htab_fd;
1265 }
1266 
1267 void close_htab_fd(sPAPRMachineState *spapr)
1268 {
1269     if (spapr->htab_fd >= 0) {
1270         close(spapr->htab_fd);
1271     }
1272     spapr->htab_fd = -1;
1273 }
1274 
1275 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1276 {
1277     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1278 
1279     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1280 }
1281 
1282 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1283 {
1284     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1285 
1286     assert(kvm_enabled());
1287 
1288     if (!spapr->htab) {
1289         return 0;
1290     }
1291 
1292     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1293 }
1294 
1295 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1296                                                 hwaddr ptex, int n)
1297 {
1298     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1299     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1300 
1301     if (!spapr->htab) {
1302         /*
1303          * HTAB is controlled by KVM. Fetch into temporary buffer
1304          */
1305         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1306         kvmppc_read_hptes(hptes, ptex, n);
1307         return hptes;
1308     }
1309 
1310     /*
1311      * HTAB is controlled by QEMU. Just point to the internally
1312      * accessible PTEG.
1313      */
1314     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1315 }
1316 
1317 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1318                               const ppc_hash_pte64_t *hptes,
1319                               hwaddr ptex, int n)
1320 {
1321     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1322 
1323     if (!spapr->htab) {
1324         g_free((void *)hptes);
1325     }
1326 
1327     /* Nothing to do for qemu managed HPT */
1328 }
1329 
1330 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1331                              uint64_t pte0, uint64_t pte1)
1332 {
1333     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1334     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1335 
1336     if (!spapr->htab) {
1337         kvmppc_write_hpte(ptex, pte0, pte1);
1338     } else {
1339         stq_p(spapr->htab + offset, pte0);
1340         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1341     }
1342 }
1343 
1344 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1345 {
1346     int shift;
1347 
1348     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1349      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1350      * that's much more than is needed for Linux guests */
1351     shift = ctz64(pow2ceil(ramsize)) - 7;
1352     shift = MAX(shift, 18); /* Minimum architected size */
1353     shift = MIN(shift, 46); /* Maximum architected size */
1354     return shift;
1355 }
1356 
1357 void spapr_free_hpt(sPAPRMachineState *spapr)
1358 {
1359     g_free(spapr->htab);
1360     spapr->htab = NULL;
1361     spapr->htab_shift = 0;
1362     close_htab_fd(spapr);
1363 }
1364 
1365 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1366                           Error **errp)
1367 {
1368     long rc;
1369 
1370     /* Clean up any HPT info from a previous boot */
1371     spapr_free_hpt(spapr);
1372 
1373     rc = kvmppc_reset_htab(shift);
1374     if (rc < 0) {
1375         /* kernel-side HPT needed, but couldn't allocate one */
1376         error_setg_errno(errp, errno,
1377                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1378                          shift);
1379         /* This is almost certainly fatal, but if the caller really
1380          * wants to carry on with shift == 0, it's welcome to try */
1381     } else if (rc > 0) {
1382         /* kernel-side HPT allocated */
1383         if (rc != shift) {
1384             error_setg(errp,
1385                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1386                        shift, rc);
1387         }
1388 
1389         spapr->htab_shift = shift;
1390         spapr->htab = NULL;
1391     } else {
1392         /* kernel-side HPT not needed, allocate in userspace instead */
1393         size_t size = 1ULL << shift;
1394         int i;
1395 
1396         spapr->htab = qemu_memalign(size, size);
1397         if (!spapr->htab) {
1398             error_setg_errno(errp, errno,
1399                              "Could not allocate HPT of order %d", shift);
1400             return;
1401         }
1402 
1403         memset(spapr->htab, 0, size);
1404         spapr->htab_shift = shift;
1405 
1406         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1407             DIRTY_HPTE(HPTE(spapr->htab, i));
1408         }
1409     }
1410     /* We're setting up a hash table, so that means we're not radix */
1411     spapr->patb_entry = 0;
1412 }
1413 
1414 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1415 {
1416     int hpt_shift;
1417 
1418     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1419         || (spapr->cas_reboot
1420             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1421         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1422     } else {
1423         uint64_t current_ram_size;
1424 
1425         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1426         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1427     }
1428     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1429 
1430     if (spapr->vrma_adjust) {
1431         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1432                                           spapr->htab_shift);
1433     }
1434 }
1435 
1436 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1437 {
1438     bool matched = false;
1439 
1440     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1441         matched = true;
1442     }
1443 
1444     if (!matched) {
1445         error_report("Device %s is not supported by this machine yet.",
1446                      qdev_fw_name(DEVICE(sbdev)));
1447         exit(1);
1448     }
1449 }
1450 
1451 static int spapr_reset_drcs(Object *child, void *opaque)
1452 {
1453     sPAPRDRConnector *drc =
1454         (sPAPRDRConnector *) object_dynamic_cast(child,
1455                                                  TYPE_SPAPR_DR_CONNECTOR);
1456 
1457     if (drc) {
1458         spapr_drc_reset(drc);
1459     }
1460 
1461     return 0;
1462 }
1463 
1464 static void spapr_machine_reset(void)
1465 {
1466     MachineState *machine = MACHINE(qdev_get_machine());
1467     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1468     PowerPCCPU *first_ppc_cpu;
1469     uint32_t rtas_limit;
1470     hwaddr rtas_addr, fdt_addr;
1471     void *fdt;
1472     int rc;
1473 
1474     /* Check for unknown sysbus devices */
1475     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1476 
1477     spapr_caps_reset(spapr);
1478 
1479     first_ppc_cpu = POWERPC_CPU(first_cpu);
1480     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1481         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1482                          spapr->max_compat_pvr)) {
1483         /* If using KVM with radix mode available, VCPUs can be started
1484          * without a HPT because KVM will start them in radix mode.
1485          * Set the GR bit in PATB so that we know there is no HPT. */
1486         spapr->patb_entry = PATBE1_GR;
1487     } else {
1488         spapr_setup_hpt_and_vrma(spapr);
1489     }
1490 
1491     /* if this reset wasn't generated by CAS, we should reset our
1492      * negotiated options and start from scratch */
1493     if (!spapr->cas_reboot) {
1494         spapr_ovec_cleanup(spapr->ov5_cas);
1495         spapr->ov5_cas = spapr_ovec_new();
1496 
1497         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1498     }
1499 
1500     qemu_devices_reset();
1501 
1502     /* DRC reset may cause a device to be unplugged. This will cause troubles
1503      * if this device is used by another device (eg, a running vhost backend
1504      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1505      * situations, we reset DRCs after all devices have been reset.
1506      */
1507     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1508 
1509     spapr_clear_pending_events(spapr);
1510 
1511     /*
1512      * We place the device tree and RTAS just below either the top of the RMA,
1513      * or just below 2GB, whichever is lowere, so that it can be
1514      * processed with 32-bit real mode code if necessary
1515      */
1516     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1517     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1518     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1519 
1520     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1521 
1522     spapr_load_rtas(spapr, fdt, rtas_addr);
1523 
1524     rc = fdt_pack(fdt);
1525 
1526     /* Should only fail if we've built a corrupted tree */
1527     assert(rc == 0);
1528 
1529     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1530         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1531                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1532         exit(1);
1533     }
1534 
1535     /* Load the fdt */
1536     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1537     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1538     g_free(fdt);
1539 
1540     /* Set up the entry state */
1541     first_ppc_cpu->env.gpr[3] = fdt_addr;
1542     first_ppc_cpu->env.gpr[5] = 0;
1543     first_cpu->halted = 0;
1544     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1545 
1546     spapr->cas_reboot = false;
1547 }
1548 
1549 static void spapr_create_nvram(sPAPRMachineState *spapr)
1550 {
1551     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1552     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1553 
1554     if (dinfo) {
1555         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1556                             &error_fatal);
1557     }
1558 
1559     qdev_init_nofail(dev);
1560 
1561     spapr->nvram = (struct sPAPRNVRAM *)dev;
1562 }
1563 
1564 static void spapr_rtc_create(sPAPRMachineState *spapr)
1565 {
1566     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1567     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1568                               &error_fatal);
1569     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1570                               &error_fatal);
1571     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1572                               "date", &error_fatal);
1573 }
1574 
1575 /* Returns whether we want to use VGA or not */
1576 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1577 {
1578     switch (vga_interface_type) {
1579     case VGA_NONE:
1580         return false;
1581     case VGA_DEVICE:
1582         return true;
1583     case VGA_STD:
1584     case VGA_VIRTIO:
1585         return pci_vga_init(pci_bus) != NULL;
1586     default:
1587         error_setg(errp,
1588                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1589         return false;
1590     }
1591 }
1592 
1593 static int spapr_pre_load(void *opaque)
1594 {
1595     int rc;
1596 
1597     rc = spapr_caps_pre_load(opaque);
1598     if (rc) {
1599         return rc;
1600     }
1601 
1602     return 0;
1603 }
1604 
1605 static int spapr_post_load(void *opaque, int version_id)
1606 {
1607     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1608     int err = 0;
1609 
1610     err = spapr_caps_post_migration(spapr);
1611     if (err) {
1612         return err;
1613     }
1614 
1615     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1616         CPUState *cs;
1617         CPU_FOREACH(cs) {
1618             PowerPCCPU *cpu = POWERPC_CPU(cs);
1619             icp_resend(ICP(cpu->intc));
1620         }
1621     }
1622 
1623     /* In earlier versions, there was no separate qdev for the PAPR
1624      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1625      * So when migrating from those versions, poke the incoming offset
1626      * value into the RTC device */
1627     if (version_id < 3) {
1628         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1629     }
1630 
1631     if (kvm_enabled() && spapr->patb_entry) {
1632         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1633         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1634         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1635 
1636         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1637         if (err) {
1638             error_report("Process table config unsupported by the host");
1639             return -EINVAL;
1640         }
1641     }
1642 
1643     return err;
1644 }
1645 
1646 static int spapr_pre_save(void *opaque)
1647 {
1648     int rc;
1649 
1650     rc = spapr_caps_pre_save(opaque);
1651     if (rc) {
1652         return rc;
1653     }
1654 
1655     return 0;
1656 }
1657 
1658 static bool version_before_3(void *opaque, int version_id)
1659 {
1660     return version_id < 3;
1661 }
1662 
1663 static bool spapr_pending_events_needed(void *opaque)
1664 {
1665     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1666     return !QTAILQ_EMPTY(&spapr->pending_events);
1667 }
1668 
1669 static const VMStateDescription vmstate_spapr_event_entry = {
1670     .name = "spapr_event_log_entry",
1671     .version_id = 1,
1672     .minimum_version_id = 1,
1673     .fields = (VMStateField[]) {
1674         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1675         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1676         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1677                                      NULL, extended_length),
1678         VMSTATE_END_OF_LIST()
1679     },
1680 };
1681 
1682 static const VMStateDescription vmstate_spapr_pending_events = {
1683     .name = "spapr_pending_events",
1684     .version_id = 1,
1685     .minimum_version_id = 1,
1686     .needed = spapr_pending_events_needed,
1687     .fields = (VMStateField[]) {
1688         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1689                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1690         VMSTATE_END_OF_LIST()
1691     },
1692 };
1693 
1694 static bool spapr_ov5_cas_needed(void *opaque)
1695 {
1696     sPAPRMachineState *spapr = opaque;
1697     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1698     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1699     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1700     bool cas_needed;
1701 
1702     /* Prior to the introduction of sPAPROptionVector, we had two option
1703      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1704      * Both of these options encode machine topology into the device-tree
1705      * in such a way that the now-booted OS should still be able to interact
1706      * appropriately with QEMU regardless of what options were actually
1707      * negotiatied on the source side.
