xref: /openbmc/qemu/hw/ppc/spapr.c (revision 3607715a)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "mmu-hash64.h"
50 #include "mmu-book3s-v3.h"
51 #include "cpu-models.h"
52 #include "hw/core/cpu.h"
53 
54 #include "hw/boards.h"
55 #include "hw/ppc/ppc.h"
56 #include "hw/loader.h"
57 
58 #include "hw/ppc/fdt.h"
59 #include "hw/ppc/spapr.h"
60 #include "hw/ppc/spapr_vio.h"
61 #include "hw/qdev-properties.h"
62 #include "hw/pci-host/spapr.h"
63 #include "hw/pci/msi.h"
64 
65 #include "hw/pci/pci.h"
66 #include "hw/scsi/scsi.h"
67 #include "hw/virtio/virtio-scsi.h"
68 #include "hw/virtio/vhost-scsi-common.h"
69 
70 #include "exec/address-spaces.h"
71 #include "exec/ram_addr.h"
72 #include "hw/usb.h"
73 #include "qemu/config-file.h"
74 #include "qemu/error-report.h"
75 #include "trace.h"
76 #include "hw/nmi.h"
77 #include "hw/intc/intc.h"
78 
79 #include "qemu/cutils.h"
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 
84 #include "monitor/monitor.h"
85 
86 #include <libfdt.h>
87 
88 /* SLOF memory layout:
89  *
90  * SLOF raw image loaded at 0, copies its romfs right below the flat
91  * device-tree, then position SLOF itself 31M below that
92  *
93  * So we set FW_OVERHEAD to 40MB which should account for all of that
94  * and more
95  *
96  * We load our kernel at 4M, leaving space for SLOF initial image
97  */
98 #define FDT_MAX_SIZE            0x100000
99 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
100 #define FW_MAX_SIZE             0x400000
101 #define FW_FILE_NAME            "slof.bin"
102 #define FW_OVERHEAD             0x2800000
103 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
104 
105 #define MIN_RMA_SLOF            128UL
106 
107 #define PHANDLE_INTC            0x00001111
108 
109 /* These two functions implement the VCPU id numbering: one to compute them
110  * all and one to identify thread 0 of a VCORE. Any change to the first one
111  * is likely to have an impact on the second one, so let's keep them close.
112  */
113 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
114 {
115     MachineState *ms = MACHINE(spapr);
116     unsigned int smp_threads = ms->smp.threads;
117 
118     assert(spapr->vsmt);
119     return
120         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
121 }
122 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
123                                       PowerPCCPU *cpu)
124 {
125     assert(spapr->vsmt);
126     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 int spapr_max_server_number(SpaprMachineState *spapr)
164 {
165     MachineState *ms = MACHINE(spapr);
166 
167     assert(spapr->vsmt);
168     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
169 }
170 
171 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
172                                   int smt_threads)
173 {
174     int i, ret = 0;
175     uint32_t servers_prop[smt_threads];
176     uint32_t gservers_prop[smt_threads * 2];
177     int index = spapr_get_vcpu_id(cpu);
178 
179     if (cpu->compat_pvr) {
180         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
181         if (ret < 0) {
182             return ret;
183         }
184     }
185 
186     /* Build interrupt servers and gservers properties */
187     for (i = 0; i < smt_threads; i++) {
188         servers_prop[i] = cpu_to_be32(index + i);
189         /* Hack, direct the group queues back to cpu 0 */
190         gservers_prop[i*2] = cpu_to_be32(index + i);
191         gservers_prop[i*2 + 1] = 0;
192     }
193     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
194                       servers_prop, sizeof(servers_prop));
195     if (ret < 0) {
196         return ret;
197     }
198     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
199                       gservers_prop, sizeof(gservers_prop));
200 
201     return ret;
202 }
203 
204 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
205 {
206     int index = spapr_get_vcpu_id(cpu);
207     uint32_t associativity[] = {cpu_to_be32(0x5),
208                                 cpu_to_be32(0x0),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(cpu->node_id),
212                                 cpu_to_be32(index)};
213 
214     /* Advertise NUMA via ibm,associativity */
215     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
216                           sizeof(associativity));
217 }
218 
219 /* Populate the "ibm,pa-features" property */
220 static void spapr_populate_pa_features(SpaprMachineState *spapr,
221                                        PowerPCCPU *cpu,
222                                        void *fdt, int offset)
223 {
224     uint8_t pa_features_206[] = { 6, 0,
225         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
226     uint8_t pa_features_207[] = { 24, 0,
227         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
228         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
229         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
230         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
231     uint8_t pa_features_300[] = { 66, 0,
232         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
233         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
234         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
235         /* 6: DS207 */
236         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
237         /* 16: Vector */
238         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
239         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
241         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
242         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
243         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
244         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
245         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
