xref: /openbmc/qemu/hw/ppc/spapr.c (revision fe3e7b71e68ad0272d535afbc2b38a82e6c5379a)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 
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             return ret;
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     return spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
894                                   "ibm,architecture-vec-5");
895 }
896 
897 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
898 {
899     MachineState *ms = MACHINE(spapr);
900     int rtas;
901     GString *hypertas = g_string_sized_new(256);
902     GString *qemu_hypertas = g_string_sized_new(256);
903     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
904     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
905         memory_region_size(&MACHINE(spapr)->device_memory->mr);
906     uint32_t lrdr_capacity[] = {
907         cpu_to_be32(max_device_addr >> 32),
908         cpu_to_be32(max_device_addr & 0xffffffff),
909         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
910         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
911     };
912     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
913     uint32_t maxdomains[] = {
914         cpu_to_be32(4),
915         maxdomain,
916         maxdomain,
917         maxdomain,
918         cpu_to_be32(spapr->gpu_numa_id),
919     };
920 
921     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
922 
923     /* hypertas */
924     add_str(hypertas, "hcall-pft");
925     add_str(hypertas, "hcall-term");
926     add_str(hypertas, "hcall-dabr");
927     add_str(hypertas, "hcall-interrupt");
928     add_str(hypertas, "hcall-tce");
929     add_str(hypertas, "hcall-vio");
930     add_str(hypertas, "hcall-splpar");
931     add_str(hypertas, "hcall-join");
932     add_str(hypertas, "hcall-bulk");
933     add_str(hypertas, "hcall-set-mode");
934     add_str(hypertas, "hcall-sprg0");
935     add_str(hypertas, "hcall-copy");
936     add_str(hypertas, "hcall-debug");
937     add_str(hypertas, "hcall-vphn");
938     add_str(qemu_hypertas, "hcall-memop1");
939 
940     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
941         add_str(hypertas, "hcall-multi-tce");
942     }
943 
944     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
945         add_str(hypertas, "hcall-hpt-resize");
946     }
947 
948     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
949                      hypertas->str, hypertas->len));
950     g_string_free(hypertas, TRUE);
951     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
952                      qemu_hypertas->str, qemu_hypertas->len));
953     g_string_free(qemu_hypertas, TRUE);
954 
955     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
956                      refpoints, sizeof(refpoints)));
957 
958     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
959                      maxdomains, sizeof(maxdomains)));
960 
961     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
962                           RTAS_ERROR_LOG_MAX));
963     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
964                           RTAS_EVENT_SCAN_RATE));
965 
966     g_assert(msi_nonbroken);
967     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
968 
969     /*
970      * According to PAPR, rtas ibm,os-term does not guarantee a return
971      * back to the guest cpu.
972      *
973      * While an additional ibm,extended-os-term property indicates
974      * that rtas call return will always occur. Set this property.
975      */
976     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
977 
978     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
979                      lrdr_capacity, sizeof(lrdr_capacity)));
980 
981     spapr_dt_rtas_tokens(fdt, rtas);
982 }
983 
984 /*
985  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
986  * and the XIVE features that the guest may request and thus the valid
987  * values for bytes 23..26 of option vector 5:
988  */
989 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
990                                           int chosen)
991 {
992     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
993 
994     char val[2 * 4] = {
995         23, 0x00, /* XICS / XIVE mode */
996         24, 0x00, /* Hash/Radix, filled in below. */
997         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
998         26, 0x40, /* Radix options: GTSE == yes. */
999     };
1000 
1001     if (spapr->irq->xics && spapr->irq->xive) {
1002         val[1] = SPAPR_OV5_XIVE_BOTH;
1003     } else if (spapr->irq->xive) {
1004         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1005     } else {
1006         assert(spapr->irq->xics);
1007         val[1] = SPAPR_OV5_XIVE_LEGACY;
1008     }
1009 
1010     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1011                           first_ppc_cpu->compat_pvr)) {
1012         /*
1013          * If we're in a pre POWER9 compat mode then the guest should
1014          * do hash and use the legacy interrupt mode
1015          */
1016         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1017         val[3] = 0x00; /* Hash */
1018     } else if (kvm_enabled()) {
1019         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1020             val[3] = 0x80; /* OV5_MMU_BOTH */
1021         } else if (kvmppc_has_cap_mmu_radix()) {
1022             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1023         } else {
1024             val[3] = 0x00; /* Hash */
1025         }
1026     } else {
1027         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1028         val[3] = 0xC0;
1029     }
1030     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1031                      val, sizeof(val)));
1032 }
1033 
1034 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1035 {
1036     MachineState *machine = MACHINE(spapr);
1037     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1038     int chosen;
1039     const char *boot_device = machine->boot_order;
1040     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1041     size_t cb = 0;
1042     char *bootlist = get_boot_devices_list(&cb);
1043 
1044     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1045 
1046     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1047         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1048                                 machine->kernel_cmdline));
1049     }
1050     if (spapr->initrd_size) {
1051         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1052                               spapr->initrd_base));
1053         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1054                               spapr->initrd_base + spapr->initrd_size));
1055     }
1056 
1057     if (spapr->kernel_size) {
1058         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1059                               cpu_to_be64(spapr->kernel_size) };
1060 
1061         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1062                          &kprop, sizeof(kprop)));
1063         if (spapr->kernel_le) {
1064             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1065         }
1066     }
1067     if (boot_menu) {
1068         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1069     }
1070     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1071     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1072     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1073 
1074     if (cb && bootlist) {
1075         int i;
1076 
1077         for (i = 0; i < cb; i++) {
1078             if (bootlist[i] == '\n') {
1079                 bootlist[i] = ' ';
1080             }
1081         }
1082         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1083     }
1084 
1085     if (boot_device && strlen(boot_device)) {
1086         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1087     }
1088 
1089     if (!spapr->has_graphics && stdout_path) {
1090         /*
1091          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1092          * kernel. New platforms should only use the "stdout-path" property. Set
1093          * the new property and continue using older property to remain
1094          * compatible with the existing firmware.
1095          */
1096         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1097         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1098     }
1099 
1100     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1101     if (smc->linux_pci_probe) {
1102         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1103     }
1104 
1105     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1106 
1107     g_free(stdout_path);
1108     g_free(bootlist);
1109 }
1110 
1111 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1112 {
1113     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1114      * KVM to work under pHyp with some guest co-operation */
1115     int hypervisor;
1116     uint8_t hypercall[16];
1117 
1118     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1119     /* indicate KVM hypercall interface */
1120     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1121     if (kvmppc_has_cap_fixup_hcalls()) {
1122         /*
1123          * Older KVM versions with older guest kernels were broken
1124          * with the magic page, don't allow the guest to map it.
1125          */
1126         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1127                                   sizeof(hypercall))) {
1128             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1129                              hypercall, sizeof(hypercall)));
1130         }
1131     }
1132 }
1133 
1134 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1135 {
1136     MachineState *machine = MACHINE(spapr);
1137     MachineClass *mc = MACHINE_GET_CLASS(machine);
1138     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1139     int ret;
1140     void *fdt;
1141     SpaprPhbState *phb;
1142     char *buf;
1143 
1144     fdt = g_malloc0(space);
1145     _FDT((fdt_create_empty_tree(fdt, space)));
1146 
1147     /* Root node */
1148     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1149     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1150     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1151 
1152     /* Guest UUID & Name*/
1153     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1154     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1155     if (qemu_uuid_set) {
1156         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1157     }
1158     g_free(buf);
1159 
1160     if (qemu_get_vm_name()) {
1161         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1162                                 qemu_get_vm_name()));
1163     }
1164 
1165     /* Host Model & Serial Number */
1166     if (spapr->host_model) {
1167         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1168     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1169         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1170         g_free(buf);
1171     }
1172 
1173     if (spapr->host_serial) {
1174         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1175     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1176         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1177         g_free(buf);
1178     }
1179 
1180     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1181     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1182 
1183     /* /interrupt controller */
1184     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1185 
1186     ret = spapr_populate_memory(spapr, fdt);
1187     if (ret < 0) {
1188         error_report("couldn't setup memory nodes in fdt");
1189         exit(1);
1190     }
1191 
1192     /* /vdevice */
1193     spapr_dt_vdevice(spapr->vio_bus, fdt);
1194 
1195     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1196         ret = spapr_rng_populate_dt(fdt);
1197         if (ret < 0) {
1198             error_report("could not set up rng device in the fdt");
1199             exit(1);
1200         }
1201     }
1202 
1203     QLIST_FOREACH(phb, &spapr->phbs, list) {
1204         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1205         if (ret < 0) {
1206             error_report("couldn't setup PCI devices in fdt");
1207             exit(1);
1208         }
1209     }
1210 
1211     /* cpus */
1212     spapr_populate_cpus_dt_node(fdt, spapr);
1213 
1214     if (smc->dr_lmb_enabled) {
1215         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1216     }
1217 
1218     if (mc->has_hotpluggable_cpus) {
1219         int offset = fdt_path_offset(fdt, "/cpus");
1220         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1221         if (ret < 0) {
1222             error_report("Couldn't set up CPU DR device tree properties");
1223             exit(1);
1224         }
1225     }
1226 
1227     /* /event-sources */
1228     spapr_dt_events(spapr, fdt);
1229 
1230     /* /rtas */
1231     spapr_dt_rtas(spapr, fdt);
1232 
1233     /* /chosen */
1234     if (reset) {
1235         spapr_dt_chosen(spapr, fdt);
1236     }
1237 
1238     /* /hypervisor */
1239     if (kvm_enabled()) {
1240         spapr_dt_hypervisor(spapr, fdt);
1241     }
1242 
1243     /* Build memory reserve map */
1244     if (reset) {
1245         if (spapr->kernel_size) {
1246             _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1247         }
1248         if (spapr->initrd_size) {
1249             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1250                                   spapr->initrd_size)));
1251         }
1252     }
1253 
1254     /* ibm,client-architecture-support updates */
1255     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1256     if (ret < 0) {
1257         error_report("couldn't setup CAS properties fdt");
1258         exit(1);
1259     }
1260 
1261     if (smc->dr_phb_enabled) {
1262         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1263         if (ret < 0) {
1264             error_report("Couldn't set up PHB DR device tree properties");
1265             exit(1);
1266         }
1267     }
1268 
1269     return fdt;
1270 }
1271 
1272 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1273 {
1274     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1275 }
1276 
1277 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1278                                     PowerPCCPU *cpu)
1279 {
1280     CPUPPCState *env = &cpu->env;
1281 
1282     /* The TCG path should also be holding the BQL at this point */
1283     g_assert(qemu_mutex_iothread_locked());
1284 
1285     if (msr_pr) {
1286         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1287         env->gpr[3] = H_PRIVILEGE;
1288     } else {
1289         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1290     }
1291 }
1292 
1293 struct LPCRSyncState {
1294     target_ulong value;
1295     target_ulong mask;
1296 };
1297 
1298 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1299 {
1300     struct LPCRSyncState *s = arg.host_ptr;
1301     PowerPCCPU *cpu = POWERPC_CPU(cs);
1302     CPUPPCState *env = &cpu->env;
1303     target_ulong lpcr;
1304 
1305     cpu_synchronize_state(cs);
1306     lpcr = env->spr[SPR_LPCR];
1307     lpcr &= ~s->mask;
1308     lpcr |= s->value;
1309     ppc_store_lpcr(cpu, lpcr);
1310 }
1311 
1312 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1313 {
1314     CPUState *cs;
1315     struct LPCRSyncState s = {
1316         .value = value,
1317         .mask = mask
1318     };
1319     CPU_FOREACH(cs) {
1320         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1321     }
1322 }
1323 
1324 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1325 {
1326     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1327 
1328     /* Copy PATE1:GR into PATE0:HR */
1329     entry->dw0 = spapr->patb_entry & PATE0_HR;
1330     entry->dw1 = spapr->patb_entry;
1331 }
1332 
1333 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1334 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1335 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1336 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1337 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1338 
1339 /*
1340  * Get the fd to access the kernel htab, re-opening it if necessary
1341  */
1342 static int get_htab_fd(SpaprMachineState *spapr)
1343 {
1344     Error *local_err = NULL;
1345 
1346     if (spapr->htab_fd >= 0) {
1347         return spapr->htab_fd;
1348     }
1349 
1350     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1351     if (spapr->htab_fd < 0) {
1352         error_report_err(local_err);
1353     }
1354 
1355     return spapr->htab_fd;
1356 }
1357 
1358 void close_htab_fd(SpaprMachineState *spapr)
1359 {
1360     if (spapr->htab_fd >= 0) {
1361         close(spapr->htab_fd);
1362     }
1363     spapr->htab_fd = -1;
1364 }
1365 
1366 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1367 {
1368     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1369 
1370     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1371 }
1372 
1373 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1374 {
1375     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1376 
1377     assert(kvm_enabled());
1378 
1379     if (!spapr->htab) {
1380         return 0;
1381     }
1382 
1383     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1384 }
1385 
1386 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1387                                                 hwaddr ptex, int n)
1388 {
1389     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1390     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1391 
1392     if (!spapr->htab) {
1393         /*
1394          * HTAB is controlled by KVM. Fetch into temporary buffer
1395          */
1396         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1397         kvmppc_read_hptes(hptes, ptex, n);
1398         return hptes;
1399     }
1400 
1401     /*
1402      * HTAB is controlled by QEMU. Just point to the internally
1403      * accessible PTEG.