1708      *
1709      * As such, we can avoid migrating the CAS-negotiated options if these
1710      * are the only options available on the current machine/platform.
1711      * Since these are the only options available for pseries-2.7 and
1712      * earlier, this allows us to maintain old->new/new->old migration
1713      * compatibility.
1714      *
1715      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1716      * via default pseries-2.8 machines and explicit command-line parameters.
1717      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1718      * of the actual CAS-negotiated values to continue working properly. For
1719      * example, availability of memory unplug depends on knowing whether
1720      * OV5_HP_EVT was negotiated via CAS.
1721      *
1722      * Thus, for any cases where the set of available CAS-negotiatable
1723      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1724      * include the CAS-negotiated options in the migration stream.
1725      */
1726     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1727     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1728 
1729     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1730      * the mask itself since in the future it's possible "legacy" bits may be
1731      * removed via machine options, which could generate a false positive
1732      * that breaks migration.
1733      */
1734     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1735     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1736 
1737     spapr_ovec_cleanup(ov5_mask);
1738     spapr_ovec_cleanup(ov5_legacy);
1739     spapr_ovec_cleanup(ov5_removed);
1740 
1741     return cas_needed;
1742 }
1743 
1744 static const VMStateDescription vmstate_spapr_ov5_cas = {
1745     .name = "spapr_option_vector_ov5_cas",
1746     .version_id = 1,
1747     .minimum_version_id = 1,
1748     .needed = spapr_ov5_cas_needed,
1749     .fields = (VMStateField[]) {
1750         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1751                                  vmstate_spapr_ovec, sPAPROptionVector),
1752         VMSTATE_END_OF_LIST()
1753     },
1754 };
1755 
1756 static bool spapr_patb_entry_needed(void *opaque)
1757 {
1758     sPAPRMachineState *spapr = opaque;
1759 
1760     return !!spapr->patb_entry;
1761 }
1762 
1763 static const VMStateDescription vmstate_spapr_patb_entry = {
1764     .name = "spapr_patb_entry",
1765     .version_id = 1,
1766     .minimum_version_id = 1,
1767     .needed = spapr_patb_entry_needed,
1768     .fields = (VMStateField[]) {
1769         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1770         VMSTATE_END_OF_LIST()
1771     },
1772 };
1773 
1774 static const VMStateDescription vmstate_spapr = {
1775     .name = "spapr",
1776     .version_id = 3,
1777     .minimum_version_id = 1,
1778     .pre_load = spapr_pre_load,
1779     .post_load = spapr_post_load,
1780     .pre_save = spapr_pre_save,
1781     .fields = (VMStateField[]) {
1782         /* used to be @next_irq */
1783         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1784 
1785         /* RTC offset */
1786         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1787 
1788         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1789         VMSTATE_END_OF_LIST()
1790     },
1791     .subsections = (const VMStateDescription*[]) {
1792         &vmstate_spapr_ov5_cas,
1793         &vmstate_spapr_patb_entry,
1794         &vmstate_spapr_pending_events,
1795         &vmstate_spapr_cap_htm,
1796         &vmstate_spapr_cap_vsx,
1797         &vmstate_spapr_cap_dfp,
1798         &vmstate_spapr_cap_cfpc,
1799         &vmstate_spapr_cap_sbbc,
1800         &vmstate_spapr_cap_ibs,
1801         NULL
1802     }
1803 };
1804 
1805 static int htab_save_setup(QEMUFile *f, void *opaque)
1806 {
1807     sPAPRMachineState *spapr = opaque;
1808 
1809     /* "Iteration" header */
1810     if (!spapr->htab_shift) {
1811         qemu_put_be32(f, -1);
1812     } else {
1813         qemu_put_be32(f, spapr->htab_shift);
1814     }
1815 
1816     if (spapr->htab) {
1817         spapr->htab_save_index = 0;
1818         spapr->htab_first_pass = true;
1819     } else {
1820         if (spapr->htab_shift) {
1821             assert(kvm_enabled());
1822         }
1823     }
1824 
1825 
1826     return 0;
1827 }
1828 
1829 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1830                             int chunkstart, int n_valid, int n_invalid)
1831 {
1832     qemu_put_be32(f, chunkstart);
1833     qemu_put_be16(f, n_valid);
1834     qemu_put_be16(f, n_invalid);
1835     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1836                     HASH_PTE_SIZE_64 * n_valid);
1837 }
1838 
1839 static void htab_save_end_marker(QEMUFile *f)
1840 {
1841     qemu_put_be32(f, 0);
1842     qemu_put_be16(f, 0);
1843     qemu_put_be16(f, 0);
1844 }
1845 
1846 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1847                                  int64_t max_ns)
1848 {
1849     bool has_timeout = max_ns != -1;
1850     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1851     int index = spapr->htab_save_index;
1852     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1853 
1854     assert(spapr->htab_first_pass);
1855 
1856     do {
1857         int chunkstart;
1858 
1859         /* Consume invalid HPTEs */
1860         while ((index < htabslots)
1861                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1862             CLEAN_HPTE(HPTE(spapr->htab, index));
1863             index++;
1864         }
1865 
1866         /* Consume valid HPTEs */
1867         chunkstart = index;
1868         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1869                && HPTE_VALID(HPTE(spapr->htab, index))) {
1870             CLEAN_HPTE(HPTE(spapr->htab, index));
1871             index++;
1872         }
1873 
1874         if (index > chunkstart) {
1875             int n_valid = index - chunkstart;
1876 
1877             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
1878 
1879             if (has_timeout &&
1880                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1881                 break;
1882             }
1883         }
1884     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1885 
1886     if (index >= htabslots) {
1887         assert(index == htabslots);
1888         index = 0;
1889         spapr->htab_first_pass = false;
1890     }
1891     spapr->htab_save_index = index;
1892 }
1893 
1894 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1895                                 int64_t max_ns)
1896 {
1897     bool final = max_ns < 0;
1898     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1899     int examined = 0, sent = 0;
1900     int index = spapr->htab_save_index;
1901     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1902 
1903     assert(!spapr->htab_first_pass);
1904 
1905     do {
1906         int chunkstart, invalidstart;
1907 
1908         /* Consume non-dirty HPTEs */
1909         while ((index < htabslots)
1910                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1911             index++;
1912             examined++;
1913         }
1914 
1915         chunkstart = index;
1916         /* Consume valid dirty HPTEs */
1917         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1918                && HPTE_DIRTY(HPTE(spapr->htab, index))
1919                && HPTE_VALID(HPTE(spapr->htab, index))) {
1920             CLEAN_HPTE(HPTE(spapr->htab, index));
1921             index++;
1922             examined++;
1923         }
1924 
1925         invalidstart = index;
1926         /* Consume invalid dirty HPTEs */
1927         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1928                && HPTE_DIRTY(HPTE(spapr->htab, index))
1929                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1930             CLEAN_HPTE(HPTE(spapr->htab, index));
1931             index++;
1932             examined++;
1933         }
1934 
1935         if (index > chunkstart) {
1936             int n_valid = invalidstart - chunkstart;
1937             int n_invalid = index - invalidstart;
1938 
1939             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
1940             sent += index - chunkstart;
1941 
1942             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1943                 break;
1944             }
1945         }
1946 
1947         if (examined >= htabslots) {
1948             break;
1949         }
1950 
1951         if (index >= htabslots) {
1952             assert(index == htabslots);
1953             index = 0;
1954         }
1955     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1956 
1957     if (index >= htabslots) {
1958         assert(index == htabslots);
1959         index = 0;
1960     }
1961 
1962     spapr->htab_save_index = index;
1963 
1964     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1965 }
1966 
1967 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1968 #define MAX_KVM_BUF_SIZE    2048
1969 
1970 static int htab_save_iterate(QEMUFile *f, void *opaque)
1971 {
1972     sPAPRMachineState *spapr = opaque;
1973     int fd;
1974     int rc = 0;
1975 
1976     /* Iteration header */
1977     if (!spapr->htab_shift) {
1978         qemu_put_be32(f, -1);
1979         return 1;
1980     } else {
1981         qemu_put_be32(f, 0);
1982     }
1983 
1984     if (!spapr->htab) {
1985         assert(kvm_enabled());
1986 
1987         fd = get_htab_fd(spapr);
1988         if (fd < 0) {
1989             return fd;
1990         }
1991 
1992         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1993         if (rc < 0) {
1994             return rc;
1995         }
1996     } else  if (spapr->htab_first_pass) {
1997         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1998     } else {
1999         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2000     }
2001 
2002     htab_save_end_marker(f);
2003 
2004     return rc;
2005 }
2006 
2007 static int htab_save_complete(QEMUFile *f, void *opaque)
2008 {
2009     sPAPRMachineState *spapr = opaque;
2010     int fd;
2011 
2012     /* Iteration header */
2013     if (!spapr->htab_shift) {
2014         qemu_put_be32(f, -1);
2015         return 0;
2016     } else {
2017         qemu_put_be32(f, 0);
2018     }
2019 
2020     if (!spapr->htab) {
2021         int rc;
2022 
2023         assert(kvm_enabled());
2024 
2025         fd = get_htab_fd(spapr);
2026         if (fd < 0) {
2027             return fd;
2028         }
2029 
2030         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2031         if (rc < 0) {
2032             return rc;
2033         }
2034     } else {
2035         if (spapr->htab_first_pass) {
2036             htab_save_first_pass(f, spapr, -1);
2037         }
2038         htab_save_later_pass(f, spapr, -1);
2039     }
2040 
2041     /* End marker */
2042     htab_save_end_marker(f);
2043 
2044     return 0;
2045 }
2046 
2047 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2048 {
2049     sPAPRMachineState *spapr = opaque;
2050     uint32_t section_hdr;
2051     int fd = -1;
2052     Error *local_err = NULL;
2053 
2054     if (version_id < 1 || version_id > 1) {
2055         error_report("htab_load() bad version");
2056         return -EINVAL;
2057     }
2058 
2059     section_hdr = qemu_get_be32(f);
2060 
2061     if (section_hdr == -1) {
2062         spapr_free_hpt(spapr);
2063         return 0;
2064     }
2065 
2066     if (section_hdr) {
2067         /* First section gives the htab size */
2068         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2069         if (local_err) {
2070             error_report_err(local_err);
2071             return -EINVAL;
2072         }
2073         return 0;
2074     }
2075 
2076     if (!spapr->htab) {
2077         assert(kvm_enabled());
2078 
2079         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2080         if (fd < 0) {
2081             error_report_err(local_err);
2082             return fd;
2083         }
2084     }
2085 
2086     while (true) {
2087         uint32_t index;
2088         uint16_t n_valid, n_invalid;
2089 
2090         index = qemu_get_be32(f);
2091         n_valid = qemu_get_be16(f);
2092         n_invalid = qemu_get_be16(f);
2093 
2094         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2095             /* End of Stream */
2096             break;
2097         }
2098 
2099         if ((index + n_valid + n_invalid) >
2100             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2101             /* Bad index in stream */
2102             error_report(
2103                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2104                 index, n_valid, n_invalid, spapr->htab_shift);
