xref: /openbmc/qemu/hw/ppc/spapr.c (revision 4248981d5190eaf2b0ab74d8c287b6070ffc300a)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
46 #include "cpu-models.h"
47 #include "qom/cpu.h"
48 
49 #include "hw/boards.h"
50 #include "hw/ppc/ppc.h"
51 #include "hw/loader.h"
52 
53 #include "hw/ppc/fdt.h"
54 #include "hw/ppc/spapr.h"
55 #include "hw/ppc/spapr_vio.h"
56 #include "hw/pci-host/spapr.h"
57 #include "hw/ppc/xics.h"
58 #include "hw/pci/msi.h"
59 
60 #include "hw/pci/pci.h"
61 #include "hw/scsi/scsi.h"
62 #include "hw/virtio/virtio-scsi.h"
63 #include "hw/virtio/vhost-scsi-common.h"
64 
65 #include "exec/address-spaces.h"
66 #include "hw/usb.h"
67 #include "qemu/config-file.h"
68 #include "qemu/error-report.h"
69 #include "trace.h"
70 #include "hw/nmi.h"
71 #include "hw/intc/intc.h"
72 
73 #include "hw/compat.h"
74 #include "qemu/cutils.h"
75 #include "hw/ppc/spapr_cpu_core.h"
76 #include "hw/mem/memory-device.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     MachineState *machine = MACHINE(spapr);
706     uint8_t *int_buf, *cur_index, buf_len;
707     int ret;
708     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
709     uint64_t addr, cur_addr, size;
710     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
711     uint64_t mem_end = machine->device_memory->base +
712                        memory_region_size(&machine->device_memory->mr);
713     uint32_t node, nr_entries = 0;
714     sPAPRDRConnector *drc;
715     DrconfCellQueue *elem, *next;
716     MemoryDeviceInfoList *info;
717     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
718         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
719 
720     /* Entry to cover RAM and the gap area */
721     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
722                                  SPAPR_LMB_FLAGS_RESERVED |
723                                  SPAPR_LMB_FLAGS_DRC_INVALID);
724     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
725     nr_entries++;
726 
727     cur_addr = machine->device_memory->base;
728     for (info = dimms; info; info = info->next) {
729         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
730 
731         addr = di->addr;
732         size = di->size;
733         node = di->node;
734 
735         /* Entry for hot-pluggable area */
736         if (cur_addr < addr) {
737             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
738             g_assert(drc);
739             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
740                                          cur_addr, spapr_drc_index(drc), -1, 0);
741             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
742             nr_entries++;
743         }
744 
745         /* Entry for DIMM */
746         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
747         g_assert(drc);
748         elem = spapr_get_drconf_cell(size / lmb_size, addr,
749                                      spapr_drc_index(drc), node,
750                                      SPAPR_LMB_FLAGS_ASSIGNED);
751         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
752         nr_entries++;
753         cur_addr = addr + size;
754     }
755 
756     /* Entry for remaining hotpluggable area */
757     if (cur_addr < mem_end) {
758         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
759         g_assert(drc);
760         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
761                                      cur_addr, spapr_drc_index(drc), -1, 0);
762         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
763         nr_entries++;
764     }
765 
766     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
767     int_buf = cur_index = g_malloc0(buf_len);
768     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
769     cur_index += sizeof(nr_entries);
770 
771     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
772         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
773         cur_index += sizeof(elem->cell);
774         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
775         g_free(elem);
776     }
777 
778     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
779     g_free(int_buf);
780     if (ret < 0) {
781         return -1;
782     }
783     return 0;
784 }
785 
786 /* ibm,dynamic-memory */
787 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
788                                    int offset, MemoryDeviceInfoList *dimms)
789 {
790     MachineState *machine = MACHINE(spapr);
791     int i, ret;
792     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
793     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
794     uint32_t nr_lmbs = (machine->device_memory->base +
795                        memory_region_size(&machine->device_memory->mr)) /
796                        lmb_size;
797     uint32_t *int_buf, *cur_index, buf_len;
798 
799     /*
800      * Allocate enough buffer size to fit in ibm,dynamic-memory
801      */
802     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
803     cur_index = int_buf = g_malloc0(buf_len);
804     int_buf[0] = cpu_to_be32(nr_lmbs);
805     cur_index++;
806     for (i = 0; i < nr_lmbs; i++) {
807         uint64_t addr = i * lmb_size;
808         uint32_t *dynamic_memory = cur_index;
809 
810         if (i >= device_lmb_start) {
811             sPAPRDRConnector *drc;
812 
813             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
814             g_assert(drc);
815 
816             dynamic_memory[0] = cpu_to_be32(addr >> 32);
817             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
818             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
819             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
820             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
821             if (memory_region_present(get_system_memory(), addr)) {
822                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
823             } else {
824                 dynamic_memory[5] = cpu_to_be32(0);
825             }
826         } else {
827             /*
828              * LMB information for RMA, boot time RAM and gap b/n RAM and
829              * device memory region -- all these are marked as reserved
830              * and as having no valid DRC.
831              */
832             dynamic_memory[0] = cpu_to_be32(addr >> 32);
833             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
834             dynamic_memory[2] = cpu_to_be32(0);
835             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
836             dynamic_memory[4] = cpu_to_be32(-1);
837             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
838                                             SPAPR_LMB_FLAGS_DRC_INVALID);
839         }
840 
841         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
842     }
843     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
844     g_free(int_buf);
845     if (ret < 0) {
846         return -1;
847     }
848     return 0;
849 }
850 
851 /*
852  * Adds ibm,dynamic-reconfiguration-memory node.
853  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
854  * of this device tree node.
855  */
856 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
857 {
858     MachineState *machine = MACHINE(spapr);
859     int ret, i, offset;
860     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
861     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
862     uint32_t *int_buf, *cur_index, buf_len;
863     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
864     MemoryDeviceInfoList *dimms = NULL;
865 
866     /*
867      * Don't create the node if there is no device memory
868      */
869     if (machine->ram_size == machine->maxram_size) {
870         return 0;
871     }
872 
873     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
874 
875     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
876                     sizeof(prop_lmb_size));
877     if (ret < 0) {
878         return ret;
879     }
880 
881     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
882     if (ret < 0) {
883         return ret;
884     }
885 
886     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
887     if (ret < 0) {
888         return ret;
889     }
890 
891     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
892     dimms = qmp_memory_device_list();
893     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
894         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
895     } else {
896         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
897     }
898     qapi_free_MemoryDeviceInfoList(dimms);
899 
900     if (ret < 0) {
901         return ret;
902     }
903 
904     /* ibm,associativity-lookup-arrays */
905     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
906     cur_index = int_buf = g_malloc0(buf_len);
907 
908     cur_index = int_buf;
909     int_buf[0] = cpu_to_be32(nr_nodes);
910     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
911     cur_index += 2;
912     for (i = 0; i < nr_nodes; i++) {
913         uint32_t associativity[] = {
914             cpu_to_be32(0x0),
915             cpu_to_be32(0x0),
916             cpu_to_be32(0x0),
917             cpu_to_be32(i)
918         };
919         memcpy(cur_index, associativity, sizeof(associativity));
920         cur_index += 4;
921     }
922     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
923             (cur_index - int_buf) * sizeof(uint32_t));
924     g_free(int_buf);
925 
926     return ret;
927 }
928 
929 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
930                                 sPAPROptionVector *ov5_updates)
931 {
932     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
933     int ret = 0, offset;
934 
935     /* Generate ibm,dynamic-reconfiguration-memory node if required */
936     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
937         g_assert(smc->dr_lmb_enabled);
938         ret = spapr_populate_drconf_memory(spapr, fdt);
939         if (ret) {
940             goto out;
941         }
942     }
943 
944     offset = fdt_path_offset(fdt, "/chosen");
945     if (offset < 0) {
946         offset = fdt_add_subnode(fdt, 0, "chosen");
947         if (offset < 0) {
948             return offset;
949         }
950     }
951     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
952                                  "ibm,architecture-vec-5");
953 
954 out:
955     return ret;
956 }
957 
958 static bool spapr_hotplugged_dev_before_cas(void)
959 {
960     Object *drc_container, *obj;
961     ObjectProperty *prop;
962     ObjectPropertyIterator iter;
963 
964     drc_container = container_get(object_get_root(), "/dr-connector");
965     object_property_iter_init(&iter, drc_container);
966     while ((prop = object_property_iter_next(&iter))) {
967         if (!strstart(prop->type, "link<", NULL)) {
968             continue;
969         }
970         obj = object_property_get_link(drc_container, prop->name, NULL);
971         if (spapr_drc_needed(obj)) {
972             return true;
973         }
974     }
975     return false;
976 }
977 
978 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
979                                  target_ulong addr, target_ulong size,
980                                  sPAPROptionVector *ov5_updates)
981 {
982     void *fdt, *fdt_skel;
983     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
984 
985     if (spapr_hotplugged_dev_before_cas()) {
986         return 1;
987     }
988 
989     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
990         error_report("SLOF provided an unexpected CAS buffer size "
991                      TARGET_FMT_lu " (min: %zu, max: %u)",
992                      size, sizeof(hdr), FW_MAX_SIZE);
993         exit(EXIT_FAILURE);
994     }
995 
996     size -= sizeof(hdr);
997 
998     /* Create skeleton */
999     fdt_skel = g_malloc0(size);
1000     _FDT((fdt_create(fdt_skel, size)));
1001     _FDT((fdt_finish_reservemap(fdt_skel)));
1002     _FDT((fdt_begin_node(fdt_skel, "")));
1003     _FDT((fdt_end_node(fdt_skel)));
1004     _FDT((fdt_finish(fdt_skel)));
1005     fdt = g_malloc0(size);
1006     _FDT((fdt_open_into(fdt_skel, fdt, size)));
1007     g_free(fdt_skel);
1008 
1009     /* Fixup cpu nodes */
1010     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
1011 
1012     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
1013         return -1;
1014     }
1015 
1016     /* Pack resulting tree */
1017     _FDT((fdt_pack(fdt)));
1018 
1019     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1020         trace_spapr_cas_failed(size);
1021         return -1;
1022     }
1023 
1024     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1025     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1026     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1027     g_free(fdt);
1028 
1029     return 0;
1030 }
1031 
1032 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1033 {
1034     int rtas;
1035     GString *hypertas = g_string_sized_new(256);
1036     GString *qemu_hypertas = g_string_sized_new(256);
1037     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1038     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1039         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1040     uint32_t lrdr_capacity[] = {
1041         cpu_to_be32(max_device_addr >> 32),
1042         cpu_to_be32(max_device_addr & 0xffffffff),
1043         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1044         cpu_to_be32(max_cpus / smp_threads),
1045     };
1046     uint32_t maxdomains[] = {
1047         cpu_to_be32(4),
1048         cpu_to_be32(0),
1049         cpu_to_be32(0),
1050         cpu_to_be32(0),
1051         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
1052     };
1053 
1054     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1055 
1056     /* hypertas */
1057     add_str(hypertas, "hcall-pft");
1058     add_str(hypertas, "hcall-term");
1059     add_str(hypertas, "hcall-dabr");
1060     add_str(hypertas, "hcall-interrupt");
1061     add_str(hypertas, "hcall-tce");
1062     add_str(hypertas, "hcall-vio");
1063     add_str(hypertas, "hcall-splpar");
1064     add_str(hypertas, "hcall-bulk");
1065     add_str(hypertas, "hcall-set-mode");
1066     add_str(hypertas, "hcall-sprg0");
1067     add_str(hypertas, "hcall-copy");
1068     add_str(hypertas, "hcall-debug");
1069     add_str(qemu_hypertas, "hcall-memop1");
1070 
1071     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1072         add_str(hypertas, "hcall-multi-tce");
1073     }
1074 
1075     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1076         add_str(hypertas, "hcall-hpt-resize");
1077     }
1078 
1079     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1080                      hypertas->str, hypertas->len));
1081     g_string_free(hypertas, TRUE);
1082     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1083                      qemu_hypertas->str, qemu_hypertas->len));
1084     g_string_free(qemu_hypertas, TRUE);
1085 
1086     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1087                      refpoints, sizeof(refpoints)));
1088 
1089     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1090                      maxdomains, sizeof(maxdomains)));
1091 
1092     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1093                           RTAS_ERROR_LOG_MAX));
1094     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1095                           RTAS_EVENT_SCAN_RATE));
1096 
1097     g_assert(msi_nonbroken);
1098     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1099 
1100     /*
1101      * According to PAPR, rtas ibm,os-term does not guarantee a return
1102      * back to the guest cpu.
1103      *
1104      * While an additional ibm,extended-os-term property indicates
1105      * that rtas call return will always occur. Set this property.