1404      */
1405     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1406 }
1407 
1408 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1409                               const ppc_hash_pte64_t *hptes,
1410                               hwaddr ptex, int n)
1411 {
1412     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1413 
1414     if (!spapr->htab) {
1415         g_free((void *)hptes);
1416     }
1417 
1418     /* Nothing to do for qemu managed HPT */
1419 }
1420 
1421 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1422                       uint64_t pte0, uint64_t pte1)
1423 {
1424     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1425     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1426 
1427     if (!spapr->htab) {
1428         kvmppc_write_hpte(ptex, pte0, pte1);
1429     } else {
1430         if (pte0 & HPTE64_V_VALID) {
1431             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1432             /*
1433              * When setting valid, we write PTE1 first. This ensures
1434              * proper synchronization with the reading code in
1435              * ppc_hash64_pteg_search()
1436              */
1437             smp_wmb();
1438             stq_p(spapr->htab + offset, pte0);
1439         } else {
1440             stq_p(spapr->htab + offset, pte0);
1441             /*
1442              * When clearing it we set PTE0 first. This ensures proper
1443              * synchronization with the reading code in
1444              * ppc_hash64_pteg_search()
1445              */
1446             smp_wmb();
1447             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1448         }
1449     }
1450 }
1451 
1452 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1453                              uint64_t pte1)
1454 {
1455     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1456     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1457 
1458     if (!spapr->htab) {
1459         /* There should always be a hash table when this is called */
1460         error_report("spapr_hpte_set_c called with no hash table !");
1461         return;
1462     }
1463 
1464     /* The HW performs a non-atomic byte update */
1465     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1466 }
1467 
1468 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1469                              uint64_t pte1)
1470 {
1471     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1472     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1473 
1474     if (!spapr->htab) {
1475         /* There should always be a hash table when this is called */
1476         error_report("spapr_hpte_set_r called with no hash table !");
1477         return;
1478     }
1479 
1480     /* The HW performs a non-atomic byte update */
1481     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1482 }
1483 
1484 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1485 {
1486     int shift;
1487 
1488     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1489      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1490      * that's much more than is needed for Linux guests */
1491     shift = ctz64(pow2ceil(ramsize)) - 7;
1492     shift = MAX(shift, 18); /* Minimum architected size */
1493     shift = MIN(shift, 46); /* Maximum architected size */
1494     return shift;
1495 }
1496 
1497 void spapr_free_hpt(SpaprMachineState *spapr)
1498 {
1499     g_free(spapr->htab);
1500     spapr->htab = NULL;
1501     spapr->htab_shift = 0;
1502     close_htab_fd(spapr);
1503 }
1504 
1505 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1506                           Error **errp)
1507 {
1508     long rc;
1509 
1510     /* Clean up any HPT info from a previous boot */
1511     spapr_free_hpt(spapr);
1512 
1513     rc = kvmppc_reset_htab(shift);
1514     if (rc < 0) {
1515         /* kernel-side HPT needed, but couldn't allocate one */
1516         error_setg_errno(errp, errno,
1517                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1518                          shift);
1519         /* This is almost certainly fatal, but if the caller really
1520          * wants to carry on with shift == 0, it's welcome to try */
1521     } else if (rc > 0) {
1522         /* kernel-side HPT allocated */
1523         if (rc != shift) {
1524             error_setg(errp,
1525                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1526                        shift, rc);
1527         }
1528 
1529         spapr->htab_shift = shift;
1530         spapr->htab = NULL;
1531     } else {
1532         /* kernel-side HPT not needed, allocate in userspace instead */
1533         size_t size = 1ULL << shift;
1534         int i;
1535 
1536         spapr->htab = qemu_memalign(size, size);
1537         if (!spapr->htab) {
1538             error_setg_errno(errp, errno,
1539                              "Could not allocate HPT of order %d", shift);
1540             return;
1541         }
1542 
1543         memset(spapr->htab, 0, size);
1544         spapr->htab_shift = shift;
1545 
1546         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1547             DIRTY_HPTE(HPTE(spapr->htab, i));
1548         }
1549     }
1550     /* We're setting up a hash table, so that means we're not radix */
1551     spapr->patb_entry = 0;
1552     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1553 }
1554 
1555 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1556 {
1557     int hpt_shift;
1558 
1559     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1560         || (spapr->cas_reboot
1561             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1562         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1563     } else {
1564         uint64_t current_ram_size;
1565 
1566         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1567         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1568     }
1569     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1570 
1571     if (spapr->vrma_adjust) {
1572         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1573                                           spapr->htab_shift);
1574     }
1575 }
1576 
1577 static int spapr_reset_drcs(Object *child, void *opaque)
1578 {
1579     SpaprDrc *drc =
1580         (SpaprDrc *) object_dynamic_cast(child,
1581                                                  TYPE_SPAPR_DR_CONNECTOR);
1582 
1583     if (drc) {
1584         spapr_drc_reset(drc);
1585     }
1586 
1587     return 0;
1588 }
1589 
1590 static void spapr_machine_reset(MachineState *machine)
1591 {
1592     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1593     PowerPCCPU *first_ppc_cpu;
1594     hwaddr fdt_addr;
1595     void *fdt;
1596     int rc;
1597 
1598     kvmppc_svm_off(&error_fatal);
1599     spapr_caps_apply(spapr);
1600 
1601     first_ppc_cpu = POWERPC_CPU(first_cpu);
1602     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1603         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1604                               spapr->max_compat_pvr)) {
1605         /*
1606          * If using KVM with radix mode available, VCPUs can be started
1607          * without a HPT because KVM will start them in radix mode.
1608          * Set the GR bit in PATE so that we know there is no HPT.
1609          */
1610         spapr->patb_entry = PATE1_GR;
1611         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1612     } else {
1613         spapr_setup_hpt_and_vrma(spapr);
1614     }
1615 
1616     qemu_devices_reset();
1617 
1618     /*
1619      * If this reset wasn't generated by CAS, we should reset our
1620      * negotiated options and start from scratch
1621      */
1622     if (!spapr->cas_reboot) {
1623         spapr_ovec_cleanup(spapr->ov5_cas);
1624         spapr->ov5_cas = spapr_ovec_new();
1625 
1626         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1627     }
1628 
1629     /*
1630      * This is fixing some of the default configuration of the XIVE
1631      * devices. To be called after the reset of the machine devices.
1632      */
1633     spapr_irq_reset(spapr, &error_fatal);
1634 
1635     /*
1636      * There is no CAS under qtest. Simulate one to please the code that
1637      * depends on spapr->ov5_cas. This is especially needed to test device
1638      * unplug, so we do that before resetting the DRCs.
1639      */
1640     if (qtest_enabled()) {
1641         spapr_ovec_cleanup(spapr->ov5_cas);
1642         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1643     }
1644 
1645     /* DRC reset may cause a device to be unplugged. This will cause troubles
1646      * if this device is used by another device (eg, a running vhost backend
1647      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1648      * situations, we reset DRCs after all devices have been reset.
1649      */
1650     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1651 
1652     spapr_clear_pending_events(spapr);
1653 
1654     /*
1655      * We place the device tree and RTAS just below either the top of the RMA,
1656      * or just below 2GB, whichever is lower, so that it can be
1657      * processed with 32-bit real mode code if necessary
1658      */
1659     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1660 
1661     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1662 
1663     rc = fdt_pack(fdt);
1664 
1665     /* Should only fail if we've built a corrupted tree */
1666     assert(rc == 0);
1667 
1668     /* Load the fdt */
1669     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1670     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1671     g_free(spapr->fdt_blob);
1672     spapr->fdt_size = fdt_totalsize(fdt);
1673     spapr->fdt_initial_size = spapr->fdt_size;
1674     spapr->fdt_blob = fdt;
1675 
1676     /* Set up the entry state */
1677     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1678     first_ppc_cpu->env.gpr[5] = 0;
1679 
1680     spapr->cas_reboot = false;
1681 
1682     spapr->mc_status = -1;
1683     spapr->guest_machine_check_addr = -1;
1684 
1685     /* Signal all vCPUs waiting on this condition */
1686     qemu_cond_broadcast(&spapr->mc_delivery_cond);
1687 
1688     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1689 }
1690 
1691 static void spapr_create_nvram(SpaprMachineState *spapr)
1692 {
1693     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1694     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1695 
1696     if (dinfo) {
1697         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1698                             &error_fatal);
1699     }
1700 
1701     qdev_init_nofail(dev);
1702 
1703     spapr->nvram = (struct SpaprNvram *)dev;
1704 }
1705 
1706 static void spapr_rtc_create(SpaprMachineState *spapr)
1707 {
1708     object_initialize_child(OBJECT(spapr), "rtc",
1709                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1710                             &error_fatal, NULL);
1711     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1712                               &error_fatal);
1713     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1714                               "date", &error_fatal);
1715 }
1716 
1717 /* Returns whether we want to use VGA or not */
1718 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1719 {
1720     switch (vga_interface_type) {
1721     case VGA_NONE:
1722         return false;
1723     case VGA_DEVICE:
1724         return true;
1725     case VGA_STD:
1726     case VGA_VIRTIO:
1727     case VGA_CIRRUS:
1728         return pci_vga_init(pci_bus) != NULL;
1729     default:
1730         error_setg(errp,
1731                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1732         return false;
1733     }
1734 }
1735 
1736 static int spapr_pre_load(void *opaque)
1737 {
1738     int rc;
1739 
1740     rc = spapr_caps_pre_load(opaque);
1741     if (rc) {
1742         return rc;
1743     }
1744 
1745     return 0;
1746 }
1747 
1748 static int spapr_post_load(void *opaque, int version_id)
1749 {
1750     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1751     int err = 0;
1752 
1753     err = spapr_caps_post_migration(spapr);
1754     if (err) {
1755         return err;
1756     }
1757 
1758     /*
1759      * In earlier versions, there was no separate qdev for the PAPR
1760      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1761      * So when migrating from those versions, poke the incoming offset
1762      * value into the RTC device
1763      */
1764     if (version_id < 3) {
1765         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1766         if (err) {
1767             return err;
1768         }
1769     }
1770 
1771     if (kvm_enabled() && spapr->patb_entry) {
1772         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1773         bool radix = !!(spapr->patb_entry & PATE1_GR);
1774         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1775 
1776         /*
1777          * Update LPCR:HR and UPRT as they may not be set properly in
1778          * the stream
1779          */
1780         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1781                             LPCR_HR | LPCR_UPRT);
1782 
1783         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1784         if (err) {
1785             error_report("Process table config unsupported by the host");
1786             return -EINVAL;
1787         }
1788     }
1789 
1790     err = spapr_irq_post_load(spapr, version_id);
1791     if (err) {
1792         return err;
1793     }
1794 
1795     return err;
1796 }
1797 
1798 static int spapr_pre_save(void *opaque)
1799 {
1800     int rc;
1801 
1802     rc = spapr_caps_pre_save(opaque);
1803     if (rc) {
1804         return rc;
1805     }
1806 
1807     return 0;
1808 }
1809 
1810 static bool version_before_3(void *opaque, int version_id)
1811 {
1812     return version_id < 3;
1813 }
1814 
1815 static bool spapr_pending_events_needed(void *opaque)
1816 {
1817     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1818     return !QTAILQ_EMPTY(&spapr->pending_events);
1819 }
1820 
1821 static const VMStateDescription vmstate_spapr_event_entry = {
1822     .name = "spapr_event_log_entry",
1823     .version_id = 1,
1824     .minimum_version_id = 1,
1825     .fields = (VMStateField[]) {
1826         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1827         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1828         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1829                                      NULL, extended_length),
1830         VMSTATE_END_OF_LIST()
1831     },
1832 };
1833 
1834 static const VMStateDescription vmstate_spapr_pending_events = {
1835     .name = "spapr_pending_events",
1836     .version_id = 1,
1837     .minimum_version_id = 1,
1838     .needed = spapr_pending_events_needed,
1839     .fields = (VMStateField[]) {
1840         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1841                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1842         VMSTATE_END_OF_LIST()
1843     },
1844 };
1845 
1846 static bool spapr_ov5_cas_needed(void *opaque)
1847 {
1848     SpaprMachineState *spapr = opaque;
1849     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1850     bool cas_needed;
1851 
1852     /* Prior to the introduction of SpaprOptionVector, we had two option
1853      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1854      * Both of these options encode machine topology into the device-tree
1855      * in such a way that the now-booted OS should still be able to interact
1856      * appropriately with QEMU regardless of what options were actually
1857      * negotiatied on the source side.
1858      *
1859      * As such, we can avoid migrating the CAS-negotiated options if these
1860      * are the only options available on the current machine/platform.
1861      * Since these are the only options available for pseries-2.7 and
1862      * earlier, this allows us to maintain old->new/new->old migration
1863      * compatibility.
1864      *
1865      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1866      * via default pseries-2.8 machines and explicit command-line parameters.
1867      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1868      * of the actual CAS-negotiated values to continue working properly. For
1869      * example, availability of memory unplug depends on knowing whether
1870      * OV5_HP_EVT was negotiated via CAS.
1871      *
1872      * Thus, for any cases where the set of available CAS-negotiatable
1873      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1874      * include the CAS-negotiated options in the migration stream, unless
1875      * if they affect boot time behaviour only.
