xref: /qemu/hw/ppc/spapr.c (revision 328d8eb24db8ec415260ee7243adf2e3d7e81bad)
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 "mmu-hash64.h"
50 #include "mmu-book3s-v3.h"
51 #include "cpu-models.h"
52 #include "hw/core/cpu.h"
53 
54 #include "hw/boards.h"
55 #include "hw/ppc/ppc.h"
56 #include "hw/loader.h"
57 
58 #include "hw/ppc/fdt.h"
59 #include "hw/ppc/spapr.h"
60 #include "hw/ppc/spapr_vio.h"
61 #include "hw/qdev-properties.h"
62 #include "hw/pci-host/spapr.h"
63 #include "hw/pci/msi.h"
64 
65 #include "hw/pci/pci.h"
66 #include "hw/scsi/scsi.h"
67 #include "hw/virtio/virtio-scsi.h"
68 #include "hw/virtio/vhost-scsi-common.h"
69 
70 #include "exec/address-spaces.h"
71 #include "exec/ram_addr.h"
72 #include "hw/usb.h"
73 #include "qemu/config-file.h"
74 #include "qemu/error-report.h"
75 #include "trace.h"
76 #include "hw/nmi.h"
77 #include "hw/intc/intc.h"
78 
79 #include "qemu/cutils.h"
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 
84 #include "monitor/monitor.h"
85 
86 #include <libfdt.h>
87 
88 /* SLOF memory layout:
89  *
90  * SLOF raw image loaded at 0, copies its romfs right below the flat
91  * device-tree, then position SLOF itself 31M below that
92  *
93  * So we set FW_OVERHEAD to 40MB which should account for all of that
94  * and more
95  *
96  * We load our kernel at 4M, leaving space for SLOF initial image
97  */
98 #define FDT_MAX_SIZE            0x100000
99 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
100 #define FW_MAX_SIZE             0x400000
101 #define FW_FILE_NAME            "slof.bin"
102 #define FW_OVERHEAD             0x2800000
103 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
104 
105 #define MIN_RMA_SLOF            128UL
106 
107 #define PHANDLE_INTC            0x00001111
108 
109 /* These two functions implement the VCPU id numbering: one to compute them
110  * all and one to identify thread 0 of a VCORE. Any change to the first one
111  * is likely to have an impact on the second one, so let's keep them close.
112  */
113 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
114 {
115     MachineState *ms = MACHINE(spapr);
116     unsigned int smp_threads = ms->smp.threads;
117 
118     assert(spapr->vsmt);
119     return
120         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
121 }
122 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
123                                       PowerPCCPU *cpu)
124 {
125     assert(spapr->vsmt);
126     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 int spapr_max_server_number(SpaprMachineState *spapr)
164 {
165     MachineState *ms = MACHINE(spapr);
166 
167     assert(spapr->vsmt);
168     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
169 }
170 
171 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
172                                   int smt_threads)
173 {
174     int i, ret = 0;
175     uint32_t servers_prop[smt_threads];
176     uint32_t gservers_prop[smt_threads * 2];
177     int index = spapr_get_vcpu_id(cpu);
178 
179     if (cpu->compat_pvr) {
180         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
181         if (ret < 0) {
182             return ret;
183         }
184     }
185 
186     /* Build interrupt servers and gservers properties */
187     for (i = 0; i < smt_threads; i++) {
188         servers_prop[i] = cpu_to_be32(index + i);
189         /* Hack, direct the group queues back to cpu 0 */
190         gservers_prop[i*2] = cpu_to_be32(index + i);
191         gservers_prop[i*2 + 1] = 0;
192     }
193     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
194                       servers_prop, sizeof(servers_prop));
195     if (ret < 0) {
196         return ret;
197     }
198     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
199                       gservers_prop, sizeof(gservers_prop));
200 
201     return ret;
202 }
203 
204 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
205 {
206     int index = spapr_get_vcpu_id(cpu);
207     uint32_t associativity[] = {cpu_to_be32(0x5),
208                                 cpu_to_be32(0x0),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(cpu->node_id),
212                                 cpu_to_be32(index)};
213 
214     /* Advertise NUMA via ibm,associativity */
215     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
216                           sizeof(associativity));
217 }
218 
219 /* Populate the "ibm,pa-features" property */
220 static void spapr_populate_pa_features(SpaprMachineState *spapr,
221                                        PowerPCCPU *cpu,
222                                        void *fdt, int offset)
223 {
224     uint8_t pa_features_206[] = { 6, 0,
225         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
226     uint8_t pa_features_207[] = { 24, 0,
227         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
228         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
229         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
230         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
231     uint8_t pa_features_300[] = { 66, 0,
232         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
233         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
234         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
235         /* 6: DS207 */
236         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
237         /* 16: Vector */
238         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
239         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
241         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
242         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
243         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
244         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
245         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
246         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
247         /* 42: PM, 44: PC RA, 46: SC vec'd */
248         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
249         /* 48: SIMD, 50: QP BFP, 52: String */
250         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
251         /* 54: DecFP, 56: DecI, 58: SHA */
252         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
253         /* 60: NM atomic, 62: RNG */
254         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
255     };
256     uint8_t *pa_features = NULL;
257     size_t pa_size;
258 
259     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
260         pa_features = pa_features_206;
261         pa_size = sizeof(pa_features_206);
262     }
263     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
264         pa_features = pa_features_207;
265         pa_size = sizeof(pa_features_207);
266     }
267     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
268         pa_features = pa_features_300;
269         pa_size = sizeof(pa_features_300);
270     }
271     if (!pa_features) {
272         return;
273     }
274 
275     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
276         /*
277          * Note: we keep CI large pages off by default because a 64K capable
278          * guest provisioned with large pages might otherwise try to map a qemu
279          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
280          * even if that qemu runs on a 4k host.
281          * We dd this bit back here if we are confident this is not an issue
282          */
283         pa_features[3] |= 0x20;
284     }
285     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
286         pa_features[24] |= 0x80;    /* Transactional memory support */
287     }
288     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
289         /* Workaround for broken kernels that attempt (guest) radix
290          * mode when they can't handle it, if they see the radix bit set
291          * in pa-features. So hide it from them. */
292         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
293     }
294 
295     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
296 }
297 
298 static hwaddr spapr_node0_size(MachineState *machine)
299 {
300     if (machine->numa_state->num_nodes) {
301         int i;
302         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
303             if (machine->numa_state->nodes[i].node_mem) {
304                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
305                            machine->ram_size);
306             }
307         }
308     }
309     return machine->ram_size;
310 }
311 
312 static void add_str(GString *s, const gchar *s1)
313 {
314     g_string_append_len(s, s1, strlen(s1) + 1);
315 }
316 
317 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
318                                        hwaddr size)
319 {
320     uint32_t associativity[] = {
321         cpu_to_be32(0x4), /* length */
322         cpu_to_be32(0x0), cpu_to_be32(0x0),
323         cpu_to_be32(0x0), cpu_to_be32(nodeid)
324     };
325     char mem_name[32];
326     uint64_t mem_reg_property[2];
327     int off;
328 
329     mem_reg_property[0] = cpu_to_be64(start);
330     mem_reg_property[1] = cpu_to_be64(size);
331 
332     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
333     off = fdt_add_subnode(fdt, 0, mem_name);
334     _FDT(off);
335     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
336     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
337                       sizeof(mem_reg_property))));
338     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
339                       sizeof(associativity))));
340     return off;
341 }
342 
343 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
344 {
345     MachineState *machine = MACHINE(spapr);
346     hwaddr mem_start, node_size;
347     int i, nb_nodes = machine->numa_state->num_nodes;
348     NodeInfo *nodes = machine->numa_state->nodes;
349 
350     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
351         if (!nodes[i].node_mem) {
352             continue;
353         }
354         if (mem_start >= machine->ram_size) {
355             node_size = 0;
356         } else {
357             node_size = nodes[i].node_mem;
358             if (node_size > machine->ram_size - mem_start) {
359                 node_size = machine->ram_size - mem_start;
360             }
361         }
362         if (!mem_start) {
363             /* spapr_machine_init() checks for rma_size <= node0_size
364              * already */
365             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
366             mem_start += spapr->rma_size;
367             node_size -= spapr->rma_size;
368         }
369         for ( ; node_size; ) {
370             hwaddr sizetmp = pow2floor(node_size);
371 
372             /* mem_start != 0 here */
373             if (ctzl(mem_start) < ctzl(sizetmp)) {
374                 sizetmp = 1ULL << ctzl(mem_start);
375             }
376 
377             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
378             node_size -= sizetmp;
379             mem_start += sizetmp;
380         }
381     }
382 
383     return 0;
384 }
385 
386 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
387                                   SpaprMachineState *spapr)
388 {
389     MachineState *ms = MACHINE(spapr);
390     PowerPCCPU *cpu = POWERPC_CPU(cs);
391     CPUPPCState *env = &cpu->env;
392     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
393     int index = spapr_get_vcpu_id(cpu);
394     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
395                        0xffffffff, 0xffffffff};
396     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
397         : SPAPR_TIMEBASE_FREQ;
398     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
399     uint32_t page_sizes_prop[64];
400     size_t page_sizes_prop_size;
401     unsigned int smp_threads = ms->smp.threads;
402     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
403     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
404     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
405     SpaprDrc *drc;
406     int drc_index;
407     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
408     int i;
409 
410     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
411     if (drc) {
412         drc_index = spapr_drc_index(drc);
413         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
414     }
415 
416     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
417     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
418 
419     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
420     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
421                            env->dcache_line_size)));
422     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
423                            env->dcache_line_size)));
424     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
425                            env->icache_line_size)));
426     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
427                            env->icache_line_size)));
428 
429     if (pcc->l1_dcache_size) {
430         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
431                                pcc->l1_dcache_size)));
432     } else {
433         warn_report("Unknown L1 dcache size for cpu");
434     }
435     if (pcc->l1_icache_size) {
436         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
437                                pcc->l1_icache_size)));
438     } else {
439         warn_report("Unknown L1 icache size for cpu");
440     }
441 
442     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
443     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
444     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
445     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
446     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
447     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
448 
449     if (env->spr_cb[SPR_PURR].oea_read) {
450         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
451     }
452     if (env->spr_cb[SPR_SPURR].oea_read) {
453         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
454     }
455 
456     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
457         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
458                           segs, sizeof(segs))));
459     }
460 
461     /* Advertise VSX (vector extensions) if available
462      *   1               == VMX / Altivec available
463      *   2               == VSX available
464      *
465      * Only CPUs for which we create core types in spapr_cpu_core.c
466      * are possible, and all of those have VMX */
467     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
468         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
469     } else {
470         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
471     }
472 
473     /* Advertise DFP (Decimal Floating Point) if available
474      *   0 / no property == no DFP
475      *   1               == DFP available */
476     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
477         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
478     }
479 
480     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
481                                                       sizeof(page_sizes_prop));
482     if (page_sizes_prop_size) {
483         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
484                           page_sizes_prop, page_sizes_prop_size)));
485     }
486 
487     spapr_populate_pa_features(spapr, cpu, fdt, offset);
488 
489     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
490                            cs->cpu_index / vcpus_per_socket)));
491 
492     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
493                       pft_size_prop, sizeof(pft_size_prop))));
494 
495     if (ms->numa_state->num_nodes > 1) {
496         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
497     }
498 
499     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
500 
501     if (pcc->radix_page_info) {
502         for (i = 0; i < pcc->radix_page_info->count; i++) {
503             radix_AP_encodings[i] =
504                 cpu_to_be32(pcc->radix_page_info->entries[i]);
505         }
506         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
507                           radix_AP_encodings,
508                           pcc->radix_page_info->count *
509                           sizeof(radix_AP_encodings[0]))));
510     }
511 
512     /*
513      * We set this property to let the guest know that it can use the large
514      * decrementer and its width in bits.
515      */
516     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
517         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
518                               pcc->lrg_decr_bits)));
519 }
520 
521 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
522 {
523     CPUState **rev;
524     CPUState *cs;
525     int n_cpus;
526     int cpus_offset;
527     char *nodename;
528     int i;
529 
530     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
531     _FDT(cpus_offset);
532     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
533     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
534 
535     /*
536      * We walk the CPUs in reverse order to ensure that CPU DT nodes
537      * created by fdt_add_subnode() end up in the right order in FDT
538      * for the guest kernel the enumerate the CPUs correctly.
539      *
540      * The CPU list cannot be traversed in reverse order, so we need
541      * to do extra work.
542      */
543     n_cpus = 0;
544     rev = NULL;
545     CPU_FOREACH(cs) {
546         rev = g_renew(CPUState *, rev, n_cpus + 1);
547         rev[n_cpus++] = cs;
548     }
549 
550     for (i = n_cpus - 1; i >= 0; i--) {
551         CPUState *cs = rev[i];
552         PowerPCCPU *cpu = POWERPC_CPU(cs);
553         int index = spapr_get_vcpu_id(cpu);
554         DeviceClass *dc = DEVICE_GET_CLASS(cs);
555         int offset;
556 
557         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
558             continue;
559         }
560 
561         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
562         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
563         g_free(nodename);
564         _FDT(offset);
565         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
566     }
567 
568     g_free(rev);
569 }
570 
571 static int spapr_rng_populate_dt(void *fdt)
572 {
573     int node;
574     int ret;
575 
576     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
577     if (node <= 0) {
578         return -1;
579     }
580     ret = fdt_setprop_string(fdt, node, "device_type",
581                              "ibm,platform-facilities");
582     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
583     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
584 
585     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
586     if (node <= 0) {
587         return -1;
588     }
589     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
590 
591     return ret ? -1 : 0;
592 }
593 
594 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
595 {
596     MemoryDeviceInfoList *info;
597 
598     for (info = list; info; info = info->next) {
599         MemoryDeviceInfo *value = info->value;
600 
601         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
602             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
603 
604             if (addr >= pcdimm_info->addr &&
605                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
606                 return pcdimm_info->node;
607             }
608         }
609     }
610 
611     return -1;
612 }
613 
614 struct sPAPRDrconfCellV2 {
615      uint32_t seq_lmbs;
616      uint64_t base_addr;
617      uint32_t drc_index;
618      uint32_t aa_index;
619      uint32_t flags;
620 } QEMU_PACKED;
621 
622 typedef struct DrconfCellQueue {
623     struct sPAPRDrconfCellV2 cell;
624     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
625 } DrconfCellQueue;
626 
627 static DrconfCellQueue *
628 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
629                       uint32_t drc_index, uint32_t aa_index,
630                       uint32_t flags)
631 {
632     DrconfCellQueue *elem;
633 
634     elem = g_malloc0(sizeof(*elem));
635     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
636     elem->cell.base_addr = cpu_to_be64(base_addr);
637     elem->cell.drc_index = cpu_to_be32(drc_index);
638     elem->cell.aa_index = cpu_to_be32(aa_index);
639     elem->cell.flags = cpu_to_be32(flags);
640 
641     return elem;
642 }
643 
644 /* ibm,dynamic-memory-v2 */
645 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
646                                    int offset, MemoryDeviceInfoList *dimms)
647 {
648     MachineState *machine = MACHINE(spapr);
649     uint8_t *int_buf, *cur_index;
650     int ret;
651     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
652     uint64_t addr, cur_addr, size;
653     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
654     uint64_t mem_end = machine->device_memory->base +
655                        memory_region_size(&machine->device_memory->mr);
656     uint32_t node, buf_len, nr_entries = 0;
657     SpaprDrc *drc;
658     DrconfCellQueue *elem, *next;
659     MemoryDeviceInfoList *info;
660     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
661         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
662 
663     /* Entry to cover RAM and the gap area */
664     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
665                                  SPAPR_LMB_FLAGS_RESERVED |
666                                  SPAPR_LMB_FLAGS_DRC_INVALID);
667     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
668     nr_entries++;
669 
670     cur_addr = machine->device_memory->base;
671     for (info = dimms; info; info = info->next) {
672         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
673 
674         addr = di->addr;
675         size = di->size;
676         node = di->node;
677 
678         /* Entry for hot-pluggable area */
679         if (cur_addr < addr) {
680             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
681             g_assert(drc);
682             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
683                                          cur_addr, spapr_drc_index(drc), -1, 0);
684             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
685             nr_entries++;
686         }
687 
688         /* Entry for DIMM */
689         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
690         g_assert(drc);
691         elem = spapr_get_drconf_cell(size / lmb_size, addr,
692                                      spapr_drc_index(drc), node,
693                                      SPAPR_LMB_FLAGS_ASSIGNED);
694         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
695         nr_entries++;
696         cur_addr = addr + size;
697     }
698 
699     /* Entry for remaining hotpluggable area */
700     if (cur_addr < mem_end) {
701         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
702         g_assert(drc);
703         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
704                                      cur_addr, spapr_drc_index(drc), -1, 0);
705         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
706         nr_entries++;
707     }
708 
709     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
710     int_buf = cur_index = g_malloc0(buf_len);
711     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
712     cur_index += sizeof(nr_entries);
713 
714     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
715         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
716         cur_index += sizeof(elem->cell);
717         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
718         g_free(elem);
719     }
720 
721     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
722     g_free(int_buf);
723     if (ret < 0) {
724         return -1;
725     }
726     return 0;
727 }
728 
729 /* ibm,dynamic-memory */
730 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
731                                    int offset, MemoryDeviceInfoList *dimms)
732 {
733     MachineState *machine = MACHINE(spapr);
734     int i, ret;
735     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
736     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
737     uint32_t nr_lmbs = (machine->device_memory->base +
738                        memory_region_size(&machine->device_memory->mr)) /
739                        lmb_size;
740     uint32_t *int_buf, *cur_index, buf_len;
741 
742     /*
743      * Allocate enough buffer size to fit in ibm,dynamic-memory
744      */
745     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
746     cur_index = int_buf = g_malloc0(buf_len);
747     int_buf[0] = cpu_to_be32(nr_lmbs);
748     cur_index++;
749     for (i = 0; i < nr_lmbs; i++) {
750         uint64_t addr = i * lmb_size;
751         uint32_t *dynamic_memory = cur_index;
752 
753         if (i >= device_lmb_start) {
754             SpaprDrc *drc;
755 
756             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
757             g_assert(drc);
758 
759             dynamic_memory[0] = cpu_to_be32(addr >> 32);
760             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
761             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
762             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
763             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
764             if (memory_region_present(get_system_memory(), addr)) {
765                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
766             } else {
767                 dynamic_memory[5] = cpu_to_be32(0);
768             }
769         } else {
770             /*
771              * LMB information for RMA, boot time RAM and gap b/n RAM and
772              * device memory region -- all these are marked as reserved
773              * and as having no valid DRC.