1106      */
1107     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1108 
1109     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1110                      lrdr_capacity, sizeof(lrdr_capacity)));
1111 
1112     spapr_dt_rtas_tokens(fdt, rtas);
1113 }
1114 
1115 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1116  * that the guest may request and thus the valid values for bytes 24..26 of
1117  * option vector 5: */
1118 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1119 {
1120     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1121 
1122     char val[2 * 4] = {
1123         23, 0x00, /* Xive mode, filled in below. */
1124         24, 0x00, /* Hash/Radix, filled in below. */
1125         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1126         26, 0x40, /* Radix options: GTSE == yes. */
1127     };
1128 
1129     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1130                           first_ppc_cpu->compat_pvr)) {
1131         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1132         val[3] = 0x00; /* Hash */
1133     } else if (kvm_enabled()) {
1134         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1135             val[3] = 0x80; /* OV5_MMU_BOTH */
1136         } else if (kvmppc_has_cap_mmu_radix()) {
1137             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1138         } else {
1139             val[3] = 0x00; /* Hash */
1140         }
1141     } else {
1142         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1143         val[3] = 0xC0;
1144     }
1145     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1146                      val, sizeof(val)));
1147 }
1148 
1149 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1150 {
1151     MachineState *machine = MACHINE(spapr);
1152     int chosen;
1153     const char *boot_device = machine->boot_order;
1154     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1155     size_t cb = 0;
1156     char *bootlist = get_boot_devices_list(&cb, true);
1157 
1158     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1159 
1160     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1161     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1162                           spapr->initrd_base));
1163     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1164                           spapr->initrd_base + spapr->initrd_size));
1165 
1166     if (spapr->kernel_size) {
1167         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1168                               cpu_to_be64(spapr->kernel_size) };
1169 
1170         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1171                          &kprop, sizeof(kprop)));
1172         if (spapr->kernel_le) {
1173             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1174         }
1175     }
1176     if (boot_menu) {
1177         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1178     }
1179     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1180     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1181     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1182 
1183     if (cb && bootlist) {
1184         int i;
1185 
1186         for (i = 0; i < cb; i++) {
1187             if (bootlist[i] == '\n') {
1188                 bootlist[i] = ' ';
1189             }
1190         }
1191         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1192     }
1193 
1194     if (boot_device && strlen(boot_device)) {
1195         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1196     }
1197 
1198     if (!spapr->has_graphics && stdout_path) {
1199         /*
1200          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1201          * kernel. New platforms should only use the "stdout-path" property. Set
1202          * the new property and continue using older property to remain
1203          * compatible with the existing firmware.
1204          */
1205         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1206         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1207     }
1208 
1209     spapr_dt_ov5_platform_support(fdt, chosen);
1210 
1211     g_free(stdout_path);
1212     g_free(bootlist);
1213 }
1214 
1215 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1216 {
1217     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1218      * KVM to work under pHyp with some guest co-operation */
1219     int hypervisor;
1220     uint8_t hypercall[16];
1221 
1222     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1223     /* indicate KVM hypercall interface */
1224     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1225     if (kvmppc_has_cap_fixup_hcalls()) {
1226         /*
1227          * Older KVM versions with older guest kernels were broken
1228          * with the magic page, don't allow the guest to map it.
1229          */
1230         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1231                                   sizeof(hypercall))) {
1232             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1233                              hypercall, sizeof(hypercall)));
1234         }
1235     }
1236 }
1237 
1238 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1239                              hwaddr rtas_addr,
1240                              hwaddr rtas_size)
1241 {
1242     MachineState *machine = MACHINE(spapr);
1243     MachineClass *mc = MACHINE_GET_CLASS(machine);
1244     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1245     int ret;
1246     void *fdt;
1247     sPAPRPHBState *phb;
1248     char *buf;
1249 
1250     fdt = g_malloc0(FDT_MAX_SIZE);
1251     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1252 
1253     /* Root node */
1254     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1255     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1256     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1257 
1258     /*
1259      * Add info to guest to indentify which host is it being run on
1260      * and what is the uuid of the guest
1261      */
1262     if (kvmppc_get_host_model(&buf)) {
1263         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1264         g_free(buf);
1265     }
1266     if (kvmppc_get_host_serial(&buf)) {
1267         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1268         g_free(buf);
1269     }
1270 
1271     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1272 
1273     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1274     if (qemu_uuid_set) {
1275         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1276     }
1277     g_free(buf);
1278 
1279     if (qemu_get_vm_name()) {
1280         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1281                                 qemu_get_vm_name()));
1282     }
1283 
1284     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1285     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1286 
1287     /* /interrupt controller */
1288     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1289 
1290     ret = spapr_populate_memory(spapr, fdt);
1291     if (ret < 0) {
1292         error_report("couldn't setup memory nodes in fdt");
1293         exit(1);
1294     }
1295 
1296     /* /vdevice */
1297     spapr_dt_vdevice(spapr->vio_bus, fdt);
1298 
1299     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1300         ret = spapr_rng_populate_dt(fdt);
1301         if (ret < 0) {
1302             error_report("could not set up rng device in the fdt");
1303             exit(1);
1304         }
1305     }
1306 
1307     QLIST_FOREACH(phb, &spapr->phbs, list) {
1308         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1309         if (ret < 0) {
1310             error_report("couldn't setup PCI devices in fdt");
1311             exit(1);
1312         }
1313     }
1314 
1315     /* cpus */
1316     spapr_populate_cpus_dt_node(fdt, spapr);
1317 
1318     if (smc->dr_lmb_enabled) {
1319         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1320     }
1321 
1322     if (mc->has_hotpluggable_cpus) {
1323         int offset = fdt_path_offset(fdt, "/cpus");
1324         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1325                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1326         if (ret < 0) {
1327             error_report("Couldn't set up CPU DR device tree properties");
1328             exit(1);
1329         }
1330     }
1331 
1332     /* /event-sources */
1333     spapr_dt_events(spapr, fdt);
1334 
1335     /* /rtas */
1336     spapr_dt_rtas(spapr, fdt);
1337 
1338     /* /chosen */
1339     spapr_dt_chosen(spapr, fdt);
1340 
1341     /* /hypervisor */
1342     if (kvm_enabled()) {
1343         spapr_dt_hypervisor(spapr, fdt);
1344     }
1345 
1346     /* Build memory reserve map */
1347     if (spapr->kernel_size) {
1348         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1349     }
1350     if (spapr->initrd_size) {
1351         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1352     }
1353 
1354     /* ibm,client-architecture-support updates */
1355     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1356     if (ret < 0) {
1357         error_report("couldn't setup CAS properties fdt");
1358         exit(1);
1359     }
1360 
1361     return fdt;
1362 }
1363 
1364 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1365 {
1366     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1367 }
1368 
1369 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1370                                     PowerPCCPU *cpu)
1371 {
1372     CPUPPCState *env = &cpu->env;
1373 
1374     /* The TCG path should also be holding the BQL at this point */
1375     g_assert(qemu_mutex_iothread_locked());
1376 
1377     if (msr_pr) {
1378         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1379         env->gpr[3] = H_PRIVILEGE;
1380     } else {
1381         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1382     }
1383 }
1384 
1385 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1386 {
1387     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1388 
1389     return spapr->patb_entry;
1390 }
1391 
1392 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1393 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1394 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1395 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1396 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1397 
1398 /*
1399  * Get the fd to access the kernel htab, re-opening it if necessary
1400  */
1401 static int get_htab_fd(sPAPRMachineState *spapr)
1402 {
1403     Error *local_err = NULL;
1404 
1405     if (spapr->htab_fd >= 0) {
1406         return spapr->htab_fd;
1407     }
1408 
1409     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1410     if (spapr->htab_fd < 0) {
1411         error_report_err(local_err);
1412     }
1413 
1414     return spapr->htab_fd;
1415 }
1416 
1417 void close_htab_fd(sPAPRMachineState *spapr)
1418 {
1419     if (spapr->htab_fd >= 0) {
1420         close(spapr->htab_fd);
1421     }
1422     spapr->htab_fd = -1;
1423 }
1424 
1425 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1426 {
1427     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1428 
1429     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1430 }
1431 
1432 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1433 {
1434     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1435 
1436     assert(kvm_enabled());
1437 
1438     if (!spapr->htab) {
1439         return 0;
1440     }
1441 
1442     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1443 }
1444 
1445 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1446                                                 hwaddr ptex, int n)
1447 {
1448     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1449     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1450 
1451     if (!spapr->htab) {
1452         /*
1453          * HTAB is controlled by KVM. Fetch into temporary buffer
1454          */
1455         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1456         kvmppc_read_hptes(hptes, ptex, n);
1457         return hptes;
1458     }
1459 
1460     /*
1461      * HTAB is controlled by QEMU. Just point to the internally
1462      * accessible PTEG.
1463      */
1464     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1465 }
1466 
1467 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1468                               const ppc_hash_pte64_t *hptes,
1469                               hwaddr ptex, int n)
1470 {
1471     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1472 
1473     if (!spapr->htab) {
1474         g_free((void *)hptes);
1475     }
1476 
1477     /* Nothing to do for qemu managed HPT */
1478 }
1479 
1480 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1481                              uint64_t pte0, uint64_t pte1)
1482 {
1483     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1484     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1485 
1486     if (!spapr->htab) {
1487         kvmppc_write_hpte(ptex, pte0, pte1);
1488     } else {
1489         stq_p(spapr->htab + offset, pte0);
1490         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1491     }
1492 }
1493 
1494 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1495 {
1496     int shift;
1497 
1498     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1499      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1500      * that's much more than is needed for Linux guests */
1501     shift = ctz64(pow2ceil(ramsize)) - 7;
1502     shift = MAX(shift, 18); /* Minimum architected size */
1503     shift = MIN(shift, 46); /* Maximum architected size */
1504     return shift;
1505 }
1506 
1507 void spapr_free_hpt(sPAPRMachineState *spapr)
1508 {
1509     g_free(spapr->htab);
1510     spapr->htab = NULL;
1511     spapr->htab_shift = 0;
1512     close_htab_fd(spapr);
1513 }
1514 
1515 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1516                           Error **errp)
1517 {
1518     long rc;
1519 
1520     /* Clean up any HPT info from a previous boot */
1521     spapr_free_hpt(spapr);
1522 
1523     rc = kvmppc_reset_htab(shift);
1524     if (rc < 0) {
1525         /* kernel-side HPT needed, but couldn't allocate one */
1526         error_setg_errno(errp, errno,
1527                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1528                          shift);
1529         /* This is almost certainly fatal, but if the caller really
1530          * wants to carry on with shift == 0, it's welcome to try */
1531     } else if (rc > 0) {
1532         /* kernel-side HPT allocated */
1533         if (rc != shift) {
1534             error_setg(errp,
1535                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1536                        shift, rc);
1537         }
1538 
1539         spapr->htab_shift = shift;
1540         spapr->htab = NULL;
1541     } else {
1542         /* kernel-side HPT not needed, allocate in userspace instead */
1543         size_t size = 1ULL << shift;
1544         int i;
1545 
1546         spapr->htab = qemu_memalign(size, size);
1547         if (!spapr->htab) {
1548             error_setg_errno(errp, errno,
1549                              "Could not allocate HPT of order %d", shift);
1550             return;
1551         }
1552 
1553         memset(spapr->htab, 0, size);
1554         spapr->htab_shift = shift;
1555 
1556         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1557             DIRTY_HPTE(HPTE(spapr->htab, i));
1558         }
1559     }
1560     /* We're setting up a hash table, so that means we're not radix */
1561     spapr->patb_entry = 0;
1562 }
1563 
1564 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1565 {
1566     int hpt_shift;
1567 
1568     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1569         || (spapr->cas_reboot
1570             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1571         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1572     } else {
1573         uint64_t current_ram_size;
1574 
1575         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1576         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1577     }
1578     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1579 
1580     if (spapr->vrma_adjust) {
1581         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1582                                           spapr->htab_shift);
1583     }
1584 }
1585 
1586 static int spapr_reset_drcs(Object *child, void *opaque)
1587 {
1588     sPAPRDRConnector *drc =
1589         (sPAPRDRConnector *) object_dynamic_cast(child,
1590                                                  TYPE_SPAPR_DR_CONNECTOR);
1591 
1592     if (drc) {
1593         spapr_drc_reset(drc);
1594     }
1595 
1596     return 0;
1597 }
1598 
1599 static void spapr_machine_reset(void)
1600 {
1601     MachineState *machine = MACHINE(qdev_get_machine());
1602     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1603     PowerPCCPU *first_ppc_cpu;
1604     uint32_t rtas_limit;
1605     hwaddr rtas_addr, fdt_addr;
1606     void *fdt;
1607     int rc;
1608 
1609     spapr_caps_reset(spapr);
1610 
1611     first_ppc_cpu = POWERPC_CPU(first_cpu);
1612     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1613         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1614                          spapr->max_compat_pvr)) {
1615         /* If using KVM with radix mode available, VCPUs can be started
1616          * without a HPT because KVM will start them in radix mode.
1617          * Set the GR bit in PATB so that we know there is no HPT. */
1618         spapr->patb_entry = PATBE1_GR;
1619     } else {
1620         spapr_setup_hpt_and_vrma(spapr);
1621     }
1622 
1623     /* if this reset wasn't generated by CAS, we should reset our
1624      * negotiated options and start from scratch */
1625     if (!spapr->cas_reboot) {
1626         spapr_ovec_cleanup(spapr->ov5_cas);
1627         spapr->ov5_cas = spapr_ovec_new();
1628 
1629         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1630     }
1631 
1632     qemu_devices_reset();
1633 
1634     /* DRC reset may cause a device to be unplugged. This will cause troubles
1635      * if this device is used by another device (eg, a running vhost backend
1636      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1637      * situations, we reset DRCs after all devices have been reset.
