xref: /qemu/hw/ppc/spapr.c (revision 4dba87221936e18111959344b05ec3384e50fed0)
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 "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 
85 #include "monitor/monitor.h"
86 
87 #include <libfdt.h>
88 
89 /* SLOF memory layout:
90  *
91  * SLOF raw image loaded at 0, copies its romfs right below the flat
92  * device-tree, then position SLOF itself 31M below that
93  *
94  * So we set FW_OVERHEAD to 40MB which should account for all of that
95  * and more
96  *
97  * We load our kernel at 4M, leaving space for SLOF initial image
98  */
99 #define FDT_MAX_SIZE            0x100000
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            (128 * MiB)
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
206 {
207     int index = spapr_get_vcpu_id(cpu);
208     uint32_t associativity[] = {cpu_to_be32(0x5),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(0x0),
212                                 cpu_to_be32(cpu->node_id),
213                                 cpu_to_be32(index)};
214 
215     /* Advertise NUMA via ibm,associativity */
216     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217                           sizeof(associativity));
218 }
219 
220 /* Populate the "ibm,pa-features" property */
221 static void spapr_populate_pa_features(SpaprMachineState *spapr,
222                                        PowerPCCPU *cpu,
223                                        void *fdt, int offset)
224 {
225     uint8_t pa_features_206[] = { 6, 0,
226         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
227     uint8_t pa_features_207[] = { 24, 0,
228         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
229         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
230         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
231         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
232     uint8_t pa_features_300[] = { 66, 0,
233         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
234         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
235         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
236         /* 6: DS207 */
237         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
238         /* 16: Vector */
239         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
240         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
241         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
242         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
243         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
244         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
245         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
246         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
248         /* 42: PM, 44: PC RA, 46: SC vec'd */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
250         /* 48: SIMD, 50: QP BFP, 52: String */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
252         /* 54: DecFP, 56: DecI, 58: SHA */
253         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
254         /* 60: NM atomic, 62: RNG */
255         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
256     };
257     uint8_t *pa_features = NULL;
258     size_t pa_size;
259 
260     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
261         pa_features = pa_features_206;
262         pa_size = sizeof(pa_features_206);
263     }
264     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
265         pa_features = pa_features_207;
266         pa_size = sizeof(pa_features_207);
267     }
268     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
269         pa_features = pa_features_300;
270         pa_size = sizeof(pa_features_300);
271     }
272     if (!pa_features) {
273         return;
274     }
275 
276     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
277         /*
278          * Note: we keep CI large pages off by default because a 64K capable
279          * guest provisioned with large pages might otherwise try to map a qemu
280          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
281          * even if that qemu runs on a 4k host.
282          * We dd this bit back here if we are confident this is not an issue
283          */
284         pa_features[3] |= 0x20;
285     }
286     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
287         pa_features[24] |= 0x80;    /* Transactional memory support */
288     }
289     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
290         /* Workaround for broken kernels that attempt (guest) radix
291          * mode when they can't handle it, if they see the radix bit set
292          * in pa-features. So hide it from them. */
293         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
294     }
295 
296     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
297 }
298 
299 static hwaddr spapr_node0_size(MachineState *machine)
300 {
301     if (machine->numa_state->num_nodes) {
302         int i;
303         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
304             if (machine->numa_state->nodes[i].node_mem) {
305                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
306                            machine->ram_size);
307             }
308         }
309     }
310     return machine->ram_size;
311 }
312 
313 static void add_str(GString *s, const gchar *s1)
314 {
315     g_string_append_len(s, s1, strlen(s1) + 1);
316 }
317 
318 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
319                                        hwaddr size)
320 {
321     uint32_t associativity[] = {
322         cpu_to_be32(0x4), /* length */
323         cpu_to_be32(0x0), cpu_to_be32(0x0),
324         cpu_to_be32(0x0), cpu_to_be32(nodeid)
325     };
326     char mem_name[32];
327     uint64_t mem_reg_property[2];
328     int off;
329 
330     mem_reg_property[0] = cpu_to_be64(start);
331     mem_reg_property[1] = cpu_to_be64(size);
332 
333     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
334     off = fdt_add_subnode(fdt, 0, mem_name);
335     _FDT(off);
336     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
337     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
338                       sizeof(mem_reg_property))));
339     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
340                       sizeof(associativity))));
341     return off;
342 }
343 
344 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
345 {
346     MachineState *machine = MACHINE(spapr);
347     hwaddr mem_start, node_size;
348     int i, nb_nodes = machine->numa_state->num_nodes;
349     NodeInfo *nodes = machine->numa_state->nodes;
350 
351     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
352         if (!nodes[i].node_mem) {
353             continue;
354         }
355         if (mem_start >= machine->ram_size) {
356             node_size = 0;
357         } else {
358             node_size = nodes[i].node_mem;
359             if (node_size > machine->ram_size - mem_start) {
360                 node_size = machine->ram_size - mem_start;
361             }
362         }
363         if (!mem_start) {
364             /* spapr_machine_init() checks for rma_size <= node0_size
365              * already */
366             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
367             mem_start += spapr->rma_size;
368             node_size -= spapr->rma_size;
369         }
370         for ( ; node_size; ) {
371             hwaddr sizetmp = pow2floor(node_size);
372 
373             /* mem_start != 0 here */
374             if (ctzl(mem_start) < ctzl(sizetmp)) {
375                 sizetmp = 1ULL << ctzl(mem_start);
376             }
377 
378             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
379             node_size -= sizetmp;
380             mem_start += sizetmp;
381         }
382     }
383 
384     return 0;
385 }
386 
387 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
388                                   SpaprMachineState *spapr)
389 {
390     MachineState *ms = MACHINE(spapr);
391     PowerPCCPU *cpu = POWERPC_CPU(cs);
392     CPUPPCState *env = &cpu->env;
393     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
394     int index = spapr_get_vcpu_id(cpu);
395     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
396                        0xffffffff, 0xffffffff};
397     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
398         : SPAPR_TIMEBASE_FREQ;
399     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
400     uint32_t page_sizes_prop[64];
401     size_t page_sizes_prop_size;
402     unsigned int smp_threads = ms->smp.threads;
403     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
404     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
405     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
406     SpaprDrc *drc;
407     int drc_index;
408     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
409     int i;
410 
411     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
412     if (drc) {
413         drc_index = spapr_drc_index(drc);
414         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
415     }
416 
417     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
418     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
419 
420     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
421     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
422                            env->dcache_line_size)));
423     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
424                            env->dcache_line_size)));
425     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
426                            env->icache_line_size)));
427     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
428                            env->icache_line_size)));
429 
430     if (pcc->l1_dcache_size) {
431         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
432                                pcc->l1_dcache_size)));
433     } else {
434         warn_report("Unknown L1 dcache size for cpu");
435     }
436     if (pcc->l1_icache_size) {
437         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
438                                pcc->l1_icache_size)));
439     } else {
440         warn_report("Unknown L1 icache size for cpu");
441     }
442 
443     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
444     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
445     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
446     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
447     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
448     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
449 
450     if (env->spr_cb[SPR_PURR].oea_read) {
451         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
452     }
453     if (env->spr_cb[SPR_SPURR].oea_read) {
454         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
455     }
456 
457     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
458         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
459                           segs, sizeof(segs))));
460     }
461 
462     /* Advertise VSX (vector extensions) if available
463      *   1               == VMX / Altivec available
464      *   2               == VSX available
465      *
466      * Only CPUs for which we create core types in spapr_cpu_core.c
467      * are possible, and all of those have VMX */
468     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
469         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
470     } else {
471         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
472     }
473 
474     /* Advertise DFP (Decimal Floating Point) if available
475      *   0 / no property == no DFP
476      *   1               == DFP available */
477     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
478         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
479     }
480 
481     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
482                                                       sizeof(page_sizes_prop));
483     if (page_sizes_prop_size) {
484         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
485                           page_sizes_prop, page_sizes_prop_size)));
486     }
487 
488     spapr_populate_pa_features(spapr, cpu, fdt, offset);
489 
490     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
491                            cs->cpu_index / vcpus_per_socket)));
492 
493     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
494                       pft_size_prop, sizeof(pft_size_prop))));
495 
496     if (ms->numa_state->num_nodes > 1) {
497         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
498     }
499 
500     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
501 
502     if (pcc->radix_page_info) {
503         for (i = 0; i < pcc->radix_page_info->count; i++) {
504             radix_AP_encodings[i] =
505                 cpu_to_be32(pcc->radix_page_info->entries[i]);
506         }
507         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
508                           radix_AP_encodings,
509                           pcc->radix_page_info->count *
510                           sizeof(radix_AP_encodings[0]))));
511     }
512 
513     /*
514      * We set this property to let the guest know that it can use the large
515      * decrementer and its width in bits.
516      */
517     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
518         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
519                               pcc->lrg_decr_bits)));
520 }
521 
522 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
523 {
524     CPUState **rev;
525     CPUState *cs;
526     int n_cpus;
527     int cpus_offset;
528     char *nodename;
529     int i;
530 
531     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
532     _FDT(cpus_offset);
533     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
534     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
535 
536     /*
537      * We walk the CPUs in reverse order to ensure that CPU DT nodes
538      * created by fdt_add_subnode() end up in the right order in FDT
539      * for the guest kernel the enumerate the CPUs correctly.
540      *
541      * The CPU list cannot be traversed in reverse order, so we need
542      * to do extra work.
543      */
544     n_cpus = 0;
545     rev = NULL;
546     CPU_FOREACH(cs) {
547         rev = g_renew(CPUState *, rev, n_cpus + 1);
548         rev[n_cpus++] = cs;
549     }
550 
551     for (i = n_cpus - 1; i >= 0; i--) {
552         CPUState *cs = rev[i];
553         PowerPCCPU *cpu = POWERPC_CPU(cs);
554         int index = spapr_get_vcpu_id(cpu);
555         DeviceClass *dc = DEVICE_GET_CLASS(cs);
556         int offset;
557 
558         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
559             continue;
560         }
561 
562         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
563         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
564         g_free(nodename);
565         _FDT(offset);
566         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
567     }
568 
569     g_free(rev);
570 }
571 
572 static int spapr_rng_populate_dt(void *fdt)
573 {
574     int node;
575     int ret;
576 
577     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
578     if (node <= 0) {
579         return -1;
580     }
581     ret = fdt_setprop_string(fdt, node, "device_type",
582                              "ibm,platform-facilities");
583     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
584     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
585 
586     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
587     if (node <= 0) {
588         return -1;
589     }
590     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
591 
592     return ret ? -1 : 0;
593 }
594 
595 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
596 {
597     MemoryDeviceInfoList *info;
598 
599     for (info = list; info; info = info->next) {
600         MemoryDeviceInfo *value = info->value;
601 
602         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
603             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
604 
605             if (addr >= pcdimm_info->addr &&
606                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
607                 return pcdimm_info->node;
608             }
609         }
610     }
611 
612     return -1;
613 }
614 
615 struct sPAPRDrconfCellV2 {
616      uint32_t seq_lmbs;
617      uint64_t base_addr;
618      uint32_t drc_index;
619      uint32_t aa_index;
620      uint32_t flags;
621 } QEMU_PACKED;
622 
623 typedef struct DrconfCellQueue {
624     struct sPAPRDrconfCellV2 cell;
625     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
626 } DrconfCellQueue;
627 
628 static DrconfCellQueue *
629 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
630                       uint32_t drc_index, uint32_t aa_index,
631                       uint32_t flags)
632 {
633     DrconfCellQueue *elem;
634 
635     elem = g_malloc0(sizeof(*elem));
636     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
637     elem->cell.base_addr = cpu_to_be64(base_addr);
638     elem->cell.drc_index = cpu_to_be32(drc_index);
639     elem->cell.aa_index = cpu_to_be32(aa_index);
640     elem->cell.flags = cpu_to_be32(flags);
641 
642     return elem;
643 }
644 
645 /* ibm,dynamic-memory-v2 */
646 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
647                                    int offset, MemoryDeviceInfoList *dimms)
648 {
649     MachineState *machine = MACHINE(spapr);
650     uint8_t *int_buf, *cur_index;
651     int ret;
652     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
653     uint64_t addr, cur_addr, size;
654     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
655     uint64_t mem_end = machine->device_memory->base +
656                        memory_region_size(&machine->device_memory->mr);
657     uint32_t node, buf_len, nr_entries = 0;
658     SpaprDrc *drc;
659     DrconfCellQueue *elem, *next;
660     MemoryDeviceInfoList *info;
661     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
662         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
663 
664     /* Entry to cover RAM and the gap area */
665     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
666                                  SPAPR_LMB_FLAGS_RESERVED |
667                                  SPAPR_LMB_FLAGS_DRC_INVALID);
668     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
669     nr_entries++;
670 
671     cur_addr = machine->device_memory->base;
672     for (info = dimms; info; info = info->next) {
673         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
674 
675         addr = di->addr;
676         size = di->size;
677         node = di->node;
678 
679         /*
680          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
681          * area is marked hotpluggable in the next iteration for the bigger
682          * chunk including the NVDIMM occupied area.
683          */
684         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
685             continue;
686 
687         /* Entry for hot-pluggable area */
688         if (cur_addr < addr) {
689             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
690             g_assert(drc);
691             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
692                                          cur_addr, spapr_drc_index(drc), -1, 0);
693             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
694             nr_entries++;
695         }
696 
697         /* Entry for DIMM */
698         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
699         g_assert(drc);
700         elem = spapr_get_drconf_cell(size / lmb_size, addr,
701                                      spapr_drc_index(drc), node,
702                                      SPAPR_LMB_FLAGS_ASSIGNED);
703         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
704         nr_entries++;
705         cur_addr = addr + size;
706     }
707 
708     /* Entry for remaining hotpluggable area */
709     if (cur_addr < mem_end) {
710         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
711         g_assert(drc);
712         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
713                                      cur_addr, spapr_drc_index(drc), -1, 0);
714         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
715         nr_entries++;
716     }
717 
718     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
719     int_buf = cur_index = g_malloc0(buf_len);
720     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
721     cur_index += sizeof(nr_entries);
722 
723     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
724         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
725         cur_index += sizeof(elem->cell);
726         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
727         g_free(elem);
728     }
729 
730     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
731     g_free(int_buf);
732     if (ret < 0) {
733         return -1;
734     }
735     return 0;
736 }
737 
738 /* ibm,dynamic-memory */
739 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
740                                    int offset, MemoryDeviceInfoList *dimms)
741 {
742     MachineState *machine = MACHINE(spapr);
743     int i, ret;
744     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
745     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
746     uint32_t nr_lmbs = (machine->device_memory->base +
747                        memory_region_size(&machine->device_memory->mr)) /
748                        lmb_size;
749     uint32_t *int_buf, *cur_index, buf_len;
750 
751     /*
752      * Allocate enough buffer size to fit in ibm,dynamic-memory
753      */
754     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
755     cur_index = int_buf = g_malloc0(buf_len);
756     int_buf[0] = cpu_to_be32(nr_lmbs);
757     cur_index++;
758     for (i = 0; i < nr_lmbs; i++) {
759         uint64_t addr = i * lmb_size;
760         uint32_t *dynamic_memory = cur_index;
761 
762         if (i >= device_lmb_start) {
763             SpaprDrc *drc;
764 
765             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
766             g_assert(drc);
767 
768             dynamic_memory[0] = cpu_to_be32(addr >> 32);
769             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
770             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
771             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
772             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
773             if (memory_region_present(get_system_memory(), addr)) {
774                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
775             } else {
776                 dynamic_memory[5] = cpu_to_be32(0);
777             }
778         } else {
779             /*
780              * LMB information for RMA, boot time RAM and gap b/n RAM and
781              * device memory region -- all these are marked as reserved
782              * and as having no valid DRC.
783              */
784             dynamic_memory[0] = cpu_to_be32(addr >> 32);
785             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
786             dynamic_memory[2] = cpu_to_be32(0);
787             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
788             dynamic_memory[4] = cpu_to_be32(-1);
789             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
790                                             SPAPR_LMB_FLAGS_DRC_INVALID);
791         }
792 
793         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
794     }
795     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
796     g_free(int_buf);
797     if (ret < 0) {
798         return -1;
799     }
800     return 0;
801 }
802 
803 /*
804  * Adds ibm,dynamic-reconfiguration-memory node.
805  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
806  * of this device tree node.
