xref: /openbmc/qemu/hw/ppc/spapr.c (revision 0fcd2a814aa331f87fd099171ae03a61311bdfee)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/block-backend.h"
39 #include "sysemu/cpus.h"
40 #include "sysemu/hw_accel.h"
41 #include "kvm_ppc.h"
42 #include "migration/misc.h"
43 #include "migration/global_state.h"
44 #include "migration/register.h"
45 #include "mmu-hash64.h"
46 #include "mmu-book3s-v3.h"
47 #include "cpu-models.h"
48 #include "qom/cpu.h"
49 
50 #include "hw/boards.h"
51 #include "hw/ppc/ppc.h"
52 #include "hw/loader.h"
53 
54 #include "hw/ppc/fdt.h"
55 #include "hw/ppc/spapr.h"
56 #include "hw/ppc/spapr_vio.h"
57 #include "hw/pci-host/spapr.h"
58 #include "hw/ppc/xics.h"
59 #include "hw/pci/msi.h"
60 
61 #include "hw/pci/pci.h"
62 #include "hw/scsi/scsi.h"
63 #include "hw/virtio/virtio-scsi.h"
64 #include "hw/virtio/vhost-scsi-common.h"
65 
66 #include "exec/address-spaces.h"
67 #include "hw/usb.h"
68 #include "qemu/config-file.h"
69 #include "qemu/error-report.h"
70 #include "trace.h"
71 #include "hw/nmi.h"
72 #include "hw/intc/intc.h"
73 
74 #include "hw/compat.h"
75 #include "qemu/cutils.h"
76 #include "hw/ppc/spapr_cpu_core.h"
77 
78 #include <libfdt.h>
79 
80 /* SLOF memory layout:
81  *
82  * SLOF raw image loaded at 0, copies its romfs right below the flat
83  * device-tree, then position SLOF itself 31M below that
84  *
85  * So we set FW_OVERHEAD to 40MB which should account for all of that
86  * and more
87  *
88  * We load our kernel at 4M, leaving space for SLOF initial image
89  */
90 #define FDT_MAX_SIZE            0x100000
91 #define RTAS_MAX_SIZE           0x10000
92 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
93 #define FW_MAX_SIZE             0x400000
94 #define FW_FILE_NAME            "slof.bin"
95 #define FW_OVERHEAD             0x2800000
96 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
97 
98 #define MIN_RMA_SLOF            128UL
99 
100 #define PHANDLE_XICP            0x00001111
101 
102 /* These two functions implement the VCPU id numbering: one to compute them
103  * all and one to identify thread 0 of a VCORE. Any change to the first one
104  * is likely to have an impact on the second one, so let's keep them close.
105  */
106 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
107 {
108     assert(spapr->vsmt);
109     return
110         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
111 }
112 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
113                                       PowerPCCPU *cpu)
114 {
115     assert(spapr->vsmt);
116     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
117 }
118 
119 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
120                                   const char *type_ics,
121                                   int nr_irqs, Error **errp)
122 {
123     Error *local_err = NULL;
124     Object *obj;
125 
126     obj = object_new(type_ics);
127     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
128     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
129                                    &error_abort);
130     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
131     if (local_err) {
132         goto error;
133     }
134     object_property_set_bool(obj, true, "realized", &local_err);
135     if (local_err) {
136         goto error;
137     }
138 
139     return ICS_SIMPLE(obj);
140 
141 error:
142     error_propagate(errp, local_err);
143     return NULL;
144 }
145 
146 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
147 {
148     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
149      * and newer QEMUs don't even have them. In both cases, we don't want
150      * to send anything on the wire.
151      */
152     return false;
153 }
154 
155 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
156     .name = "icp/server",
157     .version_id = 1,
158     .minimum_version_id = 1,
159     .needed = pre_2_10_vmstate_dummy_icp_needed,
160     .fields = (VMStateField[]) {
161         VMSTATE_UNUSED(4), /* uint32_t xirr */
162         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
163         VMSTATE_UNUSED(1), /* uint8_t mfrr */
164         VMSTATE_END_OF_LIST()
165     },
166 };
167 
168 static void pre_2_10_vmstate_register_dummy_icp(int i)
169 {
170     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
171                      (void *)(uintptr_t) i);
172 }
173 
174 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
175 {
176     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
177                        (void *)(uintptr_t) i);
178 }
179 
180 static int xics_max_server_number(sPAPRMachineState *spapr)
181 {
182     assert(spapr->vsmt);
183     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
184 }
185 
186 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
187 {
188     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
189 
190     if (kvm_enabled()) {
191         if (machine_kernel_irqchip_allowed(machine) &&
192             !xics_kvm_init(spapr, errp)) {
193             spapr->icp_type = TYPE_KVM_ICP;
194             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
195         }
196         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
197             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
198             return;
199         }
200     }
201 
202     if (!spapr->ics) {
203         xics_spapr_init(spapr);
204         spapr->icp_type = TYPE_ICP;
205         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
206         if (!spapr->ics) {
207             return;
208         }
209     }
210 }
211 
212 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
213                                   int smt_threads)
214 {
215     int i, ret = 0;
216     uint32_t servers_prop[smt_threads];
217     uint32_t gservers_prop[smt_threads * 2];
218     int index = spapr_get_vcpu_id(cpu);
219 
220     if (cpu->compat_pvr) {
221         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
222         if (ret < 0) {
223             return ret;
224         }
225     }
226 
227     /* Build interrupt servers and gservers properties */
228     for (i = 0; i < smt_threads; i++) {
229         servers_prop[i] = cpu_to_be32(index + i);
230         /* Hack, direct the group queues back to cpu 0 */
231         gservers_prop[i*2] = cpu_to_be32(index + i);
232         gservers_prop[i*2 + 1] = 0;
233     }
234     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
235                       servers_prop, sizeof(servers_prop));
236     if (ret < 0) {
237         return ret;
238     }
239     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
240                       gservers_prop, sizeof(gservers_prop));
241 
242     return ret;
243 }
244 
245 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
246 {
247     int index = spapr_get_vcpu_id(cpu);
248     uint32_t associativity[] = {cpu_to_be32(0x5),
249                                 cpu_to_be32(0x0),
250                                 cpu_to_be32(0x0),
251                                 cpu_to_be32(0x0),
252                                 cpu_to_be32(cpu->node_id),
253                                 cpu_to_be32(index)};
254 
255     /* Advertise NUMA via ibm,associativity */
256     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
257                           sizeof(associativity));
258 }
259 
260 /* Populate the "ibm,pa-features" property */
261 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
262                                        PowerPCCPU *cpu,
263                                        void *fdt, int offset,
264                                        bool legacy_guest)
265 {
266     uint8_t pa_features_206[] = { 6, 0,
267         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
268     uint8_t pa_features_207[] = { 24, 0,
269         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
270         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
271         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
272         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
273     uint8_t pa_features_300[] = { 66, 0,
274         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
275         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
276         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
277         /* 6: DS207 */
278         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
279         /* 16: Vector */
280         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
281         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
282         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
283         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
284         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
285         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
286         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
287         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
288         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
289         /* 42: PM, 44: PC RA, 46: SC vec'd */
290         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
291         /* 48: SIMD, 50: QP BFP, 52: String */
292         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
293         /* 54: DecFP, 56: DecI, 58: SHA */
294         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
295         /* 60: NM atomic, 62: RNG */
296         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
297     };
298     uint8_t *pa_features = NULL;
299     size_t pa_size;
300 
301     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
302         pa_features = pa_features_206;
303         pa_size = sizeof(pa_features_206);
304     }
305     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
306         pa_features = pa_features_207;
307         pa_size = sizeof(pa_features_207);
308     }
309     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
310         pa_features = pa_features_300;
311         pa_size = sizeof(pa_features_300);
312     }
313     if (!pa_features) {
314         return;
315     }
316 
317     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
318         /*
319          * Note: we keep CI large pages off by default because a 64K capable
320          * guest provisioned with large pages might otherwise try to map a qemu
321          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
322          * even if that qemu runs on a 4k host.
323          * We dd this bit back here if we are confident this is not an issue
324          */
325         pa_features[3] |= 0x20;
326     }
327     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
328         pa_features[24] |= 0x80;    /* Transactional memory support */
329     }
330     if (legacy_guest && pa_size > 40) {
331         /* Workaround for broken kernels that attempt (guest) radix
332          * mode when they can't handle it, if they see the radix bit set
333          * in pa-features. So hide it from them. */
334         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
335     }
336 
337     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
338 }
339 
340 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
341 {
342     int ret = 0, offset, cpus_offset;
343     CPUState *cs;
344     char cpu_model[32];
345     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
346 
347     CPU_FOREACH(cs) {
348         PowerPCCPU *cpu = POWERPC_CPU(cs);
349         DeviceClass *dc = DEVICE_GET_CLASS(cs);
350         int index = spapr_get_vcpu_id(cpu);
351         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
352 
353         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
354             continue;
355         }
356 
357         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
358 
359         cpus_offset = fdt_path_offset(fdt, "/cpus");
360         if (cpus_offset < 0) {
361             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
362             if (cpus_offset < 0) {
363                 return cpus_offset;
364             }
365         }
366         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
367         if (offset < 0) {
368             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
369             if (offset < 0) {
370                 return offset;
371             }
372         }
373 
374         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
375                           pft_size_prop, sizeof(pft_size_prop));
376         if (ret < 0) {
377             return ret;
378         }
379 
380         if (nb_numa_nodes > 1) {
381             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
382             if (ret < 0) {
383                 return ret;
384             }
385         }
386 
387         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
388         if (ret < 0) {
389             return ret;
390         }
391 
392         spapr_populate_pa_features(spapr, cpu, fdt, offset,
393                                    spapr->cas_legacy_guest_workaround);
394     }
395     return ret;
396 }
397 
398 static hwaddr spapr_node0_size(MachineState *machine)
399 {
400     if (nb_numa_nodes) {
401         int i;
402         for (i = 0; i < nb_numa_nodes; ++i) {
403             if (numa_info[i].node_mem) {
404                 return MIN(pow2floor(numa_info[i].node_mem),
405                            machine->ram_size);
406             }
407         }
408     }
409     return machine->ram_size;
410 }
411 
412 static void add_str(GString *s, const gchar *s1)
413 {
414     g_string_append_len(s, s1, strlen(s1) + 1);
415 }
416 
417 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
418                                        hwaddr size)
419 {
420     uint32_t associativity[] = {
421         cpu_to_be32(0x4), /* length */
422         cpu_to_be32(0x0), cpu_to_be32(0x0),
423         cpu_to_be32(0x0), cpu_to_be32(nodeid)
424     };
425     char mem_name[32];
426     uint64_t mem_reg_property[2];
427     int off;
428 
429     mem_reg_property[0] = cpu_to_be64(start);
430     mem_reg_property[1] = cpu_to_be64(size);
431 
432     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
433     off = fdt_add_subnode(fdt, 0, mem_name);
434     _FDT(off);
435     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
436     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
437                       sizeof(mem_reg_property))));
438     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
439                       sizeof(associativity))));
440     return off;
441 }
442 
443 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
444 {
445     MachineState *machine = MACHINE(spapr);
446     hwaddr mem_start, node_size;
447     int i, nb_nodes = nb_numa_nodes;
448     NodeInfo *nodes = numa_info;
449     NodeInfo ramnode;
450 
451     /* No NUMA nodes, assume there is just one node with whole RAM */
452     if (!nb_numa_nodes) {
453         nb_nodes = 1;
454         ramnode.node_mem = machine->ram_size;
455         nodes = &ramnode;
456     }
457 
458     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
459         if (!nodes[i].node_mem) {
460             continue;
461         }
462         if (mem_start >= machine->ram_size) {
463             node_size = 0;
464         } else {
465             node_size = nodes[i].node_mem;
466             if (node_size > machine->ram_size - mem_start) {
467                 node_size = machine->ram_size - mem_start;
468             }
469         }
470         if (!mem_start) {
471             /* spapr_machine_init() checks for rma_size <= node0_size
472              * already */
473             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
474             mem_start += spapr->rma_size;
475             node_size -= spapr->rma_size;
476         }
477         for ( ; node_size; ) {
478             hwaddr sizetmp = pow2floor(node_size);
479 
480             /* mem_start != 0 here */
481             if (ctzl(mem_start) < ctzl(sizetmp)) {
482                 sizetmp = 1ULL << ctzl(mem_start);
483             }
484 
485             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
486             node_size -= sizetmp;
487             mem_start += sizetmp;
488         }
489     }
490 
491     return 0;
492 }
493 
494 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
495                                   sPAPRMachineState *spapr)
496 {
497     PowerPCCPU *cpu = POWERPC_CPU(cs);
498     CPUPPCState *env = &cpu->env;
499     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
500     int index = spapr_get_vcpu_id(cpu);
501     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
502                        0xffffffff, 0xffffffff};
503     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
504         : SPAPR_TIMEBASE_FREQ;
505     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
506     uint32_t page_sizes_prop[64];
507     size_t page_sizes_prop_size;
508     uint32_t vcpus_per_socket = smp_threads * smp_cores;
509     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
510     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
511     sPAPRDRConnector *drc;
512     int drc_index;
513     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
514     int i;
515 
516     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
517     if (drc) {
518         drc_index = spapr_drc_index(drc);
519         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
520     }
521 
522     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
523     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
524 
525     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
526     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
527                            env->dcache_line_size)));
528     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
529                            env->dcache_line_size)));
530     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
531                            env->icache_line_size)));
532     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
533                            env->icache_line_size)));
534 
535     if (pcc->l1_dcache_size) {
536         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
537                                pcc->l1_dcache_size)));
538     } else {
539         warn_report("Unknown L1 dcache size for cpu");
540     }
541     if (pcc->l1_icache_size) {
542         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
543                                pcc->l1_icache_size)));
544     } else {
545         warn_report("Unknown L1 icache size for cpu");
546     }
547 
548     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
549     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
550     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
551     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
552     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
553     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
554 
555     if (env->spr_cb[SPR_PURR].oea_read) {
556         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
557     }
558 
559     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
560         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
561                           segs, sizeof(segs))));
562     }
563 
564     /* Advertise VSX (vector extensions) if available
565      *   1               == VMX / Altivec available
566      *   2               == VSX available
567      *
568      * Only CPUs for which we create core types in spapr_cpu_core.c
569      * are possible, and all of those have VMX */
570     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
571         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
572     } else {
573         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
574     }
575 
576     /* Advertise DFP (Decimal Floating Point) if available
577      *   0 / no property == no DFP
578      *   1               == DFP available */
579     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
580         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
581     }
582 
583     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
584                                                       sizeof(page_sizes_prop));
585     if (page_sizes_prop_size) {
586         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
587                           page_sizes_prop, page_sizes_prop_size)));
588     }
589 
590     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
591 
592     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
593                            cs->cpu_index / vcpus_per_socket)));
594 
595     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
596                       pft_size_prop, sizeof(pft_size_prop))));
597 
598     if (nb_numa_nodes > 1) {
599         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
600     }
601 
602     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
603 
604     if (pcc->radix_page_info) {
605         for (i = 0; i < pcc->radix_page_info->count; i++) {
606             radix_AP_encodings[i] =
607                 cpu_to_be32(pcc->radix_page_info->entries[i]);
608         }
609         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
610                           radix_AP_encodings,
611                           pcc->radix_page_info->count *
612                           sizeof(radix_AP_encodings[0]))));
613     }
614 }
615 
616 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
617 {
618     CPUState *cs;
619     int cpus_offset;
620     char *nodename;
621 
622     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
623     _FDT(cpus_offset);
624     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
625     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
626 
627     /*
628      * We walk the CPUs in reverse order to ensure that CPU DT nodes
629      * created by fdt_add_subnode() end up in the right order in FDT
630      * for the guest kernel the enumerate the CPUs correctly.