1876      */
1877     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1878     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1879     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1880 
1881     /* We need extra information if we have any bits outside the mask
1882      * defined above */
1883     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1884 
1885     spapr_ovec_cleanup(ov5_mask);
1886 
1887     return cas_needed;
1888 }
1889 
1890 static const VMStateDescription vmstate_spapr_ov5_cas = {
1891     .name = "spapr_option_vector_ov5_cas",
1892     .version_id = 1,
1893     .minimum_version_id = 1,
1894     .needed = spapr_ov5_cas_needed,
1895     .fields = (VMStateField[]) {
1896         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1897                                  vmstate_spapr_ovec, SpaprOptionVector),
1898         VMSTATE_END_OF_LIST()
1899     },
1900 };
1901 
1902 static bool spapr_patb_entry_needed(void *opaque)
1903 {
1904     SpaprMachineState *spapr = opaque;
1905 
1906     return !!spapr->patb_entry;
1907 }
1908 
1909 static const VMStateDescription vmstate_spapr_patb_entry = {
1910     .name = "spapr_patb_entry",
1911     .version_id = 1,
1912     .minimum_version_id = 1,
1913     .needed = spapr_patb_entry_needed,
1914     .fields = (VMStateField[]) {
1915         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1916         VMSTATE_END_OF_LIST()
1917     },
1918 };
1919 
1920 static bool spapr_irq_map_needed(void *opaque)
1921 {
1922     SpaprMachineState *spapr = opaque;
1923 
1924     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1925 }
1926 
1927 static const VMStateDescription vmstate_spapr_irq_map = {
1928     .name = "spapr_irq_map",
1929     .version_id = 1,
1930     .minimum_version_id = 1,
1931     .needed = spapr_irq_map_needed,
1932     .fields = (VMStateField[]) {
1933         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1934         VMSTATE_END_OF_LIST()
1935     },
1936 };
1937 
1938 static bool spapr_dtb_needed(void *opaque)
1939 {
1940     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1941 
1942     return smc->update_dt_enabled;
1943 }
1944 
1945 static int spapr_dtb_pre_load(void *opaque)
1946 {
1947     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1948 
1949     g_free(spapr->fdt_blob);
1950     spapr->fdt_blob = NULL;
1951     spapr->fdt_size = 0;
1952 
1953     return 0;
1954 }
1955 
1956 static const VMStateDescription vmstate_spapr_dtb = {
1957     .name = "spapr_dtb",
1958     .version_id = 1,
1959     .minimum_version_id = 1,
1960     .needed = spapr_dtb_needed,
1961     .pre_load = spapr_dtb_pre_load,
1962     .fields = (VMStateField[]) {
1963         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1964         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1965         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1966                                      fdt_size),
1967         VMSTATE_END_OF_LIST()
1968     },
1969 };
1970 
1971 static bool spapr_fwnmi_needed(void *opaque)
1972 {
1973     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1974 
1975     return spapr->guest_machine_check_addr != -1;
1976 }
1977 
1978 static int spapr_fwnmi_pre_save(void *opaque)
1979 {
1980     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1981 
1982     /*
1983      * Check if machine check handling is in progress and print a
1984      * warning message.
1985      */
1986     if (spapr->mc_status != -1) {
1987         warn_report("A machine check is being handled during migration. The"
1988                 "handler may run and log hardware error on the destination");
1989     }
1990 
1991     return 0;
1992 }
1993 
1994 static const VMStateDescription vmstate_spapr_machine_check = {
1995     .name = "spapr_machine_check",
1996     .version_id = 1,
1997     .minimum_version_id = 1,
1998     .needed = spapr_fwnmi_needed,
1999     .pre_save = spapr_fwnmi_pre_save,
2000     .fields = (VMStateField[]) {
2001         VMSTATE_UINT64(guest_machine_check_addr, SpaprMachineState),
2002         VMSTATE_INT32(mc_status, SpaprMachineState),
2003         VMSTATE_END_OF_LIST()
2004     },
2005 };
2006 
2007 static const VMStateDescription vmstate_spapr = {
2008     .name = "spapr",
2009     .version_id = 3,
2010     .minimum_version_id = 1,
2011     .pre_load = spapr_pre_load,
2012     .post_load = spapr_post_load,
2013     .pre_save = spapr_pre_save,
2014     .fields = (VMStateField[]) {
2015         /* used to be @next_irq */
2016         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2017 
2018         /* RTC offset */
2019         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2020 
2021         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2022         VMSTATE_END_OF_LIST()
2023     },
2024     .subsections = (const VMStateDescription*[]) {
2025         &vmstate_spapr_ov5_cas,
2026         &vmstate_spapr_patb_entry,
2027         &vmstate_spapr_pending_events,
2028         &vmstate_spapr_cap_htm,
2029         &vmstate_spapr_cap_vsx,
2030         &vmstate_spapr_cap_dfp,
2031         &vmstate_spapr_cap_cfpc,
2032         &vmstate_spapr_cap_sbbc,
2033         &vmstate_spapr_cap_ibs,
2034         &vmstate_spapr_cap_hpt_maxpagesize,
2035         &vmstate_spapr_irq_map,
2036         &vmstate_spapr_cap_nested_kvm_hv,
2037         &vmstate_spapr_dtb,
2038         &vmstate_spapr_cap_large_decr,
2039         &vmstate_spapr_cap_ccf_assist,
2040         &vmstate_spapr_cap_fwnmi,
2041         &vmstate_spapr_machine_check,
2042         NULL
2043     }
2044 };
2045 
2046 static int htab_save_setup(QEMUFile *f, void *opaque)
2047 {
2048     SpaprMachineState *spapr = opaque;
2049 
2050     /* "Iteration" header */
2051     if (!spapr->htab_shift) {
2052         qemu_put_be32(f, -1);
2053     } else {
2054         qemu_put_be32(f, spapr->htab_shift);
2055     }
2056 
2057     if (spapr->htab) {
2058         spapr->htab_save_index = 0;
2059         spapr->htab_first_pass = true;
2060     } else {
2061         if (spapr->htab_shift) {
2062             assert(kvm_enabled());
2063         }
2064     }
2065 
2066 
2067     return 0;
2068 }
2069 
2070 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2071                             int chunkstart, int n_valid, int n_invalid)
2072 {
2073     qemu_put_be32(f, chunkstart);
2074     qemu_put_be16(f, n_valid);
2075     qemu_put_be16(f, n_invalid);
2076     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2077                     HASH_PTE_SIZE_64 * n_valid);
2078 }
2079 
2080 static void htab_save_end_marker(QEMUFile *f)
2081 {
2082     qemu_put_be32(f, 0);
2083     qemu_put_be16(f, 0);
2084     qemu_put_be16(f, 0);
2085 }
2086 
2087 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2088                                  int64_t max_ns)
2089 {
2090     bool has_timeout = max_ns != -1;
2091     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2092     int index = spapr->htab_save_index;
2093     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2094 
2095     assert(spapr->htab_first_pass);
2096 
2097     do {
2098         int chunkstart;
2099 
2100         /* Consume invalid HPTEs */
2101         while ((index < htabslots)
2102                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2103             CLEAN_HPTE(HPTE(spapr->htab, index));
2104             index++;
2105         }
2106 
2107         /* Consume valid HPTEs */
2108         chunkstart = index;
2109         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2110                && HPTE_VALID(HPTE(spapr->htab, index))) {
2111             CLEAN_HPTE(HPTE(spapr->htab, index));
2112             index++;
2113         }
2114 
2115         if (index > chunkstart) {
2116             int n_valid = index - chunkstart;
2117 
2118             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2119 
2120             if (has_timeout &&
2121                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2122                 break;
2123             }
2124         }
2125     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2126 
2127     if (index >= htabslots) {
2128         assert(index == htabslots);
2129         index = 0;
2130         spapr->htab_first_pass = false;
2131     }
2132     spapr->htab_save_index = index;
2133 }
2134 
2135 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2136                                 int64_t max_ns)
2137 {
2138     bool final = max_ns < 0;
2139     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2140     int examined = 0, sent = 0;
2141     int index = spapr->htab_save_index;
2142     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2143 
2144     assert(!spapr->htab_first_pass);
2145 
2146     do {
2147         int chunkstart, invalidstart;
2148 
2149         /* Consume non-dirty HPTEs */
2150         while ((index < htabslots)
2151                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2152             index++;
2153             examined++;
2154         }
2155 
2156         chunkstart = index;
2157         /* Consume valid dirty HPTEs */
2158         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2159                && HPTE_DIRTY(HPTE(spapr->htab, index))
2160                && HPTE_VALID(HPTE(spapr->htab, index))) {
2161             CLEAN_HPTE(HPTE(spapr->htab, index));
2162             index++;
2163             examined++;
2164         }
2165 
2166         invalidstart = index;
2167         /* Consume invalid dirty HPTEs */
2168         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2169                && HPTE_DIRTY(HPTE(spapr->htab, index))
2170                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2171             CLEAN_HPTE(HPTE(spapr->htab, index));
2172             index++;
2173             examined++;
2174         }
2175 
2176         if (index > chunkstart) {
2177             int n_valid = invalidstart - chunkstart;
2178             int n_invalid = index - invalidstart;
2179 
2180             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2181             sent += index - chunkstart;
2182 
2183             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2184                 break;
2185             }
2186         }
2187 
2188         if (examined >= htabslots) {
2189             break;
2190         }
2191 
2192         if (index >= htabslots) {
2193             assert(index == htabslots);
2194             index = 0;
2195         }
2196     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2197 
2198     if (index >= htabslots) {
2199         assert(index == htabslots);
2200         index = 0;
2201     }
2202 
2203     spapr->htab_save_index = index;
2204 
2205     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2206 }
2207 
2208 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2209 #define MAX_KVM_BUF_SIZE    2048
2210 
2211 static int htab_save_iterate(QEMUFile *f, void *opaque)
2212 {
2213     SpaprMachineState *spapr = opaque;
2214     int fd;
2215     int rc = 0;
2216 
2217     /* Iteration header */
2218     if (!spapr->htab_shift) {
2219         qemu_put_be32(f, -1);
2220         return 1;
2221     } else {
2222         qemu_put_be32(f, 0);
2223     }
2224 
2225     if (!spapr->htab) {
2226         assert(kvm_enabled());
2227 
2228         fd = get_htab_fd(spapr);
2229         if (fd < 0) {
2230             return fd;
2231         }
2232 
2233         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2234         if (rc < 0) {
2235             return rc;
2236         }
2237     } else  if (spapr->htab_first_pass) {
2238         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2239     } else {
2240         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2241     }
2242 
2243     htab_save_end_marker(f);
2244 
2245     return rc;
2246 }
2247 
2248 static int htab_save_complete(QEMUFile *f, void *opaque)
2249 {
2250     SpaprMachineState *spapr = opaque;
2251     int fd;
2252 
2253     /* Iteration header */
2254     if (!spapr->htab_shift) {
2255         qemu_put_be32(f, -1);
2256         return 0;
2257     } else {
2258         qemu_put_be32(f, 0);
2259     }
2260 
2261     if (!spapr->htab) {
2262         int rc;
2263 
2264         assert(kvm_enabled());
2265 
2266         fd = get_htab_fd(spapr);
2267         if (fd < 0) {
2268             return fd;
2269         }
2270 
2271         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2272         if (rc < 0) {
2273             return rc;
2274         }
2275     } else {
2276         if (spapr->htab_first_pass) {
2277             htab_save_first_pass(f, spapr, -1);
2278         }
2279         htab_save_later_pass(f, spapr, -1);
2280     }
2281 
2282     /* End marker */
2283     htab_save_end_marker(f);
2284 
2285     return 0;
2286 }
2287 
2288 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2289 {
2290     SpaprMachineState *spapr = opaque;
2291     uint32_t section_hdr;
2292     int fd = -1;
2293     Error *local_err = NULL;
2294 
2295     if (version_id < 1 || version_id > 1) {
2296         error_report("htab_load() bad version");
2297         return -EINVAL;
2298     }
2299 
2300     section_hdr = qemu_get_be32(f);
2301 
2302     if (section_hdr == -1) {
2303         spapr_free_hpt(spapr);
2304         return 0;
2305     }
2306 
2307     if (section_hdr) {
2308         /* First section gives the htab size */
2309         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2310         if (local_err) {
2311             error_report_err(local_err);
2312             return -EINVAL;
2313         }
2314         return 0;
2315     }
2316 
2317     if (!spapr->htab) {
2318         assert(kvm_enabled());
2319 
2320         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2321         if (fd < 0) {
2322             error_report_err(local_err);
2323             return fd;
2324         }
2325     }
2326 
2327     while (true) {
2328         uint32_t index;
2329         uint16_t n_valid, n_invalid;
2330 
2331         index = qemu_get_be32(f);
2332         n_valid = qemu_get_be16(f);
2333         n_invalid = qemu_get_be16(f);
2334 
2335         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2336             /* End of Stream */
2337             break;
2338         }
2339 
2340         if ((index + n_valid + n_invalid) >
2341             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2342             /* Bad index in stream */
2343             error_report(
2344                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2345                 index, n_valid, n_invalid, spapr->htab_shift);
2346             return -EINVAL;
2347         }
2348 
2349         if (spapr->htab) {
2350             if (n_valid) {
2351                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2352                                 HASH_PTE_SIZE_64 * n_valid);
2353             }
2354             if (n_invalid) {
2355                 memset(HPTE(spapr->htab, index + n_valid), 0,
2356                        HASH_PTE_SIZE_64 * n_invalid);
2357             }
2358         } else {
2359             int rc;
2360 
2361             assert(fd >= 0);
2362 
2363             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2364             if (rc < 0) {
2365                 return rc;
2366             }
2367         }
2368     }
2369 
2370     if (!spapr->htab) {
2371         assert(fd >= 0);
2372         close(fd);
2373     }
2374 
2375     return 0;
2376 }
2377 
2378 static void htab_save_cleanup(void *opaque)
2379 {
2380     SpaprMachineState *spapr = opaque;
2381 
2382     close_htab_fd(spapr);
2383 }
2384 
2385 static SaveVMHandlers savevm_htab_handlers = {
2386     .save_setup = htab_save_setup,
2387     .save_live_iterate = htab_save_iterate,
2388     .save_live_complete_precopy = htab_save_complete,
2389     .save_cleanup = htab_save_cleanup,
2390     .load_state = htab_load,
2391 };
2392 
2393 static void spapr_boot_set(void *opaque, const char *boot_device,
2394                            Error **errp)
2395 {
2396     MachineState *machine = MACHINE(opaque);
2397     machine->boot_order = g_strdup(boot_device);
2398 }
2399 
2400 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2401 {
2402     MachineState *machine = MACHINE(spapr);
2403     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2404     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2405     int i;
2406 
2407     for (i = 0; i < nr_lmbs; i++) {
2408         uint64_t addr;
2409 
2410         addr = i * lmb_size + machine->device_memory->base;
2411         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2412                                addr / lmb_size);
2413     }
2414 }
2415 
2416 /*
2417  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2418  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2419  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2420  */
2421 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2422 {
2423     int i;
2424 
2425     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2426         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2427                    " is not aligned to %" PRIu64 " MiB",
2428                    machine->ram_size,
2429                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2430         return;
2431     }
2432 
2433     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2434         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2435                    " is not aligned to %" PRIu64 " MiB",
2436                    machine->ram_size,
2437                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2438         return;
2439     }
2440 
2441     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2442         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2443             error_setg(errp,
2444                        "Node %d memory size 0x%" PRIx64
2445                        " is not aligned to %" PRIu64 " MiB",
2446                        i, machine->numa_state->nodes[i].node_mem,
2447                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2448             return;
2449         }
2450     }
2451 }
2452 
2453 /* find cpu slot in machine->possible_cpus by core_id */
2454 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2455 {
2456     int index = id / ms->smp.threads;
2457 
2458     if (index >= ms->possible_cpus->len) {
2459         return NULL;
2460     }
2461     if (idx) {
2462         *idx = index;
2463     }
2464     return &ms->possible_cpus->cpus[index];
2465 }
2466 
2467 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2468 {
2469     MachineState *ms = MACHINE(spapr);
2470     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2471     Error *local_err = NULL;
2472     bool vsmt_user = !!spapr->vsmt;
2473     int kvm_smt = kvmppc_smt_threads();
2474     int ret;
2475     unsigned int smp_threads = ms->smp.threads;
2476 
2477     if (!kvm_enabled() && (smp_threads > 1)) {
2478         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2479                      "on a pseries machine");
2480         goto out;
2481     }
2482     if (!is_power_of_2(smp_threads)) {
2483         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2484                      "machine because it must be a power of 2", smp_threads);
2485         goto out;
2486     }
2487 
2488     /* Detemine the VSMT mode to use: */
2489     if (vsmt_user) {
2490         if (spapr->vsmt < smp_threads) {
2491             error_setg(&local_err, "Cannot support VSMT mode %d"
2492                          " because it must be >= threads/core (%d)",
2493                          spapr->vsmt, smp_threads);
2494             goto out;
2495         }
2496         /* In this case, spapr->vsmt has been set by the command line */
2497     } else if (!smc->smp_threads_vsmt) {
2498         /*
2499          * Default VSMT value is tricky, because we need it to be as
2500          * consistent as possible (for migration), but this requires
2501          * changing it for at least some existing cases.  We pick 8 as
2502          * the value that we'd get with KVM on POWER8, the
2503          * overwhelmingly common case in production systems.
