xref: /openbmc/qemu/hw/ppc/spapr.c (revision 3ae8a100)
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     Error *local_err = NULL;
190 
191     if (kvm_enabled()) {
192         if (machine_kernel_irqchip_allowed(machine) &&
193             !xics_kvm_init(spapr, &local_err)) {
194             spapr->icp_type = TYPE_KVM_ICP;
195             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs,
196                                           &local_err);
197         }
198         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
199             error_prepend(&local_err,
200                           "kernel_irqchip requested but unavailable: ");
201             goto error;
202         }
203         error_free(local_err);
204         local_err = NULL;
205     }
206 
207     if (!spapr->ics) {
208         xics_spapr_init(spapr);
209         spapr->icp_type = TYPE_ICP;
210         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs,
211                                       &local_err);
212     }
213 
214 error:
215     error_propagate(errp, local_err);
216 }
217 
218 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
219                                   int smt_threads)
220 {
221     int i, ret = 0;
222     uint32_t servers_prop[smt_threads];
223     uint32_t gservers_prop[smt_threads * 2];
224     int index = spapr_get_vcpu_id(cpu);
225 
226     if (cpu->compat_pvr) {
227         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
228         if (ret < 0) {
229             return ret;
230         }
231     }
232 
233     /* Build interrupt servers and gservers properties */
234     for (i = 0; i < smt_threads; i++) {
235         servers_prop[i] = cpu_to_be32(index + i);
236         /* Hack, direct the group queues back to cpu 0 */
237         gservers_prop[i*2] = cpu_to_be32(index + i);
238         gservers_prop[i*2 + 1] = 0;
239     }
240     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
241                       servers_prop, sizeof(servers_prop));
242     if (ret < 0) {
243         return ret;
244     }
245     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
246                       gservers_prop, sizeof(gservers_prop));
247 
248     return ret;
249 }
250 
251 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
252 {
253     int index = spapr_get_vcpu_id(cpu);
254     uint32_t associativity[] = {cpu_to_be32(0x5),
255                                 cpu_to_be32(0x0),
256                                 cpu_to_be32(0x0),
257                                 cpu_to_be32(0x0),
258                                 cpu_to_be32(cpu->node_id),
259                                 cpu_to_be32(index)};
260 
261     /* Advertise NUMA via ibm,associativity */
262     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
263                           sizeof(associativity));
264 }
265 
266 /* Populate the "ibm,pa-features" property */
267 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
268                                        PowerPCCPU *cpu,
269                                        void *fdt, int offset,
270                                        bool legacy_guest)
271 {
272     uint8_t pa_features_206[] = { 6, 0,
273         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
274     uint8_t pa_features_207[] = { 24, 0,
275         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
276         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
277         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
278         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
279     uint8_t pa_features_300[] = { 66, 0,
280         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
281         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
282         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
283         /* 6: DS207 */
284         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
285         /* 16: Vector */
286         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
287         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
288         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
289         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
290         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
291         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
292         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
293         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
294         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
295         /* 42: PM, 44: PC RA, 46: SC vec'd */
296         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
297         /* 48: SIMD, 50: QP BFP, 52: String */
298         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
299         /* 54: DecFP, 56: DecI, 58: SHA */
300         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
301         /* 60: NM atomic, 62: RNG */
302         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
303     };
304     uint8_t *pa_features = NULL;
305     size_t pa_size;
306 
307     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
308         pa_features = pa_features_206;
309         pa_size = sizeof(pa_features_206);
310     }
311     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
312         pa_features = pa_features_207;
313         pa_size = sizeof(pa_features_207);
314     }
315     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
316         pa_features = pa_features_300;
317         pa_size = sizeof(pa_features_300);
318     }
319     if (!pa_features) {
320         return;
321     }
322 
323     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
324         /*
325          * Note: we keep CI large pages off by default because a 64K capable
326          * guest provisioned with large pages might otherwise try to map a qemu
327          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
328          * even if that qemu runs on a 4k host.
329          * We dd this bit back here if we are confident this is not an issue
330          */
331         pa_features[3] |= 0x20;
332     }
333     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
334         pa_features[24] |= 0x80;    /* Transactional memory support */
335     }
336     if (legacy_guest && pa_size > 40) {
337         /* Workaround for broken kernels that attempt (guest) radix
338          * mode when they can't handle it, if they see the radix bit set
339          * in pa-features. So hide it from them. */
340         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
341     }
342 
343     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
344 }
345 
346 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
347 {
348     int ret = 0, offset, cpus_offset;
349     CPUState *cs;
350     char cpu_model[32];
351     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
352 
353     CPU_FOREACH(cs) {
354         PowerPCCPU *cpu = POWERPC_CPU(cs);
355         DeviceClass *dc = DEVICE_GET_CLASS(cs);
356         int index = spapr_get_vcpu_id(cpu);
357         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
358 
359         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
360             continue;
361         }
362 
363         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
364 
365         cpus_offset = fdt_path_offset(fdt, "/cpus");
366         if (cpus_offset < 0) {
367             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
368             if (cpus_offset < 0) {
369                 return cpus_offset;
370             }
371         }
372         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
373         if (offset < 0) {
374             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
375             if (offset < 0) {
376                 return offset;
377             }
378         }
379 
380         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
381                           pft_size_prop, sizeof(pft_size_prop));
382         if (ret < 0) {
383             return ret;
384         }
385 
386         if (nb_numa_nodes > 1) {
387             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
388             if (ret < 0) {
389                 return ret;
390             }
391         }
392 
393         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
394         if (ret < 0) {
395             return ret;
396         }
397 
398         spapr_populate_pa_features(spapr, cpu, fdt, offset,
399                                    spapr->cas_legacy_guest_workaround);
400     }
401     return ret;
402 }
403 
404 static hwaddr spapr_node0_size(MachineState *machine)
405 {
406     if (nb_numa_nodes) {
407         int i;
408         for (i = 0; i < nb_numa_nodes; ++i) {
409             if (numa_info[i].node_mem) {
410                 return MIN(pow2floor(numa_info[i].node_mem),
411                            machine->ram_size);
412             }
413         }
414     }
415     return machine->ram_size;
416 }
417 
418 static void add_str(GString *s, const gchar *s1)
419 {
420     g_string_append_len(s, s1, strlen(s1) + 1);
421 }
422 
423 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
424                                        hwaddr size)
425 {
426     uint32_t associativity[] = {
427         cpu_to_be32(0x4), /* length */
428         cpu_to_be32(0x0), cpu_to_be32(0x0),
429         cpu_to_be32(0x0), cpu_to_be32(nodeid)
430     };
431     char mem_name[32];
432     uint64_t mem_reg_property[2];
433     int off;
434 
435     mem_reg_property[0] = cpu_to_be64(start);
436     mem_reg_property[1] = cpu_to_be64(size);
437 
438     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
439     off = fdt_add_subnode(fdt, 0, mem_name);
440     _FDT(off);
441     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
442     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
443                       sizeof(mem_reg_property))));
444     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
445                       sizeof(associativity))));
446     return off;
447 }
448 
449 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
450 {
451     MachineState *machine = MACHINE(spapr);
452     hwaddr mem_start, node_size;
453     int i, nb_nodes = nb_numa_nodes;
454     NodeInfo *nodes = numa_info;
455     NodeInfo ramnode;
456 
457     /* No NUMA nodes, assume there is just one node with whole RAM */
458     if (!nb_numa_nodes) {
459         nb_nodes = 1;
460         ramnode.node_mem = machine->ram_size;
461         nodes = &ramnode;
462     }
463 
464     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
465         if (!nodes[i].node_mem) {
466             continue;
467         }
468         if (mem_start >= machine->ram_size) {
469             node_size = 0;
470         } else {
471             node_size = nodes[i].node_mem;
472             if (node_size > machine->ram_size - mem_start) {
473                 node_size = machine->ram_size - mem_start;
474             }
475         }
476         if (!mem_start) {
477             /* spapr_machine_init() checks for rma_size <= node0_size
478              * already */
479             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
480             mem_start += spapr->rma_size;
481             node_size -= spapr->rma_size;
482         }
483         for ( ; node_size; ) {
484             hwaddr sizetmp = pow2floor(node_size);
485 
486             /* mem_start != 0 here */
487             if (ctzl(mem_start) < ctzl(sizetmp)) {
488                 sizetmp = 1ULL << ctzl(mem_start);
489             }
490 
491             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
492             node_size -= sizetmp;
493             mem_start += sizetmp;
494         }
495     }
496 
497     return 0;
498 }
499 
500 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
501                                   sPAPRMachineState *spapr)
502 {
503     PowerPCCPU *cpu = POWERPC_CPU(cs);
504     CPUPPCState *env = &cpu->env;
505     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
506     int index = spapr_get_vcpu_id(cpu);
507     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
508                        0xffffffff, 0xffffffff};
509     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
510         : SPAPR_TIMEBASE_FREQ;
511     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
512     uint32_t page_sizes_prop[64];
513     size_t page_sizes_prop_size;
514     uint32_t vcpus_per_socket = smp_threads * smp_cores;
515     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
516     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
517     sPAPRDRConnector *drc;
518     int drc_index;
519     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
520     int i;
521 
522     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
523     if (drc) {
524         drc_index = spapr_drc_index(drc);
525         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
526     }
527 
528     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
529     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
530 
531     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
532     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
533                            env->dcache_line_size)));
534     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
535                            env->dcache_line_size)));
536     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
537                            env->icache_line_size)));
538     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
539                            env->icache_line_size)));
540 
541     if (pcc->l1_dcache_size) {
542         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
543                                pcc->l1_dcache_size)));
544     } else {
545         warn_report("Unknown L1 dcache size for cpu");
546     }
547     if (pcc->l1_icache_size) {
548         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
549                                pcc->l1_icache_size)));
550     } else {
551         warn_report("Unknown L1 icache size for cpu");
552     }
553 
554     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
555     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
556     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
557     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
558     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
559     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
560 
561     if (env->spr_cb[SPR_PURR].oea_read) {
562         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
563     }
564 
565     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
566         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
567                           segs, sizeof(segs))));
568     }
569 
570     /* Advertise VSX (vector extensions) if available
571      *   1               == VMX / Altivec available
572      *   2               == VSX available
573      *
574      * Only CPUs for which we create core types in spapr_cpu_core.c
575      * are possible, and all of those have VMX */
576     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
577         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
578     } else {
579         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
580     }
581 
582     /* Advertise DFP (Decimal Floating Point) if available
583      *   0 / no property == no DFP
584      *   1               == DFP available */
585     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
586         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
587     }
588 
589     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
590                                                       sizeof(page_sizes_prop));
591     if (page_sizes_prop_size) {
592         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
593                           page_sizes_prop, page_sizes_prop_size)));
594     }
595 
596     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
597 
598     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
599                            cs->cpu_index / vcpus_per_socket)));
600 
601     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
602                       pft_size_prop, sizeof(pft_size_prop))));
603 
604     if (nb_numa_nodes > 1) {
605         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
606     }
607 
608     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
609 
610     if (pcc->radix_page_info) {
611         for (i = 0; i < pcc->radix_page_info->count; i++) {
612             radix_AP_encodings[i] =
613                 cpu_to_be32(pcc->radix_page_info->entries[i]);
614         }
615         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
616                           radix_AP_encodings,
617                           pcc->radix_page_info->count *
618                           sizeof(radix_AP_encodings[0]))));
619     }
620 }
621 
622 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
623 {
624     CPUState *cs;
625     int cpus_offset;
626     char *nodename;
627 
628     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
629     _FDT(cpus_offset);
630     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
631     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
632 
633     /*
634      * We walk the CPUs in reverse order to ensure that CPU DT nodes
635      * created by fdt_add_subnode() end up in the right order in FDT
636      * for the guest kernel the enumerate the CPUs correctly.
637      */
638     CPU_FOREACH_REVERSE(cs) {
639         PowerPCCPU *cpu = POWERPC_CPU(cs);
640         int index = spapr_get_vcpu_id(cpu);
641         DeviceClass *dc = DEVICE_GET_CLASS(cs);
642         int offset;
643 
644         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
645             continue;
646         }
647 
648         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
649         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
650         g_free(nodename);
651         _FDT(offset);
652         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
653     }
654 
655 }
656 
657 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
658 {
659     MemoryDeviceInfoList *info;
660 
661     for (info = list; info; info = info->next) {
662         MemoryDeviceInfo *value = info->value;
663 
664         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
665             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
666 
667             if (pcdimm_info->addr >= addr &&
668                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
669                 return pcdimm_info->node;
670             }
671         }
672     }
673 
674     return -1;
675 }
676 
677 struct sPAPRDrconfCellV2 {
678      uint32_t seq_lmbs;
679      uint64_t base_addr;
680      uint32_t drc_index;
681      uint32_t aa_index;
682      uint32_t flags;
683 } QEMU_PACKED;
684 
685 typedef struct DrconfCellQueue {
686     struct sPAPRDrconfCellV2 cell;
687     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
688 } DrconfCellQueue;
689 
690 static DrconfCellQueue *
691 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
692                       uint32_t drc_index, uint32_t aa_index,
693                       uint32_t flags)
694 {
695     DrconfCellQueue *elem;
696 
697     elem = g_malloc0(sizeof(*elem));
698     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
699     elem->cell.base_addr = cpu_to_be64(base_addr);
700     elem->cell.drc_index = cpu_to_be32(drc_index);
701     elem->cell.aa_index = cpu_to_be32(aa_index);
702     elem->cell.flags = cpu_to_be32(flags);
703 
704     return elem;
705 }
706 
707 /* ibm,dynamic-memory-v2 */
708 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
709                                    int offset, MemoryDeviceInfoList *dimms)
710 {
711     MachineState *machine = MACHINE(spapr);
712     uint8_t *int_buf, *cur_index, buf_len;
713     int ret;
714     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
715     uint64_t addr, cur_addr, size;
716     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
717     uint64_t mem_end = machine->device_memory->base +
718                        memory_region_size(&machine->device_memory->mr);
719     uint32_t node, nr_entries = 0;
720     sPAPRDRConnector *drc;
721     DrconfCellQueue *elem, *next;
722     MemoryDeviceInfoList *info;
723     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
724         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
725 
726     /* Entry to cover RAM and the gap area */
727     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
728                                  SPAPR_LMB_FLAGS_RESERVED |
729                                  SPAPR_LMB_FLAGS_DRC_INVALID);
730     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
731     nr_entries++;
732 
733     cur_addr = machine->device_memory->base;
734     for (info = dimms; info; info = info->next) {
735         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
736 
737         addr = di->addr;
738         size = di->size;
739         node = di->node;
740 
741         /* Entry for hot-pluggable area */
742         if (cur_addr < addr) {
743             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
744             g_assert(drc);
745             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
746                                          cur_addr, spapr_drc_index(drc), -1, 0);
747             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
748             nr_entries++;
749         }
750 
751         /* Entry for DIMM */
752         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
753         g_assert(drc);
754         elem = spapr_get_drconf_cell(size / lmb_size, addr,
755                                      spapr_drc_index(drc), node,
756                                      SPAPR_LMB_FLAGS_ASSIGNED);
757         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
758         nr_entries++;
759         cur_addr = addr + size;
760     }
761 
762     /* Entry for remaining hotpluggable area */
763     if (cur_addr < mem_end) {
764         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
765         g_assert(drc);
766         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
767                                      cur_addr, spapr_drc_index(drc), -1, 0);
768         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
769         nr_entries++;
770     }
771 
772     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
773     int_buf = cur_index = g_malloc0(buf_len);
774     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
775     cur_index += sizeof(nr_entries);
776 
777     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
778         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
779         cur_index += sizeof(elem->cell);
780         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
781         g_free(elem);
782     }
783 
784     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
785     g_free(int_buf);
786     if (ret < 0) {
787         return -1;
788     }
789     return 0;
790 }
791 
792 /* ibm,dynamic-memory */
793 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
794                                    int offset, MemoryDeviceInfoList *dimms)
795 {
796     MachineState *machine = MACHINE(spapr);
797     int i, ret;
798     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
799     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
800     uint32_t nr_lmbs = (machine->device_memory->base +
801                        memory_region_size(&machine->device_memory->mr)) /
802                        lmb_size;
803     uint32_t *int_buf, *cur_index, buf_len;
804 
805     /*
806      * Allocate enough buffer size to fit in ibm,dynamic-memory
807      */
808     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
809     cur_index = int_buf = g_malloc0(buf_len);
810     int_buf[0] = cpu_to_be32(nr_lmbs);
811     cur_index++;
812     for (i = 0; i < nr_lmbs; i++) {
813         uint64_t addr = i * lmb_size;
814         uint32_t *dynamic_memory = cur_index;
815 
816         if (i >= device_lmb_start) {
817             sPAPRDRConnector *drc;
818 
819             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
820             g_assert(drc);
821 
822             dynamic_memory[0] = cpu_to_be32(addr >> 32);
823             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
824             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
825             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
826             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
827             if (memory_region_present(get_system_memory(), addr)) {
828                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
829             } else {
830                 dynamic_memory[5] = cpu_to_be32(0);
831             }
832         } else {
833             /*
834              * LMB information for RMA, boot time RAM and gap b/n RAM and
835              * device memory region -- all these are marked as reserved
836              * and as having no valid DRC.