2105             return -EINVAL;
2106         }
2107 
2108         if (spapr->htab) {
2109             if (n_valid) {
2110                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2111                                 HASH_PTE_SIZE_64 * n_valid);
2112             }
2113             if (n_invalid) {
2114                 memset(HPTE(spapr->htab, index + n_valid), 0,
2115                        HASH_PTE_SIZE_64 * n_invalid);
2116             }
2117         } else {
2118             int rc;
2119 
2120             assert(fd >= 0);
2121 
2122             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2123             if (rc < 0) {
2124                 return rc;
2125             }
2126         }
2127     }
2128 
2129     if (!spapr->htab) {
2130         assert(fd >= 0);
2131         close(fd);
2132     }
2133 
2134     return 0;
2135 }
2136 
2137 static void htab_save_cleanup(void *opaque)
2138 {
2139     sPAPRMachineState *spapr = opaque;
2140 
2141     close_htab_fd(spapr);
2142 }
2143 
2144 static SaveVMHandlers savevm_htab_handlers = {
2145     .save_setup = htab_save_setup,
2146     .save_live_iterate = htab_save_iterate,
2147     .save_live_complete_precopy = htab_save_complete,
2148     .save_cleanup = htab_save_cleanup,
2149     .load_state = htab_load,
2150 };
2151 
2152 static void spapr_boot_set(void *opaque, const char *boot_device,
2153                            Error **errp)
2154 {
2155     MachineState *machine = MACHINE(opaque);
2156     machine->boot_order = g_strdup(boot_device);
2157 }
2158 
2159 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2160 {
2161     MachineState *machine = MACHINE(spapr);
2162     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2163     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2164     int i;
2165 
2166     for (i = 0; i < nr_lmbs; i++) {
2167         uint64_t addr;
2168 
2169         addr = i * lmb_size + spapr->hotplug_memory.base;
2170         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2171                                addr / lmb_size);
2172     }
2173 }
2174 
2175 /*
2176  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2177  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2178  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2179  */
2180 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2181 {
2182     int i;
2183 
2184     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2185         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2186                    " is not aligned to %llu MiB",
2187                    machine->ram_size,
2188                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2189         return;
2190     }
2191 
2192     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2193         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2194                    " is not aligned to %llu MiB",
2195                    machine->ram_size,
2196                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2197         return;
2198     }
2199 
2200     for (i = 0; i < nb_numa_nodes; i++) {
2201         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2202             error_setg(errp,
2203                        "Node %d memory size 0x%" PRIx64
2204                        " is not aligned to %llu MiB",
2205                        i, numa_info[i].node_mem,
2206                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2207             return;
2208         }
2209     }
2210 }
2211 
2212 /* find cpu slot in machine->possible_cpus by core_id */
2213 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2214 {
2215     int index = id / smp_threads;
2216 
2217     if (index >= ms->possible_cpus->len) {
2218         return NULL;
2219     }
2220     if (idx) {
2221         *idx = index;
2222     }
2223     return &ms->possible_cpus->cpus[index];
2224 }
2225 
2226 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2227 {
2228     Error *local_err = NULL;
2229     bool vsmt_user = !!spapr->vsmt;
2230     int kvm_smt = kvmppc_smt_threads();
2231     int ret;
2232 
2233     if (!kvm_enabled() && (smp_threads > 1)) {
2234         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2235                      "on a pseries machine");
2236         goto out;
2237     }
2238     if (!is_power_of_2(smp_threads)) {
2239         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2240                      "machine because it must be a power of 2", smp_threads);
2241         goto out;
2242     }
2243 
2244     /* Detemine the VSMT mode to use: */
2245     if (vsmt_user) {
2246         if (spapr->vsmt < smp_threads) {
2247             error_setg(&local_err, "Cannot support VSMT mode %d"
2248                          " because it must be >= threads/core (%d)",
2249                          spapr->vsmt, smp_threads);
2250             goto out;
2251         }
2252         /* In this case, spapr->vsmt has been set by the command line */
2253     } else {
2254         /*
2255          * Default VSMT value is tricky, because we need it to be as
2256          * consistent as possible (for migration), but this requires
2257          * changing it for at least some existing cases.  We pick 8 as
2258          * the value that we'd get with KVM on POWER8, the
2259          * overwhelmingly common case in production systems.
2260          */
2261         spapr->vsmt = MAX(8, smp_threads);
2262     }
2263 
2264     /* KVM: If necessary, set the SMT mode: */
2265     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2266         ret = kvmppc_set_smt_threads(spapr->vsmt);
2267         if (ret) {
2268             /* Looks like KVM isn't able to change VSMT mode */
2269             error_setg(&local_err,
2270                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2271                        spapr->vsmt, ret);
2272             /* We can live with that if the default one is big enough
2273              * for the number of threads, and a submultiple of the one
2274              * we want.  In this case we'll waste some vcpu ids, but
2275              * behaviour will be correct */
2276             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2277                 warn_report_err(local_err);
2278                 local_err = NULL;
2279                 goto out;
2280             } else {
2281                 if (!vsmt_user) {
2282                     error_append_hint(&local_err,
2283                                       "On PPC, a VM with %d threads/core"
2284                                       " on a host with %d threads/core"
2285                                       " requires the use of VSMT mode %d.\n",
2286                                       smp_threads, kvm_smt, spapr->vsmt);
2287                 }
2288                 kvmppc_hint_smt_possible(&local_err);
2289                 goto out;
2290             }
2291         }
2292     }
2293     /* else TCG: nothing to do currently */
2294 out:
2295     error_propagate(errp, local_err);
2296 }
2297 
2298 static void spapr_init_cpus(sPAPRMachineState *spapr)
2299 {
2300     MachineState *machine = MACHINE(spapr);
2301     MachineClass *mc = MACHINE_GET_CLASS(machine);
2302     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2303     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2304     const CPUArchIdList *possible_cpus;
2305     int boot_cores_nr = smp_cpus / smp_threads;
2306     int i;
2307 
2308     possible_cpus = mc->possible_cpu_arch_ids(machine);
2309     if (mc->has_hotpluggable_cpus) {
2310         if (smp_cpus % smp_threads) {
2311             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2312                          smp_cpus, smp_threads);
2313             exit(1);
2314         }
2315         if (max_cpus % smp_threads) {
2316             error_report("max_cpus (%u) must be multiple of threads (%u)",
2317                          max_cpus, smp_threads);
2318             exit(1);
2319         }
2320     } else {
2321         if (max_cpus != smp_cpus) {
2322             error_report("This machine version does not support CPU hotplug");
2323             exit(1);
2324         }
2325         boot_cores_nr = possible_cpus->len;
2326     }
2327 
2328     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2329      * call xics_max_server_number() or spapr_vcpu_id().
2330      */
2331     spapr_set_vsmt_mode(spapr, &error_fatal);
2332 
2333     if (smc->pre_2_10_has_unused_icps) {
2334         int i;
2335 
2336         for (i = 0; i < xics_max_server_number(spapr); i++) {
2337             /* Dummy entries get deregistered when real ICPState objects
2338              * are registered during CPU core hotplug.
2339              */
2340             pre_2_10_vmstate_register_dummy_icp(i);
2341         }
2342     }
2343 
2344     for (i = 0; i < possible_cpus->len; i++) {
2345         int core_id = i * smp_threads;
2346 
2347         if (mc->has_hotpluggable_cpus) {
2348             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2349                                    spapr_vcpu_id(spapr, core_id));
2350         }
2351 
2352         if (i < boot_cores_nr) {
2353             Object *core  = object_new(type);
2354             int nr_threads = smp_threads;
2355 
2356             /* Handle the partially filled core for older machine types */
2357             if ((i + 1) * smp_threads >= smp_cpus) {
2358                 nr_threads = smp_cpus - i * smp_threads;
2359             }
2360 
2361             object_property_set_int(core, nr_threads, "nr-threads",
2362                                     &error_fatal);
2363             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2364                                     &error_fatal);
2365             object_property_set_bool(core, true, "realized", &error_fatal);
2366         }
2367     }
2368 }
2369 
2370 /* pSeries LPAR / sPAPR hardware init */
2371 static void spapr_machine_init(MachineState *machine)
2372 {
2373     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2374     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2375     const char *kernel_filename = machine->kernel_filename;
2376     const char *initrd_filename = machine->initrd_filename;
2377     PCIHostState *phb;
2378     int i;
2379     MemoryRegion *sysmem = get_system_memory();
2380     MemoryRegion *ram = g_new(MemoryRegion, 1);
2381     MemoryRegion *rma_region;
2382     void *rma = NULL;
2383     hwaddr rma_alloc_size;
2384     hwaddr node0_size = spapr_node0_size(machine);
2385     long load_limit, fw_size;
2386     char *filename;
2387     Error *resize_hpt_err = NULL;
2388 
2389     msi_nonbroken = true;
2390 
2391     QLIST_INIT(&spapr->phbs);
2392     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2393 
2394     /* Check HPT resizing availability */
2395     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2396     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2397         /*
2398          * If the user explicitly requested a mode we should either
2399          * supply it, or fail completely (which we do below).  But if
2400          * it's not set explicitly, we reset our mode to something
2401          * that works
2402          */
2403         if (resize_hpt_err) {
2404             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2405             error_free(resize_hpt_err);
2406             resize_hpt_err = NULL;
2407         } else {
2408             spapr->resize_hpt = smc->resize_hpt_default;
2409         }
2410     }
2411 
2412     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2413 
2414     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2415         /*
2416          * User requested HPT resize, but this host can't supply it.  Bail out
2417          */
2418         error_report_err(resize_hpt_err);
2419         exit(1);
2420     }
2421 
2422     /* Allocate RMA if necessary */
2423     rma_alloc_size = kvmppc_alloc_rma(&rma);
2424 
2425     if (rma_alloc_size == -1) {
2426         error_report("Unable to create RMA");
2427         exit(1);
2428     }
2429 
2430     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2431         spapr->rma_size = rma_alloc_size;
2432     } else {
2433         spapr->rma_size = node0_size;
2434 
2435         /* With KVM, we don't actually know whether KVM supports an
2436          * unbounded RMA (PR KVM) or is limited by the hash table size
2437          * (HV KVM using VRMA), so we always assume the latter
2438          *
2439          * In that case, we also limit the initial allocations for RTAS
2440          * etc... to 256M since we have no way to know what the VRMA size
2441          * is going to be as it depends on the size of the hash table
2442          * isn't determined yet.
