xref: /openbmc/qemu/hw/ppc/spapr.c (revision 62dd4eda)
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 "qom/cpu.h"
48 
49 #include "hw/boards.h"
50 #include "hw/ppc/ppc.h"
51 #include "hw/loader.h"
52 
53 #include "hw/ppc/fdt.h"
54 #include "hw/ppc/spapr.h"
55 #include "hw/ppc/spapr_vio.h"
56 #include "hw/pci-host/spapr.h"
57 #include "hw/ppc/xics.h"
58 #include "hw/pci/msi.h"
59 
60 #include "hw/pci/pci.h"
61 #include "hw/scsi/scsi.h"
62 #include "hw/virtio/virtio-scsi.h"
63 #include "hw/virtio/vhost-scsi-common.h"
64 
65 #include "exec/address-spaces.h"
66 #include "hw/usb.h"
67 #include "qemu/config-file.h"
68 #include "qemu/error-report.h"
69 #include "trace.h"
70 #include "hw/nmi.h"
71 #include "hw/intc/intc.h"
72 
73 #include "hw/compat.h"
74 #include "qemu/cutils.h"
75 #include "hw/ppc/spapr_cpu_core.h"
76 #include "qmp-commands.h"
77 
78 #include <libfdt.h>
79 
80 /* SLOF memory layout:
81  *
82  * SLOF raw image loaded at 0, copies its romfs right below the flat
83  * device-tree, then position SLOF itself 31M below that
84  *
85  * So we set FW_OVERHEAD to 40MB which should account for all of that
86  * and more
87  *
88  * We load our kernel at 4M, leaving space for SLOF initial image
89  */
90 #define FDT_MAX_SIZE            0x100000
91 #define RTAS_MAX_SIZE           0x10000
92 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
93 #define FW_MAX_SIZE             0x400000
94 #define FW_FILE_NAME            "slof.bin"
95 #define FW_OVERHEAD             0x2800000
96 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
97 
98 #define MIN_RMA_SLOF            128UL
99 
100 #define PHANDLE_XICP            0x00001111
101 
102 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
103                                   const char *type_ics,
104                                   int nr_irqs, Error **errp)
105 {
106     Error *local_err = NULL;
107     Object *obj;
108 
109     obj = object_new(type_ics);
110     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
111     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
112                                    &error_abort);
113     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
114     if (local_err) {
115         goto error;
116     }
117     object_property_set_bool(obj, true, "realized", &local_err);
118     if (local_err) {
119         goto error;
120     }
121 
122     return ICS_SIMPLE(obj);
123 
124 error:
125     error_propagate(errp, local_err);
126     return NULL;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 static inline int xics_max_server_number(void)
164 {
165     return DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(), smp_threads);
166 }
167 
168 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
169 {
170     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
171     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
172 
173     if (kvm_enabled()) {
174         if (machine_kernel_irqchip_allowed(machine) &&
175             !xics_kvm_init(spapr, errp)) {
176             spapr->icp_type = TYPE_KVM_ICP;
177             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
178         }
179         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
180             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
181             return;
182         }
183     }
184 
185     if (!spapr->ics) {
186         xics_spapr_init(spapr);
187         spapr->icp_type = TYPE_ICP;
188         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
189         if (!spapr->ics) {
190             return;
191         }
192     }
193 
194     if (smc->pre_2_10_has_unused_icps) {
195         int i;
196 
197         for (i = 0; i < xics_max_server_number(); i++) {
198             /* Dummy entries get deregistered when real ICPState objects
199              * are registered during CPU core hotplug.
200              */
201             pre_2_10_vmstate_register_dummy_icp(i);
202         }
203     }
204 }
205 
206 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
207                                   int smt_threads)
208 {
209     int i, ret = 0;
210     uint32_t servers_prop[smt_threads];
211     uint32_t gservers_prop[smt_threads * 2];
212     int index = spapr_vcpu_id(cpu);
213 
214     if (cpu->compat_pvr) {
215         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
216         if (ret < 0) {
217             return ret;
218         }
219     }
220 
221     /* Build interrupt servers and gservers properties */
222     for (i = 0; i < smt_threads; i++) {
223         servers_prop[i] = cpu_to_be32(index + i);
224         /* Hack, direct the group queues back to cpu 0 */
225         gservers_prop[i*2] = cpu_to_be32(index + i);
226         gservers_prop[i*2 + 1] = 0;
227     }
228     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
229                       servers_prop, sizeof(servers_prop));
230     if (ret < 0) {
231         return ret;
232     }
233     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
234                       gservers_prop, sizeof(gservers_prop));
235 
236     return ret;
237 }
238 
239 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
240 {
241     int index = spapr_vcpu_id(cpu);
242     uint32_t associativity[] = {cpu_to_be32(0x5),
243                                 cpu_to_be32(0x0),
244                                 cpu_to_be32(0x0),
245                                 cpu_to_be32(0x0),
246                                 cpu_to_be32(cpu->node_id),
247                                 cpu_to_be32(index)};
248 
249     /* Advertise NUMA via ibm,associativity */
250     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
251                           sizeof(associativity));
252 }
253 
254 /* Populate the "ibm,pa-features" property */
255 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset,
256                                       bool legacy_guest)
257 {
258     uint8_t pa_features_206[] = { 6, 0,
259         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
260     uint8_t pa_features_207[] = { 24, 0,
261         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
262         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
263         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
264         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
265     uint8_t pa_features_300[] = { 66, 0,
266         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
267         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
268         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
269         /* 6: DS207 */
270         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
271         /* 16: Vector */
272         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
273         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
274         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
275         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
276         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
277         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
278         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
279         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
280         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
281         /* 42: PM, 44: PC RA, 46: SC vec'd */
282         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
283         /* 48: SIMD, 50: QP BFP, 52: String */
284         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
285         /* 54: DecFP, 56: DecI, 58: SHA */
286         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
287         /* 60: NM atomic, 62: RNG */
288         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
289     };
290     uint8_t *pa_features;
291     size_t pa_size;
292 
293     switch (POWERPC_MMU_VER(env->mmu_model)) {
294     case POWERPC_MMU_VER_2_06:
295         pa_features = pa_features_206;
296         pa_size = sizeof(pa_features_206);
297         break;
298     case POWERPC_MMU_VER_2_07:
299         pa_features = pa_features_207;
300         pa_size = sizeof(pa_features_207);
301         break;
302     case POWERPC_MMU_VER_3_00:
303         pa_features = pa_features_300;
304         pa_size = sizeof(pa_features_300);
305         break;
306     default:
307         return;
308     }
309 
310     if (env->ci_large_pages) {
311         /*
312          * Note: we keep CI large pages off by default because a 64K capable
313          * guest provisioned with large pages might otherwise try to map a qemu
314          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
315          * even if that qemu runs on a 4k host.
316          * We dd this bit back here if we are confident this is not an issue
317          */
318         pa_features[3] |= 0x20;
319     }
320     if (kvmppc_has_cap_htm() && pa_size > 24) {
321         pa_features[24] |= 0x80;    /* Transactional memory support */
322     }
323     if (legacy_guest && pa_size > 40) {
324         /* Workaround for broken kernels that attempt (guest) radix
325          * mode when they can't handle it, if they see the radix bit set
326          * in pa-features. So hide it from them. */
327         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
328     }
329 
330     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
331 }
332 
333 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
334 {
335     int ret = 0, offset, cpus_offset;
336     CPUState *cs;
337     char cpu_model[32];
338     int smt = kvmppc_smt_threads();
339     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
340 
341     CPU_FOREACH(cs) {
342         PowerPCCPU *cpu = POWERPC_CPU(cs);
343         CPUPPCState *env = &cpu->env;
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, fdt_path_offset(fdt, "/"),
357                                           "cpus");
358             if (cpus_offset < 0) {
359                 return cpus_offset;
360             }
361         }
362         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
363         if (offset < 0) {
364             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
365             if (offset < 0) {
366                 return offset;
367             }
368         }
369 
370         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
371                           pft_size_prop, sizeof(pft_size_prop));
372         if (ret < 0) {
373             return ret;
374         }
375 
376         if (nb_numa_nodes > 1) {
377             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
378             if (ret < 0) {
379                 return ret;
380             }
381         }
382 
383         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
384         if (ret < 0) {
385             return ret;
386         }
387 
388         spapr_populate_pa_features(env, fdt, offset,
389                                          spapr->cas_legacy_guest_workaround);
390     }
391     return ret;
392 }
393 
394 static hwaddr spapr_node0_size(MachineState *machine)
395 {
396     if (nb_numa_nodes) {
397         int i;
398         for (i = 0; i < nb_numa_nodes; ++i) {
399             if (numa_info[i].node_mem) {
400                 return MIN(pow2floor(numa_info[i].node_mem),
401                            machine->ram_size);
402             }
403         }
404     }
405     return machine->ram_size;
406 }
407 
408 static void add_str(GString *s, const gchar *s1)
409 {
410     g_string_append_len(s, s1, strlen(s1) + 1);
411 }
412 
413 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
414                                        hwaddr size)
415 {
416     uint32_t associativity[] = {
417         cpu_to_be32(0x4), /* length */
418         cpu_to_be32(0x0), cpu_to_be32(0x0),
419         cpu_to_be32(0x0), cpu_to_be32(nodeid)
420     };
421     char mem_name[32];
422     uint64_t mem_reg_property[2];
423     int off;
424 
425     mem_reg_property[0] = cpu_to_be64(start);
426     mem_reg_property[1] = cpu_to_be64(size);
427 
428     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
429     off = fdt_add_subnode(fdt, 0, mem_name);
430     _FDT(off);
431     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
432     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
433                       sizeof(mem_reg_property))));
434     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
435                       sizeof(associativity))));
436     return off;
437 }
438 
439 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
440 {
441     MachineState *machine = MACHINE(spapr);
442     hwaddr mem_start, node_size;
443     int i, nb_nodes = nb_numa_nodes;
444     NodeInfo *nodes = numa_info;
445     NodeInfo ramnode;
446 
447     /* No NUMA nodes, assume there is just one node with whole RAM */
448     if (!nb_numa_nodes) {
449         nb_nodes = 1;
450         ramnode.node_mem = machine->ram_size;
451         nodes = &ramnode;
452     }
453 
454     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
455         if (!nodes[i].node_mem) {
456             continue;
457         }
458         if (mem_start >= machine->ram_size) {
459             node_size = 0;
460         } else {
461             node_size = nodes[i].node_mem;
462             if (node_size > machine->ram_size - mem_start) {
463                 node_size = machine->ram_size - mem_start;
464             }
465         }
466         if (!mem_start) {
467             /* ppc_spapr_init() checks for rma_size <= node0_size already */
468             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
469             mem_start += spapr->rma_size;
470             node_size -= spapr->rma_size;
471         }
472         for ( ; node_size; ) {
473             hwaddr sizetmp = pow2floor(node_size);
474 
475             /* mem_start != 0 here */
476             if (ctzl(mem_start) < ctzl(sizetmp)) {
477                 sizetmp = 1ULL << ctzl(mem_start);
478             }
479 
480             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
481             node_size -= sizetmp;
482             mem_start += sizetmp;
483         }
484     }
485 
486     return 0;
487 }
488 
489 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
490                                   sPAPRMachineState *spapr)
491 {
492     PowerPCCPU *cpu = POWERPC_CPU(cs);
493     CPUPPCState *env = &cpu->env;
494     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
495     int index = spapr_vcpu_id(cpu);
496     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
497                        0xffffffff, 0xffffffff};
498     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
499         : SPAPR_TIMEBASE_FREQ;
500     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
501     uint32_t page_sizes_prop[64];
502     size_t page_sizes_prop_size;
503     uint32_t vcpus_per_socket = smp_threads * smp_cores;
504     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
505     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
506     sPAPRDRConnector *drc;
507     int drc_index;
508     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
509     int i;
510 
511     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
512     if (drc) {
513         drc_index = spapr_drc_index(drc);
514         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
515     }
516 
517     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
518     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
519 
520     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
521     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
522                            env->dcache_line_size)));
523     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
524                            env->dcache_line_size)));
525     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
526                            env->icache_line_size)));
527     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
528                            env->icache_line_size)));
529 
530     if (pcc->l1_dcache_size) {
531         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
532                                pcc->l1_dcache_size)));
533     } else {
534         warn_report("Unknown L1 dcache size for cpu");
535     }
536     if (pcc->l1_icache_size) {
537         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
538                                pcc->l1_icache_size)));
539     } else {
540         warn_report("Unknown L1 icache size for cpu");
541     }
542 
543     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
544     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
545     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
546     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
547     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
548     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
549 
550     if (env->spr_cb[SPR_PURR].oea_read) {
551         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
552     }
553 
554     if (env->mmu_model & POWERPC_MMU_1TSEG) {
555         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
556                           segs, sizeof(segs))));
557     }
558 
559     /* Advertise VMX/VSX (vector extensions) if available
560      *   0 / no property == no vector extensions
561      *   1               == VMX / Altivec available
562      *   2               == VSX available */
563     if (env->insns_flags & PPC_ALTIVEC) {
564         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
565 
566         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
567     }
568 
569     /* Advertise DFP (Decimal Floating Point) if available
570      *   0 / no property == no DFP
571      *   1               == DFP available */
572     if (env->insns_flags2 & PPC2_DFP) {
573         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
574     }
575 
576     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
577                                                   sizeof(page_sizes_prop));
578     if (page_sizes_prop_size) {
579         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
580                           page_sizes_prop, page_sizes_prop_size)));
581     }
582 
583     spapr_populate_pa_features(env, fdt, offset, false);
584 
585     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
586                            cs->cpu_index / vcpus_per_socket)));
587 
588     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
589                       pft_size_prop, sizeof(pft_size_prop))));
590 
591     if (nb_numa_nodes > 1) {
592         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
593     }
594 
595     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
596 
597     if (pcc->radix_page_info) {
598         for (i = 0; i < pcc->radix_page_info->count; i++) {
599             radix_AP_encodings[i] =
600                 cpu_to_be32(pcc->radix_page_info->entries[i]);
601         }
602         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
603                           radix_AP_encodings,
604                           pcc->radix_page_info->count *
605                           sizeof(radix_AP_encodings[0]))));
606     }
607 }
608 
609 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
610 {
611     CPUState *cs;
612     int cpus_offset;
613     char *nodename;
614     int smt = kvmppc_smt_threads();
615 
616     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
617     _FDT(cpus_offset);
618     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
619     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
620 
621     /*
622      * We walk the CPUs in reverse order to ensure that CPU DT nodes
623      * created by fdt_add_subnode() end up in the right order in FDT
624      * for the guest kernel the enumerate the CPUs correctly.