837              */
838             dynamic_memory[0] = cpu_to_be32(addr >> 32);
839             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
840             dynamic_memory[2] = cpu_to_be32(0);
841             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
842             dynamic_memory[4] = cpu_to_be32(-1);
843             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
844                                             SPAPR_LMB_FLAGS_DRC_INVALID);
845         }
846 
847         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
848     }
849     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
850     g_free(int_buf);
851     if (ret < 0) {
852         return -1;
853     }
854     return 0;
855 }
856 
857 /*
858  * Adds ibm,dynamic-reconfiguration-memory node.
859  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
860  * of this device tree node.
861  */
862 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
863 {
864     MachineState *machine = MACHINE(spapr);
865     int ret, i, offset;
866     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
867     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
868     uint32_t *int_buf, *cur_index, buf_len;
869     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
870     MemoryDeviceInfoList *dimms = NULL;
871 
872     /*
873      * Don't create the node if there is no device memory
874      */
875     if (machine->ram_size == machine->maxram_size) {
876         return 0;
877     }
878 
879     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
880 
881     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
882                     sizeof(prop_lmb_size));
883     if (ret < 0) {
884         return ret;
885     }
886 
887     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
888     if (ret < 0) {
889         return ret;
890     }
891 
892     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
893     if (ret < 0) {
894         return ret;
895     }
896 
897     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
898     dimms = qmp_memory_device_list();
899     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
900         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
901     } else {
902         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
903     }
904     qapi_free_MemoryDeviceInfoList(dimms);
905 
906     if (ret < 0) {
907         return ret;
908     }
909 
910     /* ibm,associativity-lookup-arrays */
911     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
912     cur_index = int_buf = g_malloc0(buf_len);
913 
914     cur_index = int_buf;
915     int_buf[0] = cpu_to_be32(nr_nodes);
916     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
917     cur_index += 2;
918     for (i = 0; i < nr_nodes; i++) {
919         uint32_t associativity[] = {
920             cpu_to_be32(0x0),
921             cpu_to_be32(0x0),
922             cpu_to_be32(0x0),
923             cpu_to_be32(i)
924         };
925         memcpy(cur_index, associativity, sizeof(associativity));
926         cur_index += 4;
927     }
928     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
929             (cur_index - int_buf) * sizeof(uint32_t));
930     g_free(int_buf);
931 
932     return ret;
933 }
934 
935 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
936                                 sPAPROptionVector *ov5_updates)
937 {
938     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
939     int ret = 0, offset;
940 
941     /* Generate ibm,dynamic-reconfiguration-memory node if required */
942     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
943         g_assert(smc->dr_lmb_enabled);
944         ret = spapr_populate_drconf_memory(spapr, fdt);
945         if (ret) {
946             goto out;
947         }
948     }
949 
950     offset = fdt_path_offset(fdt, "/chosen");
951     if (offset < 0) {
952         offset = fdt_add_subnode(fdt, 0, "chosen");
953         if (offset < 0) {
954             return offset;
955         }
956     }
957     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
958                                  "ibm,architecture-vec-5");
959 
960 out:
961     return ret;
962 }
963 
964 static bool spapr_hotplugged_dev_before_cas(void)
965 {
966     Object *drc_container, *obj;
967     ObjectProperty *prop;
968     ObjectPropertyIterator iter;
969 
970     drc_container = container_get(object_get_root(), "/dr-connector");
971     object_property_iter_init(&iter, drc_container);
972     while ((prop = object_property_iter_next(&iter))) {
973         if (!strstart(prop->type, "link<", NULL)) {
974             continue;
975         }
976         obj = object_property_get_link(drc_container, prop->name, NULL);
977         if (spapr_drc_needed(obj)) {
978             return true;
979         }
980     }
981     return false;
982 }
983 
984 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
985                                  target_ulong addr, target_ulong size,
986                                  sPAPROptionVector *ov5_updates)
987 {
988     void *fdt, *fdt_skel;
989     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
990 
991     if (spapr_hotplugged_dev_before_cas()) {
992         return 1;
993     }
994 
995     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
996         error_report("SLOF provided an unexpected CAS buffer size "
997                      TARGET_FMT_lu " (min: %zu, max: %u)",
998                      size, sizeof(hdr), FW_MAX_SIZE);
999         exit(EXIT_FAILURE);
1000     }
1001 
1002     size -= sizeof(hdr);
1003 
1004     /* Create skeleton */
1005     fdt_skel = g_malloc0(size);
1006     _FDT((fdt_create(fdt_skel, size)));
1007     _FDT((fdt_finish_reservemap(fdt_skel)));
1008     _FDT((fdt_begin_node(fdt_skel, "")));
1009     _FDT((fdt_end_node(fdt_skel)));
1010     _FDT((fdt_finish(fdt_skel)));
1011     fdt = g_malloc0(size);
1012     _FDT((fdt_open_into(fdt_skel, fdt, size)));
1013     g_free(fdt_skel);
1014 
1015     /* Fixup cpu nodes */
1016     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
1017 
1018     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
1019         return -1;
1020     }
1021 
1022     /* Pack resulting tree */
1023     _FDT((fdt_pack(fdt)));
1024 
1025     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1026         trace_spapr_cas_failed(size);
1027         return -1;
1028     }
1029 
1030     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1031     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1032     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1033     g_free(fdt);
1034 
1035     return 0;
1036 }
1037 
1038 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1039 {
1040     int rtas;
1041     GString *hypertas = g_string_sized_new(256);
1042     GString *qemu_hypertas = g_string_sized_new(256);
1043     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1044     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1045         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1046     uint32_t lrdr_capacity[] = {
1047         cpu_to_be32(max_device_addr >> 32),
1048         cpu_to_be32(max_device_addr & 0xffffffff),
1049         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1050         cpu_to_be32(max_cpus / smp_threads),
1051     };
1052     uint32_t maxdomains[] = {
1053         cpu_to_be32(4),
1054         cpu_to_be32(0),
1055         cpu_to_be32(0),
1056         cpu_to_be32(0),
1057         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
1058     };
1059 
1060     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1061 
1062     /* hypertas */
1063     add_str(hypertas, "hcall-pft");
1064     add_str(hypertas, "hcall-term");
1065     add_str(hypertas, "hcall-dabr");
1066     add_str(hypertas, "hcall-interrupt");
1067     add_str(hypertas, "hcall-tce");
1068     add_str(hypertas, "hcall-vio");
1069     add_str(hypertas, "hcall-splpar");
1070     add_str(hypertas, "hcall-bulk");
1071     add_str(hypertas, "hcall-set-mode");
1072     add_str(hypertas, "hcall-sprg0");
1073     add_str(hypertas, "hcall-copy");
1074     add_str(hypertas, "hcall-debug");
1075     add_str(qemu_hypertas, "hcall-memop1");
1076 
1077     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1078         add_str(hypertas, "hcall-multi-tce");
1079     }
1080 
1081     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1082         add_str(hypertas, "hcall-hpt-resize");
1083     }
1084 
1085     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1086                      hypertas->str, hypertas->len));
1087     g_string_free(hypertas, TRUE);
1088     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1089                      qemu_hypertas->str, qemu_hypertas->len));
1090     g_string_free(qemu_hypertas, TRUE);
1091 
1092     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1093                      refpoints, sizeof(refpoints)));
1094 
1095     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1096                      maxdomains, sizeof(maxdomains)));
1097 
1098     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1099                           RTAS_ERROR_LOG_MAX));
1100     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1101                           RTAS_EVENT_SCAN_RATE));
1102 
1103     g_assert(msi_nonbroken);
1104     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1105 
1106     /*
1107      * According to PAPR, rtas ibm,os-term does not guarantee a return
1108      * back to the guest cpu.
1109      *
1110      * While an additional ibm,extended-os-term property indicates
1111      * that rtas call return will always occur. Set this property.
1112      */
1113     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1114 
1115     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1116                      lrdr_capacity, sizeof(lrdr_capacity)));
1117 
1118     spapr_dt_rtas_tokens(fdt, rtas);
1119 }
1120 
1121 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1122  * that the guest may request and thus the valid values for bytes 24..26 of
1123  * option vector 5: */
1124 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1125 {
1126     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1127 
1128     char val[2 * 4] = {
1129         23, 0x00, /* Xive mode, filled in below. */
1130         24, 0x00, /* Hash/Radix, filled in below. */
1131         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1132         26, 0x40, /* Radix options: GTSE == yes. */
1133     };
1134 
1135     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1136                           first_ppc_cpu->compat_pvr)) {
1137         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1138         val[3] = 0x00; /* Hash */
1139     } else if (kvm_enabled()) {
1140         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1141             val[3] = 0x80; /* OV5_MMU_BOTH */
1142         } else if (kvmppc_has_cap_mmu_radix()) {
1143             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1144         } else {
1145             val[3] = 0x00; /* Hash */
1146         }
1147     } else {
1148         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1149         val[3] = 0xC0;
1150     }
1151     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1152                      val, sizeof(val)));
1153 }
1154 
1155 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1156 {
1157     MachineState *machine = MACHINE(spapr);
1158     int chosen;
1159     const char *boot_device = machine->boot_order;
1160     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1161     size_t cb = 0;
1162     char *bootlist = get_boot_devices_list(&cb, true);
1163 
1164     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1165 
1166     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1167     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1168                           spapr->initrd_base));
1169     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1170                           spapr->initrd_base + spapr->initrd_size));
1171 
1172     if (spapr->kernel_size) {
1173         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1174                               cpu_to_be64(spapr->kernel_size) };
1175 
1176         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1177                          &kprop, sizeof(kprop)));
1178         if (spapr->kernel_le) {
1179             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1180         }
1181     }
1182     if (boot_menu) {
1183         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1184     }
1185     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1186     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1187     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1188 
1189     if (cb && bootlist) {
1190         int i;
1191 
1192         for (i = 0; i < cb; i++) {
1193             if (bootlist[i] == '\n') {
1194                 bootlist[i] = ' ';
1195             }
1196         }
1197         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1198     }
1199 
1200     if (boot_device && strlen(boot_device)) {
1201         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1202     }
1203 
1204     if (!spapr->has_graphics && stdout_path) {
1205         /*
1206          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1207          * kernel. New platforms should only use the "stdout-path" property. Set
1208          * the new property and continue using older property to remain
1209          * compatible with the existing firmware.
1210          */
1211         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1212         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1213     }
1214 
1215     spapr_dt_ov5_platform_support(fdt, chosen);
1216 
1217     g_free(stdout_path);
1218     g_free(bootlist);
1219 }
1220 
1221 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1222 {
1223     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1224      * KVM to work under pHyp with some guest co-operation */
1225     int hypervisor;
1226     uint8_t hypercall[16];
1227 
1228     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1229     /* indicate KVM hypercall interface */
1230     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1231     if (kvmppc_has_cap_fixup_hcalls()) {
1232         /*
1233          * Older KVM versions with older guest kernels were broken
1234          * with the magic page, don't allow the guest to map it.