774              */
775             dynamic_memory[0] = cpu_to_be32(addr >> 32);
776             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
777             dynamic_memory[2] = cpu_to_be32(0);
778             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
779             dynamic_memory[4] = cpu_to_be32(-1);
780             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
781                                             SPAPR_LMB_FLAGS_DRC_INVALID);
782         }
783 
784         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
785     }
786     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
787     g_free(int_buf);
788     if (ret < 0) {
789         return -1;
790     }
791     return 0;
792 }
793 
794 /*
795  * Adds ibm,dynamic-reconfiguration-memory node.
796  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
797  * of this device tree node.
798  */
799 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
800 {
801     MachineState *machine = MACHINE(spapr);
802     int nb_numa_nodes = machine->numa_state->num_nodes;
803     int ret, i, offset;
804     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
805     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
806     uint32_t *int_buf, *cur_index, buf_len;
807     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
808     MemoryDeviceInfoList *dimms = NULL;
809 
810     /*
811      * Don't create the node if there is no device memory
812      */
813     if (machine->ram_size == machine->maxram_size) {
814         return 0;
815     }
816 
817     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
818 
819     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
820                     sizeof(prop_lmb_size));
821     if (ret < 0) {
822         return ret;
823     }
824 
825     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
826     if (ret < 0) {
827         return ret;
828     }
829 
830     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
831     if (ret < 0) {
832         return ret;
833     }
834 
835     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
836     dimms = qmp_memory_device_list();
837     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
838         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
839     } else {
840         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
841     }
842     qapi_free_MemoryDeviceInfoList(dimms);
843 
844     if (ret < 0) {
845         return ret;
846     }
847 
848     /* ibm,associativity-lookup-arrays */
849     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
850     cur_index = int_buf = g_malloc0(buf_len);
851     int_buf[0] = cpu_to_be32(nr_nodes);
852     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
853     cur_index += 2;
854     for (i = 0; i < nr_nodes; i++) {
855         uint32_t associativity[] = {
856             cpu_to_be32(0x0),
857             cpu_to_be32(0x0),
858             cpu_to_be32(0x0),
859             cpu_to_be32(i)
860         };
861         memcpy(cur_index, associativity, sizeof(associativity));
862         cur_index += 4;
863     }
864     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
865             (cur_index - int_buf) * sizeof(uint32_t));
866     g_free(int_buf);
867 
868     return ret;
869 }
870 
871 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
872                                 SpaprOptionVector *ov5_updates)
873 {
874     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
875     int ret = 0, offset;
876 
877     /* Generate ibm,dynamic-reconfiguration-memory node if required */
878     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
879         g_assert(smc->dr_lmb_enabled);
880         ret = spapr_populate_drconf_memory(spapr, fdt);
881         if (ret) {
882             goto out;
883         }
884     }
885 
886     offset = fdt_path_offset(fdt, "/chosen");
887     if (offset < 0) {
888         offset = fdt_add_subnode(fdt, 0, "chosen");
889         if (offset < 0) {
890             return offset;
891         }
892     }
893     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
894                                  "ibm,architecture-vec-5");
895 
896 out:
897     return ret;
898 }
899 
900 static bool spapr_hotplugged_dev_before_cas(void)
901 {
902     Object *drc_container, *obj;
903     ObjectProperty *prop;
904     ObjectPropertyIterator iter;
905 
906     drc_container = container_get(object_get_root(), "/dr-connector");
907     object_property_iter_init(&iter, drc_container);
908     while ((prop = object_property_iter_next(&iter))) {
909         if (!strstart(prop->type, "link<", NULL)) {
910             continue;
911         }
912         obj = object_property_get_link(drc_container, prop->name, NULL);
913         if (spapr_drc_needed(obj)) {
914             return true;
915         }
916     }
917     return false;
918 }
919 
920 static void *spapr_build_fdt(SpaprMachineState *spapr);
921 
922 int spapr_h_cas_compose_response(SpaprMachineState *spapr,
923                                  target_ulong addr, target_ulong size,
924                                  SpaprOptionVector *ov5_updates)
925 {
926     void *fdt;
927     SpaprDeviceTreeUpdateHeader hdr = { .version_id = 1 };
928 
929     if (spapr_hotplugged_dev_before_cas()) {
930         return 1;
931     }
932 
933     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
934         error_report("SLOF provided an unexpected CAS buffer size "
935                      TARGET_FMT_lu " (min: %zu, max: %u)",
936                      size, sizeof(hdr), FW_MAX_SIZE);
937         exit(EXIT_FAILURE);
938     }
939 
940     size -= sizeof(hdr);
941 
942     fdt = spapr_build_fdt(spapr);
943     _FDT((fdt_pack(fdt)));
944 
945     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
946         g_free(fdt);
947         trace_spapr_cas_failed(size);
948         return -1;
949     }
950 
951     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
952     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
953     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
954 
955     g_free(spapr->fdt_blob);
956     spapr->fdt_size = fdt_totalsize(fdt);
957     spapr->fdt_initial_size = spapr->fdt_size;
958     spapr->fdt_blob = fdt;
959 
960     return 0;
961 }
962 
963 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
964 {
965     MachineState *ms = MACHINE(spapr);
966     int rtas;
967     GString *hypertas = g_string_sized_new(256);
968     GString *qemu_hypertas = g_string_sized_new(256);
969     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
970     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
971         memory_region_size(&MACHINE(spapr)->device_memory->mr);
972     uint32_t lrdr_capacity[] = {
973         cpu_to_be32(max_device_addr >> 32),
974         cpu_to_be32(max_device_addr & 0xffffffff),
975         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
976         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
977     };
978     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
979     uint32_t maxdomains[] = {
980         cpu_to_be32(4),
981         maxdomain,
982         maxdomain,
983         maxdomain,
984         cpu_to_be32(spapr->gpu_numa_id),
985     };
986 
987     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
988 
989     /* hypertas */
990     add_str(hypertas, "hcall-pft");
991     add_str(hypertas, "hcall-term");
992     add_str(hypertas, "hcall-dabr");
993     add_str(hypertas, "hcall-interrupt");
994     add_str(hypertas, "hcall-tce");
995     add_str(hypertas, "hcall-vio");
996     add_str(hypertas, "hcall-splpar");
997     add_str(hypertas, "hcall-join");
998     add_str(hypertas, "hcall-bulk");
999     add_str(hypertas, "hcall-set-mode");
1000     add_str(hypertas, "hcall-sprg0");
1001     add_str(hypertas, "hcall-copy");
1002     add_str(hypertas, "hcall-debug");
1003     add_str(hypertas, "hcall-vphn");
1004     add_str(qemu_hypertas, "hcall-memop1");
1005 
1006     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1007         add_str(hypertas, "hcall-multi-tce");
1008     }
1009 
1010     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1011         add_str(hypertas, "hcall-hpt-resize");
1012     }
1013 
1014     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1015                      hypertas->str, hypertas->len));
1016     g_string_free(hypertas, TRUE);
1017     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1018                      qemu_hypertas->str, qemu_hypertas->len));
1019     g_string_free(qemu_hypertas, TRUE);
1020 
1021     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1022                      refpoints, sizeof(refpoints)));
1023 
1024     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1025                      maxdomains, sizeof(maxdomains)));
1026 
1027     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1028                           RTAS_ERROR_LOG_MAX));
1029     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1030                           RTAS_EVENT_SCAN_RATE));
1031 
1032     g_assert(msi_nonbroken);
1033     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1034 
1035     /*
1036      * According to PAPR, rtas ibm,os-term does not guarantee a return
1037      * back to the guest cpu.
1038      *
1039      * While an additional ibm,extended-os-term property indicates
1040      * that rtas call return will always occur. Set this property.
1041      */
1042     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1043 
1044     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1045                      lrdr_capacity, sizeof(lrdr_capacity)));
1046 
1047     spapr_dt_rtas_tokens(fdt, rtas);
1048 }
1049 
1050 /*
1051  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1052  * and the XIVE features that the guest may request and thus the valid
1053  * values for bytes 23..26 of option vector 5:
1054  */
1055 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1056                                           int chosen)
1057 {
1058     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1059 
1060     char val[2 * 4] = {
1061         23, 0x00, /* XICS / XIVE mode */
1062         24, 0x00, /* Hash/Radix, filled in below. */
1063         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1064         26, 0x40, /* Radix options: GTSE == yes. */
1065     };
1066 
1067     if (spapr->irq->xics && spapr->irq->xive) {
1068         val[1] = SPAPR_OV5_XIVE_BOTH;
1069     } else if (spapr->irq->xive) {
1070         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1071     } else {
1072         assert(spapr->irq->xics);
1073         val[1] = SPAPR_OV5_XIVE_LEGACY;
1074     }
1075 
1076     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1077                           first_ppc_cpu->compat_pvr)) {
1078         /*
1079          * If we're in a pre POWER9 compat mode then the guest should
1080          * do hash and use the legacy interrupt mode
1081          */
1082         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1083         val[3] = 0x00; /* Hash */
1084     } else if (kvm_enabled()) {
1085         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1086             val[3] = 0x80; /* OV5_MMU_BOTH */
1087         } else if (kvmppc_has_cap_mmu_radix()) {
1088             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1089         } else {
1090             val[3] = 0x00; /* Hash */
1091         }
1092     } else {
1093         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1094         val[3] = 0xC0;
1095     }
1096     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1097                      val, sizeof(val)));
1098 }
1099 
1100 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1101 {
1102     MachineState *machine = MACHINE(spapr);
1103     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1104     int chosen;
1105     const char *boot_device = machine->boot_order;
1106     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1107     size_t cb = 0;
1108     char *bootlist = get_boot_devices_list(&cb);
1109 
1110     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1111 
1112     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1113         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1114                                 machine->kernel_cmdline));
1115     }
1116     if (spapr->initrd_size) {
1117         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1118                               spapr->initrd_base));
1119         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1120                               spapr->initrd_base + spapr->initrd_size));
1121     }
1122 
1123     if (spapr->kernel_size) {
1124         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1125                               cpu_to_be64(spapr->kernel_size) };
1126 
1127         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1128                          &kprop, sizeof(kprop)));
1129         if (spapr->kernel_le) {
1130             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1131         }
1132     }
1133     if (boot_menu) {
1134         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1135     }
1136     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1137     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1138     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1139 
1140     if (cb && bootlist) {
1141         int i;
1142 
1143         for (i = 0; i < cb; i++) {
1144             if (bootlist[i] == '\n') {
1145                 bootlist[i] = ' ';
1146             }
1147         }
1148         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1149     }
1150 
1151     if (boot_device && strlen(boot_device)) {
1152         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1153     }
1154 
1155     if (!spapr->has_graphics && stdout_path) {
1156         /*
1157          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1158          * kernel. New platforms should only use the "stdout-path" property. Set
1159          * the new property and continue using older property to remain
1160          * compatible with the existing firmware.
1161          */
1162         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1163         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1164     }
1165 
1166     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1167     if (smc->linux_pci_probe) {
1168         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1169     }
1170 
1171     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1172 
1173     g_free(stdout_path);
1174     g_free(bootlist);
1175 }
1176 
1177 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1178 {
1179     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1180      * KVM to work under pHyp with some guest co-operation */
1181     int hypervisor;
1182     uint8_t hypercall[16];
1183 
1184     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1185     /* indicate KVM hypercall interface */
1186     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1187     if (kvmppc_has_cap_fixup_hcalls()) {
1188         /*
1189          * Older KVM versions with older guest kernels were broken
1190          * with the magic page, don't allow the guest to map it.