625      */
626     CPU_FOREACH_REVERSE(cs) {
627         PowerPCCPU *cpu = POWERPC_CPU(cs);
628         int index = spapr_vcpu_id(cpu);
629         DeviceClass *dc = DEVICE_GET_CLASS(cs);
630         int offset;
631 
632         if ((index % smt) != 0) {
633             continue;
634         }
635 
636         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
637         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
638         g_free(nodename);
639         _FDT(offset);
640         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
641     }
642 
643 }
644 
645 /*
646  * Adds ibm,dynamic-reconfiguration-memory node.
647  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
648  * of this device tree node.
649  */
650 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
651 {
652     MachineState *machine = MACHINE(spapr);
653     int ret, i, offset;
654     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
655     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
656     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
657     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
658                        memory_region_size(&spapr->hotplug_memory.mr)) /
659                        lmb_size;
660     uint32_t *int_buf, *cur_index, buf_len;
661     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
662 
663     /*
664      * Don't create the node if there is no hotpluggable memory
665      */
666     if (machine->ram_size == machine->maxram_size) {
667         return 0;
668     }
669 
670     /*
671      * Allocate enough buffer size to fit in ibm,dynamic-memory
672      * or ibm,associativity-lookup-arrays
673      */
674     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
675               * sizeof(uint32_t);
676     cur_index = int_buf = g_malloc0(buf_len);
677 
678     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
679 
680     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
681                     sizeof(prop_lmb_size));
682     if (ret < 0) {
683         goto out;
684     }
685 
686     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
687     if (ret < 0) {
688         goto out;
689     }
690 
691     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
692     if (ret < 0) {
693         goto out;
694     }
695 
696     /* ibm,dynamic-memory */
697     int_buf[0] = cpu_to_be32(nr_lmbs);
698     cur_index++;
699     for (i = 0; i < nr_lmbs; i++) {
700         uint64_t addr = i * lmb_size;
701         uint32_t *dynamic_memory = cur_index;
702 
703         if (i >= hotplug_lmb_start) {
704             sPAPRDRConnector *drc;
705 
706             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
707             g_assert(drc);
708 
709             dynamic_memory[0] = cpu_to_be32(addr >> 32);
710             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
711             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
712             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
713             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
714             if (memory_region_present(get_system_memory(), addr)) {
715                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
716             } else {
717                 dynamic_memory[5] = cpu_to_be32(0);
718             }
719         } else {
720             /*
721              * LMB information for RMA, boot time RAM and gap b/n RAM and
722              * hotplug memory region -- all these are marked as reserved
723              * and as having no valid DRC.
724              */
725             dynamic_memory[0] = cpu_to_be32(addr >> 32);
726             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
727             dynamic_memory[2] = cpu_to_be32(0);
728             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
729             dynamic_memory[4] = cpu_to_be32(-1);
730             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
731                                             SPAPR_LMB_FLAGS_DRC_INVALID);
732         }
733 
734         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
735     }
736     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
737     if (ret < 0) {
738         goto out;
739     }
740 
741     /* ibm,associativity-lookup-arrays */
742     cur_index = int_buf;
743     int_buf[0] = cpu_to_be32(nr_nodes);
744     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
745     cur_index += 2;
746     for (i = 0; i < nr_nodes; i++) {
747         uint32_t associativity[] = {
748             cpu_to_be32(0x0),
749             cpu_to_be32(0x0),
750             cpu_to_be32(0x0),
751             cpu_to_be32(i)
752         };
753         memcpy(cur_index, associativity, sizeof(associativity));
754         cur_index += 4;
755     }
756     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
757             (cur_index - int_buf) * sizeof(uint32_t));
758 out:
759     g_free(int_buf);
760     return ret;
761 }
762 
763 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
764                                 sPAPROptionVector *ov5_updates)
765 {
766     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
767     int ret = 0, offset;
768 
769     /* Generate ibm,dynamic-reconfiguration-memory node if required */
770     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
771         g_assert(smc->dr_lmb_enabled);
772         ret = spapr_populate_drconf_memory(spapr, fdt);
773         if (ret) {
774             goto out;
775         }
776     }
777 
778     offset = fdt_path_offset(fdt, "/chosen");
779     if (offset < 0) {
780         offset = fdt_add_subnode(fdt, 0, "chosen");
781         if (offset < 0) {
782             return offset;
783         }
784     }
785     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
786                                  "ibm,architecture-vec-5");
787 
788 out:
789     return ret;
790 }
791 
792 static bool spapr_hotplugged_dev_before_cas(void)
793 {
794     Object *drc_container, *obj;
795     ObjectProperty *prop;
796     ObjectPropertyIterator iter;
797 
798     drc_container = container_get(object_get_root(), "/dr-connector");
799     object_property_iter_init(&iter, drc_container);
800     while ((prop = object_property_iter_next(&iter))) {
801         if (!strstart(prop->type, "link<", NULL)) {
802             continue;
803         }
804         obj = object_property_get_link(drc_container, prop->name, NULL);
805         if (spapr_drc_needed(obj)) {
806             return true;
807         }
808     }
809     return false;
810 }
811 
812 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
813                                  target_ulong addr, target_ulong size,
814                                  sPAPROptionVector *ov5_updates)
815 {
816     void *fdt, *fdt_skel;
817     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
818 
819     if (spapr_hotplugged_dev_before_cas()) {
820         return 1;
821     }
822 
823     size -= sizeof(hdr);
824 
825     /* Create skeleton */
826     fdt_skel = g_malloc0(size);
827     _FDT((fdt_create(fdt_skel, size)));
828     _FDT((fdt_begin_node(fdt_skel, "")));
829     _FDT((fdt_end_node(fdt_skel)));
830     _FDT((fdt_finish(fdt_skel)));
831     fdt = g_malloc0(size);
832     _FDT((fdt_open_into(fdt_skel, fdt, size)));
833     g_free(fdt_skel);
834 
835     /* Fixup cpu nodes */
836     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
837 
838     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
839         return -1;
840     }
841 
842     /* Pack resulting tree */
843     _FDT((fdt_pack(fdt)));
844 
845     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
846         trace_spapr_cas_failed(size);
847         return -1;
848     }
849 
850     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
851     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
852     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
853     g_free(fdt);
854 
855     return 0;
856 }
857 
858 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
859 {
860     int rtas;
861     GString *hypertas = g_string_sized_new(256);
862     GString *qemu_hypertas = g_string_sized_new(256);
863     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
864     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
865         memory_region_size(&spapr->hotplug_memory.mr);
866     uint32_t lrdr_capacity[] = {
867         cpu_to_be32(max_hotplug_addr >> 32),
868         cpu_to_be32(max_hotplug_addr & 0xffffffff),
869         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
870         cpu_to_be32(max_cpus / smp_threads),
871     };
872 
873     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
874 
875     /* hypertas */
876     add_str(hypertas, "hcall-pft");
877     add_str(hypertas, "hcall-term");
878     add_str(hypertas, "hcall-dabr");
879     add_str(hypertas, "hcall-interrupt");
880     add_str(hypertas, "hcall-tce");
881     add_str(hypertas, "hcall-vio");
882     add_str(hypertas, "hcall-splpar");
883     add_str(hypertas, "hcall-bulk");
884     add_str(hypertas, "hcall-set-mode");
885     add_str(hypertas, "hcall-sprg0");
886     add_str(hypertas, "hcall-copy");
887     add_str(hypertas, "hcall-debug");
888     add_str(qemu_hypertas, "hcall-memop1");
889 
890     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
891         add_str(hypertas, "hcall-multi-tce");
892     }
893 
894     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
895         add_str(hypertas, "hcall-hpt-resize");
896     }
897 
898     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
899                      hypertas->str, hypertas->len));
900     g_string_free(hypertas, TRUE);
901     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
902                      qemu_hypertas->str, qemu_hypertas->len));
903     g_string_free(qemu_hypertas, TRUE);
904 
905     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
906                      refpoints, sizeof(refpoints)));
907 
908     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
909                           RTAS_ERROR_LOG_MAX));
910     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
911                           RTAS_EVENT_SCAN_RATE));
912 
913     if (msi_nonbroken) {
914         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
915     }
916 
917     /*
918      * According to PAPR, rtas ibm,os-term does not guarantee a return
919      * back to the guest cpu.
920      *
921      * While an additional ibm,extended-os-term property indicates
922      * that rtas call return will always occur. Set this property.
923      */
924     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
925 
926     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
927                      lrdr_capacity, sizeof(lrdr_capacity)));
928 
929     spapr_dt_rtas_tokens(fdt, rtas);
930 }
931 
932 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
933  * that the guest may request and thus the valid values for bytes 24..26 of
934  * option vector 5: */
935 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
936 {
937     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
938 
939     char val[2 * 4] = {
940         23, 0x00, /* Xive mode, filled in below. */
941         24, 0x00, /* Hash/Radix, filled in below. */
942         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
943         26, 0x40, /* Radix options: GTSE == yes. */
944     };
945 
946     if (kvm_enabled()) {
947         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
948             val[3] = 0x80; /* OV5_MMU_BOTH */
949         } else if (kvmppc_has_cap_mmu_radix()) {
950             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
951         } else {
952             val[3] = 0x00; /* Hash */
953         }
954     } else {
955         if (first_ppc_cpu->env.mmu_model & POWERPC_MMU_V3) {
956             /* V3 MMU supports both hash and radix (with dynamic switching) */
957             val[3] = 0xC0;
958         } else {
959             /* Otherwise we can only do hash */
960             val[3] = 0x00;
961         }
962     }
963     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
964                      val, sizeof(val)));
965 }
966 
967 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
968 {
969     MachineState *machine = MACHINE(spapr);
970     int chosen;
971     const char *boot_device = machine->boot_order;
972     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
973     size_t cb = 0;
974     char *bootlist = get_boot_devices_list(&cb, true);
975 
976     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
977 
978     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
979     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
980                           spapr->initrd_base));
981     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
982                           spapr->initrd_base + spapr->initrd_size));
983 
984     if (spapr->kernel_size) {
985         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
986                               cpu_to_be64(spapr->kernel_size) };
987 
988         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
989                          &kprop, sizeof(kprop)));
990         if (spapr->kernel_le) {
991             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
992         }
993     }
994     if (boot_menu) {
995         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
996     }
997     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
998     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
999     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1000 
1001     if (cb && bootlist) {
1002         int i;
1003 
1004         for (i = 0; i < cb; i++) {
1005             if (bootlist[i] == '\n') {
1006                 bootlist[i] = ' ';
1007             }
1008         }
1009         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1010     }
1011 
1012     if (boot_device && strlen(boot_device)) {
1013         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1014     }
1015 
1016     if (!spapr->has_graphics && stdout_path) {
1017         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1018     }
1019 
1020     spapr_dt_ov5_platform_support(fdt, chosen);
1021 
1022     g_free(stdout_path);
1023     g_free(bootlist);
1024 }
1025 
1026 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1027 {
1028     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1029      * KVM to work under pHyp with some guest co-operation */
1030     int hypervisor;
1031     uint8_t hypercall[16];
1032 
1033     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1034     /* indicate KVM hypercall interface */
1035     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1036     if (kvmppc_has_cap_fixup_hcalls()) {
1037         /*
1038          * Older KVM versions with older guest kernels were broken
1039          * with the magic page, don't allow the guest to map it.