1235          */
1236         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1237                                   sizeof(hypercall))) {
1238             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1239                              hypercall, sizeof(hypercall)));
1240         }
1241     }
1242 }
1243 
1244 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1245                              hwaddr rtas_addr,
1246                              hwaddr rtas_size)
1247 {
1248     MachineState *machine = MACHINE(spapr);
1249     MachineClass *mc = MACHINE_GET_CLASS(machine);
1250     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1251     int ret;
1252     void *fdt;
1253     sPAPRPHBState *phb;
1254     char *buf;
1255 
1256     fdt = g_malloc0(FDT_MAX_SIZE);
1257     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1258 
1259     /* Root node */
1260     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1261     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1262     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1263 
1264     /*
1265      * Add info to guest to indentify which host is it being run on
1266      * and what is the uuid of the guest
1267      */
1268     if (kvmppc_get_host_model(&buf)) {
1269         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1270         g_free(buf);
1271     }
1272     if (kvmppc_get_host_serial(&buf)) {
1273         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1274         g_free(buf);
1275     }
1276 
1277     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1278 
1279     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1280     if (qemu_uuid_set) {
1281         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1282     }
1283     g_free(buf);
1284 
1285     if (qemu_get_vm_name()) {
1286         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1287                                 qemu_get_vm_name()));
1288     }
1289 
1290     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1291     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1292 
1293     /* /interrupt controller */
1294     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1295 
1296     ret = spapr_populate_memory(spapr, fdt);
1297     if (ret < 0) {
1298         error_report("couldn't setup memory nodes in fdt");
1299         exit(1);
1300     }
1301 
1302     /* /vdevice */
1303     spapr_dt_vdevice(spapr->vio_bus, fdt);
1304 
1305     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1306         ret = spapr_rng_populate_dt(fdt);
1307         if (ret < 0) {
1308             error_report("could not set up rng device in the fdt");
1309             exit(1);
1310         }
1311     }
1312 
1313     QLIST_FOREACH(phb, &spapr->phbs, list) {
1314         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1315         if (ret < 0) {
1316             error_report("couldn't setup PCI devices in fdt");
1317             exit(1);
1318         }
1319     }
1320 
1321     /* cpus */
1322     spapr_populate_cpus_dt_node(fdt, spapr);
1323 
1324     if (smc->dr_lmb_enabled) {
1325         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1326     }
1327 
1328     if (mc->has_hotpluggable_cpus) {
1329         int offset = fdt_path_offset(fdt, "/cpus");
1330         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1331                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1332         if (ret < 0) {
1333             error_report("Couldn't set up CPU DR device tree properties");
1334             exit(1);
1335         }
1336     }
1337 
1338     /* /event-sources */
1339     spapr_dt_events(spapr, fdt);
1340 
1341     /* /rtas */
1342     spapr_dt_rtas(spapr, fdt);
1343 
1344     /* /chosen */
1345     spapr_dt_chosen(spapr, fdt);
1346 
1347     /* /hypervisor */
1348     if (kvm_enabled()) {
1349         spapr_dt_hypervisor(spapr, fdt);
1350     }
1351 
1352     /* Build memory reserve map */
1353     if (spapr->kernel_size) {
1354         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1355     }
1356     if (spapr->initrd_size) {
1357         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1358     }
1359 
1360     /* ibm,client-architecture-support updates */
1361     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1362     if (ret < 0) {
1363         error_report("couldn't setup CAS properties fdt");
1364         exit(1);
1365     }
1366 
1367     return fdt;
1368 }
1369 
1370 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1371 {
1372     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1373 }
1374 
1375 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1376                                     PowerPCCPU *cpu)
1377 {
1378     CPUPPCState *env = &cpu->env;
1379 
1380     /* The TCG path should also be holding the BQL at this point */
1381     g_assert(qemu_mutex_iothread_locked());
1382 
1383     if (msr_pr) {
1384         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1385         env->gpr[3] = H_PRIVILEGE;
1386     } else {
1387         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1388     }
1389 }
1390 
1391 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1392 {
1393     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1394 
1395     return spapr->patb_entry;
1396 }
1397 
1398 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1399 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1400 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1401 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1402 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1403 
1404 /*
1405  * Get the fd to access the kernel htab, re-opening it if necessary
1406  */
1407 static int get_htab_fd(sPAPRMachineState *spapr)
1408 {
1409     Error *local_err = NULL;
1410 
1411     if (spapr->htab_fd >= 0) {
1412         return spapr->htab_fd;
1413     }
1414 
1415     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1416     if (spapr->htab_fd < 0) {
1417         error_report_err(local_err);
1418     }
1419 
1420     return spapr->htab_fd;
1421 }
1422 
1423 void close_htab_fd(sPAPRMachineState *spapr)
1424 {
1425     if (spapr->htab_fd >= 0) {
1426         close(spapr->htab_fd);
1427     }
1428     spapr->htab_fd = -1;
1429 }
1430 
1431 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1432 {
1433     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1434 
1435     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1436 }
1437 
1438 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1439 {
1440     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1441 
1442     assert(kvm_enabled());
1443 
1444     if (!spapr->htab) {
1445         return 0;
1446     }
1447 
1448     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1449 }
1450 
1451 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1452                                                 hwaddr ptex, int n)
1453 {
1454     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1455     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1456 
1457     if (!spapr->htab) {
1458         /*
1459          * HTAB is controlled by KVM. Fetch into temporary buffer
1460          */
1461         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1462         kvmppc_read_hptes(hptes, ptex, n);
1463         return hptes;
1464     }
1465 
1466     /*
1467      * HTAB is controlled by QEMU. Just point to the internally
1468      * accessible PTEG.
1469      */
1470     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1471 }
1472 
1473 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1474                               const ppc_hash_pte64_t *hptes,
1475                               hwaddr ptex, int n)
1476 {
1477     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1478 
1479     if (!spapr->htab) {
1480         g_free((void *)hptes);
1481     }
1482 
1483     /* Nothing to do for qemu managed HPT */
1484 }
1485 
1486 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1487                              uint64_t pte0, uint64_t pte1)
1488 {
1489     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1490     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1491 
1492     if (!spapr->htab) {
1493         kvmppc_write_hpte(ptex, pte0, pte1);
1494     } else {
1495         stq_p(spapr->htab + offset, pte0);
1496         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1497     }
1498 }
1499 
1500 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1501 {
1502     int shift;
1503 
1504     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1505      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1506      * that's much more than is needed for Linux guests */
1507     shift = ctz64(pow2ceil(ramsize)) - 7;
1508     shift = MAX(shift, 18); /* Minimum architected size */
1509     shift = MIN(shift, 46); /* Maximum architected size */
1510     return shift;
1511 }
1512 
1513 void spapr_free_hpt(sPAPRMachineState *spapr)
1514 {
1515     g_free(spapr->htab);
1516     spapr->htab = NULL;
1517     spapr->htab_shift = 0;
1518     close_htab_fd(spapr);
1519 }
1520 
1521 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1522                           Error **errp)
1523 {
1524     long rc;
1525 
1526     /* Clean up any HPT info from a previous boot */
1527     spapr_free_hpt(spapr);
1528 
1529     rc = kvmppc_reset_htab(shift);
1530     if (rc < 0) {
1531         /* kernel-side HPT needed, but couldn't allocate one */
1532         error_setg_errno(errp, errno,
1533                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1534                          shift);
1535         /* This is almost certainly fatal, but if the caller really
1536          * wants to carry on with shift == 0, it's welcome to try */
1537     } else if (rc > 0) {
1538         /* kernel-side HPT allocated */
1539         if (rc != shift) {
1540             error_setg(errp,
1541                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1542                        shift, rc);
1543         }
1544 
1545         spapr->htab_shift = shift;
1546         spapr->htab = NULL;
1547     } else {
1548         /* kernel-side HPT not needed, allocate in userspace instead */
1549         size_t size = 1ULL << shift;
1550         int i;
1551 
1552         spapr->htab = qemu_memalign(size, size);
1553         if (!spapr->htab) {
1554             error_setg_errno(errp, errno,
1555                              "Could not allocate HPT of order %d", shift);
1556             return;
1557         }
1558 
1559         memset(spapr->htab, 0, size);
1560         spapr->htab_shift = shift;
1561 
1562         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1563             DIRTY_HPTE(HPTE(spapr->htab, i));
1564         }
1565     }
1566     /* We're setting up a hash table, so that means we're not radix */
1567     spapr->patb_entry = 0;
1568 }
1569 
1570 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1571 {
1572     int hpt_shift;
1573 
1574     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1575         || (spapr->cas_reboot
1576             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1577         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1578     } else {
1579         uint64_t current_ram_size;
1580 
1581         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1582         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1583     }
1584     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1585 
1586     if (spapr->vrma_adjust) {
1587         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1588                                           spapr->htab_shift);
1589     }
1590 }
1591 
1592 static int spapr_reset_drcs(Object *child, void *opaque)
1593 {
1594     sPAPRDRConnector *drc =
1595         (sPAPRDRConnector *) object_dynamic_cast(child,
1596                                                  TYPE_SPAPR_DR_CONNECTOR);
1597 
1598     if (drc) {
1599         spapr_drc_reset(drc);
1600     }
1601 
1602     return 0;
1603 }
1604 
1605 static void spapr_machine_reset(void)
1606 {
1607     MachineState *machine = MACHINE(qdev_get_machine());
1608     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1609     PowerPCCPU *first_ppc_cpu;
1610     uint32_t rtas_limit;
1611     hwaddr rtas_addr, fdt_addr;
1612     void *fdt;
1613     int rc;
1614 
1615     spapr_caps_reset(spapr);
1616 
1617     first_ppc_cpu = POWERPC_CPU(first_cpu);
1618     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1619         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1620                          spapr->max_compat_pvr)) {
1621         /* If using KVM with radix mode available, VCPUs can be started
1622          * without a HPT because KVM will start them in radix mode.
1623          * Set the GR bit in PATB so that we know there is no HPT. */
1624         spapr->patb_entry = PATBE1_GR;
1625     } else {
1626         spapr_setup_hpt_and_vrma(spapr);
1627     }
1628 
1629     /* if this reset wasn't generated by CAS, we should reset our
1630      * negotiated options and start from scratch */
1631     if (!spapr->cas_reboot) {
1632         spapr_ovec_cleanup(spapr->ov5_cas);
1633         spapr->ov5_cas = spapr_ovec_new();
1634 
1635         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1636     }
1637 
1638     qemu_devices_reset();
1639 
1640     /* DRC reset may cause a device to be unplugged. This will cause troubles
1641      * if this device is used by another device (eg, a running vhost backend
1642      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1643      * situations, we reset DRCs after all devices have been reset.
1644      */
1645     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1646 
1647     spapr_clear_pending_events(spapr);
1648 
1649     /*
1650      * We place the device tree and RTAS just below either the top of the RMA,
1651      * or just below 2GB, whichever is lowere, so that it can be
1652      * processed with 32-bit real mode code if necessary
1653      */
1654     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1655     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1656     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1657 
1658     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1659 
1660     spapr_load_rtas(spapr, fdt, rtas_addr);
1661 
1662     rc = fdt_pack(fdt);
1663 
1664     /* Should only fail if we've built a corrupted tree */
1665     assert(rc == 0);
1666 
1667     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1668         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1669                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1670         exit(1);
1671     }
1672 
1673     /* Load the fdt */
1674     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1675     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1676     g_free(fdt);
1677 
1678     /* Set up the entry state */
1679     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1680     first_ppc_cpu->env.gpr[5] = 0;
1681 
1682     spapr->cas_reboot = false;
1683 }
1684 
1685 static void spapr_create_nvram(sPAPRMachineState *spapr)
1686 {
1687     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1688     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1689 
1690     if (dinfo) {
1691         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1692                             &error_fatal);
1693     }
1694 
1695     qdev_init_nofail(dev);
1696 
1697     spapr->nvram = (struct sPAPRNVRAM *)dev;
1698 }
1699 
1700 static void spapr_rtc_create(sPAPRMachineState *spapr)
1701 {
1702     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1703     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1704                               &error_fatal);
1705     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1706                               &error_fatal);
1707     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1708                               "date", &error_fatal);
1709 }
1710 
1711 /* Returns whether we want to use VGA or not */
1712 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1713 {
1714     switch (vga_interface_type) {
1715     case VGA_NONE:
1716         return false;
1717     case VGA_DEVICE:
1718         return true;
1719     case VGA_STD:
1720     case VGA_VIRTIO:
1721         return pci_vga_init(pci_bus) != NULL;
1722     default:
1723         error_setg(errp,
1724                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1725         return false;
1726     }
1727 }
1728 
1729 static int spapr_pre_load(void *opaque)
1730 {
1731     int rc;
1732 
1733     rc = spapr_caps_pre_load(opaque);
1734     if (rc) {
1735         return rc;
1736     }
1737 
1738     return 0;
1739 }
1740 
1741 static int spapr_post_load(void *opaque, int version_id)
1742 {
1743     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1744     int err = 0;
1745 
1746     err = spapr_caps_post_migration(spapr);
1747     if (err) {
1748         return err;
1749     }
1750 
1751     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1752         CPUState *cs;
1753         CPU_FOREACH(cs) {
1754             PowerPCCPU *cpu = POWERPC_CPU(cs);
1755             icp_resend(ICP(cpu->intc));
1756         }
1757     }
1758 
1759     /* In earlier versions, there was no separate qdev for the PAPR
1760      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1761      * So when migrating from those versions, poke the incoming offset
1762      * value into the RTC device */
1763     if (version_id < 3) {
1764         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1765     }
1766 
1767     if (kvm_enabled() && spapr->patb_entry) {
1768         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1769         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1770         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1771 
1772         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1773         if (err) {
1774             error_report("Process table config unsupported by the host");
1775             return -EINVAL;
1776         }
1777     }
1778 
1779     return err;
1780 }
1781 
1782 static int spapr_pre_save(void *opaque)
1783 {
1784     int rc;
1785 
1786     rc = spapr_caps_pre_save(opaque);
1787     if (rc) {
1788         return rc;
1789     }
1790 
1791     return 0;
1792 }
1793 
1794 static bool version_before_3(void *opaque, int version_id)
1795 {
1796     return version_id < 3;
1797 }
1798 
1799 static bool spapr_pending_events_needed(void *opaque)
1800 {
1801     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1802     return !QTAILQ_EMPTY(&spapr->pending_events);
1803 }
1804 
1805 static const VMStateDescription vmstate_spapr_event_entry = {
1806     .name = "spapr_event_log_entry",
1807     .version_id = 1,
1808     .minimum_version_id = 1,
1809     .fields = (VMStateField[]) {
1810         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1811         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1812         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1813                                      NULL, extended_length),
1814         VMSTATE_END_OF_LIST()
1815     },
1816 };
1817 
1818 static const VMStateDescription vmstate_spapr_pending_events = {
1819     .name = "spapr_pending_events",
1820     .version_id = 1,
1821     .minimum_version_id = 1,
1822     .needed = spapr_pending_events_needed,
1823     .fields = (VMStateField[]) {
1824         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1825                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1826         VMSTATE_END_OF_LIST()
1827     },
1828 };
1829 
1830 static bool spapr_ov5_cas_needed(void *opaque)
1831 {
1832     sPAPRMachineState *spapr = opaque;
1833     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1834     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1835     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1836     bool cas_needed;
1837 
1838     /* Prior to the introduction of sPAPROptionVector, we had two option
1839      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1840      * Both of these options encode machine topology into the device-tree
1841      * in such a way that the now-booted OS should still be able to interact
1842      * appropriately with QEMU regardless of what options were actually
1843      * negotiatied on the source side.
1844      *
1845      * As such, we can avoid migrating the CAS-negotiated options if these
1846      * are the only options available on the current machine/platform.
1847      * Since these are the only options available for pseries-2.7 and
1848      * earlier, this allows us to maintain old->new/new->old migration
1849      * compatibility.
1850      *
1851      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1852      * via default pseries-2.8 machines and explicit command-line parameters.
1853      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1854      * of the actual CAS-negotiated values to continue working properly. For
1855      * example, availability of memory unplug depends on knowing whether
1856      * OV5_HP_EVT was negotiated via CAS.
1857      *
1858      * Thus, for any cases where the set of available CAS-negotiatable
1859      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1860      * include the CAS-negotiated options in the migration stream, unless
1861      * if they affect boot time behaviour only.
