xref: /openbmc/qemu/hw/ppc/spapr.c (revision 502d0f36)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/block-backend.h"
39 #include "sysemu/cpus.h"
40 #include "sysemu/hw_accel.h"
41 #include "kvm_ppc.h"
42 #include "migration/misc.h"
43 #include "migration/global_state.h"
44 #include "migration/register.h"
45 #include "mmu-hash64.h"
46 #include "mmu-book3s-v3.h"
47 #include "cpu-models.h"
48 #include "qom/cpu.h"
49 
50 #include "hw/boards.h"
51 #include "hw/ppc/ppc.h"
52 #include "hw/loader.h"
53 
54 #include "hw/ppc/fdt.h"
55 #include "hw/ppc/spapr.h"
56 #include "hw/ppc/spapr_vio.h"
57 #include "hw/pci-host/spapr.h"
58 #include "hw/ppc/xics.h"
59 #include "hw/pci/msi.h"
60 
61 #include "hw/pci/pci.h"
62 #include "hw/scsi/scsi.h"
63 #include "hw/virtio/virtio-scsi.h"
64 #include "hw/virtio/vhost-scsi-common.h"
65 
66 #include "exec/address-spaces.h"
67 #include "hw/usb.h"
68 #include "qemu/config-file.h"
69 #include "qemu/error-report.h"
70 #include "trace.h"
71 #include "hw/nmi.h"
72 #include "hw/intc/intc.h"
73 
74 #include "hw/compat.h"
75 #include "qemu/cutils.h"
76 #include "hw/ppc/spapr_cpu_core.h"
77 #include "qmp-commands.h"
78 
79 #include <libfdt.h>
80 
81 /* SLOF memory layout:
82  *
83  * SLOF raw image loaded at 0, copies its romfs right below the flat
84  * device-tree, then position SLOF itself 31M below that
85  *
86  * So we set FW_OVERHEAD to 40MB which should account for all of that
87  * and more
88  *
89  * We load our kernel at 4M, leaving space for SLOF initial image
90  */
91 #define FDT_MAX_SIZE            0x100000
92 #define RTAS_MAX_SIZE           0x10000
93 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
94 #define FW_MAX_SIZE             0x400000
95 #define FW_FILE_NAME            "slof.bin"
96 #define FW_OVERHEAD             0x2800000
97 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
98 
99 #define MIN_RMA_SLOF            128UL
100 
101 #define PHANDLE_XICP            0x00001111
102 
103 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
104                                   const char *type_ics,
105                                   int nr_irqs, Error **errp)
106 {
107     Error *local_err = NULL;
108     Object *obj;
109 
110     obj = object_new(type_ics);
111     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
112     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
113                                    &error_abort);
114     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
115     if (local_err) {
116         goto error;
117     }
118     object_property_set_bool(obj, true, "realized", &local_err);
119     if (local_err) {
120         goto error;
121     }
122 
123     return ICS_SIMPLE(obj);
124 
125 error:
126     error_propagate(errp, local_err);
127     return NULL;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 static inline int xics_max_server_number(void)
165 {
166     return DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(), smp_threads);
167 }
168 
169 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
170 {
171     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
172     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
173 
174     if (kvm_enabled()) {
175         if (machine_kernel_irqchip_allowed(machine) &&
176             !xics_kvm_init(spapr, errp)) {
177             spapr->icp_type = TYPE_KVM_ICP;
178             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
179         }
180         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
181             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
182             return;
183         }
184     }
185 
186     if (!spapr->ics) {
187         xics_spapr_init(spapr);
188         spapr->icp_type = TYPE_ICP;
189         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
190         if (!spapr->ics) {
191             return;
192         }
193     }
194 
195     if (smc->pre_2_10_has_unused_icps) {
196         int i;
197 
198         for (i = 0; i < xics_max_server_number(); i++) {
199             /* Dummy entries get deregistered when real ICPState objects
200              * are registered during CPU core hotplug.
201              */
202             pre_2_10_vmstate_register_dummy_icp(i);
203         }
204     }
205 }
206 
207 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
208                                   int smt_threads)
209 {
210     int i, ret = 0;
211     uint32_t servers_prop[smt_threads];
212     uint32_t gservers_prop[smt_threads * 2];
213     int index = spapr_vcpu_id(cpu);
214 
215     if (cpu->compat_pvr) {
216         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
217         if (ret < 0) {
218             return ret;
219         }
220     }
221 
222     /* Build interrupt servers and gservers properties */
223     for (i = 0; i < smt_threads; i++) {
224         servers_prop[i] = cpu_to_be32(index + i);
225         /* Hack, direct the group queues back to cpu 0 */
226         gservers_prop[i*2] = cpu_to_be32(index + i);
227         gservers_prop[i*2 + 1] = 0;
228     }
229     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
230                       servers_prop, sizeof(servers_prop));
231     if (ret < 0) {
232         return ret;
233     }
234     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
235                       gservers_prop, sizeof(gservers_prop));
236 
237     return ret;
238 }
239 
240 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
241 {
242     int index = spapr_vcpu_id(cpu);
243     uint32_t associativity[] = {cpu_to_be32(0x5),
244                                 cpu_to_be32(0x0),
245                                 cpu_to_be32(0x0),
246                                 cpu_to_be32(0x0),
247                                 cpu_to_be32(cpu->node_id),
248                                 cpu_to_be32(index)};
249 
250     /* Advertise NUMA via ibm,associativity */
251     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
252                           sizeof(associativity));
253 }
254 
255 /* Populate the "ibm,pa-features" property */
256 static void spapr_populate_pa_features(PowerPCCPU *cpu, void *fdt, int offset,
257                                        bool legacy_guest)
258 {
259     CPUPPCState *env = &cpu->env;
260     uint8_t pa_features_206[] = { 6, 0,
261         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
262     uint8_t pa_features_207[] = { 24, 0,
263         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
264         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
265         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
266         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
267     uint8_t pa_features_300[] = { 66, 0,
268         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
269         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
270         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
271         /* 6: DS207 */
272         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
273         /* 16: Vector */
274         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
275         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
276         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
277         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
278         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
279         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
280         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
281         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
282         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
283         /* 42: PM, 44: PC RA, 46: SC vec'd */
284         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
285         /* 48: SIMD, 50: QP BFP, 52: String */
286         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
287         /* 54: DecFP, 56: DecI, 58: SHA */
288         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
289         /* 60: NM atomic, 62: RNG */
290         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
291     };
292     uint8_t *pa_features = NULL;
293     size_t pa_size;
294 
295     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
296         pa_features = pa_features_206;
297         pa_size = sizeof(pa_features_206);
298     }
299     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
300         pa_features = pa_features_207;
301         pa_size = sizeof(pa_features_207);
302     }
303     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
304         pa_features = pa_features_300;
305         pa_size = sizeof(pa_features_300);
306     }
307     if (!pa_features) {
308         return;
309     }
310 
311     if (env->ci_large_pages) {
312         /*
313          * Note: we keep CI large pages off by default because a 64K capable
314          * guest provisioned with large pages might otherwise try to map a qemu
315          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
316          * even if that qemu runs on a 4k host.
317          * We dd this bit back here if we are confident this is not an issue
318          */
319         pa_features[3] |= 0x20;
320     }
321     if (kvmppc_has_cap_htm() && pa_size > 24) {
322         pa_features[24] |= 0x80;    /* Transactional memory support */
323     }
324     if (legacy_guest && pa_size > 40) {
325         /* Workaround for broken kernels that attempt (guest) radix
326          * mode when they can't handle it, if they see the radix bit set
327          * in pa-features. So hide it from them. */
328         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
329     }
330 
331     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
332 }
333 
334 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
335 {
336     int ret = 0, offset, cpus_offset;
337     CPUState *cs;
338     char cpu_model[32];
339     int smt = kvmppc_smt_threads();
340     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
341 
342     CPU_FOREACH(cs) {
343         PowerPCCPU *cpu = POWERPC_CPU(cs);
344         DeviceClass *dc = DEVICE_GET_CLASS(cs);
345         int index = spapr_vcpu_id(cpu);
346         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
347 
348         if ((index % smt) != 0) {
349             continue;
350         }
351 
352         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
353 
354         cpus_offset = fdt_path_offset(fdt, "/cpus");
355         if (cpus_offset < 0) {
356             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
357             if (cpus_offset < 0) {
358                 return cpus_offset;
359             }
360         }
361         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
362         if (offset < 0) {
363             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
364             if (offset < 0) {
365                 return offset;
366             }
367         }
368 
369         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
370                           pft_size_prop, sizeof(pft_size_prop));
371         if (ret < 0) {
372             return ret;
373         }
374 
375         if (nb_numa_nodes > 1) {
376             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
377             if (ret < 0) {
378                 return ret;
379             }
380         }
381 
382         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
383         if (ret < 0) {
384             return ret;
385         }
386 
387         spapr_populate_pa_features(cpu, fdt, offset,
388                                          spapr->cas_legacy_guest_workaround);
389     }
390     return ret;
391 }
392 
393 static hwaddr spapr_node0_size(MachineState *machine)
394 {
395     if (nb_numa_nodes) {
396         int i;
397         for (i = 0; i < nb_numa_nodes; ++i) {
398             if (numa_info[i].node_mem) {
399                 return MIN(pow2floor(numa_info[i].node_mem),
400                            machine->ram_size);
401             }
402         }
403     }
404     return machine->ram_size;
405 }
406 
407 static void add_str(GString *s, const gchar *s1)
408 {
409     g_string_append_len(s, s1, strlen(s1) + 1);
410 }
411 
412 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
413                                        hwaddr size)
414 {
415     uint32_t associativity[] = {
416         cpu_to_be32(0x4), /* length */
417         cpu_to_be32(0x0), cpu_to_be32(0x0),
418         cpu_to_be32(0x0), cpu_to_be32(nodeid)
419     };
420     char mem_name[32];
421     uint64_t mem_reg_property[2];
422     int off;
423 
424     mem_reg_property[0] = cpu_to_be64(start);
425     mem_reg_property[1] = cpu_to_be64(size);
426 
427     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
428     off = fdt_add_subnode(fdt, 0, mem_name);
429     _FDT(off);
430     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
431     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
432                       sizeof(mem_reg_property))));
433     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
434                       sizeof(associativity))));
435     return off;
436 }
437 
438 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
439 {
440     MachineState *machine = MACHINE(spapr);
441     hwaddr mem_start, node_size;
442     int i, nb_nodes = nb_numa_nodes;
443     NodeInfo *nodes = numa_info;
444     NodeInfo ramnode;
445 
446     /* No NUMA nodes, assume there is just one node with whole RAM */
447     if (!nb_numa_nodes) {
448         nb_nodes = 1;
449         ramnode.node_mem = machine->ram_size;
450         nodes = &ramnode;
451     }
452 
453     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
454         if (!nodes[i].node_mem) {
455             continue;
456         }
457         if (mem_start >= machine->ram_size) {
458             node_size = 0;
459         } else {
460             node_size = nodes[i].node_mem;
461             if (node_size > machine->ram_size - mem_start) {
462                 node_size = machine->ram_size - mem_start;
463             }
464         }
465         if (!mem_start) {
466             /* ppc_spapr_init() checks for rma_size <= node0_size already */
467             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
468             mem_start += spapr->rma_size;
469             node_size -= spapr->rma_size;
470         }
471         for ( ; node_size; ) {
472             hwaddr sizetmp = pow2floor(node_size);
473 
474             /* mem_start != 0 here */
475             if (ctzl(mem_start) < ctzl(sizetmp)) {
476                 sizetmp = 1ULL << ctzl(mem_start);
477             }
478 
479             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
480             node_size -= sizetmp;
481             mem_start += sizetmp;
482         }
483     }
484 
485     return 0;
486 }
487 
488 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
489                                   sPAPRMachineState *spapr)
490 {
491     PowerPCCPU *cpu = POWERPC_CPU(cs);
492     CPUPPCState *env = &cpu->env;
493     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
494     int index = spapr_vcpu_id(cpu);
495     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
496                        0xffffffff, 0xffffffff};
497     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
498         : SPAPR_TIMEBASE_FREQ;
499     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
500     uint32_t page_sizes_prop[64];
501     size_t page_sizes_prop_size;
502     uint32_t vcpus_per_socket = smp_threads * smp_cores;
503     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
504     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
505     sPAPRDRConnector *drc;
506     int drc_index;
507     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
508     int i;
509 
510     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
511     if (drc) {
512         drc_index = spapr_drc_index(drc);
513         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
514     }
515 
516     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
517     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
518 
519     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
520     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
521                            env->dcache_line_size)));
522     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
523                            env->dcache_line_size)));
524     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
525                            env->icache_line_size)));
526     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
527                            env->icache_line_size)));
528 
529     if (pcc->l1_dcache_size) {
530         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
531                                pcc->l1_dcache_size)));
532     } else {
533         warn_report("Unknown L1 dcache size for cpu");
534     }
535     if (pcc->l1_icache_size) {
536         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
537                                pcc->l1_icache_size)));
538     } else {
539         warn_report("Unknown L1 icache size for cpu");
540     }
541 
542     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
543     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
544     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
545     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
546     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
547     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
548 
549     if (env->spr_cb[SPR_PURR].oea_read) {
550         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
551     }
552 
553     if (env->mmu_model & POWERPC_MMU_1TSEG) {
554         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
555                           segs, sizeof(segs))));
556     }
557 
558     /* Advertise VMX/VSX (vector extensions) if available
559      *   0 / no property == no vector extensions
560      *   1               == VMX / Altivec available
561      *   2               == VSX available */
562     if (env->insns_flags & PPC_ALTIVEC) {
563         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
564 
565         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
566     }
567 
568     /* Advertise DFP (Decimal Floating Point) if available
569      *   0 / no property == no DFP
570      *   1               == DFP available */
571     if (env->insns_flags2 & PPC2_DFP) {
572         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
573     }
574 
575     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
576                                                   sizeof(page_sizes_prop));
577     if (page_sizes_prop_size) {
578         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
579                           page_sizes_prop, page_sizes_prop_size)));
580     }
581 
582     spapr_populate_pa_features(cpu, fdt, offset, false);
583 
584     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
585                            cs->cpu_index / vcpus_per_socket)));
586 
587     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
588                       pft_size_prop, sizeof(pft_size_prop))));
589 
590     if (nb_numa_nodes > 1) {
591         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
592     }
593 
594     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
595 
596     if (pcc->radix_page_info) {
597         for (i = 0; i < pcc->radix_page_info->count; i++) {
598             radix_AP_encodings[i] =
599                 cpu_to_be32(pcc->radix_page_info->entries[i]);
600         }
601         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
602                           radix_AP_encodings,
603                           pcc->radix_page_info->count *
604                           sizeof(radix_AP_encodings[0]))));
605     }
606 }
607 
608 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
609 {
610     CPUState *cs;
611     int cpus_offset;
612     char *nodename;
613     int smt = kvmppc_smt_threads();
614 
615     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
616     _FDT(cpus_offset);
617     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
618     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
619 
620     /*
621      * We walk the CPUs in reverse order to ensure that CPU DT nodes
622      * created by fdt_add_subnode() end up in the right order in FDT
623      * for the guest kernel the enumerate the CPUs correctly.