1040          */
1041         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1042                                   sizeof(hypercall))) {
1043             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1044                              hypercall, sizeof(hypercall)));
1045         }
1046     }
1047 }
1048 
1049 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1050                              hwaddr rtas_addr,
1051                              hwaddr rtas_size)
1052 {
1053     MachineState *machine = MACHINE(spapr);
1054     MachineClass *mc = MACHINE_GET_CLASS(machine);
1055     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1056     int ret;
1057     void *fdt;
1058     sPAPRPHBState *phb;
1059     char *buf;
1060 
1061     fdt = g_malloc0(FDT_MAX_SIZE);
1062     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1063 
1064     /* Root node */
1065     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1066     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1067     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1068 
1069     /*
1070      * Add info to guest to indentify which host is it being run on
1071      * and what is the uuid of the guest
1072      */
1073     if (kvmppc_get_host_model(&buf)) {
1074         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1075         g_free(buf);
1076     }
1077     if (kvmppc_get_host_serial(&buf)) {
1078         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1079         g_free(buf);
1080     }
1081 
1082     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1083 
1084     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1085     if (qemu_uuid_set) {
1086         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1087     }
1088     g_free(buf);
1089 
1090     if (qemu_get_vm_name()) {
1091         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1092                                 qemu_get_vm_name()));
1093     }
1094 
1095     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1096     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1097 
1098     /* /interrupt controller */
1099     spapr_dt_xics(xics_max_server_number(), fdt, PHANDLE_XICP);
1100 
1101     ret = spapr_populate_memory(spapr, fdt);
1102     if (ret < 0) {
1103         error_report("couldn't setup memory nodes in fdt");
1104         exit(1);
1105     }
1106 
1107     /* /vdevice */
1108     spapr_dt_vdevice(spapr->vio_bus, fdt);
1109 
1110     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1111         ret = spapr_rng_populate_dt(fdt);
1112         if (ret < 0) {
1113             error_report("could not set up rng device in the fdt");
1114             exit(1);
1115         }
1116     }
1117 
1118     QLIST_FOREACH(phb, &spapr->phbs, list) {
1119         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1120         if (ret < 0) {
1121             error_report("couldn't setup PCI devices in fdt");
1122             exit(1);
1123         }
1124     }
1125 
1126     /* cpus */
1127     spapr_populate_cpus_dt_node(fdt, spapr);
1128 
1129     if (smc->dr_lmb_enabled) {
1130         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1131     }
1132 
1133     if (mc->has_hotpluggable_cpus) {
1134         int offset = fdt_path_offset(fdt, "/cpus");
1135         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1136                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1137         if (ret < 0) {
1138             error_report("Couldn't set up CPU DR device tree properties");
1139             exit(1);
1140         }
1141     }
1142 
1143     /* /event-sources */
1144     spapr_dt_events(spapr, fdt);
1145 
1146     /* /rtas */
1147     spapr_dt_rtas(spapr, fdt);
1148 
1149     /* /chosen */
1150     spapr_dt_chosen(spapr, fdt);
1151 
1152     /* /hypervisor */
1153     if (kvm_enabled()) {
1154         spapr_dt_hypervisor(spapr, fdt);
1155     }
1156 
1157     /* Build memory reserve map */
1158     if (spapr->kernel_size) {
1159         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1160     }
1161     if (spapr->initrd_size) {
1162         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1163     }
1164 
1165     /* ibm,client-architecture-support updates */
1166     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1167     if (ret < 0) {
1168         error_report("couldn't setup CAS properties fdt");
1169         exit(1);
1170     }
1171 
1172     return fdt;
1173 }
1174 
1175 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1176 {
1177     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1178 }
1179 
1180 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1181                                     PowerPCCPU *cpu)
1182 {
1183     CPUPPCState *env = &cpu->env;
1184 
1185     /* The TCG path should also be holding the BQL at this point */
1186     g_assert(qemu_mutex_iothread_locked());
1187 
1188     if (msr_pr) {
1189         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1190         env->gpr[3] = H_PRIVILEGE;
1191     } else {
1192         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1193     }
1194 }
1195 
1196 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1197 {
1198     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1199 
1200     return spapr->patb_entry;
1201 }
1202 
1203 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1204 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1205 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1206 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1207 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1208 
1209 /*
1210  * Get the fd to access the kernel htab, re-opening it if necessary
1211  */
1212 static int get_htab_fd(sPAPRMachineState *spapr)
1213 {
1214     Error *local_err = NULL;
1215 
1216     if (spapr->htab_fd >= 0) {
1217         return spapr->htab_fd;
1218     }
1219 
1220     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1221     if (spapr->htab_fd < 0) {
1222         error_report_err(local_err);
1223     }
1224 
1225     return spapr->htab_fd;
1226 }
1227 
1228 void close_htab_fd(sPAPRMachineState *spapr)
1229 {
1230     if (spapr->htab_fd >= 0) {
1231         close(spapr->htab_fd);
1232     }
1233     spapr->htab_fd = -1;
1234 }
1235 
1236 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1237 {
1238     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1239 
1240     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1241 }
1242 
1243 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1244 {
1245     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1246 
1247     assert(kvm_enabled());
1248 
1249     if (!spapr->htab) {
1250         return 0;
1251     }
1252 
1253     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1254 }
1255 
1256 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1257                                                 hwaddr ptex, int n)
1258 {
1259     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1260     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1261 
1262     if (!spapr->htab) {
1263         /*
1264          * HTAB is controlled by KVM. Fetch into temporary buffer
1265          */
1266         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1267         kvmppc_read_hptes(hptes, ptex, n);
1268         return hptes;
1269     }
1270 
1271     /*
1272      * HTAB is controlled by QEMU. Just point to the internally
1273      * accessible PTEG.
1274      */
1275     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1276 }
1277 
1278 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1279                               const ppc_hash_pte64_t *hptes,
1280                               hwaddr ptex, int n)
1281 {
1282     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1283 
1284     if (!spapr->htab) {
1285         g_free((void *)hptes);
1286     }
1287 
1288     /* Nothing to do for qemu managed HPT */
1289 }
1290 
1291 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1292                              uint64_t pte0, uint64_t pte1)
1293 {
1294     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1295     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1296 
1297     if (!spapr->htab) {
1298         kvmppc_write_hpte(ptex, pte0, pte1);
1299     } else {
1300         stq_p(spapr->htab + offset, pte0);
1301         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1302     }
1303 }
1304 
1305 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1306 {
1307     int shift;
1308 
1309     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1310      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1311      * that's much more than is needed for Linux guests */
1312     shift = ctz64(pow2ceil(ramsize)) - 7;
1313     shift = MAX(shift, 18); /* Minimum architected size */
1314     shift = MIN(shift, 46); /* Maximum architected size */
1315     return shift;
1316 }
1317 
1318 void spapr_free_hpt(sPAPRMachineState *spapr)
1319 {
1320     g_free(spapr->htab);
1321     spapr->htab = NULL;
1322     spapr->htab_shift = 0;
1323     close_htab_fd(spapr);
1324 }
1325 
1326 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1327                           Error **errp)
1328 {
1329     long rc;
1330 
1331     /* Clean up any HPT info from a previous boot */
1332     spapr_free_hpt(spapr);
1333 
1334     rc = kvmppc_reset_htab(shift);
1335     if (rc < 0) {
1336         /* kernel-side HPT needed, but couldn't allocate one */
1337         error_setg_errno(errp, errno,
1338                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1339                          shift);
1340         /* This is almost certainly fatal, but if the caller really
1341          * wants to carry on with shift == 0, it's welcome to try */
1342     } else if (rc > 0) {
1343         /* kernel-side HPT allocated */
1344         if (rc != shift) {
1345             error_setg(errp,
1346                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1347                        shift, rc);
1348         }
1349 
1350         spapr->htab_shift = shift;
1351         spapr->htab = NULL;
1352     } else {
1353         /* kernel-side HPT not needed, allocate in userspace instead */
1354         size_t size = 1ULL << shift;
1355         int i;
1356 
1357         spapr->htab = qemu_memalign(size, size);
1358         if (!spapr->htab) {
1359             error_setg_errno(errp, errno,
1360                              "Could not allocate HPT of order %d", shift);
1361             return;
1362         }
1363 
1364         memset(spapr->htab, 0, size);
1365         spapr->htab_shift = shift;
1366 
1367         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1368             DIRTY_HPTE(HPTE(spapr->htab, i));
1369         }
1370     }
1371 }
1372 
1373 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1374 {
1375     int hpt_shift;
1376 
1377     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1378         || (spapr->cas_reboot
1379             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1380         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1381     } else {
1382         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->ram_size);
1383     }
1384     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1385 
1386     if (spapr->vrma_adjust) {
1387         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1388                                           spapr->htab_shift);
1389     }
1390     /* We're setting up a hash table, so that means we're not radix */
1391     spapr->patb_entry = 0;
1392 }
1393 
1394 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1395 {
1396     bool matched = false;
1397 
1398     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1399         matched = true;
1400     }
1401 
1402     if (!matched) {
1403         error_report("Device %s is not supported by this machine yet.",
1404                      qdev_fw_name(DEVICE(sbdev)));
1405         exit(1);
1406     }
1407 }
1408 
1409 static void ppc_spapr_reset(void)
1410 {
1411     MachineState *machine = MACHINE(qdev_get_machine());
1412     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1413     PowerPCCPU *first_ppc_cpu;
1414     uint32_t rtas_limit;
1415     hwaddr rtas_addr, fdt_addr;
1416     void *fdt;
1417     int rc;
1418 
1419     /* Check for unknown sysbus devices */
1420     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1421 
1422     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1423         /* If using KVM with radix mode available, VCPUs can be started
1424          * without a HPT because KVM will start them in radix mode.
1425          * Set the GR bit in PATB so that we know there is no HPT. */
1426         spapr->patb_entry = PATBE1_GR;
1427     } else {
1428         spapr_setup_hpt_and_vrma(spapr);
1429     }
1430 
1431     qemu_devices_reset();
1432     spapr_clear_pending_events(spapr);
1433 
1434     /*
1435      * We place the device tree and RTAS just below either the top of the RMA,
1436      * or just below 2GB, whichever is lowere, so that it can be
1437      * processed with 32-bit real mode code if necessary
1438      */
1439     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1440     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1441     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1442 
1443     /* if this reset wasn't generated by CAS, we should reset our
1444      * negotiated options and start from scratch */
1445     if (!spapr->cas_reboot) {
1446         spapr_ovec_cleanup(spapr->ov5_cas);
1447         spapr->ov5_cas = spapr_ovec_new();
1448 
1449         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1450     }
1451 
1452     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1453 
1454     spapr_load_rtas(spapr, fdt, rtas_addr);
1455 
1456     rc = fdt_pack(fdt);
1457 
1458     /* Should only fail if we've built a corrupted tree */
1459     assert(rc == 0);
1460 
1461     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1462         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1463                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1464         exit(1);
1465     }
1466 
1467     /* Load the fdt */
1468     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1469     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1470     g_free(fdt);
1471 
1472     /* Set up the entry state */
1473     first_ppc_cpu = POWERPC_CPU(first_cpu);
1474     first_ppc_cpu->env.gpr[3] = fdt_addr;
1475     first_ppc_cpu->env.gpr[5] = 0;
1476     first_cpu->halted = 0;
1477     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1478 
1479     spapr->cas_reboot = false;
1480 }
1481 
1482 static void spapr_create_nvram(sPAPRMachineState *spapr)
1483 {
1484     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1485     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1486 
1487     if (dinfo) {
1488         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1489                             &error_fatal);
1490     }
1491 
1492     qdev_init_nofail(dev);
1493 
1494     spapr->nvram = (struct sPAPRNVRAM *)dev;
1495 }
1496 
1497 static void spapr_rtc_create(sPAPRMachineState *spapr)
1498 {
1499     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1500     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1501                               &error_fatal);
1502     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1503                               &error_fatal);
1504     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1505                               "date", &error_fatal);
1506 }
1507 
1508 /* Returns whether we want to use VGA or not */
1509 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1510 {
1511     switch (vga_interface_type) {
1512     case VGA_NONE:
1513         return false;
1514     case VGA_DEVICE:
1515         return true;
1516     case VGA_STD:
1517     case VGA_VIRTIO:
1518         return pci_vga_init(pci_bus) != NULL;
1519     default:
1520         error_setg(errp,
1521                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1522         return false;
1523     }
1524 }
1525 
1526 static int spapr_post_load(void *opaque, int version_id)
1527 {
1528     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1529     int err = 0;
1530 
1531     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1532         CPUState *cs;
1533         CPU_FOREACH(cs) {
1534             PowerPCCPU *cpu = POWERPC_CPU(cs);
1535             icp_resend(ICP(cpu->intc));
1536         }
1537     }
1538 
1539     /* In earlier versions, there was no separate qdev for the PAPR
1540      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1541      * So when migrating from those versions, poke the incoming offset
1542      * value into the RTC device */
1543     if (version_id < 3) {
1544         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1545     }
1546 
1547     if (spapr->patb_entry) {
1548         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1549         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1550         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1551 
1552         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1553         if (err) {
1554             error_report("Process table config unsupported by the host");
1555             return -EINVAL;
1556         }
1557     }
1558 
1559     return err;
1560 }
1561 
1562 static bool version_before_3(void *opaque, int version_id)
1563 {
1564     return version_id < 3;
1565 }
1566 
1567 static bool spapr_pending_events_needed(void *opaque)
1568 {
1569     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1570     return !QTAILQ_EMPTY(&spapr->pending_events);
1571 }
1572 
1573 static const VMStateDescription vmstate_spapr_event_entry = {
1574     .name = "spapr_event_log_entry",
1575     .version_id = 1,
1576     .minimum_version_id = 1,
1577     .fields = (VMStateField[]) {
1578         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1579         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1580         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1581                                      NULL, extended_length),
1582         VMSTATE_END_OF_LIST()
1583     },
1584 };
1585 
1586 static const VMStateDescription vmstate_spapr_pending_events = {
1587     .name = "spapr_pending_events",
1588     .version_id = 1,
1589     .minimum_version_id = 1,
1590     .needed = spapr_pending_events_needed,
1591     .fields = (VMStateField[]) {
1592         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1593                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1594         VMSTATE_END_OF_LIST()
1595     },
1596 };
1597 
1598 static bool spapr_ov5_cas_needed(void *opaque)
1599 {
1600     sPAPRMachineState *spapr = opaque;
1601     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1602     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1603     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1604     bool cas_needed;
1605 
1606     /* Prior to the introduction of sPAPROptionVector, we had two option
1607      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1608      * Both of these options encode machine topology into the device-tree
1609      * in such a way that the now-booted OS should still be able to interact
1610      * appropriately with QEMU regardless of what options were actually
1611      * negotiatied on the source side.
1612      *
1613      * As such, we can avoid migrating the CAS-negotiated options if these
1614      * are the only options available on the current machine/platform.
1615      * Since these are the only options available for pseries-2.7 and
1616      * earlier, this allows us to maintain old->new/new->old migration
1617      * compatibility.