246         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
247         /* 42: PM, 44: PC RA, 46: SC vec'd */
248         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
249         /* 48: SIMD, 50: QP BFP, 52: String */
250         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
251         /* 54: DecFP, 56: DecI, 58: SHA */
252         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
253         /* 60: NM atomic, 62: RNG */
254         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
255     };
256     uint8_t *pa_features = NULL;
257     size_t pa_size;
258 
259     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
260         pa_features = pa_features_206;
261         pa_size = sizeof(pa_features_206);
262     }
263     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
264         pa_features = pa_features_207;
265         pa_size = sizeof(pa_features_207);
266     }
267     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
268         pa_features = pa_features_300;
269         pa_size = sizeof(pa_features_300);
270     }
271     if (!pa_features) {
272         return;
273     }
274 
275     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
276         /*
277          * Note: we keep CI large pages off by default because a 64K capable
278          * guest provisioned with large pages might otherwise try to map a qemu
279          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
280          * even if that qemu runs on a 4k host.
281          * We dd this bit back here if we are confident this is not an issue
282          */
283         pa_features[3] |= 0x20;
284     }
285     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
286         pa_features[24] |= 0x80;    /* Transactional memory support */
287     }
288     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
289         /* Workaround for broken kernels that attempt (guest) radix
290          * mode when they can't handle it, if they see the radix bit set
291          * in pa-features. So hide it from them. */
292         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
293     }
294 
295     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
296 }
297 
298 static hwaddr spapr_node0_size(MachineState *machine)
299 {
300     if (machine->numa_state->num_nodes) {
301         int i;
302         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
303             if (machine->numa_state->nodes[i].node_mem) {
304                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
305                            machine->ram_size);
306             }
307         }
308     }
309     return machine->ram_size;
310 }
311 
312 static void add_str(GString *s, const gchar *s1)
313 {
314     g_string_append_len(s, s1, strlen(s1) + 1);
315 }
316 
317 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
318                                        hwaddr size)
319 {
320     uint32_t associativity[] = {
321         cpu_to_be32(0x4), /* length */
322         cpu_to_be32(0x0), cpu_to_be32(0x0),
323         cpu_to_be32(0x0), cpu_to_be32(nodeid)
324     };
325     char mem_name[32];
326     uint64_t mem_reg_property[2];
327     int off;
328 
329     mem_reg_property[0] = cpu_to_be64(start);
330     mem_reg_property[1] = cpu_to_be64(size);
331 
332     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
333     off = fdt_add_subnode(fdt, 0, mem_name);
334     _FDT(off);
335     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
336     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
337                       sizeof(mem_reg_property))));
338     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
339                       sizeof(associativity))));
340     return off;
341 }
342 
343 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
344 {
345     MachineState *machine = MACHINE(spapr);
346     hwaddr mem_start, node_size;
347     int i, nb_nodes = machine->numa_state->num_nodes;
348     NodeInfo *nodes = machine->numa_state->nodes;
349 
350     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
351         if (!nodes[i].node_mem) {
352             continue;
353         }
354         if (mem_start >= machine->ram_size) {
355             node_size = 0;
356         } else {
357             node_size = nodes[i].node_mem;
358             if (node_size > machine->ram_size - mem_start) {
359                 node_size = machine->ram_size - mem_start;
360             }
361         }
362         if (!mem_start) {
363             /* spapr_machine_init() checks for rma_size <= node0_size
364              * already */
365             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
366             mem_start += spapr->rma_size;
367             node_size -= spapr->rma_size;
368         }
369         for ( ; node_size; ) {
370             hwaddr sizetmp = pow2floor(node_size);
371 
372             /* mem_start != 0 here */
373             if (ctzl(mem_start) < ctzl(sizetmp)) {
374                 sizetmp = 1ULL << ctzl(mem_start);
375             }
376 
377             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
378             node_size -= sizetmp;
379             mem_start += sizetmp;
380         }
381     }
382 
383     return 0;
384 }
385 
386 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
387                                   SpaprMachineState *spapr)
388 {
389     MachineState *ms = MACHINE(spapr);
390     PowerPCCPU *cpu = POWERPC_CPU(cs);
391     CPUPPCState *env = &cpu->env;
392     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
393     int index = spapr_get_vcpu_id(cpu);
394     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
395                        0xffffffff, 0xffffffff};
396     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
397         : SPAPR_TIMEBASE_FREQ;