1638      */
1639     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1640 
1641     spapr_clear_pending_events(spapr);
1642 
1643     /*
1644      * We place the device tree and RTAS just below either the top of the RMA,
1645      * or just below 2GB, whichever is lowere, so that it can be
1646      * processed with 32-bit real mode code if necessary
1647      */
1648     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1649     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1650     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1651 
1652     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1653 
1654     spapr_load_rtas(spapr, fdt, rtas_addr);
1655 
1656     rc = fdt_pack(fdt);
1657 
1658     /* Should only fail if we've built a corrupted tree */
1659     assert(rc == 0);
1660 
1661     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1662         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1663                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1664         exit(1);
1665     }
1666 
1667     /* Load the fdt */
1668     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1669     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1670     g_free(fdt);
1671 
1672     /* Set up the entry state */
1673     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1674     first_ppc_cpu->env.gpr[5] = 0;
1675 
1676     spapr->cas_reboot = false;
1677 }
1678 
1679 static void spapr_create_nvram(sPAPRMachineState *spapr)
1680 {
1681     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1682     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1683 
1684     if (dinfo) {
1685         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1686                             &error_fatal);
1687     }
1688 
1689     qdev_init_nofail(dev);
1690 
1691     spapr->nvram = (struct sPAPRNVRAM *)dev;
1692 }
1693 
1694 static void spapr_rtc_create(sPAPRMachineState *spapr)
1695 {
1696     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1697     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1698                               &error_fatal);
1699     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1700                               &error_fatal);
1701     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1702                               "date", &error_fatal);
1703 }
1704 
1705 /* Returns whether we want to use VGA or not */
1706 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1707 {
1708     switch (vga_interface_type) {
1709     case VGA_NONE:
1710         return false;
1711     case VGA_DEVICE:
1712         return true;
1713     case VGA_STD:
1714     case VGA_VIRTIO:
1715         return pci_vga_init(pci_bus) != NULL;
1716     default:
1717         error_setg(errp,
1718                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1719         return false;
1720     }
1721 }
1722 
1723 static int spapr_pre_load(void *opaque)
1724 {
1725     int rc;
1726 
1727     rc = spapr_caps_pre_load(opaque);
1728     if (rc) {
1729         return rc;
1730     }
1731 
1732     return 0;
1733 }
1734 
1735 static int spapr_post_load(void *opaque, int version_id)
1736 {
1737     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1738     int err = 0;
1739 
1740     err = spapr_caps_post_migration(spapr);
1741     if (err) {
1742         return err;
1743     }
1744 
1745     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1746         CPUState *cs;
1747         CPU_FOREACH(cs) {
1748             PowerPCCPU *cpu = POWERPC_CPU(cs);
1749             icp_resend(ICP(cpu->intc));
1750         }
1751     }
1752 
1753     /* In earlier versions, there was no separate qdev for the PAPR
1754      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1755      * So when migrating from those versions, poke the incoming offset
1756      * value into the RTC device */
1757     if (version_id < 3) {
1758         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1759     }
1760 
1761     if (kvm_enabled() && spapr->patb_entry) {
1762         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1763         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1764         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1765 
1766         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1767         if (err) {
1768             error_report("Process table config unsupported by the host");
1769             return -EINVAL;
1770         }
1771     }
1772 
1773     return err;
1774 }
1775 
1776 static int spapr_pre_save(void *opaque)
1777 {
1778     int rc;
1779 
1780     rc = spapr_caps_pre_save(opaque);
1781     if (rc) {
1782         return rc;
1783     }
1784 
1785     return 0;
1786 }
1787 
1788 static bool version_before_3(void *opaque, int version_id)
1789 {
1790     return version_id < 3;
1791 }
1792 
1793 static bool spapr_pending_events_needed(void *opaque)
1794 {
1795     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1796     return !QTAILQ_EMPTY(&spapr->pending_events);
1797 }
1798 
1799 static const VMStateDescription vmstate_spapr_event_entry = {
1800     .name = "spapr_event_log_entry",
1801     .version_id = 1,
1802     .minimum_version_id = 1,
1803     .fields = (VMStateField[]) {
1804         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1805         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1806         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1807                                      NULL, extended_length),
1808         VMSTATE_END_OF_LIST()
1809     },
1810 };
1811 
1812 static const VMStateDescription vmstate_spapr_pending_events = {
1813     .name = "spapr_pending_events",
1814     .version_id = 1,
1815     .minimum_version_id = 1,
1816     .needed = spapr_pending_events_needed,
1817     .fields = (VMStateField[]) {
1818         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1819                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1820         VMSTATE_END_OF_LIST()
1821     },
1822 };
1823 
1824 static bool spapr_ov5_cas_needed(void *opaque)
1825 {
1826     sPAPRMachineState *spapr = opaque;
1827     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1828     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1829     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1830     bool cas_needed;
1831 
1832     /* Prior to the introduction of sPAPROptionVector, we had two option
1833      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1834      * Both of these options encode machine topology into the device-tree
1835      * in such a way that the now-booted OS should still be able to interact
1836      * appropriately with QEMU regardless of what options were actually
1837      * negotiatied on the source side.
1838      *
1839      * As such, we can avoid migrating the CAS-negotiated options if these
1840      * are the only options available on the current machine/platform.
1841      * Since these are the only options available for pseries-2.7 and
1842      * earlier, this allows us to maintain old->new/new->old migration
1843      * compatibility.
1844      *
1845      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1846      * via default pseries-2.8 machines and explicit command-line parameters.
1847      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1848      * of the actual CAS-negotiated values to continue working properly. For
1849      * example, availability of memory unplug depends on knowing whether
1850      * OV5_HP_EVT was negotiated via CAS.
1851      *
1852      * Thus, for any cases where the set of available CAS-negotiatable
1853      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1854      * include the CAS-negotiated options in the migration stream, unless
1855      * if they affect boot time behaviour only.
1856      */
1857     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1858     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1859     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1860 
1861     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1862      * the mask itself since in the future it's possible "legacy" bits may be
1863      * removed via machine options, which could generate a false positive
1864      * that breaks migration.
1865      */
1866     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1867     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1868 
1869     spapr_ovec_cleanup(ov5_mask);
1870     spapr_ovec_cleanup(ov5_legacy);
1871     spapr_ovec_cleanup(ov5_removed);
1872 
1873     return cas_needed;
1874 }
1875 
1876 static const VMStateDescription vmstate_spapr_ov5_cas = {
1877     .name = "spapr_option_vector_ov5_cas",
1878     .version_id = 1,
1879     .minimum_version_id = 1,
1880     .needed = spapr_ov5_cas_needed,
1881     .fields = (VMStateField[]) {
1882         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1883                                  vmstate_spapr_ovec, sPAPROptionVector),
1884         VMSTATE_END_OF_LIST()
1885     },
1886 };
1887 
1888 static bool spapr_patb_entry_needed(void *opaque)
1889 {
1890     sPAPRMachineState *spapr = opaque;
1891 
1892     return !!spapr->patb_entry;
1893 }
1894 
1895 static const VMStateDescription vmstate_spapr_patb_entry = {
1896     .name = "spapr_patb_entry",
1897     .version_id = 1,
1898     .minimum_version_id = 1,
1899     .needed = spapr_patb_entry_needed,
1900     .fields = (VMStateField[]) {
1901         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1902         VMSTATE_END_OF_LIST()
1903     },
1904 };
1905 
1906 static const VMStateDescription vmstate_spapr = {
1907     .name = "spapr",
1908     .version_id = 3,
1909     .minimum_version_id = 1,
1910     .pre_load = spapr_pre_load,
1911     .post_load = spapr_post_load,
1912     .pre_save = spapr_pre_save,
1913     .fields = (VMStateField[]) {
1914         /* used to be @next_irq */
1915         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1916 
1917         /* RTC offset */
1918         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1919 
1920         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1921         VMSTATE_END_OF_LIST()
1922     },
1923     .subsections = (const VMStateDescription*[]) {
1924         &vmstate_spapr_ov5_cas,
1925         &vmstate_spapr_patb_entry,
1926         &vmstate_spapr_pending_events,
1927         &vmstate_spapr_cap_htm,
1928         &vmstate_spapr_cap_vsx,
1929         &vmstate_spapr_cap_dfp,
1930         &vmstate_spapr_cap_cfpc,
1931         &vmstate_spapr_cap_sbbc,
1932         &vmstate_spapr_cap_ibs,
1933         NULL
1934     }
1935 };
1936 
1937 static int htab_save_setup(QEMUFile *f, void *opaque)
1938 {
1939     sPAPRMachineState *spapr = opaque;
1940 
1941     /* "Iteration" header */
1942     if (!spapr->htab_shift) {
1943         qemu_put_be32(f, -1);
1944     } else {
1945         qemu_put_be32(f, spapr->htab_shift);
1946     }
1947 
1948     if (spapr->htab) {
1949         spapr->htab_save_index = 0;
1950         spapr->htab_first_pass = true;
1951     } else {
1952         if (spapr->htab_shift) {
1953             assert(kvm_enabled());
1954         }
1955     }
1956 
1957 
1958     return 0;
1959 }
1960 
1961 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1962                             int chunkstart, int n_valid, int n_invalid)
1963 {
1964     qemu_put_be32(f, chunkstart);
1965     qemu_put_be16(f, n_valid);
1966     qemu_put_be16(f, n_invalid);
1967     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1968                     HASH_PTE_SIZE_64 * n_valid);
1969 }
1970 
1971 static void htab_save_end_marker(QEMUFile *f)
1972 {
1973     qemu_put_be32(f, 0);
1974     qemu_put_be16(f, 0);
1975     qemu_put_be16(f, 0);
1976 }
1977 
1978 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1979                                  int64_t max_ns)
1980 {
1981     bool has_timeout = max_ns != -1;
1982     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1983     int index = spapr->htab_save_index;
1984     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1985 
1986     assert(spapr->htab_first_pass);
1987 
1988     do {
1989         int chunkstart;
1990 
1991         /* Consume invalid HPTEs */
1992         while ((index < htabslots)
1993                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1994             CLEAN_HPTE(HPTE(spapr->htab, index));
1995             index++;
1996         }
1997 
1998         /* Consume valid HPTEs */
1999         chunkstart = index;
2000         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2001                && HPTE_VALID(HPTE(spapr->htab, index))) {
2002             CLEAN_HPTE(HPTE(spapr->htab, index));
2003             index++;
2004         }
2005 
2006         if (index > chunkstart) {
2007             int n_valid = index - chunkstart;
2008 
2009             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2010 
2011             if (has_timeout &&
2012                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2013                 break;
2014             }
2015         }
2016     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2017 
2018     if (index >= htabslots) {
2019         assert(index == htabslots);
2020         index = 0;
2021         spapr->htab_first_pass = false;
2022     }
2023     spapr->htab_save_index = index;
2024 }
2025 
2026 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2027                                 int64_t max_ns)
2028 {
2029     bool final = max_ns < 0;
2030     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2031     int examined = 0, sent = 0;
2032     int index = spapr->htab_save_index;
2033     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2034 
2035     assert(!spapr->htab_first_pass);
2036 
2037     do {
2038         int chunkstart, invalidstart;
2039 
2040         /* Consume non-dirty HPTEs */
2041         while ((index < htabslots)
2042                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2043             index++;
2044             examined++;
2045         }
2046 
2047         chunkstart = index;
2048         /* Consume valid dirty HPTEs */
2049         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2050                && HPTE_DIRTY(HPTE(spapr->htab, index))
2051                && HPTE_VALID(HPTE(spapr->htab, index))) {
2052             CLEAN_HPTE(HPTE(spapr->htab, index));
2053             index++;
2054             examined++;
2055         }
2056 
2057         invalidstart = index;
2058         /* Consume invalid dirty HPTEs */
2059         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2060                && HPTE_DIRTY(HPTE(spapr->htab, index))
2061                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2062             CLEAN_HPTE(HPTE(spapr->htab, index));
2063             index++;
2064             examined++;
2065         }
2066 
2067         if (index > chunkstart) {
2068             int n_valid = invalidstart - chunkstart;
2069             int n_invalid = index - invalidstart;
2070 
2071             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2072             sent += index - chunkstart;
2073 
2074             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2075                 break;
2076             }
2077         }
2078 
2079         if (examined >= htabslots) {
2080             break;
2081         }
2082 
2083         if (index >= htabslots) {
2084             assert(index == htabslots);
2085             index = 0;
2086         }
2087     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2088 
2089     if (index >= htabslots) {
2090         assert(index == htabslots);
2091         index = 0;
2092     }
2093 
2094     spapr->htab_save_index = index;
2095 
2096     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2097 }
2098 
2099 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2100 #define MAX_KVM_BUF_SIZE    2048
2101 
2102 static int htab_save_iterate(QEMUFile *f, void *opaque)