1191          */
1192         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1193                                   sizeof(hypercall))) {
1194             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1195                              hypercall, sizeof(hypercall)));
1196         }
1197     }
1198 }
1199 
1200 static void *spapr_build_fdt(SpaprMachineState *spapr)
1201 {
1202     MachineState *machine = MACHINE(spapr);
1203     MachineClass *mc = MACHINE_GET_CLASS(machine);
1204     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1205     int ret;
1206     void *fdt;
1207     SpaprPhbState *phb;
1208     char *buf;
1209 
1210     fdt = g_malloc0(FDT_MAX_SIZE);
1211     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1212 
1213     /* Root node */
1214     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1215     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1216     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1217 
1218     /* Guest UUID & Name*/
1219     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1220     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1221     if (qemu_uuid_set) {
1222         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1223     }
1224     g_free(buf);
1225 
1226     if (qemu_get_vm_name()) {
1227         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1228                                 qemu_get_vm_name()));
1229     }
1230 
1231     /* Host Model & Serial Number */
1232     if (spapr->host_model) {
1233         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1234     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1235         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1236         g_free(buf);
1237     }
1238 
1239     if (spapr->host_serial) {
1240         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1241     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1242         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1243         g_free(buf);
1244     }
1245 
1246     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1247     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1248 
1249     /* /interrupt controller */
1250     spapr->irq->dt_populate(spapr, spapr_max_server_number(spapr), fdt,
1251                           PHANDLE_INTC);
1252 
1253     ret = spapr_populate_memory(spapr, fdt);
1254     if (ret < 0) {
1255         error_report("couldn't setup memory nodes in fdt");
1256         exit(1);
1257     }
1258 
1259     /* /vdevice */
1260     spapr_dt_vdevice(spapr->vio_bus, fdt);
1261 
1262     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1263         ret = spapr_rng_populate_dt(fdt);
1264         if (ret < 0) {
1265             error_report("could not set up rng device in the fdt");
1266             exit(1);
1267         }
1268     }
1269 
1270     QLIST_FOREACH(phb, &spapr->phbs, list) {
1271         ret = spapr_dt_phb(phb, PHANDLE_INTC, fdt, spapr->irq->nr_msis, NULL);
1272         if (ret < 0) {
1273             error_report("couldn't setup PCI devices in fdt");
1274             exit(1);
1275         }
1276     }
1277 
1278     /* cpus */
1279     spapr_populate_cpus_dt_node(fdt, spapr);
1280 
1281     if (smc->dr_lmb_enabled) {
1282         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1283     }
1284 
1285     if (mc->has_hotpluggable_cpus) {
1286         int offset = fdt_path_offset(fdt, "/cpus");
1287         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1288         if (ret < 0) {
1289             error_report("Couldn't set up CPU DR device tree properties");
1290             exit(1);
1291         }
1292     }
1293 
1294     /* /event-sources */
1295     spapr_dt_events(spapr, fdt);
1296 
1297     /* /rtas */
1298     spapr_dt_rtas(spapr, fdt);
1299 
1300     /* /chosen */
1301     spapr_dt_chosen(spapr, fdt);
1302 
1303     /* /hypervisor */
1304     if (kvm_enabled()) {
1305         spapr_dt_hypervisor(spapr, fdt);
1306     }
1307 
1308     /* Build memory reserve map */
1309     if (spapr->kernel_size) {
1310         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1311     }
1312     if (spapr->initrd_size) {
1313         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1314     }
1315 
1316     /* ibm,client-architecture-support updates */
1317     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1318     if (ret < 0) {
1319         error_report("couldn't setup CAS properties fdt");
1320         exit(1);
1321     }
1322 
1323     if (smc->dr_phb_enabled) {
1324         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1325         if (ret < 0) {
1326             error_report("Couldn't set up PHB DR device tree properties");
1327             exit(1);
1328         }
1329     }
1330 
1331     return fdt;
1332 }
1333 
1334 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1335 {
1336     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1337 }
1338 
1339 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1340                                     PowerPCCPU *cpu)
1341 {
1342     CPUPPCState *env = &cpu->env;
1343 
1344     /* The TCG path should also be holding the BQL at this point */
1345     g_assert(qemu_mutex_iothread_locked());
1346 
1347     if (msr_pr) {
1348         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1349         env->gpr[3] = H_PRIVILEGE;
1350     } else {
1351         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1352     }
1353 }
1354 
1355 struct LPCRSyncState {
1356     target_ulong value;
1357     target_ulong mask;
1358 };
1359 
1360 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1361 {
1362     struct LPCRSyncState *s = arg.host_ptr;
1363     PowerPCCPU *cpu = POWERPC_CPU(cs);
1364     CPUPPCState *env = &cpu->env;
1365     target_ulong lpcr;
1366 
1367     cpu_synchronize_state(cs);
1368     lpcr = env->spr[SPR_LPCR];
1369     lpcr &= ~s->mask;
1370     lpcr |= s->value;
1371     ppc_store_lpcr(cpu, lpcr);
1372 }
1373 
1374 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1375 {
1376     CPUState *cs;
1377     struct LPCRSyncState s = {
1378         .value = value,
1379         .mask = mask
1380     };
1381     CPU_FOREACH(cs) {
1382         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1383     }
1384 }
1385 
1386 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1387 {
1388     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1389 
1390     /* Copy PATE1:GR into PATE0:HR */
1391     entry->dw0 = spapr->patb_entry & PATE0_HR;
1392     entry->dw1 = spapr->patb_entry;
1393 }
1394 
1395 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1396 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1397 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1398 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1399 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1400 
1401 /*
1402  * Get the fd to access the kernel htab, re-opening it if necessary
1403  */
1404 static int get_htab_fd(SpaprMachineState *spapr)
1405 {
1406     Error *local_err = NULL;
1407 
1408     if (spapr->htab_fd >= 0) {
1409         return spapr->htab_fd;
1410     }
1411 
1412     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1413     if (spapr->htab_fd < 0) {
1414         error_report_err(local_err);
1415     }
1416 
1417     return spapr->htab_fd;
1418 }
1419 
1420 void close_htab_fd(SpaprMachineState *spapr)
1421 {
1422     if (spapr->htab_fd >= 0) {
1423         close(spapr->htab_fd);
1424     }
1425     spapr->htab_fd = -1;
1426 }
1427 
1428 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1429 {
1430     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1431 
1432     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1433 }
1434 
1435 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1436 {
1437     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1438 
1439     assert(kvm_enabled());
1440 
1441     if (!spapr->htab) {
1442         return 0;
1443     }
1444 
1445     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1446 }
1447 
1448 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1449                                                 hwaddr ptex, int n)
1450 {
1451     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1452     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1453 
1454     if (!spapr->htab) {
1455         /*
1456          * HTAB is controlled by KVM. Fetch into temporary buffer
1457          */
1458         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1459         kvmppc_read_hptes(hptes, ptex, n);
1460         return hptes;
1461     }
1462 
1463     /*
1464      * HTAB is controlled by QEMU. Just point to the internally
1465      * accessible PTEG.
1466      */
1467     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1468 }
1469 
1470 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1471                               const ppc_hash_pte64_t *hptes,
1472                               hwaddr ptex, int n)
1473 {
1474     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1475 
1476     if (!spapr->htab) {
1477         g_free((void *)hptes);
1478     }
1479 
1480     /* Nothing to do for qemu managed HPT */
1481 }
1482 
1483 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1484                       uint64_t pte0, uint64_t pte1)
1485 {
1486     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1487     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1488 
1489     if (!spapr->htab) {
1490         kvmppc_write_hpte(ptex, pte0, pte1);
1491     } else {
1492         if (pte0 & HPTE64_V_VALID) {
1493             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1494             /*
1495              * When setting valid, we write PTE1 first. This ensures
1496              * proper synchronization with the reading code in
1497              * ppc_hash64_pteg_search()
1498              */
1499             smp_wmb();
1500             stq_p(spapr->htab + offset, pte0);
1501         } else {
1502             stq_p(spapr->htab + offset, pte0);
1503             /*
1504              * When clearing it we set PTE0 first. This ensures proper
1505              * synchronization with the reading code in
1506              * ppc_hash64_pteg_search()
1507              */
1508             smp_wmb();
1509             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1510         }
1511     }
1512 }
1513 
1514 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1515                              uint64_t pte1)
1516 {
1517     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1518     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1519 
1520     if (!spapr->htab) {
1521         /* There should always be a hash table when this is called */
1522         error_report("spapr_hpte_set_c called with no hash table !");
1523         return;
1524     }
1525 
1526     /* The HW performs a non-atomic byte update */
1527     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1528 }
1529 
1530 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1531                              uint64_t pte1)
1532 {
1533     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1534     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1535 
1536     if (!spapr->htab) {
1537         /* There should always be a hash table when this is called */
1538         error_report("spapr_hpte_set_r called with no hash table !");
1539         return;
1540     }
1541 
1542     /* The HW performs a non-atomic byte update */
1543     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1544 }
1545 
1546 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1547 {
1548     int shift;
1549 
1550     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1551      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1552      * that's much more than is needed for Linux guests */
1553     shift = ctz64(pow2ceil(ramsize)) - 7;
1554     shift = MAX(shift, 18); /* Minimum architected size */
1555     shift = MIN(shift, 46); /* Maximum architected size */
1556     return shift;
1557 }
1558 
1559 void spapr_free_hpt(SpaprMachineState *spapr)
1560 {
1561     g_free(spapr->htab);
1562     spapr->htab = NULL;
1563     spapr->htab_shift = 0;
1564     close_htab_fd(spapr);
1565 }
1566 
1567 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1568                           Error **errp)
1569 {
1570     long rc;
1571 
1572     /* Clean up any HPT info from a previous boot */
1573     spapr_free_hpt(spapr);
1574 
1575     rc = kvmppc_reset_htab(shift);
1576     if (rc < 0) {
1577         /* kernel-side HPT needed, but couldn't allocate one */
1578         error_setg_errno(errp, errno,
1579                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1580                          shift);
1581         /* This is almost certainly fatal, but if the caller really
1582          * wants to carry on with shift == 0, it's welcome to try */
1583     } else if (rc > 0) {
1584         /* kernel-side HPT allocated */
1585         if (rc != shift) {
1586             error_setg(errp,
1587                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1588                        shift, rc);
1589         }
1590 
1591         spapr->htab_shift = shift;
1592         spapr->htab = NULL;
1593     } else {
1594         /* kernel-side HPT not needed, allocate in userspace instead */
1595         size_t size = 1ULL << shift;
1596         int i;
1597 
1598         spapr->htab = qemu_memalign(size, size);
1599         if (!spapr->htab) {
1600             error_setg_errno(errp, errno,
1601                              "Could not allocate HPT of order %d", shift);
1602             return;
1603         }
1604 
1605         memset(spapr->htab, 0, size);
1606         spapr->htab_shift = shift;
1607 
1608         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1609             DIRTY_HPTE(HPTE(spapr->htab, i));
1610         }
1611     }
1612     /* We're setting up a hash table, so that means we're not radix */
1613     spapr->patb_entry = 0;
1614     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1615 }
1616 
1617 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1618 {
1619     int hpt_shift;
1620 
1621     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1622         || (spapr->cas_reboot
1623             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1624         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1625     } else {
1626         uint64_t current_ram_size;
1627 
1628         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1629         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1630     }
1631     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1632 
1633     if (spapr->vrma_adjust) {
1634         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1635                                           spapr->htab_shift);
1636     }
1637 }
1638 
1639 static int spapr_reset_drcs(Object *child, void *opaque)
1640 {
1641     SpaprDrc *drc =
1642         (SpaprDrc *) object_dynamic_cast(child,
1643                                                  TYPE_SPAPR_DR_CONNECTOR);
1644 
1645     if (drc) {
1646         spapr_drc_reset(drc);
1647     }
1648 
1649     return 0;
1650 }
1651 
1652 static void spapr_machine_reset(MachineState *machine)
1653 {
1654     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1655     PowerPCCPU *first_ppc_cpu;
1656     hwaddr fdt_addr;
1657     void *fdt;
1658     int rc;
1659 
1660     spapr_caps_apply(spapr);
1661 
1662     first_ppc_cpu = POWERPC_CPU(first_cpu);
1663     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1664         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1665                               spapr->max_compat_pvr)) {
1666         /*
1667          * If using KVM with radix mode available, VCPUs can be started
1668          * without a HPT because KVM will start them in radix mode.
1669          * Set the GR bit in PATE so that we know there is no HPT.
1670          */
1671         spapr->patb_entry = PATE1_GR;
1672         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1673     } else {
1674         spapr_setup_hpt_and_vrma(spapr);
1675     }
1676 
1677     qemu_devices_reset();
1678 
1679     /*
1680      * If this reset wasn't generated by CAS, we should reset our
1681      * negotiated options and start from scratch
1682      */
1683     if (!spapr->cas_reboot) {
1684         spapr_ovec_cleanup(spapr->ov5_cas);
1685         spapr->ov5_cas = spapr_ovec_new();
1686 
1687         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1688     }
1689 
1690     /*
1691      * This is fixing some of the default configuration of the XIVE
1692      * devices. To be called after the reset of the machine devices.
1693      */
1694     spapr_irq_reset(spapr, &error_fatal);
1695 
1696     /*
1697      * There is no CAS under qtest. Simulate one to please the code that
1698      * depends on spapr->ov5_cas. This is especially needed to test device
1699      * unplug, so we do that before resetting the DRCs.
1700      */
1701     if (qtest_enabled()) {
1702         spapr_ovec_cleanup(spapr->ov5_cas);
1703         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1704     }
1705 
1706     /* DRC reset may cause a device to be unplugged. This will cause troubles
1707      * if this device is used by another device (eg, a running vhost backend
1708      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1709      * situations, we reset DRCs after all devices have been reset.
1710      */
1711     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1712 
1713     spapr_clear_pending_events(spapr);
1714 
1715     /*
1716      * We place the device tree and RTAS just below either the top of the RMA,
1717      * or just below 2GB, whichever is lower, so that it can be
1718      * processed with 32-bit real mode code if necessary
1719      */
1720     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1721 
1722     fdt = spapr_build_fdt(spapr);
1723 
1724     rc = fdt_pack(fdt);
1725 
1726     /* Should only fail if we've built a corrupted tree */
1727     assert(rc == 0);
1728 
1729     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1730         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1731                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1732         exit(1);
1733     }
1734 
1735     /* Load the fdt */
1736     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1737     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1738     g_free(spapr->fdt_blob);
1739     spapr->fdt_size = fdt_totalsize(fdt);
1740     spapr->fdt_initial_size = spapr->fdt_size;
1741     spapr->fdt_blob = fdt;
1742 
1743     /* Set up the entry state */
1744     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1745     first_ppc_cpu->env.gpr[5] = 0;
1746 
1747     spapr->cas_reboot = false;
1748 }
1749 
1750 static void spapr_create_nvram(SpaprMachineState *spapr)
1751 {
1752     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1753     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1754 
1755     if (dinfo) {
1756         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1757                             &error_fatal);
1758     }
1759 
1760     qdev_init_nofail(dev);
1761 
1762     spapr->nvram = (struct SpaprNvram *)dev;
1763 }
1764 
1765 static void spapr_rtc_create(SpaprMachineState *spapr)
1766 {
1767     object_initialize_child(OBJECT(spapr), "rtc",
1768                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1769                             &error_fatal, NULL);
1770     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1771                               &error_fatal);
1772     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1773                               "date", &error_fatal);
1774 }
1775 
1776 /* Returns whether we want to use VGA or not */
1777 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1778 {
1779     switch (vga_interface_type) {
1780     case VGA_NONE:
1781         return false;
1782     case VGA_DEVICE:
1783         return true;
1784     case VGA_STD:
1785     case VGA_VIRTIO:
1786     case VGA_CIRRUS:
1787         return pci_vga_init(pci_bus) != NULL;
1788     default:
1789         error_setg(errp,
1790                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1791         return false;
1792     }
1793 }
1794 
1795 static int spapr_pre_load(void *opaque)
1796 {
1797     int rc;
1798 
1799     rc = spapr_caps_pre_load(opaque);
1800     if (rc) {
1801         return rc;
1802     }
1803 
1804     return 0;
1805 }
1806 
1807 static int spapr_post_load(void *opaque, int version_id)
1808 {
1809     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1810     int err = 0;
1811 
1812     err = spapr_caps_post_migration(spapr);
1813     if (err) {
1814         return err;
1815     }
1816 
1817     /*
1818      * In earlier versions, there was no separate qdev for the PAPR
1819      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1820      * So when migrating from those versions, poke the incoming offset
1821      * value into the RTC device
1822      */
1823     if (version_id < 3) {
1824         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1825         if (err) {
1826             return err;
1827         }
1828     }
1829 
1830     if (kvm_enabled() && spapr->patb_entry) {
1831         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1832         bool radix = !!(spapr->patb_entry & PATE1_GR);
1833         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1834 
1835         /*
1836          * Update LPCR:HR and UPRT as they may not be set properly in
1837          * the stream
1838          */
1839         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1840                             LPCR_HR | LPCR_UPRT);
1841 
1842         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1843         if (err) {
1844             error_report("Process table config unsupported by the host");
1845             return -EINVAL;
1846         }
1847     }
1848 
1849     err = spapr_irq_post_load(spapr, version_id);
1850     if (err) {
1851         return err;
1852     }
1853 
1854     return err;
1855 }
1856 
1857 static int spapr_pre_save(void *opaque)
1858 {
1859     int rc;
1860 
1861     rc = spapr_caps_pre_save(opaque);
1862     if (rc) {
1863         return rc;
1864     }
1865 
1866     return 0;
1867 }
1868 
1869 static bool version_before_3(void *opaque, int version_id)
1870 {
1871     return version_id < 3;
1872 }
1873 
1874 static bool spapr_pending_events_needed(void *opaque)
1875 {
1876     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1877     return !QTAILQ_EMPTY(&spapr->pending_events);
1878 }
1879 
1880 static const VMStateDescription vmstate_spapr_event_entry = {
1881     .name = "spapr_event_log_entry",
1882     .version_id = 1,
1883     .minimum_version_id = 1,
1884     .fields = (VMStateField[]) {
1885         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1886         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1887         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1888                                      NULL, extended_length),
1889         VMSTATE_END_OF_LIST()
1890     },
1891 };
1892 
1893 static const VMStateDescription vmstate_spapr_pending_events = {
1894     .name = "spapr_pending_events",
1895     .version_id = 1,
1896     .minimum_version_id = 1,
1897     .needed = spapr_pending_events_needed,
1898     .fields = (VMStateField[]) {
1899         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1900                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1901         VMSTATE_END_OF_LIST()
1902     },
1903 };
1904 
1905 static bool spapr_ov5_cas_needed(void *opaque)
1906 {
1907     SpaprMachineState *spapr = opaque;
1908     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1909     SpaprOptionVector *ov5_legacy = spapr_ovec_new();
1910     SpaprOptionVector *ov5_removed = spapr_ovec_new();
1911     bool cas_needed;
1912 
1913     /* Prior to the introduction of SpaprOptionVector, we had two option
1914      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1915      * Both of these options encode machine topology into the device-tree
1916      * in such a way that the now-booted OS should still be able to interact
1917      * appropriately with QEMU regardless of what options were actually
1918      * negotiatied on the source side.