807  */
808 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
809 {
810     MachineState *machine = MACHINE(spapr);
811     int nb_numa_nodes = machine->numa_state->num_nodes;
812     int ret, i, offset;
813     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
814     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
815     uint32_t *int_buf, *cur_index, buf_len;
816     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
817     MemoryDeviceInfoList *dimms = NULL;
818 
819     /*
820      * Don't create the node if there is no device memory
821      */
822     if (machine->ram_size == machine->maxram_size) {
823         return 0;
824     }
825 
826     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
827 
828     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
829                     sizeof(prop_lmb_size));
830     if (ret < 0) {
831         return ret;
832     }
833 
834     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
835     if (ret < 0) {
836         return ret;
837     }
838 
839     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
840     if (ret < 0) {
841         return ret;
842     }
843 
844     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
845     dimms = qmp_memory_device_list();
846     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
847         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
848     } else {
849         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
850     }
851     qapi_free_MemoryDeviceInfoList(dimms);
852 
853     if (ret < 0) {
854         return ret;
855     }
856 
857     /* ibm,associativity-lookup-arrays */
858     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
859     cur_index = int_buf = g_malloc0(buf_len);
860     int_buf[0] = cpu_to_be32(nr_nodes);
861     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
862     cur_index += 2;
863     for (i = 0; i < nr_nodes; i++) {
864         uint32_t associativity[] = {
865             cpu_to_be32(0x0),
866             cpu_to_be32(0x0),
867             cpu_to_be32(0x0),
868             cpu_to_be32(i)
869         };
870         memcpy(cur_index, associativity, sizeof(associativity));
871         cur_index += 4;
872     }
873     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
874             (cur_index - int_buf) * sizeof(uint32_t));
875     g_free(int_buf);
876 
877     return ret;
878 }
879 
880 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
881                                 SpaprOptionVector *ov5_updates)
882 {
883     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
884     int ret = 0, offset;
885 
886     /* Generate ibm,dynamic-reconfiguration-memory node if required */
887     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
888         g_assert(smc->dr_lmb_enabled);
889         ret = spapr_populate_drconf_memory(spapr, fdt);
890         if (ret) {
891             return ret;
892         }
893     }
894 
895     offset = fdt_path_offset(fdt, "/chosen");
896     if (offset < 0) {
897         offset = fdt_add_subnode(fdt, 0, "chosen");
898         if (offset < 0) {
899             return offset;
900         }
901     }
902     return spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
903                                   "ibm,architecture-vec-5");
904 }
905 
906 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
907 {
908     MachineState *ms = MACHINE(spapr);
909     int rtas;
910     GString *hypertas = g_string_sized_new(256);
911     GString *qemu_hypertas = g_string_sized_new(256);
912     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
913     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
914         memory_region_size(&MACHINE(spapr)->device_memory->mr);
915     uint32_t lrdr_capacity[] = {
916         cpu_to_be32(max_device_addr >> 32),
917         cpu_to_be32(max_device_addr & 0xffffffff),
918         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
919         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
920     };
921     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
922     uint32_t maxdomains[] = {
923         cpu_to_be32(4),
924         maxdomain,
925         maxdomain,
926         maxdomain,
927         cpu_to_be32(spapr->gpu_numa_id),
928     };
929 
930     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
931 
932     /* hypertas */
933     add_str(hypertas, "hcall-pft");
934     add_str(hypertas, "hcall-term");
935     add_str(hypertas, "hcall-dabr");
936     add_str(hypertas, "hcall-interrupt");
937     add_str(hypertas, "hcall-tce");
938     add_str(hypertas, "hcall-vio");
939     add_str(hypertas, "hcall-splpar");
940     add_str(hypertas, "hcall-join");
941     add_str(hypertas, "hcall-bulk");
942     add_str(hypertas, "hcall-set-mode");
943     add_str(hypertas, "hcall-sprg0");
944     add_str(hypertas, "hcall-copy");
945     add_str(hypertas, "hcall-debug");
946     add_str(hypertas, "hcall-vphn");
947     add_str(qemu_hypertas, "hcall-memop1");
948 
949     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
950         add_str(hypertas, "hcall-multi-tce");
951     }
952 
953     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
954         add_str(hypertas, "hcall-hpt-resize");
955     }
956 
957     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
958                      hypertas->str, hypertas->len));
959     g_string_free(hypertas, TRUE);
960     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
961                      qemu_hypertas->str, qemu_hypertas->len));
962     g_string_free(qemu_hypertas, TRUE);
963 
964     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
965                      refpoints, sizeof(refpoints)));
966 
967     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
968                      maxdomains, sizeof(maxdomains)));
969 
970     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-size", RTAS_SIZE));
971     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
972                           RTAS_ERROR_LOG_MAX));
973     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
974                           RTAS_EVENT_SCAN_RATE));
975 
976     g_assert(msi_nonbroken);
977     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
978 
979     /*
980      * According to PAPR, rtas ibm,os-term does not guarantee a return
981      * back to the guest cpu.
982      *
983      * While an additional ibm,extended-os-term property indicates
984      * that rtas call return will always occur. Set this property.
985      */
986     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
987 
988     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
989                      lrdr_capacity, sizeof(lrdr_capacity)));
990 
991     spapr_dt_rtas_tokens(fdt, rtas);
992 }
993 
994 /*
995  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
996  * and the XIVE features that the guest may request and thus the valid
997  * values for bytes 23..26 of option vector 5:
998  */
999 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1000                                           int chosen)
1001 {
1002     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1003 
1004     char val[2 * 4] = {
1005         23, 0x00, /* XICS / XIVE mode */
1006         24, 0x00, /* Hash/Radix, filled in below. */
1007         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1008         26, 0x40, /* Radix options: GTSE == yes. */
1009     };
1010 
1011     if (spapr->irq->xics && spapr->irq->xive) {
1012         val[1] = SPAPR_OV5_XIVE_BOTH;
1013     } else if (spapr->irq->xive) {
1014         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1015     } else {
1016         assert(spapr->irq->xics);
1017         val[1] = SPAPR_OV5_XIVE_LEGACY;
1018     }
1019 
1020     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1021                           first_ppc_cpu->compat_pvr)) {
1022         /*
1023          * If we're in a pre POWER9 compat mode then the guest should
1024          * do hash and use the legacy interrupt mode
1025          */
1026         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1027         val[3] = 0x00; /* Hash */
1028     } else if (kvm_enabled()) {
1029         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1030             val[3] = 0x80; /* OV5_MMU_BOTH */
1031         } else if (kvmppc_has_cap_mmu_radix()) {
1032             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1033         } else {
1034             val[3] = 0x00; /* Hash */
1035         }
1036     } else {
1037         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1038         val[3] = 0xC0;
1039     }
1040     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1041                      val, sizeof(val)));
1042 }
1043 
1044 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1045 {
1046     MachineState *machine = MACHINE(spapr);
1047     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1048     int chosen;
1049     const char *boot_device = machine->boot_order;
1050     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1051     size_t cb = 0;
1052     char *bootlist = get_boot_devices_list(&cb);
1053 
1054     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1055 
1056     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1057         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1058                                 machine->kernel_cmdline));
1059     }
1060     if (spapr->initrd_size) {
1061         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1062                               spapr->initrd_base));
1063         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1064                               spapr->initrd_base + spapr->initrd_size));
1065     }
1066 
1067     if (spapr->kernel_size) {
1068         uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1069                               cpu_to_be64(spapr->kernel_size) };
1070 
1071         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1072                          &kprop, sizeof(kprop)));
1073         if (spapr->kernel_le) {
1074             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1075         }
1076     }
1077     if (boot_menu) {
1078         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1079     }
1080     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1081     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1082     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1083 
1084     if (cb && bootlist) {
1085         int i;
1086 
1087         for (i = 0; i < cb; i++) {
1088             if (bootlist[i] == '\n') {
1089                 bootlist[i] = ' ';
1090             }
1091         }
1092         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1093     }
1094 
1095     if (boot_device && strlen(boot_device)) {
1096         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1097     }
1098 
1099     if (!spapr->has_graphics && stdout_path) {
1100         /*
1101          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1102          * kernel. New platforms should only use the "stdout-path" property. Set
1103          * the new property and continue using older property to remain
1104          * compatible with the existing firmware.
1105          */
1106         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1107         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1108     }
1109 
1110     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1111     if (smc->linux_pci_probe) {
1112         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1113     }
1114 
1115     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1116 
1117     g_free(stdout_path);
1118     g_free(bootlist);
1119 }
1120 
1121 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1122 {
1123     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1124      * KVM to work under pHyp with some guest co-operation */
1125     int hypervisor;
1126     uint8_t hypercall[16];
1127 
1128     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1129     /* indicate KVM hypercall interface */
1130     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1131     if (kvmppc_has_cap_fixup_hcalls()) {
1132         /*
1133          * Older KVM versions with older guest kernels were broken
1134          * with the magic page, don't allow the guest to map it.
1135          */
1136         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1137                                   sizeof(hypercall))) {
1138             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1139                              hypercall, sizeof(hypercall)));
1140         }
1141     }
1142 }
1143 
1144 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1145 {
1146     MachineState *machine = MACHINE(spapr);
1147     MachineClass *mc = MACHINE_GET_CLASS(machine);
1148     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1149     int ret;
1150     void *fdt;
1151     SpaprPhbState *phb;
1152     char *buf;
1153 
1154     fdt = g_malloc0(space);
1155     _FDT((fdt_create_empty_tree(fdt, space)));
1156 
1157     /* Root node */
1158     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1159     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1160     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1161 
1162     /* Guest UUID & Name*/
1163     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1164     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1165     if (qemu_uuid_set) {
1166         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1167     }
1168     g_free(buf);
1169 
1170     if (qemu_get_vm_name()) {
1171         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1172                                 qemu_get_vm_name()));
1173     }
1174 
1175     /* Host Model & Serial Number */
1176     if (spapr->host_model) {
1177         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1178     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1179         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1180         g_free(buf);
1181     }
1182 
1183     if (spapr->host_serial) {
1184         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1185     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1186         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1187         g_free(buf);
1188     }
1189 
1190     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1191     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1192 
1193     /* /interrupt controller */
1194     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1195 
1196     ret = spapr_populate_memory(spapr, fdt);
1197     if (ret < 0) {
1198         error_report("couldn't setup memory nodes in fdt");
1199         exit(1);
1200     }
1201 
1202     /* /vdevice */
1203     spapr_dt_vdevice(spapr->vio_bus, fdt);
1204 
1205     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1206         ret = spapr_rng_populate_dt(fdt);
1207         if (ret < 0) {
1208             error_report("could not set up rng device in the fdt");
1209             exit(1);
1210         }
1211     }
1212 
1213     QLIST_FOREACH(phb, &spapr->phbs, list) {
1214         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1215         if (ret < 0) {
1216             error_report("couldn't setup PCI devices in fdt");
1217             exit(1);
1218         }
1219     }
1220 
1221     /* cpus */
1222     spapr_populate_cpus_dt_node(fdt, spapr);
1223 
1224     if (smc->dr_lmb_enabled) {
1225         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1226     }
1227 
1228     if (mc->has_hotpluggable_cpus) {
1229         int offset = fdt_path_offset(fdt, "/cpus");
1230         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1231         if (ret < 0) {
1232             error_report("Couldn't set up CPU DR device tree properties");
1233             exit(1);
1234         }
1235     }
1236 
1237     /* /event-sources */
1238     spapr_dt_events(spapr, fdt);
1239 
1240     /* /rtas */
1241     spapr_dt_rtas(spapr, fdt);
1242 
1243     /* /chosen */
1244     if (reset) {
1245         spapr_dt_chosen(spapr, fdt);
1246     }
1247 
1248     /* /hypervisor */
1249     if (kvm_enabled()) {
1250         spapr_dt_hypervisor(spapr, fdt);
1251     }
1252 
1253     /* Build memory reserve map */
1254     if (reset) {
1255         if (spapr->kernel_size) {
1256             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1257                                   spapr->kernel_size)));
1258         }
1259         if (spapr->initrd_size) {
1260             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1261                                   spapr->initrd_size)));
1262         }
1263     }
1264 
1265     /* ibm,client-architecture-support updates */
1266     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1267     if (ret < 0) {
1268         error_report("couldn't setup CAS properties fdt");
1269         exit(1);
1270     }
1271 
1272     if (smc->dr_phb_enabled) {
1273         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1274         if (ret < 0) {
1275             error_report("Couldn't set up PHB DR device tree properties");
1276             exit(1);
1277         }
1278     }
1279 
1280     /* NVDIMM devices */
1281     if (mc->nvdimm_supported) {
1282         spapr_dt_persistent_memory(fdt);
1283     }
1284 
1285     return fdt;
1286 }
1287 
1288 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1289 {
1290     SpaprMachineState *spapr = opaque;
1291 
1292     return (addr & 0x0fffffff) + spapr->kernel_addr;
1293 }
1294 
1295 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1296                                     PowerPCCPU *cpu)
1297 {
1298     CPUPPCState *env = &cpu->env;
1299 
1300     /* The TCG path should also be holding the BQL at this point */
1301     g_assert(qemu_mutex_iothread_locked());
1302 
1303     if (msr_pr) {
1304         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1305         env->gpr[3] = H_PRIVILEGE;
1306     } else {
1307         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1308     }
1309 }
1310 
1311 struct LPCRSyncState {
1312     target_ulong value;
1313     target_ulong mask;
1314 };
1315 
1316 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1317 {
1318     struct LPCRSyncState *s = arg.host_ptr;
1319     PowerPCCPU *cpu = POWERPC_CPU(cs);
1320     CPUPPCState *env = &cpu->env;
1321     target_ulong lpcr;
1322 
1323     cpu_synchronize_state(cs);
1324     lpcr = env->spr[SPR_LPCR];
1325     lpcr &= ~s->mask;
1326     lpcr |= s->value;
1327     ppc_store_lpcr(cpu, lpcr);
1328 }
1329 
1330 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1331 {
1332     CPUState *cs;
1333     struct LPCRSyncState s = {
1334         .value = value,
1335         .mask = mask
1336     };
1337     CPU_FOREACH(cs) {
1338         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1339     }
1340 }
1341 
1342 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1343 {
1344     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1345 
1346     /* Copy PATE1:GR into PATE0:HR */
1347     entry->dw0 = spapr->patb_entry & PATE0_HR;
1348     entry->dw1 = spapr->patb_entry;
1349 }
1350 
1351 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1352 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1353 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1354 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1355 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1356 
1357 /*
1358  * Get the fd to access the kernel htab, re-opening it if necessary
1359  */
1360 static int get_htab_fd(SpaprMachineState *spapr)
1361 {
1362     Error *local_err = NULL;
1363 
1364     if (spapr->htab_fd >= 0) {
1365         return spapr->htab_fd;
1366     }
1367 
1368     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1369     if (spapr->htab_fd < 0) {
1370         error_report_err(local_err);
1371     }
1372 
1373     return spapr->htab_fd;
1374 }
1375 
1376 void close_htab_fd(SpaprMachineState *spapr)
1377 {
1378     if (spapr->htab_fd >= 0) {
1379         close(spapr->htab_fd);
1380     }
1381     spapr->htab_fd = -1;
1382 }
1383 
1384 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1385 {
1386     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1387 
1388     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1389 }
1390 
1391 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1392 {
1393     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1394 
1395     assert(kvm_enabled());
1396 
1397     if (!spapr->htab) {
1398         return 0;
1399     }
1400 
1401     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1402 }
1403 
1404 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1405                                                 hwaddr ptex, int n)
1406 {
1407     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1408     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1409 
1410     if (!spapr->htab) {
1411         /*
1412          * HTAB is controlled by KVM. Fetch into temporary buffer
1413          */
1414         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1415         kvmppc_read_hptes(hptes, ptex, n);
1416         return hptes;
1417     }
1418 
1419     /*
1420      * HTAB is controlled by QEMU. Just point to the internally
1421      * accessible PTEG.