2103 {
2104     sPAPRMachineState *spapr = opaque;
2105     int fd;
2106     int rc = 0;
2107 
2108     /* Iteration header */
2109     if (!spapr->htab_shift) {
2110         qemu_put_be32(f, -1);
2111         return 1;
2112     } else {
2113         qemu_put_be32(f, 0);
2114     }
2115 
2116     if (!spapr->htab) {
2117         assert(kvm_enabled());
2118 
2119         fd = get_htab_fd(spapr);
2120         if (fd < 0) {
2121             return fd;
2122         }
2123 
2124         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2125         if (rc < 0) {
2126             return rc;
2127         }
2128     } else  if (spapr->htab_first_pass) {
2129         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2130     } else {
2131         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2132     }
2133 
2134     htab_save_end_marker(f);
2135 
2136     return rc;
2137 }
2138 
2139 static int htab_save_complete(QEMUFile *f, void *opaque)
2140 {
2141     sPAPRMachineState *spapr = opaque;
2142     int fd;
2143 
2144     /* Iteration header */
2145     if (!spapr->htab_shift) {
2146         qemu_put_be32(f, -1);
2147         return 0;
2148     } else {
2149         qemu_put_be32(f, 0);
2150     }
2151 
2152     if (!spapr->htab) {
2153         int rc;
2154 
2155         assert(kvm_enabled());
2156 
2157         fd = get_htab_fd(spapr);
2158         if (fd < 0) {
2159             return fd;
2160         }
2161 
2162         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2163         if (rc < 0) {
2164             return rc;
2165         }
2166     } else {
2167         if (spapr->htab_first_pass) {
2168             htab_save_first_pass(f, spapr, -1);
2169         }
2170         htab_save_later_pass(f, spapr, -1);
2171     }
2172 
2173     /* End marker */
2174     htab_save_end_marker(f);
2175 
2176     return 0;
2177 }
2178 
2179 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2180 {
2181     sPAPRMachineState *spapr = opaque;
2182     uint32_t section_hdr;
2183     int fd = -1;
2184     Error *local_err = NULL;
2185 
2186     if (version_id < 1 || version_id > 1) {
2187         error_report("htab_load() bad version");
2188         return -EINVAL;
2189     }
2190 
2191     section_hdr = qemu_get_be32(f);
2192 
2193     if (section_hdr == -1) {
2194         spapr_free_hpt(spapr);
2195         return 0;
2196     }
2197 
2198     if (section_hdr) {
2199         /* First section gives the htab size */
2200         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2201         if (local_err) {
2202             error_report_err(local_err);
2203             return -EINVAL;
2204         }
2205         return 0;
2206     }
2207 
2208     if (!spapr->htab) {
2209         assert(kvm_enabled());
2210 
2211         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2212         if (fd < 0) {
2213             error_report_err(local_err);
2214             return fd;
2215         }
2216     }
2217 
2218     while (true) {
2219         uint32_t index;
2220         uint16_t n_valid, n_invalid;
2221 
2222         index = qemu_get_be32(f);
2223         n_valid = qemu_get_be16(f);
2224         n_invalid = qemu_get_be16(f);
2225 
2226         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2227             /* End of Stream */
2228             break;
2229         }
2230 
2231         if ((index + n_valid + n_invalid) >
2232             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2233             /* Bad index in stream */
2234             error_report(
2235                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2236                 index, n_valid, n_invalid, spapr->htab_shift);
2237             return -EINVAL;
2238         }
2239 
2240         if (spapr->htab) {
2241             if (n_valid) {
2242                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2243                                 HASH_PTE_SIZE_64 * n_valid);
2244             }
2245             if (n_invalid) {
2246                 memset(HPTE(spapr->htab, index + n_valid), 0,
2247                        HASH_PTE_SIZE_64 * n_invalid);
2248             }
2249         } else {
2250             int rc;
2251 
2252             assert(fd >= 0);
2253 
2254             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2255             if (rc < 0) {
2256                 return rc;
2257             }
2258         }
2259     }
2260 
2261     if (!spapr->htab) {
2262         assert(fd >= 0);
2263         close(fd);
2264     }
2265 
2266     return 0;
2267 }
2268 
2269 static void htab_save_cleanup(void *opaque)
2270 {
2271     sPAPRMachineState *spapr = opaque;
2272 
2273     close_htab_fd(spapr);
2274 }
2275 
2276 static SaveVMHandlers savevm_htab_handlers = {
2277     .save_setup = htab_save_setup,
2278     .save_live_iterate = htab_save_iterate,
2279     .save_live_complete_precopy = htab_save_complete,
2280     .save_cleanup = htab_save_cleanup,
2281     .load_state = htab_load,
2282 };
2283 
2284 static void spapr_boot_set(void *opaque, const char *boot_device,
2285                            Error **errp)
2286 {
2287     MachineState *machine = MACHINE(opaque);
2288     machine->boot_order = g_strdup(boot_device);
2289 }
2290 
2291 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2292 {
2293     MachineState *machine = MACHINE(spapr);
2294     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2295     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2296     int i;
2297 
2298     for (i = 0; i < nr_lmbs; i++) {
2299         uint64_t addr;
2300 
2301         addr = i * lmb_size + machine->device_memory->base;
2302         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2303                                addr / lmb_size);
2304     }
2305 }
2306 
2307 /*
2308  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2309  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2310  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2311  */
2312 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2313 {
2314     int i;
2315 
2316     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2317         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2318                    " is not aligned to %llu MiB",
2319                    machine->ram_size,
2320                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2321         return;
2322     }
2323 
2324     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2325         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2326                    " is not aligned to %llu MiB",
2327                    machine->ram_size,
2328                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2329         return;
2330     }
2331 
2332     for (i = 0; i < nb_numa_nodes; i++) {
2333         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2334             error_setg(errp,
2335                        "Node %d memory size 0x%" PRIx64
2336                        " is not aligned to %llu MiB",
2337                        i, numa_info[i].node_mem,
2338                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2339             return;
2340         }
2341     }
2342 }
2343 
2344 /* find cpu slot in machine->possible_cpus by core_id */
2345 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2346 {
2347     int index = id / smp_threads;
2348 
2349     if (index >= ms->possible_cpus->len) {
2350         return NULL;
2351     }
2352     if (idx) {
2353         *idx = index;
2354     }
2355     return &ms->possible_cpus->cpus[index];
2356 }
2357 
2358 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2359 {
2360     Error *local_err = NULL;
2361     bool vsmt_user = !!spapr->vsmt;
2362     int kvm_smt = kvmppc_smt_threads();
2363     int ret;
2364 
2365     if (!kvm_enabled() && (smp_threads > 1)) {
2366         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2367                      "on a pseries machine");
2368         goto out;
2369     }
2370     if (!is_power_of_2(smp_threads)) {
2371         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2372                      "machine because it must be a power of 2", smp_threads);
2373         goto out;
2374     }
2375 
2376     /* Detemine the VSMT mode to use: */
2377     if (vsmt_user) {
2378         if (spapr->vsmt < smp_threads) {
2379             error_setg(&local_err, "Cannot support VSMT mode %d"
2380                          " because it must be >= threads/core (%d)",
2381                          spapr->vsmt, smp_threads);
2382             goto out;
2383         }
2384         /* In this case, spapr->vsmt has been set by the command line */
2385     } else {
2386         /*
2387          * Default VSMT value is tricky, because we need it to be as
2388          * consistent as possible (for migration), but this requires
2389          * changing it for at least some existing cases.  We pick 8 as
2390          * the value that we'd get with KVM on POWER8, the
2391          * overwhelmingly common case in production systems.
2392          */
2393         spapr->vsmt = MAX(8, smp_threads);
2394     }
2395 
2396     /* KVM: If necessary, set the SMT mode: */
2397     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2398         ret = kvmppc_set_smt_threads(spapr->vsmt);
2399         if (ret) {
2400             /* Looks like KVM isn't able to change VSMT mode */
2401             error_setg(&local_err,
2402                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2403                        spapr->vsmt, ret);
2404             /* We can live with that if the default one is big enough
2405              * for the number of threads, and a submultiple of the one
2406              * we want.  In this case we'll waste some vcpu ids, but
2407              * behaviour will be correct */
2408             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2409                 warn_report_err(local_err);
2410                 local_err = NULL;
2411                 goto out;
2412             } else {
2413                 if (!vsmt_user) {
2414                     error_append_hint(&local_err,
2415                                       "On PPC, a VM with %d threads/core"
2416                                       " on a host with %d threads/core"
2417                                       " requires the use of VSMT mode %d.\n",
2418                                       smp_threads, kvm_smt, spapr->vsmt);
2419                 }
2420                 kvmppc_hint_smt_possible(&local_err);
2421                 goto out;
2422             }
2423         }
2424     }
2425     /* else TCG: nothing to do currently */
2426 out:
2427     error_propagate(errp, local_err);
2428 }
2429 
2430 static void spapr_init_cpus(sPAPRMachineState *spapr)
2431 {
2432     MachineState *machine = MACHINE(spapr);
2433     MachineClass *mc = MACHINE_GET_CLASS(machine);
2434     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2435     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2436     const CPUArchIdList *possible_cpus;
2437     int boot_cores_nr = smp_cpus / smp_threads;
2438     int i;
2439 
2440     possible_cpus = mc->possible_cpu_arch_ids(machine);
2441     if (mc->has_hotpluggable_cpus) {
2442         if (smp_cpus % smp_threads) {
2443             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2444                          smp_cpus, smp_threads);
2445             exit(1);
2446         }
2447         if (max_cpus % smp_threads) {
2448             error_report("max_cpus (%u) must be multiple of threads (%u)",
2449                          max_cpus, smp_threads);
2450             exit(1);
2451         }
2452     } else {
2453         if (max_cpus != smp_cpus) {
2454             error_report("This machine version does not support CPU hotplug");
2455             exit(1);
2456         }
2457         boot_cores_nr = possible_cpus->len;
2458     }
2459 
2460     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2461      * call xics_max_server_number() or spapr_vcpu_id().
2462      */
2463     spapr_set_vsmt_mode(spapr, &error_fatal);
2464 
2465     if (smc->pre_2_10_has_unused_icps) {
2466         int i;
2467 
2468         for (i = 0; i < xics_max_server_number(spapr); i++) {
2469             /* Dummy entries get deregistered when real ICPState objects
2470              * are registered during CPU core hotplug.
2471              */
2472             pre_2_10_vmstate_register_dummy_icp(i);
2473         }
2474     }
2475 
2476     for (i = 0; i < possible_cpus->len; i++) {
2477         int core_id = i * smp_threads;
2478 
2479         if (mc->has_hotpluggable_cpus) {
2480             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2481                                    spapr_vcpu_id(spapr, core_id));
2482         }
2483 
2484         if (i < boot_cores_nr) {
2485             Object *core  = object_new(type);
2486             int nr_threads = smp_threads;
2487 
2488             /* Handle the partially filled core for older machine types */
2489             if ((i + 1) * smp_threads >= smp_cpus) {
2490                 nr_threads = smp_cpus - i * smp_threads;
2491             }
2492 
2493             object_property_set_int(core, nr_threads, "nr-threads",
2494                                     &error_fatal);
2495             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2496                                     &error_fatal);
2497             object_property_set_bool(core, true, "realized", &error_fatal);
2498         }
2499     }
2500 }
2501 
2502 /* pSeries LPAR / sPAPR hardware init */
2503 static void spapr_machine_init(MachineState *machine)
2504 {
2505     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2506     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2507     const char *kernel_filename = machine->kernel_filename;
2508     const char *initrd_filename = machine->initrd_filename;
2509     PCIHostState *phb;
2510     int i;
2511     MemoryRegion *sysmem = get_system_memory();
2512     MemoryRegion *ram = g_new(MemoryRegion, 1);
2513     hwaddr node0_size = spapr_node0_size(machine);
2514     long load_limit, fw_size;
2515     char *filename;
2516     Error *resize_hpt_err = NULL;
2517     PowerPCCPU *first_ppc_cpu;
2518 
2519     msi_nonbroken = true;
2520 
2521     QLIST_INIT(&spapr->phbs);
2522     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2523 
2524     /* Check HPT resizing availability */
2525     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2526     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2527         /*
2528          * If the user explicitly requested a mode we should either
2529          * supply it, or fail completely (which we do below).  But if
2530          * it's not set explicitly, we reset our mode to something
2531          * that works
2532          */
2533         if (resize_hpt_err) {
2534             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2535             error_free(resize_hpt_err);
2536             resize_hpt_err = NULL;
2537         } else {
2538             spapr->resize_hpt = smc->resize_hpt_default;
2539         }
2540     }
2541 
2542     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2543 
2544     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2545         /*
2546          * User requested HPT resize, but this host can't supply it.  Bail out
2547          */
2548         error_report_err(resize_hpt_err);
2549         exit(1);
2550     }
2551 
2552     spapr->rma_size = node0_size;
2553 
2554     /* With KVM, we don't actually know whether KVM supports an
2555      * unbounded RMA (PR KVM) or is limited by the hash table size
2556      * (HV KVM using VRMA), so we always assume the latter
2557      *
2558      * In that case, we also limit the initial allocations for RTAS
2559      * etc... to 256M since we have no way to know what the VRMA size
2560      * is going to be as it depends on the size of the hash table
2561      * which isn't determined yet.