2443          */
2444         if (kvm_enabled()) {
2445             spapr->vrma_adjust = 1;
2446             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2447         }
2448 
2449         /* Actually we don't support unbounded RMA anymore since we
2450          * added proper emulation of HV mode. The max we can get is
2451          * 16G which also happens to be what we configure for PAPR
2452          * mode so make sure we don't do anything bigger than that
2453          */
2454         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2455     }
2456 
2457     if (spapr->rma_size > node0_size) {
2458         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2459                      spapr->rma_size);
2460         exit(1);
2461     }
2462 
2463     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2464     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2465 
2466     /* Set up Interrupt Controller before we create the VCPUs */
2467     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2468 
2469     /* Set up containers for ibm,client-architecture-support negotiated options
2470      */
2471     spapr->ov5 = spapr_ovec_new();
2472     spapr->ov5_cas = spapr_ovec_new();
2473 
2474     if (smc->dr_lmb_enabled) {
2475         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2476         spapr_validate_node_memory(machine, &error_fatal);
2477     }
2478 
2479     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2480     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2481         /* KVM and TCG always allow GTSE with radix... */
2482         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2483     }
2484     /* ... but not with hash (currently). */
2485 
2486     /* advertise support for dedicated HP event source to guests */
2487     if (spapr->use_hotplug_event_source) {
2488         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2489     }
2490 
2491     /* advertise support for HPT resizing */
2492     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2493         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2494     }
2495 
2496     /* init CPUs */
2497     spapr_init_cpus(spapr);
2498 
2499     if (kvm_enabled()) {
2500         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2501         kvmppc_enable_logical_ci_hcalls();
2502         kvmppc_enable_set_mode_hcall();
2503 
2504         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2505         kvmppc_enable_clear_ref_mod_hcalls();
2506     }
2507 
2508     /* allocate RAM */
2509     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2510                                          machine->ram_size);
2511     memory_region_add_subregion(sysmem, 0, ram);
2512 
2513     if (rma_alloc_size && rma) {
2514         rma_region = g_new(MemoryRegion, 1);
2515         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2516                                    rma_alloc_size, rma);
2517         vmstate_register_ram_global(rma_region);
2518         memory_region_add_subregion(sysmem, 0, rma_region);
2519     }
2520 
2521     /* initialize hotplug memory address space */
2522     if (machine->ram_size < machine->maxram_size) {
2523         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2524         /*
2525          * Limit the number of hotpluggable memory slots to half the number
2526          * slots that KVM supports, leaving the other half for PCI and other
2527          * devices. However ensure that number of slots doesn't drop below 32.
2528          */
2529         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2530                            SPAPR_MAX_RAM_SLOTS;
2531 
2532         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2533             max_memslots = SPAPR_MAX_RAM_SLOTS;
2534         }
2535         if (machine->ram_slots > max_memslots) {
2536             error_report("Specified number of memory slots %"
2537                          PRIu64" exceeds max supported %d",
2538                          machine->ram_slots, max_memslots);
2539             exit(1);
2540         }
2541 
2542         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2543                                               SPAPR_HOTPLUG_MEM_ALIGN);
2544         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2545                            "hotplug-memory", hotplug_mem_size);
2546         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2547                                     &spapr->hotplug_memory.mr);
2548     }
2549 
2550     if (smc->dr_lmb_enabled) {
2551         spapr_create_lmb_dr_connectors(spapr);
2552     }
2553 
2554     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2555     if (!filename) {
2556         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2557         exit(1);
2558     }
2559     spapr->rtas_size = get_image_size(filename);
2560     if (spapr->rtas_size < 0) {
2561         error_report("Could not get size of LPAR rtas '%s'", filename);
2562         exit(1);
2563     }
2564     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2565     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2566         error_report("Could not load LPAR rtas '%s'", filename);
2567         exit(1);
2568     }
2569     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2570         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2571                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2572         exit(1);
2573     }
2574     g_free(filename);
2575 
2576     /* Set up RTAS event infrastructure */
2577     spapr_events_init(spapr);
2578 
2579     /* Set up the RTC RTAS interfaces */
2580     spapr_rtc_create(spapr);
2581 
2582     /* Set up VIO bus */
2583     spapr->vio_bus = spapr_vio_bus_init();
2584 
2585     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2586         if (serial_hds[i]) {
2587             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2588         }
2589     }
2590 
2591     /* We always have at least the nvram device on VIO */
2592     spapr_create_nvram(spapr);
2593 
2594     /* Set up PCI */
2595     spapr_pci_rtas_init();
2596 
2597     phb = spapr_create_phb(spapr, 0);
2598 
2599     for (i = 0; i < nb_nics; i++) {
2600         NICInfo *nd = &nd_table[i];
2601 
2602         if (!nd->model) {
2603             nd->model = g_strdup("ibmveth");
2604         }
2605 
2606         if (strcmp(nd->model, "ibmveth") == 0) {
2607             spapr_vlan_create(spapr->vio_bus, nd);
2608         } else {
2609             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2610         }
2611     }
2612 
2613     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2614         spapr_vscsi_create(spapr->vio_bus);
2615     }
2616 
2617     /* Graphics */
2618     if (spapr_vga_init(phb->bus, &error_fatal)) {
2619         spapr->has_graphics = true;
2620         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2621     }
2622 
2623     if (machine->usb) {
2624         if (smc->use_ohci_by_default) {
2625             pci_create_simple(phb->bus, -1, "pci-ohci");
2626         } else {
2627             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2628         }
2629 
2630         if (spapr->has_graphics) {
2631             USBBus *usb_bus = usb_bus_find(-1);
2632 
2633             usb_create_simple(usb_bus, "usb-kbd");
2634             usb_create_simple(usb_bus, "usb-mouse");
2635         }
2636     }
2637 
2638     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2639         error_report(
2640             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2641             MIN_RMA_SLOF);
2642         exit(1);
2643     }
2644 
2645     if (kernel_filename) {
2646         uint64_t lowaddr = 0;
2647 
2648         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2649                                       NULL, NULL, &lowaddr, NULL, 1,
2650                                       PPC_ELF_MACHINE, 0, 0);
2651         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2652             spapr->kernel_size = load_elf(kernel_filename,
2653                                           translate_kernel_address, NULL, NULL,
2654                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2655                                           0, 0);
2656             spapr->kernel_le = spapr->kernel_size > 0;
2657         }
2658         if (spapr->kernel_size < 0) {
2659             error_report("error loading %s: %s", kernel_filename,
2660                          load_elf_strerror(spapr->kernel_size));
2661             exit(1);
2662         }
2663 
2664         /* load initrd */
2665         if (initrd_filename) {
2666             /* Try to locate the initrd in the gap between the kernel
2667              * and the firmware. Add a bit of space just in case
2668              */
2669             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2670                                   + 0x1ffff) & ~0xffff;
2671             spapr->initrd_size = load_image_targphys(initrd_filename,
2672                                                      spapr->initrd_base,
2673                                                      load_limit
2674                                                      - spapr->initrd_base);
2675             if (spapr->initrd_size < 0) {
2676                 error_report("could not load initial ram disk '%s'",
2677                              initrd_filename);
2678                 exit(1);
2679             }
2680         }
2681     }
2682 
2683     if (bios_name == NULL) {
2684         bios_name = FW_FILE_NAME;
2685     }
2686     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2687     if (!filename) {
2688         error_report("Could not find LPAR firmware '%s'", bios_name);
2689         exit(1);
2690     }
2691     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2692     if (fw_size <= 0) {
2693         error_report("Could not load LPAR firmware '%s'", filename);
2694         exit(1);
2695     }
2696     g_free(filename);
2697 
2698     /* FIXME: Should register things through the MachineState's qdev
2699      * interface, this is a legacy from the sPAPREnvironment structure
2700      * which predated MachineState but had a similar function */
2701     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2702     register_savevm_live(NULL, "spapr/htab", -1, 1,
2703                          &savevm_htab_handlers, spapr);
2704 
2705     qemu_register_boot_set(spapr_boot_set, spapr);
2706 
2707     if (kvm_enabled()) {
2708         /* to stop and start vmclock */
2709         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2710                                          &spapr->tb);
2711 
2712         kvmppc_spapr_enable_inkernel_multitce();
2713     }
2714 }
2715 
2716 static int spapr_kvm_type(const char *vm_type)
2717 {
2718     if (!vm_type) {
2719         return 0;
2720     }
2721 
2722     if (!strcmp(vm_type, "HV")) {
2723         return 1;
2724     }
2725 
2726     if (!strcmp(vm_type, "PR")) {
2727         return 2;
2728     }
2729 
2730     error_report("Unknown kvm-type specified '%s'", vm_type);
2731     exit(1);
2732 }
2733 
2734 /*
2735  * Implementation of an interface to adjust firmware path
2736  * for the bootindex property handling.