1422      */
1423     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1424 }
1425 
1426 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1427                               const ppc_hash_pte64_t *hptes,
1428                               hwaddr ptex, int n)
1429 {
1430     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1431 
1432     if (!spapr->htab) {
1433         g_free((void *)hptes);
1434     }
1435 
1436     /* Nothing to do for qemu managed HPT */
1437 }
1438 
1439 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1440                       uint64_t pte0, uint64_t pte1)
1441 {
1442     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1443     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1444 
1445     if (!spapr->htab) {
1446         kvmppc_write_hpte(ptex, pte0, pte1);
1447     } else {
1448         if (pte0 & HPTE64_V_VALID) {
1449             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1450             /*
1451              * When setting valid, we write PTE1 first. This ensures
1452              * proper synchronization with the reading code in
1453              * ppc_hash64_pteg_search()
1454              */
1455             smp_wmb();
1456             stq_p(spapr->htab + offset, pte0);
1457         } else {
1458             stq_p(spapr->htab + offset, pte0);
1459             /*
1460              * When clearing it we set PTE0 first. This ensures proper
1461              * synchronization with the reading code in
1462              * ppc_hash64_pteg_search()
1463              */
1464             smp_wmb();
1465             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1466         }
1467     }
1468 }
1469 
1470 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1471                              uint64_t pte1)
1472 {
1473     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1474     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1475 
1476     if (!spapr->htab) {
1477         /* There should always be a hash table when this is called */
1478         error_report("spapr_hpte_set_c called with no hash table !");
1479         return;
1480     }
1481 
1482     /* The HW performs a non-atomic byte update */
1483     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1484 }
1485 
1486 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1487                              uint64_t pte1)
1488 {
1489     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1490     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1491 
1492     if (!spapr->htab) {
1493         /* There should always be a hash table when this is called */
1494         error_report("spapr_hpte_set_r called with no hash table !");
1495         return;
1496     }
1497 
1498     /* The HW performs a non-atomic byte update */
1499     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1500 }
1501 
1502 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1503 {
1504     int shift;
1505 
1506     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1507      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1508      * that's much more than is needed for Linux guests */
1509     shift = ctz64(pow2ceil(ramsize)) - 7;
1510     shift = MAX(shift, 18); /* Minimum architected size */
1511     shift = MIN(shift, 46); /* Maximum architected size */
1512     return shift;
1513 }
1514 
1515 void spapr_free_hpt(SpaprMachineState *spapr)
1516 {
1517     g_free(spapr->htab);
1518     spapr->htab = NULL;
1519     spapr->htab_shift = 0;
1520     close_htab_fd(spapr);
1521 }
1522 
1523 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1524                           Error **errp)
1525 {
1526     long rc;
1527 
1528     /* Clean up any HPT info from a previous boot */
1529     spapr_free_hpt(spapr);
1530 
1531     rc = kvmppc_reset_htab(shift);
1532     if (rc < 0) {
1533         /* kernel-side HPT needed, but couldn't allocate one */
1534         error_setg_errno(errp, errno,
1535                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1536                          shift);
1537         /* This is almost certainly fatal, but if the caller really
1538          * wants to carry on with shift == 0, it's welcome to try */
1539     } else if (rc > 0) {
1540         /* kernel-side HPT allocated */
1541         if (rc != shift) {
1542             error_setg(errp,
1543                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1544                        shift, rc);
1545         }
1546 
1547         spapr->htab_shift = shift;
1548         spapr->htab = NULL;
1549     } else {
1550         /* kernel-side HPT not needed, allocate in userspace instead */
1551         size_t size = 1ULL << shift;
1552         int i;
1553 
1554         spapr->htab = qemu_memalign(size, size);
1555         if (!spapr->htab) {
1556             error_setg_errno(errp, errno,
1557                              "Could not allocate HPT of order %d", shift);
1558             return;
1559         }
1560 
1561         memset(spapr->htab, 0, size);
1562         spapr->htab_shift = shift;
1563 
1564         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1565             DIRTY_HPTE(HPTE(spapr->htab, i));
1566         }
1567     }
1568     /* We're setting up a hash table, so that means we're not radix */
1569     spapr->patb_entry = 0;
1570     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1571 }
1572 
1573 void spapr_setup_hpt(SpaprMachineState *spapr)
1574 {
1575     int hpt_shift;
1576 
1577     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1578         || (spapr->cas_reboot
1579             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1580         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1581     } else {
1582         uint64_t current_ram_size;
1583 
1584         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1585         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1586     }
1587     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1588 
1589     if (kvm_enabled()) {
1590         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1591 
1592         /* Check our RMA fits in the possible VRMA */
1593         if (vrma_limit < spapr->rma_size) {
1594             error_report("Unable to create %" HWADDR_PRIu
1595                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1596                          spapr->rma_size / MiB, vrma_limit / MiB);
1597             exit(EXIT_FAILURE);
1598         }
1599     }
1600 }
1601 
1602 static int spapr_reset_drcs(Object *child, void *opaque)
1603 {
1604     SpaprDrc *drc =
1605         (SpaprDrc *) object_dynamic_cast(child,
1606                                                  TYPE_SPAPR_DR_CONNECTOR);
1607 
1608     if (drc) {
1609         spapr_drc_reset(drc);
1610     }
1611 
1612     return 0;
1613 }
1614 
1615 static void spapr_machine_reset(MachineState *machine)
1616 {
1617     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1618     PowerPCCPU *first_ppc_cpu;
1619     hwaddr fdt_addr;
1620     void *fdt;
1621     int rc;
1622 
1623     kvmppc_svm_off(&error_fatal);
1624     spapr_caps_apply(spapr);
1625 
1626     first_ppc_cpu = POWERPC_CPU(first_cpu);
1627     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1628         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1629                               spapr->max_compat_pvr)) {
1630         /*
1631          * If using KVM with radix mode available, VCPUs can be started
1632          * without a HPT because KVM will start them in radix mode.
1633          * Set the GR bit in PATE so that we know there is no HPT.
1634          */
1635         spapr->patb_entry = PATE1_GR;
1636         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1637     } else {
1638         spapr_setup_hpt(spapr);
1639     }
1640 
1641     qemu_devices_reset();
1642 
1643     /*
1644      * If this reset wasn't generated by CAS, we should reset our
1645      * negotiated options and start from scratch
1646      */
1647     if (!spapr->cas_reboot) {
1648         spapr_ovec_cleanup(spapr->ov5_cas);
1649         spapr->ov5_cas = spapr_ovec_new();
1650 
1651         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1652     }
1653 
1654     /*
1655      * This is fixing some of the default configuration of the XIVE
1656      * devices. To be called after the reset of the machine devices.
1657      */
1658     spapr_irq_reset(spapr, &error_fatal);
1659 
1660     /*
1661      * There is no CAS under qtest. Simulate one to please the code that
1662      * depends on spapr->ov5_cas. This is especially needed to test device
1663      * unplug, so we do that before resetting the DRCs.
1664      */
1665     if (qtest_enabled()) {
1666         spapr_ovec_cleanup(spapr->ov5_cas);
1667         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1668     }
1669 
1670     /* DRC reset may cause a device to be unplugged. This will cause troubles
1671      * if this device is used by another device (eg, a running vhost backend
1672      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1673      * situations, we reset DRCs after all devices have been reset.
1674      */
1675     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1676 
1677     spapr_clear_pending_events(spapr);
1678 
1679     /*
1680      * We place the device tree and RTAS just below either the top of the RMA,
1681      * or just below 2GB, whichever is lower, so that it can be
1682      * processed with 32-bit real mode code if necessary
1683      */
1684     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1685 
1686     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1687 
1688     rc = fdt_pack(fdt);
1689 
1690     /* Should only fail if we've built a corrupted tree */
1691     assert(rc == 0);
1692 
1693     /* Load the fdt */
1694     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1695     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1696     g_free(spapr->fdt_blob);
1697     spapr->fdt_size = fdt_totalsize(fdt);
1698     spapr->fdt_initial_size = spapr->fdt_size;
1699     spapr->fdt_blob = fdt;
1700 
1701     /* Set up the entry state */
1702     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, 0, fdt_addr, 0);
1703     first_ppc_cpu->env.gpr[5] = 0;
1704 
1705     spapr->cas_reboot = false;
1706 
1707     spapr->mc_status = -1;
1708     spapr->guest_machine_check_addr = -1;
1709 
1710     /* Signal all vCPUs waiting on this condition */
1711     qemu_cond_broadcast(&spapr->mc_delivery_cond);
1712 
1713     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1714 }
1715 
1716 static void spapr_create_nvram(SpaprMachineState *spapr)
1717 {
1718     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1719     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1720 
1721     if (dinfo) {
1722         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1723                             &error_fatal);
1724     }
1725 
1726     qdev_init_nofail(dev);
1727 
1728     spapr->nvram = (struct SpaprNvram *)dev;
1729 }
1730 
1731 static void spapr_rtc_create(SpaprMachineState *spapr)
1732 {
1733     object_initialize_child(OBJECT(spapr), "rtc",
1734                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1735                             &error_fatal, NULL);
1736     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1737                               &error_fatal);
1738     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1739                               "date", &error_fatal);
1740 }
1741 
1742 /* Returns whether we want to use VGA or not */
1743 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1744 {
1745     switch (vga_interface_type) {
1746     case VGA_NONE:
1747         return false;
1748     case VGA_DEVICE:
1749         return true;
1750     case VGA_STD:
1751     case VGA_VIRTIO:
1752     case VGA_CIRRUS:
1753         return pci_vga_init(pci_bus) != NULL;
1754     default:
1755         error_setg(errp,
1756                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1757         return false;
1758     }
1759 }
1760 
1761 static int spapr_pre_load(void *opaque)
1762 {
1763     int rc;
1764 
1765     rc = spapr_caps_pre_load(opaque);
1766     if (rc) {
1767         return rc;
1768     }
1769 
1770     return 0;
1771 }
1772 
1773 static int spapr_post_load(void *opaque, int version_id)
1774 {
1775     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1776     int err = 0;
1777 
1778     err = spapr_caps_post_migration(spapr);
1779     if (err) {
1780         return err;
1781     }
1782 
1783     /*
1784      * In earlier versions, there was no separate qdev for the PAPR
1785      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1786      * So when migrating from those versions, poke the incoming offset
1787      * value into the RTC device
1788      */
1789     if (version_id < 3) {
1790         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1791         if (err) {
1792             return err;
1793         }
1794     }
1795 
1796     if (kvm_enabled() && spapr->patb_entry) {
1797         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1798         bool radix = !!(spapr->patb_entry & PATE1_GR);
1799         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1800 
1801         /*
1802          * Update LPCR:HR and UPRT as they may not be set properly in
1803          * the stream
1804          */
1805         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1806                             LPCR_HR | LPCR_UPRT);
1807 
1808         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1809         if (err) {
1810             error_report("Process table config unsupported by the host");
1811             return -EINVAL;
1812         }
1813     }
1814 
1815     err = spapr_irq_post_load(spapr, version_id);
1816     if (err) {
1817         return err;
1818     }
1819 
1820     return err;
1821 }
1822 
1823 static int spapr_pre_save(void *opaque)
1824 {
1825     int rc;
1826 
1827     rc = spapr_caps_pre_save(opaque);
1828     if (rc) {
1829         return rc;
1830     }
1831 
1832     return 0;
1833 }
1834 
1835 static bool version_before_3(void *opaque, int version_id)
1836 {
1837     return version_id < 3;
1838 }
1839 
1840 static bool spapr_pending_events_needed(void *opaque)
1841 {
1842     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1843     return !QTAILQ_EMPTY(&spapr->pending_events);
1844 }
1845 
1846 static const VMStateDescription vmstate_spapr_event_entry = {
1847     .name = "spapr_event_log_entry",
1848     .version_id = 1,
1849     .minimum_version_id = 1,
1850     .fields = (VMStateField[]) {
1851         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1852         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1853         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1854                                      NULL, extended_length),
1855         VMSTATE_END_OF_LIST()
1856     },
1857 };
1858 
1859 static const VMStateDescription vmstate_spapr_pending_events = {
1860     .name = "spapr_pending_events",
1861     .version_id = 1,
1862     .minimum_version_id = 1,
1863     .needed = spapr_pending_events_needed,
1864     .fields = (VMStateField[]) {
1865         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1866                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1867         VMSTATE_END_OF_LIST()
1868     },
1869 };
1870 
1871 static bool spapr_ov5_cas_needed(void *opaque)
1872 {
1873     SpaprMachineState *spapr = opaque;
1874     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1875     bool cas_needed;
1876 
1877     /* Prior to the introduction of SpaprOptionVector, we had two option
1878      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1879      * Both of these options encode machine topology into the device-tree
1880      * in such a way that the now-booted OS should still be able to interact
1881      * appropriately with QEMU regardless of what options were actually
1882      * negotiatied on the source side.
1883      *
1884      * As such, we can avoid migrating the CAS-negotiated options if these
1885      * are the only options available on the current machine/platform.
1886      * Since these are the only options available for pseries-2.7 and
1887      * earlier, this allows us to maintain old->new/new->old migration
1888      * compatibility.
1889      *
1890      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1891      * via default pseries-2.8 machines and explicit command-line parameters.
1892      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1893      * of the actual CAS-negotiated values to continue working properly. For
1894      * example, availability of memory unplug depends on knowing whether
1895      * OV5_HP_EVT was negotiated via CAS.
1896      *
1897      * Thus, for any cases where the set of available CAS-negotiatable
1898      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1899      * include the CAS-negotiated options in the migration stream, unless
1900      * if they affect boot time behaviour only.
1901      */
1902     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1903     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1904     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1905 
1906     /* We need extra information if we have any bits outside the mask
1907      * defined above */
1908     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1909 
1910     spapr_ovec_cleanup(ov5_mask);
1911 
1912     return cas_needed;
1913 }
1914 
1915 static const VMStateDescription vmstate_spapr_ov5_cas = {
1916     .name = "spapr_option_vector_ov5_cas",
1917     .version_id = 1,
1918     .minimum_version_id = 1,
1919     .needed = spapr_ov5_cas_needed,
1920     .fields = (VMStateField[]) {
1921         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1922                                  vmstate_spapr_ovec, SpaprOptionVector),
1923         VMSTATE_END_OF_LIST()
1924     },
1925 };
1926 
1927 static bool spapr_patb_entry_needed(void *opaque)
1928 {
1929     SpaprMachineState *spapr = opaque;
1930 
1931     return !!spapr->patb_entry;
1932 }
1933 
1934 static const VMStateDescription vmstate_spapr_patb_entry = {
1935     .name = "spapr_patb_entry",
1936     .version_id = 1,
1937     .minimum_version_id = 1,
1938     .needed = spapr_patb_entry_needed,
1939     .fields = (VMStateField[]) {
1940         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1941         VMSTATE_END_OF_LIST()
1942     },
1943 };
1944 
1945 static bool spapr_irq_map_needed(void *opaque)
1946 {
1947     SpaprMachineState *spapr = opaque;
1948 
1949     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1950 }
1951 
1952 static const VMStateDescription vmstate_spapr_irq_map = {
1953     .name = "spapr_irq_map",
1954     .version_id = 1,
1955     .minimum_version_id = 1,
1956     .needed = spapr_irq_map_needed,
1957     .fields = (VMStateField[]) {
1958         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1959         VMSTATE_END_OF_LIST()
1960     },
1961 };
1962 
1963 static bool spapr_dtb_needed(void *opaque)
1964 {
1965     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1966 
1967     return smc->update_dt_enabled;
1968 }
1969 
1970 static int spapr_dtb_pre_load(void *opaque)
1971 {
1972     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1973 
1974     g_free(spapr->fdt_blob);
1975     spapr->fdt_blob = NULL;
1976     spapr->fdt_size = 0;
1977 
1978     return 0;
1979 }
1980 
1981 static const VMStateDescription vmstate_spapr_dtb = {
1982     .name = "spapr_dtb",
1983     .version_id = 1,
1984     .minimum_version_id = 1,
1985     .needed = spapr_dtb_needed,
1986     .pre_load = spapr_dtb_pre_load,
1987     .fields = (VMStateField[]) {
1988         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1989         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1990         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1991                                      fdt_size),
1992         VMSTATE_END_OF_LIST()
1993     },
1994 };
1995 
1996 static bool spapr_fwnmi_needed(void *opaque)
1997 {
1998     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1999 
2000     return spapr->guest_machine_check_addr != -1;
2001 }
2002 
2003 static int spapr_fwnmi_pre_save(void *opaque)
2004 {
2005     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2006 
2007     /*
2008      * Check if machine check handling is in progress and print a
2009      * warning message.