624      */
625     CPU_FOREACH_REVERSE(cs) {
626         PowerPCCPU *cpu = POWERPC_CPU(cs);
627         int index = spapr_vcpu_id(cpu);
628         DeviceClass *dc = DEVICE_GET_CLASS(cs);
629         int offset;
630 
631         if ((index % smt) != 0) {
632             continue;
633         }
634 
635         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
636         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
637         g_free(nodename);
638         _FDT(offset);
639         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
640     }
641 
642 }
643 
644 /*
645  * Adds ibm,dynamic-reconfiguration-memory node.
646  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
647  * of this device tree node.
648  */
649 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
650 {
651     MachineState *machine = MACHINE(spapr);
652     int ret, i, offset;
653     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
654     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
655     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
656     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
657                        memory_region_size(&spapr->hotplug_memory.mr)) /
658                        lmb_size;
659     uint32_t *int_buf, *cur_index, buf_len;
660     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
661 
662     /*
663      * Don't create the node if there is no hotpluggable memory
664      */
665     if (machine->ram_size == machine->maxram_size) {
666         return 0;
667     }
668 
669     /*
670      * Allocate enough buffer size to fit in ibm,dynamic-memory
671      * or ibm,associativity-lookup-arrays
672      */
673     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
674               * sizeof(uint32_t);
675     cur_index = int_buf = g_malloc0(buf_len);
676 
677     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
678 
679     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
680                     sizeof(prop_lmb_size));
681     if (ret < 0) {
682         goto out;
683     }
684 
685     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
686     if (ret < 0) {
687         goto out;
688     }
689 
690     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
691     if (ret < 0) {
692         goto out;
693     }
694 
695     /* ibm,dynamic-memory */
696     int_buf[0] = cpu_to_be32(nr_lmbs);
697     cur_index++;
698     for (i = 0; i < nr_lmbs; i++) {
699         uint64_t addr = i * lmb_size;
700         uint32_t *dynamic_memory = cur_index;
701 
702         if (i >= hotplug_lmb_start) {
703             sPAPRDRConnector *drc;
704 
705             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
706             g_assert(drc);
707 
708             dynamic_memory[0] = cpu_to_be32(addr >> 32);
709             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
710             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
711             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
712             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
713             if (memory_region_present(get_system_memory(), addr)) {
714                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
715             } else {
716                 dynamic_memory[5] = cpu_to_be32(0);
717             }
718         } else {
719             /*
720              * LMB information for RMA, boot time RAM and gap b/n RAM and
721              * hotplug memory region -- all these are marked as reserved
722              * and as having no valid DRC.
723              */
724             dynamic_memory[0] = cpu_to_be32(addr >> 32);
725             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
726             dynamic_memory[2] = cpu_to_be32(0);
727             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
728             dynamic_memory[4] = cpu_to_be32(-1);
729             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
730                                             SPAPR_LMB_FLAGS_DRC_INVALID);
731         }
732 
733         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
734     }
735     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
736     if (ret < 0) {
737         goto out;
738     }
739 
740     /* ibm,associativity-lookup-arrays */
741     cur_index = int_buf;
742     int_buf[0] = cpu_to_be32(nr_nodes);
743     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
744     cur_index += 2;
745     for (i = 0; i < nr_nodes; i++) {
746         uint32_t associativity[] = {
747             cpu_to_be32(0x0),
748             cpu_to_be32(0x0),
749             cpu_to_be32(0x0),
750             cpu_to_be32(i)
751         };
752         memcpy(cur_index, associativity, sizeof(associativity));
753         cur_index += 4;
754     }
755     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
756             (cur_index - int_buf) * sizeof(uint32_t));
757 out:
758     g_free(int_buf);
759     return ret;
760 }
761 
762 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
763                                 sPAPROptionVector *ov5_updates)
764 {
765     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
766     int ret = 0, offset;
767 
768     /* Generate ibm,dynamic-reconfiguration-memory node if required */
769     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
770         g_assert(smc->dr_lmb_enabled);
771         ret = spapr_populate_drconf_memory(spapr, fdt);
772         if (ret) {
773             goto out;
774         }
775     }
776 
777     offset = fdt_path_offset(fdt, "/chosen");
778     if (offset < 0) {
779         offset = fdt_add_subnode(fdt, 0, "chosen");
780         if (offset < 0) {
781             return offset;
782         }
783     }
784     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
785                                  "ibm,architecture-vec-5");
786 
787 out:
788     return ret;
789 }
790 
791 static bool spapr_hotplugged_dev_before_cas(void)
792 {
793     Object *drc_container, *obj;
794     ObjectProperty *prop;
795     ObjectPropertyIterator iter;
796 
797     drc_container = container_get(object_get_root(), "/dr-connector");
798     object_property_iter_init(&iter, drc_container);
799     while ((prop = object_property_iter_next(&iter))) {
800         if (!strstart(prop->type, "link<", NULL)) {
801             continue;
802         }
803         obj = object_property_get_link(drc_container, prop->name, NULL);
804         if (spapr_drc_needed(obj)) {
805             return true;
806         }
807     }
808     return false;
809 }
810 
811 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
812                                  target_ulong addr, target_ulong size,
813                                  sPAPROptionVector *ov5_updates)
814 {
815     void *fdt, *fdt_skel;
816     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
817 
818     if (spapr_hotplugged_dev_before_cas()) {
819         return 1;
820     }
821 
822     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
823         error_report("SLOF provided an unexpected CAS buffer size "
824                      TARGET_FMT_lu " (min: %zu, max: %u)",
825                      size, sizeof(hdr), FW_MAX_SIZE);
826         exit(EXIT_FAILURE);
827     }
828 
829     size -= sizeof(hdr);
830 
831     /* Create skeleton */
832     fdt_skel = g_malloc0(size);
833     _FDT((fdt_create(fdt_skel, size)));
834     _FDT((fdt_begin_node(fdt_skel, "")));
835     _FDT((fdt_end_node(fdt_skel)));
836     _FDT((fdt_finish(fdt_skel)));
837     fdt = g_malloc0(size);
838     _FDT((fdt_open_into(fdt_skel, fdt, size)));
839     g_free(fdt_skel);
840 
841     /* Fixup cpu nodes */
842     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
843 
844     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
845         return -1;
846     }
847 
848     /* Pack resulting tree */
849     _FDT((fdt_pack(fdt)));
850 
851     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
852         trace_spapr_cas_failed(size);
853         return -1;
854     }
855 
856     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
857     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
858     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
859     g_free(fdt);
860 
861     return 0;
862 }
863 
864 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
865 {
866     int rtas;
867     GString *hypertas = g_string_sized_new(256);
868     GString *qemu_hypertas = g_string_sized_new(256);
869     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
870     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
871         memory_region_size(&spapr->hotplug_memory.mr);
872     uint32_t lrdr_capacity[] = {
873         cpu_to_be32(max_hotplug_addr >> 32),
874         cpu_to_be32(max_hotplug_addr & 0xffffffff),
875         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
876         cpu_to_be32(max_cpus / smp_threads),
877     };
878 
879     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
880 
881     /* hypertas */
882     add_str(hypertas, "hcall-pft");
883     add_str(hypertas, "hcall-term");
884     add_str(hypertas, "hcall-dabr");
885     add_str(hypertas, "hcall-interrupt");
886     add_str(hypertas, "hcall-tce");
887     add_str(hypertas, "hcall-vio");
888     add_str(hypertas, "hcall-splpar");
889     add_str(hypertas, "hcall-bulk");
890     add_str(hypertas, "hcall-set-mode");
891     add_str(hypertas, "hcall-sprg0");
892     add_str(hypertas, "hcall-copy");
893     add_str(hypertas, "hcall-debug");
894     add_str(qemu_hypertas, "hcall-memop1");
895 
896     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
897         add_str(hypertas, "hcall-multi-tce");
898     }
899 
900     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
901         add_str(hypertas, "hcall-hpt-resize");
902     }
903 
904     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
905                      hypertas->str, hypertas->len));
906     g_string_free(hypertas, TRUE);
907     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
908                      qemu_hypertas->str, qemu_hypertas->len));
909     g_string_free(qemu_hypertas, TRUE);
910 
911     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
912                      refpoints, sizeof(refpoints)));
913 
914     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
915                           RTAS_ERROR_LOG_MAX));
916     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
917                           RTAS_EVENT_SCAN_RATE));
918 
919     if (msi_nonbroken) {
920         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
921     }
922 
923     /*
924      * According to PAPR, rtas ibm,os-term does not guarantee a return
925      * back to the guest cpu.
926      *
927      * While an additional ibm,extended-os-term property indicates
928      * that rtas call return will always occur. Set this property.
929      */
930     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
931 
932     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
933                      lrdr_capacity, sizeof(lrdr_capacity)));
934 
935     spapr_dt_rtas_tokens(fdt, rtas);
936 }
937 
938 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
939  * that the guest may request and thus the valid values for bytes 24..26 of
940  * option vector 5: */
941 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
942 {
943     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
944 
945     char val[2 * 4] = {
946         23, 0x00, /* Xive mode, filled in below. */
947         24, 0x00, /* Hash/Radix, filled in below. */
948         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
949         26, 0x40, /* Radix options: GTSE == yes. */
950     };
951 
952     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
953                           first_ppc_cpu->compat_pvr)) {
954         /* If we're in a pre POWER9 compat mode then the guest should do hash */
955         val[3] = 0x00; /* Hash */
956     } else if (kvm_enabled()) {
957         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
958             val[3] = 0x80; /* OV5_MMU_BOTH */
959         } else if (kvmppc_has_cap_mmu_radix()) {
960             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
961         } else {
962             val[3] = 0x00; /* Hash */
963         }
964     } else {
965         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
966         val[3] = 0xC0;
967     }
968     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
969                      val, sizeof(val)));
970 }
971 
972 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
973 {
974     MachineState *machine = MACHINE(spapr);
975     int chosen;
976     const char *boot_device = machine->boot_order;
977     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
978     size_t cb = 0;
979     char *bootlist = get_boot_devices_list(&cb, true);
980 
981     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
982 
983     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
984     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
985                           spapr->initrd_base));
986     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
987                           spapr->initrd_base + spapr->initrd_size));
988 
989     if (spapr->kernel_size) {
990         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
991                               cpu_to_be64(spapr->kernel_size) };
992 
993         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
994                          &kprop, sizeof(kprop)));
995         if (spapr->kernel_le) {
996             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
997         }
998     }
999     if (boot_menu) {
1000         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1001     }
1002     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1003     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1004     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1005 
1006     if (cb && bootlist) {
1007         int i;
1008 
1009         for (i = 0; i < cb; i++) {
1010             if (bootlist[i] == '\n') {
1011                 bootlist[i] = ' ';
1012             }
1013         }
1014         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1015     }
1016 
1017     if (boot_device && strlen(boot_device)) {
1018         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1019     }
1020 
1021     if (!spapr->has_graphics && stdout_path) {
1022         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1023     }
1024 
1025     spapr_dt_ov5_platform_support(fdt, chosen);
1026 
1027     g_free(stdout_path);
1028     g_free(bootlist);
1029 }
1030 
1031 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1032 {
1033     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1034      * KVM to work under pHyp with some guest co-operation */
1035     int hypervisor;
1036     uint8_t hypercall[16];
1037 
1038     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1039     /* indicate KVM hypercall interface */
1040     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1041     if (kvmppc_has_cap_fixup_hcalls()) {
1042         /*
1043          * Older KVM versions with older guest kernels were broken
1044          * with the magic page, don't allow the guest to map it.