398     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
399     uint32_t page_sizes_prop[64];
400     size_t page_sizes_prop_size;
401     unsigned int smp_threads = ms->smp.threads;
402     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
403     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
404     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
405     SpaprDrc *drc;
406     int drc_index;
407     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
408     int i;
409 
410     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
411     if (drc) {
412         drc_index = spapr_drc_index(drc);
413         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
414     }
415 
416     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
417     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
418 
419     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
420     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
421                            env->dcache_line_size)));
422     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
423                            env->dcache_line_size)));
424     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
425                            env->icache_line_size)));
426     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
427                            env->icache_line_size)));
428 
429     if (pcc->l1_dcache_size) {
430         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
431                                pcc->l1_dcache_size)));
432     } else {
433         warn_report("Unknown L1 dcache size for cpu");
434     }
435     if (pcc->l1_icache_size) {
436         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
437                                pcc->l1_icache_size)));
438     } else {
439         warn_report("Unknown L1 icache size for cpu");
440     }
441 
442     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
443     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
444     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
445     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
446     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
447     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
448 
449     if (env->spr_cb[SPR_PURR].oea_read) {
450         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
451     }
452     if (env->spr_cb[SPR_SPURR].oea_read) {
453         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
454     }
455 
456     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
457         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
458                           segs, sizeof(segs))));
459     }
460 
461     /* Advertise VSX (vector extensions) if available
462      *   1               == VMX / Altivec available
463      *   2               == VSX available
464      *
465      * Only CPUs for which we create core types in spapr_cpu_core.c
466      * are possible, and all of those have VMX */
467     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
468         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
469     } else {
470         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
471     }
472 
473     /* Advertise DFP (Decimal Floating Point) if available
474      *   0 / no property == no DFP
475      *   1               == DFP available */
476     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
477         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
478     }
479 
480     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
481                                                       sizeof(page_sizes_prop));
482     if (page_sizes_prop_size) {
483         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
484                           page_sizes_prop, page_sizes_prop_size)));
485     }
486 
487     spapr_populate_pa_features(spapr, cpu, fdt, offset);
488 
489     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
490                            cs->cpu_index / vcpus_per_socket)));
491 
492     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
493                       pft_size_prop, sizeof(pft_size_prop))));
494 
495     if (ms->numa_state->num_nodes > 1) {
496         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
497     }
498 
499     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
500 
501     if (pcc->radix_page_info) {
502         for (i = 0; i < pcc->radix_page_info->count; i++) {
503             radix_AP_encodings[i] =
504                 cpu_to_be32(pcc->radix_page_info->entries[i]);
505         }
506         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
507                           radix_AP_encodings,
508                           pcc->radix_page_info->count *
509                           sizeof(radix_AP_encodings[0]))));
510     }
511 
512     /*
513      * We set this property to let the guest know that it can use the large
514      * decrementer and its width in bits.
515      */
516     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
517         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
518                               pcc->lrg_decr_bits)));
519 }
520 
521 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
522 {
523     CPUState **rev;
524     CPUState *cs;
525     int n_cpus;
526     int cpus_offset;
527     char *nodename;
528     int i;
529 
530     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
531     _FDT(cpus_offset);
532     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
533     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
534 
535     /*
536      * We walk the CPUs in reverse order to ensure that CPU DT nodes
537      * created by fdt_add_subnode() end up in the right order in FDT
538      * for the guest kernel the enumerate the CPUs correctly.