2010      */
2011     if (spapr->mc_status != -1) {
2012         warn_report("A machine check is being handled during migration. The"
2013                 "handler may run and log hardware error on the destination");
2014     }
2015 
2016     return 0;
2017 }
2018 
2019 static const VMStateDescription vmstate_spapr_machine_check = {
2020     .name = "spapr_machine_check",
2021     .version_id = 1,
2022     .minimum_version_id = 1,
2023     .needed = spapr_fwnmi_needed,
2024     .pre_save = spapr_fwnmi_pre_save,
2025     .fields = (VMStateField[]) {
2026         VMSTATE_UINT64(guest_machine_check_addr, SpaprMachineState),
2027         VMSTATE_INT32(mc_status, SpaprMachineState),
2028         VMSTATE_END_OF_LIST()
2029     },
2030 };
2031 
2032 static const VMStateDescription vmstate_spapr = {
2033     .name = "spapr",
2034     .version_id = 3,
2035     .minimum_version_id = 1,
2036     .pre_load = spapr_pre_load,
2037     .post_load = spapr_post_load,
2038     .pre_save = spapr_pre_save,
2039     .fields = (VMStateField[]) {
2040         /* used to be @next_irq */
2041         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2042 
2043         /* RTC offset */
2044         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2045 
2046         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2047         VMSTATE_END_OF_LIST()
2048     },
2049     .subsections = (const VMStateDescription*[]) {
2050         &vmstate_spapr_ov5_cas,
2051         &vmstate_spapr_patb_entry,
2052         &vmstate_spapr_pending_events,
2053         &vmstate_spapr_cap_htm,
2054         &vmstate_spapr_cap_vsx,
2055         &vmstate_spapr_cap_dfp,
2056         &vmstate_spapr_cap_cfpc,
2057         &vmstate_spapr_cap_sbbc,
2058         &vmstate_spapr_cap_ibs,
2059         &vmstate_spapr_cap_hpt_maxpagesize,
2060         &vmstate_spapr_irq_map,
2061         &vmstate_spapr_cap_nested_kvm_hv,
2062         &vmstate_spapr_dtb,
2063         &vmstate_spapr_cap_large_decr,
2064         &vmstate_spapr_cap_ccf_assist,
2065         &vmstate_spapr_cap_fwnmi,
2066         &vmstate_spapr_machine_check,
2067         NULL
2068     }
2069 };
2070 
2071 static int htab_save_setup(QEMUFile *f, void *opaque)
2072 {
2073     SpaprMachineState *spapr = opaque;
2074 
2075     /* "Iteration" header */
2076     if (!spapr->htab_shift) {
2077         qemu_put_be32(f, -1);
2078     } else {
2079         qemu_put_be32(f, spapr->htab_shift);
2080     }
2081 
2082     if (spapr->htab) {
2083         spapr->htab_save_index = 0;
2084         spapr->htab_first_pass = true;
2085     } else {
2086         if (spapr->htab_shift) {
2087             assert(kvm_enabled());
2088         }
2089     }
2090 
2091 
2092     return 0;
2093 }
2094 
2095 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2096                             int chunkstart, int n_valid, int n_invalid)
2097 {
2098     qemu_put_be32(f, chunkstart);
2099     qemu_put_be16(f, n_valid);
2100     qemu_put_be16(f, n_invalid);
2101     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2102                     HASH_PTE_SIZE_64 * n_valid);
2103 }
2104 
2105 static void htab_save_end_marker(QEMUFile *f)
2106 {
2107     qemu_put_be32(f, 0);
2108     qemu_put_be16(f, 0);
2109     qemu_put_be16(f, 0);
2110 }
2111 
2112 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2113                                  int64_t max_ns)
2114 {
2115     bool has_timeout = max_ns != -1;
2116     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2117     int index = spapr->htab_save_index;
2118     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2119 
2120     assert(spapr->htab_first_pass);
2121 
2122     do {
2123         int chunkstart;
2124 
2125         /* Consume invalid HPTEs */
2126         while ((index < htabslots)
2127                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2128             CLEAN_HPTE(HPTE(spapr->htab, index));
2129             index++;
2130         }
2131 
2132         /* Consume valid HPTEs */
2133         chunkstart = index;
2134         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2135                && HPTE_VALID(HPTE(spapr->htab, index))) {
2136             CLEAN_HPTE(HPTE(spapr->htab, index));
2137             index++;
2138         }
2139 
2140         if (index > chunkstart) {
2141             int n_valid = index - chunkstart;
2142 
2143             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2144 
2145             if (has_timeout &&
2146                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2147                 break;
2148             }
2149         }
2150     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2151 
2152     if (index >= htabslots) {
2153         assert(index == htabslots);
2154         index = 0;
2155         spapr->htab_first_pass = false;
2156     }
2157     spapr->htab_save_index = index;
2158 }
2159 
2160 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2161                                 int64_t max_ns)
2162 {
2163     bool final = max_ns < 0;
2164     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2165     int examined = 0, sent = 0;
2166     int index = spapr->htab_save_index;
2167     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2168 
2169     assert(!spapr->htab_first_pass);
2170 
2171     do {
2172         int chunkstart, invalidstart;
2173 
2174         /* Consume non-dirty HPTEs */
2175         while ((index < htabslots)
2176                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2177             index++;
2178             examined++;
2179         }
2180 
2181         chunkstart = index;
2182         /* Consume valid dirty HPTEs */
2183         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2184                && HPTE_DIRTY(HPTE(spapr->htab, index))
2185                && HPTE_VALID(HPTE(spapr->htab, index))) {
2186             CLEAN_HPTE(HPTE(spapr->htab, index));
2187             index++;
2188             examined++;
2189         }
2190 
2191         invalidstart = index;
2192         /* Consume invalid dirty HPTEs */
2193         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2194                && HPTE_DIRTY(HPTE(spapr->htab, index))
2195                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2196             CLEAN_HPTE(HPTE(spapr->htab, index));
2197             index++;
2198             examined++;
2199         }
2200 
2201         if (index > chunkstart) {
2202             int n_valid = invalidstart - chunkstart;
2203             int n_invalid = index - invalidstart;
2204 
2205             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2206             sent += index - chunkstart;
2207 
2208             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2209                 break;
2210             }
2211         }
2212 
2213         if (examined >= htabslots) {
2214             break;
2215         }
2216 
2217         if (index >= htabslots) {
2218             assert(index == htabslots);
2219             index = 0;
2220         }
2221     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2222 
2223     if (index >= htabslots) {
2224         assert(index == htabslots);
2225         index = 0;
2226     }
2227 
2228     spapr->htab_save_index = index;
2229 
2230     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2231 }
2232 
2233 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2234 #define MAX_KVM_BUF_SIZE    2048
2235 
2236 static int htab_save_iterate(QEMUFile *f, void *opaque)
2237 {
2238     SpaprMachineState *spapr = opaque;
2239     int fd;
2240     int rc = 0;
2241 
2242     /* Iteration header */
2243     if (!spapr->htab_shift) {
2244         qemu_put_be32(f, -1);
2245         return 1;
2246     } else {
2247         qemu_put_be32(f, 0);
2248     }
2249 
2250     if (!spapr->htab) {
2251         assert(kvm_enabled());
2252 
2253         fd = get_htab_fd(spapr);
2254         if (fd < 0) {
2255             return fd;
2256         }
2257 
2258         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2259         if (rc < 0) {
2260             return rc;
2261         }
2262     } else  if (spapr->htab_first_pass) {
2263         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2264     } else {
2265         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2266     }
2267 
2268     htab_save_end_marker(f);
2269 
2270     return rc;
2271 }
2272 
2273 static int htab_save_complete(QEMUFile *f, void *opaque)
2274 {
2275     SpaprMachineState *spapr = opaque;
2276     int fd;
2277 
2278     /* Iteration header */
2279     if (!spapr->htab_shift) {
2280         qemu_put_be32(f, -1);
2281         return 0;
2282     } else {
2283         qemu_put_be32(f, 0);
2284     }
2285 
2286     if (!spapr->htab) {
2287         int rc;
2288 
2289         assert(kvm_enabled());
2290 
2291         fd = get_htab_fd(spapr);
2292         if (fd < 0) {
2293             return fd;
2294         }
2295 
2296         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2297         if (rc < 0) {
2298             return rc;
2299         }
2300     } else {
2301         if (spapr->htab_first_pass) {
2302             htab_save_first_pass(f, spapr, -1);
2303         }
2304         htab_save_later_pass(f, spapr, -1);
2305     }
2306 
2307     /* End marker */
2308     htab_save_end_marker(f);
2309 
2310     return 0;
2311 }
2312 
2313 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2314 {
2315     SpaprMachineState *spapr = opaque;
2316     uint32_t section_hdr;
2317     int fd = -1;
2318     Error *local_err = NULL;
2319 
2320     if (version_id < 1 || version_id > 1) {
2321         error_report("htab_load() bad version");
2322         return -EINVAL;
2323     }
2324 
2325     section_hdr = qemu_get_be32(f);
2326 
2327     if (section_hdr == -1) {
2328         spapr_free_hpt(spapr);
2329         return 0;
2330     }
2331 
2332     if (section_hdr) {
2333         /* First section gives the htab size */
2334         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2335         if (local_err) {
2336             error_report_err(local_err);
2337             return -EINVAL;
2338         }
2339         return 0;
2340     }
2341 
2342     if (!spapr->htab) {
2343         assert(kvm_enabled());
2344 
2345         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2346         if (fd < 0) {
2347             error_report_err(local_err);
2348             return fd;
2349         }
2350     }
2351 
2352     while (true) {
2353         uint32_t index;
2354         uint16_t n_valid, n_invalid;
2355 
2356         index = qemu_get_be32(f);
2357         n_valid = qemu_get_be16(f);
2358         n_invalid = qemu_get_be16(f);
2359 
2360         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2361             /* End of Stream */
2362             break;
2363         }
2364 
2365         if ((index + n_valid + n_invalid) >
2366             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2367             /* Bad index in stream */
2368             error_report(
2369                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2370                 index, n_valid, n_invalid, spapr->htab_shift);
2371             return -EINVAL;
2372         }
2373 
2374         if (spapr->htab) {
2375             if (n_valid) {
2376                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2377                                 HASH_PTE_SIZE_64 * n_valid);
2378             }
2379             if (n_invalid) {
2380                 memset(HPTE(spapr->htab, index + n_valid), 0,
2381                        HASH_PTE_SIZE_64 * n_invalid);
2382             }
2383         } else {
2384             int rc;
2385 
2386             assert(fd >= 0);
2387 
2388             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2389             if (rc < 0) {
2390                 return rc;
2391             }
2392         }
2393     }
2394 
2395     if (!spapr->htab) {
2396         assert(fd >= 0);
2397         close(fd);
2398     }
2399 
2400     return 0;
2401 }
2402 
2403 static void htab_save_cleanup(void *opaque)
2404 {
2405     SpaprMachineState *spapr = opaque;
2406 
2407     close_htab_fd(spapr);
2408 }
2409 
2410 static SaveVMHandlers savevm_htab_handlers = {
2411     .save_setup = htab_save_setup,
2412     .save_live_iterate = htab_save_iterate,
2413     .save_live_complete_precopy = htab_save_complete,
2414     .save_cleanup = htab_save_cleanup,
2415     .load_state = htab_load,
2416 };
2417 
2418 static void spapr_boot_set(void *opaque, const char *boot_device,
2419                            Error **errp)
2420 {
2421     MachineState *machine = MACHINE(opaque);
2422     machine->boot_order = g_strdup(boot_device);
2423 }
2424 
2425 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2426 {
2427     MachineState *machine = MACHINE(spapr);
2428     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2429     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2430     int i;
2431 
2432     for (i = 0; i < nr_lmbs; i++) {
2433         uint64_t addr;
2434 
2435         addr = i * lmb_size + machine->device_memory->base;
2436         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2437                                addr / lmb_size);
2438     }
2439 }
2440 
2441 /*
2442  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2443  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2444  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2445  */
2446 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2447 {
2448     int i;
2449 
2450     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2451         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2452                    " is not aligned to %" PRIu64 " MiB",
2453                    machine->ram_size,
2454                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2455         return;
2456     }
2457 
2458     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2459         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2460                    " is not aligned to %" PRIu64 " MiB",
2461                    machine->ram_size,
2462                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2463         return;
2464     }
2465 
2466     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2467         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2468             error_setg(errp,
2469                        "Node %d memory size 0x%" PRIx64
2470                        " is not aligned to %" PRIu64 " MiB",
2471                        i, machine->numa_state->nodes[i].node_mem,
2472                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2473             return;
2474         }
2475     }
2476 }
2477 
2478 /* find cpu slot in machine->possible_cpus by core_id */
2479 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2480 {
2481     int index = id / ms->smp.threads;
2482 
2483     if (index >= ms->possible_cpus->len) {
2484         return NULL;
2485     }
2486     if (idx) {
2487         *idx = index;
2488     }
2489     return &ms->possible_cpus->cpus[index];
2490 }
2491 
2492 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2493 {
2494     MachineState *ms = MACHINE(spapr);
2495     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2496     Error *local_err = NULL;
2497     bool vsmt_user = !!spapr->vsmt;
2498     int kvm_smt = kvmppc_smt_threads();
2499     int ret;
2500     unsigned int smp_threads = ms->smp.threads;
2501 
2502     if (!kvm_enabled() && (smp_threads > 1)) {
2503         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2504                      "on a pseries machine");
2505         goto out;
2506     }
2507     if (!is_power_of_2(smp_threads)) {
2508         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2509                      "machine because it must be a power of 2", smp_threads);
2510         goto out;
2511     }
2512 
2513     /* Detemine the VSMT mode to use: */
2514     if (vsmt_user) {
2515         if (spapr->vsmt < smp_threads) {
2516             error_setg(&local_err, "Cannot support VSMT mode %d"
2517                          " because it must be >= threads/core (%d)",
2518                          spapr->vsmt, smp_threads);
2519             goto out;
2520         }
2521         /* In this case, spapr->vsmt has been set by the command line */
2522     } else if (!smc->smp_threads_vsmt) {
2523         /*
2524          * Default VSMT value is tricky, because we need it to be as
2525          * consistent as possible (for migration), but this requires
2526          * changing it for at least some existing cases.  We pick 8 as
2527          * the value that we'd get with KVM on POWER8, the
2528          * overwhelmingly common case in production systems.
2529          */
2530         spapr->vsmt = MAX(8, smp_threads);
2531     } else {
2532         spapr->vsmt = smp_threads;
2533     }
2534 
2535     /* KVM: If necessary, set the SMT mode: */
2536     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2537         ret = kvmppc_set_smt_threads(spapr->vsmt);
2538         if (ret) {
2539             /* Looks like KVM isn't able to change VSMT mode */
2540             error_setg(&local_err,
2541                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2542                        spapr->vsmt, ret);
2543             /* We can live with that if the default one is big enough
2544              * for the number of threads, and a submultiple of the one
2545              * we want.  In this case we'll waste some vcpu ids, but
2546              * behaviour will be correct */
2547             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2548                 warn_report_err(local_err);
2549                 local_err = NULL;
2550                 goto out;
2551             } else {
2552                 if (!vsmt_user) {
2553                     error_append_hint(&local_err,
2554                                       "On PPC, a VM with %d threads/core"
2555                                       " on a host with %d threads/core"
2556                                       " requires the use of VSMT mode %d.\n",
2557                                       smp_threads, kvm_smt, spapr->vsmt);
2558                 }
2559                 kvmppc_error_append_smt_possible_hint(&local_err);
2560                 goto out;
2561             }
2562         }
2563     }
2564     /* else TCG: nothing to do currently */
2565 out:
2566     error_propagate(errp, local_err);
2567 }
2568 
2569 static void spapr_init_cpus(SpaprMachineState *spapr)
2570 {
2571     MachineState *machine = MACHINE(spapr);
2572     MachineClass *mc = MACHINE_GET_CLASS(machine);
2573     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2574     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2575     const CPUArchIdList *possible_cpus;
2576     unsigned int smp_cpus = machine->smp.cpus;
2577     unsigned int smp_threads = machine->smp.threads;
2578     unsigned int max_cpus = machine->smp.max_cpus;
2579     int boot_cores_nr = smp_cpus / smp_threads;
2580     int i;
2581 
2582     possible_cpus = mc->possible_cpu_arch_ids(machine);
2583     if (mc->has_hotpluggable_cpus) {
2584         if (smp_cpus % smp_threads) {
2585             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2586                          smp_cpus, smp_threads);
2587             exit(1);
2588         }
2589         if (max_cpus % smp_threads) {
2590             error_report("max_cpus (%u) must be multiple of threads (%u)",
2591                          max_cpus, smp_threads);
2592             exit(1);
2593         }
2594     } else {
2595         if (max_cpus != smp_cpus) {
2596             error_report("This machine version does not support CPU hotplug");
2597             exit(1);
2598         }
2599         boot_cores_nr = possible_cpus->len;
2600     }
2601 
2602     if (smc->pre_2_10_has_unused_icps) {
2603         int i;
2604 
2605         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2606             /* Dummy entries get deregistered when real ICPState objects
2607              * are registered during CPU core hotplug.