1045          */
1046         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1047                                   sizeof(hypercall))) {
1048             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1049                              hypercall, sizeof(hypercall)));
1050         }
1051     }
1052 }
1053 
1054 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1055                              hwaddr rtas_addr,
1056                              hwaddr rtas_size)
1057 {
1058     MachineState *machine = MACHINE(spapr);
1059     MachineClass *mc = MACHINE_GET_CLASS(machine);
1060     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1061     int ret;
1062     void *fdt;
1063     sPAPRPHBState *phb;
1064     char *buf;
1065 
1066     fdt = g_malloc0(FDT_MAX_SIZE);
1067     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1068 
1069     /* Root node */
1070     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1071     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1072     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1073 
1074     /*
1075      * Add info to guest to indentify which host is it being run on
1076      * and what is the uuid of the guest
1077      */
1078     if (kvmppc_get_host_model(&buf)) {
1079         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1080         g_free(buf);
1081     }
1082     if (kvmppc_get_host_serial(&buf)) {
1083         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1084         g_free(buf);
1085     }
1086 
1087     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1088 
1089     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1090     if (qemu_uuid_set) {
1091         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1092     }
1093     g_free(buf);
1094 
1095     if (qemu_get_vm_name()) {
1096         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1097                                 qemu_get_vm_name()));
1098     }
1099 
1100     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1101     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1102 
1103     /* /interrupt controller */
1104     spapr_dt_xics(xics_max_server_number(), fdt, PHANDLE_XICP);
1105 
1106     ret = spapr_populate_memory(spapr, fdt);
1107     if (ret < 0) {
1108         error_report("couldn't setup memory nodes in fdt");
1109         exit(1);
1110     }
1111 
1112     /* /vdevice */
1113     spapr_dt_vdevice(spapr->vio_bus, fdt);
1114 
1115     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1116         ret = spapr_rng_populate_dt(fdt);
1117         if (ret < 0) {
1118             error_report("could not set up rng device in the fdt");
1119             exit(1);
1120         }
1121     }
1122 
1123     QLIST_FOREACH(phb, &spapr->phbs, list) {
1124         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1125         if (ret < 0) {
1126             error_report("couldn't setup PCI devices in fdt");
1127             exit(1);
1128         }
1129     }
1130 
1131     /* cpus */
1132     spapr_populate_cpus_dt_node(fdt, spapr);
1133 
1134     if (smc->dr_lmb_enabled) {
1135         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1136     }
1137 
1138     if (mc->has_hotpluggable_cpus) {
1139         int offset = fdt_path_offset(fdt, "/cpus");
1140         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1141                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1142         if (ret < 0) {
1143             error_report("Couldn't set up CPU DR device tree properties");
1144             exit(1);
1145         }
1146     }
1147 
1148     /* /event-sources */
1149     spapr_dt_events(spapr, fdt);
1150 
1151     /* /rtas */
1152     spapr_dt_rtas(spapr, fdt);
1153 
1154     /* /chosen */
1155     spapr_dt_chosen(spapr, fdt);
1156 
1157     /* /hypervisor */
1158     if (kvm_enabled()) {
1159         spapr_dt_hypervisor(spapr, fdt);
1160     }
1161 
1162     /* Build memory reserve map */
1163     if (spapr->kernel_size) {
1164         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1165     }
1166     if (spapr->initrd_size) {
1167         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1168     }
1169 
1170     /* ibm,client-architecture-support updates */
1171     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1172     if (ret < 0) {
1173         error_report("couldn't setup CAS properties fdt");
1174         exit(1);
1175     }
1176 
1177     return fdt;
1178 }
1179 
1180 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1181 {
1182     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1183 }
1184 
1185 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1186                                     PowerPCCPU *cpu)
1187 {
1188     CPUPPCState *env = &cpu->env;
1189 
1190     /* The TCG path should also be holding the BQL at this point */
1191     g_assert(qemu_mutex_iothread_locked());
1192 
1193     if (msr_pr) {
1194         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1195         env->gpr[3] = H_PRIVILEGE;
1196     } else {
1197         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1198     }
1199 }
1200 
1201 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1202 {
1203     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1204 
1205     return spapr->patb_entry;
1206 }
1207 
1208 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1209 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1210 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1211 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1212 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1213 
1214 /*
1215  * Get the fd to access the kernel htab, re-opening it if necessary
1216  */
1217 static int get_htab_fd(sPAPRMachineState *spapr)
1218 {
1219     Error *local_err = NULL;
1220 
1221     if (spapr->htab_fd >= 0) {
1222         return spapr->htab_fd;
1223     }
1224 
1225     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1226     if (spapr->htab_fd < 0) {
1227         error_report_err(local_err);
1228     }
1229 
1230     return spapr->htab_fd;
1231 }
1232 
1233 void close_htab_fd(sPAPRMachineState *spapr)
1234 {
1235     if (spapr->htab_fd >= 0) {
1236         close(spapr->htab_fd);
1237     }
1238     spapr->htab_fd = -1;
1239 }
1240 
1241 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1242 {
1243     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1244 
1245     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1246 }
1247 
1248 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1249 {
1250     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1251 
1252     assert(kvm_enabled());
1253 
1254     if (!spapr->htab) {
1255         return 0;
1256     }
1257 
1258     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1259 }
1260 
1261 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1262                                                 hwaddr ptex, int n)
1263 {
1264     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1265     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1266 
1267     if (!spapr->htab) {
1268         /*
1269          * HTAB is controlled by KVM. Fetch into temporary buffer
1270          */
1271         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1272         kvmppc_read_hptes(hptes, ptex, n);
1273         return hptes;
1274     }
1275 
1276     /*
1277      * HTAB is controlled by QEMU. Just point to the internally
1278      * accessible PTEG.
1279      */
1280     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1281 }
1282 
1283 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1284                               const ppc_hash_pte64_t *hptes,
1285                               hwaddr ptex, int n)
1286 {
1287     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1288 
1289     if (!spapr->htab) {
1290         g_free((void *)hptes);
1291     }
1292 
1293     /* Nothing to do for qemu managed HPT */
1294 }
1295 
1296 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1297                              uint64_t pte0, uint64_t pte1)
1298 {
1299     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1300     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1301 
1302     if (!spapr->htab) {
1303         kvmppc_write_hpte(ptex, pte0, pte1);
1304     } else {
1305         stq_p(spapr->htab + offset, pte0);
1306         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1307     }
1308 }
1309 
1310 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1311 {
1312     int shift;
1313 
1314     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1315      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1316      * that's much more than is needed for Linux guests */
1317     shift = ctz64(pow2ceil(ramsize)) - 7;
1318     shift = MAX(shift, 18); /* Minimum architected size */
1319     shift = MIN(shift, 46); /* Maximum architected size */
1320     return shift;
1321 }
1322 
1323 void spapr_free_hpt(sPAPRMachineState *spapr)
1324 {
1325     g_free(spapr->htab);
1326     spapr->htab = NULL;
1327     spapr->htab_shift = 0;
1328     close_htab_fd(spapr);
1329 }
1330 
1331 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1332                           Error **errp)
1333 {
1334     long rc;
1335 
1336     /* Clean up any HPT info from a previous boot */
1337     spapr_free_hpt(spapr);
1338 
1339     rc = kvmppc_reset_htab(shift);
1340     if (rc < 0) {
1341         /* kernel-side HPT needed, but couldn't allocate one */
1342         error_setg_errno(errp, errno,
1343                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1344                          shift);
1345         /* This is almost certainly fatal, but if the caller really
1346          * wants to carry on with shift == 0, it's welcome to try */
1347     } else if (rc > 0) {
1348         /* kernel-side HPT allocated */
1349         if (rc != shift) {
1350             error_setg(errp,
1351                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1352                        shift, rc);
1353         }
1354 
1355         spapr->htab_shift = shift;
1356         spapr->htab = NULL;
1357     } else {
1358         /* kernel-side HPT not needed, allocate in userspace instead */
1359         size_t size = 1ULL << shift;
1360         int i;
1361 
1362         spapr->htab = qemu_memalign(size, size);
1363         if (!spapr->htab) {
1364             error_setg_errno(errp, errno,
1365                              "Could not allocate HPT of order %d", shift);
1366             return;
1367         }
1368 
1369         memset(spapr->htab, 0, size);
1370         spapr->htab_shift = shift;
1371 
1372         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1373             DIRTY_HPTE(HPTE(spapr->htab, i));
1374         }
1375     }
1376     /* We're setting up a hash table, so that means we're not radix */
1377     spapr->patb_entry = 0;
1378 }
1379 
1380 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1381 {
1382     int hpt_shift;
1383 
1384     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1385         || (spapr->cas_reboot
1386             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1387         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1388     } else {
1389         uint64_t current_ram_size;
1390 
1391         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1392         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1393     }
1394     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1395 
1396     if (spapr->vrma_adjust) {
1397         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1398                                           spapr->htab_shift);
1399     }
1400 }
1401 
1402 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1403 {
1404     bool matched = false;
1405 
1406     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1407         matched = true;
1408     }
1409 
1410     if (!matched) {
1411         error_report("Device %s is not supported by this machine yet.",
1412                      qdev_fw_name(DEVICE(sbdev)));
1413         exit(1);
1414     }
1415 }
1416 
1417 static int spapr_reset_drcs(Object *child, void *opaque)
1418 {
1419     sPAPRDRConnector *drc =
1420         (sPAPRDRConnector *) object_dynamic_cast(child,
1421                                                  TYPE_SPAPR_DR_CONNECTOR);
1422 
1423     if (drc) {
1424         spapr_drc_reset(drc);
1425     }
1426 
1427     return 0;
1428 }
1429 
1430 static void ppc_spapr_reset(void)
1431 {
1432     MachineState *machine = MACHINE(qdev_get_machine());
1433     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1434     PowerPCCPU *first_ppc_cpu;
1435     uint32_t rtas_limit;
1436     hwaddr rtas_addr, fdt_addr;
1437     void *fdt;
1438     int rc;
1439 
1440     /* Check for unknown sysbus devices */
1441     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1442 
1443     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1444         /* If using KVM with radix mode available, VCPUs can be started
1445          * without a HPT because KVM will start them in radix mode.
1446          * Set the GR bit in PATB so that we know there is no HPT. */
1447         spapr->patb_entry = PATBE1_GR;
1448     } else {
1449         spapr_setup_hpt_and_vrma(spapr);
1450     }
1451 
1452     qemu_devices_reset();
1453 
1454     /* DRC reset may cause a device to be unplugged. This will cause troubles
1455      * if this device is used by another device (eg, a running vhost backend
1456      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1457      * situations, we reset DRCs after all devices have been reset.
1458      */
1459     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1460 
1461     spapr_clear_pending_events(spapr);
1462 
1463     /*
1464      * We place the device tree and RTAS just below either the top of the RMA,
1465      * or just below 2GB, whichever is lowere, so that it can be
1466      * processed with 32-bit real mode code if necessary
1467      */
1468     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1469     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1470     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1471 
1472     /* if this reset wasn't generated by CAS, we should reset our
1473      * negotiated options and start from scratch */
1474     if (!spapr->cas_reboot) {
1475         spapr_ovec_cleanup(spapr->ov5_cas);
1476         spapr->ov5_cas = spapr_ovec_new();
1477 
1478         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1479     }
1480 
1481     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1482 
1483     spapr_load_rtas(spapr, fdt, rtas_addr);
1484 
1485     rc = fdt_pack(fdt);
1486 
1487     /* Should only fail if we've built a corrupted tree */
1488     assert(rc == 0);
1489 
1490     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1491         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1492                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1493         exit(1);
1494     }
1495 
1496     /* Load the fdt */
1497     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1498     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1499     g_free(fdt);
1500 
1501     /* Set up the entry state */
1502     first_ppc_cpu = POWERPC_CPU(first_cpu);
1503     first_ppc_cpu->env.gpr[3] = fdt_addr;
1504     first_ppc_cpu->env.gpr[5] = 0;
1505     first_cpu->halted = 0;
1506     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1507 
1508     spapr->cas_reboot = false;
1509 }
1510 
1511 static void spapr_create_nvram(sPAPRMachineState *spapr)
1512 {
1513     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1514     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1515 
1516     if (dinfo) {
1517         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1518                             &error_fatal);
1519     }
1520 
1521     qdev_init_nofail(dev);
1522 
1523     spapr->nvram = (struct sPAPRNVRAM *)dev;
1524 }
1525 
1526 static void spapr_rtc_create(sPAPRMachineState *spapr)
1527 {
1528     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1529     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1530                               &error_fatal);
1531     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1532                               &error_fatal);
1533     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1534                               "date", &error_fatal);
1535 }
1536 
1537 /* Returns whether we want to use VGA or not */
1538 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1539 {
1540     switch (vga_interface_type) {
1541     case VGA_NONE:
1542         return false;
1543     case VGA_DEVICE:
1544         return true;
1545     case VGA_STD:
1546     case VGA_VIRTIO:
1547         return pci_vga_init(pci_bus) != NULL;
1548     default:
1549         error_setg(errp,
1550                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1551         return false;
1552     }
1553 }
1554 
1555 static int spapr_post_load(void *opaque, int version_id)
1556 {
1557     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1558     int err = 0;
1559 
1560     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1561         CPUState *cs;
1562         CPU_FOREACH(cs) {
1563             PowerPCCPU *cpu = POWERPC_CPU(cs);
1564             icp_resend(ICP(cpu->intc));
1565         }
1566     }
1567 
1568     /* In earlier versions, there was no separate qdev for the PAPR
1569      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1570      * So when migrating from those versions, poke the incoming offset
1571      * value into the RTC device */
1572     if (version_id < 3) {
1573         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1574     }
1575 
1576     if (kvm_enabled() && spapr->patb_entry) {
1577         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1578         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1579         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1580 
1581         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1582         if (err) {
1583             error_report("Process table config unsupported by the host");
1584             return -EINVAL;
1585         }
1586     }
1587 
1588     return err;
1589 }
1590 
1591 static bool version_before_3(void *opaque, int version_id)
1592 {
1593     return version_id < 3;
1594 }
1595 
1596 static bool spapr_pending_events_needed(void *opaque)
1597 {
1598     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1599     return !QTAILQ_EMPTY(&spapr->pending_events);
1600 }
1601 
1602 static const VMStateDescription vmstate_spapr_event_entry = {
1603     .name = "spapr_event_log_entry",
1604     .version_id = 1,
1605     .minimum_version_id = 1,
1606     .fields = (VMStateField[]) {
1607         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1608         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1609         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1610                                      NULL, extended_length),
1611         VMSTATE_END_OF_LIST()
1612     },
1613 };
1614 
1615 static const VMStateDescription vmstate_spapr_pending_events = {
1616     .name = "spapr_pending_events",
1617     .version_id = 1,
1618     .minimum_version_id = 1,
1619     .needed = spapr_pending_events_needed,
1620     .fields = (VMStateField[]) {
1621         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1622                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1623         VMSTATE_END_OF_LIST()
1624     },
1625 };
1626 
1627 static bool spapr_ov5_cas_needed(void *opaque)
1628 {
1629     sPAPRMachineState *spapr = opaque;
1630     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1631     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1632     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1633     bool cas_needed;
1634 
1635     /* Prior to the introduction of sPAPROptionVector, we had two option
1636      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1637      * Both of these options encode machine topology into the device-tree
1638      * in such a way that the now-booted OS should still be able to interact
1639      * appropriately with QEMU regardless of what options were actually
1640      * negotiatied on the source side.