1919      *
1920      * As such, we can avoid migrating the CAS-negotiated options if these
1921      * are the only options available on the current machine/platform.
1922      * Since these are the only options available for pseries-2.7 and
1923      * earlier, this allows us to maintain old->new/new->old migration
1924      * compatibility.
1925      *
1926      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1927      * via default pseries-2.8 machines and explicit command-line parameters.
1928      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1929      * of the actual CAS-negotiated values to continue working properly. For
1930      * example, availability of memory unplug depends on knowing whether
1931      * OV5_HP_EVT was negotiated via CAS.
1932      *
1933      * Thus, for any cases where the set of available CAS-negotiatable
1934      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1935      * include the CAS-negotiated options in the migration stream, unless
1936      * if they affect boot time behaviour only.
1937      */
1938     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1939     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1940     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1941 
1942     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1943      * the mask itself since in the future it's possible "legacy" bits may be
1944      * removed via machine options, which could generate a false positive
1945      * that breaks migration.
1946      */
1947     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1948     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1949 
1950     spapr_ovec_cleanup(ov5_mask);
1951     spapr_ovec_cleanup(ov5_legacy);
1952     spapr_ovec_cleanup(ov5_removed);
1953 
1954     return cas_needed;
1955 }
1956 
1957 static const VMStateDescription vmstate_spapr_ov5_cas = {
1958     .name = "spapr_option_vector_ov5_cas",
1959     .version_id = 1,
1960     .minimum_version_id = 1,
1961     .needed = spapr_ov5_cas_needed,
1962     .fields = (VMStateField[]) {
1963         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1964                                  vmstate_spapr_ovec, SpaprOptionVector),
1965         VMSTATE_END_OF_LIST()
1966     },
1967 };
1968 
1969 static bool spapr_patb_entry_needed(void *opaque)
1970 {
1971     SpaprMachineState *spapr = opaque;
1972 
1973     return !!spapr->patb_entry;
1974 }
1975 
1976 static const VMStateDescription vmstate_spapr_patb_entry = {
1977     .name = "spapr_patb_entry",
1978     .version_id = 1,
1979     .minimum_version_id = 1,
1980     .needed = spapr_patb_entry_needed,
1981     .fields = (VMStateField[]) {
1982         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1983         VMSTATE_END_OF_LIST()
1984     },
1985 };
1986 
1987 static bool spapr_irq_map_needed(void *opaque)
1988 {
1989     SpaprMachineState *spapr = opaque;
1990 
1991     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1992 }
1993 
1994 static const VMStateDescription vmstate_spapr_irq_map = {
1995     .name = "spapr_irq_map",
1996     .version_id = 1,
1997     .minimum_version_id = 1,
1998     .needed = spapr_irq_map_needed,
1999     .fields = (VMStateField[]) {
2000         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2001         VMSTATE_END_OF_LIST()
2002     },
2003 };
2004 
2005 static bool spapr_dtb_needed(void *opaque)
2006 {
2007     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2008 
2009     return smc->update_dt_enabled;
2010 }
2011 
2012 static int spapr_dtb_pre_load(void *opaque)
2013 {
2014     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2015 
2016     g_free(spapr->fdt_blob);
2017     spapr->fdt_blob = NULL;
2018     spapr->fdt_size = 0;
2019 
2020     return 0;
2021 }
2022 
2023 static const VMStateDescription vmstate_spapr_dtb = {
2024     .name = "spapr_dtb",
2025     .version_id = 1,
2026     .minimum_version_id = 1,
2027     .needed = spapr_dtb_needed,
2028     .pre_load = spapr_dtb_pre_load,
2029     .fields = (VMStateField[]) {
2030         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2031         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2032         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2033                                      fdt_size),
2034         VMSTATE_END_OF_LIST()
2035     },
2036 };
2037 
2038 static const VMStateDescription vmstate_spapr = {
2039     .name = "spapr",
2040     .version_id = 3,
2041     .minimum_version_id = 1,
2042     .pre_load = spapr_pre_load,
2043     .post_load = spapr_post_load,
2044     .pre_save = spapr_pre_save,
2045     .fields = (VMStateField[]) {
2046         /* used to be @next_irq */
2047         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2048 
2049         /* RTC offset */
2050         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2051 
2052         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2053         VMSTATE_END_OF_LIST()
2054     },
2055     .subsections = (const VMStateDescription*[]) {
2056         &vmstate_spapr_ov5_cas,
2057         &vmstate_spapr_patb_entry,
2058         &vmstate_spapr_pending_events,
2059         &vmstate_spapr_cap_htm,
2060         &vmstate_spapr_cap_vsx,
2061         &vmstate_spapr_cap_dfp,
2062         &vmstate_spapr_cap_cfpc,
2063         &vmstate_spapr_cap_sbbc,
2064         &vmstate_spapr_cap_ibs,
2065         &vmstate_spapr_cap_hpt_maxpagesize,
2066         &vmstate_spapr_irq_map,
2067         &vmstate_spapr_cap_nested_kvm_hv,
2068         &vmstate_spapr_dtb,
2069         &vmstate_spapr_cap_large_decr,
2070         &vmstate_spapr_cap_ccf_assist,
2071         NULL
2072     }
2073 };
2074 
2075 static int htab_save_setup(QEMUFile *f, void *opaque)
2076 {
2077     SpaprMachineState *spapr = opaque;
2078 
2079     /* "Iteration" header */
2080     if (!spapr->htab_shift) {
2081         qemu_put_be32(f, -1);
2082     } else {
2083         qemu_put_be32(f, spapr->htab_shift);
2084     }
2085 
2086     if (spapr->htab) {
2087         spapr->htab_save_index = 0;
2088         spapr->htab_first_pass = true;
2089     } else {
2090         if (spapr->htab_shift) {
2091             assert(kvm_enabled());
2092         }
2093     }
2094 
2095 
2096     return 0;
2097 }
2098 
2099 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2100                             int chunkstart, int n_valid, int n_invalid)
2101 {
2102     qemu_put_be32(f, chunkstart);
2103     qemu_put_be16(f, n_valid);
2104     qemu_put_be16(f, n_invalid);
2105     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2106                     HASH_PTE_SIZE_64 * n_valid);
2107 }
2108 
2109 static void htab_save_end_marker(QEMUFile *f)
2110 {
2111     qemu_put_be32(f, 0);
2112     qemu_put_be16(f, 0);
2113     qemu_put_be16(f, 0);
2114 }
2115 
2116 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2117                                  int64_t max_ns)
2118 {
2119     bool has_timeout = max_ns != -1;
2120     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2121     int index = spapr->htab_save_index;
2122     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2123 
2124     assert(spapr->htab_first_pass);
2125 
2126     do {
2127         int chunkstart;
2128 
2129         /* Consume invalid HPTEs */
2130         while ((index < htabslots)
2131                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2132             CLEAN_HPTE(HPTE(spapr->htab, index));
2133             index++;
2134         }
2135 
2136         /* Consume valid HPTEs */
2137         chunkstart = index;
2138         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2139                && HPTE_VALID(HPTE(spapr->htab, index))) {
2140             CLEAN_HPTE(HPTE(spapr->htab, index));
2141             index++;
2142         }
2143 
2144         if (index > chunkstart) {
2145             int n_valid = index - chunkstart;
2146 
2147             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2148 
2149             if (has_timeout &&
2150                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2151                 break;
2152             }
2153         }
2154     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2155 
2156     if (index >= htabslots) {
2157         assert(index == htabslots);
2158         index = 0;
2159         spapr->htab_first_pass = false;
2160     }
2161     spapr->htab_save_index = index;
2162 }
2163 
2164 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2165                                 int64_t max_ns)
2166 {
2167     bool final = max_ns < 0;
2168     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2169     int examined = 0, sent = 0;
2170     int index = spapr->htab_save_index;
2171     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2172 
2173     assert(!spapr->htab_first_pass);
2174 
2175     do {
2176         int chunkstart, invalidstart;
2177 
2178         /* Consume non-dirty HPTEs */
2179         while ((index < htabslots)
2180                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2181             index++;
2182             examined++;
2183         }
2184 
2185         chunkstart = index;
2186         /* Consume valid dirty HPTEs */
2187         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2188                && HPTE_DIRTY(HPTE(spapr->htab, index))
2189                && HPTE_VALID(HPTE(spapr->htab, index))) {
2190             CLEAN_HPTE(HPTE(spapr->htab, index));
2191             index++;
2192             examined++;
2193         }
2194 
2195         invalidstart = index;
2196         /* Consume invalid dirty HPTEs */
2197         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2198                && HPTE_DIRTY(HPTE(spapr->htab, index))
2199                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2200             CLEAN_HPTE(HPTE(spapr->htab, index));
2201             index++;
2202             examined++;
2203         }
2204 
2205         if (index > chunkstart) {
2206             int n_valid = invalidstart - chunkstart;
2207             int n_invalid = index - invalidstart;
2208 
2209             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2210             sent += index - chunkstart;
2211 
2212             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2213                 break;
2214             }
2215         }
2216 
2217         if (examined >= htabslots) {
2218             break;
2219         }
2220 
2221         if (index >= htabslots) {
2222             assert(index == htabslots);
2223             index = 0;
2224         }
2225     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2226 
2227     if (index >= htabslots) {
2228         assert(index == htabslots);
2229         index = 0;
2230     }
2231 
2232     spapr->htab_save_index = index;
2233 
2234     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2235 }
2236 
2237 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2238 #define MAX_KVM_BUF_SIZE    2048
2239 
2240 static int htab_save_iterate(QEMUFile *f, void *opaque)
2241 {
2242     SpaprMachineState *spapr = opaque;
2243     int fd;
2244     int rc = 0;
2245 
2246     /* Iteration header */
2247     if (!spapr->htab_shift) {
2248         qemu_put_be32(f, -1);
2249         return 1;
2250     } else {
2251         qemu_put_be32(f, 0);
2252     }
2253 
2254     if (!spapr->htab) {
2255         assert(kvm_enabled());
2256 
2257         fd = get_htab_fd(spapr);
2258         if (fd < 0) {
2259             return fd;
2260         }
2261 
2262         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2263         if (rc < 0) {
2264             return rc;
2265         }
2266     } else  if (spapr->htab_first_pass) {
2267         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2268     } else {
2269         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2270     }
2271 
2272     htab_save_end_marker(f);
2273 
2274     return rc;
2275 }
2276 
2277 static int htab_save_complete(QEMUFile *f, void *opaque)
2278 {
2279     SpaprMachineState *spapr = opaque;
2280     int fd;
2281 
2282     /* Iteration header */
2283     if (!spapr->htab_shift) {
2284         qemu_put_be32(f, -1);
2285         return 0;
2286     } else {
2287         qemu_put_be32(f, 0);
2288     }
2289 
2290     if (!spapr->htab) {
2291         int rc;
2292 
2293         assert(kvm_enabled());
2294 
2295         fd = get_htab_fd(spapr);
2296         if (fd < 0) {
2297             return fd;
2298         }
2299 
2300         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2301         if (rc < 0) {
2302             return rc;
2303         }
2304     } else {
2305         if (spapr->htab_first_pass) {
2306             htab_save_first_pass(f, spapr, -1);
2307         }
2308         htab_save_later_pass(f, spapr, -1);
2309     }
2310 
2311     /* End marker */
2312     htab_save_end_marker(f);
2313 
2314     return 0;
2315 }
2316 
2317 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2318 {
2319     SpaprMachineState *spapr = opaque;
2320     uint32_t section_hdr;
2321     int fd = -1;
2322     Error *local_err = NULL;
2323 
2324     if (version_id < 1 || version_id > 1) {
2325         error_report("htab_load() bad version");
2326         return -EINVAL;
2327     }
2328 
2329     section_hdr = qemu_get_be32(f);
2330 
2331     if (section_hdr == -1) {
2332         spapr_free_hpt(spapr);
2333         return 0;
2334     }
2335 
2336     if (section_hdr) {
2337         /* First section gives the htab size */
2338         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2339         if (local_err) {
2340             error_report_err(local_err);
2341             return -EINVAL;
2342         }
2343         return 0;
2344     }
2345 
2346     if (!spapr->htab) {
2347         assert(kvm_enabled());
2348 
2349         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2350         if (fd < 0) {
2351             error_report_err(local_err);
2352             return fd;
2353         }
2354     }
2355 
2356     while (true) {
2357         uint32_t index;
2358         uint16_t n_valid, n_invalid;
2359 
2360         index = qemu_get_be32(f);
2361         n_valid = qemu_get_be16(f);
2362         n_invalid = qemu_get_be16(f);
2363 
2364         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2365             /* End of Stream */
2366             break;
2367         }
2368 
2369         if ((index + n_valid + n_invalid) >
2370             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2371             /* Bad index in stream */
2372             error_report(
2373                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2374                 index, n_valid, n_invalid, spapr->htab_shift);
2375             return -EINVAL;
2376         }
2377 
2378         if (spapr->htab) {
2379             if (n_valid) {
2380                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2381                                 HASH_PTE_SIZE_64 * n_valid);
2382             }
2383             if (n_invalid) {
2384                 memset(HPTE(spapr->htab, index + n_valid), 0,
2385                        HASH_PTE_SIZE_64 * n_invalid);
2386             }
2387         } else {
2388             int rc;
2389 
2390             assert(fd >= 0);
2391 
2392             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2393             if (rc < 0) {
2394                 return rc;
2395             }
2396         }
2397     }
2398 
2399     if (!spapr->htab) {
2400         assert(fd >= 0);
2401         close(fd);
2402     }
2403 
2404     return 0;
2405 }
2406 
2407 static void htab_save_cleanup(void *opaque)
2408 {
2409     SpaprMachineState *spapr = opaque;
2410 
2411     close_htab_fd(spapr);
2412 }
2413 
2414 static SaveVMHandlers savevm_htab_handlers = {
2415     .save_setup = htab_save_setup,
2416     .save_live_iterate = htab_save_iterate,
2417     .save_live_complete_precopy = htab_save_complete,
2418     .save_cleanup = htab_save_cleanup,
2419     .load_state = htab_load,
2420 };
2421 
2422 static void spapr_boot_set(void *opaque, const char *boot_device,
2423                            Error **errp)
2424 {
2425     MachineState *machine = MACHINE(opaque);
2426     machine->boot_order = g_strdup(boot_device);
2427 }
2428 
2429 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2430 {
2431     MachineState *machine = MACHINE(spapr);
2432     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2433     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2434     int i;
2435 
2436     for (i = 0; i < nr_lmbs; i++) {
2437         uint64_t addr;
2438 
2439         addr = i * lmb_size + machine->device_memory->base;
2440         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2441                                addr / lmb_size);
2442     }
2443 }
2444 
2445 /*
2446  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2447  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2448  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2449  */
2450 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2451 {
2452     int i;
2453 
2454     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2455         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2456                    " is not aligned to %" PRIu64 " MiB",
2457                    machine->ram_size,
2458                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2459         return;
2460     }
2461 
2462     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2463         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2464                    " is not aligned to %" PRIu64 " MiB",
2465                    machine->ram_size,
2466                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2467         return;
2468     }
2469 
2470     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2471         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2472             error_setg(errp,
2473                        "Node %d memory size 0x%" PRIx64
2474                        " is not aligned to %" PRIu64 " MiB",
2475                        i, machine->numa_state->nodes[i].node_mem,
2476                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2477             return;
2478         }
2479     }
2480 }
2481 
2482 /* find cpu slot in machine->possible_cpus by core_id */
2483 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2484 {
2485     int index = id / ms->smp.threads;
2486 
2487     if (index >= ms->possible_cpus->len) {
2488         return NULL;
2489     }
2490     if (idx) {
2491         *idx = index;
2492     }
2493     return &ms->possible_cpus->cpus[index];
2494 }
2495 
2496 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2497 {
2498     MachineState *ms = MACHINE(spapr);
2499     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2500     Error *local_err = NULL;
2501     bool vsmt_user = !!spapr->vsmt;
2502     int kvm_smt = kvmppc_smt_threads();
2503     int ret;
2504     unsigned int smp_threads = ms->smp.threads;
2505 
2506     if (!kvm_enabled() && (smp_threads > 1)) {
2507         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2508                      "on a pseries machine");
2509         goto out;
2510     }
2511     if (!is_power_of_2(smp_threads)) {
2512         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2513                      "machine because it must be a power of 2", smp_threads);
2514         goto out;
2515     }
2516 
2517     /* Detemine the VSMT mode to use: */
2518     if (vsmt_user) {
2519         if (spapr->vsmt < smp_threads) {
2520             error_setg(&local_err, "Cannot support VSMT mode %d"
2521                          " because it must be >= threads/core (%d)",
2522                          spapr->vsmt, smp_threads);
2523             goto out;
2524         }
2525         /* In this case, spapr->vsmt has been set by the command line */
2526     } else if (!smc->smp_threads_vsmt) {
2527         /*
2528          * Default VSMT value is tricky, because we need it to be as
2529          * consistent as possible (for migration), but this requires
2530          * changing it for at least some existing cases.  We pick 8 as
2531          * the value that we'd get with KVM on POWER8, the
2532          * overwhelmingly common case in production systems.