2608              */
2609             pre_2_10_vmstate_register_dummy_icp(i);
2610         }
2611     }
2612 
2613     for (i = 0; i < possible_cpus->len; i++) {
2614         int core_id = i * smp_threads;
2615 
2616         if (mc->has_hotpluggable_cpus) {
2617             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2618                                    spapr_vcpu_id(spapr, core_id));
2619         }
2620 
2621         if (i < boot_cores_nr) {
2622             Object *core  = object_new(type);
2623             int nr_threads = smp_threads;
2624 
2625             /* Handle the partially filled core for older machine types */
2626             if ((i + 1) * smp_threads >= smp_cpus) {
2627                 nr_threads = smp_cpus - i * smp_threads;
2628             }
2629 
2630             object_property_set_int(core, nr_threads, "nr-threads",
2631                                     &error_fatal);
2632             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2633                                     &error_fatal);
2634             object_property_set_bool(core, true, "realized", &error_fatal);
2635 
2636             object_unref(core);
2637         }
2638     }
2639 }
2640 
2641 static PCIHostState *spapr_create_default_phb(void)
2642 {
2643     DeviceState *dev;
2644 
2645     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2646     qdev_prop_set_uint32(dev, "index", 0);
2647     qdev_init_nofail(dev);
2648 
2649     return PCI_HOST_BRIDGE(dev);
2650 }
2651 
2652 static hwaddr spapr_rma_size(SpaprMachineState *spapr, Error **errp)
2653 {
2654     MachineState *machine = MACHINE(spapr);
2655     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2656     hwaddr rma_size = machine->ram_size;
2657     hwaddr node0_size = spapr_node0_size(machine);
2658 
2659     /* RMA has to fit in the first NUMA node */
2660     rma_size = MIN(rma_size, node0_size);
2661 
2662     /*
2663      * VRMA access is via a special 1TiB SLB mapping, so the RMA can
2664      * never exceed that
2665      */
2666     rma_size = MIN(rma_size, 1 * TiB);
2667 
2668     /*
2669      * Clamp the RMA size based on machine type.  This is for
2670      * migration compatibility with older qemu versions, which limited
2671      * the RMA size for complicated and mostly bad reasons.
2672      */
2673     if (smc->rma_limit) {
2674         rma_size = MIN(rma_size, smc->rma_limit);
2675     }
2676 
2677     if (rma_size < MIN_RMA_SLOF) {
2678         error_setg(errp,
2679                    "pSeries SLOF firmware requires >= %" HWADDR_PRIx
2680                    "ldMiB guest RMA (Real Mode Area memory)",
2681                    MIN_RMA_SLOF / MiB);
2682         return 0;
2683     }
2684 
2685     return rma_size;
2686 }
2687 
2688 /* pSeries LPAR / sPAPR hardware init */
2689 static void spapr_machine_init(MachineState *machine)
2690 {
2691     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2692     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2693     MachineClass *mc = MACHINE_GET_CLASS(machine);
2694     const char *kernel_filename = machine->kernel_filename;
2695     const char *initrd_filename = machine->initrd_filename;
2696     PCIHostState *phb;
2697     int i;
2698     MemoryRegion *sysmem = get_system_memory();
2699     long load_limit, fw_size;
2700     char *filename;
2701     Error *resize_hpt_err = NULL;
2702 
2703     msi_nonbroken = true;
2704 
2705     QLIST_INIT(&spapr->phbs);
2706     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2707 
2708     /* Determine capabilities to run with */
2709     spapr_caps_init(spapr);
2710 
2711     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2712     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2713         /*
2714          * If the user explicitly requested a mode we should either
2715          * supply it, or fail completely (which we do below).  But if
2716          * it's not set explicitly, we reset our mode to something
2717          * that works
2718          */
2719         if (resize_hpt_err) {
2720             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2721             error_free(resize_hpt_err);
2722             resize_hpt_err = NULL;
2723         } else {
2724             spapr->resize_hpt = smc->resize_hpt_default;
2725         }
2726     }
2727 
2728     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2729 
2730     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2731         /*
2732          * User requested HPT resize, but this host can't supply it.  Bail out
2733          */
2734         error_report_err(resize_hpt_err);
2735         exit(1);
2736     }
2737 
2738     spapr->rma_size = spapr_rma_size(spapr, &error_fatal);
2739 
2740     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2741     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2742 
2743     /*
2744      * VSMT must be set in order to be able to compute VCPU ids, ie to
2745      * call spapr_max_server_number() or spapr_vcpu_id().
2746      */
2747     spapr_set_vsmt_mode(spapr, &error_fatal);
2748 
2749     /* Set up Interrupt Controller before we create the VCPUs */
2750     spapr_irq_init(spapr, &error_fatal);
2751 
2752     /* Set up containers for ibm,client-architecture-support negotiated options
2753      */
2754     spapr->ov5 = spapr_ovec_new();
2755     spapr->ov5_cas = spapr_ovec_new();
2756 
2757     if (smc->dr_lmb_enabled) {
2758         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2759         spapr_validate_node_memory(machine, &error_fatal);
2760     }
2761 
2762     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2763 
2764     /* advertise support for dedicated HP event source to guests */
2765     if (spapr->use_hotplug_event_source) {
2766         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2767     }
2768 
2769     /* advertise support for HPT resizing */
2770     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2771         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2772     }
2773 
2774     /* advertise support for ibm,dyamic-memory-v2 */
2775     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2776 
2777     /* advertise XIVE on POWER9 machines */
2778     if (spapr->irq->xive) {
2779         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2780     }
2781 
2782     /* init CPUs */
2783     spapr_init_cpus(spapr);
2784 
2785     /*
2786      * check we don't have a memory-less/cpu-less NUMA node
2787      * Firmware relies on the existing memory/cpu topology to provide the
2788      * NUMA topology to the kernel.
2789      * And the linux kernel needs to know the NUMA topology at start
2790      * to be able to hotplug CPUs later.
2791      */
2792     if (machine->numa_state->num_nodes) {
2793         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2794             /* check for memory-less node */
2795             if (machine->numa_state->nodes[i].node_mem == 0) {
2796                 CPUState *cs;
2797                 int found = 0;
2798                 /* check for cpu-less node */
2799                 CPU_FOREACH(cs) {
2800                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2801                     if (cpu->node_id == i) {
2802                         found = 1;
2803                         break;
2804                     }
2805                 }
2806                 /* memory-less and cpu-less node */
2807                 if (!found) {
2808                     error_report(
2809                        "Memory-less/cpu-less nodes are not supported (node %d)",
2810                                  i);
2811                     exit(1);
2812                 }
2813             }
2814         }
2815 
2816     }
2817 
2818     /*
2819      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2820      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2821      * called from vPHB reset handler so we initialize the counter here.
2822      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2823      * must be equally distant from any other node.
2824      * The final value of spapr->gpu_numa_id is going to be written to
2825      * max-associativity-domains in spapr_build_fdt().
2826      */
2827     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2828 
2829     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2830         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2831                               spapr->max_compat_pvr)) {
2832         /* KVM and TCG always allow GTSE with radix... */
2833         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2834     }
2835     /* ... but not with hash (currently). */
2836 
2837     if (kvm_enabled()) {
2838         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2839         kvmppc_enable_logical_ci_hcalls();
2840         kvmppc_enable_set_mode_hcall();
2841 
2842         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2843         kvmppc_enable_clear_ref_mod_hcalls();
2844 
2845         /* Enable H_PAGE_INIT */
2846         kvmppc_enable_h_page_init();
2847     }
2848 
2849     /* map RAM */
2850     memory_region_add_subregion(sysmem, 0, machine->ram);
2851 
2852     /* always allocate the device memory information */
2853     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2854 
2855     /* initialize hotplug memory address space */
2856     if (machine->ram_size < machine->maxram_size) {
2857         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2858         /*
2859          * Limit the number of hotpluggable memory slots to half the number
2860          * slots that KVM supports, leaving the other half for PCI and other
2861          * devices. However ensure that number of slots doesn't drop below 32.
2862          */
2863         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2864                            SPAPR_MAX_RAM_SLOTS;
2865 
2866         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2867             max_memslots = SPAPR_MAX_RAM_SLOTS;
2868         }
2869         if (machine->ram_slots > max_memslots) {
2870             error_report("Specified number of memory slots %"
2871                          PRIu64" exceeds max supported %d",
2872                          machine->ram_slots, max_memslots);
2873             exit(1);
2874         }
2875 
2876         machine->device_memory->base = ROUND_UP(machine->ram_size,
2877                                                 SPAPR_DEVICE_MEM_ALIGN);
2878         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2879                            "device-memory", device_mem_size);
2880         memory_region_add_subregion(sysmem, machine->device_memory->base,
2881                                     &machine->device_memory->mr);
2882     }
2883 
2884     if (smc->dr_lmb_enabled) {
2885         spapr_create_lmb_dr_connectors(spapr);
2886     }
2887 
2888     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI_MCE) == SPAPR_CAP_ON) {
2889         /* Create the error string for live migration blocker */
2890         error_setg(&spapr->fwnmi_migration_blocker,
2891             "A machine check is being handled during migration. The handler"
2892             "may run and log hardware error on the destination");
2893     }
2894 
2895     if (mc->nvdimm_supported) {
2896         spapr_create_nvdimm_dr_connectors(spapr);
2897     }
2898 
2899     /* Set up RTAS event infrastructure */
2900     spapr_events_init(spapr);
2901 
2902     /* Set up the RTC RTAS interfaces */
2903     spapr_rtc_create(spapr);
2904 
2905     /* Set up VIO bus */
2906     spapr->vio_bus = spapr_vio_bus_init();
2907 
2908     for (i = 0; i < serial_max_hds(); i++) {
2909         if (serial_hd(i)) {
2910             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2911         }
2912     }
2913 
2914     /* We always have at least the nvram device on VIO */
2915     spapr_create_nvram(spapr);
2916 
2917     /*
2918      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2919      * connectors (described in root DT node's "ibm,drc-types" property)
2920      * are pre-initialized here. additional child connectors (such as
2921      * connectors for a PHBs PCI slots) are added as needed during their
2922      * parent's realization.
2923      */
2924     if (smc->dr_phb_enabled) {
2925         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2926             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2927         }
2928     }
2929 
2930     /* Set up PCI */
2931     spapr_pci_rtas_init();
2932 
2933     phb = spapr_create_default_phb();
2934 
2935     for (i = 0; i < nb_nics; i++) {
2936         NICInfo *nd = &nd_table[i];
2937 
2938         if (!nd->model) {
2939             nd->model = g_strdup("spapr-vlan");
2940         }
2941 
2942         if (g_str_equal(nd->model, "spapr-vlan") ||
2943             g_str_equal(nd->model, "ibmveth")) {
2944             spapr_vlan_create(spapr->vio_bus, nd);
2945         } else {
2946             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2947         }
2948     }
2949 
2950     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2951         spapr_vscsi_create(spapr->vio_bus);
2952     }
2953 
2954     /* Graphics */
2955     if (spapr_vga_init(phb->bus, &error_fatal)) {
2956         spapr->has_graphics = true;
2957         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2958     }
2959 
2960     if (machine->usb) {
2961         if (smc->use_ohci_by_default) {
2962             pci_create_simple(phb->bus, -1, "pci-ohci");
2963         } else {
2964             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2965         }
2966 
2967         if (spapr->has_graphics) {
2968             USBBus *usb_bus = usb_bus_find(-1);
2969 
2970             usb_create_simple(usb_bus, "usb-kbd");
2971             usb_create_simple(usb_bus, "usb-mouse");
2972         }
2973     }
2974 
2975     if (kernel_filename) {
2976         uint64_t lowaddr = 0;
2977 
2978         spapr->kernel_size = load_elf(kernel_filename, NULL,
2979                                       translate_kernel_address, spapr,
2980                                       NULL, &lowaddr, NULL, NULL, 1,
2981                                       PPC_ELF_MACHINE, 0, 0);
2982         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2983             spapr->kernel_size = load_elf(kernel_filename, NULL,
2984                                           translate_kernel_address, spapr, NULL,
2985                                           &lowaddr, NULL, NULL, 0,
2986                                           PPC_ELF_MACHINE,
2987                                           0, 0);
2988             spapr->kernel_le = spapr->kernel_size > 0;
2989         }
2990         if (spapr->kernel_size < 0) {
2991             error_report("error loading %s: %s", kernel_filename,
2992                          load_elf_strerror(spapr->kernel_size));
2993             exit(1);
2994         }
2995 
2996         /* load initrd */
2997         if (initrd_filename) {
2998             /* Try to locate the initrd in the gap between the kernel
2999              * and the firmware. Add a bit of space just in case
3000              */
3001             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
3002                                   + 0x1ffff) & ~0xffff;
3003             spapr->initrd_size = load_image_targphys(initrd_filename,
3004                                                      spapr->initrd_base,
3005                                                      load_limit
3006                                                      - spapr->initrd_base);
3007             if (spapr->initrd_size < 0) {
3008                 error_report("could not load initial ram disk '%s'",
3009                              initrd_filename);
3010                 exit(1);
3011             }
3012         }
3013     }
3014 
3015     if (bios_name == NULL) {
3016         bios_name = FW_FILE_NAME;
3017     }
3018     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3019     if (!filename) {
3020         error_report("Could not find LPAR firmware '%s'", bios_name);
3021         exit(1);
3022     }
3023     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3024     if (fw_size <= 0) {
3025         error_report("Could not load LPAR firmware '%s'", filename);
3026         exit(1);
3027     }
3028     g_free(filename);
3029 
3030     /* FIXME: Should register things through the MachineState's qdev
3031      * interface, this is a legacy from the sPAPREnvironment structure
3032      * which predated MachineState but had a similar function */
3033     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3034     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3035                          &savevm_htab_handlers, spapr);
3036 
3037     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3038                              &error_fatal);
3039 
3040     qemu_register_boot_set(spapr_boot_set, spapr);
3041 
3042     /*
3043      * Nothing needs to be done to resume a suspended guest because
3044      * suspending does not change the machine state, so no need for
3045      * a ->wakeup method.
3046      */
3047     qemu_register_wakeup_support();
3048 
3049     if (kvm_enabled()) {
3050         /* to stop and start vmclock */
3051         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3052                                          &spapr->tb);
3053 
3054         kvmppc_spapr_enable_inkernel_multitce();
3055     }
3056 
3057     qemu_cond_init(&spapr->mc_delivery_cond);
3058 }
3059 
3060 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3061 {
3062     if (!vm_type) {
3063         return 0;
3064     }
3065 
3066     if (!strcmp(vm_type, "HV")) {
3067         return 1;
3068     }
3069 
3070     if (!strcmp(vm_type, "PR")) {
3071         return 2;
3072     }
3073 
3074     error_report("Unknown kvm-type specified '%s'", vm_type);
3075     exit(1);
3076 }
3077 
3078 /*
3079  * Implementation of an interface to adjust firmware path
3080  * for the bootindex property handling.