1641      *
1642      * As such, we can avoid migrating the CAS-negotiated options if these
1643      * are the only options available on the current machine/platform.
1644      * Since these are the only options available for pseries-2.7 and
1645      * earlier, this allows us to maintain old->new/new->old migration
1646      * compatibility.
1647      *
1648      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1649      * via default pseries-2.8 machines and explicit command-line parameters.
1650      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1651      * of the actual CAS-negotiated values to continue working properly. For
1652      * example, availability of memory unplug depends on knowing whether
1653      * OV5_HP_EVT was negotiated via CAS.
1654      *
1655      * Thus, for any cases where the set of available CAS-negotiatable
1656      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1657      * include the CAS-negotiated options in the migration stream.
1658      */
1659     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1660     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1661 
1662     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1663      * the mask itself since in the future it's possible "legacy" bits may be
1664      * removed via machine options, which could generate a false positive
1665      * that breaks migration.
1666      */
1667     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1668     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1669 
1670     spapr_ovec_cleanup(ov5_mask);
1671     spapr_ovec_cleanup(ov5_legacy);
1672     spapr_ovec_cleanup(ov5_removed);
1673 
1674     return cas_needed;
1675 }
1676 
1677 static const VMStateDescription vmstate_spapr_ov5_cas = {
1678     .name = "spapr_option_vector_ov5_cas",
1679     .version_id = 1,
1680     .minimum_version_id = 1,
1681     .needed = spapr_ov5_cas_needed,
1682     .fields = (VMStateField[]) {
1683         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1684                                  vmstate_spapr_ovec, sPAPROptionVector),
1685         VMSTATE_END_OF_LIST()
1686     },
1687 };
1688 
1689 static bool spapr_patb_entry_needed(void *opaque)
1690 {
1691     sPAPRMachineState *spapr = opaque;
1692 
1693     return !!spapr->patb_entry;
1694 }
1695 
1696 static const VMStateDescription vmstate_spapr_patb_entry = {
1697     .name = "spapr_patb_entry",
1698     .version_id = 1,
1699     .minimum_version_id = 1,
1700     .needed = spapr_patb_entry_needed,
1701     .fields = (VMStateField[]) {
1702         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1703         VMSTATE_END_OF_LIST()
1704     },
1705 };
1706 
1707 static const VMStateDescription vmstate_spapr = {
1708     .name = "spapr",
1709     .version_id = 3,
1710     .minimum_version_id = 1,
1711     .post_load = spapr_post_load,
1712     .fields = (VMStateField[]) {
1713         /* used to be @next_irq */
1714         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1715 
1716         /* RTC offset */
1717         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1718 
1719         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1720         VMSTATE_END_OF_LIST()
1721     },
1722     .subsections = (const VMStateDescription*[]) {
1723         &vmstate_spapr_ov5_cas,
1724         &vmstate_spapr_patb_entry,
1725         &vmstate_spapr_pending_events,
1726         NULL
1727     }
1728 };
1729 
1730 static int htab_save_setup(QEMUFile *f, void *opaque)
1731 {
1732     sPAPRMachineState *spapr = opaque;
1733 
1734     /* "Iteration" header */
1735     if (!spapr->htab_shift) {
1736         qemu_put_be32(f, -1);
1737     } else {
1738         qemu_put_be32(f, spapr->htab_shift);
1739     }
1740 
1741     if (spapr->htab) {
1742         spapr->htab_save_index = 0;
1743         spapr->htab_first_pass = true;
1744     } else {
1745         if (spapr->htab_shift) {
1746             assert(kvm_enabled());
1747         }
1748     }
1749 
1750 
1751     return 0;
1752 }
1753 
1754 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1755                             int chunkstart, int n_valid, int n_invalid)
1756 {
1757     qemu_put_be32(f, chunkstart);
1758     qemu_put_be16(f, n_valid);
1759     qemu_put_be16(f, n_invalid);
1760     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1761                     HASH_PTE_SIZE_64 * n_valid);
1762 }
1763 
1764 static void htab_save_end_marker(QEMUFile *f)
1765 {
1766     qemu_put_be32(f, 0);
1767     qemu_put_be16(f, 0);
1768     qemu_put_be16(f, 0);
1769 }
1770 
1771 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1772                                  int64_t max_ns)
1773 {
1774     bool has_timeout = max_ns != -1;
1775     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1776     int index = spapr->htab_save_index;
1777     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1778 
1779     assert(spapr->htab_first_pass);
1780 
1781     do {
1782         int chunkstart;
1783 
1784         /* Consume invalid HPTEs */
1785         while ((index < htabslots)
1786                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1787             CLEAN_HPTE(HPTE(spapr->htab, index));
1788             index++;
1789         }
1790 
1791         /* Consume valid HPTEs */
1792         chunkstart = index;
1793         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1794                && HPTE_VALID(HPTE(spapr->htab, index))) {
1795             CLEAN_HPTE(HPTE(spapr->htab, index));
1796             index++;
1797         }
1798 
1799         if (index > chunkstart) {
1800             int n_valid = index - chunkstart;
1801 
1802             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
1803 
1804             if (has_timeout &&
1805                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1806                 break;
1807             }
1808         }
1809     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1810 
1811     if (index >= htabslots) {
1812         assert(index == htabslots);
1813         index = 0;
1814         spapr->htab_first_pass = false;
1815     }
1816     spapr->htab_save_index = index;
1817 }
1818 
1819 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1820                                 int64_t max_ns)
1821 {
1822     bool final = max_ns < 0;
1823     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1824     int examined = 0, sent = 0;
1825     int index = spapr->htab_save_index;
1826     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1827 
1828     assert(!spapr->htab_first_pass);
1829 
1830     do {
1831         int chunkstart, invalidstart;
1832 
1833         /* Consume non-dirty HPTEs */
1834         while ((index < htabslots)
1835                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1836             index++;
1837             examined++;
1838         }
1839 
1840         chunkstart = index;
1841         /* Consume valid dirty HPTEs */
1842         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1843                && HPTE_DIRTY(HPTE(spapr->htab, index))
1844                && HPTE_VALID(HPTE(spapr->htab, index))) {
1845             CLEAN_HPTE(HPTE(spapr->htab, index));
1846             index++;
1847             examined++;
1848         }
1849 
1850         invalidstart = index;
1851         /* Consume invalid dirty HPTEs */
1852         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1853                && HPTE_DIRTY(HPTE(spapr->htab, index))
1854                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1855             CLEAN_HPTE(HPTE(spapr->htab, index));
1856             index++;
1857             examined++;
1858         }
1859 
1860         if (index > chunkstart) {
1861             int n_valid = invalidstart - chunkstart;
1862             int n_invalid = index - invalidstart;
1863 
1864             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
1865             sent += index - chunkstart;
1866 
1867             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1868                 break;
1869             }
1870         }
1871 
1872         if (examined >= htabslots) {
1873             break;
1874         }
1875 
1876         if (index >= htabslots) {
1877             assert(index == htabslots);
1878             index = 0;
1879         }
1880     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1881 
1882     if (index >= htabslots) {
1883         assert(index == htabslots);
1884         index = 0;
1885     }
1886 
1887     spapr->htab_save_index = index;
1888 
1889     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1890 }
1891 
1892 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1893 #define MAX_KVM_BUF_SIZE    2048
1894 
1895 static int htab_save_iterate(QEMUFile *f, void *opaque)
1896 {
1897     sPAPRMachineState *spapr = opaque;
1898     int fd;
1899     int rc = 0;
1900 
1901     /* Iteration header */
1902     if (!spapr->htab_shift) {
1903         qemu_put_be32(f, -1);
1904         return 1;
1905     } else {
1906         qemu_put_be32(f, 0);
1907     }
1908 
1909     if (!spapr->htab) {
1910         assert(kvm_enabled());
1911 
1912         fd = get_htab_fd(spapr);
1913         if (fd < 0) {
1914             return fd;
1915         }
1916 
1917         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1918         if (rc < 0) {
1919             return rc;
1920         }
1921     } else  if (spapr->htab_first_pass) {
1922         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1923     } else {
1924         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1925     }
1926 
1927     htab_save_end_marker(f);
1928 
1929     return rc;
1930 }
1931 
1932 static int htab_save_complete(QEMUFile *f, void *opaque)
1933 {
1934     sPAPRMachineState *spapr = opaque;
1935     int fd;
1936 
1937     /* Iteration header */
1938     if (!spapr->htab_shift) {
1939         qemu_put_be32(f, -1);
1940         return 0;
1941     } else {
1942         qemu_put_be32(f, 0);
1943     }
1944 
1945     if (!spapr->htab) {
1946         int rc;
1947 
1948         assert(kvm_enabled());
1949 
1950         fd = get_htab_fd(spapr);
1951         if (fd < 0) {
1952             return fd;
1953         }
1954 
1955         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1956         if (rc < 0) {
1957             return rc;
1958         }
1959     } else {
1960         if (spapr->htab_first_pass) {
1961             htab_save_first_pass(f, spapr, -1);
1962         }
1963         htab_save_later_pass(f, spapr, -1);
1964     }
1965 
1966     /* End marker */
1967     htab_save_end_marker(f);
1968 
1969     return 0;
1970 }
1971 
1972 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1973 {
1974     sPAPRMachineState *spapr = opaque;
1975     uint32_t section_hdr;
1976     int fd = -1;
1977     Error *local_err = NULL;
1978 
1979     if (version_id < 1 || version_id > 1) {
1980         error_report("htab_load() bad version");
1981         return -EINVAL;
1982     }
1983 
1984     section_hdr = qemu_get_be32(f);
1985 
1986     if (section_hdr == -1) {
1987         spapr_free_hpt(spapr);
1988         return 0;
1989     }
1990 
1991     if (section_hdr) {
1992         /* First section gives the htab size */
1993         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1994         if (local_err) {
1995             error_report_err(local_err);
1996             return -EINVAL;
1997         }
1998         return 0;
1999     }
2000 
2001     if (!spapr->htab) {
2002         assert(kvm_enabled());
2003 
2004         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2005         if (fd < 0) {
2006             error_report_err(local_err);
2007             return fd;
2008         }
2009     }
2010 
2011     while (true) {
2012         uint32_t index;
2013         uint16_t n_valid, n_invalid;
2014 
2015         index = qemu_get_be32(f);
2016         n_valid = qemu_get_be16(f);
2017         n_invalid = qemu_get_be16(f);
2018 
2019         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2020             /* End of Stream */
2021             break;
2022         }
2023 
2024         if ((index + n_valid + n_invalid) >
2025             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2026             /* Bad index in stream */
2027             error_report(
2028                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2029                 index, n_valid, n_invalid, spapr->htab_shift);
2030             return -EINVAL;
2031         }
2032 
2033         if (spapr->htab) {
2034             if (n_valid) {
2035                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2036                                 HASH_PTE_SIZE_64 * n_valid);
2037             }
2038             if (n_invalid) {
2039                 memset(HPTE(spapr->htab, index + n_valid), 0,
2040                        HASH_PTE_SIZE_64 * n_invalid);
2041             }
2042         } else {
2043             int rc;
2044 
2045             assert(fd >= 0);
2046 
2047             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2048             if (rc < 0) {
2049                 return rc;
2050             }
2051         }
2052     }
2053 
2054     if (!spapr->htab) {
2055         assert(fd >= 0);
2056         close(fd);
2057     }
2058 
2059     return 0;
2060 }
2061 
2062 static void htab_save_cleanup(void *opaque)
2063 {
2064     sPAPRMachineState *spapr = opaque;
2065 
2066     close_htab_fd(spapr);
2067 }
2068 
2069 static SaveVMHandlers savevm_htab_handlers = {
2070     .save_setup = htab_save_setup,
2071     .save_live_iterate = htab_save_iterate,
2072     .save_live_complete_precopy = htab_save_complete,
2073     .save_cleanup = htab_save_cleanup,
2074     .load_state = htab_load,
2075 };
2076 
2077 static void spapr_boot_set(void *opaque, const char *boot_device,
2078                            Error **errp)
2079 {
2080     MachineState *machine = MACHINE(opaque);
2081     machine->boot_order = g_strdup(boot_device);
2082 }
2083 
2084 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2085 {
2086     MachineState *machine = MACHINE(spapr);
2087     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2088     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2089     int i;
2090 
2091     for (i = 0; i < nr_lmbs; i++) {
2092         uint64_t addr;
2093 
2094         addr = i * lmb_size + spapr->hotplug_memory.base;
2095         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2096                                addr / lmb_size);
2097     }
2098 }
2099 
2100 /*
2101  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2102  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2103  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2104  */
2105 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2106 {
2107     int i;
2108 
2109     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2110         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2111                    " is not aligned to %llu MiB",
2112                    machine->ram_size,
2113                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2114         return;
2115     }
2116 
2117     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2118         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2119                    " is not aligned to %llu MiB",
2120                    machine->ram_size,
2121                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2122         return;
2123     }
2124 
2125     for (i = 0; i < nb_numa_nodes; i++) {
2126         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2127             error_setg(errp,
2128                        "Node %d memory size 0x%" PRIx64
2129                        " is not aligned to %llu MiB",
2130                        i, numa_info[i].node_mem,
2131                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2132             return;
2133         }
2134     }
2135 }
2136 
2137 /* find cpu slot in machine->possible_cpus by core_id */
2138 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2139 {
2140     int index = id / smp_threads;
2141 
2142     if (index >= ms->possible_cpus->len) {
2143         return NULL;
2144     }
2145     if (idx) {
2146         *idx = index;
2147     }
2148     return &ms->possible_cpus->cpus[index];
2149 }
2150 
2151 static void spapr_init_cpus(sPAPRMachineState *spapr)
2152 {
2153     MachineState *machine = MACHINE(spapr);
2154     MachineClass *mc = MACHINE_GET_CLASS(machine);
2155     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2156     int smt = kvmppc_smt_threads();
2157     const CPUArchIdList *possible_cpus;
2158     int boot_cores_nr = smp_cpus / smp_threads;
2159     int i;
2160 
2161     if (!type) {
2162         error_report("Unable to find sPAPR CPU Core definition");
2163         exit(1);
2164     }
2165 
2166     possible_cpus = mc->possible_cpu_arch_ids(machine);
2167     if (mc->has_hotpluggable_cpus) {
2168         if (smp_cpus % smp_threads) {
2169             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2170                          smp_cpus, smp_threads);
2171             exit(1);
2172         }
2173         if (max_cpus % smp_threads) {
2174             error_report("max_cpus (%u) must be multiple of threads (%u)",
2175                          max_cpus, smp_threads);
2176             exit(1);
2177         }
2178     } else {
2179         if (max_cpus != smp_cpus) {
2180             error_report("This machine version does not support CPU hotplug");
2181             exit(1);
2182         }