1862      */
1863     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1864     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1865     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1866 
1867     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1868      * the mask itself since in the future it's possible "legacy" bits may be
1869      * removed via machine options, which could generate a false positive
1870      * that breaks migration.
1871      */
1872     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1873     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1874 
1875     spapr_ovec_cleanup(ov5_mask);
1876     spapr_ovec_cleanup(ov5_legacy);
1877     spapr_ovec_cleanup(ov5_removed);
1878 
1879     return cas_needed;
1880 }
1881 
1882 static const VMStateDescription vmstate_spapr_ov5_cas = {
1883     .name = "spapr_option_vector_ov5_cas",
1884     .version_id = 1,
1885     .minimum_version_id = 1,
1886     .needed = spapr_ov5_cas_needed,
1887     .fields = (VMStateField[]) {
1888         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1889                                  vmstate_spapr_ovec, sPAPROptionVector),
1890         VMSTATE_END_OF_LIST()
1891     },
1892 };
1893 
1894 static bool spapr_patb_entry_needed(void *opaque)
1895 {
1896     sPAPRMachineState *spapr = opaque;
1897 
1898     return !!spapr->patb_entry;
1899 }
1900 
1901 static const VMStateDescription vmstate_spapr_patb_entry = {
1902     .name = "spapr_patb_entry",
1903     .version_id = 1,
1904     .minimum_version_id = 1,
1905     .needed = spapr_patb_entry_needed,
1906     .fields = (VMStateField[]) {
1907         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1908         VMSTATE_END_OF_LIST()
1909     },
1910 };
1911 
1912 static const VMStateDescription vmstate_spapr = {
1913     .name = "spapr",
1914     .version_id = 3,
1915     .minimum_version_id = 1,
1916     .pre_load = spapr_pre_load,
1917     .post_load = spapr_post_load,
1918     .pre_save = spapr_pre_save,
1919     .fields = (VMStateField[]) {
1920         /* used to be @next_irq */
1921         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1922 
1923         /* RTC offset */
1924         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1925 
1926         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1927         VMSTATE_END_OF_LIST()
1928     },
1929     .subsections = (const VMStateDescription*[]) {
1930         &vmstate_spapr_ov5_cas,
1931         &vmstate_spapr_patb_entry,
1932         &vmstate_spapr_pending_events,
1933         &vmstate_spapr_cap_htm,
1934         &vmstate_spapr_cap_vsx,
1935         &vmstate_spapr_cap_dfp,
1936         &vmstate_spapr_cap_cfpc,
1937         &vmstate_spapr_cap_sbbc,
1938         &vmstate_spapr_cap_ibs,
1939         NULL
1940     }
1941 };
1942 
1943 static int htab_save_setup(QEMUFile *f, void *opaque)
1944 {
1945     sPAPRMachineState *spapr = opaque;
1946 
1947     /* "Iteration" header */
1948     if (!spapr->htab_shift) {
1949         qemu_put_be32(f, -1);
1950     } else {
1951         qemu_put_be32(f, spapr->htab_shift);
1952     }
1953 
1954     if (spapr->htab) {
1955         spapr->htab_save_index = 0;
1956         spapr->htab_first_pass = true;
1957     } else {
1958         if (spapr->htab_shift) {
1959             assert(kvm_enabled());
1960         }
1961     }
1962 
1963 
1964     return 0;
1965 }
1966 
1967 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1968                             int chunkstart, int n_valid, int n_invalid)
1969 {
1970     qemu_put_be32(f, chunkstart);
1971     qemu_put_be16(f, n_valid);
1972     qemu_put_be16(f, n_invalid);
1973     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1974                     HASH_PTE_SIZE_64 * n_valid);
1975 }
1976 
1977 static void htab_save_end_marker(QEMUFile *f)
1978 {
1979     qemu_put_be32(f, 0);
1980     qemu_put_be16(f, 0);
1981     qemu_put_be16(f, 0);
1982 }
1983 
1984 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1985                                  int64_t max_ns)
1986 {
1987     bool has_timeout = max_ns != -1;
1988     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1989     int index = spapr->htab_save_index;
1990     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1991 
1992     assert(spapr->htab_first_pass);
1993 
1994     do {
1995         int chunkstart;
1996 
1997         /* Consume invalid HPTEs */
1998         while ((index < htabslots)
1999                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2000             CLEAN_HPTE(HPTE(spapr->htab, index));
2001             index++;
2002         }
2003 
2004         /* Consume valid HPTEs */
2005         chunkstart = index;
2006         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2007                && HPTE_VALID(HPTE(spapr->htab, index))) {
2008             CLEAN_HPTE(HPTE(spapr->htab, index));
2009             index++;
2010         }
2011 
2012         if (index > chunkstart) {
2013             int n_valid = index - chunkstart;
2014 
2015             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2016 
2017             if (has_timeout &&
2018                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2019                 break;
2020             }
2021         }
2022     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2023 
2024     if (index >= htabslots) {
2025         assert(index == htabslots);
2026         index = 0;
2027         spapr->htab_first_pass = false;
2028     }
2029     spapr->htab_save_index = index;
2030 }
2031 
2032 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2033                                 int64_t max_ns)
2034 {
2035     bool final = max_ns < 0;
2036     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2037     int examined = 0, sent = 0;
2038     int index = spapr->htab_save_index;
2039     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2040 
2041     assert(!spapr->htab_first_pass);
2042 
2043     do {
2044         int chunkstart, invalidstart;
2045 
2046         /* Consume non-dirty HPTEs */
2047         while ((index < htabslots)
2048                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2049             index++;
2050             examined++;
2051         }
2052 
2053         chunkstart = index;
2054         /* Consume valid dirty HPTEs */
2055         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2056                && HPTE_DIRTY(HPTE(spapr->htab, index))
2057                && HPTE_VALID(HPTE(spapr->htab, index))) {
2058             CLEAN_HPTE(HPTE(spapr->htab, index));
2059             index++;
2060             examined++;
2061         }
2062 
2063         invalidstart = index;
2064         /* Consume invalid dirty HPTEs */
2065         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2066                && HPTE_DIRTY(HPTE(spapr->htab, index))
2067                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2068             CLEAN_HPTE(HPTE(spapr->htab, index));
2069             index++;
2070             examined++;
2071         }
2072 
2073         if (index > chunkstart) {
2074             int n_valid = invalidstart - chunkstart;
2075             int n_invalid = index - invalidstart;
2076 
2077             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2078             sent += index - chunkstart;
2079 
2080             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2081                 break;
2082             }
2083         }
2084 
2085         if (examined >= htabslots) {
2086             break;
2087         }
2088 
2089         if (index >= htabslots) {
2090             assert(index == htabslots);
2091             index = 0;
2092         }
2093     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2094 
2095     if (index >= htabslots) {
2096         assert(index == htabslots);
2097         index = 0;
2098     }
2099 
2100     spapr->htab_save_index = index;
2101 
2102     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2103 }
2104 
2105 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2106 #define MAX_KVM_BUF_SIZE    2048
2107 
2108 static int htab_save_iterate(QEMUFile *f, void *opaque)
2109 {
2110     sPAPRMachineState *spapr = opaque;
2111     int fd;
2112     int rc = 0;
2113 
2114     /* Iteration header */
2115     if (!spapr->htab_shift) {
2116         qemu_put_be32(f, -1);
2117         return 1;
2118     } else {
2119         qemu_put_be32(f, 0);
2120     }
2121 
2122     if (!spapr->htab) {
2123         assert(kvm_enabled());
2124 
2125         fd = get_htab_fd(spapr);
2126         if (fd < 0) {
2127             return fd;
2128         }
2129 
2130         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2131         if (rc < 0) {
2132             return rc;
2133         }
2134     } else  if (spapr->htab_first_pass) {
2135         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2136     } else {
2137         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2138     }
2139 
2140     htab_save_end_marker(f);
2141 
2142     return rc;
2143 }
2144 
2145 static int htab_save_complete(QEMUFile *f, void *opaque)
2146 {
2147     sPAPRMachineState *spapr = opaque;
2148     int fd;
2149 
2150     /* Iteration header */
2151     if (!spapr->htab_shift) {
2152         qemu_put_be32(f, -1);
2153         return 0;
2154     } else {
2155         qemu_put_be32(f, 0);
2156     }
2157 
2158     if (!spapr->htab) {
2159         int rc;
2160 
2161         assert(kvm_enabled());
2162 
2163         fd = get_htab_fd(spapr);
2164         if (fd < 0) {
2165             return fd;
2166         }
2167 
2168         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2169         if (rc < 0) {
2170             return rc;
2171         }
2172     } else {
2173         if (spapr->htab_first_pass) {
2174             htab_save_first_pass(f, spapr, -1);
2175         }
2176         htab_save_later_pass(f, spapr, -1);
2177     }
2178 
2179     /* End marker */
2180     htab_save_end_marker(f);
2181 
2182     return 0;
2183 }
2184 
2185 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2186 {
2187     sPAPRMachineState *spapr = opaque;
2188     uint32_t section_hdr;
2189     int fd = -1;
2190     Error *local_err = NULL;
2191 
2192     if (version_id < 1 || version_id > 1) {
2193         error_report("htab_load() bad version");
2194         return -EINVAL;
2195     }
2196 
2197     section_hdr = qemu_get_be32(f);
2198 
2199     if (section_hdr == -1) {
2200         spapr_free_hpt(spapr);
2201         return 0;
2202     }
2203 
2204     if (section_hdr) {
2205         /* First section gives the htab size */
2206         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2207         if (local_err) {
2208             error_report_err(local_err);
2209             return -EINVAL;
2210         }
2211         return 0;
2212     }
2213 
2214     if (!spapr->htab) {
2215         assert(kvm_enabled());
2216 
2217         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2218         if (fd < 0) {
2219             error_report_err(local_err);
2220             return fd;
2221         }
2222     }
2223 
2224     while (true) {
2225         uint32_t index;
2226         uint16_t n_valid, n_invalid;
2227 
2228         index = qemu_get_be32(f);
2229         n_valid = qemu_get_be16(f);
2230         n_invalid = qemu_get_be16(f);
2231 
2232         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2233             /* End of Stream */
2234             break;
2235         }
2236 
2237         if ((index + n_valid + n_invalid) >
2238             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2239             /* Bad index in stream */
2240             error_report(
2241                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2242                 index, n_valid, n_invalid, spapr->htab_shift);
2243             return -EINVAL;
2244         }
2245 
2246         if (spapr->htab) {
2247             if (n_valid) {
2248                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2249                                 HASH_PTE_SIZE_64 * n_valid);
2250             }
2251             if (n_invalid) {
2252                 memset(HPTE(spapr->htab, index + n_valid), 0,
2253                        HASH_PTE_SIZE_64 * n_invalid);
2254             }
2255         } else {
2256             int rc;
2257 
2258             assert(fd >= 0);
2259 
2260             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2261             if (rc < 0) {
2262                 return rc;
2263             }
2264         }
2265     }
2266 
2267     if (!spapr->htab) {
2268         assert(fd >= 0);
2269         close(fd);
2270     }
2271 
2272     return 0;
2273 }
2274 
2275 static void htab_save_cleanup(void *opaque)
2276 {
2277     sPAPRMachineState *spapr = opaque;
2278 
2279     close_htab_fd(spapr);
2280 }
2281 
2282 static SaveVMHandlers savevm_htab_handlers = {
2283     .save_setup = htab_save_setup,
2284     .save_live_iterate = htab_save_iterate,
2285     .save_live_complete_precopy = htab_save_complete,
2286     .save_cleanup = htab_save_cleanup,
2287     .load_state = htab_load,
2288 };
2289 
2290 static void spapr_boot_set(void *opaque, const char *boot_device,
2291                            Error **errp)
2292 {
2293     MachineState *machine = MACHINE(opaque);
2294     machine->boot_order = g_strdup(boot_device);
2295 }
2296 
2297 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2298 {
2299     MachineState *machine = MACHINE(spapr);
2300     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2301     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2302     int i;
2303 
2304     for (i = 0; i < nr_lmbs; i++) {
2305         uint64_t addr;
2306 
2307         addr = i * lmb_size + machine->device_memory->base;
2308         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2309                                addr / lmb_size);
2310     }
2311 }
2312 
2313 /*
2314  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2315  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2316  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2317  */
2318 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2319 {
2320     int i;
2321 
2322     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2323         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2324                    " is not aligned to %llu MiB",
2325                    machine->ram_size,
2326                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2327         return;
2328     }
2329 
2330     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2331         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2332                    " is not aligned to %llu MiB",
2333                    machine->ram_size,
2334                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2335         return;
2336     }
2337 
2338     for (i = 0; i < nb_numa_nodes; i++) {
2339         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2340             error_setg(errp,
2341                        "Node %d memory size 0x%" PRIx64
2342                        " is not aligned to %llu MiB",
2343                        i, numa_info[i].node_mem,
2344                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2345             return;
2346         }
2347     }
2348 }
2349 
2350 /* find cpu slot in machine->possible_cpus by core_id */
2351 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2352 {
2353     int index = id / smp_threads;
2354 
2355     if (index >= ms->possible_cpus->len) {
2356         return NULL;
2357     }
2358     if (idx) {
2359         *idx = index;
2360     }
2361     return &ms->possible_cpus->cpus[index];
2362 }
2363 
2364 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2365 {
2366     Error *local_err = NULL;
2367     bool vsmt_user = !!spapr->vsmt;
2368     int kvm_smt = kvmppc_smt_threads();
2369     int ret;
2370 
2371     if (!kvm_enabled() && (smp_threads > 1)) {
2372         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2373                      "on a pseries machine");
2374         goto out;
2375     }
2376     if (!is_power_of_2(smp_threads)) {
2377         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2378                      "machine because it must be a power of 2", smp_threads);
2379         goto out;
2380     }
2381 
2382     /* Detemine the VSMT mode to use: */
2383     if (vsmt_user) {
2384         if (spapr->vsmt < smp_threads) {
2385             error_setg(&local_err, "Cannot support VSMT mode %d"
2386                          " because it must be >= threads/core (%d)",
2387                          spapr->vsmt, smp_threads);
2388             goto out;
2389         }
2390         /* In this case, spapr->vsmt has been set by the command line */
2391     } else {
2392         /*
2393          * Default VSMT value is tricky, because we need it to be as
2394          * consistent as possible (for migration), but this requires
2395          * changing it for at least some existing cases.  We pick 8 as
2396          * the value that we'd get with KVM on POWER8, the
2397          * overwhelmingly common case in production systems.