2504          */
2505         spapr->vsmt = MAX(8, smp_threads);
2506     } else {
2507         spapr->vsmt = smp_threads;
2508     }
2509 
2510     /* KVM: If necessary, set the SMT mode: */
2511     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2512         ret = kvmppc_set_smt_threads(spapr->vsmt);
2513         if (ret) {
2514             /* Looks like KVM isn't able to change VSMT mode */
2515             error_setg(&local_err,
2516                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2517                        spapr->vsmt, ret);
2518             /* We can live with that if the default one is big enough
2519              * for the number of threads, and a submultiple of the one
2520              * we want.  In this case we'll waste some vcpu ids, but
2521              * behaviour will be correct */
2522             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2523                 warn_report_err(local_err);
2524                 local_err = NULL;
2525                 goto out;
2526             } else {
2527                 if (!vsmt_user) {
2528                     error_append_hint(&local_err,
2529                                       "On PPC, a VM with %d threads/core"
2530                                       " on a host with %d threads/core"
2531                                       " requires the use of VSMT mode %d.\n",
2532                                       smp_threads, kvm_smt, spapr->vsmt);
2533                 }
2534                 kvmppc_error_append_smt_possible_hint(&local_err);
2535                 goto out;
2536             }
2537         }
2538     }
2539     /* else TCG: nothing to do currently */
2540 out:
2541     error_propagate(errp, local_err);
2542 }
2543 
2544 static void spapr_init_cpus(SpaprMachineState *spapr)
2545 {
2546     MachineState *machine = MACHINE(spapr);
2547     MachineClass *mc = MACHINE_GET_CLASS(machine);
2548     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2549     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2550     const CPUArchIdList *possible_cpus;
2551     unsigned int smp_cpus = machine->smp.cpus;
2552     unsigned int smp_threads = machine->smp.threads;
2553     unsigned int max_cpus = machine->smp.max_cpus;
2554     int boot_cores_nr = smp_cpus / smp_threads;
2555     int i;
2556 
2557     possible_cpus = mc->possible_cpu_arch_ids(machine);
2558     if (mc->has_hotpluggable_cpus) {
2559         if (smp_cpus % smp_threads) {
2560             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2561                          smp_cpus, smp_threads);
2562             exit(1);
2563         }
2564         if (max_cpus % smp_threads) {
2565             error_report("max_cpus (%u) must be multiple of threads (%u)",
2566                          max_cpus, smp_threads);
2567             exit(1);
2568         }
2569     } else {
2570         if (max_cpus != smp_cpus) {
2571             error_report("This machine version does not support CPU hotplug");
2572             exit(1);
2573         }
2574         boot_cores_nr = possible_cpus->len;
2575     }
2576 
2577     if (smc->pre_2_10_has_unused_icps) {
2578         int i;
2579 
2580         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2581             /* Dummy entries get deregistered when real ICPState objects
2582              * are registered during CPU core hotplug.
2583              */
2584             pre_2_10_vmstate_register_dummy_icp(i);
2585         }
2586     }
2587 
2588     for (i = 0; i < possible_cpus->len; i++) {
2589         int core_id = i * smp_threads;
2590 
2591         if (mc->has_hotpluggable_cpus) {
2592             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2593                                    spapr_vcpu_id(spapr, core_id));
2594         }
2595 
2596         if (i < boot_cores_nr) {
2597             Object *core  = object_new(type);
2598             int nr_threads = smp_threads;
2599 
2600             /* Handle the partially filled core for older machine types */
2601             if ((i + 1) * smp_threads >= smp_cpus) {
2602                 nr_threads = smp_cpus - i * smp_threads;
2603             }
2604 
2605             object_property_set_int(core, nr_threads, "nr-threads",
2606                                     &error_fatal);
2607             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2608                                     &error_fatal);
2609             object_property_set_bool(core, true, "realized", &error_fatal);
2610 
2611             object_unref(core);
2612         }
2613     }
2614 }
2615 
2616 static PCIHostState *spapr_create_default_phb(void)
2617 {
2618     DeviceState *dev;
2619 
2620     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2621     qdev_prop_set_uint32(dev, "index", 0);
2622     qdev_init_nofail(dev);
2623 
2624     return PCI_HOST_BRIDGE(dev);
2625 }
2626 
2627 /* pSeries LPAR / sPAPR hardware init */
2628 static void spapr_machine_init(MachineState *machine)
2629 {
2630     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2631     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2632     const char *kernel_filename = machine->kernel_filename;
2633     const char *initrd_filename = machine->initrd_filename;
2634     PCIHostState *phb;
2635     int i;
2636     MemoryRegion *sysmem = get_system_memory();
2637     hwaddr node0_size = spapr_node0_size(machine);
2638     long load_limit, fw_size;
2639     char *filename;
2640     Error *resize_hpt_err = NULL;
2641 
2642     msi_nonbroken = true;
2643 
2644     QLIST_INIT(&spapr->phbs);
2645     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2646 
2647     /* Determine capabilities to run with */
2648     spapr_caps_init(spapr);
2649 
2650     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2651     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2652         /*
2653          * If the user explicitly requested a mode we should either
2654          * supply it, or fail completely (which we do below).  But if
2655          * it's not set explicitly, we reset our mode to something
2656          * that works
2657          */
2658         if (resize_hpt_err) {
2659             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2660             error_free(resize_hpt_err);
2661             resize_hpt_err = NULL;
2662         } else {
2663             spapr->resize_hpt = smc->resize_hpt_default;
2664         }
2665     }
2666 
2667     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2668 
2669     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2670         /*
2671          * User requested HPT resize, but this host can't supply it.  Bail out
2672          */
2673         error_report_err(resize_hpt_err);
2674         exit(1);
2675     }
2676 
2677     spapr->rma_size = node0_size;
2678 
2679     /* With KVM, we don't actually know whether KVM supports an
2680      * unbounded RMA (PR KVM) or is limited by the hash table size
2681      * (HV KVM using VRMA), so we always assume the latter
2682      *
2683      * In that case, we also limit the initial allocations for RTAS
2684      * etc... to 256M since we have no way to know what the VRMA size
2685      * is going to be as it depends on the size of the hash table
2686      * which isn't determined yet.
2687      */
2688     if (kvm_enabled()) {
2689         spapr->vrma_adjust = 1;
2690         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2691     }
2692 
2693     /* Actually we don't support unbounded RMA anymore since we added
2694      * proper emulation of HV mode. The max we can get is 16G which
2695      * also happens to be what we configure for PAPR mode so make sure
2696      * we don't do anything bigger than that
2697      */
2698     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2699 
2700     if (spapr->rma_size > node0_size) {
2701         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2702                      spapr->rma_size);
2703         exit(1);
2704     }
2705 
2706     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2707     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2708 
2709     /*
2710      * VSMT must be set in order to be able to compute VCPU ids, ie to
2711      * call spapr_max_server_number() or spapr_vcpu_id().
2712      */
2713     spapr_set_vsmt_mode(spapr, &error_fatal);
2714 
2715     /* Set up Interrupt Controller before we create the VCPUs */
2716     spapr_irq_init(spapr, &error_fatal);
2717 
2718     /* Set up containers for ibm,client-architecture-support negotiated options
2719      */
2720     spapr->ov5 = spapr_ovec_new();
2721     spapr->ov5_cas = spapr_ovec_new();
2722 
2723     if (smc->dr_lmb_enabled) {
2724         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2725         spapr_validate_node_memory(machine, &error_fatal);
2726     }
2727 
2728     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2729 
2730     /* advertise support for dedicated HP event source to guests */
2731     if (spapr->use_hotplug_event_source) {
2732         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2733     }
2734 
2735     /* advertise support for HPT resizing */
2736     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2737         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2738     }
2739 
2740     /* advertise support for ibm,dyamic-memory-v2 */
2741     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2742 
2743     /* advertise XIVE on POWER9 machines */
2744     if (spapr->irq->xive) {
2745         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2746     }
2747 
2748     /* init CPUs */
2749     spapr_init_cpus(spapr);
2750 
2751     /*
2752      * check we don't have a memory-less/cpu-less NUMA node
2753      * Firmware relies on the existing memory/cpu topology to provide the
2754      * NUMA topology to the kernel.
2755      * And the linux kernel needs to know the NUMA topology at start
2756      * to be able to hotplug CPUs later.
2757      */
2758     if (machine->numa_state->num_nodes) {
2759         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2760             /* check for memory-less node */
2761             if (machine->numa_state->nodes[i].node_mem == 0) {
2762                 CPUState *cs;
2763                 int found = 0;
2764                 /* check for cpu-less node */
2765                 CPU_FOREACH(cs) {
2766                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2767                     if (cpu->node_id == i) {
2768                         found = 1;
2769                         break;
2770                     }
2771                 }
2772                 /* memory-less and cpu-less node */
2773                 if (!found) {
2774                     error_report(
2775                        "Memory-less/cpu-less nodes are not supported (node %d)",
2776                                  i);
2777                     exit(1);
2778                 }
2779             }
2780         }
2781 
2782     }
2783 
2784     /*
2785      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2786      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2787      * called from vPHB reset handler so we initialize the counter here.
2788      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2789      * must be equally distant from any other node.
2790      * The final value of spapr->gpu_numa_id is going to be written to
2791      * max-associativity-domains in spapr_build_fdt().
2792      */
2793     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2794 
2795     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2796         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2797                               spapr->max_compat_pvr)) {
2798         /* KVM and TCG always allow GTSE with radix... */
2799         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2800     }
2801     /* ... but not with hash (currently). */
2802 
2803     if (kvm_enabled()) {
2804         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2805         kvmppc_enable_logical_ci_hcalls();
2806         kvmppc_enable_set_mode_hcall();
2807 
2808         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2809         kvmppc_enable_clear_ref_mod_hcalls();
2810 
2811         /* Enable H_PAGE_INIT */
2812         kvmppc_enable_h_page_init();
2813     }
2814 
2815     /* map RAM */
2816     memory_region_add_subregion(sysmem, 0, machine->ram);
2817 
2818     /* always allocate the device memory information */
2819     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2820 
2821     /* initialize hotplug memory address space */
2822     if (machine->ram_size < machine->maxram_size) {
2823         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2824         /*
2825          * Limit the number of hotpluggable memory slots to half the number
2826          * slots that KVM supports, leaving the other half for PCI and other
2827          * devices. However ensure that number of slots doesn't drop below 32.