2183         boot_cores_nr = possible_cpus->len;
2184     }
2185 
2186     for (i = 0; i < possible_cpus->len; i++) {
2187         int core_id = i * smp_threads;
2188 
2189         if (mc->has_hotpluggable_cpus) {
2190             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2191                                    (core_id / smp_threads) * smt);
2192         }
2193 
2194         if (i < boot_cores_nr) {
2195             Object *core  = object_new(type);
2196             int nr_threads = smp_threads;
2197 
2198             /* Handle the partially filled core for older machine types */
2199             if ((i + 1) * smp_threads >= smp_cpus) {
2200                 nr_threads = smp_cpus - i * smp_threads;
2201             }
2202 
2203             object_property_set_int(core, nr_threads, "nr-threads",
2204                                     &error_fatal);
2205             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2206                                     &error_fatal);
2207             object_property_set_bool(core, true, "realized", &error_fatal);
2208         }
2209     }
2210 }
2211 
2212 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2213 {
2214     Error *local_err = NULL;
2215     bool vsmt_user = !!spapr->vsmt;
2216     int kvm_smt = kvmppc_smt_threads();
2217     int ret;
2218 
2219     if (!kvm_enabled() && (smp_threads > 1)) {
2220         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2221                      "on a pseries machine");
2222         goto out;
2223     }
2224     if (!is_power_of_2(smp_threads)) {
2225         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2226                      "machine because it must be a power of 2", smp_threads);
2227         goto out;
2228     }
2229 
2230     /* Detemine the VSMT mode to use: */
2231     if (vsmt_user) {
2232         if (spapr->vsmt < smp_threads) {
2233             error_setg(&local_err, "Cannot support VSMT mode %d"
2234                          " because it must be >= threads/core (%d)",
2235                          spapr->vsmt, smp_threads);
2236             goto out;
2237         }
2238         /* In this case, spapr->vsmt has been set by the command line */
2239     } else {
2240         /* Choose a VSMT mode that may be higher than necessary but is
2241          * likely to be compatible with hosts that don't have VSMT. */
2242         spapr->vsmt = MAX(kvm_smt, smp_threads);
2243     }
2244 
2245     /* KVM: If necessary, set the SMT mode: */
2246     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2247         ret = kvmppc_set_smt_threads(spapr->vsmt);
2248         if (ret) {
2249             error_setg(&local_err,
2250                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2251                        spapr->vsmt, ret);
2252             if (!vsmt_user) {
2253                 error_append_hint(&local_err, "On PPC, a VM with %d threads/"
2254                              "core on a host with %d threads/core requires "
2255                              " the use of VSMT mode %d.\n",
2256                              smp_threads, kvm_smt, spapr->vsmt);
2257             }
2258             kvmppc_hint_smt_possible(&local_err);
2259             goto out;
2260         }
2261     }
2262     /* else TCG: nothing to do currently */
2263 out:
2264     error_propagate(errp, local_err);
2265 }
2266 
2267 /* pSeries LPAR / sPAPR hardware init */
2268 static void ppc_spapr_init(MachineState *machine)
2269 {
2270     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2271     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2272     const char *kernel_filename = machine->kernel_filename;
2273     const char *initrd_filename = machine->initrd_filename;
2274     PCIHostState *phb;
2275     int i;
2276     MemoryRegion *sysmem = get_system_memory();
2277     MemoryRegion *ram = g_new(MemoryRegion, 1);
2278     MemoryRegion *rma_region;
2279     void *rma = NULL;
2280     hwaddr rma_alloc_size;
2281     hwaddr node0_size = spapr_node0_size(machine);
2282     long load_limit, fw_size;
2283     char *filename;
2284     Error *resize_hpt_err = NULL;
2285 
2286     msi_nonbroken = true;
2287 
2288     QLIST_INIT(&spapr->phbs);
2289     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2290 
2291     /* Check HPT resizing availability */
2292     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2293     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2294         /*
2295          * If the user explicitly requested a mode we should either
2296          * supply it, or fail completely (which we do below).  But if
2297          * it's not set explicitly, we reset our mode to something
2298          * that works
2299          */
2300         if (resize_hpt_err) {
2301             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2302             error_free(resize_hpt_err);
2303             resize_hpt_err = NULL;
2304         } else {
2305             spapr->resize_hpt = smc->resize_hpt_default;
2306         }
2307     }
2308 
2309     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2310 
2311     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2312         /*
2313          * User requested HPT resize, but this host can't supply it.  Bail out
2314          */
2315         error_report_err(resize_hpt_err);
2316         exit(1);
2317     }
2318 
2319     /* Allocate RMA if necessary */
2320     rma_alloc_size = kvmppc_alloc_rma(&rma);
2321 
2322     if (rma_alloc_size == -1) {
2323         error_report("Unable to create RMA");
2324         exit(1);
2325     }
2326 
2327     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2328         spapr->rma_size = rma_alloc_size;
2329     } else {
2330         spapr->rma_size = node0_size;
2331 
2332         /* With KVM, we don't actually know whether KVM supports an
2333          * unbounded RMA (PR KVM) or is limited by the hash table size
2334          * (HV KVM using VRMA), so we always assume the latter
2335          *
2336          * In that case, we also limit the initial allocations for RTAS
2337          * etc... to 256M since we have no way to know what the VRMA size
2338          * is going to be as it depends on the size of the hash table
2339          * isn't determined yet.
2340          */
2341         if (kvm_enabled()) {
2342             spapr->vrma_adjust = 1;
2343             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2344         }
2345 
2346         /* Actually we don't support unbounded RMA anymore since we
2347          * added proper emulation of HV mode. The max we can get is
2348          * 16G which also happens to be what we configure for PAPR
2349          * mode so make sure we don't do anything bigger than that
2350          */
2351         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2352     }
2353 
2354     if (spapr->rma_size > node0_size) {
2355         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2356                      spapr->rma_size);
2357         exit(1);
2358     }
2359 
2360     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2361     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2362 
2363     /* Set up Interrupt Controller before we create the VCPUs */
2364     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2365 
2366     /* Set up containers for ibm,client-architecture-support negotiated options
2367      */
2368     spapr->ov5 = spapr_ovec_new();
2369     spapr->ov5_cas = spapr_ovec_new();
2370 
2371     if (smc->dr_lmb_enabled) {
2372         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2373         spapr_validate_node_memory(machine, &error_fatal);
2374     }
2375 
2376     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2377     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2378         /* KVM and TCG always allow GTSE with radix... */
2379         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2380     }
2381     /* ... but not with hash (currently). */
2382 
2383     /* advertise support for dedicated HP event source to guests */
2384     if (spapr->use_hotplug_event_source) {
2385         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2386     }
2387 
2388     /* advertise support for HPT resizing */
2389     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2390         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2391     }
2392 
2393     /* init CPUs */
2394     spapr_set_vsmt_mode(spapr, &error_fatal);
2395 
2396     spapr_init_cpus(spapr);
2397 
2398     if (kvm_enabled()) {
2399         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2400         kvmppc_enable_logical_ci_hcalls();
2401         kvmppc_enable_set_mode_hcall();
2402 
2403         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2404         kvmppc_enable_clear_ref_mod_hcalls();
2405     }
2406 
2407     /* allocate RAM */
2408     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2409                                          machine->ram_size);
2410     memory_region_add_subregion(sysmem, 0, ram);
2411 
2412     if (rma_alloc_size && rma) {
2413         rma_region = g_new(MemoryRegion, 1);
2414         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2415                                    rma_alloc_size, rma);
2416         vmstate_register_ram_global(rma_region);
2417         memory_region_add_subregion(sysmem, 0, rma_region);
2418     }
2419 
2420     /* initialize hotplug memory address space */
2421     if (machine->ram_size < machine->maxram_size) {
2422         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2423         /*
2424          * Limit the number of hotpluggable memory slots to half the number
2425          * slots that KVM supports, leaving the other half for PCI and other
2426          * devices. However ensure that number of slots doesn't drop below 32.
2427          */
2428         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2429                            SPAPR_MAX_RAM_SLOTS;
2430 
2431         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2432             max_memslots = SPAPR_MAX_RAM_SLOTS;
2433         }
2434         if (machine->ram_slots > max_memslots) {
2435             error_report("Specified number of memory slots %"
2436                          PRIu64" exceeds max supported %d",
2437                          machine->ram_slots, max_memslots);
2438             exit(1);
2439         }
2440 
2441         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2442                                               SPAPR_HOTPLUG_MEM_ALIGN);
2443         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2444                            "hotplug-memory", hotplug_mem_size);
2445         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2446                                     &spapr->hotplug_memory.mr);
2447     }
2448 
2449     if (smc->dr_lmb_enabled) {
2450         spapr_create_lmb_dr_connectors(spapr);
2451     }
2452 
2453     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2454     if (!filename) {
2455         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2456         exit(1);
2457     }
2458     spapr->rtas_size = get_image_size(filename);
2459     if (spapr->rtas_size < 0) {
2460         error_report("Could not get size of LPAR rtas '%s'", filename);
2461         exit(1);
2462     }
2463     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2464     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2465         error_report("Could not load LPAR rtas '%s'", filename);
2466         exit(1);
2467     }
2468     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2469         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2470                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2471         exit(1);
2472     }
2473     g_free(filename);
2474 
2475     /* Set up RTAS event infrastructure */
2476     spapr_events_init(spapr);
2477 
2478     /* Set up the RTC RTAS interfaces */
2479     spapr_rtc_create(spapr);
2480 
2481     /* Set up VIO bus */
2482     spapr->vio_bus = spapr_vio_bus_init();
2483 
2484     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2485         if (serial_hds[i]) {
2486             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2487         }
2488     }
2489 
2490     /* We always have at least the nvram device on VIO */
2491     spapr_create_nvram(spapr);
2492 
2493     /* Set up PCI */
2494     spapr_pci_rtas_init();
2495 
2496     phb = spapr_create_phb(spapr, 0);
2497 
2498     for (i = 0; i < nb_nics; i++) {
2499         NICInfo *nd = &nd_table[i];
2500 
2501         if (!nd->model) {
2502             nd->model = g_strdup("ibmveth");
2503         }
2504 
2505         if (strcmp(nd->model, "ibmveth") == 0) {
2506             spapr_vlan_create(spapr->vio_bus, nd);
2507         } else {
2508             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2509         }
2510     }
2511 
2512     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2513         spapr_vscsi_create(spapr->vio_bus);
2514     }
2515 
2516     /* Graphics */
2517     if (spapr_vga_init(phb->bus, &error_fatal)) {
2518         spapr->has_graphics = true;
2519         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2520     }
2521 
2522     if (machine->usb) {
2523         if (smc->use_ohci_by_default) {
2524             pci_create_simple(phb->bus, -1, "pci-ohci");
2525         } else {
2526             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2527         }
2528 
2529         if (spapr->has_graphics) {
2530             USBBus *usb_bus = usb_bus_find(-1);
2531 
2532             usb_create_simple(usb_bus, "usb-kbd");
2533             usb_create_simple(usb_bus, "usb-mouse");
2534         }
2535     }
2536 
2537     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2538         error_report(
2539             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2540             MIN_RMA_SLOF);
2541         exit(1);
2542     }
2543 
2544     if (kernel_filename) {
2545         uint64_t lowaddr = 0;
2546 
2547         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2548                                       NULL, NULL, &lowaddr, NULL, 1,
2549                                       PPC_ELF_MACHINE, 0, 0);
2550         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2551             spapr->kernel_size = load_elf(kernel_filename,
2552                                           translate_kernel_address, NULL, NULL,
2553                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2554                                           0, 0);
2555             spapr->kernel_le = spapr->kernel_size > 0;
2556         }
2557         if (spapr->kernel_size < 0) {
2558             error_report("error loading %s: %s", kernel_filename,
2559                          load_elf_strerror(spapr->kernel_size));
2560             exit(1);
2561         }
2562 
2563         /* load initrd */
2564         if (initrd_filename) {
2565             /* Try to locate the initrd in the gap between the kernel
2566              * and the firmware. Add a bit of space just in case
2567              */
2568             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2569                                   + 0x1ffff) & ~0xffff;
2570             spapr->initrd_size = load_image_targphys(initrd_filename,
2571                                                      spapr->initrd_base,
2572                                                      load_limit
2573                                                      - spapr->initrd_base);
2574             if (spapr->initrd_size < 0) {
2575                 error_report("could not load initial ram disk '%s'",
2576                              initrd_filename);
2577                 exit(1);
2578             }
2579         }
2580     }
2581 
2582     if (bios_name == NULL) {
2583         bios_name = FW_FILE_NAME;
2584     }
2585     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2586     if (!filename) {
2587         error_report("Could not find LPAR firmware '%s'", bios_name);
2588         exit(1);
2589     }
2590     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2591     if (fw_size <= 0) {
2592         error_report("Could not load LPAR firmware '%s'", filename);
2593         exit(1);
2594     }
2595     g_free(filename);
2596 
2597     /* FIXME: Should register things through the MachineState's qdev
2598      * interface, this is a legacy from the sPAPREnvironment structure
2599      * which predated MachineState but had a similar function */
2600     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2601     register_savevm_live(NULL, "spapr/htab", -1, 1,
2602                          &savevm_htab_handlers, spapr);
2603 
2604     qemu_register_boot_set(spapr_boot_set, spapr);
2605 
2606     if (kvm_enabled()) {
2607         /* to stop and start vmclock */
2608         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2609                                          &spapr->tb);
2610 
2611         kvmppc_spapr_enable_inkernel_multitce();
2612     }
2613 }
2614 
2615 static int spapr_kvm_type(const char *vm_type)
2616 {
2617     if (!vm_type) {
2618         return 0;
2619     }
2620 
2621     if (!strcmp(vm_type, "HV")) {
2622         return 1;
2623     }
2624 
2625     if (!strcmp(vm_type, "PR")) {
2626         return 2;
2627     }
2628 
2629     error_report("Unknown kvm-type specified '%s'", vm_type);
2630     exit(1);
2631 }
2632 
2633 /*
2634  * Implementation of an interface to adjust firmware path
2635  * for the bootindex property handling.