1618      *
1619      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1620      * via default pseries-2.8 machines and explicit command-line parameters.
1621      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1622      * of the actual CAS-negotiated values to continue working properly. For
1623      * example, availability of memory unplug depends on knowing whether
1624      * OV5_HP_EVT was negotiated via CAS.
1625      *
1626      * Thus, for any cases where the set of available CAS-negotiatable
1627      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1628      * include the CAS-negotiated options in the migration stream.
1629      */
1630     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1631     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1632 
1633     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1634      * the mask itself since in the future it's possible "legacy" bits may be
1635      * removed via machine options, which could generate a false positive
1636      * that breaks migration.
1637      */
1638     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1639     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1640 
1641     spapr_ovec_cleanup(ov5_mask);
1642     spapr_ovec_cleanup(ov5_legacy);
1643     spapr_ovec_cleanup(ov5_removed);
1644 
1645     return cas_needed;
1646 }
1647 
1648 static const VMStateDescription vmstate_spapr_ov5_cas = {
1649     .name = "spapr_option_vector_ov5_cas",
1650     .version_id = 1,
1651     .minimum_version_id = 1,
1652     .needed = spapr_ov5_cas_needed,
1653     .fields = (VMStateField[]) {
1654         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1655                                  vmstate_spapr_ovec, sPAPROptionVector),
1656         VMSTATE_END_OF_LIST()
1657     },
1658 };
1659 
1660 static bool spapr_patb_entry_needed(void *opaque)
1661 {
1662     sPAPRMachineState *spapr = opaque;
1663 
1664     return !!spapr->patb_entry;
1665 }
1666 
1667 static const VMStateDescription vmstate_spapr_patb_entry = {
1668     .name = "spapr_patb_entry",
1669     .version_id = 1,
1670     .minimum_version_id = 1,
1671     .needed = spapr_patb_entry_needed,
1672     .fields = (VMStateField[]) {
1673         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1674         VMSTATE_END_OF_LIST()
1675     },
1676 };
1677 
1678 static const VMStateDescription vmstate_spapr = {
1679     .name = "spapr",
1680     .version_id = 3,
1681     .minimum_version_id = 1,
1682     .post_load = spapr_post_load,
1683     .fields = (VMStateField[]) {
1684         /* used to be @next_irq */
1685         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1686 
1687         /* RTC offset */
1688         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1689 
1690         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1691         VMSTATE_END_OF_LIST()
1692     },
1693     .subsections = (const VMStateDescription*[]) {
1694         &vmstate_spapr_ov5_cas,
1695         &vmstate_spapr_patb_entry,
1696         &vmstate_spapr_pending_events,
1697         NULL
1698     }
1699 };
1700 
1701 static int htab_save_setup(QEMUFile *f, void *opaque)
1702 {
1703     sPAPRMachineState *spapr = opaque;
1704 
1705     /* "Iteration" header */
1706     if (!spapr->htab_shift) {
1707         qemu_put_be32(f, -1);
1708     } else {
1709         qemu_put_be32(f, spapr->htab_shift);
1710     }
1711 
1712     if (spapr->htab) {
1713         spapr->htab_save_index = 0;
1714         spapr->htab_first_pass = true;
1715     } else {
1716         if (spapr->htab_shift) {
1717             assert(kvm_enabled());
1718         }
1719     }
1720 
1721 
1722     return 0;
1723 }
1724 
1725 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1726                             int chunkstart, int n_valid, int n_invalid)
1727 {
1728     qemu_put_be32(f, chunkstart);
1729     qemu_put_be16(f, n_valid);
1730     qemu_put_be16(f, n_invalid);
1731     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1732                     HASH_PTE_SIZE_64 * n_valid);
1733 }
1734 
1735 static void htab_save_end_marker(QEMUFile *f)
1736 {
1737     qemu_put_be32(f, 0);
1738     qemu_put_be16(f, 0);
1739     qemu_put_be16(f, 0);
1740 }
1741 
1742 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1743                                  int64_t max_ns)
1744 {
1745     bool has_timeout = max_ns != -1;
1746     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1747     int index = spapr->htab_save_index;
1748     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1749 
1750     assert(spapr->htab_first_pass);
1751 
1752     do {
1753         int chunkstart;
1754 
1755         /* Consume invalid HPTEs */
1756         while ((index < htabslots)
1757                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1758             CLEAN_HPTE(HPTE(spapr->htab, index));
1759             index++;
1760         }
1761 
1762         /* Consume valid HPTEs */
1763         chunkstart = index;
1764         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1765                && HPTE_VALID(HPTE(spapr->htab, index))) {
1766             CLEAN_HPTE(HPTE(spapr->htab, index));
1767             index++;
1768         }
1769 
1770         if (index > chunkstart) {
1771             int n_valid = index - chunkstart;
1772 
1773             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
1774 
1775             if (has_timeout &&
1776                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1777                 break;
1778             }
1779         }
1780     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1781 
1782     if (index >= htabslots) {
1783         assert(index == htabslots);
1784         index = 0;
1785         spapr->htab_first_pass = false;
1786     }
1787     spapr->htab_save_index = index;
1788 }
1789 
1790 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1791                                 int64_t max_ns)
1792 {
1793     bool final = max_ns < 0;
1794     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1795     int examined = 0, sent = 0;
1796     int index = spapr->htab_save_index;
1797     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1798 
1799     assert(!spapr->htab_first_pass);
1800 
1801     do {
1802         int chunkstart, invalidstart;
1803 
1804         /* Consume non-dirty HPTEs */
1805         while ((index < htabslots)
1806                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1807             index++;
1808             examined++;
1809         }
1810 
1811         chunkstart = index;
1812         /* Consume valid dirty HPTEs */
1813         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1814                && HPTE_DIRTY(HPTE(spapr->htab, index))
1815                && HPTE_VALID(HPTE(spapr->htab, index))) {
1816             CLEAN_HPTE(HPTE(spapr->htab, index));
1817             index++;
1818             examined++;
1819         }
1820 
1821         invalidstart = index;
1822         /* Consume invalid dirty HPTEs */
1823         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1824                && HPTE_DIRTY(HPTE(spapr->htab, index))
1825                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1826             CLEAN_HPTE(HPTE(spapr->htab, index));
1827             index++;
1828             examined++;
1829         }
1830 
1831         if (index > chunkstart) {
1832             int n_valid = invalidstart - chunkstart;
1833             int n_invalid = index - invalidstart;
1834 
1835             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
1836             sent += index - chunkstart;
1837 
1838             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1839                 break;
1840             }
1841         }
1842 
1843         if (examined >= htabslots) {
1844             break;
1845         }
1846 
1847         if (index >= htabslots) {
1848             assert(index == htabslots);
1849             index = 0;
1850         }
1851     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1852 
1853     if (index >= htabslots) {
1854         assert(index == htabslots);
1855         index = 0;
1856     }
1857 
1858     spapr->htab_save_index = index;
1859 
1860     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1861 }
1862 
1863 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1864 #define MAX_KVM_BUF_SIZE    2048
1865 
1866 static int htab_save_iterate(QEMUFile *f, void *opaque)
1867 {
1868     sPAPRMachineState *spapr = opaque;
1869     int fd;
1870     int rc = 0;
1871 
1872     /* Iteration header */
1873     if (!spapr->htab_shift) {
1874         qemu_put_be32(f, -1);
1875         return 1;
1876     } else {
1877         qemu_put_be32(f, 0);
1878     }
1879 
1880     if (!spapr->htab) {
1881         assert(kvm_enabled());
1882 
1883         fd = get_htab_fd(spapr);
1884         if (fd < 0) {
1885             return fd;
1886         }
1887 
1888         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1889         if (rc < 0) {
1890             return rc;
1891         }
1892     } else  if (spapr->htab_first_pass) {
1893         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1894     } else {
1895         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1896     }
1897 
1898     htab_save_end_marker(f);
1899 
1900     return rc;
1901 }
1902 
1903 static int htab_save_complete(QEMUFile *f, void *opaque)
1904 {
1905     sPAPRMachineState *spapr = opaque;
1906     int fd;
1907 
1908     /* Iteration header */
1909     if (!spapr->htab_shift) {
1910         qemu_put_be32(f, -1);
1911         return 0;
1912     } else {
1913         qemu_put_be32(f, 0);
1914     }
1915 
1916     if (!spapr->htab) {
1917         int rc;
1918 
1919         assert(kvm_enabled());
1920 
1921         fd = get_htab_fd(spapr);
1922         if (fd < 0) {
1923             return fd;
1924         }
1925 
1926         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1927         if (rc < 0) {
1928             return rc;
1929         }
1930     } else {
1931         if (spapr->htab_first_pass) {
1932             htab_save_first_pass(f, spapr, -1);
1933         }
1934         htab_save_later_pass(f, spapr, -1);
1935     }
1936 
1937     /* End marker */
1938     htab_save_end_marker(f);
1939 
1940     return 0;
1941 }
1942 
1943 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1944 {
1945     sPAPRMachineState *spapr = opaque;
1946     uint32_t section_hdr;
1947     int fd = -1;
1948     Error *local_err = NULL;
1949 
1950     if (version_id < 1 || version_id > 1) {
1951         error_report("htab_load() bad version");
1952         return -EINVAL;
1953     }
1954 
1955     section_hdr = qemu_get_be32(f);
1956 
1957     if (section_hdr == -1) {
1958         spapr_free_hpt(spapr);
1959         return 0;
1960     }
1961 
1962     if (section_hdr) {
1963         /* First section gives the htab size */
1964         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1965         if (local_err) {
1966             error_report_err(local_err);
1967             return -EINVAL;
1968         }
1969         return 0;
1970     }
1971 
1972     if (!spapr->htab) {
1973         assert(kvm_enabled());
1974 
1975         fd = kvmppc_get_htab_fd(true, 0, &local_err);
1976         if (fd < 0) {
1977             error_report_err(local_err);
1978             return fd;
1979         }
1980     }
1981 
1982     while (true) {
1983         uint32_t index;
1984         uint16_t n_valid, n_invalid;
1985 
1986         index = qemu_get_be32(f);
1987         n_valid = qemu_get_be16(f);
1988         n_invalid = qemu_get_be16(f);
1989 
1990         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1991             /* End of Stream */
1992             break;
1993         }
1994 
1995         if ((index + n_valid + n_invalid) >
1996             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1997             /* Bad index in stream */
1998             error_report(
1999                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2000                 index, n_valid, n_invalid, spapr->htab_shift);
2001             return -EINVAL;
2002         }
2003 
2004         if (spapr->htab) {
2005             if (n_valid) {
2006                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2007                                 HASH_PTE_SIZE_64 * n_valid);
2008             }
2009             if (n_invalid) {
2010                 memset(HPTE(spapr->htab, index + n_valid), 0,
2011                        HASH_PTE_SIZE_64 * n_invalid);
2012             }
2013         } else {
2014             int rc;
2015 
2016             assert(fd >= 0);
2017 
2018             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2019             if (rc < 0) {
2020                 return rc;
2021             }
2022         }
2023     }
2024 
2025     if (!spapr->htab) {
2026         assert(fd >= 0);
2027         close(fd);
2028     }
2029 
2030     return 0;
2031 }
2032 
2033 static void htab_save_cleanup(void *opaque)
2034 {
2035     sPAPRMachineState *spapr = opaque;
2036 
2037     close_htab_fd(spapr);
2038 }
2039 
2040 static SaveVMHandlers savevm_htab_handlers = {
2041     .save_setup = htab_save_setup,
2042     .save_live_iterate = htab_save_iterate,
2043     .save_live_complete_precopy = htab_save_complete,
2044     .save_cleanup = htab_save_cleanup,
2045     .load_state = htab_load,
2046 };
2047 
2048 static void spapr_boot_set(void *opaque, const char *boot_device,
2049                            Error **errp)
2050 {
2051     MachineState *machine = MACHINE(opaque);
2052     machine->boot_order = g_strdup(boot_device);
2053 }
2054 
2055 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2056 {
2057     MachineState *machine = MACHINE(spapr);
2058     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2059     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2060     int i;
2061 
2062     for (i = 0; i < nr_lmbs; i++) {
2063         uint64_t addr;
2064 
2065         addr = i * lmb_size + spapr->hotplug_memory.base;
2066         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2067                                addr / lmb_size);
2068     }
2069 }
2070 
2071 /*
2072  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2073  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2074  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2075  */
2076 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2077 {
2078     int i;
2079 
2080     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2081         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2082                    " is not aligned to %llu MiB",
2083                    machine->ram_size,
2084                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2085         return;
2086     }
2087 
2088     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2089         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2090                    " is not aligned to %llu MiB",
2091                    machine->ram_size,
2092                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2093         return;
2094     }
2095 
2096     for (i = 0; i < nb_numa_nodes; i++) {
2097         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2098             error_setg(errp,
2099                        "Node %d memory size 0x%" PRIx64
2100                        " is not aligned to %llu MiB",
2101                        i, numa_info[i].node_mem,
2102                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2103             return;
2104         }
2105     }
2106 }
2107 
2108 /* find cpu slot in machine->possible_cpus by core_id */
2109 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2110 {
2111     int index = id / smp_threads;
2112 
2113     if (index >= ms->possible_cpus->len) {
2114         return NULL;
2115     }
2116     if (idx) {
2117         *idx = index;
2118     }
2119     return &ms->possible_cpus->cpus[index];
2120 }
2121 
2122 static void spapr_init_cpus(sPAPRMachineState *spapr)
2123 {
2124     MachineState *machine = MACHINE(spapr);
2125     MachineClass *mc = MACHINE_GET_CLASS(machine);
2126     char *type = spapr_get_cpu_core_type(machine->cpu_model);
2127     int smt = kvmppc_smt_threads();
2128     const CPUArchIdList *possible_cpus;
2129     int boot_cores_nr = smp_cpus / smp_threads;
2130     int i;
2131 
2132     if (!type) {
2133         error_report("Unable to find sPAPR CPU Core definition");
2134         exit(1);
2135     }
2136 
2137     possible_cpus = mc->possible_cpu_arch_ids(machine);
2138     if (mc->has_hotpluggable_cpus) {
2139         if (smp_cpus % smp_threads) {
2140             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2141                          smp_cpus, smp_threads);
2142             exit(1);
2143         }
2144         if (max_cpus % smp_threads) {
2145             error_report("max_cpus (%u) must be multiple of threads (%u)",
2146                          max_cpus, smp_threads);
2147             exit(1);
2148         }
2149     } else {
2150         if (max_cpus != smp_cpus) {
2151             error_report("This machine version does not support CPU hotplug");
2152             exit(1);
2153         }
2154         boot_cores_nr = possible_cpus->len;
2155     }
2156 
2157     for (i = 0; i < possible_cpus->len; i++) {
2158         int core_id = i * smp_threads;
2159 
2160         if (mc->has_hotpluggable_cpus) {