2562      */
2563     if (kvm_enabled()) {
2564         spapr->vrma_adjust = 1;
2565         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2566     }
2567 
2568     /* Actually we don't support unbounded RMA anymore since we added
2569      * proper emulation of HV mode. The max we can get is 16G which
2570      * also happens to be what we configure for PAPR mode so make sure
2571      * we don't do anything bigger than that
2572      */
2573     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2574 
2575     if (spapr->rma_size > node0_size) {
2576         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2577                      spapr->rma_size);
2578         exit(1);
2579     }
2580 
2581     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2582     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2583 
2584     /* Set up Interrupt Controller before we create the VCPUs */
2585     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2586 
2587     /* Set up containers for ibm,client-architecture-support negotiated options
2588      */
2589     spapr->ov5 = spapr_ovec_new();
2590     spapr->ov5_cas = spapr_ovec_new();
2591 
2592     if (smc->dr_lmb_enabled) {
2593         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2594         spapr_validate_node_memory(machine, &error_fatal);
2595     }
2596 
2597     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
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     first_ppc_cpu = POWERPC_CPU(first_cpu);
2616     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2617         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
2618                          spapr->max_compat_pvr)) {
2619         /* KVM and TCG always allow GTSE with radix... */
2620         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2621     }
2622     /* ... but not with hash (currently). */
2623 
2624     if (kvm_enabled()) {
2625         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2626         kvmppc_enable_logical_ci_hcalls();
2627         kvmppc_enable_set_mode_hcall();
2628 
2629         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2630         kvmppc_enable_clear_ref_mod_hcalls();
2631     }
2632 
2633     /* allocate RAM */
2634     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2635                                          machine->ram_size);
2636     memory_region_add_subregion(sysmem, 0, ram);
2637 
2638     /* always allocate the device memory information */
2639     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2640 
2641     /* initialize hotplug memory address space */
2642     if (machine->ram_size < machine->maxram_size) {
2643         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2644         /*
2645          * Limit the number of hotpluggable memory slots to half the number
2646          * slots that KVM supports, leaving the other half for PCI and other
2647          * devices. However ensure that number of slots doesn't drop below 32.
2648          */
2649         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2650                            SPAPR_MAX_RAM_SLOTS;
2651 
2652         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2653             max_memslots = SPAPR_MAX_RAM_SLOTS;
2654         }
2655         if (machine->ram_slots > max_memslots) {
2656             error_report("Specified number of memory slots %"
2657                          PRIu64" exceeds max supported %d",
2658                          machine->ram_slots, max_memslots);
2659             exit(1);
2660         }
2661 
2662         machine->device_memory->base = ROUND_UP(machine->ram_size,
2663                                                 SPAPR_DEVICE_MEM_ALIGN);
2664         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2665                            "device-memory", device_mem_size);
2666         memory_region_add_subregion(sysmem, machine->device_memory->base,
2667                                     &machine->device_memory->mr);
2668     }
2669 
2670     if (smc->dr_lmb_enabled) {
2671         spapr_create_lmb_dr_connectors(spapr);
2672     }
2673 
2674     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2675     if (!filename) {
2676         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2677         exit(1);
2678     }
2679     spapr->rtas_size = get_image_size(filename);
2680     if (spapr->rtas_size < 0) {
2681         error_report("Could not get size of LPAR rtas '%s'", filename);
2682         exit(1);
2683     }
2684     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2685     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2686         error_report("Could not load LPAR rtas '%s'", filename);
2687         exit(1);
2688     }
2689     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2690         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2691                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2692         exit(1);
2693     }
2694     g_free(filename);
2695 
2696     /* Set up RTAS event infrastructure */
2697     spapr_events_init(spapr);
2698 
2699     /* Set up the RTC RTAS interfaces */
2700     spapr_rtc_create(spapr);
2701 
2702     /* Set up VIO bus */
2703     spapr->vio_bus = spapr_vio_bus_init();
2704 
2705     for (i = 0; i < serial_max_hds(); i++) {
2706         if (serial_hd(i)) {
2707             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2708         }
2709     }
2710 
2711     /* We always have at least the nvram device on VIO */
2712     spapr_create_nvram(spapr);
2713 
2714     /* Set up PCI */
2715     spapr_pci_rtas_init();
2716 
2717     phb = spapr_create_phb(spapr, 0);
2718 
2719     for (i = 0; i < nb_nics; i++) {
2720         NICInfo *nd = &nd_table[i];
2721 
2722         if (!nd->model) {
2723             nd->model = g_strdup("spapr-vlan");
2724         }
2725 
2726         if (g_str_equal(nd->model, "spapr-vlan") ||
2727             g_str_equal(nd->model, "ibmveth")) {
2728             spapr_vlan_create(spapr->vio_bus, nd);
2729         } else {
2730             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2731         }
2732     }
2733 
2734     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2735         spapr_vscsi_create(spapr->vio_bus);
2736     }
2737 
2738     /* Graphics */
2739     if (spapr_vga_init(phb->bus, &error_fatal)) {
2740         spapr->has_graphics = true;
2741         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2742     }
2743 
2744     if (machine->usb) {
2745         if (smc->use_ohci_by_default) {
2746             pci_create_simple(phb->bus, -1, "pci-ohci");
2747         } else {
2748             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2749         }
2750 
2751         if (spapr->has_graphics) {
2752             USBBus *usb_bus = usb_bus_find(-1);
2753 
2754             usb_create_simple(usb_bus, "usb-kbd");
2755             usb_create_simple(usb_bus, "usb-mouse");
2756         }
2757     }
2758 
2759     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2760         error_report(
2761             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2762             MIN_RMA_SLOF);
2763         exit(1);
2764     }
2765 
2766     if (kernel_filename) {
2767         uint64_t lowaddr = 0;
2768 
2769         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2770                                       NULL, NULL, &lowaddr, NULL, 1,
2771                                       PPC_ELF_MACHINE, 0, 0);
2772         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2773             spapr->kernel_size = load_elf(kernel_filename,
2774                                           translate_kernel_address, NULL, NULL,
2775                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2776                                           0, 0);
2777             spapr->kernel_le = spapr->kernel_size > 0;
2778         }
2779         if (spapr->kernel_size < 0) {
2780             error_report("error loading %s: %s", kernel_filename,
2781                          load_elf_strerror(spapr->kernel_size));
2782             exit(1);
2783         }
2784 
2785         /* load initrd */
2786         if (initrd_filename) {
2787             /* Try to locate the initrd in the gap between the kernel
2788              * and the firmware. Add a bit of space just in case
2789              */
2790             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2791                                   + 0x1ffff) & ~0xffff;
2792             spapr->initrd_size = load_image_targphys(initrd_filename,
2793                                                      spapr->initrd_base,
2794                                                      load_limit
2795                                                      - spapr->initrd_base);
2796             if (spapr->initrd_size < 0) {
2797                 error_report("could not load initial ram disk '%s'",
2798                              initrd_filename);
2799                 exit(1);
2800             }
2801         }
2802     }
2803 
2804     if (bios_name == NULL) {
2805         bios_name = FW_FILE_NAME;
2806     }
2807     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2808     if (!filename) {
2809         error_report("Could not find LPAR firmware '%s'", bios_name);
2810         exit(1);
2811     }
2812     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2813     if (fw_size <= 0) {
2814         error_report("Could not load LPAR firmware '%s'", filename);
2815         exit(1);
2816     }
2817     g_free(filename);
2818 
2819     /* FIXME: Should register things through the MachineState's qdev
2820      * interface, this is a legacy from the sPAPREnvironment structure
2821      * which predated MachineState but had a similar function */
2822     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2823     register_savevm_live(NULL, "spapr/htab", -1, 1,
2824                          &savevm_htab_handlers, spapr);
2825 
2826     qemu_register_boot_set(spapr_boot_set, spapr);
2827 
2828     if (kvm_enabled()) {
2829         /* to stop and start vmclock */
2830         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2831                                          &spapr->tb);
2832 
2833         kvmppc_spapr_enable_inkernel_multitce();
2834     }
2835 }
2836 
2837 static int spapr_kvm_type(const char *vm_type)
2838 {
2839     if (!vm_type) {
2840         return 0;
2841     }
2842 
2843     if (!strcmp(vm_type, "HV")) {
2844         return 1;
2845     }
2846 
2847     if (!strcmp(vm_type, "PR")) {
2848         return 2;
2849     }
2850 
2851     error_report("Unknown kvm-type specified '%s'", vm_type);
2852     exit(1);
2853 }
2854 
2855 /*
2856  * Implementation of an interface to adjust firmware path
2857  * for the bootindex property handling.
2858  */
2859 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2860                                    DeviceState *dev)
2861 {
2862 #define CAST(type, obj, name) \
2863     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2864     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2865     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2866     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2867 
2868     if (d) {
2869         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2870         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2871         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2872 
2873         if (spapr) {
2874             /*
2875              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2876              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2877              * in the top 16 bits of the 64-bit LUN
2878              */
2879             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2880             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2881                                    (uint64_t)id << 48);
2882         } else if (virtio) {
2883             /*
2884              * We use SRP luns of the form 01000000 | (target << 8) | lun
2885              * in the top 32 bits of the 64-bit LUN
2886              * Note: the quote above is from SLOF and it is wrong,
2887              * the actual binding is:
2888              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2889              */
2890             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2891             if (d->lun >= 256) {
2892                 /* Use the LUN "flat space addressing method" */
2893                 id |= 0x4000;
2894             }
2895             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2896                                    (uint64_t)id << 32);
2897         } else if (usb) {
2898             /*
2899              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2900              * in the top 32 bits of the 64-bit LUN
2901              */
2902             unsigned usb_port = atoi(usb->port->path);
2903             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2904             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2905                                    (uint64_t)id << 32);
2906         }
2907     }
2908 
2909     /*
2910      * SLOF probes the USB devices, and if it recognizes that the device is a
2911      * storage device, it changes its name to "storage" instead of "usb-host",
2912      * and additionally adds a child node for the SCSI LUN, so the correct
2913      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2914      */
2915     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2916         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2917         if (usb_host_dev_is_scsi_storage(usbdev)) {
2918             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2919         }
2920     }
2921 
2922     if (phb) {
2923         /* Replace "pci" with "pci@800000020000000" */
2924         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2925     }
2926 
2927     if (vsc) {
2928         /* Same logic as virtio above */
2929         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2930         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2931     }
2932 
2933     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2934         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2935         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2936         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2937     }
2938 
2939     return NULL;
2940 }
2941 
2942 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2943 {
2944     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2945 
2946     return g_strdup(spapr->kvm_type);
2947 }
2948 
2949 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2950 {
2951     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2952 
2953     g_free(spapr->kvm_type);
2954     spapr->kvm_type = g_strdup(value);
2955 }
2956 
2957 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2958 {
2959     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2960 
2961     return spapr->use_hotplug_event_source;
2962 }
2963 
2964 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2965                                             Error **errp)
2966 {
2967     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2968 
2969     spapr->use_hotplug_event_source = value;
2970 }
2971 
2972 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2973 {
2974     return true;
2975 }
2976 
2977 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2978 {
2979     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2980 
2981     switch (spapr->resize_hpt) {
2982     case SPAPR_RESIZE_HPT_DEFAULT:
2983         return g_strdup("default");
2984     case SPAPR_RESIZE_HPT_DISABLED:
2985         return g_strdup("disabled");
2986     case SPAPR_RESIZE_HPT_ENABLED:
2987         return g_strdup("enabled");
2988     case SPAPR_RESIZE_HPT_REQUIRED:
2989         return g_strdup("required");
2990     }
2991     g_assert_not_reached();
2992 }
2993 
2994 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2995 {
2996     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2997 
2998     if (strcmp(value, "default") == 0) {
2999         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3000     } else if (strcmp(value, "disabled") == 0) {
3001         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3002     } else if (strcmp(value, "enabled") == 0) {
3003         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3004     } else if (strcmp(value, "required") == 0) {
3005         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3006     } else {
3007         error_setg(errp, "Bad value for \"resize-hpt\" property");
3008     }
3009 }
3010 
3011 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3012                                    void *opaque, Error **errp)
3013 {
3014     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3015 }
3016 
3017 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3018                                    void *opaque, Error **errp)
3019 {
3020     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3021 }
3022 
3023 static void spapr_instance_init(Object *obj)
3024 {
3025     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3026 
3027     spapr->htab_fd = -1;
3028     spapr->use_hotplug_event_source = true;
3029     object_property_add_str(obj, "kvm-type",
3030                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3031     object_property_set_description(obj, "kvm-type",
3032                                     "Specifies the KVM virtualization mode (HV, PR)",
3033                                     NULL);
3034     object_property_add_bool(obj, "modern-hotplug-events",
3035                             spapr_get_modern_hotplug_events,
3036                             spapr_set_modern_hotplug_events,
3037                             NULL);
3038     object_property_set_description(obj, "modern-hotplug-events",
3039                                     "Use dedicated hotplug event mechanism in"
3040                                     " place of standard EPOW events when possible"
3041                                     " (required for memory hot-unplug support)",
3042                                     NULL);
3043     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3044                             "Maximum permitted CPU compatibility mode",
3045                             &error_fatal);
3046 
3047     object_property_add_str(obj, "resize-hpt",