2828          */
2829         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2830                            SPAPR_MAX_RAM_SLOTS;
2831 
2832         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2833             max_memslots = SPAPR_MAX_RAM_SLOTS;
2834         }
2835         if (machine->ram_slots > max_memslots) {
2836             error_report("Specified number of memory slots %"
2837                          PRIu64" exceeds max supported %d",
2838                          machine->ram_slots, max_memslots);
2839             exit(1);
2840         }
2841 
2842         machine->device_memory->base = ROUND_UP(machine->ram_size,
2843                                                 SPAPR_DEVICE_MEM_ALIGN);
2844         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2845                            "device-memory", device_mem_size);
2846         memory_region_add_subregion(sysmem, machine->device_memory->base,
2847                                     &machine->device_memory->mr);
2848     }
2849 
2850     if (smc->dr_lmb_enabled) {
2851         spapr_create_lmb_dr_connectors(spapr);
2852     }
2853 
2854     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI_MCE) == SPAPR_CAP_ON) {
2855         /* Create the error string for live migration blocker */
2856         error_setg(&spapr->fwnmi_migration_blocker,
2857             "A machine check is being handled during migration. The handler"
2858             "may run and log hardware error on the destination");
2859     }
2860 
2861     /* Set up RTAS event infrastructure */
2862     spapr_events_init(spapr);
2863 
2864     /* Set up the RTC RTAS interfaces */
2865     spapr_rtc_create(spapr);
2866 
2867     /* Set up VIO bus */
2868     spapr->vio_bus = spapr_vio_bus_init();
2869 
2870     for (i = 0; i < serial_max_hds(); i++) {
2871         if (serial_hd(i)) {
2872             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2873         }
2874     }
2875 
2876     /* We always have at least the nvram device on VIO */
2877     spapr_create_nvram(spapr);
2878 
2879     /*
2880      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2881      * connectors (described in root DT node's "ibm,drc-types" property)
2882      * are pre-initialized here. additional child connectors (such as
2883      * connectors for a PHBs PCI slots) are added as needed during their
2884      * parent's realization.
2885      */
2886     if (smc->dr_phb_enabled) {
2887         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2888             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2889         }
2890     }
2891 
2892     /* Set up PCI */
2893     spapr_pci_rtas_init();
2894 
2895     phb = spapr_create_default_phb();
2896 
2897     for (i = 0; i < nb_nics; i++) {
2898         NICInfo *nd = &nd_table[i];
2899 
2900         if (!nd->model) {
2901             nd->model = g_strdup("spapr-vlan");
2902         }
2903 
2904         if (g_str_equal(nd->model, "spapr-vlan") ||
2905             g_str_equal(nd->model, "ibmveth")) {
2906             spapr_vlan_create(spapr->vio_bus, nd);
2907         } else {
2908             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2909         }
2910     }
2911 
2912     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2913         spapr_vscsi_create(spapr->vio_bus);
2914     }
2915 
2916     /* Graphics */
2917     if (spapr_vga_init(phb->bus, &error_fatal)) {
2918         spapr->has_graphics = true;
2919         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2920     }
2921 
2922     if (machine->usb) {
2923         if (smc->use_ohci_by_default) {
2924             pci_create_simple(phb->bus, -1, "pci-ohci");
2925         } else {
2926             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2927         }
2928 
2929         if (spapr->has_graphics) {
2930             USBBus *usb_bus = usb_bus_find(-1);
2931 
2932             usb_create_simple(usb_bus, "usb-kbd");
2933             usb_create_simple(usb_bus, "usb-mouse");
2934         }
2935     }
2936 
2937     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2938         error_report(
2939             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2940             MIN_RMA_SLOF);
2941         exit(1);
2942     }
2943 
2944     if (kernel_filename) {
2945         uint64_t lowaddr = 0;
2946 
2947         spapr->kernel_size = load_elf(kernel_filename, NULL,
2948                                       translate_kernel_address, NULL,
2949                                       NULL, &lowaddr, NULL, NULL, 1,
2950                                       PPC_ELF_MACHINE, 0, 0);
2951         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2952             spapr->kernel_size = load_elf(kernel_filename, NULL,
2953                                           translate_kernel_address, NULL, NULL,
2954                                           &lowaddr, NULL, NULL, 0,
2955                                           PPC_ELF_MACHINE, 0, 0);
2956             spapr->kernel_le = spapr->kernel_size > 0;
2957         }
2958         if (spapr->kernel_size < 0) {
2959             error_report("error loading %s: %s", kernel_filename,
2960                          load_elf_strerror(spapr->kernel_size));
2961             exit(1);
2962         }
2963 
2964         /* load initrd */
2965         if (initrd_filename) {
2966             /* Try to locate the initrd in the gap between the kernel
2967              * and the firmware. Add a bit of space just in case
2968              */
2969             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2970                                   + 0x1ffff) & ~0xffff;
2971             spapr->initrd_size = load_image_targphys(initrd_filename,
2972                                                      spapr->initrd_base,
2973                                                      load_limit
2974                                                      - spapr->initrd_base);
2975             if (spapr->initrd_size < 0) {
2976                 error_report("could not load initial ram disk '%s'",
2977                              initrd_filename);
2978                 exit(1);
2979             }
2980         }
2981     }
2982 
2983     if (bios_name == NULL) {
2984         bios_name = FW_FILE_NAME;
2985     }
2986     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2987     if (!filename) {
2988         error_report("Could not find LPAR firmware '%s'", bios_name);
2989         exit(1);
2990     }
2991     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2992     if (fw_size <= 0) {
2993         error_report("Could not load LPAR firmware '%s'", filename);
2994         exit(1);
2995     }
2996     g_free(filename);
2997 
2998     /* FIXME: Should register things through the MachineState's qdev
2999      * interface, this is a legacy from the sPAPREnvironment structure
3000      * which predated MachineState but had a similar function */
3001     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3002     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3003                          &savevm_htab_handlers, spapr);
3004 
3005     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3006                              &error_fatal);
3007 
3008     qemu_register_boot_set(spapr_boot_set, spapr);
3009 
3010     /*
3011      * Nothing needs to be done to resume a suspended guest because
3012      * suspending does not change the machine state, so no need for
3013      * a ->wakeup method.
3014      */
3015     qemu_register_wakeup_support();
3016 
3017     if (kvm_enabled()) {
3018         /* to stop and start vmclock */
3019         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3020                                          &spapr->tb);
3021 
3022         kvmppc_spapr_enable_inkernel_multitce();
3023     }
3024 
3025     qemu_cond_init(&spapr->mc_delivery_cond);
3026 }
3027 
3028 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3029 {
3030     if (!vm_type) {
3031         return 0;
3032     }
3033 
3034     if (!strcmp(vm_type, "HV")) {
3035         return 1;
3036     }
3037 
3038     if (!strcmp(vm_type, "PR")) {
3039         return 2;
3040     }
3041 
3042     error_report("Unknown kvm-type specified '%s'", vm_type);
3043     exit(1);
3044 }
3045 
3046 /*
3047  * Implementation of an interface to adjust firmware path
3048  * for the bootindex property handling.
3049  */
3050 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3051                                    DeviceState *dev)
3052 {
3053 #define CAST(type, obj, name) \
3054     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3055     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3056     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3057     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3058 
3059     if (d) {
3060         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3061         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3062         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3063 
3064         if (spapr) {
3065             /*
3066              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3067              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3068              * 0x8000 | (target << 8) | (bus << 5) | lun
3069              * (see the "Logical unit addressing format" table in SAM5)
3070              */
3071             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3072             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3073                                    (uint64_t)id << 48);
3074         } else if (virtio) {
3075             /*
3076              * We use SRP luns of the form 01000000 | (target << 8) | lun
3077              * in the top 32 bits of the 64-bit LUN
3078              * Note: the quote above is from SLOF and it is wrong,
3079              * the actual binding is:
3080              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3081              */
3082             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3083             if (d->lun >= 256) {
3084                 /* Use the LUN "flat space addressing method" */
3085                 id |= 0x4000;
3086             }
3087             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3088                                    (uint64_t)id << 32);
3089         } else if (usb) {
3090             /*
3091              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3092              * in the top 32 bits of the 64-bit LUN
3093              */
3094             unsigned usb_port = atoi(usb->port->path);
3095             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3096             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3097                                    (uint64_t)id << 32);
3098         }
3099     }
3100 
3101     /*
3102      * SLOF probes the USB devices, and if it recognizes that the device is a
3103      * storage device, it changes its name to "storage" instead of "usb-host",
3104      * and additionally adds a child node for the SCSI LUN, so the correct
3105      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3106      */
3107     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3108         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3109         if (usb_host_dev_is_scsi_storage(usbdev)) {
3110             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3111         }
3112     }
3113 
3114     if (phb) {
3115         /* Replace "pci" with "pci@800000020000000" */
3116         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3117     }
3118 
3119     if (vsc) {
3120         /* Same logic as virtio above */
3121         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3122         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3123     }
3124 
3125     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3126         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3127         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3128         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3129     }
3130 
3131     return NULL;
3132 }
3133 
3134 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3135 {
3136     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3137 
3138     return g_strdup(spapr->kvm_type);
3139 }
3140 
3141 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3142 {
3143     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3144 
3145     g_free(spapr->kvm_type);
3146     spapr->kvm_type = g_strdup(value);
3147 }
3148 
3149 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3150 {
3151     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3152 
3153     return spapr->use_hotplug_event_source;
3154 }
3155 
3156 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3157                                             Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     spapr->use_hotplug_event_source = value;
3162 }
3163 
3164 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3165 {
3166     return true;
3167 }
3168 
3169 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3170 {
3171     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3172 
3173     switch (spapr->resize_hpt) {
3174     case SPAPR_RESIZE_HPT_DEFAULT:
3175         return g_strdup("default");
3176     case SPAPR_RESIZE_HPT_DISABLED:
3177         return g_strdup("disabled");
3178     case SPAPR_RESIZE_HPT_ENABLED:
3179         return g_strdup("enabled");
3180     case SPAPR_RESIZE_HPT_REQUIRED:
3181         return g_strdup("required");
3182     }
3183     g_assert_not_reached();
3184 }
3185 
3186 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3187 {
3188     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3189 
3190     if (strcmp(value, "default") == 0) {
3191         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3192     } else if (strcmp(value, "disabled") == 0) {
3193         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3194     } else if (strcmp(value, "enabled") == 0) {
3195         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3196     } else if (strcmp(value, "required") == 0) {
3197         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3198     } else {
3199         error_setg(errp, "Bad value for \"resize-hpt\" property");
3200     }
3201 }
3202 
3203 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3204                                    void *opaque, Error **errp)
3205 {
3206     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3207 }
3208 
3209 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3210                                    void *opaque, Error **errp)
3211 {
3212     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3213 }
3214 
3215 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3216 {
3217     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3218 
3219     if (spapr->irq == &spapr_irq_xics_legacy) {
3220         return g_strdup("legacy");
3221     } else if (spapr->irq == &spapr_irq_xics) {
3222         return g_strdup("xics");
3223     } else if (spapr->irq == &spapr_irq_xive) {
3224         return g_strdup("xive");
3225     } else if (spapr->irq == &spapr_irq_dual) {
3226         return g_strdup("dual");
3227     }
3228     g_assert_not_reached();
3229 }
3230 
3231 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3232 {
3233     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3234 
3235     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3236         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3237         return;
3238     }
3239 
3240     /* The legacy IRQ backend can not be set */
3241     if (strcmp(value, "xics") == 0) {
3242         spapr->irq = &spapr_irq_xics;
3243     } else if (strcmp(value, "xive") == 0) {
3244         spapr->irq = &spapr_irq_xive;
3245     } else if (strcmp(value, "dual") == 0) {
3246         spapr->irq = &spapr_irq_dual;
3247     } else {
3248         error_setg(errp, "Bad value for \"ic-mode\" property");
3249     }
3250 }
3251 
3252 static char *spapr_get_host_model(Object *obj, Error **errp)
3253 {
3254     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3255 
3256     return g_strdup(spapr->host_model);
3257 }
3258 
3259 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3260 {
3261     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3262 
3263     g_free(spapr->host_model);
3264     spapr->host_model = g_strdup(value);
3265 }
3266 
3267 static char *spapr_get_host_serial(Object *obj, Error **errp)
3268 {
3269     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3270 
3271     return g_strdup(spapr->host_serial);
3272 }
3273 
3274 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3275 {
3276     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3277 
3278     g_free(spapr->host_serial);
3279     spapr->host_serial = g_strdup(value);
3280 }
3281 
3282 static void spapr_instance_init(Object *obj)
3283 {
3284     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3285     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3286 
3287     spapr->htab_fd = -1;
3288     spapr->use_hotplug_event_source = true;
3289     object_property_add_str(obj, "kvm-type",
3290                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3291     object_property_set_description(obj, "kvm-type",
3292                                     "Specifies the KVM virtualization mode (HV, PR)",
3293                                     NULL);
3294     object_property_add_bool(obj, "modern-hotplug-events",
3295                             spapr_get_modern_hotplug_events,
3296                             spapr_set_modern_hotplug_events,
3297                             NULL);
3298     object_property_set_description(obj, "modern-hotplug-events",
3299                                     "Use dedicated hotplug event mechanism in"
3300                                     " place of standard EPOW events when possible"
3301                                     " (required for memory hot-unplug support)",
3302                                     NULL);
3303     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3304                             "Maximum permitted CPU compatibility mode",
3305                             &error_fatal);
3306 
3307     object_property_add_str(obj, "resize-hpt",
3308                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3309     object_property_set_description(obj, "resize-hpt",
3310                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3311                                     NULL);
3312     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3313                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3314     object_property_set_description(obj, "vsmt",
3315                                     "Virtual SMT: KVM behaves as if this were"
3316                                     " the host's SMT mode", &error_abort);
3317     object_property_add_bool(obj, "vfio-no-msix-emulation",
3318                              spapr_get_msix_emulation, NULL, NULL);