631      */
632     CPU_FOREACH_REVERSE(cs) {
633         PowerPCCPU *cpu = POWERPC_CPU(cs);
634         int index = spapr_get_vcpu_id(cpu);
635         DeviceClass *dc = DEVICE_GET_CLASS(cs);
636         int offset;
637 
638         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
639             continue;
640         }
641 
642         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
643         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
644         g_free(nodename);
645         _FDT(offset);
646         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
647     }
648 
649 }
650 
651 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
652 {
653     MemoryDeviceInfoList *info;
654 
655     for (info = list; info; info = info->next) {
656         MemoryDeviceInfo *value = info->value;
657 
658         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
659             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
660 
661             if (pcdimm_info->addr >= addr &&
662                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
663                 return pcdimm_info->node;
664             }
665         }
666     }
667 
668     return -1;
669 }
670 
671 struct sPAPRDrconfCellV2 {
672      uint32_t seq_lmbs;
673      uint64_t base_addr;
674      uint32_t drc_index;
675      uint32_t aa_index;
676      uint32_t flags;
677 } QEMU_PACKED;
678 
679 typedef struct DrconfCellQueue {
680     struct sPAPRDrconfCellV2 cell;
681     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
682 } DrconfCellQueue;
683 
684 static DrconfCellQueue *
685 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
686                       uint32_t drc_index, uint32_t aa_index,
687                       uint32_t flags)
688 {
689     DrconfCellQueue *elem;
690 
691     elem = g_malloc0(sizeof(*elem));
692     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
693     elem->cell.base_addr = cpu_to_be64(base_addr);
694     elem->cell.drc_index = cpu_to_be32(drc_index);
695     elem->cell.aa_index = cpu_to_be32(aa_index);
696     elem->cell.flags = cpu_to_be32(flags);
697 
698     return elem;
699 }
700 
701 /* ibm,dynamic-memory-v2 */
702 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
703                                    int offset, MemoryDeviceInfoList *dimms)
704 {
705     uint8_t *int_buf, *cur_index, buf_len;
706     int ret;
707     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
708     uint64_t addr, cur_addr, size;
709     uint32_t nr_boot_lmbs = (spapr->hotplug_memory.base / lmb_size);
710     uint64_t mem_end = spapr->hotplug_memory.base +
711                        memory_region_size(&spapr->hotplug_memory.mr);
712     uint32_t node, nr_entries = 0;
713     sPAPRDRConnector *drc;
714     DrconfCellQueue *elem, *next;
715     MemoryDeviceInfoList *info;
716     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
717         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
718 
719     /* Entry to cover RAM and the gap area */
720     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
721                                  SPAPR_LMB_FLAGS_RESERVED |
722                                  SPAPR_LMB_FLAGS_DRC_INVALID);
723     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
724     nr_entries++;
725 
726     cur_addr = spapr->hotplug_memory.base;
727     for (info = dimms; info; info = info->next) {
728         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
729 
730         addr = di->addr;
731         size = di->size;
732         node = di->node;
733 
734         /* Entry for hot-pluggable area */
735         if (cur_addr < addr) {
736             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
737             g_assert(drc);
738             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
739                                          cur_addr, spapr_drc_index(drc), -1, 0);
740             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
741             nr_entries++;
742         }
743 
744         /* Entry for DIMM */
745         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
746         g_assert(drc);
747         elem = spapr_get_drconf_cell(size / lmb_size, addr,
748                                      spapr_drc_index(drc), node,
749                                      SPAPR_LMB_FLAGS_ASSIGNED);
750         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
751         nr_entries++;
752         cur_addr = addr + size;
753     }
754 
755     /* Entry for remaining hotpluggable area */
756     if (cur_addr < mem_end) {
757         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
758         g_assert(drc);
759         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
760                                      cur_addr, spapr_drc_index(drc), -1, 0);
761         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
762         nr_entries++;
763     }
764 
765     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
766     int_buf = cur_index = g_malloc0(buf_len);
767     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
768     cur_index += sizeof(nr_entries);
769 
770     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
771         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
772         cur_index += sizeof(elem->cell);
773         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
774         g_free(elem);
775     }
776 
777     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
778     g_free(int_buf);
779     if (ret < 0) {
780         return -1;
781     }
782     return 0;
783 }
784 
785 /* ibm,dynamic-memory */
786 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
787                                    int offset, MemoryDeviceInfoList *dimms)
788 {
789     int i, ret;
790     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
791     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
792     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
793                        memory_region_size(&spapr->hotplug_memory.mr)) /
794                        lmb_size;
795     uint32_t *int_buf, *cur_index, buf_len;
796 
797     /*
798      * Allocate enough buffer size to fit in ibm,dynamic-memory
799      */
800     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
801     cur_index = int_buf = g_malloc0(buf_len);
802     int_buf[0] = cpu_to_be32(nr_lmbs);
803     cur_index++;
804     for (i = 0; i < nr_lmbs; i++) {
805         uint64_t addr = i * lmb_size;
806         uint32_t *dynamic_memory = cur_index;
807 
808         if (i >= hotplug_lmb_start) {
809             sPAPRDRConnector *drc;
810 
811             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
812             g_assert(drc);
813 
814             dynamic_memory[0] = cpu_to_be32(addr >> 32);
815             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
816             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
817             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
818             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
819             if (memory_region_present(get_system_memory(), addr)) {
820                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
821             } else {
822                 dynamic_memory[5] = cpu_to_be32(0);
823             }
824         } else {
825             /*
826              * LMB information for RMA, boot time RAM and gap b/n RAM and
827              * hotplug memory region -- all these are marked as reserved
828              * and as having no valid DRC.
829              */
830             dynamic_memory[0] = cpu_to_be32(addr >> 32);
831             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
832             dynamic_memory[2] = cpu_to_be32(0);
833             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
834             dynamic_memory[4] = cpu_to_be32(-1);
835             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
836                                             SPAPR_LMB_FLAGS_DRC_INVALID);
837         }
838 
839         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
840     }
841     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
842     g_free(int_buf);
843     if (ret < 0) {
844         return -1;
845     }
846     return 0;
847 }
848 
849 /*
850  * Adds ibm,dynamic-reconfiguration-memory node.
851  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
852  * of this device tree node.
853  */
854 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
855 {
856     MachineState *machine = MACHINE(spapr);
857     int ret, i, offset;
858     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
859     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
860     uint32_t *int_buf, *cur_index, buf_len;
861     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
862     MemoryDeviceInfoList *dimms = NULL;
863 
864     /*
865      * Don't create the node if there is no hotpluggable memory
866      */
867     if (machine->ram_size == machine->maxram_size) {
868         return 0;
869     }
870 
871     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
872 
873     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
874                     sizeof(prop_lmb_size));
875     if (ret < 0) {
876         return ret;
877     }
878 
879     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
880     if (ret < 0) {
881         return ret;
882     }
883 
884     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
885     if (ret < 0) {
886         return ret;
887     }
888 
889     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
890     dimms = qmp_pc_dimm_device_list();
891     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
892         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
893     } else {
894         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
895     }
896     qapi_free_MemoryDeviceInfoList(dimms);
897 
898     if (ret < 0) {
899         return ret;
900     }
901 
902     /* ibm,associativity-lookup-arrays */
903     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
904     cur_index = int_buf = g_malloc0(buf_len);
905 
906     cur_index = int_buf;
907     int_buf[0] = cpu_to_be32(nr_nodes);
908     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
909     cur_index += 2;
910     for (i = 0; i < nr_nodes; i++) {
911         uint32_t associativity[] = {
912             cpu_to_be32(0x0),
913             cpu_to_be32(0x0),
914             cpu_to_be32(0x0),
915             cpu_to_be32(i)
916         };
917         memcpy(cur_index, associativity, sizeof(associativity));
918         cur_index += 4;
919     }
920     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
921             (cur_index - int_buf) * sizeof(uint32_t));
922     g_free(int_buf);
923 
924     return ret;
925 }
926 
927 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
928                                 sPAPROptionVector *ov5_updates)
929 {
930     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
931     int ret = 0, offset;
932 
933     /* Generate ibm,dynamic-reconfiguration-memory node if required */
934     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
935         g_assert(smc->dr_lmb_enabled);
936         ret = spapr_populate_drconf_memory(spapr, fdt);
937         if (ret) {
938             goto out;
939         }
940     }
941 
942     offset = fdt_path_offset(fdt, "/chosen");
943     if (offset < 0) {
944         offset = fdt_add_subnode(fdt, 0, "chosen");
945         if (offset < 0) {
946             return offset;
947         }
948     }
949     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
950                                  "ibm,architecture-vec-5");
951 
952 out:
953     return ret;
954 }
955 
956 static bool spapr_hotplugged_dev_before_cas(void)
957 {
958     Object *drc_container, *obj;
959     ObjectProperty *prop;
960     ObjectPropertyIterator iter;
961 
962     drc_container = container_get(object_get_root(), "/dr-connector");
963     object_property_iter_init(&iter, drc_container);
964     while ((prop = object_property_iter_next(&iter))) {
965         if (!strstart(prop->type, "link<", NULL)) {
966             continue;
967         }
968         obj = object_property_get_link(drc_container, prop->name, NULL);
969         if (spapr_drc_needed(obj)) {
970             return true;
971         }
972     }
973     return false;
974 }
975 
976 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
977                                  target_ulong addr, target_ulong size,
978                                  sPAPROptionVector *ov5_updates)
979 {
980     void *fdt, *fdt_skel;
981     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
982 
983     if (spapr_hotplugged_dev_before_cas()) {
984         return 1;
985     }
986 
987     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
988         error_report("SLOF provided an unexpected CAS buffer size "
989                      TARGET_FMT_lu " (min: %zu, max: %u)",
990                      size, sizeof(hdr), FW_MAX_SIZE);
991         exit(EXIT_FAILURE);
992     }
993 
994     size -= sizeof(hdr);
995 
996     /* Create skeleton */
997     fdt_skel = g_malloc0(size);
998     _FDT((fdt_create(fdt_skel, size)));
999     _FDT((fdt_finish_reservemap(fdt_skel)));
1000     _FDT((fdt_begin_node(fdt_skel, "")));
1001     _FDT((fdt_end_node(fdt_skel)));
1002     _FDT((fdt_finish(fdt_skel)));
1003     fdt = g_malloc0(size);
1004     _FDT((fdt_open_into(fdt_skel, fdt, size)));
1005     g_free(fdt_skel);
1006 
1007     /* Fixup cpu nodes */
1008     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
1009 
1010     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
1011         return -1;
1012     }
1013 
1014     /* Pack resulting tree */
1015     _FDT((fdt_pack(fdt)));
1016 
1017     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1018         trace_spapr_cas_failed(size);
1019         return -1;
1020     }
1021 
1022     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1023     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1024     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1025     g_free(fdt);
1026 
1027     return 0;
1028 }
1029 
1030 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1031 {
1032     int rtas;
1033     GString *hypertas = g_string_sized_new(256);
1034     GString *qemu_hypertas = g_string_sized_new(256);
1035     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1036     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
1037         memory_region_size(&spapr->hotplug_memory.mr);
1038     uint32_t lrdr_capacity[] = {
1039         cpu_to_be32(max_hotplug_addr >> 32),
1040         cpu_to_be32(max_hotplug_addr & 0xffffffff),
1041         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1042         cpu_to_be32(max_cpus / smp_threads),
1043     };
1044     uint32_t maxdomains[] = {
1045         cpu_to_be32(4),
1046         cpu_to_be32(0),
1047         cpu_to_be32(0),
1048         cpu_to_be32(0),
1049         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
1050     };
1051 
1052     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1053 
1054     /* hypertas */
1055     add_str(hypertas, "hcall-pft");
1056     add_str(hypertas, "hcall-term");
1057     add_str(hypertas, "hcall-dabr");
1058     add_str(hypertas, "hcall-interrupt");
1059     add_str(hypertas, "hcall-tce");
1060     add_str(hypertas, "hcall-vio");
1061     add_str(hypertas, "hcall-splpar");
1062     add_str(hypertas, "hcall-bulk");
1063     add_str(hypertas, "hcall-set-mode");
1064     add_str(hypertas, "hcall-sprg0");
1065     add_str(hypertas, "hcall-copy");
1066     add_str(hypertas, "hcall-debug");
1067     add_str(qemu_hypertas, "hcall-memop1");
1068 
1069     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1070         add_str(hypertas, "hcall-multi-tce");
1071     }
1072 
1073     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1074         add_str(hypertas, "hcall-hpt-resize");
1075     }
1076 
1077     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1078                      hypertas->str, hypertas->len));
1079     g_string_free(hypertas, TRUE);
1080     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1081                      qemu_hypertas->str, qemu_hypertas->len));
1082     g_string_free(qemu_hypertas, TRUE);
1083 
1084     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1085                      refpoints, sizeof(refpoints)));
1086 
1087     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1088                      maxdomains, sizeof(maxdomains)));
1089 
1090     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1091                           RTAS_ERROR_LOG_MAX));
1092     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1093                           RTAS_EVENT_SCAN_RATE));
1094 
1095     g_assert(msi_nonbroken);
1096     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1097 
1098     /*
1099      * According to PAPR, rtas ibm,os-term does not guarantee a return
1100      * back to the guest cpu.
1101      *
1102      * While an additional ibm,extended-os-term property indicates
1103      * that rtas call return will always occur. Set this property.
1104      */
1105     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1106 
1107     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1108                      lrdr_capacity, sizeof(lrdr_capacity)));
1109 
1110     spapr_dt_rtas_tokens(fdt, rtas);
1111 }
1112 
1113 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1114  * that the guest may request and thus the valid values for bytes 24..26 of
1115  * option vector 5: */
1116 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1117 {
1118     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1119 
1120     char val[2 * 4] = {
1121         23, 0x00, /* Xive mode, filled in below. */
1122         24, 0x00, /* Hash/Radix, filled in below. */
1123         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1124         26, 0x40, /* Radix options: GTSE == yes. */
1125     };
1126 
1127     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1128                           first_ppc_cpu->compat_pvr)) {
1129         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1130         val[3] = 0x00; /* Hash */
1131     } else if (kvm_enabled()) {
1132         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1133             val[3] = 0x80; /* OV5_MMU_BOTH */
1134         } else if (kvmppc_has_cap_mmu_radix()) {
1135             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1136         } else {
1137             val[3] = 0x00; /* Hash */
1138         }
1139     } else {
1140         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1141         val[3] = 0xC0;
1142     }
1143     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1144                      val, sizeof(val)));
1145 }
1146 
1147 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1148 {
1149     MachineState *machine = MACHINE(spapr);
1150     int chosen;
1151     const char *boot_device = machine->boot_order;
1152     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1153     size_t cb = 0;
1154     char *bootlist = get_boot_devices_list(&cb, true);
1155 
1156     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1157 
1158     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1159     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1160                           spapr->initrd_base));
1161     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1162                           spapr->initrd_base + spapr->initrd_size));
1163 
1164     if (spapr->kernel_size) {
1165         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1166                               cpu_to_be64(spapr->kernel_size) };
1167 
1168         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1169                          &kprop, sizeof(kprop)));
1170         if (spapr->kernel_le) {
1171             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1172         }
1173     }
1174     if (boot_menu) {
1175         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1176     }
1177     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1178     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1179     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1180 
1181     if (cb && bootlist) {
1182         int i;
1183 
1184         for (i = 0; i < cb; i++) {
1185             if (bootlist[i] == '\n') {
1186                 bootlist[i] = ' ';
1187             }
1188         }
1189         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1190     }
1191 
1192     if (boot_device && strlen(boot_device)) {
1193         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1194     }
1195 
1196     if (!spapr->has_graphics && stdout_path) {
1197         /*
1198          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1199          * kernel. New platforms should only use the "stdout-path" property. Set
1200          * the new property and continue using older property to remain
1201          * compatible with the existing firmware.