3048                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3049     object_property_set_description(obj, "resize-hpt",
3050                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3051                                     NULL);
3052     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3053                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3054     object_property_set_description(obj, "vsmt",
3055                                     "Virtual SMT: KVM behaves as if this were"
3056                                     " the host's SMT mode", &error_abort);
3057     object_property_add_bool(obj, "vfio-no-msix-emulation",
3058                              spapr_get_msix_emulation, NULL, NULL);
3059 }
3060 
3061 static void spapr_machine_finalizefn(Object *obj)
3062 {
3063     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3064 
3065     g_free(spapr->kvm_type);
3066 }
3067 
3068 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3069 {
3070     cpu_synchronize_state(cs);
3071     ppc_cpu_do_system_reset(cs);
3072 }
3073 
3074 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3075 {
3076     CPUState *cs;
3077 
3078     CPU_FOREACH(cs) {
3079         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3080     }
3081 }
3082 
3083 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3084                            uint32_t node, bool dedicated_hp_event_source,
3085                            Error **errp)
3086 {
3087     sPAPRDRConnector *drc;
3088     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3089     int i, fdt_offset, fdt_size;
3090     void *fdt;
3091     uint64_t addr = addr_start;
3092     bool hotplugged = spapr_drc_hotplugged(dev);
3093     Error *local_err = NULL;
3094 
3095     for (i = 0; i < nr_lmbs; i++) {
3096         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3097                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3098         g_assert(drc);
3099 
3100         fdt = create_device_tree(&fdt_size);
3101         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3102                                                 SPAPR_MEMORY_BLOCK_SIZE);
3103 
3104         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3105         if (local_err) {
3106             while (addr > addr_start) {
3107                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3108                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3109                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3110                 spapr_drc_detach(drc);
3111             }
3112             g_free(fdt);
3113             error_propagate(errp, local_err);
3114             return;
3115         }
3116         if (!hotplugged) {
3117             spapr_drc_reset(drc);
3118         }
3119         addr += SPAPR_MEMORY_BLOCK_SIZE;
3120     }
3121     /* send hotplug notification to the
3122      * guest only in case of hotplugged memory
3123      */
3124     if (hotplugged) {
3125         if (dedicated_hp_event_source) {
3126             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3127                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3128             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3129                                                    nr_lmbs,
3130                                                    spapr_drc_index(drc));
3131         } else {
3132             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3133                                            nr_lmbs);
3134         }
3135     }
3136 }
3137 
3138 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3139                               Error **errp)
3140 {
3141     Error *local_err = NULL;
3142     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3143     PCDIMMDevice *dimm = PC_DIMM(dev);
3144     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3145     MemoryRegion *mr;
3146     uint64_t align, size, addr;
3147     uint32_t node;
3148 
3149     mr = ddc->get_memory_region(dimm, &local_err);
3150     if (local_err) {
3151         goto out;
3152     }
3153     align = memory_region_get_alignment(mr);
3154     size = memory_region_size(mr);
3155 
3156     pc_dimm_memory_plug(dev, MACHINE(ms), align, &local_err);
3157     if (local_err) {
3158         goto out;
3159     }
3160 
3161     addr = object_property_get_uint(OBJECT(dimm),
3162                                     PC_DIMM_ADDR_PROP, &local_err);
3163     if (local_err) {
3164         goto out_unplug;
3165     }
3166 
3167     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3168                                     &error_abort);
3169     spapr_add_lmbs(dev, addr, size, node,
3170                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3171                    &local_err);
3172     if (local_err) {
3173         goto out_unplug;
3174     }
3175 
3176     return;
3177 
3178 out_unplug:
3179     pc_dimm_memory_unplug(dev, MACHINE(ms));
3180 out:
3181     error_propagate(errp, local_err);
3182 }
3183 
3184 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3185                                   Error **errp)
3186 {
3187     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3188     PCDIMMDevice *dimm = PC_DIMM(dev);
3189     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3190     MemoryRegion *mr;
3191     uint64_t size;
3192     char *mem_dev;
3193 
3194     if (!smc->dr_lmb_enabled) {
3195         error_setg(errp, "Memory hotplug not supported for this machine");
3196         return;
3197     }
3198 
3199     mr = ddc->get_memory_region(dimm, errp);
3200     if (!mr) {
3201         return;
3202     }
3203     size = memory_region_size(mr);
3204 
3205     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3206         error_setg(errp, "Hotplugged memory size must be a multiple of "
3207                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
3208         return;
3209     }
3210 
3211     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
3212     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
3213         error_setg(errp, "Memory backend has bad page size. "
3214                    "Use 'memory-backend-file' with correct mem-path.");
3215         goto out;
3216     }
3217 
3218 out:
3219     g_free(mem_dev);
3220 }
3221 
3222 struct sPAPRDIMMState {
3223     PCDIMMDevice *dimm;
3224     uint32_t nr_lmbs;
3225     QTAILQ_ENTRY(sPAPRDIMMState) next;
3226 };
3227 
3228 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3229                                                        PCDIMMDevice *dimm)
3230 {
3231     sPAPRDIMMState *dimm_state = NULL;
3232 
3233     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3234         if (dimm_state->dimm == dimm) {
3235             break;
3236         }
3237     }
3238     return dimm_state;
3239 }
3240 
3241 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3242                                                       uint32_t nr_lmbs,
3243                                                       PCDIMMDevice *dimm)
3244 {
3245     sPAPRDIMMState *ds = NULL;
3246 
3247     /*
3248      * If this request is for a DIMM whose removal had failed earlier
3249      * (due to guest's refusal to remove the LMBs), we would have this
3250      * dimm already in the pending_dimm_unplugs list. In that
3251      * case don't add again.
3252      */
3253     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3254     if (!ds) {
3255         ds = g_malloc0(sizeof(sPAPRDIMMState));
3256         ds->nr_lmbs = nr_lmbs;
3257         ds->dimm = dimm;
3258         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3259     }
3260     return ds;
3261 }
3262 
3263 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3264                                               sPAPRDIMMState *dimm_state)
3265 {
3266     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3267     g_free(dimm_state);
3268 }
3269 
3270 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3271                                                         PCDIMMDevice *dimm)
3272 {
3273     sPAPRDRConnector *drc;
3274     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3275     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3276     uint64_t size = memory_region_size(mr);
3277     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3278     uint32_t avail_lmbs = 0;
3279     uint64_t addr_start, addr;
3280     int i;
3281 
3282     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3283                                          &error_abort);
3284 
3285     addr = addr_start;
3286     for (i = 0; i < nr_lmbs; i++) {
3287         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3288                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3289         g_assert(drc);
3290         if (drc->dev) {
3291             avail_lmbs++;
3292         }
3293         addr += SPAPR_MEMORY_BLOCK_SIZE;
3294     }
3295 
3296     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3297 }
3298 
3299 /* Callback to be called during DRC release. */
3300 void spapr_lmb_release(DeviceState *dev)
3301 {
3302     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3303     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3304     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3305 
3306     /* This information will get lost if a migration occurs
3307      * during the unplug process. In this case recover it. */
3308     if (ds == NULL) {
3309         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3310         g_assert(ds);
3311         /* The DRC being examined by the caller at least must be counted */
3312         g_assert(ds->nr_lmbs);
3313     }
3314 
3315     if (--ds->nr_lmbs) {
3316         return;
3317     }
3318 
3319     /*
3320      * Now that all the LMBs have been removed by the guest, call the
3321      * unplug handler chain. This can never fail.
3322      */
3323     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3324 }
3325 
3326 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3327 {
3328     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3329     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3330 
3331     pc_dimm_memory_unplug(dev, MACHINE(hotplug_dev));
3332     object_unparent(OBJECT(dev));
3333     spapr_pending_dimm_unplugs_remove(spapr, ds);
3334 }
3335 
3336 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3337                                         DeviceState *dev, Error **errp)
3338 {
3339     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3340     Error *local_err = NULL;
3341     PCDIMMDevice *dimm = PC_DIMM(dev);
3342     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3343     MemoryRegion *mr;
3344     uint32_t nr_lmbs;
3345     uint64_t size, addr_start, addr;
3346     int i;
3347     sPAPRDRConnector *drc;
3348 
3349     mr = ddc->get_memory_region(dimm, &local_err);
3350     if (local_err) {
3351         goto out;
3352     }
3353     size = memory_region_size(mr);
3354     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3355 
3356     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3357                                          &local_err);
3358     if (local_err) {
3359         goto out;
3360     }
3361 
3362     /*
3363      * An existing pending dimm state for this DIMM means that there is an
3364      * unplug operation in progress, waiting for the spapr_lmb_release
3365      * callback to complete the job (BQL can't cover that far). In this case,
3366      * bail out to avoid detaching DRCs that were already released.
3367      */
3368     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3369         error_setg(&local_err,
3370                    "Memory unplug already in progress for device %s",
3371                    dev->id);
3372         goto out;
3373     }
3374 
3375     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3376 
3377     addr = addr_start;
3378     for (i = 0; i < nr_lmbs; i++) {
3379         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3380                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3381         g_assert(drc);
3382 
3383         spapr_drc_detach(drc);
3384         addr += SPAPR_MEMORY_BLOCK_SIZE;
3385     }
3386 
3387     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3388                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3389     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3390                                               nr_lmbs, spapr_drc_index(drc));
3391 out:
3392     error_propagate(errp, local_err);
3393 }
3394 
3395 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3396                                            sPAPRMachineState *spapr)
3397 {
3398     PowerPCCPU *cpu = POWERPC_CPU(cs);
3399     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3400     int id = spapr_get_vcpu_id(cpu);
3401     void *fdt;
3402     int offset, fdt_size;
3403     char *nodename;
3404 
3405     fdt = create_device_tree(&fdt_size);
3406     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3407     offset = fdt_add_subnode(fdt, 0, nodename);
3408 
3409     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3410     g_free(nodename);
3411 
3412     *fdt_offset = offset;
3413     return fdt;
3414 }
3415 
3416 /* Callback to be called during DRC release. */
3417 void spapr_core_release(DeviceState *dev)
3418 {
3419     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3420 
3421     /* Call the unplug handler chain. This can never fail. */
3422     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3423 }
3424 
3425 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3426 {
3427     MachineState *ms = MACHINE(hotplug_dev);
3428     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3429     CPUCore *cc = CPU_CORE(dev);
3430     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3431 
3432     if (smc->pre_2_10_has_unused_icps) {
3433         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3434         int i;
3435 
3436         for (i = 0; i < cc->nr_threads; i++) {
3437             CPUState *cs = CPU(sc->threads[i]);
3438 
3439             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3440         }
3441     }
3442 
3443     assert(core_slot);
3444     core_slot->cpu = NULL;
3445     object_unparent(OBJECT(dev));
3446 }
3447 
3448 static
3449 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3450                                Error **errp)
3451 {
3452     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3453     int index;
3454     sPAPRDRConnector *drc;
3455     CPUCore *cc = CPU_CORE(dev);
3456 
3457     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3458         error_setg(errp, "Unable to find CPU core with core-id: %d",
3459                    cc->core_id);
3460         return;
3461     }
3462     if (index == 0) {
3463         error_setg(errp, "Boot CPU core may not be unplugged");
3464         return;
3465     }
3466 
3467     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3468                           spapr_vcpu_id(spapr, cc->core_id));
3469     g_assert(drc);
3470 
3471     spapr_drc_detach(drc);
3472 
3473     spapr_hotplug_req_remove_by_index(drc);
3474 }
3475 
3476 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3477                             Error **errp)
3478 {
3479     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3480     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3481     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3482     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3483     CPUCore *cc = CPU_CORE(dev);
3484     CPUState *cs = CPU(core->threads[0]);
3485     sPAPRDRConnector *drc;
3486     Error *local_err = NULL;
3487     CPUArchId *core_slot;
3488     int index;
3489     bool hotplugged = spapr_drc_hotplugged(dev);
3490 
3491     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3492     if (!core_slot) {
3493         error_setg(errp, "Unable to find CPU core with core-id: %d",
3494                    cc->core_id);
3495         return;
3496     }
3497     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3498                           spapr_vcpu_id(spapr, cc->core_id));
3499 
3500     g_assert(drc || !mc->has_hotpluggable_cpus);
3501 
3502     if (drc) {
3503         void *fdt;
3504         int fdt_offset;
3505 
3506         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3507 
3508         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3509         if (local_err) {
3510             g_free(fdt);
3511             error_propagate(errp, local_err);
3512             return;
3513         }
3514 
3515         if (hotplugged) {
3516             /*
3517              * Send hotplug notification interrupt to the guest only
3518              * in case of hotplugged CPUs.
3519              */
3520             spapr_hotplug_req_add_by_index(drc);
3521         } else {
3522             spapr_drc_reset(drc);
3523         }
3524     }
3525 
3526     core_slot->cpu = OBJECT(dev);
3527 
3528     if (smc->pre_2_10_has_unused_icps) {
3529         int i;
3530 
3531         for (i = 0; i < cc->nr_threads; i++) {
3532             cs = CPU(core->threads[i]);
3533             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3534         }
3535     }
3536 }
3537 
3538 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3539                                 Error **errp)
3540 {
3541     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3542     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3543     Error *local_err = NULL;
3544     CPUCore *cc = CPU_CORE(dev);
3545     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3546     const char *type = object_get_typename(OBJECT(dev));
3547     CPUArchId *core_slot;
3548     int index;
3549 
3550     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3551         error_setg(&local_err, "CPU hotplug not supported for this machine");
3552         goto out;
3553     }
3554 
3555     if (strcmp(base_core_type, type)) {
3556         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3557         goto out;
3558     }
3559 
3560     if (cc->core_id % smp_threads) {
3561         error_setg(&local_err, "invalid core id %d", cc->core_id);
3562         goto out;
3563     }
3564 
3565     /*
3566      * In general we should have homogeneous threads-per-core, but old
3567      * (pre hotplug support) machine types allow the last core to have
3568      * reduced threads as a compatibility hack for when we allowed
3569      * total vcpus not a multiple of threads-per-core.