2636  */
2637 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2638                                    DeviceState *dev)
2639 {
2640 #define CAST(type, obj, name) \
2641     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2642     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2643     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2644     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2645 
2646     if (d) {
2647         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2648         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2649         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2650 
2651         if (spapr) {
2652             /*
2653              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2654              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2655              * in the top 16 bits of the 64-bit LUN
2656              */
2657             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2658             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2659                                    (uint64_t)id << 48);
2660         } else if (virtio) {
2661             /*
2662              * We use SRP luns of the form 01000000 | (target << 8) | lun
2663              * in the top 32 bits of the 64-bit LUN
2664              * Note: the quote above is from SLOF and it is wrong,
2665              * the actual binding is:
2666              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2667              */
2668             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2669             if (d->lun >= 256) {
2670                 /* Use the LUN "flat space addressing method" */
2671                 id |= 0x4000;
2672             }
2673             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2674                                    (uint64_t)id << 32);
2675         } else if (usb) {
2676             /*
2677              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2678              * in the top 32 bits of the 64-bit LUN
2679              */
2680             unsigned usb_port = atoi(usb->port->path);
2681             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2682             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2683                                    (uint64_t)id << 32);
2684         }
2685     }
2686 
2687     /*
2688      * SLOF probes the USB devices, and if it recognizes that the device is a
2689      * storage device, it changes its name to "storage" instead of "usb-host",
2690      * and additionally adds a child node for the SCSI LUN, so the correct
2691      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2692      */
2693     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2694         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2695         if (usb_host_dev_is_scsi_storage(usbdev)) {
2696             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2697         }
2698     }
2699 
2700     if (phb) {
2701         /* Replace "pci" with "pci@800000020000000" */
2702         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2703     }
2704 
2705     if (vsc) {
2706         /* Same logic as virtio above */
2707         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2708         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2709     }
2710 
2711     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2712         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2713         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2714         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2715     }
2716 
2717     return NULL;
2718 }
2719 
2720 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2721 {
2722     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2723 
2724     return g_strdup(spapr->kvm_type);
2725 }
2726 
2727 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2728 {
2729     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2730 
2731     g_free(spapr->kvm_type);
2732     spapr->kvm_type = g_strdup(value);
2733 }
2734 
2735 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2736 {
2737     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2738 
2739     return spapr->use_hotplug_event_source;
2740 }
2741 
2742 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2743                                             Error **errp)
2744 {
2745     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2746 
2747     spapr->use_hotplug_event_source = value;
2748 }
2749 
2750 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2751 {
2752     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2753 
2754     switch (spapr->resize_hpt) {
2755     case SPAPR_RESIZE_HPT_DEFAULT:
2756         return g_strdup("default");
2757     case SPAPR_RESIZE_HPT_DISABLED:
2758         return g_strdup("disabled");
2759     case SPAPR_RESIZE_HPT_ENABLED:
2760         return g_strdup("enabled");
2761     case SPAPR_RESIZE_HPT_REQUIRED:
2762         return g_strdup("required");
2763     }
2764     g_assert_not_reached();
2765 }
2766 
2767 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2768 {
2769     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2770 
2771     if (strcmp(value, "default") == 0) {
2772         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2773     } else if (strcmp(value, "disabled") == 0) {
2774         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2775     } else if (strcmp(value, "enabled") == 0) {
2776         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2777     } else if (strcmp(value, "required") == 0) {
2778         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2779     } else {
2780         error_setg(errp, "Bad value for \"resize-hpt\" property");
2781     }
2782 }
2783 
2784 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
2785                                    void *opaque, Error **errp)
2786 {
2787     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2788 }
2789 
2790 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
2791                                    void *opaque, Error **errp)
2792 {
2793     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2794 }
2795 
2796 static void spapr_machine_initfn(Object *obj)
2797 {
2798     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2799 
2800     spapr->htab_fd = -1;
2801     spapr->use_hotplug_event_source = true;
2802     object_property_add_str(obj, "kvm-type",
2803                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2804     object_property_set_description(obj, "kvm-type",
2805                                     "Specifies the KVM virtualization mode (HV, PR)",
2806                                     NULL);
2807     object_property_add_bool(obj, "modern-hotplug-events",
2808                             spapr_get_modern_hotplug_events,
2809                             spapr_set_modern_hotplug_events,
2810                             NULL);
2811     object_property_set_description(obj, "modern-hotplug-events",
2812                                     "Use dedicated hotplug event mechanism in"
2813                                     " place of standard EPOW events when possible"
2814                                     " (required for memory hot-unplug support)",
2815                                     NULL);
2816 
2817     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
2818                             "Maximum permitted CPU compatibility mode",
2819                             &error_fatal);
2820 
2821     object_property_add_str(obj, "resize-hpt",
2822                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
2823     object_property_set_description(obj, "resize-hpt",
2824                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
2825                                     NULL);
2826     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
2827                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
2828     object_property_set_description(obj, "vsmt",
2829                                     "Virtual SMT: KVM behaves as if this were"
2830                                     " the host's SMT mode", &error_abort);
2831 }
2832 
2833 static void spapr_machine_finalizefn(Object *obj)
2834 {
2835     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2836 
2837     g_free(spapr->kvm_type);
2838 }
2839 
2840 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2841 {
2842     cpu_synchronize_state(cs);
2843     ppc_cpu_do_system_reset(cs);
2844 }
2845 
2846 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2847 {
2848     CPUState *cs;
2849 
2850     CPU_FOREACH(cs) {
2851         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2852     }
2853 }
2854 
2855 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2856                            uint32_t node, bool dedicated_hp_event_source,
2857                            Error **errp)
2858 {
2859     sPAPRDRConnector *drc;
2860     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2861     int i, fdt_offset, fdt_size;
2862     void *fdt;
2863     uint64_t addr = addr_start;
2864     bool hotplugged = spapr_drc_hotplugged(dev);
2865     Error *local_err = NULL;
2866 
2867     for (i = 0; i < nr_lmbs; i++) {
2868         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2869                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2870         g_assert(drc);
2871 
2872         fdt = create_device_tree(&fdt_size);
2873         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2874                                                 SPAPR_MEMORY_BLOCK_SIZE);
2875 
2876         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
2877         if (local_err) {
2878             while (addr > addr_start) {
2879                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
2880                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2881                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
2882                 spapr_drc_detach(drc);
2883             }
2884             g_free(fdt);
2885             error_propagate(errp, local_err);
2886             return;
2887         }
2888         if (!hotplugged) {
2889             spapr_drc_reset(drc);
2890         }
2891         addr += SPAPR_MEMORY_BLOCK_SIZE;
2892     }
2893     /* send hotplug notification to the
2894      * guest only in case of hotplugged memory
2895      */
2896     if (hotplugged) {
2897         if (dedicated_hp_event_source) {
2898             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2899                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2900             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2901                                                    nr_lmbs,
2902                                                    spapr_drc_index(drc));
2903         } else {
2904             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2905                                            nr_lmbs);
2906         }
2907     }
2908 }
2909 
2910 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2911                               uint32_t node, Error **errp)
2912 {
2913     Error *local_err = NULL;
2914     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2915     PCDIMMDevice *dimm = PC_DIMM(dev);
2916     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2917     MemoryRegion *mr;
2918     uint64_t align, size, addr;
2919 
2920     mr = ddc->get_memory_region(dimm, &local_err);
2921     if (local_err) {
2922         goto out;
2923     }
2924     align = memory_region_get_alignment(mr);
2925     size = memory_region_size(mr);
2926 
2927     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2928     if (local_err) {
2929         goto out;
2930     }
2931 
2932     addr = object_property_get_uint(OBJECT(dimm),
2933                                     PC_DIMM_ADDR_PROP, &local_err);
2934     if (local_err) {
2935         goto out_unplug;
2936     }
2937 
2938     spapr_add_lmbs(dev, addr, size, node,
2939                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2940                    &local_err);
2941     if (local_err) {
2942         goto out_unplug;
2943     }
2944 
2945     return;
2946 
2947 out_unplug:
2948     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2949 out:
2950     error_propagate(errp, local_err);
2951 }
2952 
2953 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2954                                   Error **errp)
2955 {
2956     PCDIMMDevice *dimm = PC_DIMM(dev);
2957     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2958     MemoryRegion *mr;
2959     uint64_t size;
2960     char *mem_dev;
2961 
2962     mr = ddc->get_memory_region(dimm, errp);
2963     if (!mr) {
2964         return;
2965     }
2966     size = memory_region_size(mr);
2967 
2968     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2969         error_setg(errp, "Hotplugged memory size must be a multiple of "
2970                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2971         return;
2972     }
2973 
2974     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2975     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2976         error_setg(errp, "Memory backend has bad page size. "
2977                    "Use 'memory-backend-file' with correct mem-path.");
2978         goto out;
2979     }
2980 
2981 out:
2982     g_free(mem_dev);
2983 }
2984 
2985 struct sPAPRDIMMState {
2986     PCDIMMDevice *dimm;
2987     uint32_t nr_lmbs;
2988     QTAILQ_ENTRY(sPAPRDIMMState) next;
2989 };
2990 
2991 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
2992                                                        PCDIMMDevice *dimm)
2993 {
2994     sPAPRDIMMState *dimm_state = NULL;
2995 
2996     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
2997         if (dimm_state->dimm == dimm) {
2998             break;
2999         }
3000     }
3001     return dimm_state;
3002 }
3003 
3004 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3005                                                       uint32_t nr_lmbs,
3006                                                       PCDIMMDevice *dimm)
3007 {
3008     sPAPRDIMMState *ds = NULL;
3009 
3010     /*
3011      * If this request is for a DIMM whose removal had failed earlier
3012      * (due to guest's refusal to remove the LMBs), we would have this
3013      * dimm already in the pending_dimm_unplugs list. In that
3014      * case don't add again.
3015      */
3016     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3017     if (!ds) {
3018         ds = g_malloc0(sizeof(sPAPRDIMMState));
3019         ds->nr_lmbs = nr_lmbs;
3020         ds->dimm = dimm;
3021         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3022     }
3023     return ds;
3024 }
3025 
3026 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3027                                               sPAPRDIMMState *dimm_state)
3028 {
3029     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3030     g_free(dimm_state);
3031 }
3032 
3033 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3034                                                         PCDIMMDevice *dimm)
3035 {
3036     sPAPRDRConnector *drc;
3037     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3038     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3039     uint64_t size = memory_region_size(mr);
3040     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3041     uint32_t avail_lmbs = 0;
3042     uint64_t addr_start, addr;
3043     int i;
3044 
3045     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3046                                          &error_abort);
3047 
3048     addr = addr_start;
3049     for (i = 0; i < nr_lmbs; i++) {
3050         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3051                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3052         g_assert(drc);
3053         if (drc->dev) {
3054             avail_lmbs++;
3055         }
3056         addr += SPAPR_MEMORY_BLOCK_SIZE;
3057     }
3058 
3059     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3060 }
3061 
3062 /* Callback to be called during DRC release. */
3063 void spapr_lmb_release(DeviceState *dev)
3064 {
3065     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_hotplug_handler(dev));
3066     PCDIMMDevice *dimm = PC_DIMM(dev);
3067     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3068     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3069     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3070 
3071     /* This information will get lost if a migration occurs
3072      * during the unplug process. In this case recover it. */
3073     if (ds == NULL) {
3074         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3075         g_assert(ds);
3076         /* The DRC being examined by the caller at least must be counted */
3077         g_assert(ds->nr_lmbs);
3078     }
3079 
3080     if (--ds->nr_lmbs) {
3081         return;
3082     }
3083 
3084     /*
3085      * Now that all the LMBs have been removed by the guest, call the
3086      * pc-dimm unplug handler to cleanup up the pc-dimm device.