1202          */
1203         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1204         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1205     }
1206 
1207     spapr_dt_ov5_platform_support(fdt, chosen);
1208 
1209     g_free(stdout_path);
1210     g_free(bootlist);
1211 }
1212 
1213 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1214 {
1215     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1216      * KVM to work under pHyp with some guest co-operation */
1217     int hypervisor;
1218     uint8_t hypercall[16];
1219 
1220     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1221     /* indicate KVM hypercall interface */
1222     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1223     if (kvmppc_has_cap_fixup_hcalls()) {
1224         /*
1225          * Older KVM versions with older guest kernels were broken
1226          * with the magic page, don't allow the guest to map it.
1227          */
1228         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1229                                   sizeof(hypercall))) {
1230             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1231                              hypercall, sizeof(hypercall)));
1232         }
1233     }
1234 }
1235 
1236 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1237                              hwaddr rtas_addr,
1238                              hwaddr rtas_size)
1239 {
1240     MachineState *machine = MACHINE(spapr);
1241     MachineClass *mc = MACHINE_GET_CLASS(machine);
1242     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1243     int ret;
1244     void *fdt;
1245     sPAPRPHBState *phb;
1246     char *buf;
1247 
1248     fdt = g_malloc0(FDT_MAX_SIZE);
1249     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1250 
1251     /* Root node */
1252     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1253     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1254     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1255 
1256     /*
1257      * Add info to guest to indentify which host is it being run on
1258      * and what is the uuid of the guest
1259      */
1260     if (kvmppc_get_host_model(&buf)) {
1261         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1262         g_free(buf);
1263     }
1264     if (kvmppc_get_host_serial(&buf)) {
1265         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1266         g_free(buf);
1267     }
1268 
1269     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1270 
1271     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1272     if (qemu_uuid_set) {
1273         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1274     }
1275     g_free(buf);
1276 
1277     if (qemu_get_vm_name()) {
1278         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1279                                 qemu_get_vm_name()));
1280     }
1281 
1282     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1283     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1284 
1285     /* /interrupt controller */
1286     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1287 
1288     ret = spapr_populate_memory(spapr, fdt);
1289     if (ret < 0) {
1290         error_report("couldn't setup memory nodes in fdt");
1291         exit(1);
1292     }
1293 
1294     /* /vdevice */
1295     spapr_dt_vdevice(spapr->vio_bus, fdt);
1296 
1297     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1298         ret = spapr_rng_populate_dt(fdt);
1299         if (ret < 0) {
1300             error_report("could not set up rng device in the fdt");
1301             exit(1);
1302         }
1303     }
1304 
1305     QLIST_FOREACH(phb, &spapr->phbs, list) {
1306         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1307         if (ret < 0) {
1308             error_report("couldn't setup PCI devices in fdt");
1309             exit(1);
1310         }
1311     }
1312 
1313     /* cpus */
1314     spapr_populate_cpus_dt_node(fdt, spapr);
1315 
1316     if (smc->dr_lmb_enabled) {
1317         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1318     }
1319 
1320     if (mc->has_hotpluggable_cpus) {
1321         int offset = fdt_path_offset(fdt, "/cpus");
1322         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1323                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1324         if (ret < 0) {
1325             error_report("Couldn't set up CPU DR device tree properties");
1326             exit(1);
1327         }
1328     }
1329 
1330     /* /event-sources */
1331     spapr_dt_events(spapr, fdt);
1332 
1333     /* /rtas */
1334     spapr_dt_rtas(spapr, fdt);
1335 
1336     /* /chosen */
1337     spapr_dt_chosen(spapr, fdt);
1338 
1339     /* /hypervisor */
1340     if (kvm_enabled()) {
1341         spapr_dt_hypervisor(spapr, fdt);
1342     }
1343 
1344     /* Build memory reserve map */
1345     if (spapr->kernel_size) {
1346         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1347     }
1348     if (spapr->initrd_size) {
1349         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1350     }
1351 
1352     /* ibm,client-architecture-support updates */
1353     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1354     if (ret < 0) {
1355         error_report("couldn't setup CAS properties fdt");
1356         exit(1);
1357     }
1358 
1359     return fdt;
1360 }
1361 
1362 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1363 {
1364     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1365 }
1366 
1367 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1368                                     PowerPCCPU *cpu)
1369 {
1370     CPUPPCState *env = &cpu->env;
1371 
1372     /* The TCG path should also be holding the BQL at this point */
1373     g_assert(qemu_mutex_iothread_locked());
1374 
1375     if (msr_pr) {
1376         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1377         env->gpr[3] = H_PRIVILEGE;
1378     } else {
1379         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1380     }
1381 }
1382 
1383 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1384 {
1385     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1386 
1387     return spapr->patb_entry;
1388 }
1389 
1390 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1391 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1392 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1393 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1394 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1395 
1396 /*
1397  * Get the fd to access the kernel htab, re-opening it if necessary
1398  */
1399 static int get_htab_fd(sPAPRMachineState *spapr)
1400 {
1401     Error *local_err = NULL;
1402 
1403     if (spapr->htab_fd >= 0) {
1404         return spapr->htab_fd;
1405     }
1406 
1407     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1408     if (spapr->htab_fd < 0) {
1409         error_report_err(local_err);
1410     }
1411 
1412     return spapr->htab_fd;
1413 }
1414 
1415 void close_htab_fd(sPAPRMachineState *spapr)
1416 {
1417     if (spapr->htab_fd >= 0) {
1418         close(spapr->htab_fd);
1419     }
1420     spapr->htab_fd = -1;
1421 }
1422 
1423 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1424 {
1425     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1426 
1427     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1428 }
1429 
1430 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1431 {
1432     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1433 
1434     assert(kvm_enabled());
1435 
1436     if (!spapr->htab) {
1437         return 0;
1438     }
1439 
1440     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1441 }
1442 
1443 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1444                                                 hwaddr ptex, int n)
1445 {
1446     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1447     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1448 
1449     if (!spapr->htab) {
1450         /*
1451          * HTAB is controlled by KVM. Fetch into temporary buffer
1452          */
1453         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1454         kvmppc_read_hptes(hptes, ptex, n);
1455         return hptes;
1456     }
1457 
1458     /*
1459      * HTAB is controlled by QEMU. Just point to the internally
1460      * accessible PTEG.
1461      */
1462     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1463 }
1464 
1465 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1466                               const ppc_hash_pte64_t *hptes,
1467                               hwaddr ptex, int n)
1468 {
1469     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1470 
1471     if (!spapr->htab) {
1472         g_free((void *)hptes);
1473     }
1474 
1475     /* Nothing to do for qemu managed HPT */
1476 }
1477 
1478 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1479                              uint64_t pte0, uint64_t pte1)
1480 {
1481     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1482     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1483 
1484     if (!spapr->htab) {
1485         kvmppc_write_hpte(ptex, pte0, pte1);
1486     } else {
1487         stq_p(spapr->htab + offset, pte0);
1488         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1489     }
1490 }
1491 
1492 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1493 {
1494     int shift;
1495 
1496     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1497      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1498      * that's much more than is needed for Linux guests */
1499     shift = ctz64(pow2ceil(ramsize)) - 7;
1500     shift = MAX(shift, 18); /* Minimum architected size */
1501     shift = MIN(shift, 46); /* Maximum architected size */
1502     return shift;
1503 }
1504 
1505 void spapr_free_hpt(sPAPRMachineState *spapr)
1506 {
1507     g_free(spapr->htab);
1508     spapr->htab = NULL;
1509     spapr->htab_shift = 0;
1510     close_htab_fd(spapr);
1511 }
1512 
1513 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1514                           Error **errp)
1515 {
1516     long rc;
1517 
1518     /* Clean up any HPT info from a previous boot */
1519     spapr_free_hpt(spapr);
1520 
1521     rc = kvmppc_reset_htab(shift);
1522     if (rc < 0) {
1523         /* kernel-side HPT needed, but couldn't allocate one */
1524         error_setg_errno(errp, errno,
1525                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1526                          shift);
1527         /* This is almost certainly fatal, but if the caller really
1528          * wants to carry on with shift == 0, it's welcome to try */
1529     } else if (rc > 0) {
1530         /* kernel-side HPT allocated */
1531         if (rc != shift) {
1532             error_setg(errp,
1533                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1534                        shift, rc);
1535         }
1536 
1537         spapr->htab_shift = shift;
1538         spapr->htab = NULL;
1539     } else {
1540         /* kernel-side HPT not needed, allocate in userspace instead */
1541         size_t size = 1ULL << shift;
1542         int i;
1543 
1544         spapr->htab = qemu_memalign(size, size);
1545         if (!spapr->htab) {
1546             error_setg_errno(errp, errno,
1547                              "Could not allocate HPT of order %d", shift);
1548             return;
1549         }
1550 
1551         memset(spapr->htab, 0, size);
1552         spapr->htab_shift = shift;
1553 
1554         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1555             DIRTY_HPTE(HPTE(spapr->htab, i));
1556         }
1557     }
1558     /* We're setting up a hash table, so that means we're not radix */
1559     spapr->patb_entry = 0;
1560 }
1561 
1562 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1563 {
1564     int hpt_shift;
1565 
1566     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1567         || (spapr->cas_reboot
1568             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1569         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1570     } else {
1571         uint64_t current_ram_size;
1572 
1573         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1574         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1575     }
1576     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1577 
1578     if (spapr->vrma_adjust) {
1579         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1580                                           spapr->htab_shift);
1581     }
1582 }
1583 
1584 static int spapr_reset_drcs(Object *child, void *opaque)
1585 {
1586     sPAPRDRConnector *drc =
1587         (sPAPRDRConnector *) object_dynamic_cast(child,
1588                                                  TYPE_SPAPR_DR_CONNECTOR);
1589 
1590     if (drc) {
1591         spapr_drc_reset(drc);
1592     }
1593 
1594     return 0;
1595 }
1596 
1597 static void spapr_machine_reset(void)
1598 {
1599     MachineState *machine = MACHINE(qdev_get_machine());
1600     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1601     PowerPCCPU *first_ppc_cpu;
1602     uint32_t rtas_limit;
1603     hwaddr rtas_addr, fdt_addr;
1604     void *fdt;
1605     int rc;
1606 
1607     spapr_caps_reset(spapr);
1608 
1609     first_ppc_cpu = POWERPC_CPU(first_cpu);
1610     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1611         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1612                          spapr->max_compat_pvr)) {
1613         /* If using KVM with radix mode available, VCPUs can be started
1614          * without a HPT because KVM will start them in radix mode.
1615          * Set the GR bit in PATB so that we know there is no HPT. */
1616         spapr->patb_entry = PATBE1_GR;
1617     } else {
1618         spapr_setup_hpt_and_vrma(spapr);
1619     }
1620 
1621     /* if this reset wasn't generated by CAS, we should reset our
1622      * negotiated options and start from scratch */
1623     if (!spapr->cas_reboot) {
1624         spapr_ovec_cleanup(spapr->ov5_cas);
1625         spapr->ov5_cas = spapr_ovec_new();
1626 
1627         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1628     }
1629 
1630     qemu_devices_reset();
1631 
1632     /* DRC reset may cause a device to be unplugged. This will cause troubles
1633      * if this device is used by another device (eg, a running vhost backend
1634      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1635      * situations, we reset DRCs after all devices have been reset.
1636      */
1637     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1638 
1639     spapr_clear_pending_events(spapr);
1640 
1641     /*
1642      * We place the device tree and RTAS just below either the top of the RMA,
1643      * or just below 2GB, whichever is lowere, so that it can be
1644      * processed with 32-bit real mode code if necessary
1645      */
1646     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1647     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1648     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1649 
1650     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1651 
1652     spapr_load_rtas(spapr, fdt, rtas_addr);
1653 
1654     rc = fdt_pack(fdt);
1655 
1656     /* Should only fail if we've built a corrupted tree */
1657     assert(rc == 0);
1658 
1659     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1660         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1661                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1662         exit(1);
1663     }
1664 
1665     /* Load the fdt */
1666     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1667     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1668     g_free(fdt);
1669 
1670     /* Set up the entry state */
1671     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1672     first_ppc_cpu->env.gpr[5] = 0;
1673 
1674     spapr->cas_reboot = false;
1675 }
1676 
1677 static void spapr_create_nvram(sPAPRMachineState *spapr)
1678 {
1679     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1680     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1681 
1682     if (dinfo) {
1683         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1684                             &error_fatal);
1685     }
1686 
1687     qdev_init_nofail(dev);
1688 
1689     spapr->nvram = (struct sPAPRNVRAM *)dev;
1690 }
1691 
1692 static void spapr_rtc_create(sPAPRMachineState *spapr)
1693 {
1694     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1695     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1696                               &error_fatal);
1697     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1698                               &error_fatal);
1699     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1700                               "date", &error_fatal);
1701 }
1702 
1703 /* Returns whether we want to use VGA or not */
1704 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1705 {
1706     switch (vga_interface_type) {
1707     case VGA_NONE:
1708         return false;
1709     case VGA_DEVICE:
1710         return true;
1711     case VGA_STD:
1712     case VGA_VIRTIO:
1713         return pci_vga_init(pci_bus) != NULL;
1714     default:
1715         error_setg(errp,
1716                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1717         return false;
1718     }
1719 }
1720 
1721 static int spapr_pre_load(void *opaque)
1722 {
1723     int rc;
1724 
1725     rc = spapr_caps_pre_load(opaque);
1726     if (rc) {
1727         return rc;
1728     }
1729 
1730     return 0;
1731 }
1732 
1733 static int spapr_post_load(void *opaque, int version_id)
1734 {
1735     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1736     int err = 0;
1737 
1738     err = spapr_caps_post_migration(spapr);
1739     if (err) {
1740         return err;
1741     }
1742 
1743     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1744         CPUState *cs;
1745         CPU_FOREACH(cs) {
1746             PowerPCCPU *cpu = POWERPC_CPU(cs);
1747             icp_resend(ICP(cpu->intc));
1748         }
1749     }
1750 
1751     /* In earlier versions, there was no separate qdev for the PAPR
1752      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1753      * So when migrating from those versions, poke the incoming offset
1754      * value into the RTC device */
1755     if (version_id < 3) {
1756         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1757     }
1758 
1759     if (kvm_enabled() && spapr->patb_entry) {
1760         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1761         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1762         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1763 
1764         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1765         if (err) {
1766             error_report("Process table config unsupported by the host");
1767             return -EINVAL;
1768         }
1769     }
1770 
1771     return err;
1772 }
1773 
1774 static int spapr_pre_save(void *opaque)
1775 {
1776     int rc;
1777 
1778     rc = spapr_caps_pre_save(opaque);
1779     if (rc) {
1780         return rc;
1781     }
1782 
1783     return 0;
1784 }
1785 
1786 static bool version_before_3(void *opaque, int version_id)
1787 {
1788     return version_id < 3;
1789 }
1790 
1791 static bool spapr_pending_events_needed(void *opaque)
1792 {
1793     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1794     return !QTAILQ_EMPTY(&spapr->pending_events);
1795 }
1796 
1797 static const VMStateDescription vmstate_spapr_event_entry = {
1798     .name = "spapr_event_log_entry",
1799     .version_id = 1,
1800     .minimum_version_id = 1,
1801     .fields = (VMStateField[]) {
1802         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1803         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1804         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1805                                      NULL, extended_length),
1806         VMSTATE_END_OF_LIST()
1807     },
1808 };
1809 
1810 static const VMStateDescription vmstate_spapr_pending_events = {
1811     .name = "spapr_pending_events",
1812     .version_id = 1,
1813     .minimum_version_id = 1,
1814     .needed = spapr_pending_events_needed,
1815     .fields = (VMStateField[]) {
1816         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1817                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1818         VMSTATE_END_OF_LIST()
1819     },
1820 };
1821 
1822 static bool spapr_ov5_cas_needed(void *opaque)
1823 {
1824     sPAPRMachineState *spapr = opaque;
1825     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1826     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1827     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1828     bool cas_needed;
1829 
1830     /* Prior to the introduction of sPAPROptionVector, we had two option
1831      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1832      * Both of these options encode machine topology into the device-tree
1833      * in such a way that the now-booted OS should still be able to interact
1834      * appropriately with QEMU regardless of what options were actually
1835      * negotiatied on the source side.