2398          */
2399         spapr->vsmt = MAX(8, smp_threads);
2400     }
2401 
2402     /* KVM: If necessary, set the SMT mode: */
2403     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2404         ret = kvmppc_set_smt_threads(spapr->vsmt);
2405         if (ret) {
2406             /* Looks like KVM isn't able to change VSMT mode */
2407             error_setg(&local_err,
2408                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2409                        spapr->vsmt, ret);
2410             /* We can live with that if the default one is big enough
2411              * for the number of threads, and a submultiple of the one
2412              * we want.  In this case we'll waste some vcpu ids, but
2413              * behaviour will be correct */
2414             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2415                 warn_report_err(local_err);
2416                 local_err = NULL;
2417                 goto out;
2418             } else {
2419                 if (!vsmt_user) {
2420                     error_append_hint(&local_err,
2421                                       "On PPC, a VM with %d threads/core"
2422                                       " on a host with %d threads/core"
2423                                       " requires the use of VSMT mode %d.\n",
2424                                       smp_threads, kvm_smt, spapr->vsmt);
2425                 }
2426                 kvmppc_hint_smt_possible(&local_err);
2427                 goto out;
2428             }
2429         }
2430     }
2431     /* else TCG: nothing to do currently */
2432 out:
2433     error_propagate(errp, local_err);
2434 }
2435 
2436 static void spapr_init_cpus(sPAPRMachineState *spapr)
2437 {
2438     MachineState *machine = MACHINE(spapr);
2439     MachineClass *mc = MACHINE_GET_CLASS(machine);
2440     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2441     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2442     const CPUArchIdList *possible_cpus;
2443     int boot_cores_nr = smp_cpus / smp_threads;
2444     int i;
2445 
2446     possible_cpus = mc->possible_cpu_arch_ids(machine);
2447     if (mc->has_hotpluggable_cpus) {
2448         if (smp_cpus % smp_threads) {
2449             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2450                          smp_cpus, smp_threads);
2451             exit(1);
2452         }
2453         if (max_cpus % smp_threads) {
2454             error_report("max_cpus (%u) must be multiple of threads (%u)",
2455                          max_cpus, smp_threads);
2456             exit(1);
2457         }
2458     } else {
2459         if (max_cpus != smp_cpus) {
2460             error_report("This machine version does not support CPU hotplug");
2461             exit(1);
2462         }
2463         boot_cores_nr = possible_cpus->len;
2464     }
2465 
2466     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2467      * call xics_max_server_number() or spapr_vcpu_id().
2468      */
2469     spapr_set_vsmt_mode(spapr, &error_fatal);
2470 
2471     if (smc->pre_2_10_has_unused_icps) {
2472         int i;
2473 
2474         for (i = 0; i < xics_max_server_number(spapr); i++) {
2475             /* Dummy entries get deregistered when real ICPState objects
2476              * are registered during CPU core hotplug.
2477              */
2478             pre_2_10_vmstate_register_dummy_icp(i);
2479         }
2480     }
2481 
2482     for (i = 0; i < possible_cpus->len; i++) {
2483         int core_id = i * smp_threads;
2484 
2485         if (mc->has_hotpluggable_cpus) {
2486             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2487                                    spapr_vcpu_id(spapr, core_id));
2488         }
2489 
2490         if (i < boot_cores_nr) {
2491             Object *core  = object_new(type);
2492             int nr_threads = smp_threads;
2493 
2494             /* Handle the partially filled core for older machine types */
2495             if ((i + 1) * smp_threads >= smp_cpus) {
2496                 nr_threads = smp_cpus - i * smp_threads;
2497             }
2498 
2499             object_property_set_int(core, nr_threads, "nr-threads",
2500                                     &error_fatal);
2501             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2502                                     &error_fatal);
2503             object_property_set_bool(core, true, "realized", &error_fatal);
2504         }
2505     }
2506 }
2507 
2508 /* pSeries LPAR / sPAPR hardware init */
2509 static void spapr_machine_init(MachineState *machine)
2510 {
2511     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2512     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2513     const char *kernel_filename = machine->kernel_filename;
2514     const char *initrd_filename = machine->initrd_filename;
2515     PCIHostState *phb;
2516     int i;
2517     MemoryRegion *sysmem = get_system_memory();
2518     MemoryRegion *ram = g_new(MemoryRegion, 1);
2519     hwaddr node0_size = spapr_node0_size(machine);
2520     long load_limit, fw_size;
2521     char *filename;
2522     Error *resize_hpt_err = NULL;
2523     PowerPCCPU *first_ppc_cpu;
2524 
2525     msi_nonbroken = true;
2526 
2527     QLIST_INIT(&spapr->phbs);
2528     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2529 
2530     /* Check HPT resizing availability */
2531     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2532     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2533         /*
2534          * If the user explicitly requested a mode we should either
2535          * supply it, or fail completely (which we do below).  But if
2536          * it's not set explicitly, we reset our mode to something
2537          * that works
2538          */
2539         if (resize_hpt_err) {
2540             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2541             error_free(resize_hpt_err);
2542             resize_hpt_err = NULL;
2543         } else {
2544             spapr->resize_hpt = smc->resize_hpt_default;
2545         }
2546     }
2547 
2548     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2549 
2550     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2551         /*
2552          * User requested HPT resize, but this host can't supply it.  Bail out
2553          */
2554         error_report_err(resize_hpt_err);
2555         exit(1);
2556     }
2557 
2558     spapr->rma_size = node0_size;
2559 
2560     /* With KVM, we don't actually know whether KVM supports an
2561      * unbounded RMA (PR KVM) or is limited by the hash table size
2562      * (HV KVM using VRMA), so we always assume the latter
2563      *
2564      * In that case, we also limit the initial allocations for RTAS
2565      * etc... to 256M since we have no way to know what the VRMA size
2566      * is going to be as it depends on the size of the hash table
2567      * which isn't determined yet.
2568      */
2569     if (kvm_enabled()) {
2570         spapr->vrma_adjust = 1;
2571         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2572     }
2573 
2574     /* Actually we don't support unbounded RMA anymore since we added
2575      * proper emulation of HV mode. The max we can get is 16G which
2576      * also happens to be what we configure for PAPR mode so make sure
2577      * we don't do anything bigger than that
2578      */
2579     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2580 
2581     if (spapr->rma_size > node0_size) {
2582         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2583                      spapr->rma_size);
2584         exit(1);
2585     }
2586 
2587     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2588     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2589 
2590     /* Set up Interrupt Controller before we create the VCPUs */
2591     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2592 
2593     /* Set up containers for ibm,client-architecture-support negotiated options
2594      */
2595     spapr->ov5 = spapr_ovec_new();
2596     spapr->ov5_cas = spapr_ovec_new();
2597 
2598     if (smc->dr_lmb_enabled) {
2599         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2600         spapr_validate_node_memory(machine, &error_fatal);
2601     }
2602 
2603     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2604 
2605     /* advertise support for dedicated HP event source to guests */
2606     if (spapr->use_hotplug_event_source) {
2607         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2608     }
2609 
2610     /* advertise support for HPT resizing */
2611     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2612         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2613     }
2614 
2615     /* advertise support for ibm,dyamic-memory-v2 */
2616     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2617 
2618     /* init CPUs */
2619     spapr_init_cpus(spapr);
2620 
2621     first_ppc_cpu = POWERPC_CPU(first_cpu);
2622     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2623         ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
2624                          spapr->max_compat_pvr)) {
2625         /* KVM and TCG always allow GTSE with radix... */
2626         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2627     }
2628     /* ... but not with hash (currently). */
2629 
2630     if (kvm_enabled()) {
2631         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2632         kvmppc_enable_logical_ci_hcalls();
2633         kvmppc_enable_set_mode_hcall();
2634 
2635         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2636         kvmppc_enable_clear_ref_mod_hcalls();
2637     }
2638 
2639     /* allocate RAM */
2640     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2641                                          machine->ram_size);
2642     memory_region_add_subregion(sysmem, 0, ram);
2643 
2644     /* always allocate the device memory information */
2645     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2646 
2647     /* initialize hotplug memory address space */
2648     if (machine->ram_size < machine->maxram_size) {
2649         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2650         /*
2651          * Limit the number of hotpluggable memory slots to half the number
2652          * slots that KVM supports, leaving the other half for PCI and other
2653          * devices. However ensure that number of slots doesn't drop below 32.
2654          */
2655         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2656                            SPAPR_MAX_RAM_SLOTS;
2657 
2658         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2659             max_memslots = SPAPR_MAX_RAM_SLOTS;
2660         }
2661         if (machine->ram_slots > max_memslots) {
2662             error_report("Specified number of memory slots %"
2663                          PRIu64" exceeds max supported %d",
2664                          machine->ram_slots, max_memslots);
2665             exit(1);
2666         }
2667 
2668         machine->device_memory->base = ROUND_UP(machine->ram_size,
2669                                                 SPAPR_DEVICE_MEM_ALIGN);
2670         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2671                            "device-memory", device_mem_size);
2672         memory_region_add_subregion(sysmem, machine->device_memory->base,
2673                                     &machine->device_memory->mr);
2674     }
2675 
2676     if (smc->dr_lmb_enabled) {
2677         spapr_create_lmb_dr_connectors(spapr);
2678     }
2679 
2680     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2681     if (!filename) {
2682         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2683         exit(1);
2684     }
2685     spapr->rtas_size = get_image_size(filename);
2686     if (spapr->rtas_size < 0) {
2687         error_report("Could not get size of LPAR rtas '%s'", filename);
2688         exit(1);
2689     }
2690     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2691     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2692         error_report("Could not load LPAR rtas '%s'", filename);
2693         exit(1);
2694     }
2695     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2696         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2697                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2698         exit(1);
2699     }
2700     g_free(filename);
2701 
2702     /* Set up RTAS event infrastructure */
2703     spapr_events_init(spapr);
2704 
2705     /* Set up the RTC RTAS interfaces */
2706     spapr_rtc_create(spapr);
2707 
2708     /* Set up VIO bus */
2709     spapr->vio_bus = spapr_vio_bus_init();
2710 
2711     for (i = 0; i < serial_max_hds(); i++) {
2712         if (serial_hd(i)) {
2713             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2714         }
2715     }
2716 
2717     /* We always have at least the nvram device on VIO */
2718     spapr_create_nvram(spapr);
2719 
2720     /* Set up PCI */
2721     spapr_pci_rtas_init();
2722 
2723     phb = spapr_create_phb(spapr, 0);
2724 
2725     for (i = 0; i < nb_nics; i++) {
2726         NICInfo *nd = &nd_table[i];
2727 
2728         if (!nd->model) {
2729             nd->model = g_strdup("spapr-vlan");
2730         }
2731 
2732         if (g_str_equal(nd->model, "spapr-vlan") ||
2733             g_str_equal(nd->model, "ibmveth")) {
2734             spapr_vlan_create(spapr->vio_bus, nd);
2735         } else {
2736             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2737         }
2738     }
2739 
2740     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2741         spapr_vscsi_create(spapr->vio_bus);
2742     }
2743 
2744     /* Graphics */
2745     if (spapr_vga_init(phb->bus, &error_fatal)) {
2746         spapr->has_graphics = true;
2747         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2748     }
2749 
2750     if (machine->usb) {
2751         if (smc->use_ohci_by_default) {
2752             pci_create_simple(phb->bus, -1, "pci-ohci");
2753         } else {
2754             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2755         }
2756 
2757         if (spapr->has_graphics) {
2758             USBBus *usb_bus = usb_bus_find(-1);
2759 
2760             usb_create_simple(usb_bus, "usb-kbd");
2761             usb_create_simple(usb_bus, "usb-mouse");
2762         }
2763     }
2764 
2765     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2766         error_report(
2767             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2768             MIN_RMA_SLOF);
2769         exit(1);
2770     }
2771 
2772     if (kernel_filename) {
2773         uint64_t lowaddr = 0;
2774 
2775         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2776                                       NULL, NULL, &lowaddr, NULL, 1,
2777                                       PPC_ELF_MACHINE, 0, 0);
2778         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2779             spapr->kernel_size = load_elf(kernel_filename,
2780                                           translate_kernel_address, NULL, NULL,
2781                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2782                                           0, 0);
2783             spapr->kernel_le = spapr->kernel_size > 0;
2784         }
2785         if (spapr->kernel_size < 0) {
2786             error_report("error loading %s: %s", kernel_filename,
2787                          load_elf_strerror(spapr->kernel_size));
2788             exit(1);
2789         }
2790 
2791         /* load initrd */
2792         if (initrd_filename) {
2793             /* Try to locate the initrd in the gap between the kernel
2794              * and the firmware. Add a bit of space just in case
2795              */
2796             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2797                                   + 0x1ffff) & ~0xffff;
2798             spapr->initrd_size = load_image_targphys(initrd_filename,
2799                                                      spapr->initrd_base,
2800                                                      load_limit
2801                                                      - spapr->initrd_base);
2802             if (spapr->initrd_size < 0) {
2803                 error_report("could not load initial ram disk '%s'",
2804                              initrd_filename);
2805                 exit(1);
2806             }
2807         }
2808     }
2809 
2810     if (bios_name == NULL) {
2811         bios_name = FW_FILE_NAME;
2812     }
2813     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2814     if (!filename) {
2815         error_report("Could not find LPAR firmware '%s'", bios_name);
2816         exit(1);
2817     }
2818     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2819     if (fw_size <= 0) {
2820         error_report("Could not load LPAR firmware '%s'", filename);
2821         exit(1);
2822     }
2823     g_free(filename);
2824 
2825     /* FIXME: Should register things through the MachineState's qdev
2826      * interface, this is a legacy from the sPAPREnvironment structure
2827      * which predated MachineState but had a similar function */
2828     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2829     register_savevm_live(NULL, "spapr/htab", -1, 1,
2830                          &savevm_htab_handlers, spapr);
2831 
2832     qemu_register_boot_set(spapr_boot_set, spapr);
2833 
2834     if (kvm_enabled()) {
2835         /* to stop and start vmclock */
2836         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2837                                          &spapr->tb);
2838 
2839         kvmppc_spapr_enable_inkernel_multitce();
2840     }
2841 }
2842 
2843 static int spapr_kvm_type(const char *vm_type)
2844 {
2845     if (!vm_type) {
2846         return 0;
2847     }
2848 
2849     if (!strcmp(vm_type, "HV")) {
2850         return 1;
2851     }
2852 
2853     if (!strcmp(vm_type, "PR")) {
2854         return 2;
2855     }
2856 
2857     error_report("Unknown kvm-type specified '%s'", vm_type);
2858     exit(1);
2859 }
2860 
2861 /*
2862  * Implementation of an interface to adjust firmware path
2863  * for the bootindex property handling.