3319 
3320     /* The machine class defines the default interrupt controller mode */
3321     spapr->irq = smc->irq;
3322     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3323                             spapr_set_ic_mode, NULL);
3324     object_property_set_description(obj, "ic-mode",
3325                  "Specifies the interrupt controller mode (xics, xive, dual)",
3326                  NULL);
3327 
3328     object_property_add_str(obj, "host-model",
3329         spapr_get_host_model, spapr_set_host_model,
3330         &error_abort);
3331     object_property_set_description(obj, "host-model",
3332         "Host model to advertise in guest device tree", &error_abort);
3333     object_property_add_str(obj, "host-serial",
3334         spapr_get_host_serial, spapr_set_host_serial,
3335         &error_abort);
3336     object_property_set_description(obj, "host-serial",
3337         "Host serial number to advertise in guest device tree", &error_abort);
3338 }
3339 
3340 static void spapr_machine_finalizefn(Object *obj)
3341 {
3342     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3343 
3344     g_free(spapr->kvm_type);
3345 }
3346 
3347 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3348 {
3349     cpu_synchronize_state(cs);
3350     ppc_cpu_do_system_reset(cs);
3351 }
3352 
3353 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3354 {
3355     CPUState *cs;
3356 
3357     CPU_FOREACH(cs) {
3358         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3359     }
3360 }
3361 
3362 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3363                           void *fdt, int *fdt_start_offset, Error **errp)
3364 {
3365     uint64_t addr;
3366     uint32_t node;
3367 
3368     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3369     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3370                                     &error_abort);
3371     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3372                                                    SPAPR_MEMORY_BLOCK_SIZE);
3373     return 0;
3374 }
3375 
3376 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3377                            bool dedicated_hp_event_source, Error **errp)
3378 {
3379     SpaprDrc *drc;
3380     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3381     int i;
3382     uint64_t addr = addr_start;
3383     bool hotplugged = spapr_drc_hotplugged(dev);
3384     Error *local_err = NULL;
3385 
3386     for (i = 0; i < nr_lmbs; i++) {
3387         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3388                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3389         g_assert(drc);
3390 
3391         spapr_drc_attach(drc, dev, &local_err);
3392         if (local_err) {
3393             while (addr > addr_start) {
3394                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3395                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3396                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3397                 spapr_drc_detach(drc);
3398             }
3399             error_propagate(errp, local_err);
3400             return;
3401         }
3402         if (!hotplugged) {
3403             spapr_drc_reset(drc);
3404         }
3405         addr += SPAPR_MEMORY_BLOCK_SIZE;
3406     }
3407     /* send hotplug notification to the
3408      * guest only in case of hotplugged memory
3409      */
3410     if (hotplugged) {
3411         if (dedicated_hp_event_source) {
3412             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3413                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3414             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3415                                                    nr_lmbs,
3416                                                    spapr_drc_index(drc));
3417         } else {
3418             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3419                                            nr_lmbs);
3420         }
3421     }
3422 }
3423 
3424 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3425                               Error **errp)
3426 {
3427     Error *local_err = NULL;
3428     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3429     PCDIMMDevice *dimm = PC_DIMM(dev);
3430     uint64_t size, addr;
3431 
3432     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3433 
3434     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3435     if (local_err) {
3436         goto out;
3437     }
3438 
3439     addr = object_property_get_uint(OBJECT(dimm),
3440                                     PC_DIMM_ADDR_PROP, &local_err);
3441     if (local_err) {
3442         goto out_unplug;
3443     }
3444 
3445     spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3446                    &local_err);
3447     if (local_err) {
3448         goto out_unplug;
3449     }
3450 
3451     return;
3452 
3453 out_unplug:
3454     pc_dimm_unplug(dimm, MACHINE(ms));
3455 out:
3456     error_propagate(errp, local_err);
3457 }
3458 
3459 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3460                                   Error **errp)
3461 {
3462     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3463     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3464     PCDIMMDevice *dimm = PC_DIMM(dev);
3465     Error *local_err = NULL;
3466     uint64_t size;
3467     Object *memdev;
3468     hwaddr pagesize;
3469 
3470     if (!smc->dr_lmb_enabled) {
3471         error_setg(errp, "Memory hotplug not supported for this machine");
3472         return;
3473     }
3474 
3475     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3476     if (local_err) {
3477         error_propagate(errp, local_err);
3478         return;
3479     }
3480 
3481     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3482         error_setg(errp, "Hotplugged memory size must be a multiple of "
3483                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3484         return;
3485     }
3486 
3487     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3488                                       &error_abort);
3489     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3490     spapr_check_pagesize(spapr, pagesize, &local_err);
3491     if (local_err) {
3492         error_propagate(errp, local_err);
3493         return;
3494     }
3495 
3496     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3497 }
3498 
3499 struct SpaprDimmState {
3500     PCDIMMDevice *dimm;
3501     uint32_t nr_lmbs;
3502     QTAILQ_ENTRY(SpaprDimmState) next;
3503 };
3504 
3505 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3506                                                        PCDIMMDevice *dimm)
3507 {
3508     SpaprDimmState *dimm_state = NULL;
3509 
3510     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3511         if (dimm_state->dimm == dimm) {
3512             break;
3513         }
3514     }
3515     return dimm_state;
3516 }
3517 
3518 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3519                                                       uint32_t nr_lmbs,
3520                                                       PCDIMMDevice *dimm)
3521 {
3522     SpaprDimmState *ds = NULL;
3523 
3524     /*
3525      * If this request is for a DIMM whose removal had failed earlier
3526      * (due to guest's refusal to remove the LMBs), we would have this
3527      * dimm already in the pending_dimm_unplugs list. In that
3528      * case don't add again.
3529      */
3530     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3531     if (!ds) {
3532         ds = g_malloc0(sizeof(SpaprDimmState));
3533         ds->nr_lmbs = nr_lmbs;
3534         ds->dimm = dimm;
3535         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3536     }
3537     return ds;
3538 }
3539 
3540 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3541                                               SpaprDimmState *dimm_state)
3542 {
3543     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3544     g_free(dimm_state);
3545 }
3546 
3547 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3548                                                         PCDIMMDevice *dimm)
3549 {
3550     SpaprDrc *drc;
3551     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3552                                                   &error_abort);
3553     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3554     uint32_t avail_lmbs = 0;
3555     uint64_t addr_start, addr;
3556     int i;
3557 
3558     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3559                                          &error_abort);
3560 
3561     addr = addr_start;
3562     for (i = 0; i < nr_lmbs; i++) {
3563         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3564                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3565         g_assert(drc);
3566         if (drc->dev) {
3567             avail_lmbs++;
3568         }
3569         addr += SPAPR_MEMORY_BLOCK_SIZE;
3570     }
3571 
3572     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3573 }
3574 
3575 /* Callback to be called during DRC release. */
3576 void spapr_lmb_release(DeviceState *dev)
3577 {
3578     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3579     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3580     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3581 
3582     /* This information will get lost if a migration occurs
3583      * during the unplug process. In this case recover it. */
3584     if (ds == NULL) {
3585         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3586         g_assert(ds);
3587         /* The DRC being examined by the caller at least must be counted */
3588         g_assert(ds->nr_lmbs);
3589     }
3590 
3591     if (--ds->nr_lmbs) {
3592         return;
3593     }
3594 
3595     /*
3596      * Now that all the LMBs have been removed by the guest, call the
3597      * unplug handler chain. This can never fail.
3598      */
3599     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3600     object_unparent(OBJECT(dev));
3601 }
3602 
3603 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3604 {
3605     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3606     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3607 
3608     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3609     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3610     spapr_pending_dimm_unplugs_remove(spapr, ds);
3611 }
3612 
3613 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3614                                         DeviceState *dev, Error **errp)
3615 {
3616     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3617     Error *local_err = NULL;
3618     PCDIMMDevice *dimm = PC_DIMM(dev);
3619     uint32_t nr_lmbs;
3620     uint64_t size, addr_start, addr;
3621     int i;
3622     SpaprDrc *drc;
3623 
3624     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3625     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3626 
3627     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3628                                          &local_err);
3629     if (local_err) {
3630         goto out;
3631     }
3632 
3633     /*
3634      * An existing pending dimm state for this DIMM means that there is an
3635      * unplug operation in progress, waiting for the spapr_lmb_release
3636      * callback to complete the job (BQL can't cover that far). In this case,
3637      * bail out to avoid detaching DRCs that were already released.
3638      */
3639     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3640         error_setg(&local_err,
3641                    "Memory unplug already in progress for device %s",
3642                    dev->id);
3643         goto out;
3644     }
3645 
3646     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3647 
3648     addr = addr_start;
3649     for (i = 0; i < nr_lmbs; i++) {
3650         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3651                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3652         g_assert(drc);
3653 
3654         spapr_drc_detach(drc);
3655         addr += SPAPR_MEMORY_BLOCK_SIZE;
3656     }
3657 
3658     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3659                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3660     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3661                                               nr_lmbs, spapr_drc_index(drc));
3662 out:
3663     error_propagate(errp, local_err);
3664 }
3665 
3666 /* Callback to be called during DRC release. */
3667 void spapr_core_release(DeviceState *dev)
3668 {
3669     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3670 
3671     /* Call the unplug handler chain. This can never fail. */
3672     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3673     object_unparent(OBJECT(dev));
3674 }
3675 
3676 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3677 {
3678     MachineState *ms = MACHINE(hotplug_dev);
3679     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3680     CPUCore *cc = CPU_CORE(dev);
3681     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3682 
3683     if (smc->pre_2_10_has_unused_icps) {
3684         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3685         int i;
3686 
3687         for (i = 0; i < cc->nr_threads; i++) {
3688             CPUState *cs = CPU(sc->threads[i]);
3689 
3690             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3691         }
3692     }
3693 
3694     assert(core_slot);
3695     core_slot->cpu = NULL;
3696     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3697 }
3698 
3699 static
3700 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3701                                Error **errp)
3702 {
3703     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3704     int index;
3705     SpaprDrc *drc;
3706     CPUCore *cc = CPU_CORE(dev);
3707 
3708     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3709         error_setg(errp, "Unable to find CPU core with core-id: %d",
3710                    cc->core_id);
3711         return;
3712     }
3713     if (index == 0) {
3714         error_setg(errp, "Boot CPU core may not be unplugged");
3715         return;
3716     }
3717 
3718     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3719                           spapr_vcpu_id(spapr, cc->core_id));
3720     g_assert(drc);
3721 
3722     if (!spapr_drc_unplug_requested(drc)) {
3723         spapr_drc_detach(drc);
3724         spapr_hotplug_req_remove_by_index(drc);
3725     }
3726 }
3727 
3728 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3729                            void *fdt, int *fdt_start_offset, Error **errp)
3730 {
3731     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3732     CPUState *cs = CPU(core->threads[0]);
3733     PowerPCCPU *cpu = POWERPC_CPU(cs);
3734     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3735     int id = spapr_get_vcpu_id(cpu);
3736     char *nodename;
3737     int offset;
3738 
3739     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3740     offset = fdt_add_subnode(fdt, 0, nodename);
3741     g_free(nodename);
3742 
3743     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3744 
3745     *fdt_start_offset = offset;
3746     return 0;
3747 }
3748 
3749 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3750                             Error **errp)
3751 {
3752     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3753     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3754     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3755     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3756     CPUCore *cc = CPU_CORE(dev);
3757     CPUState *cs;
3758     SpaprDrc *drc;
3759     Error *local_err = NULL;
3760     CPUArchId *core_slot;
3761     int index;
3762     bool hotplugged = spapr_drc_hotplugged(dev);
3763     int i;
3764 
3765     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3766     if (!core_slot) {
3767         error_setg(errp, "Unable to find CPU core with core-id: %d",
3768                    cc->core_id);
3769         return;
3770     }
3771     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3772                           spapr_vcpu_id(spapr, cc->core_id));
3773 
3774     g_assert(drc || !mc->has_hotpluggable_cpus);
3775 
3776     if (drc) {
3777         spapr_drc_attach(drc, dev, &local_err);
3778         if (local_err) {
3779             error_propagate(errp, local_err);
3780             return;
3781         }
3782 
3783         if (hotplugged) {
3784             /*
3785              * Send hotplug notification interrupt to the guest only
3786              * in case of hotplugged CPUs.
3787              */
3788             spapr_hotplug_req_add_by_index(drc);
3789         } else {
3790             spapr_drc_reset(drc);
3791         }
3792     }
3793 
3794     core_slot->cpu = OBJECT(dev);
3795 
3796     if (smc->pre_2_10_has_unused_icps) {
3797         for (i = 0; i < cc->nr_threads; i++) {
3798             cs = CPU(core->threads[i]);
3799             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3800         }
3801     }
3802 
3803     /*
3804      * Set compatibility mode to match the boot CPU, which was either set
3805      * by the machine reset code or by CAS.
3806      */
3807     if (hotplugged) {
3808         for (i = 0; i < cc->nr_threads; i++) {
3809             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3810                            &local_err);
3811             if (local_err) {
3812                 error_propagate(errp, local_err);
3813                 return;
3814             }
3815         }
3816     }
3817 }
3818 
3819 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3820                                 Error **errp)
3821 {
3822     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3823     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3824     Error *local_err = NULL;
3825     CPUCore *cc = CPU_CORE(dev);
3826     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3827     const char *type = object_get_typename(OBJECT(dev));
3828     CPUArchId *core_slot;
3829     int index;
3830     unsigned int smp_threads = machine->smp.threads;
3831 
3832     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3833         error_setg(&local_err, "CPU hotplug not supported for this machine");
3834         goto out;
3835     }
3836 
3837     if (strcmp(base_core_type, type)) {
3838         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3839         goto out;
3840     }
3841 
3842     if (cc->core_id % smp_threads) {
3843         error_setg(&local_err, "invalid core id %d", cc->core_id);
3844         goto out;
3845     }
3846 
3847     /*
3848      * In general we should have homogeneous threads-per-core, but old
3849      * (pre hotplug support) machine types allow the last core to have
3850      * reduced threads as a compatibility hack for when we allowed
3851      * total vcpus not a multiple of threads-per-core.