3081  */
3082 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3083                                    DeviceState *dev)
3084 {
3085 #define CAST(type, obj, name) \
3086     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3087     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3088     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3089     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3090 
3091     if (d) {
3092         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3093         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3094         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3095 
3096         if (spapr) {
3097             /*
3098              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3099              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3100              * 0x8000 | (target << 8) | (bus << 5) | lun
3101              * (see the "Logical unit addressing format" table in SAM5)
3102              */
3103             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3104             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3105                                    (uint64_t)id << 48);
3106         } else if (virtio) {
3107             /*
3108              * We use SRP luns of the form 01000000 | (target << 8) | lun
3109              * in the top 32 bits of the 64-bit LUN
3110              * Note: the quote above is from SLOF and it is wrong,
3111              * the actual binding is:
3112              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3113              */
3114             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3115             if (d->lun >= 256) {
3116                 /* Use the LUN "flat space addressing method" */
3117                 id |= 0x4000;
3118             }
3119             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3120                                    (uint64_t)id << 32);
3121         } else if (usb) {
3122             /*
3123              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3124              * in the top 32 bits of the 64-bit LUN
3125              */
3126             unsigned usb_port = atoi(usb->port->path);
3127             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3128             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3129                                    (uint64_t)id << 32);
3130         }
3131     }
3132 
3133     /*
3134      * SLOF probes the USB devices, and if it recognizes that the device is a
3135      * storage device, it changes its name to "storage" instead of "usb-host",
3136      * and additionally adds a child node for the SCSI LUN, so the correct
3137      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3138      */
3139     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3140         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3141         if (usb_host_dev_is_scsi_storage(usbdev)) {
3142             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3143         }
3144     }
3145 
3146     if (phb) {
3147         /* Replace "pci" with "pci@800000020000000" */
3148         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3149     }
3150 
3151     if (vsc) {
3152         /* Same logic as virtio above */
3153         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3154         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3155     }
3156 
3157     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3158         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3159         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3160         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3161     }
3162 
3163     return NULL;
3164 }
3165 
3166 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3167 {
3168     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3169 
3170     return g_strdup(spapr->kvm_type);
3171 }
3172 
3173 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3174 {
3175     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3176 
3177     g_free(spapr->kvm_type);
3178     spapr->kvm_type = g_strdup(value);
3179 }
3180 
3181 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3182 {
3183     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3184 
3185     return spapr->use_hotplug_event_source;
3186 }
3187 
3188 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3189                                             Error **errp)
3190 {
3191     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3192 
3193     spapr->use_hotplug_event_source = value;
3194 }
3195 
3196 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3197 {
3198     return true;
3199 }
3200 
3201 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3202 {
3203     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3204 
3205     switch (spapr->resize_hpt) {
3206     case SPAPR_RESIZE_HPT_DEFAULT:
3207         return g_strdup("default");
3208     case SPAPR_RESIZE_HPT_DISABLED:
3209         return g_strdup("disabled");
3210     case SPAPR_RESIZE_HPT_ENABLED:
3211         return g_strdup("enabled");
3212     case SPAPR_RESIZE_HPT_REQUIRED:
3213         return g_strdup("required");
3214     }
3215     g_assert_not_reached();
3216 }
3217 
3218 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3219 {
3220     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3221 
3222     if (strcmp(value, "default") == 0) {
3223         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3224     } else if (strcmp(value, "disabled") == 0) {
3225         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3226     } else if (strcmp(value, "enabled") == 0) {
3227         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3228     } else if (strcmp(value, "required") == 0) {
3229         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3230     } else {
3231         error_setg(errp, "Bad value for \"resize-hpt\" property");
3232     }
3233 }
3234 
3235 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3236                                    void *opaque, Error **errp)
3237 {
3238     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3239 }
3240 
3241 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3242                                    void *opaque, Error **errp)
3243 {
3244     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3245 }
3246 
3247 static void spapr_get_kernel_addr(Object *obj, Visitor *v, const char *name,
3248                                   void *opaque, Error **errp)
3249 {
3250     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3251 }
3252 
3253 static void spapr_set_kernel_addr(Object *obj, Visitor *v, const char *name,
3254                                   void *opaque, Error **errp)
3255 {
3256     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3257 }
3258 
3259 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3260 {
3261     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3262 
3263     if (spapr->irq == &spapr_irq_xics_legacy) {
3264         return g_strdup("legacy");
3265     } else if (spapr->irq == &spapr_irq_xics) {
3266         return g_strdup("xics");
3267     } else if (spapr->irq == &spapr_irq_xive) {
3268         return g_strdup("xive");
3269     } else if (spapr->irq == &spapr_irq_dual) {
3270         return g_strdup("dual");
3271     }
3272     g_assert_not_reached();
3273 }
3274 
3275 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3276 {
3277     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3278 
3279     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3280         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3281         return;
3282     }
3283 
3284     /* The legacy IRQ backend can not be set */
3285     if (strcmp(value, "xics") == 0) {
3286         spapr->irq = &spapr_irq_xics;
3287     } else if (strcmp(value, "xive") == 0) {
3288         spapr->irq = &spapr_irq_xive;
3289     } else if (strcmp(value, "dual") == 0) {
3290         spapr->irq = &spapr_irq_dual;
3291     } else {
3292         error_setg(errp, "Bad value for \"ic-mode\" property");
3293     }
3294 }
3295 
3296 static char *spapr_get_host_model(Object *obj, Error **errp)
3297 {
3298     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3299 
3300     return g_strdup(spapr->host_model);
3301 }
3302 
3303 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3304 {
3305     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3306 
3307     g_free(spapr->host_model);
3308     spapr->host_model = g_strdup(value);
3309 }
3310 
3311 static char *spapr_get_host_serial(Object *obj, Error **errp)
3312 {
3313     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3314 
3315     return g_strdup(spapr->host_serial);
3316 }
3317 
3318 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3319 {
3320     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3321 
3322     g_free(spapr->host_serial);
3323     spapr->host_serial = g_strdup(value);
3324 }
3325 
3326 static void spapr_instance_init(Object *obj)
3327 {
3328     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3329     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3330 
3331     spapr->htab_fd = -1;
3332     spapr->use_hotplug_event_source = true;
3333     object_property_add_str(obj, "kvm-type",
3334                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3335     object_property_set_description(obj, "kvm-type",
3336                                     "Specifies the KVM virtualization mode (HV, PR)",
3337                                     NULL);
3338     object_property_add_bool(obj, "modern-hotplug-events",
3339                             spapr_get_modern_hotplug_events,
3340                             spapr_set_modern_hotplug_events,
3341                             NULL);
3342     object_property_set_description(obj, "modern-hotplug-events",
3343                                     "Use dedicated hotplug event mechanism in"
3344                                     " place of standard EPOW events when possible"
3345                                     " (required for memory hot-unplug support)",
3346                                     NULL);
3347     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3348                             "Maximum permitted CPU compatibility mode",
3349                             &error_fatal);
3350 
3351     object_property_add_str(obj, "resize-hpt",
3352                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3353     object_property_set_description(obj, "resize-hpt",
3354                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3355                                     NULL);
3356     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3357                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3358     object_property_set_description(obj, "vsmt",
3359                                     "Virtual SMT: KVM behaves as if this were"
3360                                     " the host's SMT mode", &error_abort);
3361     object_property_add_bool(obj, "vfio-no-msix-emulation",
3362                              spapr_get_msix_emulation, NULL, NULL);
3363 
3364     object_property_add(obj, "kernel-addr", "uint64", spapr_get_kernel_addr,
3365                         spapr_set_kernel_addr, NULL, &spapr->kernel_addr,
3366                         &error_abort);
3367     object_property_set_description(obj, "kernel-addr",
3368                                     stringify(KERNEL_LOAD_ADDR)
3369                                     " for -kernel is the default",
3370                                     NULL);
3371     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3372     /* The machine class defines the default interrupt controller mode */
3373     spapr->irq = smc->irq;
3374     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3375                             spapr_set_ic_mode, NULL);
3376     object_property_set_description(obj, "ic-mode",
3377                  "Specifies the interrupt controller mode (xics, xive, dual)",
3378                  NULL);
3379 
3380     object_property_add_str(obj, "host-model",
3381         spapr_get_host_model, spapr_set_host_model,
3382         &error_abort);
3383     object_property_set_description(obj, "host-model",
3384         "Host model to advertise in guest device tree", &error_abort);
3385     object_property_add_str(obj, "host-serial",
3386         spapr_get_host_serial, spapr_set_host_serial,
3387         &error_abort);
3388     object_property_set_description(obj, "host-serial",
3389         "Host serial number to advertise in guest device tree", &error_abort);
3390 }
3391 
3392 static void spapr_machine_finalizefn(Object *obj)
3393 {
3394     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3395 
3396     g_free(spapr->kvm_type);
3397 }
3398 
3399 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3400 {
3401     cpu_synchronize_state(cs);
3402     ppc_cpu_do_system_reset(cs);
3403 }
3404 
3405 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3406 {
3407     CPUState *cs;
3408 
3409     CPU_FOREACH(cs) {
3410         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3411     }
3412 }
3413 
3414 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3415                           void *fdt, int *fdt_start_offset, Error **errp)
3416 {
3417     uint64_t addr;
3418     uint32_t node;
3419 
3420     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3421     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3422                                     &error_abort);
3423     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3424                                                    SPAPR_MEMORY_BLOCK_SIZE);
3425     return 0;
3426 }
3427 
3428 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3429                            bool dedicated_hp_event_source, Error **errp)
3430 {
3431     SpaprDrc *drc;
3432     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3433     int i;
3434     uint64_t addr = addr_start;
3435     bool hotplugged = spapr_drc_hotplugged(dev);
3436     Error *local_err = NULL;
3437 
3438     for (i = 0; i < nr_lmbs; i++) {
3439         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3440                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3441         g_assert(drc);
3442 
3443         spapr_drc_attach(drc, dev, &local_err);
3444         if (local_err) {
3445             while (addr > addr_start) {
3446                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3447                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3448                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3449                 spapr_drc_detach(drc);
3450             }
3451             error_propagate(errp, local_err);
3452             return;
3453         }
3454         if (!hotplugged) {
3455             spapr_drc_reset(drc);
3456         }
3457         addr += SPAPR_MEMORY_BLOCK_SIZE;
3458     }
3459     /* send hotplug notification to the
3460      * guest only in case of hotplugged memory
3461      */
3462     if (hotplugged) {
3463         if (dedicated_hp_event_source) {
3464             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3465                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3466             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3467                                                    nr_lmbs,
3468                                                    spapr_drc_index(drc));
3469         } else {
3470             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3471                                            nr_lmbs);
3472         }
3473     }
3474 }
3475 
3476 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3477                               Error **errp)
3478 {
3479     Error *local_err = NULL;
3480     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3481     PCDIMMDevice *dimm = PC_DIMM(dev);
3482     uint64_t size, addr, slot;
3483     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3484 
3485     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3486 
3487     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3488     if (local_err) {
3489         goto out;
3490     }
3491 
3492     if (!is_nvdimm) {
3493         addr = object_property_get_uint(OBJECT(dimm),
3494                                         PC_DIMM_ADDR_PROP, &local_err);
3495         if (local_err) {
3496             goto out_unplug;
3497         }
3498         spapr_add_lmbs(dev, addr, size,
3499                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3500                        &local_err);
3501     } else {
3502         slot = object_property_get_uint(OBJECT(dimm),
3503                                         PC_DIMM_SLOT_PROP, &local_err);
3504         if (local_err) {
3505             goto out_unplug;
3506         }
3507         spapr_add_nvdimm(dev, slot, &local_err);
3508     }
3509 
3510     if (local_err) {
3511         goto out_unplug;
3512     }
3513 
3514     return;
3515 
3516 out_unplug:
3517     pc_dimm_unplug(dimm, MACHINE(ms));
3518 out:
3519     error_propagate(errp, local_err);
3520 }
3521 
3522 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3523                                   Error **errp)
3524 {
3525     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3526     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3527     const MachineClass *mc = MACHINE_CLASS(smc);
3528     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3529     PCDIMMDevice *dimm = PC_DIMM(dev);
3530     Error *local_err = NULL;
3531     uint64_t size;
3532     Object *memdev;
3533     hwaddr pagesize;
3534 
3535     if (!smc->dr_lmb_enabled) {
3536         error_setg(errp, "Memory hotplug not supported for this machine");
3537         return;
3538     }
3539 
3540     if (is_nvdimm && !mc->nvdimm_supported) {
3541         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3542         return;
3543     }
3544 
3545     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3546     if (local_err) {
3547         error_propagate(errp, local_err);
3548         return;
3549     }
3550 
3551     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3552         error_setg(errp, "Hotplugged memory size must be a multiple of "
3553                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3554         return;
3555     } else if (is_nvdimm) {
3556         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3557         if (local_err) {
3558             error_propagate(errp, local_err);
3559             return;
3560         }
3561     }
3562 
3563     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3564                                       &error_abort);
3565     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3566     spapr_check_pagesize(spapr, pagesize, &local_err);
3567     if (local_err) {
3568         error_propagate(errp, local_err);
3569         return;
3570     }
3571 
3572     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3573 }
3574 
3575 struct SpaprDimmState {
3576     PCDIMMDevice *dimm;
3577     uint32_t nr_lmbs;
3578     QTAILQ_ENTRY(SpaprDimmState) next;
3579 };
3580 
3581 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3582                                                        PCDIMMDevice *dimm)
3583 {
3584     SpaprDimmState *dimm_state = NULL;
3585 
3586     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3587         if (dimm_state->dimm == dimm) {
3588             break;
3589         }
3590     }
3591     return dimm_state;
3592 }
3593 
3594 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3595                                                       uint32_t nr_lmbs,
3596                                                       PCDIMMDevice *dimm)
3597 {
3598     SpaprDimmState *ds = NULL;
3599 
3600     /*
3601      * If this request is for a DIMM whose removal had failed earlier
3602      * (due to guest's refusal to remove the LMBs), we would have this
3603      * dimm already in the pending_dimm_unplugs list. In that
3604      * case don't add again.
3605      */
3606     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3607     if (!ds) {
3608         ds = g_malloc0(sizeof(SpaprDimmState));
3609         ds->nr_lmbs = nr_lmbs;
3610         ds->dimm = dimm;
3611         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3612     }
3613     return ds;
3614 }
3615 
3616 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3617                                               SpaprDimmState *dimm_state)
3618 {
3619     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3620     g_free(dimm_state);
3621 }
3622 
3623 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3624                                                         PCDIMMDevice *dimm)
3625 {
3626     SpaprDrc *drc;
3627     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3628                                                   &error_abort);
3629     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3630     uint32_t avail_lmbs = 0;
3631     uint64_t addr_start, addr;
3632     int i;
3633 
3634     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3635                                          &error_abort);
3636 
3637     addr = addr_start;
3638     for (i = 0; i < nr_lmbs; i++) {
3639         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3640                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3641         g_assert(drc);
3642         if (drc->dev) {
3643             avail_lmbs++;
3644         }
3645         addr += SPAPR_MEMORY_BLOCK_SIZE;
3646     }
3647 
3648     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3649 }
3650 
3651 /* Callback to be called during DRC release. */
3652 void spapr_lmb_release(DeviceState *dev)
3653 {
3654     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3655     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3656     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3657 
3658     /* This information will get lost if a migration occurs
3659      * during the unplug process. In this case recover it. */
3660     if (ds == NULL) {
3661         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3662         g_assert(ds);
3663         /* The DRC being examined by the caller at least must be counted */
3664         g_assert(ds->nr_lmbs);
3665     }
3666 
3667     if (--ds->nr_lmbs) {
3668         return;
3669     }
3670 
3671     /*
3672      * Now that all the LMBs have been removed by the guest, call the
3673      * unplug handler chain. This can never fail.
3674      */
3675     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3676     object_unparent(OBJECT(dev));
3677 }
3678 
3679 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3680 {
3681     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3682     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3683 
3684     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3685     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3686     spapr_pending_dimm_unplugs_remove(spapr, ds);
3687 }
3688 
3689 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3690                                         DeviceState *dev, Error **errp)
3691 {
3692     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3693     Error *local_err = NULL;
3694     PCDIMMDevice *dimm = PC_DIMM(dev);
3695     uint32_t nr_lmbs;
3696     uint64_t size, addr_start, addr;
3697     int i;
3698     SpaprDrc *drc;
3699 
3700     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3701         error_setg(&local_err,
3702                    "nvdimm device hot unplug is not supported yet.");
3703         goto out;
3704     }
3705 
3706     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3707     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3708 
3709     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3710                                          &local_err);
3711     if (local_err) {
3712         goto out;
3713     }
3714 
3715     /*
3716      * An existing pending dimm state for this DIMM means that there is an
3717      * unplug operation in progress, waiting for the spapr_lmb_release
3718      * callback to complete the job (BQL can't cover that far). In this case,
3719      * bail out to avoid detaching DRCs that were already released.