2161             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2162                                    (core_id / smp_threads) * smt);
2163         }
2164 
2165         if (i < boot_cores_nr) {
2166             Object *core  = object_new(type);
2167             int nr_threads = smp_threads;
2168 
2169             /* Handle the partially filled core for older machine types */
2170             if ((i + 1) * smp_threads >= smp_cpus) {
2171                 nr_threads = smp_cpus - i * smp_threads;
2172             }
2173 
2174             object_property_set_int(core, nr_threads, "nr-threads",
2175                                     &error_fatal);
2176             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2177                                     &error_fatal);
2178             object_property_set_bool(core, true, "realized", &error_fatal);
2179         }
2180     }
2181     g_free(type);
2182 }
2183 
2184 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2185 {
2186     Error *local_err = NULL;
2187     bool vsmt_user = !!spapr->vsmt;
2188     int kvm_smt = kvmppc_smt_threads();
2189     int ret;
2190 
2191     if (!kvm_enabled() && (smp_threads > 1)) {
2192         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2193                      "on a pseries machine");
2194         goto out;
2195     }
2196     if (!is_power_of_2(smp_threads)) {
2197         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2198                      "machine because it must be a power of 2", smp_threads);
2199         goto out;
2200     }
2201 
2202     /* Detemine the VSMT mode to use: */
2203     if (vsmt_user) {
2204         if (spapr->vsmt < smp_threads) {
2205             error_setg(&local_err, "Cannot support VSMT mode %d"
2206                          " because it must be >= threads/core (%d)",
2207                          spapr->vsmt, smp_threads);
2208             goto out;
2209         }
2210         /* In this case, spapr->vsmt has been set by the command line */
2211     } else {
2212         /* Choose a VSMT mode that may be higher than necessary but is
2213          * likely to be compatible with hosts that don't have VSMT. */
2214         spapr->vsmt = MAX(kvm_smt, smp_threads);
2215     }
2216 
2217     /* KVM: If necessary, set the SMT mode: */
2218     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2219         ret = kvmppc_set_smt_threads(spapr->vsmt);
2220         if (ret) {
2221             error_setg(&local_err,
2222                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2223                        spapr->vsmt, ret);
2224             if (!vsmt_user) {
2225                 error_append_hint(&local_err, "On PPC, a VM with %d threads/"
2226                              "core on a host with %d threads/core requires "
2227                              " the use of VSMT mode %d.\n",
2228                              smp_threads, kvm_smt, spapr->vsmt);
2229             }
2230             kvmppc_hint_smt_possible(&local_err);
2231             goto out;
2232         }
2233     }
2234     /* else TCG: nothing to do currently */
2235 out:
2236     error_propagate(errp, local_err);
2237 }
2238 
2239 /* pSeries LPAR / sPAPR hardware init */
2240 static void ppc_spapr_init(MachineState *machine)
2241 {
2242     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2243     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2244     const char *kernel_filename = machine->kernel_filename;
2245     const char *initrd_filename = machine->initrd_filename;
2246     PCIHostState *phb;
2247     int i;
2248     MemoryRegion *sysmem = get_system_memory();
2249     MemoryRegion *ram = g_new(MemoryRegion, 1);
2250     MemoryRegion *rma_region;
2251     void *rma = NULL;
2252     hwaddr rma_alloc_size;
2253     hwaddr node0_size = spapr_node0_size(machine);
2254     long load_limit, fw_size;
2255     char *filename;
2256     Error *resize_hpt_err = NULL;
2257 
2258     msi_nonbroken = true;
2259 
2260     QLIST_INIT(&spapr->phbs);
2261     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2262 
2263     /* Check HPT resizing availability */
2264     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2265     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2266         /*
2267          * If the user explicitly requested a mode we should either
2268          * supply it, or fail completely (which we do below).  But if
2269          * it's not set explicitly, we reset our mode to something
2270          * that works
2271          */
2272         if (resize_hpt_err) {
2273             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2274             error_free(resize_hpt_err);
2275             resize_hpt_err = NULL;
2276         } else {
2277             spapr->resize_hpt = smc->resize_hpt_default;
2278         }
2279     }
2280 
2281     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2282 
2283     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2284         /*
2285          * User requested HPT resize, but this host can't supply it.  Bail out
2286          */
2287         error_report_err(resize_hpt_err);
2288         exit(1);
2289     }
2290 
2291     /* Allocate RMA if necessary */
2292     rma_alloc_size = kvmppc_alloc_rma(&rma);
2293 
2294     if (rma_alloc_size == -1) {
2295         error_report("Unable to create RMA");
2296         exit(1);
2297     }
2298 
2299     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2300         spapr->rma_size = rma_alloc_size;
2301     } else {
2302         spapr->rma_size = node0_size;
2303 
2304         /* With KVM, we don't actually know whether KVM supports an
2305          * unbounded RMA (PR KVM) or is limited by the hash table size
2306          * (HV KVM using VRMA), so we always assume the latter
2307          *
2308          * In that case, we also limit the initial allocations for RTAS
2309          * etc... to 256M since we have no way to know what the VRMA size
2310          * is going to be as it depends on the size of the hash table
2311          * isn't determined yet.
2312          */
2313         if (kvm_enabled()) {
2314             spapr->vrma_adjust = 1;
2315             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2316         }
2317 
2318         /* Actually we don't support unbounded RMA anymore since we
2319          * added proper emulation of HV mode. The max we can get is
2320          * 16G which also happens to be what we configure for PAPR
2321          * mode so make sure we don't do anything bigger than that
2322          */
2323         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2324     }
2325 
2326     if (spapr->rma_size > node0_size) {
2327         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2328                      spapr->rma_size);
2329         exit(1);
2330     }
2331 
2332     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2333     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2334 
2335     /* Set up Interrupt Controller before we create the VCPUs */
2336     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2337 
2338     /* Set up containers for ibm,client-set-architecture negotiated options */
2339     spapr->ov5 = spapr_ovec_new();
2340     spapr->ov5_cas = spapr_ovec_new();
2341 
2342     if (smc->dr_lmb_enabled) {
2343         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2344         spapr_validate_node_memory(machine, &error_fatal);
2345     }
2346 
2347     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2348     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2349         /* KVM and TCG always allow GTSE with radix... */
2350         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2351     }
2352     /* ... but not with hash (currently). */
2353 
2354     /* advertise support for dedicated HP event source to guests */
2355     if (spapr->use_hotplug_event_source) {
2356         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2357     }
2358 
2359     /* advertise support for HPT resizing */
2360     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2361         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2362     }
2363 
2364     /* init CPUs */
2365     if (machine->cpu_model == NULL) {
2366         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2367     }
2368 
2369     spapr_cpu_parse_features(spapr);
2370 
2371     spapr_set_vsmt_mode(spapr, &error_fatal);
2372 
2373     spapr_init_cpus(spapr);
2374 
2375     if (kvm_enabled()) {
2376         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2377         kvmppc_enable_logical_ci_hcalls();
2378         kvmppc_enable_set_mode_hcall();
2379 
2380         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2381         kvmppc_enable_clear_ref_mod_hcalls();
2382     }
2383 
2384     /* allocate RAM */
2385     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2386                                          machine->ram_size);
2387     memory_region_add_subregion(sysmem, 0, ram);
2388 
2389     if (rma_alloc_size && rma) {
2390         rma_region = g_new(MemoryRegion, 1);
2391         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2392                                    rma_alloc_size, rma);
2393         vmstate_register_ram_global(rma_region);
2394         memory_region_add_subregion(sysmem, 0, rma_region);
2395     }
2396 
2397     /* initialize hotplug memory address space */
2398     if (machine->ram_size < machine->maxram_size) {
2399         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2400         /*
2401          * Limit the number of hotpluggable memory slots to half the number
2402          * slots that KVM supports, leaving the other half for PCI and other
2403          * devices. However ensure that number of slots doesn't drop below 32.
2404          */
2405         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2406                            SPAPR_MAX_RAM_SLOTS;
2407 
2408         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2409             max_memslots = SPAPR_MAX_RAM_SLOTS;
2410         }
2411         if (machine->ram_slots > max_memslots) {
2412             error_report("Specified number of memory slots %"
2413                          PRIu64" exceeds max supported %d",
2414                          machine->ram_slots, max_memslots);
2415             exit(1);
2416         }
2417 
2418         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2419                                               SPAPR_HOTPLUG_MEM_ALIGN);
2420         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2421                            "hotplug-memory", hotplug_mem_size);
2422         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2423                                     &spapr->hotplug_memory.mr);
2424     }
2425 
2426     if (smc->dr_lmb_enabled) {
2427         spapr_create_lmb_dr_connectors(spapr);
2428     }
2429 
2430     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2431     if (!filename) {
2432         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2433         exit(1);
2434     }
2435     spapr->rtas_size = get_image_size(filename);
2436     if (spapr->rtas_size < 0) {
2437         error_report("Could not get size of LPAR rtas '%s'", filename);
2438         exit(1);
2439     }
2440     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2441     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2442         error_report("Could not load LPAR rtas '%s'", filename);
2443         exit(1);
2444     }
2445     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2446         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2447                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2448         exit(1);
2449     }
2450     g_free(filename);
2451 
2452     /* Set up RTAS event infrastructure */
2453     spapr_events_init(spapr);
2454 
2455     /* Set up the RTC RTAS interfaces */
2456     spapr_rtc_create(spapr);
2457 
2458     /* Set up VIO bus */
2459     spapr->vio_bus = spapr_vio_bus_init();
2460 
2461     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2462         if (serial_hds[i]) {
2463             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2464         }
2465     }
2466 
2467     /* We always have at least the nvram device on VIO */
2468     spapr_create_nvram(spapr);
2469 
2470     /* Set up PCI */
2471     spapr_pci_rtas_init();
2472 
2473     phb = spapr_create_phb(spapr, 0);
2474 
2475     for (i = 0; i < nb_nics; i++) {
2476         NICInfo *nd = &nd_table[i];
2477 
2478         if (!nd->model) {
2479             nd->model = g_strdup("ibmveth");
2480         }
2481 
2482         if (strcmp(nd->model, "ibmveth") == 0) {
2483             spapr_vlan_create(spapr->vio_bus, nd);
2484         } else {
2485             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2486         }
2487     }
2488 
2489     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2490         spapr_vscsi_create(spapr->vio_bus);
2491     }
2492 
2493     /* Graphics */
2494     if (spapr_vga_init(phb->bus, &error_fatal)) {
2495         spapr->has_graphics = true;
2496         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2497     }
2498 
2499     if (machine->usb) {
2500         if (smc->use_ohci_by_default) {
2501             pci_create_simple(phb->bus, -1, "pci-ohci");
2502         } else {
2503             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2504         }
2505 
2506         if (spapr->has_graphics) {
2507             USBBus *usb_bus = usb_bus_find(-1);
2508 
2509             usb_create_simple(usb_bus, "usb-kbd");
2510             usb_create_simple(usb_bus, "usb-mouse");
2511         }
2512     }
2513 
2514     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2515         error_report(
2516             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2517             MIN_RMA_SLOF);
2518         exit(1);
2519     }
2520 
2521     if (kernel_filename) {
2522         uint64_t lowaddr = 0;
2523 
2524         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2525                                       NULL, NULL, &lowaddr, NULL, 1,
2526                                       PPC_ELF_MACHINE, 0, 0);
2527         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2528             spapr->kernel_size = load_elf(kernel_filename,
2529                                           translate_kernel_address, NULL, NULL,
2530                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2531                                           0, 0);
2532             spapr->kernel_le = spapr->kernel_size > 0;
2533         }
2534         if (spapr->kernel_size < 0) {
2535             error_report("error loading %s: %s", kernel_filename,
2536                          load_elf_strerror(spapr->kernel_size));
2537             exit(1);
2538         }
2539 
2540         /* load initrd */
2541         if (initrd_filename) {
2542             /* Try to locate the initrd in the gap between the kernel
2543              * and the firmware. Add a bit of space just in case
2544              */
2545             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2546                                   + 0x1ffff) & ~0xffff;
2547             spapr->initrd_size = load_image_targphys(initrd_filename,
2548                                                      spapr->initrd_base,
2549                                                      load_limit
2550                                                      - spapr->initrd_base);
2551             if (spapr->initrd_size < 0) {
2552                 error_report("could not load initial ram disk '%s'",
2553                              initrd_filename);
2554                 exit(1);
2555             }
2556         }
2557     }
2558 
2559     if (bios_name == NULL) {
2560         bios_name = FW_FILE_NAME;
2561     }
2562     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2563     if (!filename) {
2564         error_report("Could not find LPAR firmware '%s'", bios_name);
2565         exit(1);
2566     }
2567     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2568     if (fw_size <= 0) {
2569         error_report("Could not load LPAR firmware '%s'", filename);
2570         exit(1);
2571     }
2572     g_free(filename);
2573 
2574     /* FIXME: Should register things through the MachineState's qdev
2575      * interface, this is a legacy from the sPAPREnvironment structure
2576      * which predated MachineState but had a similar function */
2577     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2578     register_savevm_live(NULL, "spapr/htab", -1, 1,
2579                          &savevm_htab_handlers, spapr);
2580 
2581     qemu_register_boot_set(spapr_boot_set, spapr);
2582 
2583     if (kvm_enabled()) {
2584         /* to stop and start vmclock */
2585         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2586                                          &spapr->tb);
2587 
2588         kvmppc_spapr_enable_inkernel_multitce();
2589     }
2590 }
2591 
2592 static int spapr_kvm_type(const char *vm_type)
2593 {
2594     if (!vm_type) {
2595         return 0;
2596     }
2597 
2598     if (!strcmp(vm_type, "HV")) {
2599         return 1;
2600     }
2601 
2602     if (!strcmp(vm_type, "PR")) {
2603         return 2;
2604     }
2605 
2606     error_report("Unknown kvm-type specified '%s'", vm_type);
2607     exit(1);
2608 }
2609 
2610 /*
2611  * Implementation of an interface to adjust firmware path
2612  * for the bootindex property handling.