2737  */
2738 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2739                                    DeviceState *dev)
2740 {
2741 #define CAST(type, obj, name) \
2742     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2743     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2744     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2745     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2746 
2747     if (d) {
2748         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2749         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2750         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2751 
2752         if (spapr) {
2753             /*
2754              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2755              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2756              * in the top 16 bits of the 64-bit LUN
2757              */
2758             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2759             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2760                                    (uint64_t)id << 48);
2761         } else if (virtio) {
2762             /*
2763              * We use SRP luns of the form 01000000 | (target << 8) | lun
2764              * in the top 32 bits of the 64-bit LUN
2765              * Note: the quote above is from SLOF and it is wrong,
2766              * the actual binding is:
2767              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2768              */
2769             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2770             if (d->lun >= 256) {
2771                 /* Use the LUN "flat space addressing method" */
2772                 id |= 0x4000;
2773             }
2774             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2775                                    (uint64_t)id << 32);
2776         } else if (usb) {
2777             /*
2778              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2779              * in the top 32 bits of the 64-bit LUN
2780              */
2781             unsigned usb_port = atoi(usb->port->path);
2782             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2783             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2784                                    (uint64_t)id << 32);
2785         }
2786     }
2787 
2788     /*
2789      * SLOF probes the USB devices, and if it recognizes that the device is a
2790      * storage device, it changes its name to "storage" instead of "usb-host",
2791      * and additionally adds a child node for the SCSI LUN, so the correct
2792      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2793      */
2794     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2795         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2796         if (usb_host_dev_is_scsi_storage(usbdev)) {
2797             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2798         }
2799     }
2800 
2801     if (phb) {
2802         /* Replace "pci" with "pci@800000020000000" */
2803         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2804     }
2805 
2806     if (vsc) {
2807         /* Same logic as virtio above */
2808         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2809         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2810     }
2811 
2812     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2813         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2814         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2815         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2816     }
2817 
2818     return NULL;
2819 }
2820 
2821 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2822 {
2823     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2824 
2825     return g_strdup(spapr->kvm_type);
2826 }
2827 
2828 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2829 {
2830     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2831 
2832     g_free(spapr->kvm_type);
2833     spapr->kvm_type = g_strdup(value);
2834 }
2835 
2836 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2837 {
2838     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2839 
2840     return spapr->use_hotplug_event_source;
2841 }
2842 
2843 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2844                                             Error **errp)
2845 {
2846     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2847 
2848     spapr->use_hotplug_event_source = value;
2849 }
2850 
2851 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2852 {
2853     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2854 
2855     switch (spapr->resize_hpt) {
2856     case SPAPR_RESIZE_HPT_DEFAULT:
2857         return g_strdup("default");
2858     case SPAPR_RESIZE_HPT_DISABLED:
2859         return g_strdup("disabled");
2860     case SPAPR_RESIZE_HPT_ENABLED:
2861         return g_strdup("enabled");
2862     case SPAPR_RESIZE_HPT_REQUIRED:
2863         return g_strdup("required");
2864     }
2865     g_assert_not_reached();
2866 }
2867 
2868 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2869 {
2870     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2871 
2872     if (strcmp(value, "default") == 0) {
2873         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2874     } else if (strcmp(value, "disabled") == 0) {
2875         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2876     } else if (strcmp(value, "enabled") == 0) {
2877         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2878     } else if (strcmp(value, "required") == 0) {
2879         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2880     } else {
2881         error_setg(errp, "Bad value for \"resize-hpt\" property");
2882     }
2883 }
2884 
2885 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
2886                                    void *opaque, Error **errp)
2887 {
2888     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2889 }
2890 
2891 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
2892                                    void *opaque, Error **errp)
2893 {
2894     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2895 }
2896 
2897 static void spapr_instance_init(Object *obj)
2898 {
2899     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2900 
2901     spapr->htab_fd = -1;
2902     spapr->use_hotplug_event_source = true;
2903     object_property_add_str(obj, "kvm-type",
2904                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2905     object_property_set_description(obj, "kvm-type",
2906                                     "Specifies the KVM virtualization mode (HV, PR)",
2907                                     NULL);
2908     object_property_add_bool(obj, "modern-hotplug-events",
2909                             spapr_get_modern_hotplug_events,
2910                             spapr_set_modern_hotplug_events,
2911                             NULL);
2912     object_property_set_description(obj, "modern-hotplug-events",
2913                                     "Use dedicated hotplug event mechanism in"
2914                                     " place of standard EPOW events when possible"
2915                                     " (required for memory hot-unplug support)",
2916                                     NULL);
2917 
2918     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
2919                             "Maximum permitted CPU compatibility mode",
2920                             &error_fatal);
2921 
2922     object_property_add_str(obj, "resize-hpt",
2923                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
2924     object_property_set_description(obj, "resize-hpt",
2925                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
2926                                     NULL);
2927     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
2928                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
2929     object_property_set_description(obj, "vsmt",
2930                                     "Virtual SMT: KVM behaves as if this were"
2931                                     " the host's SMT mode", &error_abort);
2932 }
2933 
2934 static void spapr_machine_finalizefn(Object *obj)
2935 {
2936     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2937 
2938     g_free(spapr->kvm_type);
2939 }
2940 
2941 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2942 {
2943     cpu_synchronize_state(cs);
2944     ppc_cpu_do_system_reset(cs);
2945 }
2946 
2947 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2948 {
2949     CPUState *cs;
2950 
2951     CPU_FOREACH(cs) {
2952         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2953     }
2954 }
2955 
2956 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2957                            uint32_t node, bool dedicated_hp_event_source,
2958                            Error **errp)
2959 {
2960     sPAPRDRConnector *drc;
2961     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2962     int i, fdt_offset, fdt_size;
2963     void *fdt;
2964     uint64_t addr = addr_start;
2965     bool hotplugged = spapr_drc_hotplugged(dev);
2966     Error *local_err = NULL;
2967 
2968     for (i = 0; i < nr_lmbs; i++) {
2969         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2970                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2971         g_assert(drc);
2972 
2973         fdt = create_device_tree(&fdt_size);
2974         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2975                                                 SPAPR_MEMORY_BLOCK_SIZE);
2976 
2977         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
2978         if (local_err) {
2979             while (addr > addr_start) {
2980                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
2981                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2982                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
2983                 spapr_drc_detach(drc);
2984             }
2985             g_free(fdt);
2986             error_propagate(errp, local_err);
2987             return;
2988         }
2989         if (!hotplugged) {
2990             spapr_drc_reset(drc);
2991         }
2992         addr += SPAPR_MEMORY_BLOCK_SIZE;
2993     }
2994     /* send hotplug notification to the
2995      * guest only in case of hotplugged memory
2996      */
2997     if (hotplugged) {
2998         if (dedicated_hp_event_source) {
2999             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3000                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3001             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3002                                                    nr_lmbs,
3003                                                    spapr_drc_index(drc));
3004         } else {
3005             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3006                                            nr_lmbs);
3007         }
3008     }
3009 }
3010 
3011 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3012                               uint32_t node, Error **errp)
3013 {
3014     Error *local_err = NULL;
3015     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3016     PCDIMMDevice *dimm = PC_DIMM(dev);
3017     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3018     MemoryRegion *mr;
3019     uint64_t align, size, addr;
3020 
3021     mr = ddc->get_memory_region(dimm, &local_err);
3022     if (local_err) {
3023         goto out;
3024     }
3025     align = memory_region_get_alignment(mr);
3026     size = memory_region_size(mr);
3027 
3028     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
3029     if (local_err) {
3030         goto out;
3031     }
3032 
3033     addr = object_property_get_uint(OBJECT(dimm),
3034                                     PC_DIMM_ADDR_PROP, &local_err);
3035     if (local_err) {
3036         goto out_unplug;
3037     }
3038 
3039     spapr_add_lmbs(dev, addr, size, node,
3040                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3041                    &local_err);
3042     if (local_err) {
3043         goto out_unplug;
3044     }
3045 
3046     return;
3047 
3048 out_unplug:
3049     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
3050 out:
3051     error_propagate(errp, local_err);
3052 }
3053 
3054 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3055                                   Error **errp)
3056 {
3057     PCDIMMDevice *dimm = PC_DIMM(dev);
3058     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3059     MemoryRegion *mr;
3060     uint64_t size;
3061     char *mem_dev;
3062 
3063     mr = ddc->get_memory_region(dimm, errp);
3064     if (!mr) {
3065         return;
3066     }
3067     size = memory_region_size(mr);
3068 
3069     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3070         error_setg(errp, "Hotplugged memory size must be a multiple of "
3071                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
3072         return;
3073     }
3074 
3075     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
3076     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
3077         error_setg(errp, "Memory backend has bad page size. "
3078                    "Use 'memory-backend-file' with correct mem-path.");
3079         goto out;
3080     }
3081 
3082 out:
3083     g_free(mem_dev);
3084 }
3085 
3086 struct sPAPRDIMMState {
3087     PCDIMMDevice *dimm;
3088     uint32_t nr_lmbs;
3089     QTAILQ_ENTRY(sPAPRDIMMState) next;
3090 };
3091 
3092 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3093                                                        PCDIMMDevice *dimm)
3094 {
3095     sPAPRDIMMState *dimm_state = NULL;
3096 
3097     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3098         if (dimm_state->dimm == dimm) {
3099             break;
3100         }
3101     }
3102     return dimm_state;
3103 }
3104 
3105 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3106                                                       uint32_t nr_lmbs,
3107                                                       PCDIMMDevice *dimm)
3108 {
3109     sPAPRDIMMState *ds = NULL;
3110 
3111     /*
3112      * If this request is for a DIMM whose removal had failed earlier
3113      * (due to guest's refusal to remove the LMBs), we would have this
3114      * dimm already in the pending_dimm_unplugs list. In that
3115      * case don't add again.
3116      */
3117     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3118     if (!ds) {
3119         ds = g_malloc0(sizeof(sPAPRDIMMState));
3120         ds->nr_lmbs = nr_lmbs;
3121         ds->dimm = dimm;
3122         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3123     }
3124     return ds;
3125 }
3126 
3127 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3128                                               sPAPRDIMMState *dimm_state)
3129 {
3130     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3131     g_free(dimm_state);
3132 }
3133 
3134 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3135                                                         PCDIMMDevice *dimm)
3136 {
3137     sPAPRDRConnector *drc;
3138     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3139     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3140     uint64_t size = memory_region_size(mr);
3141     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3142     uint32_t avail_lmbs = 0;
3143     uint64_t addr_start, addr;
3144     int i;
3145 
3146     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3147                                          &error_abort);
3148 
3149     addr = addr_start;
3150     for (i = 0; i < nr_lmbs; i++) {
3151         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3152                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3153         g_assert(drc);
3154         if (drc->dev) {
3155             avail_lmbs++;
3156         }
3157         addr += SPAPR_MEMORY_BLOCK_SIZE;
3158     }
3159 
3160     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3161 }
3162 
3163 /* Callback to be called during DRC release. */
3164 void spapr_lmb_release(DeviceState *dev)
3165 {
3166     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_hotplug_handler(dev));
3167     PCDIMMDevice *dimm = PC_DIMM(dev);
3168     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3169     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3170     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3171 
3172     /* This information will get lost if a migration occurs
3173      * during the unplug process. In this case recover it. */
3174     if (ds == NULL) {
3175         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3176         g_assert(ds);
3177         /* The DRC being examined by the caller at least must be counted */
3178         g_assert(ds->nr_lmbs);
3179     }
3180 
3181     if (--ds->nr_lmbs) {
3182         return;
3183     }
3184 
3185     /*
3186      * Now that all the LMBs have been removed by the guest, call the
3187      * pc-dimm unplug handler to cleanup up the pc-dimm device.
3188      */
3189     pc_dimm_memory_unplug(dev, &spapr->hotplug_memory, mr);
3190     object_unparent(OBJECT(dev));
3191     spapr_pending_dimm_unplugs_remove(spapr, ds);
3192 }
3193 
3194 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3195                                         DeviceState *dev, Error **errp)
3196 {
3197     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3198     Error *local_err = NULL;
3199     PCDIMMDevice *dimm = PC_DIMM(dev);
3200     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3201     MemoryRegion *mr;
3202     uint32_t nr_lmbs;
3203     uint64_t size, addr_start, addr;
3204     int i;
3205     sPAPRDRConnector *drc;
3206 
3207     mr = ddc->get_memory_region(dimm, &local_err);
3208     if (local_err) {
3209         goto out;
3210     }
3211     size = memory_region_size(mr);
3212     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3213 
3214     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3215                                          &local_err);
3216     if (local_err) {
3217         goto out;
3218     }
3219 
3220     /*
3221      * An existing pending dimm state for this DIMM means that there is an
3222      * unplug operation in progress, waiting for the spapr_lmb_release
3223      * callback to complete the job (BQL can't cover that far). In this case,
3224      * bail out to avoid detaching DRCs that were already released.