539      *
540      * The CPU list cannot be traversed in reverse order, so we need
541      * to do extra work.
542      */
543     n_cpus = 0;
544     rev = NULL;
545     CPU_FOREACH(cs) {
546         rev = g_renew(CPUState *, rev, n_cpus + 1);
547         rev[n_cpus++] = cs;
548     }
549 
550     for (i = n_cpus - 1; i >= 0; i--) {
551         CPUState *cs = rev[i];
552         PowerPCCPU *cpu = POWERPC_CPU(cs);
553         int index = spapr_get_vcpu_id(cpu);
554         DeviceClass *dc = DEVICE_GET_CLASS(cs);
555         int offset;
556 
557         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
558             continue;
559         }
560 
561         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
562         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
563         g_free(nodename);
564         _FDT(offset);
565         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
566     }
567 
568     g_free(rev);
569 }
570 
571 static int spapr_rng_populate_dt(void *fdt)
572 {
573     int node;
574     int ret;
575 
576     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
577     if (node <= 0) {
578         return -1;
579     }
580     ret = fdt_setprop_string(fdt, node, "device_type",
581                              "ibm,platform-facilities");
582     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
583     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
584 
585     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
586     if (node <= 0) {
587         return -1;
588     }
589     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
590 
591     return ret ? -1 : 0;
592 }
593 
594 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
595 {
596     MemoryDeviceInfoList *info;
597 
598     for (info = list; info; info = info->next) {
599         MemoryDeviceInfo *value = info->value;
600 
601         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
602             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
603 
604             if (addr >= pcdimm_info->addr &&
605                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
606                 return pcdimm_info->node;
607             }
608         }
609     }
610 
611     return -1;
612 }
613 
614 struct sPAPRDrconfCellV2 {
615      uint32_t seq_lmbs;
616      uint64_t base_addr;
617      uint32_t drc_index;
618      uint32_t aa_index;
619      uint32_t flags;
620 } QEMU_PACKED;
621 
622 typedef struct DrconfCellQueue {
623     struct sPAPRDrconfCellV2 cell;
624     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
625 } DrconfCellQueue;
626 
627 static DrconfCellQueue *
628 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
629                       uint32_t drc_index, uint32_t aa_index,
630                       uint32_t flags)
631 {
632     DrconfCellQueue *elem;
633 
634     elem = g_malloc0(sizeof(*elem));
635     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
636     elem->cell.base_addr = cpu_to_be64(base_addr);
637     elem->cell.drc_index = cpu_to_be32(drc_index);
638     elem->cell.aa_index = cpu_to_be32(aa_index);
639     elem->cell.flags = cpu_to_be32(flags);
640 
641     return elem;
642 }
643 
644 /* ibm,dynamic-memory-v2 */
645 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
646                                    int offset, MemoryDeviceInfoList *dimms)
647 {
648     MachineState *machine = MACHINE(spapr);
649     uint8_t *int_buf, *cur_index;
650     int ret;
651     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
652     uint64_t addr, cur_addr, size;
653     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
654     uint64_t mem_end = machine->device_memory->base +
655                        memory_region_size(&machine->device_memory->mr);
656     uint32_t node, buf_len, nr_entries = 0;
657     SpaprDrc *drc;
658     DrconfCellQueue *elem, *next;
659     MemoryDeviceInfoList *info;
660     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
661         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
662 
663     /* Entry to cover RAM and the gap area */
664     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
665                                  SPAPR_LMB_FLAGS_RESERVED |
666                                  SPAPR_LMB_FLAGS_DRC_INVALID);
667     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
668     nr_entries++;
669 
670     cur_addr = machine->device_memory->base;
671     for (info = dimms; info; info = info->next) {
672         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
673 
674         addr = di->addr;
675         size = di->size;
676         node = di->node;
677 
678         /* Entry for hot-pluggable area */
679         if (cur_addr < addr) {
680             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