2613  */
2614 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2615                                    DeviceState *dev)
2616 {
2617 #define CAST(type, obj, name) \
2618     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2619     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2620     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2621     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2622 
2623     if (d) {
2624         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2625         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2626         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2627 
2628         if (spapr) {
2629             /*
2630              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2631              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2632              * in the top 16 bits of the 64-bit LUN
2633              */
2634             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2635             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2636                                    (uint64_t)id << 48);
2637         } else if (virtio) {
2638             /*
2639              * We use SRP luns of the form 01000000 | (target << 8) | lun
2640              * in the top 32 bits of the 64-bit LUN
2641              * Note: the quote above is from SLOF and it is wrong,
2642              * the actual binding is:
2643              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2644              */
2645             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2646             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2647                                    (uint64_t)id << 32);
2648         } else if (usb) {
2649             /*
2650              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2651              * in the top 32 bits of the 64-bit LUN
2652              */
2653             unsigned usb_port = atoi(usb->port->path);
2654             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2655             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2656                                    (uint64_t)id << 32);
2657         }
2658     }
2659 
2660     /*
2661      * SLOF probes the USB devices, and if it recognizes that the device is a
2662      * storage device, it changes its name to "storage" instead of "usb-host",
2663      * and additionally adds a child node for the SCSI LUN, so the correct
2664      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2665      */
2666     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2667         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2668         if (usb_host_dev_is_scsi_storage(usbdev)) {
2669             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2670         }
2671     }
2672 
2673     if (phb) {
2674         /* Replace "pci" with "pci@800000020000000" */
2675         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2676     }
2677 
2678     if (vsc) {
2679         /* Same logic as virtio above */
2680         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2681         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2682     }
2683 
2684     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2685         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2686         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2687         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2688     }
2689 
2690     return NULL;
2691 }
2692 
2693 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2694 {
2695     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2696 
2697     return g_strdup(spapr->kvm_type);
2698 }
2699 
2700 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2701 {
2702     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2703 
2704     g_free(spapr->kvm_type);
2705     spapr->kvm_type = g_strdup(value);
2706 }
2707 
2708 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2709 {
2710     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2711 
2712     return spapr->use_hotplug_event_source;
2713 }
2714 
2715 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2716                                             Error **errp)
2717 {
2718     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2719 
2720     spapr->use_hotplug_event_source = value;
2721 }
2722 
2723 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2724 {
2725     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2726 
2727     switch (spapr->resize_hpt) {
2728     case SPAPR_RESIZE_HPT_DEFAULT:
2729         return g_strdup("default");
2730     case SPAPR_RESIZE_HPT_DISABLED:
2731         return g_strdup("disabled");
2732     case SPAPR_RESIZE_HPT_ENABLED:
2733         return g_strdup("enabled");
2734     case SPAPR_RESIZE_HPT_REQUIRED:
2735         return g_strdup("required");
2736     }
2737     g_assert_not_reached();
2738 }
2739 
2740 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2741 {
2742     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2743 
2744     if (strcmp(value, "default") == 0) {
2745         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2746     } else if (strcmp(value, "disabled") == 0) {
2747         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2748     } else if (strcmp(value, "enabled") == 0) {
2749         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2750     } else if (strcmp(value, "required") == 0) {
2751         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2752     } else {
2753         error_setg(errp, "Bad value for \"resize-hpt\" property");
2754     }
2755 }
2756 
2757 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
2758                                    void *opaque, Error **errp)
2759 {
2760     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2761 }
2762 
2763 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
2764                                    void *opaque, Error **errp)
2765 {
2766     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
2767 }
2768 
2769 static void spapr_machine_initfn(Object *obj)
2770 {
2771     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2772 
2773     spapr->htab_fd = -1;
2774     spapr->use_hotplug_event_source = true;
2775     object_property_add_str(obj, "kvm-type",
2776                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2777     object_property_set_description(obj, "kvm-type",
2778                                     "Specifies the KVM virtualization mode (HV, PR)",
2779                                     NULL);
2780     object_property_add_bool(obj, "modern-hotplug-events",
2781                             spapr_get_modern_hotplug_events,
2782                             spapr_set_modern_hotplug_events,
2783                             NULL);
2784     object_property_set_description(obj, "modern-hotplug-events",
2785                                     "Use dedicated hotplug event mechanism in"
2786                                     " place of standard EPOW events when possible"
2787                                     " (required for memory hot-unplug support)",
2788                                     NULL);
2789 
2790     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
2791                             "Maximum permitted CPU compatibility mode",
2792                             &error_fatal);
2793 
2794     object_property_add_str(obj, "resize-hpt",
2795                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
2796     object_property_set_description(obj, "resize-hpt",
2797                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
2798                                     NULL);
2799     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
2800                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
2801     object_property_set_description(obj, "vsmt",
2802                                     "Virtual SMT: KVM behaves as if this were"
2803                                     " the host's SMT mode", &error_abort);
2804 }
2805 
2806 static void spapr_machine_finalizefn(Object *obj)
2807 {
2808     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2809 
2810     g_free(spapr->kvm_type);
2811 }
2812 
2813 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2814 {
2815     cpu_synchronize_state(cs);
2816     ppc_cpu_do_system_reset(cs);
2817 }
2818 
2819 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2820 {
2821     CPUState *cs;
2822 
2823     CPU_FOREACH(cs) {
2824         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2825     }
2826 }
2827 
2828 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2829                            uint32_t node, bool dedicated_hp_event_source,
2830                            Error **errp)
2831 {
2832     sPAPRDRConnector *drc;
2833     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2834     int i, fdt_offset, fdt_size;
2835     void *fdt;
2836     uint64_t addr = addr_start;
2837     bool hotplugged = spapr_drc_hotplugged(dev);
2838     Error *local_err = NULL;
2839 
2840     for (i = 0; i < nr_lmbs; i++) {
2841         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2842                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2843         g_assert(drc);
2844 
2845         fdt = create_device_tree(&fdt_size);
2846         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2847                                                 SPAPR_MEMORY_BLOCK_SIZE);
2848 
2849         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
2850         if (local_err) {
2851             while (addr > addr_start) {
2852                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
2853                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2854                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
2855                 spapr_drc_detach(drc);
2856             }
2857             g_free(fdt);
2858             error_propagate(errp, local_err);
2859             return;
2860         }
2861         if (!hotplugged) {
2862             spapr_drc_reset(drc);
2863         }
2864         addr += SPAPR_MEMORY_BLOCK_SIZE;
2865     }
2866     /* send hotplug notification to the
2867      * guest only in case of hotplugged memory
2868      */
2869     if (hotplugged) {
2870         if (dedicated_hp_event_source) {
2871             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2872                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2873             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2874                                                    nr_lmbs,
2875                                                    spapr_drc_index(drc));
2876         } else {
2877             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2878                                            nr_lmbs);
2879         }
2880     }
2881 }
2882 
2883 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2884                               uint32_t node, Error **errp)
2885 {
2886     Error *local_err = NULL;
2887     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2888     PCDIMMDevice *dimm = PC_DIMM(dev);
2889     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2890     MemoryRegion *mr;
2891     uint64_t align, size, addr;
2892 
2893     mr = ddc->get_memory_region(dimm, &local_err);
2894     if (local_err) {
2895         goto out;
2896     }
2897     align = memory_region_get_alignment(mr);
2898     size = memory_region_size(mr);
2899 
2900     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2901     if (local_err) {
2902         goto out;
2903     }
2904 
2905     addr = object_property_get_uint(OBJECT(dimm),
2906                                     PC_DIMM_ADDR_PROP, &local_err);
2907     if (local_err) {
2908         goto out_unplug;
2909     }
2910 
2911     spapr_add_lmbs(dev, addr, size, node,
2912                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2913                    &local_err);
2914     if (local_err) {
2915         goto out_unplug;
2916     }
2917 
2918     return;
2919 
2920 out_unplug:
2921     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2922 out:
2923     error_propagate(errp, local_err);
2924 }
2925 
2926 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2927                                   Error **errp)
2928 {
2929     PCDIMMDevice *dimm = PC_DIMM(dev);
2930     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2931     MemoryRegion *mr;
2932     uint64_t size;
2933     char *mem_dev;
2934 
2935     mr = ddc->get_memory_region(dimm, errp);
2936     if (!mr) {
2937         return;
2938     }
2939     size = memory_region_size(mr);
2940 
2941     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2942         error_setg(errp, "Hotplugged memory size must be a multiple of "
2943                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2944         return;
2945     }
2946 
2947     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2948     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2949         error_setg(errp, "Memory backend has bad page size. "
2950                    "Use 'memory-backend-file' with correct mem-path.");
2951         goto out;
2952     }
2953 
2954 out:
2955     g_free(mem_dev);
2956 }
2957 
2958 struct sPAPRDIMMState {
2959     PCDIMMDevice *dimm;
2960     uint32_t nr_lmbs;
2961     QTAILQ_ENTRY(sPAPRDIMMState) next;
2962 };
2963 
2964 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
2965                                                        PCDIMMDevice *dimm)
2966 {
2967     sPAPRDIMMState *dimm_state = NULL;
2968 
2969     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
2970         if (dimm_state->dimm == dimm) {
2971             break;
2972         }
2973     }
2974     return dimm_state;
2975 }
2976 
2977 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
2978                                                       uint32_t nr_lmbs,
2979                                                       PCDIMMDevice *dimm)
2980 {
2981     sPAPRDIMMState *ds = NULL;
2982 
2983     /*
2984      * If this request is for a DIMM whose removal had failed earlier
2985      * (due to guest's refusal to remove the LMBs), we would have this
2986      * dimm already in the pending_dimm_unplugs list. In that
2987      * case don't add again.
2988      */
2989     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
2990     if (!ds) {
2991         ds = g_malloc0(sizeof(sPAPRDIMMState));
2992         ds->nr_lmbs = nr_lmbs;
2993         ds->dimm = dimm;
2994         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
2995     }
2996     return ds;
2997 }
2998 
2999 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3000                                               sPAPRDIMMState *dimm_state)
3001 {
3002     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3003     g_free(dimm_state);
3004 }
3005 
3006 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3007                                                         PCDIMMDevice *dimm)
3008 {
3009     sPAPRDRConnector *drc;
3010     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3011     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3012     uint64_t size = memory_region_size(mr);
3013     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3014     uint32_t avail_lmbs = 0;
3015     uint64_t addr_start, addr;
3016     int i;
3017 
3018     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3019                                          &error_abort);
3020 
3021     addr = addr_start;
3022     for (i = 0; i < nr_lmbs; i++) {
3023         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3024                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3025         g_assert(drc);
3026         if (drc->dev) {
3027             avail_lmbs++;
3028         }
3029         addr += SPAPR_MEMORY_BLOCK_SIZE;
3030     }
3031 
3032     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3033 }
3034 
3035 /* Callback to be called during DRC release. */
3036 void spapr_lmb_release(DeviceState *dev)
3037 {
3038     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_hotplug_handler(dev));
3039     PCDIMMDevice *dimm = PC_DIMM(dev);
3040     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3041     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3042     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3043 
3044     /* This information will get lost if a migration occurs
3045      * during the unplug process. In this case recover it. */
3046     if (ds == NULL) {
3047         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3048         g_assert(ds);
3049         /* The DRC being examined by the caller at least must be counted */
3050         g_assert(ds->nr_lmbs);
3051     }
3052 
3053     if (--ds->nr_lmbs) {
3054         return;
3055     }
3056 
3057     spapr_pending_dimm_unplugs_remove(spapr, ds);
3058 
3059     /*
3060      * Now that all the LMBs have been removed by the guest, call the
3061      * pc-dimm unplug handler to cleanup up the pc-dimm device.