3225      */
3226     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3227         error_setg(&local_err,
3228                    "Memory unplug already in progress for device %s",
3229                    dev->id);
3230         goto out;
3231     }
3232 
3233     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3234 
3235     addr = addr_start;
3236     for (i = 0; i < nr_lmbs; i++) {
3237         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3238                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3239         g_assert(drc);
3240 
3241         spapr_drc_detach(drc);
3242         addr += SPAPR_MEMORY_BLOCK_SIZE;
3243     }
3244 
3245     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3246                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3247     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3248                                               nr_lmbs, spapr_drc_index(drc));
3249 out:
3250     error_propagate(errp, local_err);
3251 }
3252 
3253 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3254                                            sPAPRMachineState *spapr)
3255 {
3256     PowerPCCPU *cpu = POWERPC_CPU(cs);
3257     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3258     int id = spapr_get_vcpu_id(cpu);
3259     void *fdt;
3260     int offset, fdt_size;
3261     char *nodename;
3262 
3263     fdt = create_device_tree(&fdt_size);
3264     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3265     offset = fdt_add_subnode(fdt, 0, nodename);
3266 
3267     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3268     g_free(nodename);
3269 
3270     *fdt_offset = offset;
3271     return fdt;
3272 }
3273 
3274 /* Callback to be called during DRC release. */
3275 void spapr_core_release(DeviceState *dev)
3276 {
3277     MachineState *ms = MACHINE(qdev_get_hotplug_handler(dev));
3278     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3279     CPUCore *cc = CPU_CORE(dev);
3280     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3281 
3282     if (smc->pre_2_10_has_unused_icps) {
3283         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3284         int i;
3285 
3286         for (i = 0; i < cc->nr_threads; i++) {
3287             CPUState *cs = CPU(sc->threads[i]);
3288 
3289             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3290         }
3291     }
3292 
3293     assert(core_slot);
3294     core_slot->cpu = NULL;
3295     object_unparent(OBJECT(dev));
3296 }
3297 
3298 static
3299 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3300                                Error **errp)
3301 {
3302     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3303     int index;
3304     sPAPRDRConnector *drc;
3305     CPUCore *cc = CPU_CORE(dev);
3306 
3307     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3308         error_setg(errp, "Unable to find CPU core with core-id: %d",
3309                    cc->core_id);
3310         return;
3311     }
3312     if (index == 0) {
3313         error_setg(errp, "Boot CPU core may not be unplugged");
3314         return;
3315     }
3316 
3317     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3318                           spapr_vcpu_id(spapr, cc->core_id));
3319     g_assert(drc);
3320 
3321     spapr_drc_detach(drc);
3322 
3323     spapr_hotplug_req_remove_by_index(drc);
3324 }
3325 
3326 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3327                             Error **errp)
3328 {
3329     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3330     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3331     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3332     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3333     CPUCore *cc = CPU_CORE(dev);
3334     CPUState *cs = CPU(core->threads[0]);
3335     sPAPRDRConnector *drc;
3336     Error *local_err = NULL;
3337     CPUArchId *core_slot;
3338     int index;
3339     bool hotplugged = spapr_drc_hotplugged(dev);
3340 
3341     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3342     if (!core_slot) {
3343         error_setg(errp, "Unable to find CPU core with core-id: %d",
3344                    cc->core_id);
3345         return;
3346     }
3347     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3348                           spapr_vcpu_id(spapr, cc->core_id));
3349 
3350     g_assert(drc || !mc->has_hotpluggable_cpus);
3351 
3352     if (drc) {
3353         void *fdt;
3354         int fdt_offset;
3355 
3356         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3357 
3358         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3359         if (local_err) {
3360             g_free(fdt);
3361             error_propagate(errp, local_err);
3362             return;
3363         }
3364 
3365         if (hotplugged) {
3366             /*
3367              * Send hotplug notification interrupt to the guest only
3368              * in case of hotplugged CPUs.
3369              */
3370             spapr_hotplug_req_add_by_index(drc);
3371         } else {
3372             spapr_drc_reset(drc);
3373         }
3374     }
3375 
3376     core_slot->cpu = OBJECT(dev);
3377 
3378     if (smc->pre_2_10_has_unused_icps) {
3379         int i;
3380 
3381         for (i = 0; i < cc->nr_threads; i++) {
3382             cs = CPU(core->threads[i]);
3383             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3384         }
3385     }
3386 }
3387 
3388 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3389                                 Error **errp)
3390 {
3391     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3392     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3393     Error *local_err = NULL;
3394     CPUCore *cc = CPU_CORE(dev);
3395     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3396     const char *type = object_get_typename(OBJECT(dev));
3397     CPUArchId *core_slot;
3398     int index;
3399 
3400     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3401         error_setg(&local_err, "CPU hotplug not supported for this machine");
3402         goto out;
3403     }
3404 
3405     if (strcmp(base_core_type, type)) {
3406         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3407         goto out;
3408     }
3409 
3410     if (cc->core_id % smp_threads) {
3411         error_setg(&local_err, "invalid core id %d", cc->core_id);
3412         goto out;
3413     }
3414 
3415     /*
3416      * In general we should have homogeneous threads-per-core, but old
3417      * (pre hotplug support) machine types allow the last core to have
3418      * reduced threads as a compatibility hack for when we allowed
3419      * total vcpus not a multiple of threads-per-core.
3420      */
3421     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3422         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3423                    cc->nr_threads, smp_threads);
3424         goto out;
3425     }
3426 
3427     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3428     if (!core_slot) {
3429         error_setg(&local_err, "core id %d out of range", cc->core_id);
3430         goto out;
3431     }
3432 
3433     if (core_slot->cpu) {
3434         error_setg(&local_err, "core %d already populated", cc->core_id);
3435         goto out;
3436     }
3437 
3438     numa_cpu_pre_plug(core_slot, dev, &local_err);
3439 
3440 out:
3441     error_propagate(errp, local_err);
3442 }
3443 
3444 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3445                                       DeviceState *dev, Error **errp)
3446 {
3447     MachineState *ms = MACHINE(hotplug_dev);
3448     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3449 
3450     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3451         int node;
3452 
3453         if (!smc->dr_lmb_enabled) {
3454             error_setg(errp, "Memory hotplug not supported for this machine");
3455             return;
3456         }
3457         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3458         if (*errp) {
3459             return;
3460         }
3461         if (node < 0 || node >= MAX_NODES) {
3462             error_setg(errp, "Invaild node %d", node);
3463             return;
3464         }
3465 
3466         /*
3467          * Currently PowerPC kernel doesn't allow hot-adding memory to
3468          * memory-less node, but instead will silently add the memory
3469          * to the first node that has some memory. This causes two
3470          * unexpected behaviours for the user.
3471          *
3472          * - Memory gets hotplugged to a different node than what the user
3473          *   specified.
3474          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3475          *   to memory-less node, a reboot will set things accordingly
3476          *   and the previously hotplugged memory now ends in the right node.
3477          *   This appears as if some memory moved from one node to another.
3478          *
3479          * So until kernel starts supporting memory hotplug to memory-less
3480          * nodes, just prevent such attempts upfront in QEMU.
3481          */
3482         if (nb_numa_nodes && !numa_info[node].node_mem) {
3483             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3484                        node);
3485             return;
3486         }
3487 
3488         spapr_memory_plug(hotplug_dev, dev, node, errp);
3489     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3490         spapr_core_plug(hotplug_dev, dev, errp);
3491     }
3492 }
3493 
3494 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3495                                                 DeviceState *dev, Error **errp)
3496 {
3497     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3498     MachineClass *mc = MACHINE_GET_CLASS(sms);
3499 
3500     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3501         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3502             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3503         } else {
3504             /* NOTE: this means there is a window after guest reset, prior to
3505              * CAS negotiation, where unplug requests will fail due to the
3506              * capability not being detected yet. This is a bit different than
3507              * the case with PCI unplug, where the events will be queued and
3508              * eventually handled by the guest after boot
3509              */
3510             error_setg(errp, "Memory hot unplug not supported for this guest");
3511         }
3512     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3513         if (!mc->has_hotpluggable_cpus) {
3514             error_setg(errp, "CPU hot unplug not supported on this machine");
3515             return;
3516         }
3517         spapr_core_unplug_request(hotplug_dev, dev, errp);
3518     }
3519 }
3520 
3521 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3522                                           DeviceState *dev, Error **errp)
3523 {
3524     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3525         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3526     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3527         spapr_core_pre_plug(hotplug_dev, dev, errp);
3528     }
3529 }
3530 
3531 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3532                                                  DeviceState *dev)
3533 {
3534     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3535         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3536         return HOTPLUG_HANDLER(machine);
3537     }
3538     return NULL;
3539 }
3540 
3541 static CpuInstanceProperties
3542 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3543 {
3544     CPUArchId *core_slot;
3545     MachineClass *mc = MACHINE_GET_CLASS(machine);
3546 
3547     /* make sure possible_cpu are intialized */
3548     mc->possible_cpu_arch_ids(machine);
3549     /* get CPU core slot containing thread that matches cpu_index */
3550     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3551     assert(core_slot);
3552     return core_slot->props;
3553 }
3554 
3555 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3556 {
3557     return idx / smp_cores % nb_numa_nodes;
3558 }
3559 
3560 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3561 {
3562     int i;
3563     const char *core_type;
3564     int spapr_max_cores = max_cpus / smp_threads;
3565     MachineClass *mc = MACHINE_GET_CLASS(machine);
3566 
3567     if (!mc->has_hotpluggable_cpus) {
3568         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3569     }
3570     if (machine->possible_cpus) {
3571         assert(machine->possible_cpus->len == spapr_max_cores);
3572         return machine->possible_cpus;
3573     }
3574 
3575     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3576     if (!core_type) {
3577         error_report("Unable to find sPAPR CPU Core definition");
3578         exit(1);
3579     }
3580 
3581     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3582                              sizeof(CPUArchId) * spapr_max_cores);
3583     machine->possible_cpus->len = spapr_max_cores;
3584     for (i = 0; i < machine->possible_cpus->len; i++) {
3585         int core_id = i * smp_threads;
3586 
3587         machine->possible_cpus->cpus[i].type = core_type;
3588         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3589         machine->possible_cpus->cpus[i].arch_id = core_id;
3590         machine->possible_cpus->cpus[i].props.has_core_id = true;
3591         machine->possible_cpus->cpus[i].props.core_id = core_id;
3592     }
3593     return machine->possible_cpus;
3594 }
3595 
3596 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3597                                 uint64_t *buid, hwaddr *pio,
3598                                 hwaddr *mmio32, hwaddr *mmio64,
3599                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3600 {
3601     /*
3602      * New-style PHB window placement.