3852      */
3853     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3854         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3855                    cc->nr_threads, smp_threads);
3856         goto out;
3857     }
3858 
3859     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3860     if (!core_slot) {
3861         error_setg(&local_err, "core id %d out of range", cc->core_id);
3862         goto out;
3863     }
3864 
3865     if (core_slot->cpu) {
3866         error_setg(&local_err, "core %d already populated", cc->core_id);
3867         goto out;
3868     }
3869 
3870     numa_cpu_pre_plug(core_slot, dev, &local_err);
3871 
3872 out:
3873     error_propagate(errp, local_err);
3874 }
3875 
3876 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3877                           void *fdt, int *fdt_start_offset, Error **errp)
3878 {
3879     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3880     int intc_phandle;
3881 
3882     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3883     if (intc_phandle <= 0) {
3884         return -1;
3885     }
3886 
3887     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3888         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3889         return -1;
3890     }
3891 
3892     /* generally SLOF creates these, for hotplug it's up to QEMU */
3893     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3894 
3895     return 0;
3896 }
3897 
3898 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3899                                Error **errp)
3900 {
3901     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3902     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3903     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3904     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3905 
3906     if (dev->hotplugged && !smc->dr_phb_enabled) {
3907         error_setg(errp, "PHB hotplug not supported for this machine");
3908         return;
3909     }
3910 
3911     if (sphb->index == (uint32_t)-1) {
3912         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3913         return;
3914     }
3915 
3916     /*
3917      * This will check that sphb->index doesn't exceed the maximum number of
3918      * PHBs for the current machine type.
3919      */
3920     smc->phb_placement(spapr, sphb->index,
3921                        &sphb->buid, &sphb->io_win_addr,
3922                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3923                        windows_supported, sphb->dma_liobn,
3924                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3925                        errp);
3926 }
3927 
3928 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3929                            Error **errp)
3930 {
3931     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3932     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3933     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3934     SpaprDrc *drc;
3935     bool hotplugged = spapr_drc_hotplugged(dev);
3936     Error *local_err = NULL;
3937 
3938     if (!smc->dr_phb_enabled) {
3939         return;
3940     }
3941 
3942     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3943     /* hotplug hooks should check it's enabled before getting this far */
3944     assert(drc);
3945 
3946     spapr_drc_attach(drc, DEVICE(dev), &local_err);
3947     if (local_err) {
3948         error_propagate(errp, local_err);
3949         return;
3950     }
3951 
3952     if (hotplugged) {
3953         spapr_hotplug_req_add_by_index(drc);
3954     } else {
3955         spapr_drc_reset(drc);
3956     }
3957 }
3958 
3959 void spapr_phb_release(DeviceState *dev)
3960 {
3961     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3962 
3963     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3964     object_unparent(OBJECT(dev));
3965 }
3966 
3967 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3968 {
3969     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3970 }
3971 
3972 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3973                                      DeviceState *dev, Error **errp)
3974 {
3975     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3976     SpaprDrc *drc;
3977 
3978     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3979     assert(drc);
3980 
3981     if (!spapr_drc_unplug_requested(drc)) {
3982         spapr_drc_detach(drc);
3983         spapr_hotplug_req_remove_by_index(drc);
3984     }
3985 }
3986 
3987 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3988                                  Error **errp)
3989 {
3990     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3991     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
3992 
3993     if (spapr->tpm_proxy != NULL) {
3994         error_setg(errp, "Only one TPM proxy can be specified for this machine");
3995         return;
3996     }
3997 
3998     spapr->tpm_proxy = tpm_proxy;
3999 }
4000 
4001 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4002 {
4003     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4004 
4005     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4006     object_unparent(OBJECT(dev));
4007     spapr->tpm_proxy = NULL;
4008 }
4009 
4010 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4011                                       DeviceState *dev, Error **errp)
4012 {
4013     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4014         spapr_memory_plug(hotplug_dev, dev, errp);
4015     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4016         spapr_core_plug(hotplug_dev, dev, errp);
4017     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4018         spapr_phb_plug(hotplug_dev, dev, errp);
4019     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4020         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4021     }
4022 }
4023 
4024 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4025                                         DeviceState *dev, Error **errp)
4026 {
4027     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4028         spapr_memory_unplug(hotplug_dev, dev);
4029     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4030         spapr_core_unplug(hotplug_dev, dev);
4031     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4032         spapr_phb_unplug(hotplug_dev, dev);
4033     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4034         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4035     }
4036 }
4037 
4038 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4039                                                 DeviceState *dev, Error **errp)
4040 {
4041     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4042     MachineClass *mc = MACHINE_GET_CLASS(sms);
4043     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4044 
4045     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4046         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4047             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4048         } else {
4049             /* NOTE: this means there is a window after guest reset, prior to
4050              * CAS negotiation, where unplug requests will fail due to the
4051              * capability not being detected yet. This is a bit different than
4052              * the case with PCI unplug, where the events will be queued and
4053              * eventually handled by the guest after boot
4054              */
4055             error_setg(errp, "Memory hot unplug not supported for this guest");
4056         }
4057     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4058         if (!mc->has_hotpluggable_cpus) {
4059             error_setg(errp, "CPU hot unplug not supported on this machine");
4060             return;
4061         }
4062         spapr_core_unplug_request(hotplug_dev, dev, errp);
4063     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4064         if (!smc->dr_phb_enabled) {
4065             error_setg(errp, "PHB hot unplug not supported on this machine");
4066             return;
4067         }
4068         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4069     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4070         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4071     }
4072 }
4073 
4074 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4075                                           DeviceState *dev, Error **errp)
4076 {
4077     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4078         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4079     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4080         spapr_core_pre_plug(hotplug_dev, dev, errp);
4081     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4082         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4083     }
4084 }
4085 
4086 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4087                                                  DeviceState *dev)
4088 {
4089     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4090         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4091         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4092         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4093         return HOTPLUG_HANDLER(machine);
4094     }
4095     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4096         PCIDevice *pcidev = PCI_DEVICE(dev);
4097         PCIBus *root = pci_device_root_bus(pcidev);
4098         SpaprPhbState *phb =
4099             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4100                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4101 
4102         if (phb) {
4103             return HOTPLUG_HANDLER(phb);
4104         }
4105     }
4106     return NULL;
4107 }
4108 
4109 static CpuInstanceProperties
4110 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4111 {
4112     CPUArchId *core_slot;
4113     MachineClass *mc = MACHINE_GET_CLASS(machine);
4114 
4115     /* make sure possible_cpu are intialized */
4116     mc->possible_cpu_arch_ids(machine);
4117     /* get CPU core slot containing thread that matches cpu_index */
4118     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4119     assert(core_slot);
4120     return core_slot->props;
4121 }
4122 
4123 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4124 {
4125     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4126 }
4127 
4128 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4129 {
4130     int i;
4131     unsigned int smp_threads = machine->smp.threads;
4132     unsigned int smp_cpus = machine->smp.cpus;
4133     const char *core_type;
4134     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4135     MachineClass *mc = MACHINE_GET_CLASS(machine);
4136 
4137     if (!mc->has_hotpluggable_cpus) {
4138         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4139     }
4140     if (machine->possible_cpus) {
4141         assert(machine->possible_cpus->len == spapr_max_cores);
4142         return machine->possible_cpus;
4143     }
4144 
4145     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4146     if (!core_type) {
4147         error_report("Unable to find sPAPR CPU Core definition");
4148         exit(1);
4149     }
4150 
4151     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4152                              sizeof(CPUArchId) * spapr_max_cores);
4153     machine->possible_cpus->len = spapr_max_cores;
4154     for (i = 0; i < machine->possible_cpus->len; i++) {
4155         int core_id = i * smp_threads;
4156 
4157         machine->possible_cpus->cpus[i].type = core_type;
4158         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4159         machine->possible_cpus->cpus[i].arch_id = core_id;
4160         machine->possible_cpus->cpus[i].props.has_core_id = true;
4161         machine->possible_cpus->cpus[i].props.core_id = core_id;
4162     }
4163     return machine->possible_cpus;
4164 }
4165 
4166 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4167                                 uint64_t *buid, hwaddr *pio,
4168                                 hwaddr *mmio32, hwaddr *mmio64,
4169                                 unsigned n_dma, uint32_t *liobns,
4170                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4171 {
4172     /*
4173      * New-style PHB window placement.
4174      *
4175      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4176      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4177      * windows.
4178      *
4179      * Some guest kernels can't work with MMIO windows above 1<<46
4180      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4181      *
4182      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4183      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4184      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4185      * 1TiB 64-bit MMIO windows for each PHB.
4186      */
4187     const uint64_t base_buid = 0x800000020000000ULL;
4188     int i;
4189 
4190     /* Sanity check natural alignments */
4191     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4192     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4193     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4194     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4195     /* Sanity check bounds */
4196     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4197                       SPAPR_PCI_MEM32_WIN_SIZE);
4198     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4199                       SPAPR_PCI_MEM64_WIN_SIZE);
4200 
4201     if (index >= SPAPR_MAX_PHBS) {
4202         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4203                    SPAPR_MAX_PHBS - 1);
4204         return;
4205     }
4206 
4207     *buid = base_buid + index;
4208     for (i = 0; i < n_dma; ++i) {
4209         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4210     }
4211 
4212     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4213     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4214     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4215 
4216     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4217     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4218 }
4219 
4220 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4221 {
4222     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4223 
4224     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4225 }
4226 
4227 static void spapr_ics_resend(XICSFabric *dev)
4228 {
4229     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4230 
4231     ics_resend(spapr->ics);
4232 }
4233 
4234 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4235 {
4236     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4237 
4238     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4239 }
4240 
4241 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4242                                  Monitor *mon)
4243 {
4244     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4245 
4246     spapr_irq_print_info(spapr, mon);
4247     monitor_printf(mon, "irqchip: %s\n",
4248                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4249 }
4250 
4251 /*
4252  * This is a XIVE only operation
4253  */
4254 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4255                            uint8_t nvt_blk, uint32_t nvt_idx,
4256                            bool cam_ignore, uint8_t priority,
4257                            uint32_t logic_serv, XiveTCTXMatch *match)
4258 {
4259     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4260     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4261     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4262     int count;
4263 
4264     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4265                            priority, logic_serv, match);
4266     if (count < 0) {
4267         return count;
4268     }
4269 
4270     /*
4271      * When we implement the save and restore of the thread interrupt
4272      * contexts in the enter/exit CPU handlers of the machine and the
4273      * escalations in QEMU, we should be able to handle non dispatched
4274      * vCPUs.
4275      *
4276      * Until this is done, the sPAPR machine should find at least one
4277      * matching context always.