2533          */
2534         spapr->vsmt = MAX(8, smp_threads);
2535     } else {
2536         spapr->vsmt = smp_threads;
2537     }
2538 
2539     /* KVM: If necessary, set the SMT mode: */
2540     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2541         ret = kvmppc_set_smt_threads(spapr->vsmt);
2542         if (ret) {
2543             /* Looks like KVM isn't able to change VSMT mode */
2544             error_setg(&local_err,
2545                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2546                        spapr->vsmt, ret);
2547             /* We can live with that if the default one is big enough
2548              * for the number of threads, and a submultiple of the one
2549              * we want.  In this case we'll waste some vcpu ids, but
2550              * behaviour will be correct */
2551             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2552                 warn_report_err(local_err);
2553                 local_err = NULL;
2554                 goto out;
2555             } else {
2556                 if (!vsmt_user) {
2557                     error_append_hint(&local_err,
2558                                       "On PPC, a VM with %d threads/core"
2559                                       " on a host with %d threads/core"
2560                                       " requires the use of VSMT mode %d.\n",
2561                                       smp_threads, kvm_smt, spapr->vsmt);
2562                 }
2563                 kvmppc_hint_smt_possible(&local_err);
2564                 goto out;
2565             }
2566         }
2567     }
2568     /* else TCG: nothing to do currently */
2569 out:
2570     error_propagate(errp, local_err);
2571 }
2572 
2573 static void spapr_init_cpus(SpaprMachineState *spapr)
2574 {
2575     MachineState *machine = MACHINE(spapr);
2576     MachineClass *mc = MACHINE_GET_CLASS(machine);
2577     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2578     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2579     const CPUArchIdList *possible_cpus;
2580     unsigned int smp_cpus = machine->smp.cpus;
2581     unsigned int smp_threads = machine->smp.threads;
2582     unsigned int max_cpus = machine->smp.max_cpus;
2583     int boot_cores_nr = smp_cpus / smp_threads;
2584     int i;
2585 
2586     possible_cpus = mc->possible_cpu_arch_ids(machine);
2587     if (mc->has_hotpluggable_cpus) {
2588         if (smp_cpus % smp_threads) {
2589             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2590                          smp_cpus, smp_threads);
2591             exit(1);
2592         }
2593         if (max_cpus % smp_threads) {
2594             error_report("max_cpus (%u) must be multiple of threads (%u)",
2595                          max_cpus, smp_threads);
2596             exit(1);
2597         }
2598     } else {
2599         if (max_cpus != smp_cpus) {
2600             error_report("This machine version does not support CPU hotplug");
2601             exit(1);
2602         }
2603         boot_cores_nr = possible_cpus->len;
2604     }
2605 
2606     if (smc->pre_2_10_has_unused_icps) {
2607         int i;
2608 
2609         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2610             /* Dummy entries get deregistered when real ICPState objects
2611              * are registered during CPU core hotplug.
2612              */
2613             pre_2_10_vmstate_register_dummy_icp(i);
2614         }
2615     }
2616 
2617     for (i = 0; i < possible_cpus->len; i++) {
2618         int core_id = i * smp_threads;
2619 
2620         if (mc->has_hotpluggable_cpus) {
2621             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2622                                    spapr_vcpu_id(spapr, core_id));
2623         }
2624 
2625         if (i < boot_cores_nr) {
2626             Object *core  = object_new(type);
2627             int nr_threads = smp_threads;
2628 
2629             /* Handle the partially filled core for older machine types */
2630             if ((i + 1) * smp_threads >= smp_cpus) {
2631                 nr_threads = smp_cpus - i * smp_threads;
2632             }
2633 
2634             object_property_set_int(core, nr_threads, "nr-threads",
2635                                     &error_fatal);
2636             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2637                                     &error_fatal);
2638             object_property_set_bool(core, true, "realized", &error_fatal);
2639 
2640             object_unref(core);
2641         }
2642     }
2643 }
2644 
2645 static PCIHostState *spapr_create_default_phb(void)
2646 {
2647     DeviceState *dev;
2648 
2649     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2650     qdev_prop_set_uint32(dev, "index", 0);
2651     qdev_init_nofail(dev);
2652 
2653     return PCI_HOST_BRIDGE(dev);
2654 }
2655 
2656 /* pSeries LPAR / sPAPR hardware init */
2657 static void spapr_machine_init(MachineState *machine)
2658 {
2659     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2660     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2661     const char *kernel_filename = machine->kernel_filename;
2662     const char *initrd_filename = machine->initrd_filename;
2663     PCIHostState *phb;
2664     int i;
2665     MemoryRegion *sysmem = get_system_memory();
2666     MemoryRegion *ram = g_new(MemoryRegion, 1);
2667     hwaddr node0_size = spapr_node0_size(machine);
2668     long load_limit, fw_size;
2669     char *filename;
2670     Error *resize_hpt_err = NULL;
2671 
2672     msi_nonbroken = true;
2673 
2674     QLIST_INIT(&spapr->phbs);
2675     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2676 
2677     /* Determine capabilities to run with */
2678     spapr_caps_init(spapr);
2679 
2680     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2681     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2682         /*
2683          * If the user explicitly requested a mode we should either
2684          * supply it, or fail completely (which we do below).  But if
2685          * it's not set explicitly, we reset our mode to something
2686          * that works
2687          */
2688         if (resize_hpt_err) {
2689             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2690             error_free(resize_hpt_err);
2691             resize_hpt_err = NULL;
2692         } else {
2693             spapr->resize_hpt = smc->resize_hpt_default;
2694         }
2695     }
2696 
2697     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2698 
2699     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2700         /*
2701          * User requested HPT resize, but this host can't supply it.  Bail out
2702          */
2703         error_report_err(resize_hpt_err);
2704         exit(1);
2705     }
2706 
2707     spapr->rma_size = node0_size;
2708 
2709     /* With KVM, we don't actually know whether KVM supports an
2710      * unbounded RMA (PR KVM) or is limited by the hash table size
2711      * (HV KVM using VRMA), so we always assume the latter
2712      *
2713      * In that case, we also limit the initial allocations for RTAS
2714      * etc... to 256M since we have no way to know what the VRMA size
2715      * is going to be as it depends on the size of the hash table
2716      * which isn't determined yet.
2717      */
2718     if (kvm_enabled()) {
2719         spapr->vrma_adjust = 1;
2720         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2721     }
2722 
2723     /* Actually we don't support unbounded RMA anymore since we added
2724      * proper emulation of HV mode. The max we can get is 16G which
2725      * also happens to be what we configure for PAPR mode so make sure
2726      * we don't do anything bigger than that
2727      */
2728     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2729 
2730     if (spapr->rma_size > node0_size) {
2731         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2732                      spapr->rma_size);
2733         exit(1);
2734     }
2735 
2736     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2737     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2738 
2739     /*
2740      * VSMT must be set in order to be able to compute VCPU ids, ie to
2741      * call spapr_max_server_number() or spapr_vcpu_id().
2742      */
2743     spapr_set_vsmt_mode(spapr, &error_fatal);
2744 
2745     /* Set up Interrupt Controller before we create the VCPUs */
2746     spapr_irq_init(spapr, &error_fatal);
2747 
2748     /* Set up containers for ibm,client-architecture-support negotiated options
2749      */
2750     spapr->ov5 = spapr_ovec_new();
2751     spapr->ov5_cas = spapr_ovec_new();
2752 
2753     if (smc->dr_lmb_enabled) {
2754         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2755         spapr_validate_node_memory(machine, &error_fatal);
2756     }
2757 
2758     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2759 
2760     /* advertise support for dedicated HP event source to guests */
2761     if (spapr->use_hotplug_event_source) {
2762         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2763     }
2764 
2765     /* advertise support for HPT resizing */
2766     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2767         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2768     }
2769 
2770     /* advertise support for ibm,dyamic-memory-v2 */
2771     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2772 
2773     /* advertise XIVE on POWER9 machines */
2774     if (spapr->irq->xive) {
2775         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2776     }
2777 
2778     /* init CPUs */
2779     spapr_init_cpus(spapr);
2780 
2781     /*
2782      * check we don't have a memory-less/cpu-less NUMA node
2783      * Firmware relies on the existing memory/cpu topology to provide the
2784      * NUMA topology to the kernel.
2785      * And the linux kernel needs to know the NUMA topology at start
2786      * to be able to hotplug CPUs later.
2787      */
2788     if (machine->numa_state->num_nodes) {
2789         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2790             /* check for memory-less node */
2791             if (machine->numa_state->nodes[i].node_mem == 0) {
2792                 CPUState *cs;
2793                 int found = 0;
2794                 /* check for cpu-less node */
2795                 CPU_FOREACH(cs) {
2796                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2797                     if (cpu->node_id == i) {
2798                         found = 1;
2799                         break;
2800                     }
2801                 }
2802                 /* memory-less and cpu-less node */
2803                 if (!found) {
2804                     error_report(
2805                        "Memory-less/cpu-less nodes are not supported (node %d)",
2806                                  i);
2807                     exit(1);
2808                 }
2809             }
2810         }
2811 
2812     }
2813 
2814     /*
2815      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2816      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2817      * called from vPHB reset handler so we initialize the counter here.
2818      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2819      * must be equally distant from any other node.
2820      * The final value of spapr->gpu_numa_id is going to be written to
2821      * max-associativity-domains in spapr_build_fdt().
2822      */
2823     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2824 
2825     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2826         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2827                               spapr->max_compat_pvr)) {
2828         /* KVM and TCG always allow GTSE with radix... */
2829         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2830     }
2831     /* ... but not with hash (currently). */
2832 
2833     if (kvm_enabled()) {
2834         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2835         kvmppc_enable_logical_ci_hcalls();
2836         kvmppc_enable_set_mode_hcall();
2837 
2838         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2839         kvmppc_enable_clear_ref_mod_hcalls();
2840 
2841         /* Enable H_PAGE_INIT */
2842         kvmppc_enable_h_page_init();
2843     }
2844 
2845     /* allocate RAM */
2846     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2847                                          machine->ram_size);
2848     memory_region_add_subregion(sysmem, 0, ram);
2849 
2850     /* always allocate the device memory information */
2851     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2852 
2853     /* initialize hotplug memory address space */
2854     if (machine->ram_size < machine->maxram_size) {
2855         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2856         /*
2857          * Limit the number of hotpluggable memory slots to half the number
2858          * slots that KVM supports, leaving the other half for PCI and other
2859          * devices. However ensure that number of slots doesn't drop below 32.
2860          */
2861         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2862                            SPAPR_MAX_RAM_SLOTS;
2863 
2864         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2865             max_memslots = SPAPR_MAX_RAM_SLOTS;
2866         }
2867         if (machine->ram_slots > max_memslots) {
2868             error_report("Specified number of memory slots %"
2869                          PRIu64" exceeds max supported %d",
2870                          machine->ram_slots, max_memslots);
2871             exit(1);
2872         }
2873 
2874         machine->device_memory->base = ROUND_UP(machine->ram_size,
2875                                                 SPAPR_DEVICE_MEM_ALIGN);
2876         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2877                            "device-memory", device_mem_size);
2878         memory_region_add_subregion(sysmem, machine->device_memory->base,
2879                                     &machine->device_memory->mr);
2880     }
2881 
2882     if (smc->dr_lmb_enabled) {
2883         spapr_create_lmb_dr_connectors(spapr);
2884     }
2885 
2886     /* Set up RTAS event infrastructure */
2887     spapr_events_init(spapr);
2888 
2889     /* Set up the RTC RTAS interfaces */
2890     spapr_rtc_create(spapr);
2891 
2892     /* Set up VIO bus */
2893     spapr->vio_bus = spapr_vio_bus_init();
2894 
2895     for (i = 0; i < serial_max_hds(); i++) {
2896         if (serial_hd(i)) {
2897             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2898         }
2899     }
2900 
2901     /* We always have at least the nvram device on VIO */
2902     spapr_create_nvram(spapr);
2903 
2904     /*
2905      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2906      * connectors (described in root DT node's "ibm,drc-types" property)
2907      * are pre-initialized here. additional child connectors (such as
2908      * connectors for a PHBs PCI slots) are added as needed during their
2909      * parent's realization.
2910      */
2911     if (smc->dr_phb_enabled) {
2912         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2913             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2914         }
2915     }
2916 
2917     /* Set up PCI */
2918     spapr_pci_rtas_init();
2919 
2920     phb = spapr_create_default_phb();
2921 
2922     for (i = 0; i < nb_nics; i++) {
2923         NICInfo *nd = &nd_table[i];
2924 
2925         if (!nd->model) {
2926             nd->model = g_strdup("spapr-vlan");
2927         }
2928 
2929         if (g_str_equal(nd->model, "spapr-vlan") ||
2930             g_str_equal(nd->model, "ibmveth")) {
2931             spapr_vlan_create(spapr->vio_bus, nd);
2932         } else {
2933             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2934         }
2935     }
2936 
2937     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2938         spapr_vscsi_create(spapr->vio_bus);
2939     }
2940 
2941     /* Graphics */
2942     if (spapr_vga_init(phb->bus, &error_fatal)) {
2943         spapr->has_graphics = true;
2944         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2945     }
2946 
2947     if (machine->usb) {
2948         if (smc->use_ohci_by_default) {
2949             pci_create_simple(phb->bus, -1, "pci-ohci");
2950         } else {
2951             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2952         }
2953 
2954         if (spapr->has_graphics) {
2955             USBBus *usb_bus = usb_bus_find(-1);
2956 
2957             usb_create_simple(usb_bus, "usb-kbd");
2958             usb_create_simple(usb_bus, "usb-mouse");
2959         }
2960     }
2961 
2962     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2963         error_report(
2964             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2965             MIN_RMA_SLOF);
2966         exit(1);
2967     }
2968 
2969     if (kernel_filename) {
2970         uint64_t lowaddr = 0;
2971 
2972         spapr->kernel_size = load_elf(kernel_filename, NULL,
2973                                       translate_kernel_address, NULL,
2974                                       NULL, &lowaddr, NULL, 1,
2975                                       PPC_ELF_MACHINE, 0, 0);
2976         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2977             spapr->kernel_size = load_elf(kernel_filename, NULL,
2978                                           translate_kernel_address, NULL, NULL,
2979                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2980                                           0, 0);
2981             spapr->kernel_le = spapr->kernel_size > 0;
2982         }
2983         if (spapr->kernel_size < 0) {
2984             error_report("error loading %s: %s", kernel_filename,
2985                          load_elf_strerror(spapr->kernel_size));
2986             exit(1);
2987         }
2988 
2989         /* load initrd */
2990         if (initrd_filename) {
2991             /* Try to locate the initrd in the gap between the kernel
2992              * and the firmware. Add a bit of space just in case
2993              */
2994             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2995                                   + 0x1ffff) & ~0xffff;
2996             spapr->initrd_size = load_image_targphys(initrd_filename,
2997                                                      spapr->initrd_base,
2998                                                      load_limit
2999                                                      - spapr->initrd_base);
3000             if (spapr->initrd_size < 0) {
3001                 error_report("could not load initial ram disk '%s'",
3002                              initrd_filename);
3003                 exit(1);
3004             }
3005         }
3006     }
3007 
3008     if (bios_name == NULL) {
3009         bios_name = FW_FILE_NAME;
3010     }
3011     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3012     if (!filename) {
3013         error_report("Could not find LPAR firmware '%s'", bios_name);
3014         exit(1);
3015     }
3016     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3017     if (fw_size <= 0) {
3018         error_report("Could not load LPAR firmware '%s'", filename);
3019         exit(1);
3020     }
3021     g_free(filename);
3022 
3023     /* FIXME: Should register things through the MachineState's qdev
3024      * interface, this is a legacy from the sPAPREnvironment structure
3025      * which predated MachineState but had a similar function */
3026     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3027     register_savevm_live("spapr/htab", -1, 1,
3028                          &savevm_htab_handlers, spapr);
3029 
3030     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3031                              &error_fatal);
3032 
3033     qemu_register_boot_set(spapr_boot_set, spapr);
3034 
3035     /*
3036      * Nothing needs to be done to resume a suspended guest because
3037      * suspending does not change the machine state, so no need for
3038      * a ->wakeup method.