3720      */
3721     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3722         error_setg(&local_err,
3723                    "Memory unplug already in progress for device %s",
3724                    dev->id);
3725         goto out;
3726     }
3727 
3728     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3729 
3730     addr = addr_start;
3731     for (i = 0; i < nr_lmbs; i++) {
3732         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3733                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3734         g_assert(drc);
3735 
3736         spapr_drc_detach(drc);
3737         addr += SPAPR_MEMORY_BLOCK_SIZE;
3738     }
3739 
3740     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3741                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3742     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3743                                               nr_lmbs, spapr_drc_index(drc));
3744 out:
3745     error_propagate(errp, local_err);
3746 }
3747 
3748 /* Callback to be called during DRC release. */
3749 void spapr_core_release(DeviceState *dev)
3750 {
3751     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3752 
3753     /* Call the unplug handler chain. This can never fail. */
3754     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3755     object_unparent(OBJECT(dev));
3756 }
3757 
3758 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3759 {
3760     MachineState *ms = MACHINE(hotplug_dev);
3761     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3762     CPUCore *cc = CPU_CORE(dev);
3763     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3764 
3765     if (smc->pre_2_10_has_unused_icps) {
3766         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3767         int i;
3768 
3769         for (i = 0; i < cc->nr_threads; i++) {
3770             CPUState *cs = CPU(sc->threads[i]);
3771 
3772             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3773         }
3774     }
3775 
3776     assert(core_slot);
3777     core_slot->cpu = NULL;
3778     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3779 }
3780 
3781 static
3782 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3783                                Error **errp)
3784 {
3785     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3786     int index;
3787     SpaprDrc *drc;
3788     CPUCore *cc = CPU_CORE(dev);
3789 
3790     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3791         error_setg(errp, "Unable to find CPU core with core-id: %d",
3792                    cc->core_id);
3793         return;
3794     }
3795     if (index == 0) {
3796         error_setg(errp, "Boot CPU core may not be unplugged");
3797         return;
3798     }
3799 
3800     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3801                           spapr_vcpu_id(spapr, cc->core_id));
3802     g_assert(drc);
3803 
3804     if (!spapr_drc_unplug_requested(drc)) {
3805         spapr_drc_detach(drc);
3806         spapr_hotplug_req_remove_by_index(drc);
3807     }
3808 }
3809 
3810 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3811                            void *fdt, int *fdt_start_offset, Error **errp)
3812 {
3813     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3814     CPUState *cs = CPU(core->threads[0]);
3815     PowerPCCPU *cpu = POWERPC_CPU(cs);
3816     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3817     int id = spapr_get_vcpu_id(cpu);
3818     char *nodename;
3819     int offset;
3820 
3821     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3822     offset = fdt_add_subnode(fdt, 0, nodename);
3823     g_free(nodename);
3824 
3825     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3826 
3827     *fdt_start_offset = offset;
3828     return 0;
3829 }
3830 
3831 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3832                             Error **errp)
3833 {
3834     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3835     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3836     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3837     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3838     CPUCore *cc = CPU_CORE(dev);
3839     CPUState *cs;
3840     SpaprDrc *drc;
3841     Error *local_err = NULL;
3842     CPUArchId *core_slot;
3843     int index;
3844     bool hotplugged = spapr_drc_hotplugged(dev);
3845     int i;
3846 
3847     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3848     if (!core_slot) {
3849         error_setg(errp, "Unable to find CPU core with core-id: %d",
3850                    cc->core_id);
3851         return;
3852     }
3853     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3854                           spapr_vcpu_id(spapr, cc->core_id));
3855 
3856     g_assert(drc || !mc->has_hotpluggable_cpus);
3857 
3858     if (drc) {
3859         spapr_drc_attach(drc, dev, &local_err);
3860         if (local_err) {
3861             error_propagate(errp, local_err);
3862             return;
3863         }
3864 
3865         if (hotplugged) {
3866             /*
3867              * Send hotplug notification interrupt to the guest only
3868              * in case of hotplugged CPUs.
3869              */
3870             spapr_hotplug_req_add_by_index(drc);
3871         } else {
3872             spapr_drc_reset(drc);
3873         }
3874     }
3875 
3876     core_slot->cpu = OBJECT(dev);
3877 
3878     if (smc->pre_2_10_has_unused_icps) {
3879         for (i = 0; i < cc->nr_threads; i++) {
3880             cs = CPU(core->threads[i]);
3881             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3882         }
3883     }
3884 
3885     /*
3886      * Set compatibility mode to match the boot CPU, which was either set
3887      * by the machine reset code or by CAS.
3888      */
3889     if (hotplugged) {
3890         for (i = 0; i < cc->nr_threads; i++) {
3891             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3892                            &local_err);
3893             if (local_err) {
3894                 error_propagate(errp, local_err);
3895                 return;
3896             }
3897         }
3898     }
3899 }
3900 
3901 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3902                                 Error **errp)
3903 {
3904     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3905     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3906     Error *local_err = NULL;
3907     CPUCore *cc = CPU_CORE(dev);
3908     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3909     const char *type = object_get_typename(OBJECT(dev));
3910     CPUArchId *core_slot;
3911     int index;
3912     unsigned int smp_threads = machine->smp.threads;
3913 
3914     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3915         error_setg(&local_err, "CPU hotplug not supported for this machine");
3916         goto out;
3917     }
3918 
3919     if (strcmp(base_core_type, type)) {
3920         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3921         goto out;
3922     }
3923 
3924     if (cc->core_id % smp_threads) {
3925         error_setg(&local_err, "invalid core id %d", cc->core_id);
3926         goto out;
3927     }
3928 
3929     /*
3930      * In general we should have homogeneous threads-per-core, but old
3931      * (pre hotplug support) machine types allow the last core to have
3932      * reduced threads as a compatibility hack for when we allowed
3933      * total vcpus not a multiple of threads-per-core.
3934      */
3935     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3936         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3937                    cc->nr_threads, smp_threads);
3938         goto out;
3939     }
3940 
3941     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3942     if (!core_slot) {
3943         error_setg(&local_err, "core id %d out of range", cc->core_id);
3944         goto out;
3945     }
3946 
3947     if (core_slot->cpu) {
3948         error_setg(&local_err, "core %d already populated", cc->core_id);
3949         goto out;
3950     }
3951 
3952     numa_cpu_pre_plug(core_slot, dev, &local_err);
3953 
3954 out:
3955     error_propagate(errp, local_err);
3956 }
3957 
3958 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3959                           void *fdt, int *fdt_start_offset, Error **errp)
3960 {
3961     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3962     int intc_phandle;
3963 
3964     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3965     if (intc_phandle <= 0) {
3966         return -1;
3967     }
3968 
3969     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3970         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3971         return -1;
3972     }
3973 
3974     /* generally SLOF creates these, for hotplug it's up to QEMU */
3975     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3976 
3977     return 0;
3978 }
3979 
3980 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3981                                Error **errp)
3982 {
3983     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3984     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3985     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3986     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3987 
3988     if (dev->hotplugged && !smc->dr_phb_enabled) {
3989         error_setg(errp, "PHB hotplug not supported for this machine");
3990         return;
3991     }
3992 
3993     if (sphb->index == (uint32_t)-1) {
3994         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3995         return;
3996     }
3997 
3998     /*
3999      * This will check that sphb->index doesn't exceed the maximum number of
4000      * PHBs for the current machine type.
4001      */
4002     smc->phb_placement(spapr, sphb->index,
4003                        &sphb->buid, &sphb->io_win_addr,
4004                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
4005                        windows_supported, sphb->dma_liobn,
4006                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
4007                        errp);
4008 }
4009 
4010 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4011                            Error **errp)
4012 {
4013     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4014     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4015     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4016     SpaprDrc *drc;
4017     bool hotplugged = spapr_drc_hotplugged(dev);
4018     Error *local_err = NULL;
4019 
4020     if (!smc->dr_phb_enabled) {
4021         return;
4022     }
4023 
4024     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4025     /* hotplug hooks should check it's enabled before getting this far */
4026     assert(drc);
4027 
4028     spapr_drc_attach(drc, DEVICE(dev), &local_err);
4029     if (local_err) {
4030         error_propagate(errp, local_err);
4031         return;
4032     }
4033 
4034     if (hotplugged) {
4035         spapr_hotplug_req_add_by_index(drc);
4036     } else {
4037         spapr_drc_reset(drc);
4038     }
4039 }
4040 
4041 void spapr_phb_release(DeviceState *dev)
4042 {
4043     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4044 
4045     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4046     object_unparent(OBJECT(dev));
4047 }
4048 
4049 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4050 {
4051     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4052 }
4053 
4054 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4055                                      DeviceState *dev, Error **errp)
4056 {
4057     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4058     SpaprDrc *drc;
4059 
4060     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4061     assert(drc);
4062 
4063     if (!spapr_drc_unplug_requested(drc)) {
4064         spapr_drc_detach(drc);
4065         spapr_hotplug_req_remove_by_index(drc);
4066     }
4067 }
4068 
4069 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4070                                  Error **errp)
4071 {
4072     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4073     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4074 
4075     if (spapr->tpm_proxy != NULL) {
4076         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4077         return;
4078     }
4079 
4080     spapr->tpm_proxy = tpm_proxy;
4081 }
4082 
4083 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4084 {
4085     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4086 
4087     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4088     object_unparent(OBJECT(dev));
4089     spapr->tpm_proxy = NULL;
4090 }
4091 
4092 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4093                                       DeviceState *dev, Error **errp)
4094 {
4095     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4096         spapr_memory_plug(hotplug_dev, dev, errp);
4097     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4098         spapr_core_plug(hotplug_dev, dev, errp);
4099     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4100         spapr_phb_plug(hotplug_dev, dev, errp);
4101     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4102         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4103     }
4104 }
4105 
4106 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4107                                         DeviceState *dev, Error **errp)
4108 {
4109     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4110         spapr_memory_unplug(hotplug_dev, dev);
4111     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4112         spapr_core_unplug(hotplug_dev, dev);
4113     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4114         spapr_phb_unplug(hotplug_dev, dev);
4115     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4116         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4117     }
4118 }
4119 
4120 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4121                                                 DeviceState *dev, Error **errp)
4122 {
4123     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4124     MachineClass *mc = MACHINE_GET_CLASS(sms);
4125     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4126 
4127     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4128         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4129             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4130         } else {
4131             /* NOTE: this means there is a window after guest reset, prior to
4132              * CAS negotiation, where unplug requests will fail due to the
4133              * capability not being detected yet. This is a bit different than
4134              * the case with PCI unplug, where the events will be queued and
4135              * eventually handled by the guest after boot
4136              */
4137             error_setg(errp, "Memory hot unplug not supported for this guest");
4138         }
4139     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4140         if (!mc->has_hotpluggable_cpus) {
4141             error_setg(errp, "CPU hot unplug not supported on this machine");
4142             return;
4143         }
4144         spapr_core_unplug_request(hotplug_dev, dev, errp);
4145     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4146         if (!smc->dr_phb_enabled) {
4147             error_setg(errp, "PHB hot unplug not supported on this machine");
4148             return;
4149         }
4150         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4151     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4152         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4153     }
4154 }
4155 
4156 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4157                                           DeviceState *dev, Error **errp)
4158 {
4159     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4160         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4161     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4162         spapr_core_pre_plug(hotplug_dev, dev, errp);
4163     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4164         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4165     }
4166 }
4167 
4168 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4169                                                  DeviceState *dev)
4170 {
4171     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4172         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4173         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4174         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4175         return HOTPLUG_HANDLER(machine);
4176     }
4177     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4178         PCIDevice *pcidev = PCI_DEVICE(dev);
4179         PCIBus *root = pci_device_root_bus(pcidev);
4180         SpaprPhbState *phb =
4181             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4182                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4183 
4184         if (phb) {
4185             return HOTPLUG_HANDLER(phb);
4186         }
4187     }
4188     return NULL;
4189 }
4190 
4191 static CpuInstanceProperties
4192 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4193 {
4194     CPUArchId *core_slot;
4195     MachineClass *mc = MACHINE_GET_CLASS(machine);
4196 
4197     /* make sure possible_cpu are intialized */
4198     mc->possible_cpu_arch_ids(machine);
4199     /* get CPU core slot containing thread that matches cpu_index */
4200     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4201     assert(core_slot);
4202     return core_slot->props;
4203 }
4204 
4205 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4206 {
4207     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4208 }
4209 
4210 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4211 {
4212     int i;
4213     unsigned int smp_threads = machine->smp.threads;
4214     unsigned int smp_cpus = machine->smp.cpus;
4215     const char *core_type;
4216     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4217     MachineClass *mc = MACHINE_GET_CLASS(machine);
4218 
4219     if (!mc->has_hotpluggable_cpus) {
4220         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4221     }
4222     if (machine->possible_cpus) {
4223         assert(machine->possible_cpus->len == spapr_max_cores);
4224         return machine->possible_cpus;
4225     }
4226 
4227     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4228     if (!core_type) {
4229         error_report("Unable to find sPAPR CPU Core definition");
4230         exit(1);
4231     }
4232 
4233     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4234                              sizeof(CPUArchId) * spapr_max_cores);
4235     machine->possible_cpus->len = spapr_max_cores;
4236     for (i = 0; i < machine->possible_cpus->len; i++) {
4237         int core_id = i * smp_threads;
4238 
4239         machine->possible_cpus->cpus[i].type = core_type;
4240         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4241         machine->possible_cpus->cpus[i].arch_id = core_id;
4242         machine->possible_cpus->cpus[i].props.has_core_id = true;
4243         machine->possible_cpus->cpus[i].props.core_id = core_id;
4244     }
4245     return machine->possible_cpus;
4246 }
4247 
4248 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4249                                 uint64_t *buid, hwaddr *pio,
4250                                 hwaddr *mmio32, hwaddr *mmio64,
4251                                 unsigned n_dma, uint32_t *liobns,
4252                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4253 {
4254     /*
4255      * New-style PHB window placement.
4256      *
4257      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4258      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4259      * windows.
4260      *
4261      * Some guest kernels can't work with MMIO windows above 1<<46
4262      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4263      *
4264      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4265      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4266      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4267      * 1TiB 64-bit MMIO windows for each PHB.
4268      */
4269     const uint64_t base_buid = 0x800000020000000ULL;
4270     int i;
4271 
4272     /* Sanity check natural alignments */
4273     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4274     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4275     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4276     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4277     /* Sanity check bounds */
4278     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4279                       SPAPR_PCI_MEM32_WIN_SIZE);
4280     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4281                       SPAPR_PCI_MEM64_WIN_SIZE);
4282 
4283     if (index >= SPAPR_MAX_PHBS) {
4284         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4285                    SPAPR_MAX_PHBS - 1);
4286         return;
4287     }
4288 
4289     *buid = base_buid + index;
4290     for (i = 0; i < n_dma; ++i) {
4291         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4292     }
4293 
4294     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4295     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4296     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4297 
4298     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4299     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4300 }
4301 
4302 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4303 {
4304     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4305 
4306     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4307 }
4308 
4309 static void spapr_ics_resend(XICSFabric *dev)
4310 {
4311     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4312 
4313     ics_resend(spapr->ics);
4314 }
4315 
4316 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4317 {
4318     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4319 
4320     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4321 }
4322 
4323 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4324                                  Monitor *mon)
4325 {
4326     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4327 
4328     spapr_irq_print_info(spapr, mon);
4329     monitor_printf(mon, "irqchip: %s\n",
4330                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4331 }
4332 
4333 /*
4334  * This is a XIVE only operation
4335  */
4336 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4337                            uint8_t nvt_blk, uint32_t nvt_idx,
4338                            bool cam_ignore, uint8_t priority,
4339                            uint32_t logic_serv, XiveTCTXMatch *match)
4340 {
4341     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4342     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4343     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4344     int count;
4345 
4346     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4347                            priority, logic_serv, match);
4348     if (count < 0) {
4349         return count;
4350     }
4351 
4352     /*
4353      * When we implement the save and restore of the thread interrupt
4354      * contexts in the enter/exit CPU handlers of the machine and the
4355      * escalations in QEMU, we should be able to handle non dispatched
4356      * vCPUs.
4357      *
4358      * Until this is done, the sPAPR machine should find at least one
4359      * matching context always.