4278      */
4279     if (count == 0) {
4280         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4281                       nvt_blk, nvt_idx);
4282     }
4283 
4284     return count;
4285 }
4286 
4287 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4288 {
4289     return cpu->vcpu_id;
4290 }
4291 
4292 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4293 {
4294     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4295     MachineState *ms = MACHINE(spapr);
4296     int vcpu_id;
4297 
4298     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4299 
4300     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4301         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4302         error_append_hint(errp, "Adjust the number of cpus to %d "
4303                           "or try to raise the number of threads per core\n",
4304                           vcpu_id * ms->smp.threads / spapr->vsmt);
4305         return;
4306     }
4307 
4308     cpu->vcpu_id = vcpu_id;
4309 }
4310 
4311 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4312 {
4313     CPUState *cs;
4314 
4315     CPU_FOREACH(cs) {
4316         PowerPCCPU *cpu = POWERPC_CPU(cs);
4317 
4318         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4319             return cpu;
4320         }
4321     }
4322 
4323     return NULL;
4324 }
4325 
4326 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4327 {
4328     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4329 
4330     /* These are only called by TCG, KVM maintains dispatch state */
4331 
4332     spapr_cpu->prod = false;
4333     if (spapr_cpu->vpa_addr) {
4334         CPUState *cs = CPU(cpu);
4335         uint32_t dispatch;
4336 
4337         dispatch = ldl_be_phys(cs->as,
4338                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4339         dispatch++;
4340         if ((dispatch & 1) != 0) {
4341             qemu_log_mask(LOG_GUEST_ERROR,
4342                           "VPA: incorrect dispatch counter value for "
4343                           "dispatched partition %u, correcting.\n", dispatch);
4344             dispatch++;
4345         }
4346         stl_be_phys(cs->as,
4347                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4348     }
4349 }
4350 
4351 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4352 {
4353     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4354 
4355     if (spapr_cpu->vpa_addr) {
4356         CPUState *cs = CPU(cpu);
4357         uint32_t dispatch;
4358 
4359         dispatch = ldl_be_phys(cs->as,
4360                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4361         dispatch++;
4362         if ((dispatch & 1) != 1) {
4363             qemu_log_mask(LOG_GUEST_ERROR,
4364                           "VPA: incorrect dispatch counter value for "
4365                           "preempted partition %u, correcting.\n", dispatch);
4366             dispatch++;
4367         }
4368         stl_be_phys(cs->as,
4369                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4370     }
4371 }
4372 
4373 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4374 {
4375     MachineClass *mc = MACHINE_CLASS(oc);
4376     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4377     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4378     NMIClass *nc = NMI_CLASS(oc);
4379     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4380     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4381     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4382     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4383     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4384 
4385     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4386     mc->ignore_boot_device_suffixes = true;
4387 
4388     /*
4389      * We set up the default / latest behaviour here.  The class_init
4390      * functions for the specific versioned machine types can override
4391      * these details for backwards compatibility
4392      */
4393     mc->init = spapr_machine_init;
4394     mc->reset = spapr_machine_reset;
4395     mc->block_default_type = IF_SCSI;
4396     mc->max_cpus = 1024;
4397     mc->no_parallel = 1;
4398     mc->default_boot_order = "";
4399     mc->default_ram_size = 512 * MiB;
4400     mc->default_ram_id = "ppc_spapr.ram";
4401     mc->default_display = "std";
4402     mc->kvm_type = spapr_kvm_type;
4403     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4404     mc->pci_allow_0_address = true;
4405     assert(!mc->get_hotplug_handler);
4406     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4407     hc->pre_plug = spapr_machine_device_pre_plug;
4408     hc->plug = spapr_machine_device_plug;
4409     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4410     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4411     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4412     hc->unplug_request = spapr_machine_device_unplug_request;
4413     hc->unplug = spapr_machine_device_unplug;
4414 
4415     smc->dr_lmb_enabled = true;
4416     smc->update_dt_enabled = true;
4417     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4418     mc->has_hotpluggable_cpus = true;
4419     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4420     fwc->get_dev_path = spapr_get_fw_dev_path;
4421     nc->nmi_monitor_handler = spapr_nmi;
4422     smc->phb_placement = spapr_phb_placement;
4423     vhc->hypercall = emulate_spapr_hypercall;
4424     vhc->hpt_mask = spapr_hpt_mask;
4425     vhc->map_hptes = spapr_map_hptes;
4426     vhc->unmap_hptes = spapr_unmap_hptes;
4427     vhc->hpte_set_c = spapr_hpte_set_c;
4428     vhc->hpte_set_r = spapr_hpte_set_r;
4429     vhc->get_pate = spapr_get_pate;
4430     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4431     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4432     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4433     xic->ics_get = spapr_ics_get;
4434     xic->ics_resend = spapr_ics_resend;
4435     xic->icp_get = spapr_icp_get;
4436     ispc->print_info = spapr_pic_print_info;
4437     /* Force NUMA node memory size to be a multiple of
4438      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4439      * in which LMBs are represented and hot-added
4440      */
4441     mc->numa_mem_align_shift = 28;
4442     mc->numa_mem_supported = true;
4443     mc->auto_enable_numa = true;
4444 
4445     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4446     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4447     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4448     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4449     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4450     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4451     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4452     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4453     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4454     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4455     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_ON;
4456     spapr_caps_add_properties(smc, &error_abort);
4457     smc->irq = &spapr_irq_dual;
4458     smc->dr_phb_enabled = true;
4459     smc->linux_pci_probe = true;
4460     smc->smp_threads_vsmt = true;
4461     smc->nr_xirqs = SPAPR_NR_XIRQS;
4462     xfc->match_nvt = spapr_match_nvt;
4463 }
4464 
4465 static const TypeInfo spapr_machine_info = {
4466     .name          = TYPE_SPAPR_MACHINE,
4467     .parent        = TYPE_MACHINE,
4468     .abstract      = true,
4469     .instance_size = sizeof(SpaprMachineState),
4470     .instance_init = spapr_instance_init,
4471     .instance_finalize = spapr_machine_finalizefn,
4472     .class_size    = sizeof(SpaprMachineClass),
4473     .class_init    = spapr_machine_class_init,
4474     .interfaces = (InterfaceInfo[]) {
4475         { TYPE_FW_PATH_PROVIDER },
4476         { TYPE_NMI },
4477         { TYPE_HOTPLUG_HANDLER },
4478         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4479         { TYPE_XICS_FABRIC },
4480         { TYPE_INTERRUPT_STATS_PROVIDER },
4481         { TYPE_XIVE_FABRIC },
4482         { }
4483     },
4484 };
4485 
4486 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4487     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4488                                                     void *data)      \
4489     {                                                                \
4490         MachineClass *mc = MACHINE_CLASS(oc);                        \
4491         spapr_machine_##suffix##_class_options(mc);                  \
4492         if (latest) {                                                \
4493             mc->alias = "pseries";                                   \
4494             mc->is_default = 1;                                      \
4495         }                                                            \
4496     }                                                                \
4497     static const TypeInfo spapr_machine_##suffix##_info = {          \
4498         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4499         .parent = TYPE_SPAPR_MACHINE,                                \
4500         .class_init = spapr_machine_##suffix##_class_init,           \
4501     };                                                               \
4502     static void spapr_machine_register_##suffix(void)                \
4503     {                                                                \
4504         type_register(&spapr_machine_##suffix##_info);               \
4505     }                                                                \
4506     type_init(spapr_machine_register_##suffix)
4507 
4508 /*
4509  * pseries-5.0
4510  */
4511 static void spapr_machine_5_0_class_options(MachineClass *mc)
4512 {
4513     /* Defaults for the latest behaviour inherited from the base class */
4514 }
4515 
4516 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4517 
4518 /*
4519  * pseries-4.2
4520  */
4521 static void spapr_machine_4_2_class_options(MachineClass *mc)
4522 {
4523     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4524 
4525     spapr_machine_5_0_class_options(mc);
4526     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4527     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4528     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_OFF;
4529 }
4530 
4531 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4532 
4533 /*
4534  * pseries-4.1
4535  */
4536 static void spapr_machine_4_1_class_options(MachineClass *mc)
4537 {
4538     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4539     static GlobalProperty compat[] = {
4540         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4541         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4542     };
4543 
4544     spapr_machine_4_2_class_options(mc);
4545     smc->linux_pci_probe = false;
4546     smc->smp_threads_vsmt = false;
4547     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4548     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4549 }
4550 
4551 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4552 
4553 /*
4554  * pseries-4.0
4555  */
4556 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4557                               uint64_t *buid, hwaddr *pio,
4558                               hwaddr *mmio32, hwaddr *mmio64,
4559                               unsigned n_dma, uint32_t *liobns,
4560                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4561 {
4562     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4563                         nv2gpa, nv2atsd, errp);
4564     *nv2gpa = 0;
4565     *nv2atsd = 0;
4566 }
4567 
4568 static void spapr_machine_4_0_class_options(MachineClass *mc)
4569 {
4570     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4571 
4572     spapr_machine_4_1_class_options(mc);
4573     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4574     smc->phb_placement = phb_placement_4_0;
4575     smc->irq = &spapr_irq_xics;
4576     smc->pre_4_1_migration = true;
4577 }
4578 
4579 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4580 
4581 /*
4582  * pseries-3.1
4583  */
4584 static void spapr_machine_3_1_class_options(MachineClass *mc)
4585 {
4586     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4587 
4588     spapr_machine_4_0_class_options(mc);
4589     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4590 
4591     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4592     smc->update_dt_enabled = false;
4593     smc->dr_phb_enabled = false;
4594     smc->broken_host_serial_model = true;
4595     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4596     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4597     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4598     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4599 }
4600 
4601 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4602 
4603 /*
4604  * pseries-3.0
4605  */
4606 
4607 static void spapr_machine_3_0_class_options(MachineClass *mc)
4608 {
4609     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4610 
4611     spapr_machine_3_1_class_options(mc);
4612     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4613 
4614     smc->legacy_irq_allocation = true;
4615     smc->nr_xirqs = 0x400;
4616     smc->irq = &spapr_irq_xics_legacy;
4617 }
4618 
4619 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4620 
4621 /*
4622  * pseries-2.12
4623  */
4624 static void spapr_machine_2_12_class_options(MachineClass *mc)
4625 {
4626     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4627     static GlobalProperty compat[] = {
4628         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4629         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4630     };
4631 
4632     spapr_machine_3_0_class_options(mc);
4633     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4634     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4635 
4636     /* We depend on kvm_enabled() to choose a default value for the
4637      * hpt-max-page-size capability. Of course we can't do it here
4638      * because this is too early and the HW accelerator isn't initialzed
4639      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4640      */
4641     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4642 }
4643 
4644 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4645 
4646 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4647 {
4648     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4649 
4650     spapr_machine_2_12_class_options(mc);
4651     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4652     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4653     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4654 }
4655 
4656 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4657 
4658 /*
4659  * pseries-2.11
4660  */
4661 
4662 static void spapr_machine_2_11_class_options(MachineClass *mc)
4663 {
4664     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4665 
4666     spapr_machine_2_12_class_options(mc);
4667     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4668     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4669 }
4670 
4671 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4672 
4673 /*
4674  * pseries-2.10
4675  */
4676 
4677 static void spapr_machine_2_10_class_options(MachineClass *mc)
4678 {
4679     spapr_machine_2_11_class_options(mc);
4680     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4681 }
4682 
4683 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4684 
4685 /*
4686  * pseries-2.9
4687  */
4688 
4689 static void spapr_machine_2_9_class_options(MachineClass *mc)
4690 {
4691     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4692     static GlobalProperty compat[] = {
4693         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4694     };
4695 
4696     spapr_machine_2_10_class_options(mc);
4697     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4698     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4699     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4700     smc->pre_2_10_has_unused_icps = true;
4701     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4702 }
4703 
4704 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4705 
4706 /*
4707  * pseries-2.8
4708  */
4709 
4710 static void spapr_machine_2_8_class_options(MachineClass *mc)
4711 {
4712     static GlobalProperty compat[] = {
4713         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4714     };
4715 
4716     spapr_machine_2_9_class_options(mc);
4717     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4718     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4719     mc->numa_mem_align_shift = 23;
4720 }
4721 
4722 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4723 
4724 /*
4725  * pseries-2.7
4726  */
4727 
4728 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4729                               uint64_t *buid, hwaddr *pio,
4730                               hwaddr *mmio32, hwaddr *mmio64,
4731                               unsigned n_dma, uint32_t *liobns,
4732                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4733 {
4734     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4735     const uint64_t base_buid = 0x800000020000000ULL;
4736     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4737     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4738     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4739     const uint32_t max_index = 255;
4740     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4741 
4742     uint64_t ram_top = MACHINE(spapr)->ram_size;
4743     hwaddr phb0_base, phb_base;
4744     int i;
4745 
4746     /* Do we have device memory? */
4747     if (MACHINE(spapr)->maxram_size > ram_top) {
4748         /* Can't just use maxram_size, because there may be an
4749          * alignment gap between normal and device memory regions
4750          */
4751         ram_top = MACHINE(spapr)->device_memory->base +
4752             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4753     }
4754 
4755     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4756 
4757     if (index > max_index) {
4758         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4759                    max_index);
4760         return;
4761     }
4762 
4763     *buid = base_buid + index;
4764     for (i = 0; i < n_dma; ++i) {
4765         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4766     }
4767 
4768     phb_base = phb0_base + index * phb_spacing;
4769     *pio = phb_base + pio_offset;
4770     *mmio32 = phb_base + mmio_offset;
4771     /*
4772      * We don't set the 64-bit MMIO window, relying on the PHB's
4773      * fallback behaviour of automatically splitting a large "32-bit"
4774      * window into contiguous 32-bit and 64-bit windows
4775      */
4776 
4777     *nv2gpa = 0;
4778     *nv2atsd = 0;
4779 }
4780 
4781 static void spapr_machine_2_7_class_options(MachineClass *mc)
4782 {
4783     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4784     static GlobalProperty compat[] = {
4785         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4786         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4787         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4788         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4789     };
4790 
4791     spapr_machine_2_8_class_options(mc);
4792     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4793     mc->default_machine_opts = "modern-hotplug-events=off";
4794     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4795     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4796     smc->phb_placement = phb_placement_2_7;
4797 }
4798 
4799 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4800 
4801 /*
4802  * pseries-2.6
4803  */
4804 
4805 static void spapr_machine_2_6_class_options(MachineClass *mc)
4806 {
4807     static GlobalProperty compat[] = {
4808         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4809     };
4810 
4811     spapr_machine_2_7_class_options(mc);
4812     mc->has_hotpluggable_cpus = false;
4813     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4814     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4815 }
4816 
4817 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4818 
4819 /*
4820  * pseries-2.5
4821  */
4822 
4823 static void spapr_machine_2_5_class_options(MachineClass *mc)
4824 {
4825     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4826     static GlobalProperty compat[] = {
4827         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4828     };
4829 
4830     spapr_machine_2_6_class_options(mc);
4831     smc->use_ohci_by_default = true;
4832     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4833     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4834 }
4835 
4836 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4837 
4838 /*
4839  * pseries-2.4
4840  */
4841 
4842 static void spapr_machine_2_4_class_options(MachineClass *mc)
4843 {
4844     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4845 
4846     spapr_machine_2_5_class_options(mc);
4847     smc->dr_lmb_enabled = false;
4848     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4849 }
4850 
4851 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4852 
4853 /*
4854  * pseries-2.3
4855  */
4856 
4857 static void spapr_machine_2_3_class_options(MachineClass *mc)
4858 {
4859     static GlobalProperty compat[] = {
4860         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4861     };
4862     spapr_machine_2_4_class_options(mc);
4863     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4864     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4865 }
4866 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4867 
4868 /*
4869  * pseries-2.2
4870  */
4871 
4872 static void spapr_machine_2_2_class_options(MachineClass *mc)
4873 {
4874     static GlobalProperty compat[] = {
4875         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4876     };
4877 
4878     spapr_machine_2_3_class_options(mc);
4879     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4880     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4881     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4882 }
4883 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4884 
4885 /*
4886  * pseries-2.1
4887  */
4888 
4889 static void spapr_machine_2_1_class_options(MachineClass *mc)
4890 {
4891     spapr_machine_2_2_class_options(mc);
4892     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4893 }
4894 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4895 
4896 static void spapr_machine_register_types(void)
4897 {
4898     type_register_static(&spapr_machine_info);
4899 }
4900 
4901 type_init(spapr_machine_register_types)
4902