3039      */
3040     qemu_register_wakeup_support();
3041 
3042     if (kvm_enabled()) {
3043         /* to stop and start vmclock */
3044         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3045                                          &spapr->tb);
3046 
3047         kvmppc_spapr_enable_inkernel_multitce();
3048     }
3049 }
3050 
3051 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3052 {
3053     if (!vm_type) {
3054         return 0;
3055     }
3056 
3057     if (!strcmp(vm_type, "HV")) {
3058         return 1;
3059     }
3060 
3061     if (!strcmp(vm_type, "PR")) {
3062         return 2;
3063     }
3064 
3065     error_report("Unknown kvm-type specified '%s'", vm_type);
3066     exit(1);
3067 }
3068 
3069 /*
3070  * Implementation of an interface to adjust firmware path
3071  * for the bootindex property handling.
3072  */
3073 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3074                                    DeviceState *dev)
3075 {
3076 #define CAST(type, obj, name) \
3077     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3078     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3079     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3080     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3081 
3082     if (d) {
3083         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3084         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3085         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3086 
3087         if (spapr) {
3088             /*
3089              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3090              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3091              * 0x8000 | (target << 8) | (bus << 5) | lun
3092              * (see the "Logical unit addressing format" table in SAM5)
3093              */
3094             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3095             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3096                                    (uint64_t)id << 48);
3097         } else if (virtio) {
3098             /*
3099              * We use SRP luns of the form 01000000 | (target << 8) | lun
3100              * in the top 32 bits of the 64-bit LUN
3101              * Note: the quote above is from SLOF and it is wrong,
3102              * the actual binding is:
3103              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3104              */
3105             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3106             if (d->lun >= 256) {
3107                 /* Use the LUN "flat space addressing method" */
3108                 id |= 0x4000;
3109             }
3110             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3111                                    (uint64_t)id << 32);
3112         } else if (usb) {
3113             /*
3114              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3115              * in the top 32 bits of the 64-bit LUN
3116              */
3117             unsigned usb_port = atoi(usb->port->path);
3118             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3119             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3120                                    (uint64_t)id << 32);
3121         }
3122     }
3123 
3124     /*
3125      * SLOF probes the USB devices, and if it recognizes that the device is a
3126      * storage device, it changes its name to "storage" instead of "usb-host",
3127      * and additionally adds a child node for the SCSI LUN, so the correct
3128      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3129      */
3130     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3131         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3132         if (usb_host_dev_is_scsi_storage(usbdev)) {
3133             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3134         }
3135     }
3136 
3137     if (phb) {
3138         /* Replace "pci" with "pci@800000020000000" */
3139         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3140     }
3141 
3142     if (vsc) {
3143         /* Same logic as virtio above */
3144         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3145         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3146     }
3147 
3148     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3149         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3150         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3151         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3152     }
3153 
3154     return NULL;
3155 }
3156 
3157 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     return g_strdup(spapr->kvm_type);
3162 }
3163 
3164 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3165 {
3166     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3167 
3168     g_free(spapr->kvm_type);
3169     spapr->kvm_type = g_strdup(value);
3170 }
3171 
3172 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3173 {
3174     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3175 
3176     return spapr->use_hotplug_event_source;
3177 }
3178 
3179 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3180                                             Error **errp)
3181 {
3182     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3183 
3184     spapr->use_hotplug_event_source = value;
3185 }
3186 
3187 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3188 {
3189     return true;
3190 }
3191 
3192 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3193 {
3194     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3195 
3196     switch (spapr->resize_hpt) {
3197     case SPAPR_RESIZE_HPT_DEFAULT:
3198         return g_strdup("default");
3199     case SPAPR_RESIZE_HPT_DISABLED:
3200         return g_strdup("disabled");
3201     case SPAPR_RESIZE_HPT_ENABLED:
3202         return g_strdup("enabled");
3203     case SPAPR_RESIZE_HPT_REQUIRED:
3204         return g_strdup("required");
3205     }
3206     g_assert_not_reached();
3207 }
3208 
3209 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3210 {
3211     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3212 
3213     if (strcmp(value, "default") == 0) {
3214         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3215     } else if (strcmp(value, "disabled") == 0) {
3216         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3217     } else if (strcmp(value, "enabled") == 0) {
3218         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3219     } else if (strcmp(value, "required") == 0) {
3220         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3221     } else {
3222         error_setg(errp, "Bad value for \"resize-hpt\" property");
3223     }
3224 }
3225 
3226 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3227                                    void *opaque, Error **errp)
3228 {
3229     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3230 }
3231 
3232 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3233                                    void *opaque, Error **errp)
3234 {
3235     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3236 }
3237 
3238 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3239 {
3240     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3241 
3242     if (spapr->irq == &spapr_irq_xics_legacy) {
3243         return g_strdup("legacy");
3244     } else if (spapr->irq == &spapr_irq_xics) {
3245         return g_strdup("xics");
3246     } else if (spapr->irq == &spapr_irq_xive) {
3247         return g_strdup("xive");
3248     } else if (spapr->irq == &spapr_irq_dual) {
3249         return g_strdup("dual");
3250     }
3251     g_assert_not_reached();
3252 }
3253 
3254 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3255 {
3256     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3257 
3258     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3259         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3260         return;
3261     }
3262 
3263     /* The legacy IRQ backend can not be set */
3264     if (strcmp(value, "xics") == 0) {
3265         spapr->irq = &spapr_irq_xics;
3266     } else if (strcmp(value, "xive") == 0) {
3267         spapr->irq = &spapr_irq_xive;
3268     } else if (strcmp(value, "dual") == 0) {
3269         spapr->irq = &spapr_irq_dual;
3270     } else {
3271         error_setg(errp, "Bad value for \"ic-mode\" property");
3272     }
3273 }
3274 
3275 static char *spapr_get_host_model(Object *obj, Error **errp)
3276 {
3277     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3278 
3279     return g_strdup(spapr->host_model);
3280 }
3281 
3282 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3283 {
3284     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3285 
3286     g_free(spapr->host_model);
3287     spapr->host_model = g_strdup(value);
3288 }
3289 
3290 static char *spapr_get_host_serial(Object *obj, Error **errp)
3291 {
3292     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3293 
3294     return g_strdup(spapr->host_serial);
3295 }
3296 
3297 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3298 {
3299     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3300 
3301     g_free(spapr->host_serial);
3302     spapr->host_serial = g_strdup(value);
3303 }
3304 
3305 static void spapr_instance_init(Object *obj)
3306 {
3307     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3308     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3309 
3310     spapr->htab_fd = -1;
3311     spapr->use_hotplug_event_source = true;
3312     object_property_add_str(obj, "kvm-type",
3313                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3314     object_property_set_description(obj, "kvm-type",
3315                                     "Specifies the KVM virtualization mode (HV, PR)",
3316                                     NULL);
3317     object_property_add_bool(obj, "modern-hotplug-events",
3318                             spapr_get_modern_hotplug_events,
3319                             spapr_set_modern_hotplug_events,
3320                             NULL);
3321     object_property_set_description(obj, "modern-hotplug-events",
3322                                     "Use dedicated hotplug event mechanism in"
3323                                     " place of standard EPOW events when possible"
3324                                     " (required for memory hot-unplug support)",
3325                                     NULL);
3326     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3327                             "Maximum permitted CPU compatibility mode",
3328                             &error_fatal);
3329 
3330     object_property_add_str(obj, "resize-hpt",
3331                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3332     object_property_set_description(obj, "resize-hpt",
3333                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3334                                     NULL);
3335     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3336                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3337     object_property_set_description(obj, "vsmt",
3338                                     "Virtual SMT: KVM behaves as if this were"
3339                                     " the host's SMT mode", &error_abort);
3340     object_property_add_bool(obj, "vfio-no-msix-emulation",
3341                              spapr_get_msix_emulation, NULL, NULL);
3342 
3343     /* The machine class defines the default interrupt controller mode */
3344     spapr->irq = smc->irq;
3345     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3346                             spapr_set_ic_mode, NULL);
3347     object_property_set_description(obj, "ic-mode",
3348                  "Specifies the interrupt controller mode (xics, xive, dual)",
3349                  NULL);
3350 
3351     object_property_add_str(obj, "host-model",
3352         spapr_get_host_model, spapr_set_host_model,
3353         &error_abort);
3354     object_property_set_description(obj, "host-model",
3355         "Host model to advertise in guest device tree", &error_abort);
3356     object_property_add_str(obj, "host-serial",
3357         spapr_get_host_serial, spapr_set_host_serial,
3358         &error_abort);
3359     object_property_set_description(obj, "host-serial",
3360         "Host serial number to advertise in guest device tree", &error_abort);
3361 }
3362 
3363 static void spapr_machine_finalizefn(Object *obj)
3364 {
3365     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3366 
3367     g_free(spapr->kvm_type);
3368 }
3369 
3370 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3371 {
3372     cpu_synchronize_state(cs);
3373     ppc_cpu_do_system_reset(cs);
3374 }
3375 
3376 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3377 {
3378     CPUState *cs;
3379 
3380     CPU_FOREACH(cs) {
3381         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3382     }
3383 }
3384 
3385 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3386                           void *fdt, int *fdt_start_offset, Error **errp)
3387 {
3388     uint64_t addr;
3389     uint32_t node;
3390 
3391     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3392     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3393                                     &error_abort);
3394     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3395                                                    SPAPR_MEMORY_BLOCK_SIZE);
3396     return 0;
3397 }
3398 
3399 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3400                            bool dedicated_hp_event_source, Error **errp)
3401 {
3402     SpaprDrc *drc;
3403     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3404     int i;
3405     uint64_t addr = addr_start;
3406     bool hotplugged = spapr_drc_hotplugged(dev);
3407     Error *local_err = NULL;
3408 
3409     for (i = 0; i < nr_lmbs; i++) {
3410         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3411                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3412         g_assert(drc);
3413 
3414         spapr_drc_attach(drc, dev, &local_err);
3415         if (local_err) {
3416             while (addr > addr_start) {
3417                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3418                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3419                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3420                 spapr_drc_detach(drc);
3421             }
3422             error_propagate(errp, local_err);
3423             return;
3424         }
3425         if (!hotplugged) {
3426             spapr_drc_reset(drc);
3427         }
3428         addr += SPAPR_MEMORY_BLOCK_SIZE;
3429     }
3430     /* send hotplug notification to the
3431      * guest only in case of hotplugged memory
3432      */
3433     if (hotplugged) {
3434         if (dedicated_hp_event_source) {
3435             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3436                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3437             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3438                                                    nr_lmbs,
3439                                                    spapr_drc_index(drc));
3440         } else {
3441             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3442                                            nr_lmbs);
3443         }
3444     }
3445 }
3446 
3447 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3448                               Error **errp)
3449 {
3450     Error *local_err = NULL;
3451     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3452     PCDIMMDevice *dimm = PC_DIMM(dev);
3453     uint64_t size, addr;
3454 
3455     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3456 
3457     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3458     if (local_err) {
3459         goto out;
3460     }
3461 
3462     addr = object_property_get_uint(OBJECT(dimm),
3463                                     PC_DIMM_ADDR_PROP, &local_err);
3464     if (local_err) {
3465         goto out_unplug;
3466     }
3467 
3468     spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3469                    &local_err);
3470     if (local_err) {
3471         goto out_unplug;
3472     }
3473 
3474     return;
3475 
3476 out_unplug:
3477     pc_dimm_unplug(dimm, MACHINE(ms));
3478 out:
3479     error_propagate(errp, local_err);
3480 }
3481 
3482 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3483                                   Error **errp)
3484 {
3485     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3486     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3487     PCDIMMDevice *dimm = PC_DIMM(dev);
3488     Error *local_err = NULL;
3489     uint64_t size;
3490     Object *memdev;
3491     hwaddr pagesize;
3492 
3493     if (!smc->dr_lmb_enabled) {
3494         error_setg(errp, "Memory hotplug not supported for this machine");
3495         return;
3496     }
3497 
3498     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3499     if (local_err) {
3500         error_propagate(errp, local_err);
3501         return;
3502     }
3503 
3504     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3505         error_setg(errp, "Hotplugged memory size must be a multiple of "
3506                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3507         return;
3508     }
3509 
3510     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3511                                       &error_abort);
3512     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3513     spapr_check_pagesize(spapr, pagesize, &local_err);
3514     if (local_err) {
3515         error_propagate(errp, local_err);
3516         return;
3517     }
3518 
3519     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3520 }
3521 
3522 struct SpaprDimmState {
3523     PCDIMMDevice *dimm;
3524     uint32_t nr_lmbs;
3525     QTAILQ_ENTRY(SpaprDimmState) next;
3526 };
3527 
3528 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3529                                                        PCDIMMDevice *dimm)
3530 {
3531     SpaprDimmState *dimm_state = NULL;
3532 
3533     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3534         if (dimm_state->dimm == dimm) {
3535             break;
3536         }
3537     }
3538     return dimm_state;
3539 }
3540 
3541 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3542                                                       uint32_t nr_lmbs,
3543                                                       PCDIMMDevice *dimm)
3544 {
3545     SpaprDimmState *ds = NULL;
3546 
3547     /*
3548      * If this request is for a DIMM whose removal had failed earlier
3549      * (due to guest's refusal to remove the LMBs), we would have this
3550      * dimm already in the pending_dimm_unplugs list. In that
3551      * case don't add again.
3552      */
3553     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3554     if (!ds) {
3555         ds = g_malloc0(sizeof(SpaprDimmState));
3556         ds->nr_lmbs = nr_lmbs;
3557         ds->dimm = dimm;
3558         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3559     }
3560     return ds;
3561 }
3562 
3563 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3564                                               SpaprDimmState *dimm_state)
3565 {
3566     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3567     g_free(dimm_state);
3568 }
3569 
3570 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3571                                                         PCDIMMDevice *dimm)
3572 {
3573     SpaprDrc *drc;
3574     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3575                                                   &error_abort);
3576     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3577     uint32_t avail_lmbs = 0;
3578     uint64_t addr_start, addr;
3579     int i;
3580 
3581     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3582                                          &error_abort);
3583 
3584     addr = addr_start;
3585     for (i = 0; i < nr_lmbs; i++) {
3586         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3587                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3588         g_assert(drc);
3589         if (drc->dev) {
3590             avail_lmbs++;
3591         }
3592         addr += SPAPR_MEMORY_BLOCK_SIZE;
3593     }
3594 
3595     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3596 }
3597 
3598 /* Callback to be called during DRC release. */
3599 void spapr_lmb_release(DeviceState *dev)
3600 {
3601     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3602     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3603     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3604 
3605     /* This information will get lost if a migration occurs
3606      * during the unplug process. In this case recover it. */
3607     if (ds == NULL) {
3608         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3609         g_assert(ds);
3610         /* The DRC being examined by the caller at least must be counted */
3611         g_assert(ds->nr_lmbs);
3612     }
3613 
3614     if (--ds->nr_lmbs) {
3615         return;
3616     }
3617 
3618     /*
3619      * Now that all the LMBs have been removed by the guest, call the
3620      * unplug handler chain. This can never fail.