1836      *
1837      * As such, we can avoid migrating the CAS-negotiated options if these
1838      * are the only options available on the current machine/platform.
1839      * Since these are the only options available for pseries-2.7 and
1840      * earlier, this allows us to maintain old->new/new->old migration
1841      * compatibility.
1842      *
1843      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1844      * via default pseries-2.8 machines and explicit command-line parameters.
1845      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1846      * of the actual CAS-negotiated values to continue working properly. For
1847      * example, availability of memory unplug depends on knowing whether
1848      * OV5_HP_EVT was negotiated via CAS.
1849      *
1850      * Thus, for any cases where the set of available CAS-negotiatable
1851      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1852      * include the CAS-negotiated options in the migration stream.
1853      */
1854     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1855     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1856 
1857     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1858      * the mask itself since in the future it's possible "legacy" bits may be
1859      * removed via machine options, which could generate a false positive
1860      * that breaks migration.
1861      */
1862     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1863     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1864 
1865     spapr_ovec_cleanup(ov5_mask);
1866     spapr_ovec_cleanup(ov5_legacy);
1867     spapr_ovec_cleanup(ov5_removed);
1868 
1869     return cas_needed;
1870 }
1871 
1872 static const VMStateDescription vmstate_spapr_ov5_cas = {
1873     .name = "spapr_option_vector_ov5_cas",
1874     .version_id = 1,
1875     .minimum_version_id = 1,
1876     .needed = spapr_ov5_cas_needed,
1877     .fields = (VMStateField[]) {
1878         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1879                                  vmstate_spapr_ovec, sPAPROptionVector),
1880         VMSTATE_END_OF_LIST()
1881     },
1882 };
1883 
1884 static bool spapr_patb_entry_needed(void *opaque)
1885 {
1886     sPAPRMachineState *spapr = opaque;
1887 
1888     return !!spapr->patb_entry;
1889 }
1890 
1891 static const VMStateDescription vmstate_spapr_patb_entry = {
1892     .name = "spapr_patb_entry",
1893     .version_id = 1,
1894     .minimum_version_id = 1,
1895     .needed = spapr_patb_entry_needed,
1896     .fields = (VMStateField[]) {
1897         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1898         VMSTATE_END_OF_LIST()
1899     },
1900 };
1901 
1902 static const VMStateDescription vmstate_spapr = {
1903     .name = "spapr",
1904     .version_id = 3,
1905     .minimum_version_id = 1,
1906     .pre_load = spapr_pre_load,
1907     .post_load = spapr_post_load,
1908     .pre_save = spapr_pre_save,
1909     .fields = (VMStateField[]) {
1910         /* used to be @next_irq */
1911         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1912 
1913         /* RTC offset */
1914         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1915 
1916         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1917         VMSTATE_END_OF_LIST()
1918     },
1919     .subsections = (const VMStateDescription*[]) {
1920         &vmstate_spapr_ov5_cas,
1921         &vmstate_spapr_patb_entry,
1922         &vmstate_spapr_pending_events,
1923         &vmstate_spapr_cap_htm,
1924         &vmstate_spapr_cap_vsx,
1925         &vmstate_spapr_cap_dfp,
1926         &vmstate_spapr_cap_cfpc,
1927         &vmstate_spapr_cap_sbbc,
1928         &vmstate_spapr_cap_ibs,
1929         NULL
1930     }
1931 };
1932 
1933 static int htab_save_setup(QEMUFile *f, void *opaque)
1934 {
1935     sPAPRMachineState *spapr = opaque;
1936 
1937     /* "Iteration" header */
1938     if (!spapr->htab_shift) {
1939         qemu_put_be32(f, -1);
1940     } else {
1941         qemu_put_be32(f, spapr->htab_shift);
1942     }
1943 
1944     if (spapr->htab) {
1945         spapr->htab_save_index = 0;
1946         spapr->htab_first_pass = true;
1947     } else {
1948         if (spapr->htab_shift) {
1949             assert(kvm_enabled());
1950         }
1951     }
1952 
1953 
1954     return 0;
1955 }
1956 
1957 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1958                             int chunkstart, int n_valid, int n_invalid)
1959 {
1960     qemu_put_be32(f, chunkstart);
1961     qemu_put_be16(f, n_valid);
1962     qemu_put_be16(f, n_invalid);
1963     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1964                     HASH_PTE_SIZE_64 * n_valid);
1965 }
1966 
1967 static void htab_save_end_marker(QEMUFile *f)
1968 {
1969     qemu_put_be32(f, 0);
1970     qemu_put_be16(f, 0);
1971     qemu_put_be16(f, 0);
1972 }
1973 
1974 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1975                                  int64_t max_ns)
1976 {
1977     bool has_timeout = max_ns != -1;
1978     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1979     int index = spapr->htab_save_index;
1980     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1981 
1982     assert(spapr->htab_first_pass);
1983 
1984     do {
1985         int chunkstart;
1986 
1987         /* Consume invalid HPTEs */
1988         while ((index < htabslots)
1989                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1990             CLEAN_HPTE(HPTE(spapr->htab, index));
1991             index++;
1992         }
1993 
1994         /* Consume valid HPTEs */
1995         chunkstart = index;
1996         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1997                && HPTE_VALID(HPTE(spapr->htab, index))) {
1998             CLEAN_HPTE(HPTE(spapr->htab, index));
1999             index++;
2000         }
2001 
2002         if (index > chunkstart) {
2003             int n_valid = index - chunkstart;
2004 
2005             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2006 
2007             if (has_timeout &&
2008                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2009                 break;
2010             }
2011         }
2012     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2013 
2014     if (index >= htabslots) {
2015         assert(index == htabslots);
2016         index = 0;
2017         spapr->htab_first_pass = false;
2018     }
2019     spapr->htab_save_index = index;
2020 }
2021 
2022 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2023                                 int64_t max_ns)
2024 {
2025     bool final = max_ns < 0;
2026     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2027     int examined = 0, sent = 0;
2028     int index = spapr->htab_save_index;
2029     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2030 
2031     assert(!spapr->htab_first_pass);
2032 
2033     do {
2034         int chunkstart, invalidstart;
2035 
2036         /* Consume non-dirty HPTEs */
2037         while ((index < htabslots)
2038                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2039             index++;
2040             examined++;
2041         }
2042 
2043         chunkstart = index;
2044         /* Consume valid dirty HPTEs */
2045         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2046                && HPTE_DIRTY(HPTE(spapr->htab, index))
2047                && HPTE_VALID(HPTE(spapr->htab, index))) {
2048             CLEAN_HPTE(HPTE(spapr->htab, index));
2049             index++;
2050             examined++;
2051         }
2052 
2053         invalidstart = index;
2054         /* Consume invalid dirty HPTEs */
2055         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2056                && HPTE_DIRTY(HPTE(spapr->htab, index))
2057                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2058             CLEAN_HPTE(HPTE(spapr->htab, index));
2059             index++;
2060             examined++;
2061         }
2062 
2063         if (index > chunkstart) {
2064             int n_valid = invalidstart - chunkstart;
2065             int n_invalid = index - invalidstart;
2066 
2067             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2068             sent += index - chunkstart;
2069 
2070             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2071                 break;
2072             }
2073         }
2074 
2075         if (examined >= htabslots) {
2076             break;
2077         }
2078 
2079         if (index >= htabslots) {
2080             assert(index == htabslots);
2081             index = 0;
2082         }
2083     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2084 
2085     if (index >= htabslots) {
2086         assert(index == htabslots);
2087         index = 0;
2088     }
2089 
2090     spapr->htab_save_index = index;
2091 
2092     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2093 }
2094 
2095 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2096 #define MAX_KVM_BUF_SIZE    2048
2097 
2098 static int htab_save_iterate(QEMUFile *f, void *opaque)
2099 {
2100     sPAPRMachineState *spapr = opaque;
2101     int fd;
2102     int rc = 0;
2103 
2104     /* Iteration header */
2105     if (!spapr->htab_shift) {
2106         qemu_put_be32(f, -1);
2107         return 1;
2108     } else {
2109         qemu_put_be32(f, 0);
2110     }
2111 
2112     if (!spapr->htab) {
2113         assert(kvm_enabled());
2114 
2115         fd = get_htab_fd(spapr);
2116         if (fd < 0) {
2117             return fd;
2118         }
2119 
2120         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2121         if (rc < 0) {
2122             return rc;
2123         }
2124     } else  if (spapr->htab_first_pass) {
2125         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2126     } else {
2127         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2128     }
2129 
2130     htab_save_end_marker(f);
2131 
2132     return rc;
2133 }
2134 
2135 static int htab_save_complete(QEMUFile *f, void *opaque)
2136 {
2137     sPAPRMachineState *spapr = opaque;
2138     int fd;
2139 
2140     /* Iteration header */
2141     if (!spapr->htab_shift) {
2142         qemu_put_be32(f, -1);
2143         return 0;
2144     } else {
2145         qemu_put_be32(f, 0);
2146     }
2147 
2148     if (!spapr->htab) {
2149         int rc;
2150 
2151         assert(kvm_enabled());
2152 
2153         fd = get_htab_fd(spapr);
2154         if (fd < 0) {
2155             return fd;
2156         }
2157 
2158         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2159         if (rc < 0) {
2160             return rc;
2161         }
2162     } else {
2163         if (spapr->htab_first_pass) {
2164             htab_save_first_pass(f, spapr, -1);
2165         }
2166         htab_save_later_pass(f, spapr, -1);
2167     }
2168 
2169     /* End marker */
2170     htab_save_end_marker(f);
2171 
2172     return 0;
2173 }
2174 
2175 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2176 {
2177     sPAPRMachineState *spapr = opaque;
2178     uint32_t section_hdr;
2179     int fd = -1;
2180     Error *local_err = NULL;
2181 
2182     if (version_id < 1 || version_id > 1) {
2183         error_report("htab_load() bad version");
2184         return -EINVAL;
2185     }
2186 
2187     section_hdr = qemu_get_be32(f);
2188 
2189     if (section_hdr == -1) {
2190         spapr_free_hpt(spapr);
2191         return 0;
2192     }
2193 
2194     if (section_hdr) {
2195         /* First section gives the htab size */
2196         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2197         if (local_err) {
2198             error_report_err(local_err);
2199             return -EINVAL;
2200         }
2201         return 0;
2202     }
2203 
2204     if (!spapr->htab) {
2205         assert(kvm_enabled());
2206 
2207         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2208         if (fd < 0) {
2209             error_report_err(local_err);
2210             return fd;
2211         }
2212     }
2213 
2214     while (true) {
2215         uint32_t index;
2216         uint16_t n_valid, n_invalid;
2217 
2218         index = qemu_get_be32(f);
2219         n_valid = qemu_get_be16(f);
2220         n_invalid = qemu_get_be16(f);
2221 
2222         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2223             /* End of Stream */
2224             break;
2225         }
2226 
2227         if ((index + n_valid + n_invalid) >
2228             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2229             /* Bad index in stream */
2230             error_report(
2231                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2232                 index, n_valid, n_invalid, spapr->htab_shift);
2233             return -EINVAL;
2234         }
2235 
2236         if (spapr->htab) {
2237             if (n_valid) {
2238                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2239                                 HASH_PTE_SIZE_64 * n_valid);
2240             }
2241             if (n_invalid) {
2242                 memset(HPTE(spapr->htab, index + n_valid), 0,
2243                        HASH_PTE_SIZE_64 * n_invalid);
2244             }
2245         } else {
2246             int rc;
2247 
2248             assert(fd >= 0);
2249 
2250             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2251             if (rc < 0) {
2252                 return rc;
2253             }
2254         }
2255     }
2256 
2257     if (!spapr->htab) {
2258         assert(fd >= 0);
2259         close(fd);
2260     }
2261 
2262     return 0;
2263 }
2264 
2265 static void htab_save_cleanup(void *opaque)
2266 {
2267     sPAPRMachineState *spapr = opaque;
2268 
2269     close_htab_fd(spapr);
2270 }
2271 
2272 static SaveVMHandlers savevm_htab_handlers = {
2273     .save_setup = htab_save_setup,
2274     .save_live_iterate = htab_save_iterate,
2275     .save_live_complete_precopy = htab_save_complete,
2276     .save_cleanup = htab_save_cleanup,
2277     .load_state = htab_load,
2278 };
2279 
2280 static void spapr_boot_set(void *opaque, const char *boot_device,
2281                            Error **errp)
2282 {
2283     MachineState *machine = MACHINE(opaque);
2284     machine->boot_order = g_strdup(boot_device);
2285 }
2286 
2287 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2288 {
2289     MachineState *machine = MACHINE(spapr);
2290     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2291     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2292     int i;
2293 
2294     for (i = 0; i < nr_lmbs; i++) {
2295         uint64_t addr;
2296 
2297         addr = i * lmb_size + spapr->hotplug_memory.base;
2298         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2299                                addr / lmb_size);
2300     }
2301 }
2302 
2303 /*
2304  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2305  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2306  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2307  */
2308 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2309 {
2310     int i;
2311 
2312     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2313         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2314                    " is not aligned to %llu MiB",
2315                    machine->ram_size,
2316                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2317         return;
2318     }
2319 
2320     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2321         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2322                    " is not aligned to %llu MiB",
2323                    machine->ram_size,
2324                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2325         return;
2326     }
2327 
2328     for (i = 0; i < nb_numa_nodes; i++) {
2329         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2330             error_setg(errp,
2331                        "Node %d memory size 0x%" PRIx64
2332                        " is not aligned to %llu MiB",
2333                        i, numa_info[i].node_mem,
2334                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2335             return;
2336         }
2337     }
2338 }
2339 
2340 /* find cpu slot in machine->possible_cpus by core_id */
2341 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2342 {
2343     int index = id / smp_threads;
2344 
2345     if (index >= ms->possible_cpus->len) {
2346         return NULL;
2347     }
2348     if (idx) {
2349         *idx = index;
2350     }
2351     return &ms->possible_cpus->cpus[index];
2352 }
2353 
2354 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2355 {
2356     Error *local_err = NULL;
2357     bool vsmt_user = !!spapr->vsmt;
2358     int kvm_smt = kvmppc_smt_threads();
2359     int ret;
2360 
2361     if (!kvm_enabled() && (smp_threads > 1)) {
2362         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2363                      "on a pseries machine");
2364         goto out;
2365     }
2366     if (!is_power_of_2(smp_threads)) {
2367         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2368                      "machine because it must be a power of 2", smp_threads);
2369         goto out;
2370     }
2371 
2372     /* Detemine the VSMT mode to use: */
2373     if (vsmt_user) {
2374         if (spapr->vsmt < smp_threads) {
2375             error_setg(&local_err, "Cannot support VSMT mode %d"
2376                          " because it must be >= threads/core (%d)",
2377                          spapr->vsmt, smp_threads);
2378             goto out;
2379         }
2380         /* In this case, spapr->vsmt has been set by the command line */
2381     } else {
2382         /*
2383          * Default VSMT value is tricky, because we need it to be as
2384          * consistent as possible (for migration), but this requires
2385          * changing it for at least some existing cases.  We pick 8 as
2386          * the value that we'd get with KVM on POWER8, the
2387          * overwhelmingly common case in production systems.