681             g_assert(drc);
682             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
683                                          cur_addr, spapr_drc_index(drc), -1, 0);
684             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
685             nr_entries++;
686         }
687 
688         /* Entry for DIMM */
689         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
690         g_assert(drc);
691         elem = spapr_get_drconf_cell(size / lmb_size, addr,
692                                      spapr_drc_index(drc), node,
693                                      SPAPR_LMB_FLAGS_ASSIGNED);
694         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
695         nr_entries++;
696         cur_addr = addr + size;
697     }
698 
699     /* Entry for remaining hotpluggable area */
700     if (cur_addr < mem_end) {
701         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
702         g_assert(drc);
703         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
704                                      cur_addr, spapr_drc_index(drc), -1, 0);
705         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
706         nr_entries++;
707     }
708 
709     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
710     int_buf = cur_index = g_malloc0(buf_len);
711     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
712     cur_index += sizeof(nr_entries);
713 
714     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
715         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
716         cur_index += sizeof(elem->cell);
717         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
718         g_free(elem);
719     }
720 
721     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
722     g_free(int_buf);
723     if (ret < 0) {
724         return -1;
725     }
726     return 0;
727 }
728 
729 /* ibm,dynamic-memory */
730 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
731                                    int offset, MemoryDeviceInfoList *dimms)
732 {
733     MachineState *machine = MACHINE(spapr);
734     int i, ret;
735     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
736     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
737     uint32_t nr_lmbs = (machine->device_memory->base +
738                        memory_region_size(&machine->device_memory->mr)) /
739                        lmb_size;
740     uint32_t *int_buf, *cur_index, buf_len;
741 
742     /*
743      * Allocate enough buffer size to fit in ibm,dynamic-memory
744      */
745     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
746     cur_index = int_buf = g_malloc0(buf_len);
747     int_buf[0] = cpu_to_be32(nr_lmbs);
748     cur_index++;
749     for (i = 0; i < nr_lmbs; i++) {
750         uint64_t addr = i * lmb_size;
751         uint32_t *dynamic_memory = cur_index;
752 
753         if (i >= device_lmb_start) {
754             SpaprDrc *drc;
755 
756             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
757             g_assert(drc);
758 
759             dynamic_memory[0] = cpu_to_be32(addr >> 32);
760             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
761             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
762             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
763             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
764             if (memory_region_present(get_system_memory(), addr)) {
765                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
766             } else {
767                 dynamic_memory[5] = cpu_to_be32(0);
768             }
769         } else {
770             /*
771              * LMB information for RMA, boot time RAM and gap b/n RAM and
772              * device memory region -- all these are marked as reserved
773              * and as having no valid DRC.
774              */
775             dynamic_memory[0] = cpu_to_be32(addr >> 32);
776             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
777             dynamic_memory[2] = cpu_to_be32(0);
778             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
779             dynamic_memory[4] = cpu_to_be32(-1);
780             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
781                                             SPAPR_LMB_FLAGS_DRC_INVALID);
782         }
783 
784         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
785     }
786     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
787     g_free(int_buf);
788     if (ret < 0) {
789         return -1;
790     }
791     return 0;
792 }
793 
794 /*
795  * Adds ibm,dynamic-reconfiguration-memory node.
796  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
797  * of this device tree node.