2864  */
2865 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2866                                    DeviceState *dev)
2867 {
2868 #define CAST(type, obj, name) \
2869     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2870     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2871     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2872     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2873 
2874     if (d) {
2875         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2876         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2877         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2878 
2879         if (spapr) {
2880             /*
2881              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2882              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2883              * in the top 16 bits of the 64-bit LUN
2884              */
2885             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2886             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2887                                    (uint64_t)id << 48);
2888         } else if (virtio) {
2889             /*
2890              * We use SRP luns of the form 01000000 | (target << 8) | lun
2891              * in the top 32 bits of the 64-bit LUN
2892              * Note: the quote above is from SLOF and it is wrong,
2893              * the actual binding is:
2894              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2895              */
2896             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2897             if (d->lun >= 256) {
2898                 /* Use the LUN "flat space addressing method" */
2899                 id |= 0x4000;
2900             }
2901             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2902                                    (uint64_t)id << 32);
2903         } else if (usb) {
2904             /*
2905              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2906              * in the top 32 bits of the 64-bit LUN
2907              */
2908             unsigned usb_port = atoi(usb->port->path);
2909             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2910             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2911                                    (uint64_t)id << 32);
2912         }
2913     }
2914 
2915     /*
2916      * SLOF probes the USB devices, and if it recognizes that the device is a
2917      * storage device, it changes its name to "storage" instead of "usb-host",
2918      * and additionally adds a child node for the SCSI LUN, so the correct
2919      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2920      */
2921     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2922         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2923         if (usb_host_dev_is_scsi_storage(usbdev)) {
2924             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2925         }
2926     }
2927 
2928     if (phb) {
2929         /* Replace "pci" with "pci@800000020000000" */
2930         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2931     }
2932 
2933     if (vsc) {
2934         /* Same logic as virtio above */
2935         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2936         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2937     }
2938 
2939     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2940         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2941         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2942         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2943     }
2944 
2945     return NULL;
2946 }
2947 
2948 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2949 {
2950     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2951 
2952     return g_strdup(spapr->kvm_type);
2953 }
2954 
2955 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2956 {
2957     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2958 
2959     g_free(spapr->kvm_type);
2960     spapr->kvm_type = g_strdup(value);
2961 }
2962 
2963 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2964 {
2965     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2966 
2967     return spapr->use_hotplug_event_source;
2968 }
2969 
2970 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2971                                             Error **errp)
2972 {
2973     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2974 
2975     spapr->use_hotplug_event_source = value;
2976 }
2977 
2978 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2979 {
2980     return true;
2981 }
2982 
2983 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2984 {
2985     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2986 
2987     switch (spapr->resize_hpt) {
2988     case SPAPR_RESIZE_HPT_DEFAULT:
2989         return g_strdup("default");
2990     case SPAPR_RESIZE_HPT_DISABLED:
2991         return g_strdup("disabled");
2992     case SPAPR_RESIZE_HPT_ENABLED:
2993         return g_strdup("enabled");
2994     case SPAPR_RESIZE_HPT_REQUIRED:
2995         return g_strdup("required");
2996     }
2997     g_assert_not_reached();
2998 }
2999 
3000 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3001 {
3002     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3003 
3004     if (strcmp(value, "default") == 0) {
3005         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3006     } else if (strcmp(value, "disabled") == 0) {
3007         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3008     } else if (strcmp(value, "enabled") == 0) {
3009         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3010     } else if (strcmp(value, "required") == 0) {
3011         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3012     } else {
3013         error_setg(errp, "Bad value for \"resize-hpt\" property");
3014     }
3015 }
3016 
3017 static void spapr_get_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_set_vsmt(Object *obj, Visitor *v, const char *name,
3024                                    void *opaque, Error **errp)
3025 {
3026     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3027 }
3028 
3029 static void spapr_instance_init(Object *obj)
3030 {
3031     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3032 
3033     spapr->htab_fd = -1;
3034     spapr->use_hotplug_event_source = true;
3035     object_property_add_str(obj, "kvm-type",
3036                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3037     object_property_set_description(obj, "kvm-type",
3038                                     "Specifies the KVM virtualization mode (HV, PR)",
3039                                     NULL);
3040     object_property_add_bool(obj, "modern-hotplug-events",
3041                             spapr_get_modern_hotplug_events,
3042                             spapr_set_modern_hotplug_events,
3043                             NULL);
3044     object_property_set_description(obj, "modern-hotplug-events",
3045                                     "Use dedicated hotplug event mechanism in"
3046                                     " place of standard EPOW events when possible"
3047                                     " (required for memory hot-unplug support)",
3048                                     NULL);
3049     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3050                             "Maximum permitted CPU compatibility mode",
3051                             &error_fatal);
3052 
3053     object_property_add_str(obj, "resize-hpt",
3054                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3055     object_property_set_description(obj, "resize-hpt",
3056                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3057                                     NULL);
3058     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3059                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3060     object_property_set_description(obj, "vsmt",
3061                                     "Virtual SMT: KVM behaves as if this were"
3062                                     " the host's SMT mode", &error_abort);
3063     object_property_add_bool(obj, "vfio-no-msix-emulation",
3064                              spapr_get_msix_emulation, NULL, NULL);
3065 }
3066 
3067 static void spapr_machine_finalizefn(Object *obj)
3068 {
3069     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3070 
3071     g_free(spapr->kvm_type);
3072 }
3073 
3074 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3075 {
3076     cpu_synchronize_state(cs);
3077     ppc_cpu_do_system_reset(cs);
3078 }
3079 
3080 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3081 {
3082     CPUState *cs;
3083 
3084     CPU_FOREACH(cs) {
3085         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3086     }
3087 }
3088 
3089 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3090                            uint32_t node, bool dedicated_hp_event_source,
3091                            Error **errp)
3092 {
3093     sPAPRDRConnector *drc;
3094     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3095     int i, fdt_offset, fdt_size;
3096     void *fdt;
3097     uint64_t addr = addr_start;
3098     bool hotplugged = spapr_drc_hotplugged(dev);
3099     Error *local_err = NULL;
3100 
3101     for (i = 0; i < nr_lmbs; i++) {
3102         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3103                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3104         g_assert(drc);
3105 
3106         fdt = create_device_tree(&fdt_size);
3107         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3108                                                 SPAPR_MEMORY_BLOCK_SIZE);
3109 
3110         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3111         if (local_err) {
3112             while (addr > addr_start) {
3113                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3114                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3115                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3116                 spapr_drc_detach(drc);
3117             }
3118             g_free(fdt);
3119             error_propagate(errp, local_err);
3120             return;
3121         }
3122         if (!hotplugged) {
3123             spapr_drc_reset(drc);
3124         }
3125         addr += SPAPR_MEMORY_BLOCK_SIZE;
3126     }
3127     /* send hotplug notification to the
3128      * guest only in case of hotplugged memory
3129      */
3130     if (hotplugged) {
3131         if (dedicated_hp_event_source) {
3132             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3133                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3134             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3135                                                    nr_lmbs,
3136                                                    spapr_drc_index(drc));
3137         } else {
3138             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3139                                            nr_lmbs);
3140         }
3141     }
3142 }
3143 
3144 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3145                               Error **errp)
3146 {
3147     Error *local_err = NULL;
3148     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3149     PCDIMMDevice *dimm = PC_DIMM(dev);
3150     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3151     MemoryRegion *mr;
3152     uint64_t align, size, addr;
3153     uint32_t node;
3154 
3155     mr = ddc->get_memory_region(dimm, &local_err);
3156     if (local_err) {
3157         goto out;
3158     }
3159     align = memory_region_get_alignment(mr);
3160     size = memory_region_size(mr);
3161 
3162     pc_dimm_memory_plug(dev, MACHINE(ms), align, &local_err);
3163     if (local_err) {
3164         goto out;
3165     }
3166 
3167     addr = object_property_get_uint(OBJECT(dimm),
3168                                     PC_DIMM_ADDR_PROP, &local_err);
3169     if (local_err) {
3170         goto out_unplug;
3171     }
3172 
3173     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3174                                     &error_abort);
3175     spapr_add_lmbs(dev, addr, size, node,
3176                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3177                    &local_err);
3178     if (local_err) {
3179         goto out_unplug;
3180     }
3181 
3182     return;
3183 
3184 out_unplug:
3185     pc_dimm_memory_unplug(dev, MACHINE(ms));
3186 out:
3187     error_propagate(errp, local_err);
3188 }
3189 
3190 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3191                                   Error **errp)
3192 {
3193     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3194     PCDIMMDevice *dimm = PC_DIMM(dev);
3195     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3196     MemoryRegion *mr;
3197     uint64_t size;
3198     char *mem_dev;
3199 
3200     if (!smc->dr_lmb_enabled) {
3201         error_setg(errp, "Memory hotplug not supported for this machine");
3202         return;
3203     }
3204 
3205     mr = ddc->get_memory_region(dimm, errp);
3206     if (!mr) {
3207         return;
3208     }
3209     size = memory_region_size(mr);
3210 
3211     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3212         error_setg(errp, "Hotplugged memory size must be a multiple of "
3213                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
3214         return;
3215     }
3216 
3217     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
3218     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
3219         error_setg(errp, "Memory backend has bad page size. "
3220                    "Use 'memory-backend-file' with correct mem-path.");
3221         goto out;
3222     }
3223 
3224 out:
3225     g_free(mem_dev);
3226 }
3227 
3228 struct sPAPRDIMMState {
3229     PCDIMMDevice *dimm;
3230     uint32_t nr_lmbs;
3231     QTAILQ_ENTRY(sPAPRDIMMState) next;
3232 };
3233 
3234 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3235                                                        PCDIMMDevice *dimm)
3236 {
3237     sPAPRDIMMState *dimm_state = NULL;
3238 
3239     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3240         if (dimm_state->dimm == dimm) {
3241             break;
3242         }
3243     }
3244     return dimm_state;
3245 }
3246 
3247 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3248                                                       uint32_t nr_lmbs,
3249                                                       PCDIMMDevice *dimm)
3250 {
3251     sPAPRDIMMState *ds = NULL;
3252 
3253     /*
3254      * If this request is for a DIMM whose removal had failed earlier
3255      * (due to guest's refusal to remove the LMBs), we would have this
3256      * dimm already in the pending_dimm_unplugs list. In that
3257      * case don't add again.
3258      */
3259     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3260     if (!ds) {
3261         ds = g_malloc0(sizeof(sPAPRDIMMState));
3262         ds->nr_lmbs = nr_lmbs;
3263         ds->dimm = dimm;
3264         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3265     }
3266     return ds;
3267 }
3268 
3269 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3270                                               sPAPRDIMMState *dimm_state)
3271 {
3272     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3273     g_free(dimm_state);
3274 }
3275 
3276 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3277                                                         PCDIMMDevice *dimm)
3278 {
3279     sPAPRDRConnector *drc;
3280     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3281     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3282     uint64_t size = memory_region_size(mr);
3283     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3284     uint32_t avail_lmbs = 0;
3285     uint64_t addr_start, addr;
3286     int i;
3287 
3288     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3289                                          &error_abort);
3290 
3291     addr = addr_start;
3292     for (i = 0; i < nr_lmbs; i++) {
3293         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3294                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3295         g_assert(drc);
3296         if (drc->dev) {
3297             avail_lmbs++;
3298         }
3299         addr += SPAPR_MEMORY_BLOCK_SIZE;
3300     }
3301 
3302     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3303 }
3304 
3305 /* Callback to be called during DRC release. */
3306 void spapr_lmb_release(DeviceState *dev)
3307 {
3308     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3309     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3310     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3311 
3312     /* This information will get lost if a migration occurs
3313      * during the unplug process. In this case recover it. */
3314     if (ds == NULL) {
3315         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3316         g_assert(ds);
3317         /* The DRC being examined by the caller at least must be counted */
3318         g_assert(ds->nr_lmbs);
3319     }
3320 
3321     if (--ds->nr_lmbs) {
3322         return;
3323     }
3324 
3325     /*
3326      * Now that all the LMBs have been removed by the guest, call the
3327      * unplug handler chain. This can never fail.
3328      */
3329     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3330 }
3331 
3332 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3333 {
3334     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3335     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3336 
3337     pc_dimm_memory_unplug(dev, MACHINE(hotplug_dev));
3338     object_unparent(OBJECT(dev));
3339     spapr_pending_dimm_unplugs_remove(spapr, ds);
3340 }
3341 
3342 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3343                                         DeviceState *dev, Error **errp)
3344 {
3345     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3346     Error *local_err = NULL;
3347     PCDIMMDevice *dimm = PC_DIMM(dev);
3348     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3349     MemoryRegion *mr;
3350     uint32_t nr_lmbs;
3351     uint64_t size, addr_start, addr;
3352     int i;
3353     sPAPRDRConnector *drc;
3354 
3355     mr = ddc->get_memory_region(dimm, &local_err);
3356     if (local_err) {
3357         goto out;
3358     }
3359     size = memory_region_size(mr);
3360     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3361 
3362     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3363                                          &local_err);
3364     if (local_err) {
3365         goto out;
3366     }
3367 
3368     /*
3369      * An existing pending dimm state for this DIMM means that there is an
3370      * unplug operation in progress, waiting for the spapr_lmb_release
3371      * callback to complete the job (BQL can't cover that far). In this case,
3372      * bail out to avoid detaching DRCs that were already released.
3373      */
3374     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3375         error_setg(&local_err,
3376                    "Memory unplug already in progress for device %s",
3377                    dev->id);
3378         goto out;
3379     }
3380 
3381     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3382 
3383     addr = addr_start;
3384     for (i = 0; i < nr_lmbs; i++) {
3385         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3386                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3387         g_assert(drc);
3388 
3389         spapr_drc_detach(drc);
3390         addr += SPAPR_MEMORY_BLOCK_SIZE;
3391     }
3392 
3393     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3394                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3395     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3396                                               nr_lmbs, spapr_drc_index(drc));
3397 out:
3398     error_propagate(errp, local_err);
3399 }
3400 
3401 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3402                                            sPAPRMachineState *spapr)
3403 {
3404     PowerPCCPU *cpu = POWERPC_CPU(cs);
3405     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3406     int id = spapr_get_vcpu_id(cpu);
3407     void *fdt;
3408     int offset, fdt_size;
3409     char *nodename;
3410 
3411     fdt = create_device_tree(&fdt_size);
3412     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3413     offset = fdt_add_subnode(fdt, 0, nodename);
3414 
3415     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3416     g_free(nodename);
3417 
3418     *fdt_offset = offset;
3419     return fdt;
3420 }
3421 
3422 /* Callback to be called during DRC release. */
3423 void spapr_core_release(DeviceState *dev)
3424 {
3425     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3426 
3427     /* Call the unplug handler chain. This can never fail. */
3428     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3429 }
3430 
3431 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3432 {
3433     MachineState *ms = MACHINE(hotplug_dev);
3434     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3435     CPUCore *cc = CPU_CORE(dev);
3436     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3437 
3438     if (smc->pre_2_10_has_unused_icps) {
3439         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3440         int i;
3441 
3442         for (i = 0; i < cc->nr_threads; i++) {
3443             CPUState *cs = CPU(sc->threads[i]);
3444 
3445             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3446         }
3447     }
3448 
3449     assert(core_slot);
3450     core_slot->cpu = NULL;
3451     object_unparent(OBJECT(dev));
3452 }
3453 
3454 static
3455 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3456                                Error **errp)
3457 {
3458     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3459     int index;
3460     sPAPRDRConnector *drc;
3461     CPUCore *cc = CPU_CORE(dev);
3462 
3463     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3464         error_setg(errp, "Unable to find CPU core with core-id: %d",
3465                    cc->core_id);
3466         return;
3467     }
3468     if (index == 0) {
3469         error_setg(errp, "Boot CPU core may not be unplugged");
3470         return;
3471     }
3472 
3473     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3474                           spapr_vcpu_id(spapr, cc->core_id));
3475     g_assert(drc);
3476 
3477     spapr_drc_detach(drc);
3478 
3479     spapr_hotplug_req_remove_by_index(drc);
3480 }
3481 
3482 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3483                             Error **errp)
3484 {
3485     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3486     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3487     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3488     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3489     CPUCore *cc = CPU_CORE(dev);
3490     CPUState *cs = CPU(core->threads[0]);
3491     sPAPRDRConnector *drc;
3492     Error *local_err = NULL;
3493     CPUArchId *core_slot;
3494     int index;
3495     bool hotplugged = spapr_drc_hotplugged(dev);
3496 
3497     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3498     if (!core_slot) {
3499         error_setg(errp, "Unable to find CPU core with core-id: %d",
3500                    cc->core_id);
3501         return;
3502     }
3503     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3504                           spapr_vcpu_id(spapr, cc->core_id));
3505 
3506     g_assert(drc || !mc->has_hotpluggable_cpus);
3507 
3508     if (drc) {
3509         void *fdt;
3510         int fdt_offset;
3511 
3512         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3513 
3514         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3515         if (local_err) {
3516             g_free(fdt);
3517             error_propagate(errp, local_err);
3518             return;
3519         }
3520 
3521         if (hotplugged) {
3522             /*
3523              * Send hotplug notification interrupt to the guest only
3524              * in case of hotplugged CPUs.