2388          */
2389         spapr->vsmt = MAX(8, smp_threads);
2390     }
2391 
2392     /* KVM: If necessary, set the SMT mode: */
2393     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2394         ret = kvmppc_set_smt_threads(spapr->vsmt);
2395         if (ret) {
2396             /* Looks like KVM isn't able to change VSMT mode */
2397             error_setg(&local_err,
2398                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2399                        spapr->vsmt, ret);
2400             /* We can live with that if the default one is big enough
2401              * for the number of threads, and a submultiple of the one
2402              * we want.  In this case we'll waste some vcpu ids, but
2403              * behaviour will be correct */
2404             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2405                 warn_report_err(local_err);
2406                 local_err = NULL;
2407                 goto out;
2408             } else {
2409                 if (!vsmt_user) {
2410                     error_append_hint(&local_err,
2411                                       "On PPC, a VM with %d threads/core"
2412                                       " on a host with %d threads/core"
2413                                       " requires the use of VSMT mode %d.\n",
2414                                       smp_threads, kvm_smt, spapr->vsmt);
2415                 }
2416                 kvmppc_hint_smt_possible(&local_err);
2417                 goto out;
2418             }
2419         }
2420     }
2421     /* else TCG: nothing to do currently */
2422 out:
2423     error_propagate(errp, local_err);
2424 }
2425 
2426 static void spapr_init_cpus(sPAPRMachineState *spapr)
2427 {
2428     MachineState *machine = MACHINE(spapr);
2429     MachineClass *mc = MACHINE_GET_CLASS(machine);
2430     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2431     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2432     const CPUArchIdList *possible_cpus;
2433     int boot_cores_nr = smp_cpus / smp_threads;
2434     int i;
2435 
2436     possible_cpus = mc->possible_cpu_arch_ids(machine);
2437     if (mc->has_hotpluggable_cpus) {
2438         if (smp_cpus % smp_threads) {
2439             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2440                          smp_cpus, smp_threads);
2441             exit(1);
2442         }
2443         if (max_cpus % smp_threads) {
2444             error_report("max_cpus (%u) must be multiple of threads (%u)",
2445                          max_cpus, smp_threads);
2446             exit(1);
2447         }
2448     } else {
2449         if (max_cpus != smp_cpus) {
2450             error_report("This machine version does not support CPU hotplug");
2451             exit(1);
2452         }
2453         boot_cores_nr = possible_cpus->len;
2454     }
2455 
2456     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2457      * call xics_max_server_number() or spapr_vcpu_id().
2458      */
2459     spapr_set_vsmt_mode(spapr, &error_fatal);
2460 
2461     if (smc->pre_2_10_has_unused_icps) {
2462         int i;
2463 
2464         for (i = 0; i < xics_max_server_number(spapr); i++) {
2465             /* Dummy entries get deregistered when real ICPState objects
2466              * are registered during CPU core hotplug.
2467              */
2468             pre_2_10_vmstate_register_dummy_icp(i);
2469         }
2470     }
2471 
2472     for (i = 0; i < possible_cpus->len; i++) {
2473         int core_id = i * smp_threads;
2474 
2475         if (mc->has_hotpluggable_cpus) {
2476             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2477                                    spapr_vcpu_id(spapr, core_id));
2478         }
2479 
2480         if (i < boot_cores_nr) {
2481             Object *core  = object_new(type);
2482             int nr_threads = smp_threads;
2483 
2484             /* Handle the partially filled core for older machine types */
2485             if ((i + 1) * smp_threads >= smp_cpus) {
2486                 nr_threads = smp_cpus - i * smp_threads;
2487             }
2488 
2489             object_property_set_int(core, nr_threads, "nr-threads",
2490                                     &error_fatal);
2491             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2492                                     &error_fatal);
2493             object_property_set_bool(core, true, "realized", &error_fatal);
2494         }
2495     }
2496 }
2497 
2498 /* pSeries LPAR / sPAPR hardware init */
2499 static void spapr_machine_init(MachineState *machine)
2500 {
2501     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2502     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2503     const char *kernel_filename = machine->kernel_filename;
2504     const char *initrd_filename = machine->initrd_filename;
2505     PCIHostState *phb;
2506     int i;
2507     MemoryRegion *sysmem = get_system_memory();
2508     MemoryRegion *ram = g_new(MemoryRegion, 1);
2509     hwaddr node0_size = spapr_node0_size(machine);
2510     long load_limit, fw_size;
2511     char *filename;
2512     Error *resize_hpt_err = NULL;
2513 
2514     msi_nonbroken = true;
2515 
2516     QLIST_INIT(&spapr->phbs);
2517     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2518 
2519     /* Check HPT resizing availability */
2520     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2521     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2522         /*
2523          * If the user explicitly requested a mode we should either
2524          * supply it, or fail completely (which we do below).  But if
2525          * it's not set explicitly, we reset our mode to something
2526          * that works
2527          */
2528         if (resize_hpt_err) {
2529             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2530             error_free(resize_hpt_err);
2531             resize_hpt_err = NULL;
2532         } else {
2533             spapr->resize_hpt = smc->resize_hpt_default;
2534         }
2535     }
2536 
2537     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2538 
2539     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2540         /*
2541          * User requested HPT resize, but this host can't supply it.  Bail out
2542          */
2543         error_report_err(resize_hpt_err);
2544         exit(1);
2545     }
2546 
2547     spapr->rma_size = node0_size;
2548 
2549     /* With KVM, we don't actually know whether KVM supports an
2550      * unbounded RMA (PR KVM) or is limited by the hash table size
2551      * (HV KVM using VRMA), so we always assume the latter
2552      *
2553      * In that case, we also limit the initial allocations for RTAS
2554      * etc... to 256M since we have no way to know what the VRMA size
2555      * is going to be as it depends on the size of the hash table
2556      * which isn't determined yet.
2557      */
2558     if (kvm_enabled()) {
2559         spapr->vrma_adjust = 1;
2560         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2561     }
2562 
2563     /* Actually we don't support unbounded RMA anymore since we added
2564      * proper emulation of HV mode. The max we can get is 16G which
2565      * also happens to be what we configure for PAPR mode so make sure
2566      * we don't do anything bigger than that
2567      */
2568     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2569 
2570     if (spapr->rma_size > node0_size) {
2571         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2572                      spapr->rma_size);
2573         exit(1);
2574     }
2575 
2576     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2577     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2578 
2579     /* Set up Interrupt Controller before we create the VCPUs */
2580     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2581 
2582     /* Set up containers for ibm,client-architecture-support negotiated options
2583      */
2584     spapr->ov5 = spapr_ovec_new();
2585     spapr->ov5_cas = spapr_ovec_new();
2586 
2587     if (smc->dr_lmb_enabled) {
2588         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2589         spapr_validate_node_memory(machine, &error_fatal);
2590     }
2591 
2592     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2593     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2594         /* KVM and TCG always allow GTSE with radix... */
2595         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2596     }
2597     /* ... but not with hash (currently). */
2598 
2599     /* advertise support for dedicated HP event source to guests */
2600     if (spapr->use_hotplug_event_source) {
2601         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2602     }
2603 
2604     /* advertise support for HPT resizing */
2605     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2606         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2607     }
2608 
2609     /* advertise support for ibm,dyamic-memory-v2 */
2610     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2611 
2612     /* init CPUs */
2613     spapr_init_cpus(spapr);
2614 
2615     if (kvm_enabled()) {
2616         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2617         kvmppc_enable_logical_ci_hcalls();
2618         kvmppc_enable_set_mode_hcall();
2619 
2620         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2621         kvmppc_enable_clear_ref_mod_hcalls();
2622     }
2623 
2624     /* allocate RAM */
2625     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2626                                          machine->ram_size);
2627     memory_region_add_subregion(sysmem, 0, ram);
2628 
2629     /* initialize hotplug memory address space */
2630     if (machine->ram_size < machine->maxram_size) {
2631         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2632         /*
2633          * Limit the number of hotpluggable memory slots to half the number
2634          * slots that KVM supports, leaving the other half for PCI and other
2635          * devices. However ensure that number of slots doesn't drop below 32.
2636          */
2637         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2638                            SPAPR_MAX_RAM_SLOTS;
2639 
2640         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2641             max_memslots = SPAPR_MAX_RAM_SLOTS;
2642         }
2643         if (machine->ram_slots > max_memslots) {
2644             error_report("Specified number of memory slots %"
2645                          PRIu64" exceeds max supported %d",
2646                          machine->ram_slots, max_memslots);
2647             exit(1);
2648         }
2649 
2650         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2651                                               SPAPR_HOTPLUG_MEM_ALIGN);
2652         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2653                            "hotplug-memory", hotplug_mem_size);
2654         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2655                                     &spapr->hotplug_memory.mr);
2656     }
2657 
2658     if (smc->dr_lmb_enabled) {
2659         spapr_create_lmb_dr_connectors(spapr);
2660     }
2661 
2662     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2663     if (!filename) {
2664         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2665         exit(1);
2666     }
2667     spapr->rtas_size = get_image_size(filename);
2668     if (spapr->rtas_size < 0) {
2669         error_report("Could not get size of LPAR rtas '%s'", filename);
2670         exit(1);
2671     }
2672     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2673     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2674         error_report("Could not load LPAR rtas '%s'", filename);
2675         exit(1);
2676     }
2677     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2678         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2679                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2680         exit(1);
2681     }
2682     g_free(filename);
2683 
2684     /* Set up RTAS event infrastructure */
2685     spapr_events_init(spapr);
2686 
2687     /* Set up the RTC RTAS interfaces */
2688     spapr_rtc_create(spapr);
2689 
2690     /* Set up VIO bus */
2691     spapr->vio_bus = spapr_vio_bus_init();
2692 
2693     for (i = 0; i < serial_max_hds(); i++) {
2694         if (serial_hd(i)) {
2695             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2696         }
2697     }
2698 
2699     /* We always have at least the nvram device on VIO */
2700     spapr_create_nvram(spapr);
2701 
2702     /* Set up PCI */
2703     spapr_pci_rtas_init();
2704 
2705     phb = spapr_create_phb(spapr, 0);
2706 
2707     for (i = 0; i < nb_nics; i++) {
2708         NICInfo *nd = &nd_table[i];
2709 
2710         if (!nd->model) {
2711             nd->model = g_strdup("spapr-vlan");
2712         }
2713 
2714         if (g_str_equal(nd->model, "spapr-vlan") ||
2715             g_str_equal(nd->model, "ibmveth")) {
2716             spapr_vlan_create(spapr->vio_bus, nd);
2717         } else {
2718             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2719         }
2720     }
2721 
2722     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2723         spapr_vscsi_create(spapr->vio_bus);
2724     }
2725 
2726     /* Graphics */
2727     if (spapr_vga_init(phb->bus, &error_fatal)) {
2728         spapr->has_graphics = true;
2729         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2730     }
2731 
2732     if (machine->usb) {
2733         if (smc->use_ohci_by_default) {
2734             pci_create_simple(phb->bus, -1, "pci-ohci");
2735         } else {
2736             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2737         }
2738 
2739         if (spapr->has_graphics) {
2740             USBBus *usb_bus = usb_bus_find(-1);
2741 
2742             usb_create_simple(usb_bus, "usb-kbd");
2743             usb_create_simple(usb_bus, "usb-mouse");
2744         }
2745     }
2746 
2747     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2748         error_report(
2749             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2750             MIN_RMA_SLOF);
2751         exit(1);
2752     }
2753 
2754     if (kernel_filename) {
2755         uint64_t lowaddr = 0;
2756 
2757         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2758                                       NULL, NULL, &lowaddr, NULL, 1,
2759                                       PPC_ELF_MACHINE, 0, 0);
2760         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2761             spapr->kernel_size = load_elf(kernel_filename,
2762                                           translate_kernel_address, NULL, NULL,
2763                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2764                                           0, 0);
2765             spapr->kernel_le = spapr->kernel_size > 0;
2766         }
2767         if (spapr->kernel_size < 0) {
2768             error_report("error loading %s: %s", kernel_filename,
2769                          load_elf_strerror(spapr->kernel_size));
2770             exit(1);
2771         }
2772 
2773         /* load initrd */
2774         if (initrd_filename) {
2775             /* Try to locate the initrd in the gap between the kernel
2776              * and the firmware. Add a bit of space just in case
2777              */
2778             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2779                                   + 0x1ffff) & ~0xffff;
2780             spapr->initrd_size = load_image_targphys(initrd_filename,
2781                                                      spapr->initrd_base,
2782                                                      load_limit
2783                                                      - spapr->initrd_base);
2784             if (spapr->initrd_size < 0) {
2785                 error_report("could not load initial ram disk '%s'",
2786                              initrd_filename);
2787                 exit(1);
2788             }
2789         }
2790     }
2791 
2792     if (bios_name == NULL) {
2793         bios_name = FW_FILE_NAME;
2794     }
2795     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2796     if (!filename) {
2797         error_report("Could not find LPAR firmware '%s'", bios_name);
2798         exit(1);
2799     }
2800     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2801     if (fw_size <= 0) {
2802         error_report("Could not load LPAR firmware '%s'", filename);
2803         exit(1);
2804     }
2805     g_free(filename);
2806 
2807     /* FIXME: Should register things through the MachineState's qdev
2808      * interface, this is a legacy from the sPAPREnvironment structure
2809      * which predated MachineState but had a similar function */
2810     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2811     register_savevm_live(NULL, "spapr/htab", -1, 1,
2812                          &savevm_htab_handlers, spapr);
2813 
2814     qemu_register_boot_set(spapr_boot_set, spapr);
2815 
2816     if (kvm_enabled()) {
2817         /* to stop and start vmclock */
2818         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2819                                          &spapr->tb);
2820 
2821         kvmppc_spapr_enable_inkernel_multitce();
2822     }
2823 }
2824 
2825 static int spapr_kvm_type(const char *vm_type)
2826 {
2827     if (!vm_type) {
2828         return 0;
2829     }
2830 
2831     if (!strcmp(vm_type, "HV")) {
2832         return 1;
2833     }
2834 
2835     if (!strcmp(vm_type, "PR")) {
2836         return 2;
2837     }
2838 
2839     error_report("Unknown kvm-type specified '%s'", vm_type);
2840     exit(1);
2841 }
2842 
2843 /*
2844  * Implementation of an interface to adjust firmware path
2845  * for the bootindex property handling.