3087      */
3088     pc_dimm_memory_unplug(dev, &spapr->hotplug_memory, mr);
3089     object_unparent(OBJECT(dev));
3090     spapr_pending_dimm_unplugs_remove(spapr, ds);
3091 }
3092 
3093 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3094                                         DeviceState *dev, Error **errp)
3095 {
3096     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3097     Error *local_err = NULL;
3098     PCDIMMDevice *dimm = PC_DIMM(dev);
3099     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3100     MemoryRegion *mr;
3101     uint32_t nr_lmbs;
3102     uint64_t size, addr_start, addr;
3103     int i;
3104     sPAPRDRConnector *drc;
3105 
3106     mr = ddc->get_memory_region(dimm, &local_err);
3107     if (local_err) {
3108         goto out;
3109     }
3110     size = memory_region_size(mr);
3111     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3112 
3113     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3114                                          &local_err);
3115     if (local_err) {
3116         goto out;
3117     }
3118 
3119     /*
3120      * An existing pending dimm state for this DIMM means that there is an
3121      * unplug operation in progress, waiting for the spapr_lmb_release
3122      * callback to complete the job (BQL can't cover that far). In this case,
3123      * bail out to avoid detaching DRCs that were already released.
3124      */
3125     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3126         error_setg(&local_err,
3127                    "Memory unplug already in progress for device %s",
3128                    dev->id);
3129         goto out;
3130     }
3131 
3132     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3133 
3134     addr = addr_start;
3135     for (i = 0; i < nr_lmbs; i++) {
3136         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3137                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3138         g_assert(drc);
3139 
3140         spapr_drc_detach(drc);
3141         addr += SPAPR_MEMORY_BLOCK_SIZE;
3142     }
3143 
3144     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3145                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3146     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3147                                               nr_lmbs, spapr_drc_index(drc));
3148 out:
3149     error_propagate(errp, local_err);
3150 }
3151 
3152 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3153                                            sPAPRMachineState *spapr)
3154 {
3155     PowerPCCPU *cpu = POWERPC_CPU(cs);
3156     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3157     int id = spapr_vcpu_id(cpu);
3158     void *fdt;
3159     int offset, fdt_size;
3160     char *nodename;
3161 
3162     fdt = create_device_tree(&fdt_size);
3163     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3164     offset = fdt_add_subnode(fdt, 0, nodename);
3165 
3166     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3167     g_free(nodename);
3168 
3169     *fdt_offset = offset;
3170     return fdt;
3171 }
3172 
3173 /* Callback to be called during DRC release. */
3174 void spapr_core_release(DeviceState *dev)
3175 {
3176     MachineState *ms = MACHINE(qdev_get_hotplug_handler(dev));
3177     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3178     CPUCore *cc = CPU_CORE(dev);
3179     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3180 
3181     if (smc->pre_2_10_has_unused_icps) {
3182         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3183         sPAPRCPUCoreClass *scc = SPAPR_CPU_CORE_GET_CLASS(OBJECT(cc));
3184         size_t size = object_type_get_instance_size(scc->cpu_type);
3185         int i;
3186 
3187         for (i = 0; i < cc->nr_threads; i++) {
3188             CPUState *cs = CPU(sc->threads + i * size);
3189 
3190             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3191         }
3192     }
3193 
3194     assert(core_slot);
3195     core_slot->cpu = NULL;
3196     object_unparent(OBJECT(dev));
3197 }
3198 
3199 static
3200 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3201                                Error **errp)
3202 {
3203     int index;
3204     sPAPRDRConnector *drc;
3205     CPUCore *cc = CPU_CORE(dev);
3206     int smt = kvmppc_smt_threads();
3207 
3208     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3209         error_setg(errp, "Unable to find CPU core with core-id: %d",
3210                    cc->core_id);
3211         return;
3212     }
3213     if (index == 0) {
3214         error_setg(errp, "Boot CPU core may not be unplugged");
3215         return;
3216     }
3217 
3218     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3219     g_assert(drc);
3220 
3221     spapr_drc_detach(drc);
3222 
3223     spapr_hotplug_req_remove_by_index(drc);
3224 }
3225 
3226 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3227                             Error **errp)
3228 {
3229     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3230     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3231     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3232     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3233     CPUCore *cc = CPU_CORE(dev);
3234     CPUState *cs = CPU(core->threads);
3235     sPAPRDRConnector *drc;
3236     Error *local_err = NULL;
3237     int smt = kvmppc_smt_threads();
3238     CPUArchId *core_slot;
3239     int index;
3240     bool hotplugged = spapr_drc_hotplugged(dev);
3241 
3242     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3243     if (!core_slot) {
3244         error_setg(errp, "Unable to find CPU core with core-id: %d",
3245                    cc->core_id);
3246         return;
3247     }
3248     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3249 
3250     g_assert(drc || !mc->has_hotpluggable_cpus);
3251 
3252     if (drc) {
3253         void *fdt;
3254         int fdt_offset;
3255 
3256         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3257 
3258         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3259         if (local_err) {
3260             g_free(fdt);
3261             error_propagate(errp, local_err);
3262             return;
3263         }
3264 
3265         if (hotplugged) {
3266             /*
3267              * Send hotplug notification interrupt to the guest only
3268              * in case of hotplugged CPUs.
3269              */
3270             spapr_hotplug_req_add_by_index(drc);
3271         } else {
3272             spapr_drc_reset(drc);
3273         }
3274     }
3275 
3276     core_slot->cpu = OBJECT(dev);
3277 
3278     if (smc->pre_2_10_has_unused_icps) {
3279         sPAPRCPUCoreClass *scc = SPAPR_CPU_CORE_GET_CLASS(OBJECT(cc));
3280         size_t size = object_type_get_instance_size(scc->cpu_type);
3281         int i;
3282 
3283         for (i = 0; i < cc->nr_threads; i++) {
3284             sPAPRCPUCore *sc = SPAPR_CPU_CORE(dev);
3285             void *obj = sc->threads + i * size;
3286 
3287             cs = CPU(obj);
3288             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3289         }
3290     }
3291 }
3292 
3293 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3294                                 Error **errp)
3295 {
3296     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3297     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3298     Error *local_err = NULL;
3299     CPUCore *cc = CPU_CORE(dev);
3300     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3301     const char *type = object_get_typename(OBJECT(dev));
3302     CPUArchId *core_slot;
3303     int index;
3304 
3305     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3306         error_setg(&local_err, "CPU hotplug not supported for this machine");
3307         goto out;
3308     }
3309 
3310     if (strcmp(base_core_type, type)) {
3311         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3312         goto out;
3313     }
3314 
3315     if (cc->core_id % smp_threads) {
3316         error_setg(&local_err, "invalid core id %d", cc->core_id);
3317         goto out;
3318     }
3319 
3320     /*
3321      * In general we should have homogeneous threads-per-core, but old
3322      * (pre hotplug support) machine types allow the last core to have
3323      * reduced threads as a compatibility hack for when we allowed
3324      * total vcpus not a multiple of threads-per-core.
3325      */
3326     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3327         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3328                    cc->nr_threads, smp_threads);
3329         goto out;
3330     }
3331 
3332     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3333     if (!core_slot) {
3334         error_setg(&local_err, "core id %d out of range", cc->core_id);
3335         goto out;
3336     }
3337 
3338     if (core_slot->cpu) {
3339         error_setg(&local_err, "core %d already populated", cc->core_id);
3340         goto out;
3341     }
3342 
3343     numa_cpu_pre_plug(core_slot, dev, &local_err);
3344 
3345 out:
3346     error_propagate(errp, local_err);
3347 }
3348 
3349 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3350                                       DeviceState *dev, Error **errp)
3351 {
3352     MachineState *ms = MACHINE(hotplug_dev);
3353     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3354 
3355     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3356         int node;
3357 
3358         if (!smc->dr_lmb_enabled) {
3359             error_setg(errp, "Memory hotplug not supported for this machine");
3360             return;
3361         }
3362         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3363         if (*errp) {
3364             return;
3365         }
3366         if (node < 0 || node >= MAX_NODES) {
3367             error_setg(errp, "Invaild node %d", node);
3368             return;
3369         }
3370 
3371         /*
3372          * Currently PowerPC kernel doesn't allow hot-adding memory to
3373          * memory-less node, but instead will silently add the memory
3374          * to the first node that has some memory. This causes two
3375          * unexpected behaviours for the user.
3376          *
3377          * - Memory gets hotplugged to a different node than what the user
3378          *   specified.
3379          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3380          *   to memory-less node, a reboot will set things accordingly
3381          *   and the previously hotplugged memory now ends in the right node.
3382          *   This appears as if some memory moved from one node to another.
3383          *
3384          * So until kernel starts supporting memory hotplug to memory-less
3385          * nodes, just prevent such attempts upfront in QEMU.
3386          */
3387         if (nb_numa_nodes && !numa_info[node].node_mem) {
3388             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3389                        node);
3390             return;
3391         }
3392 
3393         spapr_memory_plug(hotplug_dev, dev, node, errp);
3394     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3395         spapr_core_plug(hotplug_dev, dev, errp);
3396     }
3397 }
3398 
3399 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3400                                                 DeviceState *dev, Error **errp)
3401 {
3402     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3403     MachineClass *mc = MACHINE_GET_CLASS(sms);
3404 
3405     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3406         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3407             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3408         } else {
3409             /* NOTE: this means there is a window after guest reset, prior to
3410              * CAS negotiation, where unplug requests will fail due to the
3411              * capability not being detected yet. This is a bit different than
3412              * the case with PCI unplug, where the events will be queued and
3413              * eventually handled by the guest after boot
3414              */
3415             error_setg(errp, "Memory hot unplug not supported for this guest");
3416         }
3417     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3418         if (!mc->has_hotpluggable_cpus) {
3419             error_setg(errp, "CPU hot unplug not supported on this machine");
3420             return;
3421         }
3422         spapr_core_unplug_request(hotplug_dev, dev, errp);
3423     }
3424 }
3425 
3426 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3427                                           DeviceState *dev, Error **errp)
3428 {
3429     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3430         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3431     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3432         spapr_core_pre_plug(hotplug_dev, dev, errp);
3433     }
3434 }
3435 
3436 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3437                                                  DeviceState *dev)
3438 {
3439     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3440         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3441         return HOTPLUG_HANDLER(machine);
3442     }
3443     return NULL;
3444 }
3445 
3446 static CpuInstanceProperties
3447 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3448 {
3449     CPUArchId *core_slot;
3450     MachineClass *mc = MACHINE_GET_CLASS(machine);
3451 
3452     /* make sure possible_cpu are intialized */
3453     mc->possible_cpu_arch_ids(machine);
3454     /* get CPU core slot containing thread that matches cpu_index */
3455     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3456     assert(core_slot);
3457     return core_slot->props;
3458 }
3459 
3460 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3461 {
3462     return idx / smp_cores % nb_numa_nodes;
3463 }
3464 
3465 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3466 {
3467     int i;
3468     int spapr_max_cores = max_cpus / smp_threads;
3469     MachineClass *mc = MACHINE_GET_CLASS(machine);
3470 
3471     if (!mc->has_hotpluggable_cpus) {
3472         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3473     }
3474     if (machine->possible_cpus) {
3475         assert(machine->possible_cpus->len == spapr_max_cores);
3476         return machine->possible_cpus;
3477     }
3478 
3479     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3480                              sizeof(CPUArchId) * spapr_max_cores);
3481     machine->possible_cpus->len = spapr_max_cores;
3482     for (i = 0; i < machine->possible_cpus->len; i++) {
3483         int core_id = i * smp_threads;
3484 
3485         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3486         machine->possible_cpus->cpus[i].arch_id = core_id;
3487         machine->possible_cpus->cpus[i].props.has_core_id = true;
3488         machine->possible_cpus->cpus[i].props.core_id = core_id;
3489     }
3490     return machine->possible_cpus;
3491 }
3492 
3493 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3494                                 uint64_t *buid, hwaddr *pio,
3495                                 hwaddr *mmio32, hwaddr *mmio64,
3496                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3497 {
3498     /*
3499      * New-style PHB window placement.
3500      *
3501      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3502      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3503      * windows.
3504      *
3505      * Some guest kernels can't work with MMIO windows above 1<<46
3506      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3507      *
3508      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3509      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3510      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3511      * 1TiB 64-bit MMIO windows for each PHB.