3525              */
3526             spapr_hotplug_req_add_by_index(drc);
3527         } else {
3528             spapr_drc_reset(drc);
3529         }
3530     }
3531 
3532     core_slot->cpu = OBJECT(dev);
3533 
3534     if (smc->pre_2_10_has_unused_icps) {
3535         int i;
3536 
3537         for (i = 0; i < cc->nr_threads; i++) {
3538             cs = CPU(core->threads[i]);
3539             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3540         }
3541     }
3542 }
3543 
3544 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3545                                 Error **errp)
3546 {
3547     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3548     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3549     Error *local_err = NULL;
3550     CPUCore *cc = CPU_CORE(dev);
3551     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3552     const char *type = object_get_typename(OBJECT(dev));
3553     CPUArchId *core_slot;
3554     int index;
3555 
3556     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3557         error_setg(&local_err, "CPU hotplug not supported for this machine");
3558         goto out;
3559     }
3560 
3561     if (strcmp(base_core_type, type)) {
3562         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3563         goto out;
3564     }
3565 
3566     if (cc->core_id % smp_threads) {
3567         error_setg(&local_err, "invalid core id %d", cc->core_id);
3568         goto out;
3569     }
3570 
3571     /*
3572      * In general we should have homogeneous threads-per-core, but old
3573      * (pre hotplug support) machine types allow the last core to have
3574      * reduced threads as a compatibility hack for when we allowed
3575      * total vcpus not a multiple of threads-per-core.
3576      */
3577     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3578         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3579                    cc->nr_threads, smp_threads);
3580         goto out;
3581     }
3582 
3583     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3584     if (!core_slot) {
3585         error_setg(&local_err, "core id %d out of range", cc->core_id);
3586         goto out;
3587     }
3588 
3589     if (core_slot->cpu) {
3590         error_setg(&local_err, "core %d already populated", cc->core_id);
3591         goto out;
3592     }
3593 
3594     numa_cpu_pre_plug(core_slot, dev, &local_err);
3595 
3596 out:
3597     error_propagate(errp, local_err);
3598 }
3599 
3600 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3601                                       DeviceState *dev, Error **errp)
3602 {
3603     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3604         spapr_memory_plug(hotplug_dev, dev, errp);
3605     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3606         spapr_core_plug(hotplug_dev, dev, errp);
3607     }
3608 }
3609 
3610 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3611                                         DeviceState *dev, Error **errp)
3612 {
3613     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3614         spapr_memory_unplug(hotplug_dev, dev);
3615     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3616         spapr_core_unplug(hotplug_dev, dev);
3617     }
3618 }
3619 
3620 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3621                                                 DeviceState *dev, Error **errp)
3622 {
3623     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3624     MachineClass *mc = MACHINE_GET_CLASS(sms);
3625 
3626     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3627         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3628             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3629         } else {
3630             /* NOTE: this means there is a window after guest reset, prior to
3631              * CAS negotiation, where unplug requests will fail due to the
3632              * capability not being detected yet. This is a bit different than
3633              * the case with PCI unplug, where the events will be queued and
3634              * eventually handled by the guest after boot
3635              */
3636             error_setg(errp, "Memory hot unplug not supported for this guest");
3637         }
3638     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3639         if (!mc->has_hotpluggable_cpus) {
3640             error_setg(errp, "CPU hot unplug not supported on this machine");
3641             return;
3642         }
3643         spapr_core_unplug_request(hotplug_dev, dev, errp);
3644     }
3645 }
3646 
3647 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3648                                           DeviceState *dev, Error **errp)
3649 {
3650     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3651         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3652     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3653         spapr_core_pre_plug(hotplug_dev, dev, errp);
3654     }
3655 }
3656 
3657 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3658                                                  DeviceState *dev)
3659 {
3660     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3661         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3662         return HOTPLUG_HANDLER(machine);
3663     }
3664     return NULL;
3665 }
3666 
3667 static CpuInstanceProperties
3668 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3669 {
3670     CPUArchId *core_slot;
3671     MachineClass *mc = MACHINE_GET_CLASS(machine);
3672 
3673     /* make sure possible_cpu are intialized */
3674     mc->possible_cpu_arch_ids(machine);
3675     /* get CPU core slot containing thread that matches cpu_index */
3676     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3677     assert(core_slot);
3678     return core_slot->props;
3679 }
3680 
3681 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3682 {
3683     return idx / smp_cores % nb_numa_nodes;
3684 }
3685 
3686 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3687 {
3688     int i;
3689     const char *core_type;
3690     int spapr_max_cores = max_cpus / smp_threads;
3691     MachineClass *mc = MACHINE_GET_CLASS(machine);
3692 
3693     if (!mc->has_hotpluggable_cpus) {
3694         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3695     }
3696     if (machine->possible_cpus) {
3697         assert(machine->possible_cpus->len == spapr_max_cores);
3698         return machine->possible_cpus;
3699     }
3700 
3701     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3702     if (!core_type) {
3703         error_report("Unable to find sPAPR CPU Core definition");
3704         exit(1);
3705     }
3706 
3707     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3708                              sizeof(CPUArchId) * spapr_max_cores);
3709     machine->possible_cpus->len = spapr_max_cores;
3710     for (i = 0; i < machine->possible_cpus->len; i++) {
3711         int core_id = i * smp_threads;
3712 
3713         machine->possible_cpus->cpus[i].type = core_type;
3714         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3715         machine->possible_cpus->cpus[i].arch_id = core_id;
3716         machine->possible_cpus->cpus[i].props.has_core_id = true;
3717         machine->possible_cpus->cpus[i].props.core_id = core_id;
3718     }
3719     return machine->possible_cpus;
3720 }
3721 
3722 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3723                                 uint64_t *buid, hwaddr *pio,
3724                                 hwaddr *mmio32, hwaddr *mmio64,
3725                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3726 {
3727     /*
3728      * New-style PHB window placement.
3729      *
3730      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3731      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3732      * windows.
3733      *
3734      * Some guest kernels can't work with MMIO windows above 1<<46
3735      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3736      *
3737      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3738      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3739      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3740      * 1TiB 64-bit MMIO windows for each PHB.
3741      */
3742     const uint64_t base_buid = 0x800000020000000ULL;
3743 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3744                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3745     int i;
3746 
3747     /* Sanity check natural alignments */
3748     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3749     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3750     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3751     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3752     /* Sanity check bounds */
3753     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3754                       SPAPR_PCI_MEM32_WIN_SIZE);
3755     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3756                       SPAPR_PCI_MEM64_WIN_SIZE);
3757 
3758     if (index >= SPAPR_MAX_PHBS) {
3759         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3760                    SPAPR_MAX_PHBS - 1);
3761         return;
3762     }
3763 
3764     *buid = base_buid + index;
3765     for (i = 0; i < n_dma; ++i) {
3766         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3767     }
3768 
3769     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3770     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3771     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3772 }
3773 
3774 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3775 {
3776     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3777 
3778     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3779 }
3780 
3781 static void spapr_ics_resend(XICSFabric *dev)
3782 {
3783     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3784 
3785     ics_resend(spapr->ics);
3786 }
3787 
3788 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3789 {
3790     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3791 
3792     return cpu ? ICP(cpu->intc) : NULL;
3793 }
3794 
3795 #define ICS_IRQ_FREE(ics, srcno)   \
3796     (!((ics)->irqs[(srcno)].flags & (XICS_FLAGS_IRQ_MASK)))
3797 
3798 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
3799 {
3800     int first, i;
3801 
3802     for (first = 0; first < ics->nr_irqs; first += alignnum) {
3803         if (num > (ics->nr_irqs - first)) {
3804             return -1;
3805         }
3806         for (i = first; i < first + num; ++i) {
3807             if (!ICS_IRQ_FREE(ics, i)) {
3808                 break;
3809             }
3810         }
3811         if (i == (first + num)) {
3812             return first;
3813         }
3814     }
3815 
3816     return -1;
3817 }
3818 
3819 /*
3820  * Allocate the IRQ number and set the IRQ type, LSI or MSI
3821  */
3822 static void spapr_irq_set_lsi(sPAPRMachineState *spapr, int irq, bool lsi)
3823 {
3824     ics_set_irq_type(spapr->ics, irq - spapr->ics->offset, lsi);
3825 }
3826 
3827 int spapr_irq_alloc(sPAPRMachineState *spapr, int irq_hint, bool lsi,
3828                     Error **errp)
3829 {
3830     ICSState *ics = spapr->ics;
3831     int irq;
3832 
3833     assert(ics);
3834 
3835     if (irq_hint) {
3836         if (!ICS_IRQ_FREE(ics, irq_hint - ics->offset)) {
3837             error_setg(errp, "can't allocate IRQ %d: already in use", irq_hint);
3838             return -1;
3839         }
3840         irq = irq_hint;
3841     } else {
3842         irq = ics_find_free_block(ics, 1, 1);
3843         if (irq < 0) {
3844             error_setg(errp, "can't allocate IRQ: no IRQ left");
3845             return -1;
3846         }
3847         irq += ics->offset;
3848     }
3849 
3850     spapr_irq_set_lsi(spapr, irq, lsi);
3851     trace_spapr_irq_alloc(irq);
3852 
3853     return irq;
3854 }
3855 
3856 /*
3857  * Allocate block of consecutive IRQs, and return the number of the first IRQ in
3858  * the block. If align==true, aligns the first IRQ number to num.
3859  */
3860 int spapr_irq_alloc_block(sPAPRMachineState *spapr, int num, bool lsi,
3861                           bool align, Error **errp)
3862 {
3863     ICSState *ics = spapr->ics;
3864     int i, first = -1;
3865 
3866     assert(ics);
3867 
3868     /*
3869      * MSIMesage::data is used for storing VIRQ so
3870      * it has to be aligned to num to support multiple
3871      * MSI vectors. MSI-X is not affected by this.
3872      * The hint is used for the first IRQ, the rest should
3873      * be allocated continuously.