3062      */
3063     pc_dimm_memory_unplug(dev, &spapr->hotplug_memory, mr);
3064     object_unparent(OBJECT(dev));
3065 }
3066 
3067 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3068                                         DeviceState *dev, Error **errp)
3069 {
3070     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3071     Error *local_err = NULL;
3072     PCDIMMDevice *dimm = PC_DIMM(dev);
3073     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3074     MemoryRegion *mr;
3075     uint32_t nr_lmbs;
3076     uint64_t size, addr_start, addr;
3077     int i;
3078     sPAPRDRConnector *drc;
3079 
3080     mr = ddc->get_memory_region(dimm, &local_err);
3081     if (local_err) {
3082         goto out;
3083     }
3084     size = memory_region_size(mr);
3085     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3086 
3087     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3088                                          &local_err);
3089     if (local_err) {
3090         goto out;
3091     }
3092 
3093     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3094 
3095     addr = addr_start;
3096     for (i = 0; i < nr_lmbs; i++) {
3097         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3098                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3099         g_assert(drc);
3100 
3101         spapr_drc_detach(drc);
3102         addr += SPAPR_MEMORY_BLOCK_SIZE;
3103     }
3104 
3105     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3106                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3107     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3108                                               nr_lmbs, spapr_drc_index(drc));
3109 out:
3110     error_propagate(errp, local_err);
3111 }
3112 
3113 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3114                                            sPAPRMachineState *spapr)
3115 {
3116     PowerPCCPU *cpu = POWERPC_CPU(cs);
3117     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3118     int id = spapr_vcpu_id(cpu);
3119     void *fdt;
3120     int offset, fdt_size;
3121     char *nodename;
3122 
3123     fdt = create_device_tree(&fdt_size);
3124     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3125     offset = fdt_add_subnode(fdt, 0, nodename);
3126 
3127     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3128     g_free(nodename);
3129 
3130     *fdt_offset = offset;
3131     return fdt;
3132 }
3133 
3134 /* Callback to be called during DRC release. */
3135 void spapr_core_release(DeviceState *dev)
3136 {
3137     MachineState *ms = MACHINE(qdev_get_hotplug_handler(dev));
3138     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3139     CPUCore *cc = CPU_CORE(dev);
3140     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3141 
3142     if (smc->pre_2_10_has_unused_icps) {
3143         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3144         sPAPRCPUCoreClass *scc = SPAPR_CPU_CORE_GET_CLASS(OBJECT(cc));
3145         const char *typename = object_class_get_name(scc->cpu_class);
3146         size_t size = object_type_get_instance_size(typename);
3147         int i;
3148 
3149         for (i = 0; i < cc->nr_threads; i++) {
3150             CPUState *cs = CPU(sc->threads + i * size);
3151 
3152             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3153         }
3154     }
3155 
3156     assert(core_slot);
3157     core_slot->cpu = NULL;
3158     object_unparent(OBJECT(dev));
3159 }
3160 
3161 static
3162 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3163                                Error **errp)
3164 {
3165     int index;
3166     sPAPRDRConnector *drc;
3167     CPUCore *cc = CPU_CORE(dev);
3168     int smt = kvmppc_smt_threads();
3169 
3170     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3171         error_setg(errp, "Unable to find CPU core with core-id: %d",
3172                    cc->core_id);
3173         return;
3174     }
3175     if (index == 0) {
3176         error_setg(errp, "Boot CPU core may not be unplugged");
3177         return;
3178     }
3179 
3180     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3181     g_assert(drc);
3182 
3183     spapr_drc_detach(drc);
3184 
3185     spapr_hotplug_req_remove_by_index(drc);
3186 }
3187 
3188 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3189                             Error **errp)
3190 {
3191     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3192     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3193     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3194     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3195     CPUCore *cc = CPU_CORE(dev);
3196     CPUState *cs = CPU(core->threads);
3197     sPAPRDRConnector *drc;
3198     Error *local_err = NULL;
3199     int smt = kvmppc_smt_threads();
3200     CPUArchId *core_slot;
3201     int index;
3202     bool hotplugged = spapr_drc_hotplugged(dev);
3203 
3204     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3205     if (!core_slot) {
3206         error_setg(errp, "Unable to find CPU core with core-id: %d",
3207                    cc->core_id);
3208         return;
3209     }
3210     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
3211 
3212     g_assert(drc || !mc->has_hotpluggable_cpus);
3213 
3214     if (drc) {
3215         void *fdt;
3216         int fdt_offset;
3217 
3218         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3219 
3220         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3221         if (local_err) {
3222             g_free(fdt);
3223             error_propagate(errp, local_err);
3224             return;
3225         }
3226 
3227         if (hotplugged) {
3228             /*
3229              * Send hotplug notification interrupt to the guest only
3230              * in case of hotplugged CPUs.
3231              */
3232             spapr_hotplug_req_add_by_index(drc);
3233         } else {
3234             spapr_drc_reset(drc);
3235         }
3236     }
3237 
3238     core_slot->cpu = OBJECT(dev);
3239 
3240     if (smc->pre_2_10_has_unused_icps) {
3241         sPAPRCPUCoreClass *scc = SPAPR_CPU_CORE_GET_CLASS(OBJECT(cc));
3242         const char *typename = object_class_get_name(scc->cpu_class);
3243         size_t size = object_type_get_instance_size(typename);
3244         int i;
3245 
3246         for (i = 0; i < cc->nr_threads; i++) {
3247             sPAPRCPUCore *sc = SPAPR_CPU_CORE(dev);
3248             void *obj = sc->threads + i * size;
3249 
3250             cs = CPU(obj);
3251             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3252         }
3253     }
3254 }
3255 
3256 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3257                                 Error **errp)
3258 {
3259     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3260     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3261     Error *local_err = NULL;
3262     CPUCore *cc = CPU_CORE(dev);
3263     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
3264     const char *type = object_get_typename(OBJECT(dev));
3265     CPUArchId *core_slot;
3266     int index;
3267 
3268     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3269         error_setg(&local_err, "CPU hotplug not supported for this machine");
3270         goto out;
3271     }
3272 
3273     if (strcmp(base_core_type, type)) {
3274         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3275         goto out;
3276     }
3277 
3278     if (cc->core_id % smp_threads) {
3279         error_setg(&local_err, "invalid core id %d", cc->core_id);
3280         goto out;
3281     }
3282 
3283     /*
3284      * In general we should have homogeneous threads-per-core, but old
3285      * (pre hotplug support) machine types allow the last core to have
3286      * reduced threads as a compatibility hack for when we allowed
3287      * total vcpus not a multiple of threads-per-core.
3288      */
3289     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3290         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3291                    cc->nr_threads, smp_threads);
3292         goto out;
3293     }
3294 
3295     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3296     if (!core_slot) {
3297         error_setg(&local_err, "core id %d out of range", cc->core_id);
3298         goto out;
3299     }
3300 
3301     if (core_slot->cpu) {
3302         error_setg(&local_err, "core %d already populated", cc->core_id);
3303         goto out;
3304     }
3305 
3306     numa_cpu_pre_plug(core_slot, dev, &local_err);
3307 
3308 out:
3309     g_free(base_core_type);
3310     error_propagate(errp, local_err);
3311 }
3312 
3313 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3314                                       DeviceState *dev, Error **errp)
3315 {
3316     MachineState *ms = MACHINE(hotplug_dev);
3317     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3318 
3319     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3320         int node;
3321 
3322         if (!smc->dr_lmb_enabled) {
3323             error_setg(errp, "Memory hotplug not supported for this machine");
3324             return;
3325         }
3326         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3327         if (*errp) {
3328             return;
3329         }
3330         if (node < 0 || node >= MAX_NODES) {
3331             error_setg(errp, "Invaild node %d", node);
3332             return;
3333         }
3334 
3335         /*
3336          * Currently PowerPC kernel doesn't allow hot-adding memory to
3337          * memory-less node, but instead will silently add the memory
3338          * to the first node that has some memory. This causes two
3339          * unexpected behaviours for the user.
3340          *
3341          * - Memory gets hotplugged to a different node than what the user
3342          *   specified.
3343          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3344          *   to memory-less node, a reboot will set things accordingly
3345          *   and the previously hotplugged memory now ends in the right node.
3346          *   This appears as if some memory moved from one node to another.
3347          *
3348          * So until kernel starts supporting memory hotplug to memory-less
3349          * nodes, just prevent such attempts upfront in QEMU.
3350          */
3351         if (nb_numa_nodes && !numa_info[node].node_mem) {
3352             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3353                        node);
3354             return;
3355         }
3356 
3357         spapr_memory_plug(hotplug_dev, dev, node, errp);
3358     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3359         spapr_core_plug(hotplug_dev, dev, errp);
3360     }
3361 }
3362 
3363 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3364                                                 DeviceState *dev, Error **errp)
3365 {
3366     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3367     MachineClass *mc = MACHINE_GET_CLASS(sms);
3368 
3369     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3370         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3371             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3372         } else {
3373             /* NOTE: this means there is a window after guest reset, prior to
3374              * CAS negotiation, where unplug requests will fail due to the
3375              * capability not being detected yet. This is a bit different than
3376              * the case with PCI unplug, where the events will be queued and
3377              * eventually handled by the guest after boot
3378              */
3379             error_setg(errp, "Memory hot unplug not supported for this guest");
3380         }
3381     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3382         if (!mc->has_hotpluggable_cpus) {
3383             error_setg(errp, "CPU hot unplug not supported on this machine");
3384             return;
3385         }
3386         spapr_core_unplug_request(hotplug_dev, dev, errp);
3387     }
3388 }
3389 
3390 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3391                                           DeviceState *dev, Error **errp)
3392 {
3393     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3394         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3395     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3396         spapr_core_pre_plug(hotplug_dev, dev, errp);
3397     }
3398 }
3399 
3400 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3401                                                  DeviceState *dev)
3402 {
3403     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3404         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3405         return HOTPLUG_HANDLER(machine);
3406     }
3407     return NULL;
3408 }
3409 
3410 static CpuInstanceProperties
3411 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3412 {
3413     CPUArchId *core_slot;
3414     MachineClass *mc = MACHINE_GET_CLASS(machine);
3415 
3416     /* make sure possible_cpu are intialized */
3417     mc->possible_cpu_arch_ids(machine);
3418     /* get CPU core slot containing thread that matches cpu_index */
3419     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3420     assert(core_slot);
3421     return core_slot->props;
3422 }
3423 
3424 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3425 {
3426     return idx / smp_cores % nb_numa_nodes;
3427 }
3428 
3429 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3430 {
3431     int i;
3432     int spapr_max_cores = max_cpus / smp_threads;
3433     MachineClass *mc = MACHINE_GET_CLASS(machine);
3434 
3435     if (!mc->has_hotpluggable_cpus) {
3436         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3437     }
3438     if (machine->possible_cpus) {
3439         assert(machine->possible_cpus->len == spapr_max_cores);
3440         return machine->possible_cpus;
3441     }
3442 
3443     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3444                              sizeof(CPUArchId) * spapr_max_cores);
3445     machine->possible_cpus->len = spapr_max_cores;
3446     for (i = 0; i < machine->possible_cpus->len; i++) {
3447         int core_id = i * smp_threads;
3448 
3449         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3450         machine->possible_cpus->cpus[i].arch_id = core_id;
3451         machine->possible_cpus->cpus[i].props.has_core_id = true;
3452         machine->possible_cpus->cpus[i].props.core_id = core_id;
3453     }
3454     return machine->possible_cpus;
3455 }
3456 
3457 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3458                                 uint64_t *buid, hwaddr *pio,
3459                                 hwaddr *mmio32, hwaddr *mmio64,
3460                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3461 {
3462     /*
3463      * New-style PHB window placement.
3464      *
3465      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3466      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3467      * windows.
3468      *
3469      * Some guest kernels can't work with MMIO windows above 1<<46
3470      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3471      *
3472      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3473      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3474      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3475      * 1TiB 64-bit MMIO windows for each PHB.