2846  */
2847 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2848                                    DeviceState *dev)
2849 {
2850 #define CAST(type, obj, name) \
2851     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2852     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2853     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2854     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2855 
2856     if (d) {
2857         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2858         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2859         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2860 
2861         if (spapr) {
2862             /*
2863              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2864              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2865              * in the top 16 bits of the 64-bit LUN
2866              */
2867             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2868             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2869                                    (uint64_t)id << 48);
2870         } else if (virtio) {
2871             /*
2872              * We use SRP luns of the form 01000000 | (target << 8) | lun
2873              * in the top 32 bits of the 64-bit LUN
2874              * Note: the quote above is from SLOF and it is wrong,
2875              * the actual binding is:
2876              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2877              */
2878             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2879             if (d->lun >= 256) {
2880                 /* Use the LUN "flat space addressing method" */
2881                 id |= 0x4000;
2882             }
2883             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2884                                    (uint64_t)id << 32);
2885         } else if (usb) {
2886             /*
2887              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2888              * in the top 32 bits of the 64-bit LUN
2889              */
2890             unsigned usb_port = atoi(usb->port->path);
2891             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2892             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2893                                    (uint64_t)id << 32);
2894         }
2895     }
2896 
2897     /*
2898      * SLOF probes the USB devices, and if it recognizes that the device is a
2899      * storage device, it changes its name to "storage" instead of "usb-host",
2900      * and additionally adds a child node for the SCSI LUN, so the correct
2901      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2902      */
2903     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2904         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2905         if (usb_host_dev_is_scsi_storage(usbdev)) {
2906             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2907         }
2908     }
2909 
2910     if (phb) {
2911         /* Replace "pci" with "pci@800000020000000" */
2912         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2913     }
2914 
2915     if (vsc) {
2916         /* Same logic as virtio above */
2917         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2918         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2919     }
2920 
2921     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2922         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2923         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2924         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2925     }
2926 
2927     return NULL;
2928 }
2929 
2930 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2931 {
2932     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2933 
2934     return g_strdup(spapr->kvm_type);
2935 }
2936 
2937 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2938 {
2939     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2940 
2941     g_free(spapr->kvm_type);
2942     spapr->kvm_type = g_strdup(value);
2943 }
2944 
2945 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2946 {
2947     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2948 
2949     return spapr->use_hotplug_event_source;
2950 }
2951 
2952 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2953                                             Error **errp)
2954 {
2955     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2956 
2957     spapr->use_hotplug_event_source = value;
2958 }
2959 
2960 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2961 {
2962     return true;
2963 }
2964 
2965 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2966 {
2967     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2968 
2969     switch (spapr->resize_hpt) {
2970     case SPAPR_RESIZE_HPT_DEFAULT:
2971         return g_strdup("default");
2972     case SPAPR_RESIZE_HPT_DISABLED:
2973         return g_strdup("disabled");
2974     case SPAPR_RESIZE_HPT_ENABLED:
2975         return g_strdup("enabled");
2976     case SPAPR_RESIZE_HPT_REQUIRED:
2977         return g_strdup("required");
2978     }
2979     g_assert_not_reached();
2980 }
2981 
2982 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2983 {
2984     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2985 
2986     if (strcmp(value, "default") == 0) {
2987         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2988     } else if (strcmp(value, "disabled") == 0) {
2989         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2990     } else if (strcmp(value, "enabled") == 0) {
2991         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2992     } else if (strcmp(value, "required") == 0) {
2993         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2994     } else {
2995         error_setg(errp, "Bad value for \"resize-hpt\" property");
2996     }
2997 }
2998 
2999 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3000                                    void *opaque, Error **errp)
3001 {
3002     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3003 }
3004 
3005 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3006                                    void *opaque, Error **errp)
3007 {
3008     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3009 }
3010 
3011 static void spapr_instance_init(Object *obj)
3012 {
3013     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3014 
3015     spapr->htab_fd = -1;
3016     spapr->use_hotplug_event_source = true;
3017     object_property_add_str(obj, "kvm-type",
3018                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3019     object_property_set_description(obj, "kvm-type",
3020                                     "Specifies the KVM virtualization mode (HV, PR)",
3021                                     NULL);
3022     object_property_add_bool(obj, "modern-hotplug-events",
3023                             spapr_get_modern_hotplug_events,
3024                             spapr_set_modern_hotplug_events,
3025                             NULL);
3026     object_property_set_description(obj, "modern-hotplug-events",
3027                                     "Use dedicated hotplug event mechanism in"
3028                                     " place of standard EPOW events when possible"
3029                                     " (required for memory hot-unplug support)",
3030                                     NULL);
3031     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3032                             "Maximum permitted CPU compatibility mode",
3033                             &error_fatal);
3034 
3035     object_property_add_str(obj, "resize-hpt",
3036                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3037     object_property_set_description(obj, "resize-hpt",
3038                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3039                                     NULL);
3040     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3041                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3042     object_property_set_description(obj, "vsmt",
3043                                     "Virtual SMT: KVM behaves as if this were"
3044                                     " the host's SMT mode", &error_abort);
3045     object_property_add_bool(obj, "vfio-no-msix-emulation",
3046                              spapr_get_msix_emulation, NULL, NULL);
3047 }
3048 
3049 static void spapr_machine_finalizefn(Object *obj)
3050 {
3051     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3052 
3053     g_free(spapr->kvm_type);
3054 }
3055 
3056 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3057 {
3058     cpu_synchronize_state(cs);
3059     ppc_cpu_do_system_reset(cs);
3060 }
3061 
3062 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3063 {
3064     CPUState *cs;
3065 
3066     CPU_FOREACH(cs) {
3067         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3068     }
3069 }
3070 
3071 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3072                            uint32_t node, bool dedicated_hp_event_source,
3073                            Error **errp)
3074 {
3075     sPAPRDRConnector *drc;
3076     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3077     int i, fdt_offset, fdt_size;
3078     void *fdt;
3079     uint64_t addr = addr_start;
3080     bool hotplugged = spapr_drc_hotplugged(dev);
3081     Error *local_err = NULL;
3082 
3083     for (i = 0; i < nr_lmbs; i++) {
3084         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3085                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3086         g_assert(drc);
3087 
3088         fdt = create_device_tree(&fdt_size);
3089         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3090                                                 SPAPR_MEMORY_BLOCK_SIZE);
3091 
3092         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3093         if (local_err) {
3094             while (addr > addr_start) {
3095                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3096                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3097                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3098                 spapr_drc_detach(drc);
3099             }
3100             g_free(fdt);
3101             error_propagate(errp, local_err);
3102             return;
3103         }
3104         if (!hotplugged) {
3105             spapr_drc_reset(drc);
3106         }
3107         addr += SPAPR_MEMORY_BLOCK_SIZE;
3108     }
3109     /* send hotplug notification to the
3110      * guest only in case of hotplugged memory
3111      */
3112     if (hotplugged) {
3113         if (dedicated_hp_event_source) {
3114             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3115                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3116             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3117                                                    nr_lmbs,
3118                                                    spapr_drc_index(drc));
3119         } else {
3120             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3121                                            nr_lmbs);
3122         }
3123     }
3124 }
3125 
3126 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3127                               uint32_t node, Error **errp)
3128 {
3129     Error *local_err = NULL;
3130     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3131     PCDIMMDevice *dimm = PC_DIMM(dev);
3132     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3133     MemoryRegion *mr;
3134     uint64_t align, size, addr;
3135 
3136     mr = ddc->get_memory_region(dimm, &local_err);
3137     if (local_err) {
3138         goto out;
3139     }
3140     align = memory_region_get_alignment(mr);
3141     size = memory_region_size(mr);
3142 
3143     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
3144     if (local_err) {
3145         goto out;
3146     }
3147 
3148     addr = object_property_get_uint(OBJECT(dimm),
3149                                     PC_DIMM_ADDR_PROP, &local_err);
3150     if (local_err) {
3151         goto out_unplug;
3152     }
3153 
3154     spapr_add_lmbs(dev, addr, size, node,
3155                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3156                    &local_err);
3157     if (local_err) {
3158         goto out_unplug;
3159     }
3160 
3161     return;
3162 
3163 out_unplug:
3164     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
3165 out:
3166     error_propagate(errp, local_err);
3167 }
3168 
3169 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3170                                   Error **errp)
3171 {
3172     PCDIMMDevice *dimm = PC_DIMM(dev);
3173     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3174     MemoryRegion *mr;
3175     uint64_t size;
3176     char *mem_dev;
3177 
3178     mr = ddc->get_memory_region(dimm, errp);
3179     if (!mr) {
3180         return;
3181     }
3182     size = memory_region_size(mr);
3183 
3184     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3185         error_setg(errp, "Hotplugged memory size must be a multiple of "
3186                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
3187         return;
3188     }
3189 
3190     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
3191     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
3192         error_setg(errp, "Memory backend has bad page size. "
3193                    "Use 'memory-backend-file' with correct mem-path.");
3194         goto out;
3195     }
3196 
3197 out:
3198     g_free(mem_dev);
3199 }
3200 
3201 struct sPAPRDIMMState {
3202     PCDIMMDevice *dimm;
3203     uint32_t nr_lmbs;
3204     QTAILQ_ENTRY(sPAPRDIMMState) next;
3205 };
3206 
3207 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3208                                                        PCDIMMDevice *dimm)
3209 {
3210     sPAPRDIMMState *dimm_state = NULL;
3211 
3212     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3213         if (dimm_state->dimm == dimm) {
3214             break;
3215         }
3216     }
3217     return dimm_state;
3218 }
3219 
3220 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3221                                                       uint32_t nr_lmbs,
3222                                                       PCDIMMDevice *dimm)
3223 {
3224     sPAPRDIMMState *ds = NULL;
3225 
3226     /*
3227      * If this request is for a DIMM whose removal had failed earlier
3228      * (due to guest's refusal to remove the LMBs), we would have this
3229      * dimm already in the pending_dimm_unplugs list. In that
3230      * case don't add again.
3231      */
3232     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3233     if (!ds) {
3234         ds = g_malloc0(sizeof(sPAPRDIMMState));
3235         ds->nr_lmbs = nr_lmbs;
3236         ds->dimm = dimm;
3237         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3238     }
3239     return ds;
3240 }
3241 
3242 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3243                                               sPAPRDIMMState *dimm_state)
3244 {
3245     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3246     g_free(dimm_state);
3247 }
3248 
3249 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3250                                                         PCDIMMDevice *dimm)
3251 {
3252     sPAPRDRConnector *drc;
3253     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3254     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3255     uint64_t size = memory_region_size(mr);
3256     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3257     uint32_t avail_lmbs = 0;
3258     uint64_t addr_start, addr;
3259     int i;
3260 
3261     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3262                                          &error_abort);
3263 
3264     addr = addr_start;
3265     for (i = 0; i < nr_lmbs; i++) {
3266         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3267                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3268         g_assert(drc);
3269         if (drc->dev) {
3270             avail_lmbs++;
3271         }
3272         addr += SPAPR_MEMORY_BLOCK_SIZE;
3273     }
3274 
3275     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3276 }
3277 
3278 /* Callback to be called during DRC release. */
3279 void spapr_lmb_release(DeviceState *dev)
3280 {
3281     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_hotplug_handler(dev));
3282     PCDIMMDevice *dimm = PC_DIMM(dev);
3283     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3284     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3285     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3286 
3287     /* This information will get lost if a migration occurs
3288      * during the unplug process. In this case recover it. */
3289     if (ds == NULL) {
3290         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3291         g_assert(ds);
3292         /* The DRC being examined by the caller at least must be counted */
3293         g_assert(ds->nr_lmbs);
3294     }
3295 
3296     if (--ds->nr_lmbs) {
3297         return;
3298     }
3299 
3300     /*
3301      * Now that all the LMBs have been removed by the guest, call the
3302      * pc-dimm unplug handler to cleanup up the pc-dimm device.
3303      */
3304     pc_dimm_memory_unplug(dev, &spapr->hotplug_memory, mr);
3305     object_unparent(OBJECT(dev));
3306     spapr_pending_dimm_unplugs_remove(spapr, ds);
3307 }
3308 
3309 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3310                                         DeviceState *dev, Error **errp)
3311 {
3312     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3313     Error *local_err = NULL;
3314     PCDIMMDevice *dimm = PC_DIMM(dev);
3315     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3316     MemoryRegion *mr;
3317     uint32_t nr_lmbs;
3318     uint64_t size, addr_start, addr;
3319     int i;
3320     sPAPRDRConnector *drc;
3321 
3322     mr = ddc->get_memory_region(dimm, &local_err);
3323     if (local_err) {
3324         goto out;
3325     }
3326     size = memory_region_size(mr);
3327     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3328 
3329     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3330                                          &local_err);
3331     if (local_err) {
3332         goto out;
3333     }
3334 
3335     /*
3336      * An existing pending dimm state for this DIMM means that there is an
3337      * unplug operation in progress, waiting for the spapr_lmb_release
3338      * callback to complete the job (BQL can't cover that far). In this case,
3339      * bail out to avoid detaching DRCs that were already released.