3874      */
3875     if (align) {
3876         assert((num == 1) || (num == 2) || (num == 4) ||
3877                (num == 8) || (num == 16) || (num == 32));
3878         first = ics_find_free_block(ics, num, num);
3879     } else {
3880         first = ics_find_free_block(ics, num, 1);
3881     }
3882     if (first < 0) {
3883         error_setg(errp, "can't find a free %d-IRQ block", num);
3884         return -1;
3885     }
3886 
3887     first += ics->offset;
3888     for (i = first; i < first + num; ++i) {
3889         spapr_irq_set_lsi(spapr, i, lsi);
3890     }
3891 
3892     trace_spapr_irq_alloc_block(first, num, lsi, align);
3893 
3894     return first;
3895 }
3896 
3897 void spapr_irq_free(sPAPRMachineState *spapr, int irq, int num)
3898 {
3899     ICSState *ics = spapr->ics;
3900     int srcno = irq - ics->offset;
3901     int i;
3902 
3903     if (ics_valid_irq(ics, irq)) {
3904         trace_spapr_irq_free(0, irq, num);
3905         for (i = srcno; i < srcno + num; ++i) {
3906             if (ICS_IRQ_FREE(ics, i)) {
3907                 trace_spapr_irq_free_warn(0, i + ics->offset);
3908             }
3909             memset(&ics->irqs[i], 0, sizeof(ICSIRQState));
3910         }
3911     }
3912 }
3913 
3914 qemu_irq spapr_qirq(sPAPRMachineState *spapr, int irq)
3915 {
3916     ICSState *ics = spapr->ics;
3917 
3918     if (ics_valid_irq(ics, irq)) {
3919         return ics->qirqs[irq - ics->offset];
3920     }
3921 
3922     return NULL;
3923 }
3924 
3925 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3926                                  Monitor *mon)
3927 {
3928     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3929     CPUState *cs;
3930 
3931     CPU_FOREACH(cs) {
3932         PowerPCCPU *cpu = POWERPC_CPU(cs);
3933 
3934         icp_pic_print_info(ICP(cpu->intc), mon);
3935     }
3936 
3937     ics_pic_print_info(spapr->ics, mon);
3938 }
3939 
3940 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3941 {
3942     return cpu->vcpu_id;
3943 }
3944 
3945 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3946 {
3947     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3948     int vcpu_id;
3949 
3950     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3951 
3952     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3953         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3954         error_append_hint(errp, "Adjust the number of cpus to %d "
3955                           "or try to raise the number of threads per core\n",
3956                           vcpu_id * smp_threads / spapr->vsmt);
3957         return;
3958     }
3959 
3960     cpu->vcpu_id = vcpu_id;
3961 }
3962 
3963 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3964 {
3965     CPUState *cs;
3966 
3967     CPU_FOREACH(cs) {
3968         PowerPCCPU *cpu = POWERPC_CPU(cs);
3969 
3970         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3971             return cpu;
3972         }
3973     }
3974 
3975     return NULL;
3976 }
3977 
3978 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3979 {
3980     MachineClass *mc = MACHINE_CLASS(oc);
3981     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3982     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3983     NMIClass *nc = NMI_CLASS(oc);
3984     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3985     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3986     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3987     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3988 
3989     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3990 
3991     /*
3992      * We set up the default / latest behaviour here.  The class_init
3993      * functions for the specific versioned machine types can override
3994      * these details for backwards compatibility
3995      */
3996     mc->init = spapr_machine_init;
3997     mc->reset = spapr_machine_reset;
3998     mc->block_default_type = IF_SCSI;
3999     mc->max_cpus = 1024;
4000     mc->no_parallel = 1;
4001     mc->default_boot_order = "";
4002     mc->default_ram_size = 512 * M_BYTE;
4003     mc->kvm_type = spapr_kvm_type;
4004     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4005     mc->pci_allow_0_address = true;
4006     assert(!mc->get_hotplug_handler);
4007     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4008     hc->pre_plug = spapr_machine_device_pre_plug;
4009     hc->plug = spapr_machine_device_plug;
4010     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4011     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4012     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4013     hc->unplug_request = spapr_machine_device_unplug_request;
4014     hc->unplug = spapr_machine_device_unplug;
4015 
4016     smc->dr_lmb_enabled = true;
4017     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4018     mc->has_hotpluggable_cpus = true;
4019     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4020     fwc->get_dev_path = spapr_get_fw_dev_path;
4021     nc->nmi_monitor_handler = spapr_nmi;
4022     smc->phb_placement = spapr_phb_placement;
4023     vhc->hypercall = emulate_spapr_hypercall;
4024     vhc->hpt_mask = spapr_hpt_mask;
4025     vhc->map_hptes = spapr_map_hptes;
4026     vhc->unmap_hptes = spapr_unmap_hptes;
4027     vhc->store_hpte = spapr_store_hpte;
4028     vhc->get_patbe = spapr_get_patbe;
4029     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4030     xic->ics_get = spapr_ics_get;
4031     xic->ics_resend = spapr_ics_resend;
4032     xic->icp_get = spapr_icp_get;
4033     ispc->print_info = spapr_pic_print_info;
4034     /* Force NUMA node memory size to be a multiple of
4035      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4036      * in which LMBs are represented and hot-added
4037      */
4038     mc->numa_mem_align_shift = 28;
4039 
4040     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4041     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4042     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4043     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4044     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4045     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4046     spapr_caps_add_properties(smc, &error_abort);
4047 }
4048 
4049 static const TypeInfo spapr_machine_info = {
4050     .name          = TYPE_SPAPR_MACHINE,
4051     .parent        = TYPE_MACHINE,
4052     .abstract      = true,
4053     .instance_size = sizeof(sPAPRMachineState),
4054     .instance_init = spapr_instance_init,
4055     .instance_finalize = spapr_machine_finalizefn,
4056     .class_size    = sizeof(sPAPRMachineClass),
4057     .class_init    = spapr_machine_class_init,
4058     .interfaces = (InterfaceInfo[]) {
4059         { TYPE_FW_PATH_PROVIDER },
4060         { TYPE_NMI },
4061         { TYPE_HOTPLUG_HANDLER },
4062         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4063         { TYPE_XICS_FABRIC },
4064         { TYPE_INTERRUPT_STATS_PROVIDER },
4065         { }
4066     },
4067 };
4068 
4069 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4070     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4071                                                     void *data)      \
4072     {                                                                \
4073         MachineClass *mc = MACHINE_CLASS(oc);                        \
4074         spapr_machine_##suffix##_class_options(mc);                  \
4075         if (latest) {                                                \
4076             mc->alias = "pseries";                                   \
4077             mc->is_default = 1;                                      \
4078         }                                                            \
4079     }                                                                \
4080     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
4081     {                                                                \
4082         MachineState *machine = MACHINE(obj);                        \
4083         spapr_machine_##suffix##_instance_options(machine);          \
4084     }                                                                \
4085     static const TypeInfo spapr_machine_##suffix##_info = {          \
4086         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4087         .parent = TYPE_SPAPR_MACHINE,                                \
4088         .class_init = spapr_machine_##suffix##_class_init,           \
4089         .instance_init = spapr_machine_##suffix##_instance_init,     \
4090     };                                                               \
4091     static void spapr_machine_register_##suffix(void)                \
4092     {                                                                \
4093         type_register(&spapr_machine_##suffix##_info);               \
4094     }                                                                \
4095     type_init(spapr_machine_register_##suffix)
4096 
4097 /*
4098  * pseries-3.0
4099  */
4100 static void spapr_machine_3_0_instance_options(MachineState *machine)
4101 {
4102 }
4103 
4104 static void spapr_machine_3_0_class_options(MachineClass *mc)
4105 {
4106     /* Defaults for the latest behaviour inherited from the base class */
4107 }
4108 
4109 DEFINE_SPAPR_MACHINE(3_0, "3.0", true);
4110 
4111 /*
4112  * pseries-2.12
4113  */
4114 #define SPAPR_COMPAT_2_12                                              \
4115     HW_COMPAT_2_12                                                     \
4116     {                                                                  \
4117         .driver = TYPE_POWERPC_CPU,                                    \
4118         .property = "pre-3.0-migration",                              \
4119         .value    = "on",                                              \
4120     },
4121 
4122 static void spapr_machine_2_12_instance_options(MachineState *machine)
4123 {
4124     spapr_machine_3_0_instance_options(machine);
4125 }
4126 
4127 static void spapr_machine_2_12_class_options(MachineClass *mc)
4128 {
4129     spapr_machine_3_0_class_options(mc);
4130     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4131 }
4132 
4133 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4134 
4135 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
4136 {
4137     spapr_machine_2_12_instance_options(machine);
4138 }
4139 
4140 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4141 {
4142     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4143 
4144     spapr_machine_2_12_class_options(mc);
4145     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4146     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4147     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4148 }
4149 
4150 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4151 
4152 /*
4153  * pseries-2.11
4154  */
4155 #define SPAPR_COMPAT_2_11                                              \
4156     HW_COMPAT_2_11
4157 
4158 static void spapr_machine_2_11_instance_options(MachineState *machine)
4159 {
4160     spapr_machine_2_12_instance_options(machine);
4161 }
4162 
4163 static void spapr_machine_2_11_class_options(MachineClass *mc)
4164 {
4165     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4166 
4167     spapr_machine_2_12_class_options(mc);
4168     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4169     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4170 }
4171 
4172 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4173 
4174 /*
4175  * pseries-2.10
4176  */
4177 #define SPAPR_COMPAT_2_10                                              \
4178     HW_COMPAT_2_10
4179 
4180 static void spapr_machine_2_10_instance_options(MachineState *machine)
4181 {
4182     spapr_machine_2_11_instance_options(machine);
4183 }
4184 
4185 static void spapr_machine_2_10_class_options(MachineClass *mc)
4186 {
4187     spapr_machine_2_11_class_options(mc);
4188     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4189 }
4190 
4191 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4192 
4193 /*
4194  * pseries-2.9
4195  */
4196 #define SPAPR_COMPAT_2_9                                               \
4197     HW_COMPAT_2_9                                                      \
4198     {                                                                  \
4199         .driver = TYPE_POWERPC_CPU,                                    \
4200         .property = "pre-2.10-migration",                              \
4201         .value    = "on",                                              \
4202     },                                                                 \
4203 
4204 static void spapr_machine_2_9_instance_options(MachineState *machine)
4205 {
4206     spapr_machine_2_10_instance_options(machine);
4207 }
4208 
4209 static void spapr_machine_2_9_class_options(MachineClass *mc)
4210 {
4211     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4212 
4213     spapr_machine_2_10_class_options(mc);
4214     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4215     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4216     smc->pre_2_10_has_unused_icps = true;
4217     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4218 }
4219 
4220 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4221 
4222 /*
4223  * pseries-2.8
4224  */
4225 #define SPAPR_COMPAT_2_8                                        \
4226     HW_COMPAT_2_8                                               \
4227     {                                                           \
4228         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4229         .property = "pcie-extended-configuration-space",        \
4230         .value    = "off",                                      \
4231     },
4232 
4233 static void spapr_machine_2_8_instance_options(MachineState *machine)
4234 {
4235     spapr_machine_2_9_instance_options(machine);
4236 }
4237 
4238 static void spapr_machine_2_8_class_options(MachineClass *mc)
4239 {
4240     spapr_machine_2_9_class_options(mc);
4241     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4242     mc->numa_mem_align_shift = 23;
4243 }
4244 
4245 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4246 
4247 /*
4248  * pseries-2.7
4249  */
4250 #define SPAPR_COMPAT_2_7                            \
4251     HW_COMPAT_2_7                                   \
4252     {                                               \
4253         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4254         .property = "mem_win_size",                 \
4255         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4256     },                                              \
4257     {                                               \
4258         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4259         .property = "mem64_win_size",               \
4260         .value    = "0",                            \
4261     },                                              \
4262     {                                               \
4263         .driver = TYPE_POWERPC_CPU,                 \
4264         .property = "pre-2.8-migration",            \
4265         .value    = "on",                           \
4266     },                                              \
4267     {                                               \
4268         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4269         .property = "pre-2.8-migration",            \
4270         .value    = "on",                           \
4271     },
4272 
4273 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4274                               uint64_t *buid, hwaddr *pio,
4275                               hwaddr *mmio32, hwaddr *mmio64,
4276                               unsigned n_dma, uint32_t *liobns, Error **errp)
4277 {
4278     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4279     const uint64_t base_buid = 0x800000020000000ULL;
4280     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4281     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4282     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4283     const uint32_t max_index = 255;
4284     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4285 
4286     uint64_t ram_top = MACHINE(spapr)->ram_size;
4287     hwaddr phb0_base, phb_base;
4288     int i;
4289 
4290     /* Do we have device memory? */
4291     if (MACHINE(spapr)->maxram_size > ram_top) {
4292         /* Can't just use maxram_size, because there may be an
4293          * alignment gap between normal and device memory regions
4294          */
4295         ram_top = MACHINE(spapr)->device_memory->base +
4296             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4297     }
4298 
4299     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4300 
4301     if (index > max_index) {
4302         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4303                    max_index);
4304         return;
4305     }
4306 
4307     *buid = base_buid + index;
4308     for (i = 0; i < n_dma; ++i) {
4309         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4310     }
4311 
4312     phb_base = phb0_base + index * phb_spacing;
4313     *pio = phb_base + pio_offset;
4314     *mmio32 = phb_base + mmio_offset;
4315     /*
4316      * We don't set the 64-bit MMIO window, relying on the PHB's
4317      * fallback behaviour of automatically splitting a large "32-bit"
4318      * window into contiguous 32-bit and 64-bit windows
4319      */
4320 }
4321 
4322 static void spapr_machine_2_7_instance_options(MachineState *machine)
4323 {
4324     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4325 
4326     spapr_machine_2_8_instance_options(machine);
4327     spapr->use_hotplug_event_source = false;
4328 }
4329 
4330 static void spapr_machine_2_7_class_options(MachineClass *mc)
4331 {
4332     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4333 
4334     spapr_machine_2_8_class_options(mc);
4335     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4336     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4337     smc->phb_placement = phb_placement_2_7;
4338 }
4339 
4340 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4341 
4342 /*
4343  * pseries-2.6
4344  */
4345 #define SPAPR_COMPAT_2_6 \
4346     HW_COMPAT_2_6 \
4347     { \
4348         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4349         .property = "ddw",\
4350         .value    = stringify(off),\
4351     },
4352 
4353 static void spapr_machine_2_6_instance_options(MachineState *machine)
4354 {
4355     spapr_machine_2_7_instance_options(machine);
4356 }
4357 
4358 static void spapr_machine_2_6_class_options(MachineClass *mc)
4359 {
4360     spapr_machine_2_7_class_options(mc);
4361     mc->has_hotpluggable_cpus = false;
4362     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4363 }
4364 
4365 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4366 
4367 /*
4368  * pseries-2.5
4369  */
4370 #define SPAPR_COMPAT_2_5 \
4371     HW_COMPAT_2_5 \
4372     { \
4373         .driver   = "spapr-vlan", \
4374         .property = "use-rx-buffer-pools", \
4375         .value    = "off", \
4376     },
4377 
4378 static void spapr_machine_2_5_instance_options(MachineState *machine)
4379 {
4380     spapr_machine_2_6_instance_options(machine);
4381 }
4382 
4383 static void spapr_machine_2_5_class_options(MachineClass *mc)
4384 {
4385     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4386 
4387     spapr_machine_2_6_class_options(mc);
4388     smc->use_ohci_by_default = true;
4389     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4390 }
4391 
4392 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4393 
4394 /*
4395  * pseries-2.4
4396  */
4397 #define SPAPR_COMPAT_2_4 \
4398         HW_COMPAT_2_4
4399 
4400 static void spapr_machine_2_4_instance_options(MachineState *machine)
4401 {
4402     spapr_machine_2_5_instance_options(machine);
4403 }
4404 
4405 static void spapr_machine_2_4_class_options(MachineClass *mc)
4406 {
4407     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4408 
4409     spapr_machine_2_5_class_options(mc);
4410     smc->dr_lmb_enabled = false;
4411     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4412 }
4413 
4414 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4415 
4416 /*
4417  * pseries-2.3
4418  */
4419 #define SPAPR_COMPAT_2_3 \
4420         HW_COMPAT_2_3 \
4421         {\
4422             .driver   = "spapr-pci-host-bridge",\
4423             .property = "dynamic-reconfiguration",\
4424             .value    = "off",\
4425         },
4426 
4427 static void spapr_machine_2_3_instance_options(MachineState *machine)
4428 {
4429     spapr_machine_2_4_instance_options(machine);
4430 }
4431 
4432 static void spapr_machine_2_3_class_options(MachineClass *mc)
4433 {
4434     spapr_machine_2_4_class_options(mc);
4435     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4436 }
4437 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4438 
4439 /*
4440  * pseries-2.2
4441  */
4442 
4443 #define SPAPR_COMPAT_2_2 \
4444         HW_COMPAT_2_2 \
4445         {\
4446             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4447             .property = "mem_win_size",\
4448             .value    = "0x20000000",\
4449         },
4450 
4451 static void spapr_machine_2_2_instance_options(MachineState *machine)
4452 {
4453     spapr_machine_2_3_instance_options(machine);
4454     machine->suppress_vmdesc = true;
4455 }
4456 
4457 static void spapr_machine_2_2_class_options(MachineClass *mc)
4458 {
4459     spapr_machine_2_3_class_options(mc);
4460     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4461 }
4462 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4463 
4464 /*
4465  * pseries-2.1
4466  */
4467 #define SPAPR_COMPAT_2_1 \
4468         HW_COMPAT_2_1
4469 
4470 static void spapr_machine_2_1_instance_options(MachineState *machine)
4471 {
4472     spapr_machine_2_2_instance_options(machine);
4473 }
4474 
4475 static void spapr_machine_2_1_class_options(MachineClass *mc)
4476 {
4477     spapr_machine_2_2_class_options(mc);
4478     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4479 }
4480 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4481 
4482 static void spapr_machine_register_types(void)
4483 {
4484     type_register_static(&spapr_machine_info);
4485 }
4486 
4487 type_init(spapr_machine_register_types)
4488