3476      */
3477     const uint64_t base_buid = 0x800000020000000ULL;
3478 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3479                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3480     int i;
3481 
3482     /* Sanity check natural alignments */
3483     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3484     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3485     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3486     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3487     /* Sanity check bounds */
3488     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3489                       SPAPR_PCI_MEM32_WIN_SIZE);
3490     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3491                       SPAPR_PCI_MEM64_WIN_SIZE);
3492 
3493     if (index >= SPAPR_MAX_PHBS) {
3494         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3495                    SPAPR_MAX_PHBS - 1);
3496         return;
3497     }
3498 
3499     *buid = base_buid + index;
3500     for (i = 0; i < n_dma; ++i) {
3501         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3502     }
3503 
3504     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3505     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3506     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3507 }
3508 
3509 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3510 {
3511     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3512 
3513     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3514 }
3515 
3516 static void spapr_ics_resend(XICSFabric *dev)
3517 {
3518     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3519 
3520     ics_resend(spapr->ics);
3521 }
3522 
3523 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3524 {
3525     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3526 
3527     return cpu ? ICP(cpu->intc) : NULL;
3528 }
3529 
3530 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3531                                  Monitor *mon)
3532 {
3533     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3534     CPUState *cs;
3535 
3536     CPU_FOREACH(cs) {
3537         PowerPCCPU *cpu = POWERPC_CPU(cs);
3538 
3539         icp_pic_print_info(ICP(cpu->intc), mon);
3540     }
3541 
3542     ics_pic_print_info(spapr->ics, mon);
3543 }
3544 
3545 int spapr_vcpu_id(PowerPCCPU *cpu)
3546 {
3547     CPUState *cs = CPU(cpu);
3548 
3549     if (kvm_enabled()) {
3550         return kvm_arch_vcpu_id(cs);
3551     } else {
3552         return cs->cpu_index;
3553     }
3554 }
3555 
3556 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3557 {
3558     CPUState *cs;
3559 
3560     CPU_FOREACH(cs) {
3561         PowerPCCPU *cpu = POWERPC_CPU(cs);
3562 
3563         if (spapr_vcpu_id(cpu) == vcpu_id) {
3564             return cpu;
3565         }
3566     }
3567 
3568     return NULL;
3569 }
3570 
3571 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3572 {
3573     MachineClass *mc = MACHINE_CLASS(oc);
3574     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3575     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3576     NMIClass *nc = NMI_CLASS(oc);
3577     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3578     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3579     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3580     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3581 
3582     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3583 
3584     /*
3585      * We set up the default / latest behaviour here.  The class_init
3586      * functions for the specific versioned machine types can override
3587      * these details for backwards compatibility
3588      */
3589     mc->init = ppc_spapr_init;
3590     mc->reset = ppc_spapr_reset;
3591     mc->block_default_type = IF_SCSI;
3592     mc->max_cpus = 1024;
3593     mc->no_parallel = 1;
3594     mc->default_boot_order = "";
3595     mc->default_ram_size = 512 * M_BYTE;
3596     mc->kvm_type = spapr_kvm_type;
3597     mc->has_dynamic_sysbus = true;
3598     mc->pci_allow_0_address = true;
3599     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3600     hc->pre_plug = spapr_machine_device_pre_plug;
3601     hc->plug = spapr_machine_device_plug;
3602     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3603     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3604     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3605     hc->unplug_request = spapr_machine_device_unplug_request;
3606 
3607     smc->dr_lmb_enabled = true;
3608     smc->tcg_default_cpu = "POWER8";
3609     mc->has_hotpluggable_cpus = true;
3610     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3611     fwc->get_dev_path = spapr_get_fw_dev_path;
3612     nc->nmi_monitor_handler = spapr_nmi;
3613     smc->phb_placement = spapr_phb_placement;
3614     vhc->hypercall = emulate_spapr_hypercall;
3615     vhc->hpt_mask = spapr_hpt_mask;
3616     vhc->map_hptes = spapr_map_hptes;
3617     vhc->unmap_hptes = spapr_unmap_hptes;
3618     vhc->store_hpte = spapr_store_hpte;
3619     vhc->get_patbe = spapr_get_patbe;
3620     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3621     xic->ics_get = spapr_ics_get;
3622     xic->ics_resend = spapr_ics_resend;
3623     xic->icp_get = spapr_icp_get;
3624     ispc->print_info = spapr_pic_print_info;
3625     /* Force NUMA node memory size to be a multiple of
3626      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3627      * in which LMBs are represented and hot-added
3628      */
3629     mc->numa_mem_align_shift = 28;
3630 }
3631 
3632 static const TypeInfo spapr_machine_info = {
3633     .name          = TYPE_SPAPR_MACHINE,
3634     .parent        = TYPE_MACHINE,
3635     .abstract      = true,
3636     .instance_size = sizeof(sPAPRMachineState),
3637     .instance_init = spapr_machine_initfn,
3638     .instance_finalize = spapr_machine_finalizefn,
3639     .class_size    = sizeof(sPAPRMachineClass),
3640     .class_init    = spapr_machine_class_init,
3641     .interfaces = (InterfaceInfo[]) {
3642         { TYPE_FW_PATH_PROVIDER },
3643         { TYPE_NMI },
3644         { TYPE_HOTPLUG_HANDLER },
3645         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3646         { TYPE_XICS_FABRIC },
3647         { TYPE_INTERRUPT_STATS_PROVIDER },
3648         { }
3649     },
3650 };
3651 
3652 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3653     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3654                                                     void *data)      \
3655     {                                                                \
3656         MachineClass *mc = MACHINE_CLASS(oc);                        \
3657         spapr_machine_##suffix##_class_options(mc);                  \
3658         if (latest) {                                                \
3659             mc->alias = "pseries";                                   \
3660             mc->is_default = 1;                                      \
3661         }                                                            \
3662     }                                                                \
3663     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3664     {                                                                \
3665         MachineState *machine = MACHINE(obj);                        \
3666         spapr_machine_##suffix##_instance_options(machine);          \
3667     }                                                                \
3668     static const TypeInfo spapr_machine_##suffix##_info = {          \
3669         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3670         .parent = TYPE_SPAPR_MACHINE,                                \
3671         .class_init = spapr_machine_##suffix##_class_init,           \
3672         .instance_init = spapr_machine_##suffix##_instance_init,     \
3673     };                                                               \
3674     static void spapr_machine_register_##suffix(void)                \
3675     {                                                                \
3676         type_register(&spapr_machine_##suffix##_info);               \
3677     }                                                                \
3678     type_init(spapr_machine_register_##suffix)
3679 
3680 /*
3681  * pseries-2.11
3682  */
3683 static void spapr_machine_2_11_instance_options(MachineState *machine)
3684 {
3685 }
3686 
3687 static void spapr_machine_2_11_class_options(MachineClass *mc)
3688 {
3689     /* Defaults for the latest behaviour inherited from the base class */
3690 }
3691 
3692 DEFINE_SPAPR_MACHINE(2_11, "2.11", true);
3693 
3694 /*
3695  * pseries-2.10
3696  */
3697 #define SPAPR_COMPAT_2_10                                              \
3698     HW_COMPAT_2_10                                                     \
3699 
3700 static void spapr_machine_2_10_instance_options(MachineState *machine)
3701 {
3702 }
3703 
3704 static void spapr_machine_2_10_class_options(MachineClass *mc)
3705 {
3706     spapr_machine_2_11_class_options(mc);
3707     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
3708 }
3709 
3710 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
3711 
3712 /*
3713  * pseries-2.9
3714  */
3715 #define SPAPR_COMPAT_2_9                                               \
3716     HW_COMPAT_2_9                                                      \
3717     {                                                                  \
3718         .driver = TYPE_POWERPC_CPU,                                    \
3719         .property = "pre-2.10-migration",                              \
3720         .value    = "on",                                              \
3721     },                                                                 \
3722 
3723 static void spapr_machine_2_9_instance_options(MachineState *machine)
3724 {
3725     spapr_machine_2_10_instance_options(machine);
3726 }
3727 
3728 static void spapr_machine_2_9_class_options(MachineClass *mc)
3729 {
3730     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3731 
3732     spapr_machine_2_10_class_options(mc);
3733     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3734     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
3735     smc->pre_2_10_has_unused_icps = true;
3736     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
3737 }
3738 
3739 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3740 
3741 /*
3742  * pseries-2.8
3743  */
3744 #define SPAPR_COMPAT_2_8                                        \
3745     HW_COMPAT_2_8                                               \
3746     {                                                           \
3747         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3748         .property = "pcie-extended-configuration-space",        \
3749         .value    = "off",                                      \
3750     },
3751 
3752 static void spapr_machine_2_8_instance_options(MachineState *machine)
3753 {
3754     spapr_machine_2_9_instance_options(machine);
3755 }
3756 
3757 static void spapr_machine_2_8_class_options(MachineClass *mc)
3758 {
3759     spapr_machine_2_9_class_options(mc);
3760     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3761     mc->numa_mem_align_shift = 23;
3762 }
3763 
3764 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3765 
3766 /*
3767  * pseries-2.7
3768  */
3769 #define SPAPR_COMPAT_2_7                            \
3770     HW_COMPAT_2_7                                   \
3771     {                                               \
3772         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3773         .property = "mem_win_size",                 \
3774         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3775     },                                              \
3776     {                                               \
3777         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3778         .property = "mem64_win_size",               \
3779         .value    = "0",                            \
3780     },                                              \
3781     {                                               \
3782         .driver = TYPE_POWERPC_CPU,                 \
3783         .property = "pre-2.8-migration",            \
3784         .value    = "on",                           \
3785     },                                              \
3786     {                                               \
3787         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3788         .property = "pre-2.8-migration",            \
3789         .value    = "on",                           \
3790     },
3791 
3792 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3793                               uint64_t *buid, hwaddr *pio,
3794                               hwaddr *mmio32, hwaddr *mmio64,
3795                               unsigned n_dma, uint32_t *liobns, Error **errp)
3796 {
3797     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3798     const uint64_t base_buid = 0x800000020000000ULL;
3799     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3800     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3801     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3802     const uint32_t max_index = 255;
3803     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3804 
3805     uint64_t ram_top = MACHINE(spapr)->ram_size;
3806     hwaddr phb0_base, phb_base;
3807     int i;
3808 
3809     /* Do we have hotpluggable memory? */
3810     if (MACHINE(spapr)->maxram_size > ram_top) {
3811         /* Can't just use maxram_size, because there may be an
3812          * alignment gap between normal and hotpluggable memory
3813          * regions */
3814         ram_top = spapr->hotplug_memory.base +
3815             memory_region_size(&spapr->hotplug_memory.mr);
3816     }
3817 
3818     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3819 
3820     if (index > max_index) {
3821         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3822                    max_index);
3823         return;
3824     }
3825 
3826     *buid = base_buid + index;
3827     for (i = 0; i < n_dma; ++i) {
3828         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3829     }
3830 
3831     phb_base = phb0_base + index * phb_spacing;
3832     *pio = phb_base + pio_offset;
3833     *mmio32 = phb_base + mmio_offset;
3834     /*
3835      * We don't set the 64-bit MMIO window, relying on the PHB's
3836      * fallback behaviour of automatically splitting a large "32-bit"
3837      * window into contiguous 32-bit and 64-bit windows
3838      */
3839 }
3840 
3841 static void spapr_machine_2_7_instance_options(MachineState *machine)
3842 {
3843     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3844 
3845     spapr_machine_2_8_instance_options(machine);
3846     spapr->use_hotplug_event_source = false;
3847 }
3848 
3849 static void spapr_machine_2_7_class_options(MachineClass *mc)
3850 {
3851     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3852 
3853     spapr_machine_2_8_class_options(mc);
3854     smc->tcg_default_cpu = "POWER7";
3855     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3856     smc->phb_placement = phb_placement_2_7;
3857 }
3858 
3859 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3860 
3861 /*
3862  * pseries-2.6
3863  */
3864 #define SPAPR_COMPAT_2_6 \
3865     HW_COMPAT_2_6 \
3866     { \
3867         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3868         .property = "ddw",\
3869         .value    = stringify(off),\
3870     },
3871 
3872 static void spapr_machine_2_6_instance_options(MachineState *machine)
3873 {
3874     spapr_machine_2_7_instance_options(machine);
3875 }
3876 
3877 static void spapr_machine_2_6_class_options(MachineClass *mc)
3878 {
3879     spapr_machine_2_7_class_options(mc);
3880     mc->has_hotpluggable_cpus = false;
3881     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3882 }
3883 
3884 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3885 
3886 /*
3887  * pseries-2.5
3888  */
3889 #define SPAPR_COMPAT_2_5 \
3890     HW_COMPAT_2_5 \
3891     { \
3892         .driver   = "spapr-vlan", \
3893         .property = "use-rx-buffer-pools", \
3894         .value    = "off", \
3895     },
3896 
3897 static void spapr_machine_2_5_instance_options(MachineState *machine)
3898 {
3899     spapr_machine_2_6_instance_options(machine);
3900 }
3901 
3902 static void spapr_machine_2_5_class_options(MachineClass *mc)
3903 {
3904     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3905 
3906     spapr_machine_2_6_class_options(mc);
3907     smc->use_ohci_by_default = true;
3908     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3909 }
3910 
3911 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3912 
3913 /*
3914  * pseries-2.4
3915  */
3916 #define SPAPR_COMPAT_2_4 \
3917         HW_COMPAT_2_4
3918 
3919 static void spapr_machine_2_4_instance_options(MachineState *machine)
3920 {
3921     spapr_machine_2_5_instance_options(machine);
3922 }
3923 
3924 static void spapr_machine_2_4_class_options(MachineClass *mc)
3925 {
3926     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3927 
3928     spapr_machine_2_5_class_options(mc);
3929     smc->dr_lmb_enabled = false;
3930     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3931 }
3932 
3933 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3934 
3935 /*
3936  * pseries-2.3
3937  */
3938 #define SPAPR_COMPAT_2_3 \
3939         HW_COMPAT_2_3 \
3940         {\
3941             .driver   = "spapr-pci-host-bridge",\
3942             .property = "dynamic-reconfiguration",\
3943             .value    = "off",\
3944         },
3945 
3946 static void spapr_machine_2_3_instance_options(MachineState *machine)
3947 {
3948     spapr_machine_2_4_instance_options(machine);
3949 }
3950 
3951 static void spapr_machine_2_3_class_options(MachineClass *mc)
3952 {
3953     spapr_machine_2_4_class_options(mc);
3954     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3955 }
3956 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3957 
3958 /*
3959  * pseries-2.2
3960  */
3961 
3962 #define SPAPR_COMPAT_2_2 \
3963         HW_COMPAT_2_2 \
3964         {\
3965             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3966             .property = "mem_win_size",\
3967             .value    = "0x20000000",\
3968         },
3969 
3970 static void spapr_machine_2_2_instance_options(MachineState *machine)
3971 {
3972     spapr_machine_2_3_instance_options(machine);
3973     machine->suppress_vmdesc = true;
3974 }
3975 
3976 static void spapr_machine_2_2_class_options(MachineClass *mc)
3977 {
3978     spapr_machine_2_3_class_options(mc);
3979     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3980 }
3981 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3982 
3983 /*
3984  * pseries-2.1
3985  */
3986 #define SPAPR_COMPAT_2_1 \
3987         HW_COMPAT_2_1
3988 
3989 static void spapr_machine_2_1_instance_options(MachineState *machine)
3990 {
3991     spapr_machine_2_2_instance_options(machine);
3992 }
3993 
3994 static void spapr_machine_2_1_class_options(MachineClass *mc)
3995 {
3996     spapr_machine_2_2_class_options(mc);
3997     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3998 }
3999 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4000 
4001 static void spapr_machine_register_types(void)
4002 {
4003     type_register_static(&spapr_machine_info);
4004 }
4005 
4006 type_init(spapr_machine_register_types)
4007