3570      */
3571     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3572         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3573                    cc->nr_threads, smp_threads);
3574         goto out;
3575     }
3576 
3577     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3578     if (!core_slot) {
3579         error_setg(&local_err, "core id %d out of range", cc->core_id);
3580         goto out;
3581     }
3582 
3583     if (core_slot->cpu) {
3584         error_setg(&local_err, "core %d already populated", cc->core_id);
3585         goto out;
3586     }
3587 
3588     numa_cpu_pre_plug(core_slot, dev, &local_err);
3589 
3590 out:
3591     error_propagate(errp, local_err);
3592 }
3593 
3594 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3595                                       DeviceState *dev, Error **errp)
3596 {
3597     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3598         spapr_memory_plug(hotplug_dev, dev, errp);
3599     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3600         spapr_core_plug(hotplug_dev, dev, errp);
3601     }
3602 }
3603 
3604 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3605                                         DeviceState *dev, Error **errp)
3606 {
3607     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3608         spapr_memory_unplug(hotplug_dev, dev);
3609     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3610         spapr_core_unplug(hotplug_dev, dev);
3611     }
3612 }
3613 
3614 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3615                                                 DeviceState *dev, Error **errp)
3616 {
3617     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3618     MachineClass *mc = MACHINE_GET_CLASS(sms);
3619 
3620     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3621         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3622             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3623         } else {
3624             /* NOTE: this means there is a window after guest reset, prior to
3625              * CAS negotiation, where unplug requests will fail due to the
3626              * capability not being detected yet. This is a bit different than
3627              * the case with PCI unplug, where the events will be queued and
3628              * eventually handled by the guest after boot
3629              */
3630             error_setg(errp, "Memory hot unplug not supported for this guest");
3631         }
3632     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3633         if (!mc->has_hotpluggable_cpus) {
3634             error_setg(errp, "CPU hot unplug not supported on this machine");
3635             return;
3636         }
3637         spapr_core_unplug_request(hotplug_dev, dev, errp);
3638     }
3639 }
3640 
3641 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3642                                           DeviceState *dev, Error **errp)
3643 {
3644     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3645         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3646     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3647         spapr_core_pre_plug(hotplug_dev, dev, errp);
3648     }
3649 }
3650 
3651 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3652                                                  DeviceState *dev)
3653 {
3654     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3655         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3656         return HOTPLUG_HANDLER(machine);
3657     }
3658     return NULL;
3659 }
3660 
3661 static CpuInstanceProperties
3662 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3663 {
3664     CPUArchId *core_slot;
3665     MachineClass *mc = MACHINE_GET_CLASS(machine);
3666 
3667     /* make sure possible_cpu are intialized */
3668     mc->possible_cpu_arch_ids(machine);
3669     /* get CPU core slot containing thread that matches cpu_index */
3670     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3671     assert(core_slot);
3672     return core_slot->props;
3673 }
3674 
3675 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3676 {
3677     return idx / smp_cores % nb_numa_nodes;
3678 }
3679 
3680 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3681 {
3682     int i;
3683     const char *core_type;
3684     int spapr_max_cores = max_cpus / smp_threads;
3685     MachineClass *mc = MACHINE_GET_CLASS(machine);
3686 
3687     if (!mc->has_hotpluggable_cpus) {
3688         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3689     }
3690     if (machine->possible_cpus) {
3691         assert(machine->possible_cpus->len == spapr_max_cores);
3692         return machine->possible_cpus;
3693     }
3694 
3695     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3696     if (!core_type) {
3697         error_report("Unable to find sPAPR CPU Core definition");
3698         exit(1);
3699     }
3700 
3701     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3702                              sizeof(CPUArchId) * spapr_max_cores);
3703     machine->possible_cpus->len = spapr_max_cores;
3704     for (i = 0; i < machine->possible_cpus->len; i++) {
3705         int core_id = i * smp_threads;
3706 
3707         machine->possible_cpus->cpus[i].type = core_type;
3708         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3709         machine->possible_cpus->cpus[i].arch_id = core_id;
3710         machine->possible_cpus->cpus[i].props.has_core_id = true;
3711         machine->possible_cpus->cpus[i].props.core_id = core_id;
3712     }
3713     return machine->possible_cpus;
3714 }
3715 
3716 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3717                                 uint64_t *buid, hwaddr *pio,
3718                                 hwaddr *mmio32, hwaddr *mmio64,
3719                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3720 {
3721     /*
3722      * New-style PHB window placement.
3723      *
3724      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3725      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3726      * windows.
3727      *
3728      * Some guest kernels can't work with MMIO windows above 1<<46
3729      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3730      *
3731      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3732      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3733      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3734      * 1TiB 64-bit MMIO windows for each PHB.
3735      */
3736     const uint64_t base_buid = 0x800000020000000ULL;
3737 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3738                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3739     int i;
3740 
3741     /* Sanity check natural alignments */
3742     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3743     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3744     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3745     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3746     /* Sanity check bounds */
3747     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3748                       SPAPR_PCI_MEM32_WIN_SIZE);
3749     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3750                       SPAPR_PCI_MEM64_WIN_SIZE);
3751 
3752     if (index >= SPAPR_MAX_PHBS) {
3753         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3754                    SPAPR_MAX_PHBS - 1);
3755         return;
3756     }
3757 
3758     *buid = base_buid + index;
3759     for (i = 0; i < n_dma; ++i) {
3760         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3761     }
3762 
3763     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3764     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3765     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3766 }
3767 
3768 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3769 {
3770     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3771 
3772     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3773 }
3774 
3775 static void spapr_ics_resend(XICSFabric *dev)
3776 {
3777     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3778 
3779     ics_resend(spapr->ics);
3780 }
3781 
3782 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3783 {
3784     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3785 
3786     return cpu ? ICP(cpu->intc) : NULL;
3787 }
3788 
3789 #define ICS_IRQ_FREE(ics, srcno)   \
3790     (!((ics)->irqs[(srcno)].flags & (XICS_FLAGS_IRQ_MASK)))
3791 
3792 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
3793 {
3794     int first, i;
3795 
3796     for (first = 0; first < ics->nr_irqs; first += alignnum) {
3797         if (num > (ics->nr_irqs - first)) {
3798             return -1;
3799         }
3800         for (i = first; i < first + num; ++i) {
3801             if (!ICS_IRQ_FREE(ics, i)) {
3802                 break;
3803             }
3804         }
3805         if (i == (first + num)) {
3806             return first;
3807         }
3808     }
3809 
3810     return -1;
3811 }
3812 
3813 /*
3814  * Allocate the IRQ number and set the IRQ type, LSI or MSI
3815  */
3816 static void spapr_irq_set_lsi(sPAPRMachineState *spapr, int irq, bool lsi)
3817 {
3818     ics_set_irq_type(spapr->ics, irq - spapr->ics->offset, lsi);
3819 }
3820 
3821 int spapr_irq_alloc(sPAPRMachineState *spapr, int irq_hint, bool lsi,
3822                     Error **errp)
3823 {
3824     ICSState *ics = spapr->ics;
3825     int irq;
3826 
3827     assert(ics);
3828 
3829     if (irq_hint) {
3830         if (!ICS_IRQ_FREE(ics, irq_hint - ics->offset)) {
3831             error_setg(errp, "can't allocate IRQ %d: already in use", irq_hint);
3832             return -1;
3833         }
3834         irq = irq_hint;
3835     } else {
3836         irq = ics_find_free_block(ics, 1, 1);
3837         if (irq < 0) {
3838             error_setg(errp, "can't allocate IRQ: no IRQ left");
3839             return -1;
3840         }
3841         irq += ics->offset;
3842     }
3843 
3844     spapr_irq_set_lsi(spapr, irq, lsi);
3845     trace_spapr_irq_alloc(irq);
3846 
3847     return irq;
3848 }
3849 
3850 /*
3851  * Allocate block of consecutive IRQs, and return the number of the first IRQ in
3852  * the block. If align==true, aligns the first IRQ number to num.
3853  */
3854 int spapr_irq_alloc_block(sPAPRMachineState *spapr, int num, bool lsi,
3855                           bool align, Error **errp)
3856 {
3857     ICSState *ics = spapr->ics;
3858     int i, first = -1;
3859 
3860     assert(ics);
3861 
3862     /*
3863      * MSIMesage::data is used for storing VIRQ so
3864      * it has to be aligned to num to support multiple
3865      * MSI vectors. MSI-X is not affected by this.
3866      * The hint is used for the first IRQ, the rest should
3867      * be allocated continuously.
3868      */
3869     if (align) {
3870         assert((num == 1) || (num == 2) || (num == 4) ||
3871                (num == 8) || (num == 16) || (num == 32));
3872         first = ics_find_free_block(ics, num, num);
3873     } else {
3874         first = ics_find_free_block(ics, num, 1);
3875     }
3876     if (first < 0) {
3877         error_setg(errp, "can't find a free %d-IRQ block", num);
3878         return -1;
3879     }
3880 
3881     first += ics->offset;
3882     for (i = first; i < first + num; ++i) {
3883         spapr_irq_set_lsi(spapr, i, lsi);
3884     }
3885 
3886     trace_spapr_irq_alloc_block(first, num, lsi, align);
3887 
3888     return first;
3889 }
3890 
3891 void spapr_irq_free(sPAPRMachineState *spapr, int irq, int num)
3892 {
3893     ICSState *ics = spapr->ics;
3894     int srcno = irq - ics->offset;
3895     int i;
3896 
3897     if (ics_valid_irq(ics, irq)) {
3898         trace_spapr_irq_free(0, irq, num);
3899         for (i = srcno; i < srcno + num; ++i) {
3900             if (ICS_IRQ_FREE(ics, i)) {
3901                 trace_spapr_irq_free_warn(0, i + ics->offset);
3902             }
3903             memset(&ics->irqs[i], 0, sizeof(ICSIRQState));
3904         }
3905     }
3906 }
3907 
3908 qemu_irq spapr_qirq(sPAPRMachineState *spapr, int irq)
3909 {
3910     ICSState *ics = spapr->ics;
3911 
3912     if (ics_valid_irq(ics, irq)) {
3913         return ics->qirqs[irq - ics->offset];
3914     }
3915 
3916     return NULL;
3917 }
3918 
3919 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3920                                  Monitor *mon)
3921 {
3922     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3923     CPUState *cs;
3924 
3925     CPU_FOREACH(cs) {
3926         PowerPCCPU *cpu = POWERPC_CPU(cs);
3927 
3928         icp_pic_print_info(ICP(cpu->intc), mon);
3929     }
3930 
3931     ics_pic_print_info(spapr->ics, mon);
3932 }
3933 
3934 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3935 {
3936     return cpu->vcpu_id;
3937 }
3938 
3939 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3940 {
3941     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3942     int vcpu_id;
3943 
3944     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3945 
3946     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3947         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3948         error_append_hint(errp, "Adjust the number of cpus to %d "
3949                           "or try to raise the number of threads per core\n",
3950                           vcpu_id * smp_threads / spapr->vsmt);
3951         return;
3952     }
3953 
3954     cpu->vcpu_id = vcpu_id;
3955 }
3956 
3957 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3958 {
3959     CPUState *cs;
3960 
3961     CPU_FOREACH(cs) {
3962         PowerPCCPU *cpu = POWERPC_CPU(cs);
3963 
3964         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3965             return cpu;
3966         }
3967     }
3968 
3969     return NULL;
3970 }
3971 
3972 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3973 {
3974     MachineClass *mc = MACHINE_CLASS(oc);
3975     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3976     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3977     NMIClass *nc = NMI_CLASS(oc);
3978     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3979     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3980     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3981     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3982 
3983     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3984 
3985     /*
3986      * We set up the default / latest behaviour here.  The class_init
3987      * functions for the specific versioned machine types can override
3988      * these details for backwards compatibility
3989      */
3990     mc->init = spapr_machine_init;
3991     mc->reset = spapr_machine_reset;
3992     mc->block_default_type = IF_SCSI;
3993     mc->max_cpus = 1024;
3994     mc->no_parallel = 1;
3995     mc->default_boot_order = "";
3996     mc->default_ram_size = 512 * M_BYTE;
3997     mc->kvm_type = spapr_kvm_type;
3998     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3999     mc->pci_allow_0_address = true;
4000     assert(!mc->get_hotplug_handler);
4001     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4002     hc->pre_plug = spapr_machine_device_pre_plug;
4003     hc->plug = spapr_machine_device_plug;
4004     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4005     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4006     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4007     hc->unplug_request = spapr_machine_device_unplug_request;
4008     hc->unplug = spapr_machine_device_unplug;
4009 
4010     smc->dr_lmb_enabled = true;
4011     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4012     mc->has_hotpluggable_cpus = true;
4013     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4014     fwc->get_dev_path = spapr_get_fw_dev_path;
4015     nc->nmi_monitor_handler = spapr_nmi;
4016     smc->phb_placement = spapr_phb_placement;
4017     vhc->hypercall = emulate_spapr_hypercall;
4018     vhc->hpt_mask = spapr_hpt_mask;
4019     vhc->map_hptes = spapr_map_hptes;
4020     vhc->unmap_hptes = spapr_unmap_hptes;
4021     vhc->store_hpte = spapr_store_hpte;
4022     vhc->get_patbe = spapr_get_patbe;
4023     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4024     xic->ics_get = spapr_ics_get;
4025     xic->ics_resend = spapr_ics_resend;
4026     xic->icp_get = spapr_icp_get;
4027     ispc->print_info = spapr_pic_print_info;
4028     /* Force NUMA node memory size to be a multiple of
4029      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4030      * in which LMBs are represented and hot-added
4031      */