3621      */
3622     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3623     object_unparent(OBJECT(dev));
3624 }
3625 
3626 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3627 {
3628     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3629     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3630 
3631     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3632     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3633     spapr_pending_dimm_unplugs_remove(spapr, ds);
3634 }
3635 
3636 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3637                                         DeviceState *dev, Error **errp)
3638 {
3639     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3640     Error *local_err = NULL;
3641     PCDIMMDevice *dimm = PC_DIMM(dev);
3642     uint32_t nr_lmbs;
3643     uint64_t size, addr_start, addr;
3644     int i;
3645     SpaprDrc *drc;
3646 
3647     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3648     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3649 
3650     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3651                                          &local_err);
3652     if (local_err) {
3653         goto out;
3654     }
3655 
3656     /*
3657      * An existing pending dimm state for this DIMM means that there is an
3658      * unplug operation in progress, waiting for the spapr_lmb_release
3659      * callback to complete the job (BQL can't cover that far). In this case,
3660      * bail out to avoid detaching DRCs that were already released.
3661      */
3662     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3663         error_setg(&local_err,
3664                    "Memory unplug already in progress for device %s",
3665                    dev->id);
3666         goto out;
3667     }
3668 
3669     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3670 
3671     addr = addr_start;
3672     for (i = 0; i < nr_lmbs; i++) {
3673         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3674                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3675         g_assert(drc);
3676 
3677         spapr_drc_detach(drc);
3678         addr += SPAPR_MEMORY_BLOCK_SIZE;
3679     }
3680 
3681     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3682                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3683     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3684                                               nr_lmbs, spapr_drc_index(drc));
3685 out:
3686     error_propagate(errp, local_err);
3687 }
3688 
3689 /* Callback to be called during DRC release. */
3690 void spapr_core_release(DeviceState *dev)
3691 {
3692     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3693 
3694     /* Call the unplug handler chain. This can never fail. */
3695     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3696     object_unparent(OBJECT(dev));
3697 }
3698 
3699 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3700 {
3701     MachineState *ms = MACHINE(hotplug_dev);
3702     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3703     CPUCore *cc = CPU_CORE(dev);
3704     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3705 
3706     if (smc->pre_2_10_has_unused_icps) {
3707         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3708         int i;
3709 
3710         for (i = 0; i < cc->nr_threads; i++) {
3711             CPUState *cs = CPU(sc->threads[i]);
3712 
3713             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3714         }
3715     }
3716 
3717     assert(core_slot);
3718     core_slot->cpu = NULL;
3719     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3720 }
3721 
3722 static
3723 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3724                                Error **errp)
3725 {
3726     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3727     int index;
3728     SpaprDrc *drc;
3729     CPUCore *cc = CPU_CORE(dev);
3730 
3731     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3732         error_setg(errp, "Unable to find CPU core with core-id: %d",
3733                    cc->core_id);
3734         return;
3735     }
3736     if (index == 0) {
3737         error_setg(errp, "Boot CPU core may not be unplugged");
3738         return;
3739     }
3740 
3741     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3742                           spapr_vcpu_id(spapr, cc->core_id));
3743     g_assert(drc);
3744 
3745     spapr_drc_detach(drc);
3746 
3747     spapr_hotplug_req_remove_by_index(drc);
3748 }
3749 
3750 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3751                            void *fdt, int *fdt_start_offset, Error **errp)
3752 {
3753     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3754     CPUState *cs = CPU(core->threads[0]);
3755     PowerPCCPU *cpu = POWERPC_CPU(cs);
3756     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3757     int id = spapr_get_vcpu_id(cpu);
3758     char *nodename;
3759     int offset;
3760 
3761     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3762     offset = fdt_add_subnode(fdt, 0, nodename);
3763     g_free(nodename);
3764 
3765     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3766 
3767     *fdt_start_offset = offset;
3768     return 0;
3769 }
3770 
3771 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3772                             Error **errp)
3773 {
3774     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3775     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3776     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3777     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3778     CPUCore *cc = CPU_CORE(dev);
3779     CPUState *cs;
3780     SpaprDrc *drc;
3781     Error *local_err = NULL;
3782     CPUArchId *core_slot;
3783     int index;
3784     bool hotplugged = spapr_drc_hotplugged(dev);
3785     int i;
3786 
3787     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3788     if (!core_slot) {
3789         error_setg(errp, "Unable to find CPU core with core-id: %d",
3790                    cc->core_id);
3791         return;
3792     }
3793     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3794                           spapr_vcpu_id(spapr, cc->core_id));
3795 
3796     g_assert(drc || !mc->has_hotpluggable_cpus);
3797 
3798     if (drc) {
3799         spapr_drc_attach(drc, dev, &local_err);
3800         if (local_err) {
3801             error_propagate(errp, local_err);
3802             return;
3803         }
3804 
3805         if (hotplugged) {
3806             /*
3807              * Send hotplug notification interrupt to the guest only
3808              * in case of hotplugged CPUs.
3809              */
3810             spapr_hotplug_req_add_by_index(drc);
3811         } else {
3812             spapr_drc_reset(drc);
3813         }
3814     }
3815 
3816     core_slot->cpu = OBJECT(dev);
3817 
3818     if (smc->pre_2_10_has_unused_icps) {
3819         for (i = 0; i < cc->nr_threads; i++) {
3820             cs = CPU(core->threads[i]);
3821             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3822         }
3823     }
3824 
3825     /*
3826      * Set compatibility mode to match the boot CPU, which was either set
3827      * by the machine reset code or by CAS.
3828      */
3829     if (hotplugged) {
3830         for (i = 0; i < cc->nr_threads; i++) {
3831             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3832                            &local_err);
3833             if (local_err) {
3834                 error_propagate(errp, local_err);
3835                 return;
3836             }
3837         }
3838     }
3839 }
3840 
3841 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3842                                 Error **errp)
3843 {
3844     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3845     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3846     Error *local_err = NULL;
3847     CPUCore *cc = CPU_CORE(dev);
3848     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3849     const char *type = object_get_typename(OBJECT(dev));
3850     CPUArchId *core_slot;
3851     int index;
3852     unsigned int smp_threads = machine->smp.threads;
3853 
3854     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3855         error_setg(&local_err, "CPU hotplug not supported for this machine");
3856         goto out;
3857     }
3858 
3859     if (strcmp(base_core_type, type)) {
3860         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3861         goto out;
3862     }
3863 
3864     if (cc->core_id % smp_threads) {
3865         error_setg(&local_err, "invalid core id %d", cc->core_id);
3866         goto out;
3867     }
3868 
3869     /*
3870      * In general we should have homogeneous threads-per-core, but old
3871      * (pre hotplug support) machine types allow the last core to have
3872      * reduced threads as a compatibility hack for when we allowed
3873      * total vcpus not a multiple of threads-per-core.
3874      */
3875     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3876         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3877                    cc->nr_threads, smp_threads);
3878         goto out;
3879     }
3880 
3881     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3882     if (!core_slot) {
3883         error_setg(&local_err, "core id %d out of range", cc->core_id);
3884         goto out;
3885     }
3886 
3887     if (core_slot->cpu) {
3888         error_setg(&local_err, "core %d already populated", cc->core_id);
3889         goto out;
3890     }
3891 
3892     numa_cpu_pre_plug(core_slot, dev, &local_err);
3893 
3894 out:
3895     error_propagate(errp, local_err);
3896 }
3897 
3898 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3899                           void *fdt, int *fdt_start_offset, Error **errp)
3900 {
3901     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3902     int intc_phandle;
3903 
3904     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3905     if (intc_phandle <= 0) {
3906         return -1;
3907     }
3908 
3909     if (spapr_dt_phb(sphb, intc_phandle, fdt, spapr->irq->nr_msis,
3910                      fdt_start_offset)) {
3911         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3912         return -1;
3913     }
3914 
3915     /* generally SLOF creates these, for hotplug it's up to QEMU */
3916     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3917 
3918     return 0;
3919 }
3920 
3921 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3922                                Error **errp)
3923 {
3924     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3925     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3926     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3927     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3928 
3929     if (dev->hotplugged && !smc->dr_phb_enabled) {
3930         error_setg(errp, "PHB hotplug not supported for this machine");
3931         return;
3932     }
3933 
3934     if (sphb->index == (uint32_t)-1) {
3935         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3936         return;
3937     }
3938 
3939     /*
3940      * This will check that sphb->index doesn't exceed the maximum number of
3941      * PHBs for the current machine type.
3942      */
3943     smc->phb_placement(spapr, sphb->index,
3944                        &sphb->buid, &sphb->io_win_addr,
3945                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3946                        windows_supported, sphb->dma_liobn,
3947                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3948                        errp);
3949 }
3950 
3951 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3952                            Error **errp)
3953 {
3954     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3955     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3956     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3957     SpaprDrc *drc;
3958     bool hotplugged = spapr_drc_hotplugged(dev);
3959     Error *local_err = NULL;
3960 
3961     if (!smc->dr_phb_enabled) {
3962         return;
3963     }
3964 
3965     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3966     /* hotplug hooks should check it's enabled before getting this far */
3967     assert(drc);
3968 
3969     spapr_drc_attach(drc, DEVICE(dev), &local_err);
3970     if (local_err) {
3971         error_propagate(errp, local_err);
3972         return;
3973     }
3974 
3975     if (hotplugged) {
3976         spapr_hotplug_req_add_by_index(drc);
3977     } else {
3978         spapr_drc_reset(drc);
3979     }
3980 }
3981 
3982 void spapr_phb_release(DeviceState *dev)
3983 {
3984     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3985 
3986     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3987     object_unparent(OBJECT(dev));
3988 }
3989 
3990 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3991 {
3992     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3993 }
3994 
3995 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3996                                      DeviceState *dev, Error **errp)
3997 {
3998     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3999     SpaprDrc *drc;
4000 
4001     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4002     assert(drc);
4003 
4004     if (!spapr_drc_unplug_requested(drc)) {
4005         spapr_drc_detach(drc);
4006         spapr_hotplug_req_remove_by_index(drc);
4007     }
4008 }
4009 
4010 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4011                                  Error **errp)
4012 {
4013     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4014     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4015 
4016     if (spapr->tpm_proxy != NULL) {
4017         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4018         return;
4019     }
4020 
4021     spapr->tpm_proxy = tpm_proxy;
4022 }
4023 
4024 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4025 {
4026     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4027 
4028     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4029     object_unparent(OBJECT(dev));
4030     spapr->tpm_proxy = NULL;
4031 }
4032 
4033 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4034                                       DeviceState *dev, Error **errp)
4035 {
4036     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4037         spapr_memory_plug(hotplug_dev, dev, errp);
4038     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4039         spapr_core_plug(hotplug_dev, dev, errp);
4040     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4041         spapr_phb_plug(hotplug_dev, dev, errp);
4042     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4043         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4044     }
4045 }
4046 
4047 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4048                                         DeviceState *dev, Error **errp)
4049 {
4050     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4051         spapr_memory_unplug(hotplug_dev, dev);
4052     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4053         spapr_core_unplug(hotplug_dev, dev);
4054     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4055         spapr_phb_unplug(hotplug_dev, dev);
4056     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4057         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4058     }
4059 }
4060 
4061 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4062                                                 DeviceState *dev, Error **errp)
4063 {
4064     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4065     MachineClass *mc = MACHINE_GET_CLASS(sms);
4066     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4067 
4068     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4069         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4070             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4071         } else {
4072             /* NOTE: this means there is a window after guest reset, prior to
4073              * CAS negotiation, where unplug requests will fail due to the
4074              * capability not being detected yet. This is a bit different than
4075              * the case with PCI unplug, where the events will be queued and
4076              * eventually handled by the guest after boot
4077              */
4078             error_setg(errp, "Memory hot unplug not supported for this guest");
4079         }
4080     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4081         if (!mc->has_hotpluggable_cpus) {
4082             error_setg(errp, "CPU hot unplug not supported on this machine");
4083             return;
4084         }
4085         spapr_core_unplug_request(hotplug_dev, dev, errp);
4086     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4087         if (!smc->dr_phb_enabled) {
4088             error_setg(errp, "PHB hot unplug not supported on this machine");
4089             return;
4090         }
4091         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4092     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4093         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4094     }
4095 }
4096 
4097 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4098                                           DeviceState *dev, Error **errp)
4099 {
4100     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4101         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4102     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4103         spapr_core_pre_plug(hotplug_dev, dev, errp);
4104     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4105         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4106     }
4107 }
4108 
4109 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4110                                                  DeviceState *dev)
4111 {
4112     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4113         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4114         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4115         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4116         return HOTPLUG_HANDLER(machine);
4117     }
4118     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4119         PCIDevice *pcidev = PCI_DEVICE(dev);
4120         PCIBus *root = pci_device_root_bus(pcidev);
4121         SpaprPhbState *phb =
4122             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4123                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4124 
4125         if (phb) {
4126             return HOTPLUG_HANDLER(phb);
4127         }
4128     }
4129     return NULL;
4130 }
4131 
4132 static CpuInstanceProperties
4133 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4134 {
4135     CPUArchId *core_slot;
4136     MachineClass *mc = MACHINE_GET_CLASS(machine);
4137 
4138     /* make sure possible_cpu are intialized */
4139     mc->possible_cpu_arch_ids(machine);
4140     /* get CPU core slot containing thread that matches cpu_index */
4141     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4142     assert(core_slot);
4143     return core_slot->props;
4144 }
4145 
4146 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4147 {
4148     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4149 }
4150 
4151 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4152 {
4153     int i;
4154     unsigned int smp_threads = machine->smp.threads;
4155     unsigned int smp_cpus = machine->smp.cpus;
4156     const char *core_type;
4157     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4158     MachineClass *mc = MACHINE_GET_CLASS(machine);
4159 
4160     if (!mc->has_hotpluggable_cpus) {
4161         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4162     }
4163     if (machine->possible_cpus) {
4164         assert(machine->possible_cpus->len == spapr_max_cores);
4165         return machine->possible_cpus;
4166     }
4167 
4168     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4169     if (!core_type) {
4170         error_report("Unable to find sPAPR CPU Core definition");
4171         exit(1);
4172     }
4173 
4174     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4175                              sizeof(CPUArchId) * spapr_max_cores);
4176     machine->possible_cpus->len = spapr_max_cores;
4177     for (i = 0; i < machine->possible_cpus->len; i++) {
4178         int core_id = i * smp_threads;
4179 
4180         machine->possible_cpus->cpus[i].type = core_type;
4181         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4182         machine->possible_cpus->cpus[i].arch_id = core_id;
4183         machine->possible_cpus->cpus[i].props.has_core_id = true;
4184         machine->possible_cpus->cpus[i].props.core_id = core_id;
4185     }
4186     return machine->possible_cpus;
4187 }
4188 
4189 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4190                                 uint64_t *buid, hwaddr *pio,
4191                                 hwaddr *mmio32, hwaddr *mmio64,
4192                                 unsigned n_dma, uint32_t *liobns,
4193                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4194 {
4195     /*
4196      * New-style PHB window placement.
4197      *
4198      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4199      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4200      * windows.
4201      *
4202      * Some guest kernels can't work with MMIO windows above 1<<46
4203      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4204      *
4205      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4206      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4207      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4208      * 1TiB 64-bit MMIO windows for each PHB.