3603      *
3604      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3605      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3606      * windows.
3607      *
3608      * Some guest kernels can't work with MMIO windows above 1<<46
3609      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3610      *
3611      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3612      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3613      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3614      * 1TiB 64-bit MMIO windows for each PHB.
3615      */
3616     const uint64_t base_buid = 0x800000020000000ULL;
3617 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3618                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3619     int i;
3620 
3621     /* Sanity check natural alignments */
3622     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3623     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3624     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3625     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3626     /* Sanity check bounds */
3627     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3628                       SPAPR_PCI_MEM32_WIN_SIZE);
3629     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3630                       SPAPR_PCI_MEM64_WIN_SIZE);
3631 
3632     if (index >= SPAPR_MAX_PHBS) {
3633         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3634                    SPAPR_MAX_PHBS - 1);
3635         return;
3636     }
3637 
3638     *buid = base_buid + index;
3639     for (i = 0; i < n_dma; ++i) {
3640         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3641     }
3642 
3643     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3644     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3645     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3646 }
3647 
3648 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3649 {
3650     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3651 
3652     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3653 }
3654 
3655 static void spapr_ics_resend(XICSFabric *dev)
3656 {
3657     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3658 
3659     ics_resend(spapr->ics);
3660 }
3661 
3662 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3663 {
3664     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3665 
3666     return cpu ? ICP(cpu->intc) : NULL;
3667 }
3668 
3669 #define ICS_IRQ_FREE(ics, srcno)   \
3670     (!((ics)->irqs[(srcno)].flags & (XICS_FLAGS_IRQ_MASK)))
3671 
3672 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
3673 {
3674     int first, i;
3675 
3676     for (first = 0; first < ics->nr_irqs; first += alignnum) {
3677         if (num > (ics->nr_irqs - first)) {
3678             return -1;
3679         }
3680         for (i = first; i < first + num; ++i) {
3681             if (!ICS_IRQ_FREE(ics, i)) {
3682                 break;
3683             }
3684         }
3685         if (i == (first + num)) {
3686             return first;
3687         }
3688     }
3689 
3690     return -1;
3691 }
3692 
3693 /*
3694  * Allocate the IRQ number and set the IRQ type, LSI or MSI
3695  */
3696 static void spapr_irq_set_lsi(sPAPRMachineState *spapr, int irq, bool lsi)
3697 {
3698     ics_set_irq_type(spapr->ics, irq - spapr->ics->offset, lsi);
3699 }
3700 
3701 int spapr_irq_alloc(sPAPRMachineState *spapr, int irq_hint, bool lsi,
3702                     Error **errp)
3703 {
3704     ICSState *ics = spapr->ics;
3705     int irq;
3706 
3707     if (!ics) {
3708         return -1;
3709     }
3710     if (irq_hint) {
3711         if (!ICS_IRQ_FREE(ics, irq_hint - ics->offset)) {
3712             error_setg(errp, "can't allocate IRQ %d: already in use", irq_hint);
3713             return -1;
3714         }
3715         irq = irq_hint;
3716     } else {
3717         irq = ics_find_free_block(ics, 1, 1);
3718         if (irq < 0) {
3719             error_setg(errp, "can't allocate IRQ: no IRQ left");
3720             return -1;
3721         }
3722         irq += ics->offset;
3723     }
3724 
3725     spapr_irq_set_lsi(spapr, irq, lsi);
3726     trace_spapr_irq_alloc(irq);
3727 
3728     return irq;
3729 }
3730 
3731 /*
3732  * Allocate block of consecutive IRQs, and return the number of the first IRQ in
3733  * the block. If align==true, aligns the first IRQ number to num.
3734  */
3735 int spapr_irq_alloc_block(sPAPRMachineState *spapr, int num, bool lsi,
3736                           bool align, Error **errp)
3737 {
3738     ICSState *ics = spapr->ics;
3739     int i, first = -1;
3740 
3741     if (!ics) {
3742         return -1;
3743     }
3744 
3745     /*
3746      * MSIMesage::data is used for storing VIRQ so
3747      * it has to be aligned to num to support multiple
3748      * MSI vectors. MSI-X is not affected by this.
3749      * The hint is used for the first IRQ, the rest should
3750      * be allocated continuously.
3751      */
3752     if (align) {
3753         assert((num == 1) || (num == 2) || (num == 4) ||
3754                (num == 8) || (num == 16) || (num == 32));
3755         first = ics_find_free_block(ics, num, num);
3756     } else {
3757         first = ics_find_free_block(ics, num, 1);
3758     }
3759     if (first < 0) {
3760         error_setg(errp, "can't find a free %d-IRQ block", num);
3761         return -1;
3762     }
3763 
3764     first += ics->offset;
3765     for (i = first; i < first + num; ++i) {
3766         spapr_irq_set_lsi(spapr, i, lsi);
3767     }
3768 
3769     trace_spapr_irq_alloc_block(first, num, lsi, align);
3770 
3771     return first;
3772 }
3773 
3774 void spapr_irq_free(sPAPRMachineState *spapr, int irq, int num)
3775 {
3776     ICSState *ics = spapr->ics;
3777     int srcno = irq - ics->offset;
3778     int i;
3779 
3780     if (ics_valid_irq(ics, irq)) {
3781         trace_spapr_irq_free(0, irq, num);
3782         for (i = srcno; i < srcno + num; ++i) {
3783             if (ICS_IRQ_FREE(ics, i)) {
3784                 trace_spapr_irq_free_warn(0, i + ics->offset);
3785             }
3786             memset(&ics->irqs[i], 0, sizeof(ICSIRQState));
3787         }
3788     }
3789 }
3790 
3791 qemu_irq spapr_qirq(sPAPRMachineState *spapr, int irq)
3792 {
3793     ICSState *ics = spapr->ics;
3794 
3795     if (ics_valid_irq(ics, irq)) {
3796         return ics->qirqs[irq - ics->offset];
3797     }
3798 
3799     return NULL;
3800 }
3801 
3802 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3803                                  Monitor *mon)
3804 {
3805     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3806     CPUState *cs;
3807 
3808     CPU_FOREACH(cs) {
3809         PowerPCCPU *cpu = POWERPC_CPU(cs);
3810 
3811         icp_pic_print_info(ICP(cpu->intc), mon);
3812     }
3813 
3814     ics_pic_print_info(spapr->ics, mon);
3815 }
3816 
3817 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3818 {
3819     return cpu->vcpu_id;
3820 }
3821 
3822 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3823 {
3824     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3825     int vcpu_id;
3826 
3827     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3828 
3829     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3830         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3831         error_append_hint(errp, "Adjust the number of cpus to %d "
3832                           "or try to raise the number of threads per core\n",
3833                           vcpu_id * smp_threads / spapr->vsmt);
3834         return;
3835     }
3836 
3837     cpu->vcpu_id = vcpu_id;
3838 }
3839 
3840 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3841 {
3842     CPUState *cs;
3843 
3844     CPU_FOREACH(cs) {
3845         PowerPCCPU *cpu = POWERPC_CPU(cs);
3846 
3847         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3848             return cpu;
3849         }
3850     }
3851 
3852     return NULL;
3853 }
3854 
3855 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3856 {
3857     MachineClass *mc = MACHINE_CLASS(oc);
3858     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3859     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3860     NMIClass *nc = NMI_CLASS(oc);
3861     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3862     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3863     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3864     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3865 
3866     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3867 
3868     /*
3869      * We set up the default / latest behaviour here.  The class_init
3870      * functions for the specific versioned machine types can override
3871      * these details for backwards compatibility
3872      */
3873     mc->init = spapr_machine_init;
3874     mc->reset = spapr_machine_reset;
3875     mc->block_default_type = IF_SCSI;
3876     mc->max_cpus = 1024;
3877     mc->no_parallel = 1;
3878     mc->default_boot_order = "";
3879     mc->default_ram_size = 512 * M_BYTE;
3880     mc->kvm_type = spapr_kvm_type;
3881     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3882     mc->pci_allow_0_address = true;
3883     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3884     hc->pre_plug = spapr_machine_device_pre_plug;
3885     hc->plug = spapr_machine_device_plug;
3886     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3887     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3888     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3889     hc->unplug_request = spapr_machine_device_unplug_request;
3890 
3891     smc->dr_lmb_enabled = true;
3892     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3893     mc->has_hotpluggable_cpus = true;
3894     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3895     fwc->get_dev_path = spapr_get_fw_dev_path;
3896     nc->nmi_monitor_handler = spapr_nmi;
3897     smc->phb_placement = spapr_phb_placement;
3898     vhc->hypercall = emulate_spapr_hypercall;
3899     vhc->hpt_mask = spapr_hpt_mask;
3900     vhc->map_hptes = spapr_map_hptes;
3901     vhc->unmap_hptes = spapr_unmap_hptes;
3902     vhc->store_hpte = spapr_store_hpte;
3903     vhc->get_patbe = spapr_get_patbe;
3904     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3905     xic->ics_get = spapr_ics_get;
3906     xic->ics_resend = spapr_ics_resend;
3907     xic->icp_get = spapr_icp_get;
3908     ispc->print_info = spapr_pic_print_info;
3909     /* Force NUMA node memory size to be a multiple of
3910      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3911      * in which LMBs are represented and hot-added
3912      */
3913     mc->numa_mem_align_shift = 28;
3914 
3915     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
3916     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
3917     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
3918     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
3919     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
3920     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
3921     spapr_caps_add_properties(smc, &error_abort);
3922 }
3923 
3924 static const TypeInfo spapr_machine_info = {
3925     .name          = TYPE_SPAPR_MACHINE,
3926     .parent        = TYPE_MACHINE,
3927     .abstract      = true,
3928     .instance_size = sizeof(sPAPRMachineState),
3929     .instance_init = spapr_instance_init,
3930     .instance_finalize = spapr_machine_finalizefn,
3931     .class_size    = sizeof(sPAPRMachineClass),
3932     .class_init    = spapr_machine_class_init,
3933     .interfaces = (InterfaceInfo[]) {
3934         { TYPE_FW_PATH_PROVIDER },
3935         { TYPE_NMI },
3936         { TYPE_HOTPLUG_HANDLER },
3937         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3938         { TYPE_XICS_FABRIC },
3939         { TYPE_INTERRUPT_STATS_PROVIDER },
3940         { }
3941     },
3942 };
3943 
3944 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3945     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3946                                                     void *data)      \
3947     {                                                                \
3948         MachineClass *mc = MACHINE_CLASS(oc);                        \
3949         spapr_machine_##suffix##_class_options(mc);                  \
3950         if (latest) {                                                \
3951             mc->alias = "pseries";                                   \
3952             mc->is_default = 1;                                      \
3953         }                                                            \
3954     }                                                                \
3955     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3956     {                                                                \
3957         MachineState *machine = MACHINE(obj);                        \