3340      */
3341     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3342         error_setg(&local_err,
3343                    "Memory unplug already in progress for device %s",
3344                    dev->id);
3345         goto out;
3346     }
3347 
3348     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3349 
3350     addr = addr_start;
3351     for (i = 0; i < nr_lmbs; i++) {
3352         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3353                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3354         g_assert(drc);
3355 
3356         spapr_drc_detach(drc);
3357         addr += SPAPR_MEMORY_BLOCK_SIZE;
3358     }
3359 
3360     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3361                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3362     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3363                                               nr_lmbs, spapr_drc_index(drc));
3364 out:
3365     error_propagate(errp, local_err);
3366 }
3367 
3368 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3369                                            sPAPRMachineState *spapr)
3370 {
3371     PowerPCCPU *cpu = POWERPC_CPU(cs);
3372     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3373     int id = spapr_get_vcpu_id(cpu);
3374     void *fdt;
3375     int offset, fdt_size;
3376     char *nodename;
3377 
3378     fdt = create_device_tree(&fdt_size);
3379     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3380     offset = fdt_add_subnode(fdt, 0, nodename);
3381 
3382     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3383     g_free(nodename);
3384 
3385     *fdt_offset = offset;
3386     return fdt;
3387 }
3388 
3389 /* Callback to be called during DRC release. */
3390 void spapr_core_release(DeviceState *dev)
3391 {
3392     MachineState *ms = MACHINE(qdev_get_hotplug_handler(dev));
3393     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3394     CPUCore *cc = CPU_CORE(dev);
3395     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3396 
3397     if (smc->pre_2_10_has_unused_icps) {
3398         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3399         int i;
3400 
3401         for (i = 0; i < cc->nr_threads; i++) {
3402             CPUState *cs = CPU(sc->threads[i]);
3403 
3404             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3405         }
3406     }
3407 
3408     assert(core_slot);
3409     core_slot->cpu = NULL;
3410     object_unparent(OBJECT(dev));
3411 }
3412 
3413 static
3414 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3415                                Error **errp)
3416 {
3417     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3418     int index;
3419     sPAPRDRConnector *drc;
3420     CPUCore *cc = CPU_CORE(dev);
3421 
3422     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3423         error_setg(errp, "Unable to find CPU core with core-id: %d",
3424                    cc->core_id);
3425         return;
3426     }
3427     if (index == 0) {
3428         error_setg(errp, "Boot CPU core may not be unplugged");
3429         return;
3430     }
3431 
3432     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3433                           spapr_vcpu_id(spapr, cc->core_id));
3434     g_assert(drc);
3435 
3436     spapr_drc_detach(drc);
3437 
3438     spapr_hotplug_req_remove_by_index(drc);
3439 }
3440 
3441 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3442                             Error **errp)
3443 {
3444     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3445     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3446     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3447     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3448     CPUCore *cc = CPU_CORE(dev);
3449     CPUState *cs = CPU(core->threads[0]);
3450     sPAPRDRConnector *drc;
3451     Error *local_err = NULL;
3452     CPUArchId *core_slot;
3453     int index;
3454     bool hotplugged = spapr_drc_hotplugged(dev);
3455 
3456     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3457     if (!core_slot) {
3458         error_setg(errp, "Unable to find CPU core with core-id: %d",
3459                    cc->core_id);
3460         return;
3461     }
3462     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3463                           spapr_vcpu_id(spapr, cc->core_id));
3464 
3465     g_assert(drc || !mc->has_hotpluggable_cpus);
3466 
3467     if (drc) {
3468         void *fdt;
3469         int fdt_offset;
3470 
3471         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3472 
3473         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3474         if (local_err) {
3475             g_free(fdt);
3476             error_propagate(errp, local_err);
3477             return;
3478         }
3479 
3480         if (hotplugged) {
3481             /*
3482              * Send hotplug notification interrupt to the guest only
3483              * in case of hotplugged CPUs.
3484              */
3485             spapr_hotplug_req_add_by_index(drc);
3486         } else {
3487             spapr_drc_reset(drc);
3488         }
3489     }
3490 
3491     core_slot->cpu = OBJECT(dev);
3492 
3493     if (smc->pre_2_10_has_unused_icps) {
3494         int i;
3495 
3496         for (i = 0; i < cc->nr_threads; i++) {
3497             cs = CPU(core->threads[i]);
3498             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3499         }
3500     }
3501 }
3502 
3503 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3504                                 Error **errp)
3505 {
3506     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3507     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3508     Error *local_err = NULL;
3509     CPUCore *cc = CPU_CORE(dev);
3510     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3511     const char *type = object_get_typename(OBJECT(dev));
3512     CPUArchId *core_slot;
3513     int index;
3514 
3515     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3516         error_setg(&local_err, "CPU hotplug not supported for this machine");
3517         goto out;
3518     }
3519 
3520     if (strcmp(base_core_type, type)) {
3521         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3522         goto out;
3523     }
3524 
3525     if (cc->core_id % smp_threads) {
3526         error_setg(&local_err, "invalid core id %d", cc->core_id);
3527         goto out;
3528     }
3529 
3530     /*
3531      * In general we should have homogeneous threads-per-core, but old
3532      * (pre hotplug support) machine types allow the last core to have
3533      * reduced threads as a compatibility hack for when we allowed
3534      * total vcpus not a multiple of threads-per-core.
3535      */
3536     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3537         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3538                    cc->nr_threads, smp_threads);
3539         goto out;
3540     }
3541 
3542     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3543     if (!core_slot) {
3544         error_setg(&local_err, "core id %d out of range", cc->core_id);
3545         goto out;
3546     }
3547 
3548     if (core_slot->cpu) {
3549         error_setg(&local_err, "core %d already populated", cc->core_id);
3550         goto out;
3551     }
3552 
3553     numa_cpu_pre_plug(core_slot, dev, &local_err);
3554 
3555 out:
3556     error_propagate(errp, local_err);
3557 }
3558 
3559 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3560                                       DeviceState *dev, Error **errp)
3561 {
3562     MachineState *ms = MACHINE(hotplug_dev);
3563     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3564 
3565     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3566         int node;
3567 
3568         if (!smc->dr_lmb_enabled) {
3569             error_setg(errp, "Memory hotplug not supported for this machine");
3570             return;
3571         }
3572         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3573         if (*errp) {
3574             return;
3575         }
3576         if (node < 0 || node >= MAX_NODES) {
3577             error_setg(errp, "Invaild node %d", node);
3578             return;
3579         }
3580 
3581         spapr_memory_plug(hotplug_dev, dev, node, errp);
3582     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3583         spapr_core_plug(hotplug_dev, dev, errp);
3584     }
3585 }
3586 
3587 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3588                                                 DeviceState *dev, Error **errp)
3589 {
3590     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3591     MachineClass *mc = MACHINE_GET_CLASS(sms);
3592 
3593     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3594         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3595             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3596         } else {
3597             /* NOTE: this means there is a window after guest reset, prior to
3598              * CAS negotiation, where unplug requests will fail due to the
3599              * capability not being detected yet. This is a bit different than
3600              * the case with PCI unplug, where the events will be queued and
3601              * eventually handled by the guest after boot
3602              */
3603             error_setg(errp, "Memory hot unplug not supported for this guest");
3604         }
3605     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3606         if (!mc->has_hotpluggable_cpus) {
3607             error_setg(errp, "CPU hot unplug not supported on this machine");
3608             return;
3609         }
3610         spapr_core_unplug_request(hotplug_dev, dev, errp);
3611     }
3612 }
3613 
3614 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3615                                           DeviceState *dev, Error **errp)
3616 {
3617     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3618         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3619     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3620         spapr_core_pre_plug(hotplug_dev, dev, errp);
3621     }
3622 }
3623 
3624 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3625                                                  DeviceState *dev)
3626 {
3627     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3628         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3629         return HOTPLUG_HANDLER(machine);
3630     }
3631     return NULL;
3632 }
3633 
3634 static CpuInstanceProperties
3635 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3636 {
3637     CPUArchId *core_slot;
3638     MachineClass *mc = MACHINE_GET_CLASS(machine);
3639 
3640     /* make sure possible_cpu are intialized */
3641     mc->possible_cpu_arch_ids(machine);
3642     /* get CPU core slot containing thread that matches cpu_index */
3643     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3644     assert(core_slot);
3645     return core_slot->props;
3646 }
3647 
3648 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3649 {
3650     return idx / smp_cores % nb_numa_nodes;
3651 }
3652 
3653 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3654 {
3655     int i;
3656     const char *core_type;
3657     int spapr_max_cores = max_cpus / smp_threads;
3658     MachineClass *mc = MACHINE_GET_CLASS(machine);
3659 
3660     if (!mc->has_hotpluggable_cpus) {
3661         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3662     }
3663     if (machine->possible_cpus) {
3664         assert(machine->possible_cpus->len == spapr_max_cores);
3665         return machine->possible_cpus;
3666     }
3667 
3668     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3669     if (!core_type) {
3670         error_report("Unable to find sPAPR CPU Core definition");
3671         exit(1);
3672     }
3673 
3674     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3675                              sizeof(CPUArchId) * spapr_max_cores);
3676     machine->possible_cpus->len = spapr_max_cores;
3677     for (i = 0; i < machine->possible_cpus->len; i++) {
3678         int core_id = i * smp_threads;
3679 
3680         machine->possible_cpus->cpus[i].type = core_type;
3681         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3682         machine->possible_cpus->cpus[i].arch_id = core_id;
3683         machine->possible_cpus->cpus[i].props.has_core_id = true;
3684         machine->possible_cpus->cpus[i].props.core_id = core_id;
3685     }
3686     return machine->possible_cpus;
3687 }
3688 
3689 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3690                                 uint64_t *buid, hwaddr *pio,
3691                                 hwaddr *mmio32, hwaddr *mmio64,
3692                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3693 {
3694     /*
3695      * New-style PHB window placement.
3696      *
3697      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3698      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3699      * windows.
3700      *
3701      * Some guest kernels can't work with MMIO windows above 1<<46
3702      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3703      *
3704      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3705      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3706      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3707      * 1TiB 64-bit MMIO windows for each PHB.
3708      */
3709     const uint64_t base_buid = 0x800000020000000ULL;
3710 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3711                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3712     int i;
3713 
3714     /* Sanity check natural alignments */
3715     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3716     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3717     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3718     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3719     /* Sanity check bounds */
3720     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3721                       SPAPR_PCI_MEM32_WIN_SIZE);
3722     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3723                       SPAPR_PCI_MEM64_WIN_SIZE);
3724 
3725     if (index >= SPAPR_MAX_PHBS) {
3726         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3727                    SPAPR_MAX_PHBS - 1);
3728         return;
3729     }
3730 
3731     *buid = base_buid + index;
3732     for (i = 0; i < n_dma; ++i) {
3733         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3734     }
3735 
3736     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3737     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3738     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3739 }
3740 
3741 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3742 {
3743     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3744 
3745     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3746 }
3747 
3748 static void spapr_ics_resend(XICSFabric *dev)
3749 {
3750     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3751 
3752     ics_resend(spapr->ics);
3753 }
3754 
3755 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3756 {
3757     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3758 
3759     return cpu ? ICP(cpu->intc) : NULL;
3760 }
3761 
3762 #define ICS_IRQ_FREE(ics, srcno)   \
3763     (!((ics)->irqs[(srcno)].flags & (XICS_FLAGS_IRQ_MASK)))
3764 
3765 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
3766 {
3767     int first, i;
3768 
3769     for (first = 0; first < ics->nr_irqs; first += alignnum) {
3770         if (num > (ics->nr_irqs - first)) {
3771             return -1;
3772         }
3773         for (i = first; i < first + num; ++i) {
3774             if (!ICS_IRQ_FREE(ics, i)) {
3775                 break;
3776             }
3777         }
3778         if (i == (first + num)) {
3779             return first;
3780         }
3781     }
3782 
3783     return -1;
3784 }
3785 
3786 /*
3787  * Allocate the IRQ number and set the IRQ type, LSI or MSI
3788  */
3789 static void spapr_irq_set_lsi(sPAPRMachineState *spapr, int irq, bool lsi)
3790 {
3791     ics_set_irq_type(spapr->ics, irq - spapr->ics->offset, lsi);
3792 }
3793 
3794 int spapr_irq_alloc(sPAPRMachineState *spapr, int irq_hint, bool lsi,
3795                     Error **errp)
3796 {
3797     ICSState *ics = spapr->ics;
3798     int irq;
3799 
3800     assert(ics);
3801 
3802     if (irq_hint) {
3803         if (!ICS_IRQ_FREE(ics, irq_hint - ics->offset)) {
3804             error_setg(errp, "can't allocate IRQ %d: already in use", irq_hint);
3805             return -1;
3806         }
3807         irq = irq_hint;
3808     } else {
3809         irq = ics_find_free_block(ics, 1, 1);
3810         if (irq < 0) {
3811             error_setg(errp, "can't allocate IRQ: no IRQ left");
3812             return -1;
3813         }
3814         irq += ics->offset;
3815     }
3816 
3817     spapr_irq_set_lsi(spapr, irq, lsi);
3818     trace_spapr_irq_alloc(irq);
3819 
3820     return irq;
3821 }
3822 
3823 /*
3824  * Allocate block of consecutive IRQs, and return the number of the first IRQ in
3825  * the block. If align==true, aligns the first IRQ number to num.
3826  */
3827 int spapr_irq_alloc_block(sPAPRMachineState *spapr, int num, bool lsi,
3828                           bool align, Error **errp)
3829 {
3830     ICSState *ics = spapr->ics;
3831     int i, first = -1;
3832 
3833     assert(ics);
3834 
3835     /*
3836      * MSIMesage::data is used for storing VIRQ so
3837      * it has to be aligned to num to support multiple
3838      * MSI vectors. MSI-X is not affected by this.
3839      * The hint is used for the first IRQ, the rest should
3840      * be allocated continuously.