4209      */
4210     const uint64_t base_buid = 0x800000020000000ULL;
4211     int i;
4212 
4213     /* Sanity check natural alignments */
4214     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4215     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4216     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4217     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4218     /* Sanity check bounds */
4219     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4220                       SPAPR_PCI_MEM32_WIN_SIZE);
4221     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4222                       SPAPR_PCI_MEM64_WIN_SIZE);
4223 
4224     if (index >= SPAPR_MAX_PHBS) {
4225         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4226                    SPAPR_MAX_PHBS - 1);
4227         return;
4228     }
4229 
4230     *buid = base_buid + index;
4231     for (i = 0; i < n_dma; ++i) {
4232         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4233     }
4234 
4235     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4236     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4237     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4238 
4239     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4240     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4241 }
4242 
4243 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4244 {
4245     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4246 
4247     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4248 }
4249 
4250 static void spapr_ics_resend(XICSFabric *dev)
4251 {
4252     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4253 
4254     ics_resend(spapr->ics);
4255 }
4256 
4257 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4258 {
4259     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4260 
4261     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4262 }
4263 
4264 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4265                                  Monitor *mon)
4266 {
4267     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4268 
4269     spapr_irq_print_info(spapr, mon);
4270     monitor_printf(mon, "irqchip: %s\n",
4271                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4272 }
4273 
4274 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4275 {
4276     return cpu->vcpu_id;
4277 }
4278 
4279 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4280 {
4281     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4282     MachineState *ms = MACHINE(spapr);
4283     int vcpu_id;
4284 
4285     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4286 
4287     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4288         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4289         error_append_hint(errp, "Adjust the number of cpus to %d "
4290                           "or try to raise the number of threads per core\n",
4291                           vcpu_id * ms->smp.threads / spapr->vsmt);
4292         return;
4293     }
4294 
4295     cpu->vcpu_id = vcpu_id;
4296 }
4297 
4298 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4299 {
4300     CPUState *cs;
4301 
4302     CPU_FOREACH(cs) {
4303         PowerPCCPU *cpu = POWERPC_CPU(cs);
4304 
4305         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4306             return cpu;
4307         }
4308     }
4309 
4310     return NULL;
4311 }
4312 
4313 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4314 {
4315     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4316 
4317     /* These are only called by TCG, KVM maintains dispatch state */
4318 
4319     spapr_cpu->prod = false;
4320     if (spapr_cpu->vpa_addr) {
4321         CPUState *cs = CPU(cpu);
4322         uint32_t dispatch;
4323 
4324         dispatch = ldl_be_phys(cs->as,
4325                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4326         dispatch++;
4327         if ((dispatch & 1) != 0) {
4328             qemu_log_mask(LOG_GUEST_ERROR,
4329                           "VPA: incorrect dispatch counter value for "
4330                           "dispatched partition %u, correcting.\n", dispatch);
4331             dispatch++;
4332         }
4333         stl_be_phys(cs->as,
4334                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4335     }
4336 }
4337 
4338 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4339 {
4340     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4341 
4342     if (spapr_cpu->vpa_addr) {
4343         CPUState *cs = CPU(cpu);
4344         uint32_t dispatch;
4345 
4346         dispatch = ldl_be_phys(cs->as,
4347                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4348         dispatch++;
4349         if ((dispatch & 1) != 1) {
4350             qemu_log_mask(LOG_GUEST_ERROR,
4351                           "VPA: incorrect dispatch counter value for "
4352                           "preempted partition %u, correcting.\n", dispatch);
4353             dispatch++;
4354         }
4355         stl_be_phys(cs->as,
4356                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4357     }
4358 }
4359 
4360 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4361 {
4362     MachineClass *mc = MACHINE_CLASS(oc);
4363     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4364     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4365     NMIClass *nc = NMI_CLASS(oc);
4366     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4367     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4368     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4369     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4370 
4371     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4372     mc->ignore_boot_device_suffixes = true;
4373 
4374     /*
4375      * We set up the default / latest behaviour here.  The class_init
4376      * functions for the specific versioned machine types can override
4377      * these details for backwards compatibility
4378      */
4379     mc->init = spapr_machine_init;
4380     mc->reset = spapr_machine_reset;
4381     mc->block_default_type = IF_SCSI;
4382     mc->max_cpus = 1024;
4383     mc->no_parallel = 1;
4384     mc->default_boot_order = "";
4385     mc->default_ram_size = 512 * MiB;
4386     mc->default_display = "std";
4387     mc->kvm_type = spapr_kvm_type;
4388     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4389     mc->pci_allow_0_address = true;
4390     assert(!mc->get_hotplug_handler);
4391     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4392     hc->pre_plug = spapr_machine_device_pre_plug;
4393     hc->plug = spapr_machine_device_plug;
4394     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4395     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4396     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4397     hc->unplug_request = spapr_machine_device_unplug_request;
4398     hc->unplug = spapr_machine_device_unplug;
4399 
4400     smc->dr_lmb_enabled = true;
4401     smc->update_dt_enabled = true;
4402     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4403     mc->has_hotpluggable_cpus = true;
4404     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4405     fwc->get_dev_path = spapr_get_fw_dev_path;
4406     nc->nmi_monitor_handler = spapr_nmi;
4407     smc->phb_placement = spapr_phb_placement;
4408     vhc->hypercall = emulate_spapr_hypercall;
4409     vhc->hpt_mask = spapr_hpt_mask;
4410     vhc->map_hptes = spapr_map_hptes;
4411     vhc->unmap_hptes = spapr_unmap_hptes;
4412     vhc->hpte_set_c = spapr_hpte_set_c;
4413     vhc->hpte_set_r = spapr_hpte_set_r;
4414     vhc->get_pate = spapr_get_pate;
4415     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4416     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4417     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4418     xic->ics_get = spapr_ics_get;
4419     xic->ics_resend = spapr_ics_resend;
4420     xic->icp_get = spapr_icp_get;
4421     ispc->print_info = spapr_pic_print_info;
4422     /* Force NUMA node memory size to be a multiple of
4423      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4424      * in which LMBs are represented and hot-added
4425      */
4426     mc->numa_mem_align_shift = 28;
4427     mc->numa_mem_supported = true;
4428     mc->auto_enable_numa = true;
4429 
4430     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4431     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4432     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4433     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4434     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4435     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4436     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4437     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4438     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4439     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4440     spapr_caps_add_properties(smc, &error_abort);
4441     smc->irq = &spapr_irq_dual;
4442     smc->dr_phb_enabled = true;
4443     smc->linux_pci_probe = true;
4444     smc->smp_threads_vsmt = true;
4445 }
4446 
4447 static const TypeInfo spapr_machine_info = {
4448     .name          = TYPE_SPAPR_MACHINE,
4449     .parent        = TYPE_MACHINE,
4450     .abstract      = true,
4451     .instance_size = sizeof(SpaprMachineState),
4452     .instance_init = spapr_instance_init,
4453     .instance_finalize = spapr_machine_finalizefn,
4454     .class_size    = sizeof(SpaprMachineClass),
4455     .class_init    = spapr_machine_class_init,
4456     .interfaces = (InterfaceInfo[]) {
4457         { TYPE_FW_PATH_PROVIDER },
4458         { TYPE_NMI },
4459         { TYPE_HOTPLUG_HANDLER },
4460         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4461         { TYPE_XICS_FABRIC },
4462         { TYPE_INTERRUPT_STATS_PROVIDER },
4463         { }
4464     },
4465 };
4466 
4467 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4468     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4469                                                     void *data)      \
4470     {                                                                \
4471         MachineClass *mc = MACHINE_CLASS(oc);                        \
4472         spapr_machine_##suffix##_class_options(mc);                  \
4473         if (latest) {                                                \
4474             mc->alias = "pseries";                                   \
4475             mc->is_default = 1;                                      \
4476         }                                                            \
4477     }                                                                \
4478     static const TypeInfo spapr_machine_##suffix##_info = {          \
4479         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4480         .parent = TYPE_SPAPR_MACHINE,                                \
4481         .class_init = spapr_machine_##suffix##_class_init,           \
4482     };                                                               \
4483     static void spapr_machine_register_##suffix(void)                \
4484     {                                                                \
4485         type_register(&spapr_machine_##suffix##_info);               \
4486     }                                                                \
4487     type_init(spapr_machine_register_##suffix)
4488 
4489 /*
4490  * pseries-4.2
4491  */
4492 static void spapr_machine_4_2_class_options(MachineClass *mc)
4493 {
4494     /* Defaults for the latest behaviour inherited from the base class */
4495 }
4496 
4497 DEFINE_SPAPR_MACHINE(4_2, "4.2", true);
4498 
4499 /*
4500  * pseries-4.1
4501  */
4502 static void spapr_machine_4_1_class_options(MachineClass *mc)
4503 {
4504     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4505     static GlobalProperty compat[] = {
4506         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4507         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4508     };
4509 
4510     spapr_machine_4_2_class_options(mc);
4511     smc->linux_pci_probe = false;
4512     smc->smp_threads_vsmt = false;
4513     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4514     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4515 }
4516 
4517 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4518 
4519 /*
4520  * pseries-4.0
4521  */
4522 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4523                               uint64_t *buid, hwaddr *pio,
4524                               hwaddr *mmio32, hwaddr *mmio64,
4525                               unsigned n_dma, uint32_t *liobns,
4526                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4527 {
4528     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4529                         nv2gpa, nv2atsd, errp);
4530     *nv2gpa = 0;
4531     *nv2atsd = 0;
4532 }
4533 
4534 static void spapr_machine_4_0_class_options(MachineClass *mc)
4535 {
4536     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4537 
4538     spapr_machine_4_1_class_options(mc);
4539     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4540     smc->phb_placement = phb_placement_4_0;
4541     smc->irq = &spapr_irq_xics;
4542     smc->pre_4_1_migration = true;
4543 }
4544 
4545 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4546 
4547 /*
4548  * pseries-3.1
4549  */
4550 static void spapr_machine_3_1_class_options(MachineClass *mc)
4551 {
4552     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4553 
4554     spapr_machine_4_0_class_options(mc);
4555     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4556 
4557     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4558     smc->update_dt_enabled = false;
4559     smc->dr_phb_enabled = false;
4560     smc->broken_host_serial_model = true;
4561     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4562     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4563     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4564     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4565 }
4566 
4567 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4568 
4569 /*
4570  * pseries-3.0
4571  */
4572 
4573 static void spapr_machine_3_0_class_options(MachineClass *mc)
4574 {
4575     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4576 
4577     spapr_machine_3_1_class_options(mc);
4578     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4579 
4580     smc->legacy_irq_allocation = true;
4581     smc->irq = &spapr_irq_xics_legacy;
4582 }
4583 
4584 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4585 
4586 /*
4587  * pseries-2.12
4588  */
4589 static void spapr_machine_2_12_class_options(MachineClass *mc)
4590 {
4591     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4592     static GlobalProperty compat[] = {
4593         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4594         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4595     };
4596 
4597     spapr_machine_3_0_class_options(mc);
4598     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4599     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4600 
4601     /* We depend on kvm_enabled() to choose a default value for the
4602      * hpt-max-page-size capability. Of course we can't do it here
4603      * because this is too early and the HW accelerator isn't initialzed
4604      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4605      */
4606     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4607 }
4608 
4609 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4610 
4611 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4612 {
4613     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4614 
4615     spapr_machine_2_12_class_options(mc);
4616     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4617     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4618     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4619 }
4620 
4621 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4622 
4623 /*
4624  * pseries-2.11
4625  */
4626 
4627 static void spapr_machine_2_11_class_options(MachineClass *mc)
4628 {
4629     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4630 
4631     spapr_machine_2_12_class_options(mc);
4632     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4633     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4634 }
4635 
4636 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4637 
4638 /*
4639  * pseries-2.10
4640  */
4641 
4642 static void spapr_machine_2_10_class_options(MachineClass *mc)
4643 {
4644     spapr_machine_2_11_class_options(mc);
4645     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4646 }
4647 
4648 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4649 
4650 /*
4651  * pseries-2.9
4652  */
4653 
4654 static void spapr_machine_2_9_class_options(MachineClass *mc)
4655 {
4656     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4657     static GlobalProperty compat[] = {
4658         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4659     };
4660 
4661     spapr_machine_2_10_class_options(mc);
4662     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4663     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4664     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4665     smc->pre_2_10_has_unused_icps = true;
4666     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4667 }
4668 
4669 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4670 
4671 /*
4672  * pseries-2.8
4673  */
4674 
4675 static void spapr_machine_2_8_class_options(MachineClass *mc)
4676 {
4677     static GlobalProperty compat[] = {
4678         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4679     };
4680 
4681     spapr_machine_2_9_class_options(mc);
4682     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4683     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4684     mc->numa_mem_align_shift = 23;
4685 }
4686 
4687 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4688 
4689 /*
4690  * pseries-2.7
4691  */
4692 
4693 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4694                               uint64_t *buid, hwaddr *pio,
4695                               hwaddr *mmio32, hwaddr *mmio64,
4696                               unsigned n_dma, uint32_t *liobns,
4697                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4698 {
4699     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4700     const uint64_t base_buid = 0x800000020000000ULL;
4701     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4702     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4703     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4704     const uint32_t max_index = 255;
4705     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4706 
4707     uint64_t ram_top = MACHINE(spapr)->ram_size;
4708     hwaddr phb0_base, phb_base;
4709     int i;
4710 
4711     /* Do we have device memory? */
4712     if (MACHINE(spapr)->maxram_size > ram_top) {
4713         /* Can't just use maxram_size, because there may be an
4714          * alignment gap between normal and device memory regions
4715          */
4716         ram_top = MACHINE(spapr)->device_memory->base +
4717             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4718     }
4719 
4720     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4721 
4722     if (index > max_index) {
4723         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4724                    max_index);
4725         return;
4726     }
4727 
4728     *buid = base_buid + index;
4729     for (i = 0; i < n_dma; ++i) {
4730         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4731     }
4732 
4733     phb_base = phb0_base + index * phb_spacing;
4734     *pio = phb_base + pio_offset;
4735     *mmio32 = phb_base + mmio_offset;
4736     /*
4737      * We don't set the 64-bit MMIO window, relying on the PHB's
4738      * fallback behaviour of automatically splitting a large "32-bit"
4739      * window into contiguous 32-bit and 64-bit windows
4740      */
4741 
4742     *nv2gpa = 0;
4743     *nv2atsd = 0;
4744 }
4745 
4746 static void spapr_machine_2_7_class_options(MachineClass *mc)
4747 {
4748     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4749     static GlobalProperty compat[] = {
4750         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4751         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4752         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4753         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4754     };
4755 
4756     spapr_machine_2_8_class_options(mc);
4757     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4758     mc->default_machine_opts = "modern-hotplug-events=off";
4759     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4760     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4761     smc->phb_placement = phb_placement_2_7;
4762 }
4763 
4764 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4765 
4766 /*
4767  * pseries-2.6
4768  */
4769 
4770 static void spapr_machine_2_6_class_options(MachineClass *mc)
4771 {
4772     static GlobalProperty compat[] = {
4773         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4774     };
4775 
4776     spapr_machine_2_7_class_options(mc);
4777     mc->has_hotpluggable_cpus = false;
4778     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4779     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4780 }
4781 
4782 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4783 
4784 /*
4785  * pseries-2.5
4786  */
4787 
4788 static void spapr_machine_2_5_class_options(MachineClass *mc)
4789 {
4790     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4791     static GlobalProperty compat[] = {
4792         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4793     };
4794 
4795     spapr_machine_2_6_class_options(mc);
4796     smc->use_ohci_by_default = true;
4797     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4798     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4799 }
4800 
4801 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4802 
4803 /*
4804  * pseries-2.4
4805  */
4806 
4807 static void spapr_machine_2_4_class_options(MachineClass *mc)
4808 {
4809     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4810 
4811     spapr_machine_2_5_class_options(mc);
4812     smc->dr_lmb_enabled = false;
4813     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4814 }
4815 
4816 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4817 
4818 /*
4819  * pseries-2.3
4820  */
4821 
4822 static void spapr_machine_2_3_class_options(MachineClass *mc)
4823 {
4824     static GlobalProperty compat[] = {
4825         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4826     };
4827     spapr_machine_2_4_class_options(mc);
4828     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4829     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4830 }
4831 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4832 
4833 /*
4834  * pseries-2.2
4835  */
4836 
4837 static void spapr_machine_2_2_class_options(MachineClass *mc)
4838 {
4839     static GlobalProperty compat[] = {
4840         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4841     };
4842 
4843     spapr_machine_2_3_class_options(mc);
4844     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4845     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4846     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4847 }
4848 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4849 
4850 /*
4851  * pseries-2.1
4852  */
4853 
4854 static void spapr_machine_2_1_class_options(MachineClass *mc)
4855 {
4856     spapr_machine_2_2_class_options(mc);
4857     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4858 }
4859 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4860 
4861 static void spapr_machine_register_types(void)
4862 {
4863     type_register_static(&spapr_machine_info);
4864 }
4865 
4866 type_init(spapr_machine_register_types)
4867