4360      */
4361     if (count == 0) {
4362         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4363                       nvt_blk, nvt_idx);
4364     }
4365 
4366     return count;
4367 }
4368 
4369 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4370 {
4371     return cpu->vcpu_id;
4372 }
4373 
4374 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4375 {
4376     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4377     MachineState *ms = MACHINE(spapr);
4378     int vcpu_id;
4379 
4380     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4381 
4382     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4383         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4384         error_append_hint(errp, "Adjust the number of cpus to %d "
4385                           "or try to raise the number of threads per core\n",
4386                           vcpu_id * ms->smp.threads / spapr->vsmt);
4387         return;
4388     }
4389 
4390     cpu->vcpu_id = vcpu_id;
4391 }
4392 
4393 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4394 {
4395     CPUState *cs;
4396 
4397     CPU_FOREACH(cs) {
4398         PowerPCCPU *cpu = POWERPC_CPU(cs);
4399 
4400         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4401             return cpu;
4402         }
4403     }
4404 
4405     return NULL;
4406 }
4407 
4408 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4409 {
4410     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4411 
4412     /* These are only called by TCG, KVM maintains dispatch state */
4413 
4414     spapr_cpu->prod = false;
4415     if (spapr_cpu->vpa_addr) {
4416         CPUState *cs = CPU(cpu);
4417         uint32_t dispatch;
4418 
4419         dispatch = ldl_be_phys(cs->as,
4420                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4421         dispatch++;
4422         if ((dispatch & 1) != 0) {
4423             qemu_log_mask(LOG_GUEST_ERROR,
4424                           "VPA: incorrect dispatch counter value for "
4425                           "dispatched partition %u, correcting.\n", dispatch);
4426             dispatch++;
4427         }
4428         stl_be_phys(cs->as,
4429                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4430     }
4431 }
4432 
4433 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4434 {
4435     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4436 
4437     if (spapr_cpu->vpa_addr) {
4438         CPUState *cs = CPU(cpu);
4439         uint32_t dispatch;
4440 
4441         dispatch = ldl_be_phys(cs->as,
4442                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4443         dispatch++;
4444         if ((dispatch & 1) != 1) {
4445             qemu_log_mask(LOG_GUEST_ERROR,
4446                           "VPA: incorrect dispatch counter value for "
4447                           "preempted partition %u, correcting.\n", dispatch);
4448             dispatch++;
4449         }
4450         stl_be_phys(cs->as,
4451                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4452     }
4453 }
4454 
4455 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4456 {
4457     MachineClass *mc = MACHINE_CLASS(oc);
4458     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4459     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4460     NMIClass *nc = NMI_CLASS(oc);
4461     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4462     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4463     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4464     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4465     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4466 
4467     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4468     mc->ignore_boot_device_suffixes = true;
4469 
4470     /*
4471      * We set up the default / latest behaviour here.  The class_init
4472      * functions for the specific versioned machine types can override
4473      * these details for backwards compatibility
4474      */
4475     mc->init = spapr_machine_init;
4476     mc->reset = spapr_machine_reset;
4477     mc->block_default_type = IF_SCSI;
4478     mc->max_cpus = 1024;
4479     mc->no_parallel = 1;
4480     mc->default_boot_order = "";
4481     mc->default_ram_size = 512 * MiB;
4482     mc->default_ram_id = "ppc_spapr.ram";
4483     mc->default_display = "std";
4484     mc->kvm_type = spapr_kvm_type;
4485     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4486     mc->pci_allow_0_address = true;
4487     assert(!mc->get_hotplug_handler);
4488     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4489     hc->pre_plug = spapr_machine_device_pre_plug;
4490     hc->plug = spapr_machine_device_plug;
4491     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4492     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4493     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4494     hc->unplug_request = spapr_machine_device_unplug_request;
4495     hc->unplug = spapr_machine_device_unplug;
4496 
4497     smc->dr_lmb_enabled = true;
4498     smc->update_dt_enabled = true;
4499     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4500     mc->has_hotpluggable_cpus = true;
4501     mc->nvdimm_supported = true;
4502     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4503     fwc->get_dev_path = spapr_get_fw_dev_path;
4504     nc->nmi_monitor_handler = spapr_nmi;
4505     smc->phb_placement = spapr_phb_placement;
4506     vhc->hypercall = emulate_spapr_hypercall;
4507     vhc->hpt_mask = spapr_hpt_mask;
4508     vhc->map_hptes = spapr_map_hptes;
4509     vhc->unmap_hptes = spapr_unmap_hptes;
4510     vhc->hpte_set_c = spapr_hpte_set_c;
4511     vhc->hpte_set_r = spapr_hpte_set_r;
4512     vhc->get_pate = spapr_get_pate;
4513     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4514     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4515     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4516     xic->ics_get = spapr_ics_get;
4517     xic->ics_resend = spapr_ics_resend;
4518     xic->icp_get = spapr_icp_get;
4519     ispc->print_info = spapr_pic_print_info;
4520     /* Force NUMA node memory size to be a multiple of
4521      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4522      * in which LMBs are represented and hot-added
4523      */
4524     mc->numa_mem_align_shift = 28;
4525     mc->numa_mem_supported = true;
4526     mc->auto_enable_numa = true;
4527 
4528     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4529     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4530     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4531     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4532     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4533     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4534     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4535     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4536     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4537     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4538     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_ON;
4539     spapr_caps_add_properties(smc, &error_abort);
4540     smc->irq = &spapr_irq_dual;
4541     smc->dr_phb_enabled = true;
4542     smc->linux_pci_probe = true;
4543     smc->smp_threads_vsmt = true;
4544     smc->nr_xirqs = SPAPR_NR_XIRQS;
4545     xfc->match_nvt = spapr_match_nvt;
4546 }
4547 
4548 static const TypeInfo spapr_machine_info = {
4549     .name          = TYPE_SPAPR_MACHINE,
4550     .parent        = TYPE_MACHINE,
4551     .abstract      = true,
4552     .instance_size = sizeof(SpaprMachineState),
4553     .instance_init = spapr_instance_init,
4554     .instance_finalize = spapr_machine_finalizefn,
4555     .class_size    = sizeof(SpaprMachineClass),
4556     .class_init    = spapr_machine_class_init,
4557     .interfaces = (InterfaceInfo[]) {
4558         { TYPE_FW_PATH_PROVIDER },
4559         { TYPE_NMI },
4560         { TYPE_HOTPLUG_HANDLER },
4561         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4562         { TYPE_XICS_FABRIC },
4563         { TYPE_INTERRUPT_STATS_PROVIDER },
4564         { TYPE_XIVE_FABRIC },
4565         { }
4566     },
4567 };
4568 
4569 static void spapr_machine_latest_class_options(MachineClass *mc)
4570 {
4571     mc->alias = "pseries";
4572     mc->is_default = true;
4573 }
4574 
4575 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4576     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4577                                                     void *data)      \
4578     {                                                                \
4579         MachineClass *mc = MACHINE_CLASS(oc);                        \
4580         spapr_machine_##suffix##_class_options(mc);                  \
4581         if (latest) {                                                \
4582             spapr_machine_latest_class_options(mc);                  \
4583         }                                                            \
4584     }                                                                \
4585     static const TypeInfo spapr_machine_##suffix##_info = {          \
4586         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4587         .parent = TYPE_SPAPR_MACHINE,                                \
4588         .class_init = spapr_machine_##suffix##_class_init,           \
4589     };                                                               \
4590     static void spapr_machine_register_##suffix(void)                \
4591     {                                                                \
4592         type_register(&spapr_machine_##suffix##_info);               \
4593     }                                                                \
4594     type_init(spapr_machine_register_##suffix)
4595 
4596 /*
4597  * pseries-5.0
4598  */
4599 static void spapr_machine_5_0_class_options(MachineClass *mc)
4600 {
4601     /* Defaults for the latest behaviour inherited from the base class */
4602 }
4603 
4604 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4605 
4606 /*
4607  * pseries-4.2
4608  */
4609 static void spapr_machine_4_2_class_options(MachineClass *mc)
4610 {
4611     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4612 
4613     spapr_machine_5_0_class_options(mc);
4614     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4615     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4616     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_OFF;
4617     smc->rma_limit = 16 * GiB;
4618     mc->nvdimm_supported = false;
4619 }
4620 
4621 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4622 
4623 /*
4624  * pseries-4.1
4625  */
4626 static void spapr_machine_4_1_class_options(MachineClass *mc)
4627 {
4628     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4629     static GlobalProperty compat[] = {
4630         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4631         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4632     };
4633 
4634     spapr_machine_4_2_class_options(mc);
4635     smc->linux_pci_probe = false;
4636     smc->smp_threads_vsmt = false;
4637     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4638     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4639 }
4640 
4641 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4642 
4643 /*
4644  * pseries-4.0
4645  */
4646 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4647                               uint64_t *buid, hwaddr *pio,
4648                               hwaddr *mmio32, hwaddr *mmio64,
4649                               unsigned n_dma, uint32_t *liobns,
4650                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4651 {
4652     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4653                         nv2gpa, nv2atsd, errp);
4654     *nv2gpa = 0;
4655     *nv2atsd = 0;
4656 }
4657 
4658 static void spapr_machine_4_0_class_options(MachineClass *mc)
4659 {
4660     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4661 
4662     spapr_machine_4_1_class_options(mc);
4663     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4664     smc->phb_placement = phb_placement_4_0;
4665     smc->irq = &spapr_irq_xics;
4666     smc->pre_4_1_migration = true;
4667 }
4668 
4669 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4670 
4671 /*
4672  * pseries-3.1
4673  */
4674 static void spapr_machine_3_1_class_options(MachineClass *mc)
4675 {
4676     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4677 
4678     spapr_machine_4_0_class_options(mc);
4679     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4680 
4681     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4682     smc->update_dt_enabled = false;
4683     smc->dr_phb_enabled = false;
4684     smc->broken_host_serial_model = true;
4685     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4686     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4687     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4688     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4689 }
4690 
4691 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4692 
4693 /*
4694  * pseries-3.0
4695  */
4696 
4697 static void spapr_machine_3_0_class_options(MachineClass *mc)
4698 {
4699     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4700 
4701     spapr_machine_3_1_class_options(mc);
4702     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4703 
4704     smc->legacy_irq_allocation = true;
4705     smc->nr_xirqs = 0x400;
4706     smc->irq = &spapr_irq_xics_legacy;
4707 }
4708 
4709 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4710 
4711 /*
4712  * pseries-2.12
4713  */
4714 static void spapr_machine_2_12_class_options(MachineClass *mc)
4715 {
4716     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4717     static GlobalProperty compat[] = {
4718         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4719         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4720     };
4721 
4722     spapr_machine_3_0_class_options(mc);
4723     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4724     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4725 
4726     /* We depend on kvm_enabled() to choose a default value for the
4727      * hpt-max-page-size capability. Of course we can't do it here
4728      * because this is too early and the HW accelerator isn't initialzed
4729      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4730      */
4731     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4732 }
4733 
4734 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4735 
4736 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4737 {
4738     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4739 
4740     spapr_machine_2_12_class_options(mc);
4741     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4742     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4743     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4744 }
4745 
4746 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4747 
4748 /*
4749  * pseries-2.11
4750  */
4751 
4752 static void spapr_machine_2_11_class_options(MachineClass *mc)
4753 {
4754     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4755 
4756     spapr_machine_2_12_class_options(mc);
4757     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4758     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4759 }
4760 
4761 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4762 
4763 /*
4764  * pseries-2.10
4765  */
4766 
4767 static void spapr_machine_2_10_class_options(MachineClass *mc)
4768 {
4769     spapr_machine_2_11_class_options(mc);
4770     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4771 }
4772 
4773 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4774 
4775 /*
4776  * pseries-2.9
4777  */
4778 
4779 static void spapr_machine_2_9_class_options(MachineClass *mc)
4780 {
4781     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4782     static GlobalProperty compat[] = {
4783         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4784     };
4785 
4786     spapr_machine_2_10_class_options(mc);
4787     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4788     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4789     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4790     smc->pre_2_10_has_unused_icps = true;
4791     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4792 }
4793 
4794 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4795 
4796 /*
4797  * pseries-2.8
4798  */
4799 
4800 static void spapr_machine_2_8_class_options(MachineClass *mc)
4801 {
4802     static GlobalProperty compat[] = {
4803         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4804     };
4805 
4806     spapr_machine_2_9_class_options(mc);
4807     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4808     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4809     mc->numa_mem_align_shift = 23;
4810 }
4811 
4812 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4813 
4814 /*
4815  * pseries-2.7
4816  */
4817 
4818 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4819                               uint64_t *buid, hwaddr *pio,
4820                               hwaddr *mmio32, hwaddr *mmio64,
4821                               unsigned n_dma, uint32_t *liobns,
4822                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4823 {
4824     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4825     const uint64_t base_buid = 0x800000020000000ULL;
4826     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4827     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4828     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4829     const uint32_t max_index = 255;
4830     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4831 
4832     uint64_t ram_top = MACHINE(spapr)->ram_size;
4833     hwaddr phb0_base, phb_base;
4834     int i;
4835 
4836     /* Do we have device memory? */
4837     if (MACHINE(spapr)->maxram_size > ram_top) {
4838         /* Can't just use maxram_size, because there may be an
4839          * alignment gap between normal and device memory regions
4840          */
4841         ram_top = MACHINE(spapr)->device_memory->base +
4842             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4843     }
4844 
4845     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4846 
4847     if (index > max_index) {
4848         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4849                    max_index);
4850         return;
4851     }
4852 
4853     *buid = base_buid + index;
4854     for (i = 0; i < n_dma; ++i) {
4855         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4856     }
4857 
4858     phb_base = phb0_base + index * phb_spacing;
4859     *pio = phb_base + pio_offset;
4860     *mmio32 = phb_base + mmio_offset;
4861     /*
4862      * We don't set the 64-bit MMIO window, relying on the PHB's
4863      * fallback behaviour of automatically splitting a large "32-bit"
4864      * window into contiguous 32-bit and 64-bit windows
4865      */
4866 
4867     *nv2gpa = 0;
4868     *nv2atsd = 0;
4869 }
4870 
4871 static void spapr_machine_2_7_class_options(MachineClass *mc)
4872 {
4873     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4874     static GlobalProperty compat[] = {
4875         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4876         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4877         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4878         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4879     };
4880 
4881     spapr_machine_2_8_class_options(mc);
4882     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4883     mc->default_machine_opts = "modern-hotplug-events=off";
4884     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4885     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4886     smc->phb_placement = phb_placement_2_7;
4887 }
4888 
4889 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4890 
4891 /*
4892  * pseries-2.6
4893  */
4894 
4895 static void spapr_machine_2_6_class_options(MachineClass *mc)
4896 {
4897     static GlobalProperty compat[] = {
4898         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4899     };
4900 
4901     spapr_machine_2_7_class_options(mc);
4902     mc->has_hotpluggable_cpus = false;
4903     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4904     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4905 }
4906 
4907 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4908 
4909 /*
4910  * pseries-2.5
4911  */
4912 
4913 static void spapr_machine_2_5_class_options(MachineClass *mc)
4914 {
4915     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4916     static GlobalProperty compat[] = {
4917         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4918     };
4919 
4920     spapr_machine_2_6_class_options(mc);
4921     smc->use_ohci_by_default = true;
4922     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4923     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4924 }
4925 
4926 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4927 
4928 /*
4929  * pseries-2.4
4930  */
4931 
4932 static void spapr_machine_2_4_class_options(MachineClass *mc)
4933 {
4934     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4935 
4936     spapr_machine_2_5_class_options(mc);
4937     smc->dr_lmb_enabled = false;
4938     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4939 }
4940 
4941 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4942 
4943 /*
4944  * pseries-2.3
4945  */
4946 
4947 static void spapr_machine_2_3_class_options(MachineClass *mc)
4948 {
4949     static GlobalProperty compat[] = {
4950         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4951     };
4952     spapr_machine_2_4_class_options(mc);
4953     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4954     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4955 }
4956 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4957 
4958 /*
4959  * pseries-2.2
4960  */
4961 
4962 static void spapr_machine_2_2_class_options(MachineClass *mc)
4963 {
4964     static GlobalProperty compat[] = {
4965         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4966     };
4967 
4968     spapr_machine_2_3_class_options(mc);
4969     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4970     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4971     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4972 }
4973 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4974 
4975 /*
4976  * pseries-2.1
4977  */
4978 
4979 static void spapr_machine_2_1_class_options(MachineClass *mc)
4980 {
4981     spapr_machine_2_2_class_options(mc);
4982     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4983 }
4984 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4985 
4986 static void spapr_machine_register_types(void)
4987 {
4988     type_register_static(&spapr_machine_info);
4989 }
4990 
4991 type_init(spapr_machine_register_types)
4992