3958         spapr_machine_##suffix##_instance_options(machine);          \
3959     }                                                                \
3960     static const TypeInfo spapr_machine_##suffix##_info = {          \
3961         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3962         .parent = TYPE_SPAPR_MACHINE,                                \
3963         .class_init = spapr_machine_##suffix##_class_init,           \
3964         .instance_init = spapr_machine_##suffix##_instance_init,     \
3965     };                                                               \
3966     static void spapr_machine_register_##suffix(void)                \
3967     {                                                                \
3968         type_register(&spapr_machine_##suffix##_info);               \
3969     }                                                                \
3970     type_init(spapr_machine_register_##suffix)
3971 
3972 /*
3973  * pseries-2.12
3974  */
3975 static void spapr_machine_2_12_instance_options(MachineState *machine)
3976 {
3977 }
3978 
3979 static void spapr_machine_2_12_class_options(MachineClass *mc)
3980 {
3981     /* Defaults for the latest behaviour inherited from the base class */
3982 }
3983 
3984 DEFINE_SPAPR_MACHINE(2_12, "2.12", true);
3985 
3986 /*
3987  * pseries-2.11
3988  */
3989 #define SPAPR_COMPAT_2_11                                              \
3990     HW_COMPAT_2_11
3991 
3992 static void spapr_machine_2_11_instance_options(MachineState *machine)
3993 {
3994     spapr_machine_2_12_instance_options(machine);
3995 }
3996 
3997 static void spapr_machine_2_11_class_options(MachineClass *mc)
3998 {
3999     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4000 
4001     spapr_machine_2_12_class_options(mc);
4002     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4003     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4004 }
4005 
4006 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4007 
4008 /*
4009  * pseries-2.10
4010  */
4011 #define SPAPR_COMPAT_2_10                                              \
4012     HW_COMPAT_2_10
4013 
4014 static void spapr_machine_2_10_instance_options(MachineState *machine)
4015 {
4016     spapr_machine_2_11_instance_options(machine);
4017 }
4018 
4019 static void spapr_machine_2_10_class_options(MachineClass *mc)
4020 {
4021     spapr_machine_2_11_class_options(mc);
4022     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4023 }
4024 
4025 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4026 
4027 /*
4028  * pseries-2.9
4029  */
4030 #define SPAPR_COMPAT_2_9                                               \
4031     HW_COMPAT_2_9                                                      \
4032     {                                                                  \
4033         .driver = TYPE_POWERPC_CPU,                                    \
4034         .property = "pre-2.10-migration",                              \
4035         .value    = "on",                                              \
4036     },                                                                 \
4037 
4038 static void spapr_machine_2_9_instance_options(MachineState *machine)
4039 {
4040     spapr_machine_2_10_instance_options(machine);
4041 }
4042 
4043 static void spapr_machine_2_9_class_options(MachineClass *mc)
4044 {
4045     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4046 
4047     spapr_machine_2_10_class_options(mc);
4048     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4049     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4050     smc->pre_2_10_has_unused_icps = true;
4051     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4052 }
4053 
4054 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4055 
4056 /*
4057  * pseries-2.8
4058  */
4059 #define SPAPR_COMPAT_2_8                                        \
4060     HW_COMPAT_2_8                                               \
4061     {                                                           \
4062         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4063         .property = "pcie-extended-configuration-space",        \
4064         .value    = "off",                                      \
4065     },
4066 
4067 static void spapr_machine_2_8_instance_options(MachineState *machine)
4068 {
4069     spapr_machine_2_9_instance_options(machine);
4070 }
4071 
4072 static void spapr_machine_2_8_class_options(MachineClass *mc)
4073 {
4074     spapr_machine_2_9_class_options(mc);
4075     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4076     mc->numa_mem_align_shift = 23;
4077 }
4078 
4079 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4080 
4081 /*
4082  * pseries-2.7
4083  */
4084 #define SPAPR_COMPAT_2_7                            \
4085     HW_COMPAT_2_7                                   \
4086     {                                               \
4087         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4088         .property = "mem_win_size",                 \
4089         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4090     },                                              \
4091     {                                               \
4092         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4093         .property = "mem64_win_size",               \
4094         .value    = "0",                            \
4095     },                                              \
4096     {                                               \
4097         .driver = TYPE_POWERPC_CPU,                 \
4098         .property = "pre-2.8-migration",            \
4099         .value    = "on",                           \
4100     },                                              \
4101     {                                               \
4102         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4103         .property = "pre-2.8-migration",            \
4104         .value    = "on",                           \
4105     },
4106 
4107 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4108                               uint64_t *buid, hwaddr *pio,
4109                               hwaddr *mmio32, hwaddr *mmio64,
4110                               unsigned n_dma, uint32_t *liobns, Error **errp)
4111 {
4112     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4113     const uint64_t base_buid = 0x800000020000000ULL;
4114     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4115     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4116     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4117     const uint32_t max_index = 255;
4118     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4119 
4120     uint64_t ram_top = MACHINE(spapr)->ram_size;
4121     hwaddr phb0_base, phb_base;
4122     int i;
4123 
4124     /* Do we have hotpluggable memory? */
4125     if (MACHINE(spapr)->maxram_size > ram_top) {
4126         /* Can't just use maxram_size, because there may be an
4127          * alignment gap between normal and hotpluggable memory
4128          * regions */
4129         ram_top = spapr->hotplug_memory.base +
4130             memory_region_size(&spapr->hotplug_memory.mr);
4131     }
4132 
4133     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4134 
4135     if (index > max_index) {
4136         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4137                    max_index);
4138         return;
4139     }
4140 
4141     *buid = base_buid + index;
4142     for (i = 0; i < n_dma; ++i) {
4143         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4144     }
4145 
4146     phb_base = phb0_base + index * phb_spacing;
4147     *pio = phb_base + pio_offset;
4148     *mmio32 = phb_base + mmio_offset;
4149     /*
4150      * We don't set the 64-bit MMIO window, relying on the PHB's
4151      * fallback behaviour of automatically splitting a large "32-bit"
4152      * window into contiguous 32-bit and 64-bit windows
4153      */
4154 }
4155 
4156 static void spapr_machine_2_7_instance_options(MachineState *machine)
4157 {
4158     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4159 
4160     spapr_machine_2_8_instance_options(machine);
4161     spapr->use_hotplug_event_source = false;
4162 }
4163 
4164 static void spapr_machine_2_7_class_options(MachineClass *mc)
4165 {
4166     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4167 
4168     spapr_machine_2_8_class_options(mc);
4169     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4170     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4171     smc->phb_placement = phb_placement_2_7;
4172 }
4173 
4174 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4175 
4176 /*
4177  * pseries-2.6
4178  */
4179 #define SPAPR_COMPAT_2_6 \
4180     HW_COMPAT_2_6 \
4181     { \
4182         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4183         .property = "ddw",\
4184         .value    = stringify(off),\
4185     },
4186 
4187 static void spapr_machine_2_6_instance_options(MachineState *machine)
4188 {
4189     spapr_machine_2_7_instance_options(machine);
4190 }
4191 
4192 static void spapr_machine_2_6_class_options(MachineClass *mc)
4193 {
4194     spapr_machine_2_7_class_options(mc);
4195     mc->has_hotpluggable_cpus = false;
4196     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4197 }
4198 
4199 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4200 
4201 /*
4202  * pseries-2.5
4203  */
4204 #define SPAPR_COMPAT_2_5 \
4205     HW_COMPAT_2_5 \
4206     { \
4207         .driver   = "spapr-vlan", \
4208         .property = "use-rx-buffer-pools", \
4209         .value    = "off", \
4210     },
4211 
4212 static void spapr_machine_2_5_instance_options(MachineState *machine)
4213 {
4214     spapr_machine_2_6_instance_options(machine);
4215 }
4216 
4217 static void spapr_machine_2_5_class_options(MachineClass *mc)
4218 {
4219     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4220 
4221     spapr_machine_2_6_class_options(mc);
4222     smc->use_ohci_by_default = true;
4223     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4224 }
4225 
4226 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4227 
4228 /*
4229  * pseries-2.4
4230  */
4231 #define SPAPR_COMPAT_2_4 \
4232         HW_COMPAT_2_4
4233 
4234 static void spapr_machine_2_4_instance_options(MachineState *machine)
4235 {
4236     spapr_machine_2_5_instance_options(machine);
4237 }
4238 
4239 static void spapr_machine_2_4_class_options(MachineClass *mc)
4240 {
4241     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4242 
4243     spapr_machine_2_5_class_options(mc);
4244     smc->dr_lmb_enabled = false;
4245     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4246 }
4247 
4248 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4249 
4250 /*
4251  * pseries-2.3
4252  */
4253 #define SPAPR_COMPAT_2_3 \
4254         HW_COMPAT_2_3 \
4255         {\
4256             .driver   = "spapr-pci-host-bridge",\
4257             .property = "dynamic-reconfiguration",\
4258             .value    = "off",\
4259         },
4260 
4261 static void spapr_machine_2_3_instance_options(MachineState *machine)
4262 {
4263     spapr_machine_2_4_instance_options(machine);
4264 }
4265 
4266 static void spapr_machine_2_3_class_options(MachineClass *mc)
4267 {
4268     spapr_machine_2_4_class_options(mc);
4269     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4270 }
4271 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4272 
4273 /*
4274  * pseries-2.2
4275  */
4276 
4277 #define SPAPR_COMPAT_2_2 \
4278         HW_COMPAT_2_2 \
4279         {\
4280             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4281             .property = "mem_win_size",\
4282             .value    = "0x20000000",\
4283         },
4284 
4285 static void spapr_machine_2_2_instance_options(MachineState *machine)
4286 {
4287     spapr_machine_2_3_instance_options(machine);
4288     machine->suppress_vmdesc = true;
4289 }
4290 
4291 static void spapr_machine_2_2_class_options(MachineClass *mc)
4292 {
4293     spapr_machine_2_3_class_options(mc);
4294     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4295 }
4296 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4297 
4298 /*
4299  * pseries-2.1
4300  */
4301 #define SPAPR_COMPAT_2_1 \
4302         HW_COMPAT_2_1
4303 
4304 static void spapr_machine_2_1_instance_options(MachineState *machine)
4305 {
4306     spapr_machine_2_2_instance_options(machine);
4307 }
4308 
4309 static void spapr_machine_2_1_class_options(MachineClass *mc)
4310 {
4311     spapr_machine_2_2_class_options(mc);
4312     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4313 }
4314 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4315 
4316 static void spapr_machine_register_types(void)
4317 {
4318     type_register_static(&spapr_machine_info);
4319 }
4320 
4321 type_init(spapr_machine_register_types)
4322