4032     mc->numa_mem_align_shift = 28;
4033 
4034     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4035     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4036     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4037     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4038     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4039     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4040     spapr_caps_add_properties(smc, &error_abort);
4041 }
4042 
4043 static const TypeInfo spapr_machine_info = {
4044     .name          = TYPE_SPAPR_MACHINE,
4045     .parent        = TYPE_MACHINE,
4046     .abstract      = true,
4047     .instance_size = sizeof(sPAPRMachineState),
4048     .instance_init = spapr_instance_init,
4049     .instance_finalize = spapr_machine_finalizefn,
4050     .class_size    = sizeof(sPAPRMachineClass),
4051     .class_init    = spapr_machine_class_init,
4052     .interfaces = (InterfaceInfo[]) {
4053         { TYPE_FW_PATH_PROVIDER },
4054         { TYPE_NMI },
4055         { TYPE_HOTPLUG_HANDLER },
4056         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4057         { TYPE_XICS_FABRIC },
4058         { TYPE_INTERRUPT_STATS_PROVIDER },
4059         { }
4060     },
4061 };
4062 
4063 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4064     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4065                                                     void *data)      \
4066     {                                                                \
4067         MachineClass *mc = MACHINE_CLASS(oc);                        \
4068         spapr_machine_##suffix##_class_options(mc);                  \
4069         if (latest) {                                                \
4070             mc->alias = "pseries";                                   \
4071             mc->is_default = 1;                                      \
4072         }                                                            \
4073     }                                                                \
4074     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
4075     {                                                                \
4076         MachineState *machine = MACHINE(obj);                        \
4077         spapr_machine_##suffix##_instance_options(machine);          \
4078     }                                                                \
4079     static const TypeInfo spapr_machine_##suffix##_info = {          \
4080         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4081         .parent = TYPE_SPAPR_MACHINE,                                \
4082         .class_init = spapr_machine_##suffix##_class_init,           \
4083         .instance_init = spapr_machine_##suffix##_instance_init,     \
4084     };                                                               \
4085     static void spapr_machine_register_##suffix(void)                \
4086     {                                                                \
4087         type_register(&spapr_machine_##suffix##_info);               \
4088     }                                                                \
4089     type_init(spapr_machine_register_##suffix)
4090 
4091 /*
4092  * pseries-3.0
4093  */
4094 static void spapr_machine_3_0_instance_options(MachineState *machine)
4095 {
4096 }
4097 
4098 static void spapr_machine_3_0_class_options(MachineClass *mc)
4099 {
4100     /* Defaults for the latest behaviour inherited from the base class */
4101 }
4102 
4103 DEFINE_SPAPR_MACHINE(3_0, "3.0", true);
4104 
4105 /*
4106  * pseries-2.12
4107  */
4108 #define SPAPR_COMPAT_2_12                                              \
4109     HW_COMPAT_2_12                                                     \
4110     {                                                                  \
4111         .driver = TYPE_POWERPC_CPU,                                    \
4112         .property = "pre-3.0-migration",                              \
4113         .value    = "on",                                              \
4114     },
4115 
4116 static void spapr_machine_2_12_instance_options(MachineState *machine)
4117 {
4118     spapr_machine_3_0_instance_options(machine);
4119 }
4120 
4121 static void spapr_machine_2_12_class_options(MachineClass *mc)
4122 {
4123     spapr_machine_3_0_class_options(mc);
4124     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4125 }
4126 
4127 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4128 
4129 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
4130 {
4131     spapr_machine_2_12_instance_options(machine);
4132 }
4133 
4134 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4135 {
4136     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4137 
4138     spapr_machine_2_12_class_options(mc);
4139     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4140     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4141     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4142 }
4143 
4144 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4145 
4146 /*
4147  * pseries-2.11
4148  */
4149 #define SPAPR_COMPAT_2_11                                              \
4150     HW_COMPAT_2_11
4151 
4152 static void spapr_machine_2_11_instance_options(MachineState *machine)
4153 {
4154     spapr_machine_2_12_instance_options(machine);
4155 }
4156 
4157 static void spapr_machine_2_11_class_options(MachineClass *mc)
4158 {
4159     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4160 
4161     spapr_machine_2_12_class_options(mc);
4162     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4163     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4164 }
4165 
4166 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4167 
4168 /*
4169  * pseries-2.10
4170  */
4171 #define SPAPR_COMPAT_2_10                                              \
4172     HW_COMPAT_2_10
4173 
4174 static void spapr_machine_2_10_instance_options(MachineState *machine)
4175 {
4176     spapr_machine_2_11_instance_options(machine);
4177 }
4178 
4179 static void spapr_machine_2_10_class_options(MachineClass *mc)
4180 {
4181     spapr_machine_2_11_class_options(mc);
4182     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4183 }
4184 
4185 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4186 
4187 /*
4188  * pseries-2.9
4189  */
4190 #define SPAPR_COMPAT_2_9                                               \
4191     HW_COMPAT_2_9                                                      \
4192     {                                                                  \
4193         .driver = TYPE_POWERPC_CPU,                                    \
4194         .property = "pre-2.10-migration",                              \
4195         .value    = "on",                                              \
4196     },                                                                 \
4197 
4198 static void spapr_machine_2_9_instance_options(MachineState *machine)
4199 {
4200     spapr_machine_2_10_instance_options(machine);
4201 }
4202 
4203 static void spapr_machine_2_9_class_options(MachineClass *mc)
4204 {
4205     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4206 
4207     spapr_machine_2_10_class_options(mc);
4208     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4209     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4210     smc->pre_2_10_has_unused_icps = true;
4211     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4212 }
4213 
4214 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4215 
4216 /*
4217  * pseries-2.8
4218  */
4219 #define SPAPR_COMPAT_2_8                                        \
4220     HW_COMPAT_2_8                                               \
4221     {                                                           \
4222         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4223         .property = "pcie-extended-configuration-space",        \
4224         .value    = "off",                                      \
4225     },
4226 
4227 static void spapr_machine_2_8_instance_options(MachineState *machine)
4228 {
4229     spapr_machine_2_9_instance_options(machine);
4230 }
4231 
4232 static void spapr_machine_2_8_class_options(MachineClass *mc)
4233 {
4234     spapr_machine_2_9_class_options(mc);
4235     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4236     mc->numa_mem_align_shift = 23;
4237 }
4238 
4239 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4240 
4241 /*
4242  * pseries-2.7
4243  */
4244 #define SPAPR_COMPAT_2_7                            \
4245     HW_COMPAT_2_7                                   \
4246     {                                               \
4247         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4248         .property = "mem_win_size",                 \
4249         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4250     },                                              \
4251     {                                               \
4252         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4253         .property = "mem64_win_size",               \
4254         .value    = "0",                            \
4255     },                                              \
4256     {                                               \
4257         .driver = TYPE_POWERPC_CPU,                 \
4258         .property = "pre-2.8-migration",            \
4259         .value    = "on",                           \
4260     },                                              \
4261     {                                               \
4262         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4263         .property = "pre-2.8-migration",            \
4264         .value    = "on",                           \
4265     },
4266 
4267 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4268                               uint64_t *buid, hwaddr *pio,
4269                               hwaddr *mmio32, hwaddr *mmio64,
4270                               unsigned n_dma, uint32_t *liobns, Error **errp)
4271 {
4272     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4273     const uint64_t base_buid = 0x800000020000000ULL;
4274     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4275     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4276     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4277     const uint32_t max_index = 255;
4278     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4279 
4280     uint64_t ram_top = MACHINE(spapr)->ram_size;
4281     hwaddr phb0_base, phb_base;
4282     int i;
4283 
4284     /* Do we have device memory? */
4285     if (MACHINE(spapr)->maxram_size > ram_top) {
4286         /* Can't just use maxram_size, because there may be an
4287          * alignment gap between normal and device memory regions
4288          */
4289         ram_top = MACHINE(spapr)->device_memory->base +
4290             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4291     }
4292 
4293     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4294 
4295     if (index > max_index) {
4296         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4297                    max_index);
4298         return;
4299     }
4300 
4301     *buid = base_buid + index;
4302     for (i = 0; i < n_dma; ++i) {
4303         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4304     }
4305 
4306     phb_base = phb0_base + index * phb_spacing;
4307     *pio = phb_base + pio_offset;
4308     *mmio32 = phb_base + mmio_offset;
4309     /*
4310      * We don't set the 64-bit MMIO window, relying on the PHB's
4311      * fallback behaviour of automatically splitting a large "32-bit"
4312      * window into contiguous 32-bit and 64-bit windows
4313      */
4314 }
4315 
4316 static void spapr_machine_2_7_instance_options(MachineState *machine)
4317 {
4318     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4319 
4320     spapr_machine_2_8_instance_options(machine);
4321     spapr->use_hotplug_event_source = false;
4322 }
4323 
4324 static void spapr_machine_2_7_class_options(MachineClass *mc)
4325 {
4326     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4327 
4328     spapr_machine_2_8_class_options(mc);
4329     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4330     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4331     smc->phb_placement = phb_placement_2_7;
4332 }
4333 
4334 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4335 
4336 /*
4337  * pseries-2.6
4338  */
4339 #define SPAPR_COMPAT_2_6 \
4340     HW_COMPAT_2_6 \
4341     { \
4342         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4343         .property = "ddw",\
4344         .value    = stringify(off),\
4345     },
4346 
4347 static void spapr_machine_2_6_instance_options(MachineState *machine)
4348 {
4349     spapr_machine_2_7_instance_options(machine);
4350 }
4351 
4352 static void spapr_machine_2_6_class_options(MachineClass *mc)
4353 {
4354     spapr_machine_2_7_class_options(mc);
4355     mc->has_hotpluggable_cpus = false;
4356     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4357 }
4358 
4359 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4360 
4361 /*
4362  * pseries-2.5
4363  */
4364 #define SPAPR_COMPAT_2_5 \
4365     HW_COMPAT_2_5 \
4366     { \
4367         .driver   = "spapr-vlan", \
4368         .property = "use-rx-buffer-pools", \
4369         .value    = "off", \
4370     },
4371 
4372 static void spapr_machine_2_5_instance_options(MachineState *machine)
4373 {
4374     spapr_machine_2_6_instance_options(machine);
4375 }
4376 
4377 static void spapr_machine_2_5_class_options(MachineClass *mc)
4378 {
4379     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4380 
4381     spapr_machine_2_6_class_options(mc);
4382     smc->use_ohci_by_default = true;
4383     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4384 }
4385 
4386 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4387 
4388 /*
4389  * pseries-2.4
4390  */
4391 #define SPAPR_COMPAT_2_4 \
4392         HW_COMPAT_2_4
4393 
4394 static void spapr_machine_2_4_instance_options(MachineState *machine)
4395 {
4396     spapr_machine_2_5_instance_options(machine);
4397 }
4398 
4399 static void spapr_machine_2_4_class_options(MachineClass *mc)
4400 {
4401     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4402 
4403     spapr_machine_2_5_class_options(mc);
4404     smc->dr_lmb_enabled = false;
4405     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4406 }
4407 
4408 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4409 
4410 /*
4411  * pseries-2.3
4412  */
4413 #define SPAPR_COMPAT_2_3 \
4414         HW_COMPAT_2_3 \
4415         {\
4416             .driver   = "spapr-pci-host-bridge",\
4417             .property = "dynamic-reconfiguration",\
4418             .value    = "off",\
4419         },
4420 
4421 static void spapr_machine_2_3_instance_options(MachineState *machine)
4422 {
4423     spapr_machine_2_4_instance_options(machine);
4424 }
4425 
4426 static void spapr_machine_2_3_class_options(MachineClass *mc)
4427 {
4428     spapr_machine_2_4_class_options(mc);
4429     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4430 }
4431 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4432 
4433 /*
4434  * pseries-2.2
4435  */
4436 
4437 #define SPAPR_COMPAT_2_2 \
4438         HW_COMPAT_2_2 \
4439         {\
4440             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4441             .property = "mem_win_size",\
4442             .value    = "0x20000000",\
4443         },
4444 
4445 static void spapr_machine_2_2_instance_options(MachineState *machine)
4446 {
4447     spapr_machine_2_3_instance_options(machine);
4448     machine->suppress_vmdesc = true;
4449 }
4450 
4451 static void spapr_machine_2_2_class_options(MachineClass *mc)
4452 {
4453     spapr_machine_2_3_class_options(mc);
4454     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4455 }
4456 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4457 
4458 /*
4459  * pseries-2.1
4460  */
4461 #define SPAPR_COMPAT_2_1 \
4462         HW_COMPAT_2_1
4463 
4464 static void spapr_machine_2_1_instance_options(MachineState *machine)
4465 {
4466     spapr_machine_2_2_instance_options(machine);
4467 }
4468 
4469 static void spapr_machine_2_1_class_options(MachineClass *mc)
4470 {
4471     spapr_machine_2_2_class_options(mc);
4472     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4473 }
4474 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4475 
4476 static void spapr_machine_register_types(void)
4477 {
4478     type_register_static(&spapr_machine_info);
4479 }
4480 
4481 type_init(spapr_machine_register_types)
4482