3841      */
3842     if (align) {
3843         assert((num == 1) || (num == 2) || (num == 4) ||
3844                (num == 8) || (num == 16) || (num == 32));
3845         first = ics_find_free_block(ics, num, num);
3846     } else {
3847         first = ics_find_free_block(ics, num, 1);
3848     }
3849     if (first < 0) {
3850         error_setg(errp, "can't find a free %d-IRQ block", num);
3851         return -1;
3852     }
3853 
3854     first += ics->offset;
3855     for (i = first; i < first + num; ++i) {
3856         spapr_irq_set_lsi(spapr, i, lsi);
3857     }
3858 
3859     trace_spapr_irq_alloc_block(first, num, lsi, align);
3860 
3861     return first;
3862 }
3863 
3864 void spapr_irq_free(sPAPRMachineState *spapr, int irq, int num)
3865 {
3866     ICSState *ics = spapr->ics;
3867     int srcno = irq - ics->offset;
3868     int i;
3869 
3870     if (ics_valid_irq(ics, irq)) {
3871         trace_spapr_irq_free(0, irq, num);
3872         for (i = srcno; i < srcno + num; ++i) {
3873             if (ICS_IRQ_FREE(ics, i)) {
3874                 trace_spapr_irq_free_warn(0, i + ics->offset);
3875             }
3876             memset(&ics->irqs[i], 0, sizeof(ICSIRQState));
3877         }
3878     }
3879 }
3880 
3881 qemu_irq spapr_qirq(sPAPRMachineState *spapr, int irq)
3882 {
3883     ICSState *ics = spapr->ics;
3884 
3885     if (ics_valid_irq(ics, irq)) {
3886         return ics->qirqs[irq - ics->offset];
3887     }
3888 
3889     return NULL;
3890 }
3891 
3892 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3893                                  Monitor *mon)
3894 {
3895     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3896     CPUState *cs;
3897 
3898     CPU_FOREACH(cs) {
3899         PowerPCCPU *cpu = POWERPC_CPU(cs);
3900 
3901         icp_pic_print_info(ICP(cpu->intc), mon);
3902     }
3903 
3904     ics_pic_print_info(spapr->ics, mon);
3905 }
3906 
3907 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3908 {
3909     return cpu->vcpu_id;
3910 }
3911 
3912 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3913 {
3914     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3915     int vcpu_id;
3916 
3917     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3918 
3919     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3920         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3921         error_append_hint(errp, "Adjust the number of cpus to %d "
3922                           "or try to raise the number of threads per core\n",
3923                           vcpu_id * smp_threads / spapr->vsmt);
3924         return;
3925     }
3926 
3927     cpu->vcpu_id = vcpu_id;
3928 }
3929 
3930 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3931 {
3932     CPUState *cs;
3933 
3934     CPU_FOREACH(cs) {
3935         PowerPCCPU *cpu = POWERPC_CPU(cs);
3936 
3937         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3938             return cpu;
3939         }
3940     }
3941 
3942     return NULL;
3943 }
3944 
3945 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3946 {
3947     MachineClass *mc = MACHINE_CLASS(oc);
3948     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3949     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3950     NMIClass *nc = NMI_CLASS(oc);
3951     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3952     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3953     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3954     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3955 
3956     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3957 
3958     /*
3959      * We set up the default / latest behaviour here.  The class_init
3960      * functions for the specific versioned machine types can override
3961      * these details for backwards compatibility
3962      */
3963     mc->init = spapr_machine_init;
3964     mc->reset = spapr_machine_reset;
3965     mc->block_default_type = IF_SCSI;
3966     mc->max_cpus = 1024;
3967     mc->no_parallel = 1;
3968     mc->default_boot_order = "";
3969     mc->default_ram_size = 512 * M_BYTE;
3970     mc->kvm_type = spapr_kvm_type;
3971     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3972     mc->pci_allow_0_address = true;
3973     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3974     hc->pre_plug = spapr_machine_device_pre_plug;
3975     hc->plug = spapr_machine_device_plug;
3976     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3977     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3978     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3979     hc->unplug_request = spapr_machine_device_unplug_request;
3980 
3981     smc->dr_lmb_enabled = true;
3982     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3983     mc->has_hotpluggable_cpus = true;
3984     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3985     fwc->get_dev_path = spapr_get_fw_dev_path;
3986     nc->nmi_monitor_handler = spapr_nmi;
3987     smc->phb_placement = spapr_phb_placement;
3988     vhc->hypercall = emulate_spapr_hypercall;
3989     vhc->hpt_mask = spapr_hpt_mask;
3990     vhc->map_hptes = spapr_map_hptes;
3991     vhc->unmap_hptes = spapr_unmap_hptes;
3992     vhc->store_hpte = spapr_store_hpte;
3993     vhc->get_patbe = spapr_get_patbe;
3994     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3995     xic->ics_get = spapr_ics_get;
3996     xic->ics_resend = spapr_ics_resend;
3997     xic->icp_get = spapr_icp_get;
3998     ispc->print_info = spapr_pic_print_info;
3999     /* Force NUMA node memory size to be a multiple of
4000      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4001      * in which LMBs are represented and hot-added
4002      */
4003     mc->numa_mem_align_shift = 28;
4004 
4005     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4006     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4007     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4008     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4009     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4010     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4011     spapr_caps_add_properties(smc, &error_abort);
4012 }
4013 
4014 static const TypeInfo spapr_machine_info = {
4015     .name          = TYPE_SPAPR_MACHINE,
4016     .parent        = TYPE_MACHINE,
4017     .abstract      = true,
4018     .instance_size = sizeof(sPAPRMachineState),
4019     .instance_init = spapr_instance_init,
4020     .instance_finalize = spapr_machine_finalizefn,
4021     .class_size    = sizeof(sPAPRMachineClass),
4022     .class_init    = spapr_machine_class_init,
4023     .interfaces = (InterfaceInfo[]) {
4024         { TYPE_FW_PATH_PROVIDER },
4025         { TYPE_NMI },
4026         { TYPE_HOTPLUG_HANDLER },
4027         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4028         { TYPE_XICS_FABRIC },
4029         { TYPE_INTERRUPT_STATS_PROVIDER },
4030         { }
4031     },
4032 };
4033 
4034 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4035     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4036                                                     void *data)      \
4037     {                                                                \
4038         MachineClass *mc = MACHINE_CLASS(oc);                        \
4039         spapr_machine_##suffix##_class_options(mc);                  \
4040         if (latest) {                                                \
4041             mc->alias = "pseries";                                   \
4042             mc->is_default = 1;                                      \
4043         }                                                            \
4044     }                                                                \
4045     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
4046     {                                                                \
4047         MachineState *machine = MACHINE(obj);                        \
4048         spapr_machine_##suffix##_instance_options(machine);          \
4049     }                                                                \
4050     static const TypeInfo spapr_machine_##suffix##_info = {          \
4051         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4052         .parent = TYPE_SPAPR_MACHINE,                                \
4053         .class_init = spapr_machine_##suffix##_class_init,           \
4054         .instance_init = spapr_machine_##suffix##_instance_init,     \
4055     };                                                               \
4056     static void spapr_machine_register_##suffix(void)                \
4057     {                                                                \
4058         type_register(&spapr_machine_##suffix##_info);               \
4059     }                                                                \
4060     type_init(spapr_machine_register_##suffix)
4061 
4062 /*
4063  * pseries-2.13
4064  */
4065 static void spapr_machine_2_13_instance_options(MachineState *machine)
4066 {
4067 }
4068 
4069 static void spapr_machine_2_13_class_options(MachineClass *mc)
4070 {
4071     /* Defaults for the latest behaviour inherited from the base class */
4072 }
4073 
4074 DEFINE_SPAPR_MACHINE(2_13, "2.13", true);
4075 
4076 /*
4077  * pseries-2.12
4078  */
4079 #define SPAPR_COMPAT_2_12                                              \
4080     HW_COMPAT_2_12                                                     \
4081     {                                                                  \
4082         .driver = TYPE_POWERPC_CPU,                                    \
4083         .property = "pre-2.13-migration",                              \
4084         .value    = "on",                                              \
4085     },
4086 
4087 static void spapr_machine_2_12_instance_options(MachineState *machine)
4088 {
4089     spapr_machine_2_13_instance_options(machine);
4090 }
4091 
4092 static void spapr_machine_2_12_class_options(MachineClass *mc)
4093 {
4094     spapr_machine_2_13_class_options(mc);
4095     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4096 }
4097 
4098 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4099 
4100 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
4101 {
4102     spapr_machine_2_12_instance_options(machine);
4103 }
4104 
4105 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4106 {
4107     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4108 
4109     spapr_machine_2_12_class_options(mc);
4110     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4111     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4112     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4113 }
4114 
4115 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4116 
4117 /*
4118  * pseries-2.11
4119  */
4120 #define SPAPR_COMPAT_2_11                                              \
4121     HW_COMPAT_2_11
4122 
4123 static void spapr_machine_2_11_instance_options(MachineState *machine)
4124 {
4125     spapr_machine_2_12_instance_options(machine);
4126 }
4127 
4128 static void spapr_machine_2_11_class_options(MachineClass *mc)
4129 {
4130     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4131 
4132     spapr_machine_2_12_class_options(mc);
4133     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4134     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4135 }
4136 
4137 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4138 
4139 /*
4140  * pseries-2.10
4141  */
4142 #define SPAPR_COMPAT_2_10                                              \
4143     HW_COMPAT_2_10
4144 
4145 static void spapr_machine_2_10_instance_options(MachineState *machine)
4146 {
4147     spapr_machine_2_11_instance_options(machine);
4148 }
4149 
4150 static void spapr_machine_2_10_class_options(MachineClass *mc)
4151 {
4152     spapr_machine_2_11_class_options(mc);
4153     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4154 }
4155 
4156 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4157 
4158 /*
4159  * pseries-2.9
4160  */
4161 #define SPAPR_COMPAT_2_9                                               \
4162     HW_COMPAT_2_9                                                      \
4163     {                                                                  \
4164         .driver = TYPE_POWERPC_CPU,                                    \
4165         .property = "pre-2.10-migration",                              \
4166         .value    = "on",                                              \
4167     },                                                                 \
4168 
4169 static void spapr_machine_2_9_instance_options(MachineState *machine)
4170 {
4171     spapr_machine_2_10_instance_options(machine);
4172 }
4173 
4174 static void spapr_machine_2_9_class_options(MachineClass *mc)
4175 {
4176     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4177 
4178     spapr_machine_2_10_class_options(mc);
4179     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4180     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4181     smc->pre_2_10_has_unused_icps = true;
4182     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4183 }
4184 
4185 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4186 
4187 /*
4188  * pseries-2.8
4189  */
4190 #define SPAPR_COMPAT_2_8                                        \
4191     HW_COMPAT_2_8                                               \
4192     {                                                           \
4193         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4194         .property = "pcie-extended-configuration-space",        \
4195         .value    = "off",                                      \
4196     },
4197 
4198 static void spapr_machine_2_8_instance_options(MachineState *machine)
4199 {
4200     spapr_machine_2_9_instance_options(machine);
4201 }
4202 
4203 static void spapr_machine_2_8_class_options(MachineClass *mc)
4204 {
4205     spapr_machine_2_9_class_options(mc);
4206     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4207     mc->numa_mem_align_shift = 23;
4208 }
4209 
4210 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4211 
4212 /*
4213  * pseries-2.7
4214  */
4215 #define SPAPR_COMPAT_2_7                            \
4216     HW_COMPAT_2_7                                   \
4217     {                                               \
4218         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4219         .property = "mem_win_size",                 \
4220         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4221     },                                              \
4222     {                                               \
4223         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4224         .property = "mem64_win_size",               \
4225         .value    = "0",                            \
4226     },                                              \
4227     {                                               \
4228         .driver = TYPE_POWERPC_CPU,                 \
4229         .property = "pre-2.8-migration",            \
4230         .value    = "on",                           \
4231     },                                              \
4232     {                                               \
4233         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4234         .property = "pre-2.8-migration",            \
4235         .value    = "on",                           \
4236     },
4237 
4238 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4239                               uint64_t *buid, hwaddr *pio,
4240                               hwaddr *mmio32, hwaddr *mmio64,
4241                               unsigned n_dma, uint32_t *liobns, Error **errp)
4242 {
4243     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4244     const uint64_t base_buid = 0x800000020000000ULL;
4245     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4246     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4247     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4248     const uint32_t max_index = 255;
4249     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4250 
4251     uint64_t ram_top = MACHINE(spapr)->ram_size;
4252     hwaddr phb0_base, phb_base;
4253     int i;
4254 
4255     /* Do we have hotpluggable memory? */
4256     if (MACHINE(spapr)->maxram_size > ram_top) {
4257         /* Can't just use maxram_size, because there may be an
4258          * alignment gap between normal and hotpluggable memory
4259          * regions */
4260         ram_top = spapr->hotplug_memory.base +
4261             memory_region_size(&spapr->hotplug_memory.mr);
4262     }
4263 
4264     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4265 
4266     if (index > max_index) {
4267         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4268                    max_index);
4269         return;
4270     }
4271 
4272     *buid = base_buid + index;
4273     for (i = 0; i < n_dma; ++i) {
4274         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4275     }
4276 
4277     phb_base = phb0_base + index * phb_spacing;
4278     *pio = phb_base + pio_offset;
4279     *mmio32 = phb_base + mmio_offset;
4280     /*
4281      * We don't set the 64-bit MMIO window, relying on the PHB's
4282      * fallback behaviour of automatically splitting a large "32-bit"
4283      * window into contiguous 32-bit and 64-bit windows
4284      */
4285 }
4286 
4287 static void spapr_machine_2_7_instance_options(MachineState *machine)
4288 {
4289     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4290 
4291     spapr_machine_2_8_instance_options(machine);
4292     spapr->use_hotplug_event_source = false;
4293 }
4294 
4295 static void spapr_machine_2_7_class_options(MachineClass *mc)
4296 {
4297     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4298 
4299     spapr_machine_2_8_class_options(mc);
4300     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4301     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4302     smc->phb_placement = phb_placement_2_7;
4303 }
4304 
4305 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4306 
4307 /*
4308  * pseries-2.6
4309  */
4310 #define SPAPR_COMPAT_2_6 \
4311     HW_COMPAT_2_6 \
4312     { \
4313         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4314         .property = "ddw",\
4315         .value    = stringify(off),\
4316     },
4317 
4318 static void spapr_machine_2_6_instance_options(MachineState *machine)
4319 {
4320     spapr_machine_2_7_instance_options(machine);
4321 }
4322 
4323 static void spapr_machine_2_6_class_options(MachineClass *mc)
4324 {
4325     spapr_machine_2_7_class_options(mc);
4326     mc->has_hotpluggable_cpus = false;
4327     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4328 }
4329 
4330 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4331 
4332 /*
4333  * pseries-2.5
4334  */
4335 #define SPAPR_COMPAT_2_5 \
4336     HW_COMPAT_2_5 \
4337     { \
4338         .driver   = "spapr-vlan", \
4339         .property = "use-rx-buffer-pools", \
4340         .value    = "off", \
4341     },
4342 
4343 static void spapr_machine_2_5_instance_options(MachineState *machine)
4344 {
4345     spapr_machine_2_6_instance_options(machine);
4346 }
4347 
4348 static void spapr_machine_2_5_class_options(MachineClass *mc)
4349 {
4350     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4351 
4352     spapr_machine_2_6_class_options(mc);
4353     smc->use_ohci_by_default = true;
4354     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4355 }
4356 
4357 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4358 
4359 /*
4360  * pseries-2.4
4361  */
4362 #define SPAPR_COMPAT_2_4 \
4363         HW_COMPAT_2_4
4364 
4365 static void spapr_machine_2_4_instance_options(MachineState *machine)
4366 {
4367     spapr_machine_2_5_instance_options(machine);
4368 }
4369 
4370 static void spapr_machine_2_4_class_options(MachineClass *mc)
4371 {
4372     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4373 
4374     spapr_machine_2_5_class_options(mc);
4375     smc->dr_lmb_enabled = false;
4376     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4377 }
4378 
4379 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4380 
4381 /*
4382  * pseries-2.3
4383  */
4384 #define SPAPR_COMPAT_2_3 \
4385         HW_COMPAT_2_3 \
4386         {\
4387             .driver   = "spapr-pci-host-bridge",\
4388             .property = "dynamic-reconfiguration",\
4389             .value    = "off",\
4390         },
4391 
4392 static void spapr_machine_2_3_instance_options(MachineState *machine)
4393 {
4394     spapr_machine_2_4_instance_options(machine);
4395 }
4396 
4397 static void spapr_machine_2_3_class_options(MachineClass *mc)
4398 {
4399     spapr_machine_2_4_class_options(mc);
4400     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4401 }
4402 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4403 
4404 /*
4405  * pseries-2.2
4406  */
4407 
4408 #define SPAPR_COMPAT_2_2 \
4409         HW_COMPAT_2_2 \
4410         {\
4411             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4412             .property = "mem_win_size",\
4413             .value    = "0x20000000",\
4414         },
4415 
4416 static void spapr_machine_2_2_instance_options(MachineState *machine)
4417 {
4418     spapr_machine_2_3_instance_options(machine);
4419     machine->suppress_vmdesc = true;
4420 }
4421 
4422 static void spapr_machine_2_2_class_options(MachineClass *mc)
4423 {
4424     spapr_machine_2_3_class_options(mc);
4425     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4426 }
4427 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4428 
4429 /*
4430  * pseries-2.1
4431  */
4432 #define SPAPR_COMPAT_2_1 \
4433         HW_COMPAT_2_1
4434 
4435 static void spapr_machine_2_1_instance_options(MachineState *machine)
4436 {
4437     spapr_machine_2_2_instance_options(machine);
4438 }
4439 
4440 static void spapr_machine_2_1_class_options(MachineClass *mc)
4441 {
4442     spapr_machine_2_2_class_options(mc);
4443     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4444 }
4445 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4446 
4447 static void spapr_machine_register_types(void)
4448 {
4449     type_register_static(&spapr_machine_info);
4450 }
4451 
4452 type_init(spapr_machine_register_types)
4453