3512      */
3513     const uint64_t base_buid = 0x800000020000000ULL;
3514 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3515                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3516     int i;
3517 
3518     /* Sanity check natural alignments */
3519     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3520     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3521     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3522     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3523     /* Sanity check bounds */
3524     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3525                       SPAPR_PCI_MEM32_WIN_SIZE);
3526     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3527                       SPAPR_PCI_MEM64_WIN_SIZE);
3528 
3529     if (index >= SPAPR_MAX_PHBS) {
3530         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3531                    SPAPR_MAX_PHBS - 1);
3532         return;
3533     }
3534 
3535     *buid = base_buid + index;
3536     for (i = 0; i < n_dma; ++i) {
3537         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3538     }
3539 
3540     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3541     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3542     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3543 }
3544 
3545 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3546 {
3547     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3548 
3549     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3550 }
3551 
3552 static void spapr_ics_resend(XICSFabric *dev)
3553 {
3554     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3555 
3556     ics_resend(spapr->ics);
3557 }
3558 
3559 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3560 {
3561     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3562 
3563     return cpu ? ICP(cpu->intc) : NULL;
3564 }
3565 
3566 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3567                                  Monitor *mon)
3568 {
3569     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3570     CPUState *cs;
3571 
3572     CPU_FOREACH(cs) {
3573         PowerPCCPU *cpu = POWERPC_CPU(cs);
3574 
3575         icp_pic_print_info(ICP(cpu->intc), mon);
3576     }
3577 
3578     ics_pic_print_info(spapr->ics, mon);
3579 }
3580 
3581 int spapr_vcpu_id(PowerPCCPU *cpu)
3582 {
3583     CPUState *cs = CPU(cpu);
3584 
3585     if (kvm_enabled()) {
3586         return kvm_arch_vcpu_id(cs);
3587     } else {
3588         return cs->cpu_index;
3589     }
3590 }
3591 
3592 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3593 {
3594     CPUState *cs;
3595 
3596     CPU_FOREACH(cs) {
3597         PowerPCCPU *cpu = POWERPC_CPU(cs);
3598 
3599         if (spapr_vcpu_id(cpu) == vcpu_id) {
3600             return cpu;
3601         }
3602     }
3603 
3604     return NULL;
3605 }
3606 
3607 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3608 {
3609     MachineClass *mc = MACHINE_CLASS(oc);
3610     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3611     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3612     NMIClass *nc = NMI_CLASS(oc);
3613     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3614     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3615     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3616     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3617 
3618     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3619 
3620     /*
3621      * We set up the default / latest behaviour here.  The class_init
3622      * functions for the specific versioned machine types can override
3623      * these details for backwards compatibility
3624      */
3625     mc->init = ppc_spapr_init;
3626     mc->reset = ppc_spapr_reset;
3627     mc->block_default_type = IF_SCSI;
3628     mc->max_cpus = 1024;
3629     mc->no_parallel = 1;
3630     mc->default_boot_order = "";
3631     mc->default_ram_size = 512 * M_BYTE;
3632     mc->kvm_type = spapr_kvm_type;
3633     mc->has_dynamic_sysbus = true;
3634     mc->pci_allow_0_address = true;
3635     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3636     hc->pre_plug = spapr_machine_device_pre_plug;
3637     hc->plug = spapr_machine_device_plug;
3638     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3639     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3640     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3641     hc->unplug_request = spapr_machine_device_unplug_request;
3642 
3643     smc->dr_lmb_enabled = true;
3644     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3645     mc->has_hotpluggable_cpus = true;
3646     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3647     fwc->get_dev_path = spapr_get_fw_dev_path;
3648     nc->nmi_monitor_handler = spapr_nmi;
3649     smc->phb_placement = spapr_phb_placement;
3650     vhc->hypercall = emulate_spapr_hypercall;
3651     vhc->hpt_mask = spapr_hpt_mask;
3652     vhc->map_hptes = spapr_map_hptes;
3653     vhc->unmap_hptes = spapr_unmap_hptes;
3654     vhc->store_hpte = spapr_store_hpte;
3655     vhc->get_patbe = spapr_get_patbe;
3656     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3657     xic->ics_get = spapr_ics_get;
3658     xic->ics_resend = spapr_ics_resend;
3659     xic->icp_get = spapr_icp_get;
3660     ispc->print_info = spapr_pic_print_info;
3661     /* Force NUMA node memory size to be a multiple of
3662      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3663      * in which LMBs are represented and hot-added
3664      */
3665     mc->numa_mem_align_shift = 28;
3666 }
3667 
3668 static const TypeInfo spapr_machine_info = {
3669     .name          = TYPE_SPAPR_MACHINE,
3670     .parent        = TYPE_MACHINE,
3671     .abstract      = true,
3672     .instance_size = sizeof(sPAPRMachineState),
3673     .instance_init = spapr_machine_initfn,
3674     .instance_finalize = spapr_machine_finalizefn,
3675     .class_size    = sizeof(sPAPRMachineClass),
3676     .class_init    = spapr_machine_class_init,
3677     .interfaces = (InterfaceInfo[]) {
3678         { TYPE_FW_PATH_PROVIDER },
3679         { TYPE_NMI },
3680         { TYPE_HOTPLUG_HANDLER },
3681         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3682         { TYPE_XICS_FABRIC },
3683         { TYPE_INTERRUPT_STATS_PROVIDER },
3684         { }
3685     },
3686 };
3687 
3688 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3689     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3690                                                     void *data)      \
3691     {                                                                \
3692         MachineClass *mc = MACHINE_CLASS(oc);                        \
3693         spapr_machine_##suffix##_class_options(mc);                  \
3694         if (latest) {                                                \
3695             mc->alias = "pseries";                                   \
3696             mc->is_default = 1;                                      \
3697         }                                                            \
3698     }                                                                \
3699     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3700     {                                                                \
3701         MachineState *machine = MACHINE(obj);                        \
3702         spapr_machine_##suffix##_instance_options(machine);          \
3703     }                                                                \
3704     static const TypeInfo spapr_machine_##suffix##_info = {          \
3705         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3706         .parent = TYPE_SPAPR_MACHINE,                                \
3707         .class_init = spapr_machine_##suffix##_class_init,           \
3708         .instance_init = spapr_machine_##suffix##_instance_init,     \
3709     };                                                               \
3710     static void spapr_machine_register_##suffix(void)                \
3711     {                                                                \
3712         type_register(&spapr_machine_##suffix##_info);               \
3713     }                                                                \
3714     type_init(spapr_machine_register_##suffix)
3715 
3716 /*
3717  * pseries-2.11
3718  */
3719 static void spapr_machine_2_11_instance_options(MachineState *machine)
3720 {
3721 }
3722 
3723 static void spapr_machine_2_11_class_options(MachineClass *mc)
3724 {
3725     /* Defaults for the latest behaviour inherited from the base class */
3726 }
3727 
3728 DEFINE_SPAPR_MACHINE(2_11, "2.11", true);
3729 
3730 /*
3731  * pseries-2.10
3732  */
3733 #define SPAPR_COMPAT_2_10                                              \
3734     HW_COMPAT_2_10                                                     \
3735 
3736 static void spapr_machine_2_10_instance_options(MachineState *machine)
3737 {
3738 }
3739 
3740 static void spapr_machine_2_10_class_options(MachineClass *mc)
3741 {
3742     spapr_machine_2_11_class_options(mc);
3743     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
3744 }
3745 
3746 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
3747 
3748 /*
3749  * pseries-2.9
3750  */
3751 #define SPAPR_COMPAT_2_9                                               \
3752     HW_COMPAT_2_9                                                      \
3753     {                                                                  \
3754         .driver = TYPE_POWERPC_CPU,                                    \
3755         .property = "pre-2.10-migration",                              \
3756         .value    = "on",                                              \
3757     },                                                                 \
3758 
3759 static void spapr_machine_2_9_instance_options(MachineState *machine)
3760 {
3761     spapr_machine_2_10_instance_options(machine);
3762 }
3763 
3764 static void spapr_machine_2_9_class_options(MachineClass *mc)
3765 {
3766     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3767 
3768     spapr_machine_2_10_class_options(mc);
3769     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3770     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
3771     smc->pre_2_10_has_unused_icps = true;
3772     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
3773 }
3774 
3775 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3776 
3777 /*
3778  * pseries-2.8
3779  */
3780 #define SPAPR_COMPAT_2_8                                        \
3781     HW_COMPAT_2_8                                               \
3782     {                                                           \
3783         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3784         .property = "pcie-extended-configuration-space",        \
3785         .value    = "off",                                      \
3786     },
3787 
3788 static void spapr_machine_2_8_instance_options(MachineState *machine)
3789 {
3790     spapr_machine_2_9_instance_options(machine);
3791 }
3792 
3793 static void spapr_machine_2_8_class_options(MachineClass *mc)
3794 {
3795     spapr_machine_2_9_class_options(mc);
3796     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3797     mc->numa_mem_align_shift = 23;
3798 }
3799 
3800 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3801 
3802 /*
3803  * pseries-2.7
3804  */
3805 #define SPAPR_COMPAT_2_7                            \
3806     HW_COMPAT_2_7                                   \
3807     {                                               \
3808         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3809         .property = "mem_win_size",                 \
3810         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3811     },                                              \
3812     {                                               \
3813         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3814         .property = "mem64_win_size",               \
3815         .value    = "0",                            \
3816     },                                              \
3817     {                                               \
3818         .driver = TYPE_POWERPC_CPU,                 \
3819         .property = "pre-2.8-migration",            \
3820         .value    = "on",                           \
3821     },                                              \
3822     {                                               \
3823         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3824         .property = "pre-2.8-migration",            \
3825         .value    = "on",                           \
3826     },
3827 
3828 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3829                               uint64_t *buid, hwaddr *pio,
3830                               hwaddr *mmio32, hwaddr *mmio64,
3831                               unsigned n_dma, uint32_t *liobns, Error **errp)
3832 {
3833     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3834     const uint64_t base_buid = 0x800000020000000ULL;
3835     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3836     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3837     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3838     const uint32_t max_index = 255;
3839     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3840 
3841     uint64_t ram_top = MACHINE(spapr)->ram_size;
3842     hwaddr phb0_base, phb_base;
3843     int i;
3844 
3845     /* Do we have hotpluggable memory? */
3846     if (MACHINE(spapr)->maxram_size > ram_top) {
3847         /* Can't just use maxram_size, because there may be an
3848          * alignment gap between normal and hotpluggable memory
3849          * regions */
3850         ram_top = spapr->hotplug_memory.base +
3851             memory_region_size(&spapr->hotplug_memory.mr);
3852     }
3853 
3854     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3855 
3856     if (index > max_index) {
3857         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3858                    max_index);
3859         return;
3860     }
3861 
3862     *buid = base_buid + index;
3863     for (i = 0; i < n_dma; ++i) {
3864         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3865     }
3866 
3867     phb_base = phb0_base + index * phb_spacing;
3868     *pio = phb_base + pio_offset;
3869     *mmio32 = phb_base + mmio_offset;
3870     /*
3871      * We don't set the 64-bit MMIO window, relying on the PHB's
3872      * fallback behaviour of automatically splitting a large "32-bit"
3873      * window into contiguous 32-bit and 64-bit windows
3874      */
3875 }
3876 
3877 static void spapr_machine_2_7_instance_options(MachineState *machine)
3878 {
3879     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3880 
3881     spapr_machine_2_8_instance_options(machine);
3882     spapr->use_hotplug_event_source = false;
3883 }
3884 
3885 static void spapr_machine_2_7_class_options(MachineClass *mc)
3886 {
3887     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3888 
3889     spapr_machine_2_8_class_options(mc);
3890     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
3891     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3892     smc->phb_placement = phb_placement_2_7;
3893 }
3894 
3895 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3896 
3897 /*
3898  * pseries-2.6
3899  */
3900 #define SPAPR_COMPAT_2_6 \
3901     HW_COMPAT_2_6 \
3902     { \
3903         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3904         .property = "ddw",\
3905         .value    = stringify(off),\
3906     },
3907 
3908 static void spapr_machine_2_6_instance_options(MachineState *machine)
3909 {
3910     spapr_machine_2_7_instance_options(machine);
3911 }
3912 
3913 static void spapr_machine_2_6_class_options(MachineClass *mc)
3914 {
3915     spapr_machine_2_7_class_options(mc);
3916     mc->has_hotpluggable_cpus = false;
3917     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3918 }
3919 
3920 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3921 
3922 /*
3923  * pseries-2.5
3924  */
3925 #define SPAPR_COMPAT_2_5 \
3926     HW_COMPAT_2_5 \
3927     { \
3928         .driver   = "spapr-vlan", \
3929         .property = "use-rx-buffer-pools", \
3930         .value    = "off", \
3931     },
3932 
3933 static void spapr_machine_2_5_instance_options(MachineState *machine)
3934 {
3935     spapr_machine_2_6_instance_options(machine);
3936 }
3937 
3938 static void spapr_machine_2_5_class_options(MachineClass *mc)
3939 {
3940     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3941 
3942     spapr_machine_2_6_class_options(mc);
3943     smc->use_ohci_by_default = true;
3944     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3945 }
3946 
3947 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3948 
3949 /*
3950  * pseries-2.4
3951  */
3952 #define SPAPR_COMPAT_2_4 \
3953         HW_COMPAT_2_4
3954 
3955 static void spapr_machine_2_4_instance_options(MachineState *machine)
3956 {
3957     spapr_machine_2_5_instance_options(machine);
3958 }
3959 
3960 static void spapr_machine_2_4_class_options(MachineClass *mc)
3961 {
3962     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3963 
3964     spapr_machine_2_5_class_options(mc);
3965     smc->dr_lmb_enabled = false;
3966     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3967 }
3968 
3969 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3970 
3971 /*
3972  * pseries-2.3
3973  */
3974 #define SPAPR_COMPAT_2_3 \
3975         HW_COMPAT_2_3 \
3976         {\
3977             .driver   = "spapr-pci-host-bridge",\
3978             .property = "dynamic-reconfiguration",\
3979             .value    = "off",\
3980         },
3981 
3982 static void spapr_machine_2_3_instance_options(MachineState *machine)
3983 {
3984     spapr_machine_2_4_instance_options(machine);
3985 }
3986 
3987 static void spapr_machine_2_3_class_options(MachineClass *mc)
3988 {
3989     spapr_machine_2_4_class_options(mc);
3990     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3991 }
3992 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3993 
3994 /*
3995  * pseries-2.2
3996  */
3997 
3998 #define SPAPR_COMPAT_2_2 \
3999         HW_COMPAT_2_2 \
4000         {\
4001             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4002             .property = "mem_win_size",\
4003             .value    = "0x20000000",\
4004         },
4005 
4006 static void spapr_machine_2_2_instance_options(MachineState *machine)
4007 {
4008     spapr_machine_2_3_instance_options(machine);
4009     machine->suppress_vmdesc = true;
4010 }
4011 
4012 static void spapr_machine_2_2_class_options(MachineClass *mc)
4013 {
4014     spapr_machine_2_3_class_options(mc);
4015     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4016 }
4017 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4018 
4019 /*
4020  * pseries-2.1
4021  */
4022 #define SPAPR_COMPAT_2_1 \
4023         HW_COMPAT_2_1
4024 
4025 static void spapr_machine_2_1_instance_options(MachineState *machine)
4026 {
4027     spapr_machine_2_2_instance_options(machine);
4028 }
4029 
4030 static void spapr_machine_2_1_class_options(MachineClass *mc)
4031 {
4032     spapr_machine_2_2_class_options(mc);
4033     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4034 }
4035 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4036 
4037 static void spapr_machine_register_types(void)
4038 {
4039     type_register_static(&spapr_machine_info);
4040 }
4041 
4042 type_init(spapr_machine_register_types)
4043