798  */
799 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
800 {
801     MachineState *machine = MACHINE(spapr);
802     int nb_numa_nodes = machine->numa_state->num_nodes;
803     int ret, i, offset;
804     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
805     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
806     uint32_t *int_buf, *cur_index, buf_len;
807     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
808     MemoryDeviceInfoList *dimms = NULL;
809 
810     /*
811      * Don't create the node if there is no device memory
812      */
813     if (machine->ram_size == machine->maxram_size) {
814         return 0;
815     }
816 
817     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
818 
819     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
820                     sizeof(prop_lmb_size));
821     if (ret < 0) {
822         return ret;
823     }
824 
825     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
826     if (ret < 0) {
827         return ret;
828     }
829 
830     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
831     if (ret < 0) {
832         return ret;
833     }
834 
835     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
836     dimms = qmp_memory_device_list();
837     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
838         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
839     } else {
840         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
841     }
842     qapi_free_MemoryDeviceInfoList(dimms);
843 
844     if (ret < 0) {
845         return ret;
846     }
847 
848     /* ibm,associativity-lookup-arrays */
849     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
850     cur_index = int_buf = g_malloc0(buf_len);
851     int_buf[0] = cpu_to_be32(nr_nodes);
852     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
853     cur_index += 2;
854     for (i = 0; i < nr_nodes; i++) {
855         uint32_t associativity[] = {
856             cpu_to_be32(0x0),
857             cpu_to_be32(0x0),
858             cpu_to_be32(0x0),
859             cpu_to_be32(i)
860         };
861         memcpy(cur_index, associativity, sizeof(associativity));
862         cur_index += 4;
863     }
864     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
865             (cur_index - int_buf) * sizeof(uint32_t));
866     g_free(int_buf);
867 
868     return ret;
869 }
870 
871 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
872                                 SpaprOptionVector *ov5_updates)
873 {
874     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
875     int ret = 0, offset;
876 
877     /* Generate ibm,dynamic-reconfiguration-memory node if required */
878     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
879         g_assert(smc->dr_lmb_enabled);
880         ret = spapr_populate_drconf_memory(spapr, fdt);
881         if (ret) {
882             goto out;
883         }
884     }
885 
886     offset = fdt_path_offset(fdt, "/chosen");
887     if (offset < 0) {
888         offset = fdt_add_subnode(fdt, 0, "chosen");
889         if (offset < 0) {
890             return offset;
891         }
892     }
893     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
894                                  "ibm,architecture-vec-5");
895 
896 out:
897     return ret;
898 }
899 
900 static bool spapr_hotplugged_dev_before_cas(void)
901 {
902     Object *drc_container, *obj;
903     ObjectProperty *prop;
904     ObjectPropertyIterator iter;
905 
906     drc_container = container_get(object_get_root(), "/dr-connector");
907     object_property_iter_init(&iter, drc_container);
908     while ((prop = object_property_iter_next(&iter))) {
909         if (!strstart(prop->type, "link<", NULL)) {
910             continue;
911         }
912         obj = object_property_get_link(drc_container, prop->name, NULL);
913         if (spapr_drc_needed(obj)) {
914             return true;
915         }
916     }
917     return false;
918 }
919 
920 static void *spapr_build_fdt(SpaprMachineState *spapr, bool reset);
921 
922 int spapr_h_cas_compose_response(SpaprMachineState *spapr,
923                                  target_ulong addr, target_ulong size,
924                                  SpaprOptionVector *ov5_updates)
925 {
926     void *fdt;
927     SpaprDeviceTreeUpdateHeader hdr = { .version_id = 1 };
928 
929     if (spapr_hotplugged_dev_before_cas()) {
930         return 1;
931     }
932 
933     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
934         error_report("SLOF provided an unexpected CAS buffer size "
935                      TARGET_FMT_lu " (min: %zu, max: %u)",
936                      size, sizeof(hdr), FW_MAX_SIZE);
937         exit(EXIT_FAILURE);
938     }
939 
940     size -= sizeof(hdr);
941 
942     fdt = spapr_build_fdt(spapr, false);
943     _FDT((fdt_pack(fdt)));
944 
945     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
946         g_free(fdt);
947         trace_spapr_cas_failed(size);
948         return -1;
949     }
950 
951     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
952     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
953     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
954 
955     g_free(spapr->fdt_blob);
956     spapr->fdt_size = fdt_totalsize(fdt);
957     spapr->fdt_initial_size = spapr->fdt_size;
958     spapr->fdt_blob = fdt;
959 
960     return 0;
961 }
962 
963 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
964 {
965     MachineState *ms = MACHINE(spapr);
966     int rtas;
967     GString *hypertas = g_string_sized_new(256);
968     GString *qemu_hypertas = g_string_sized_new(256);
969     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
970     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
971         memory_region_size(&MACHINE(spapr)->device_memory->mr);
972     uint32_t lrdr_capacity[] = {
973         cpu_to_be32(max_device_addr >> 32),
974         cpu_to_be32(max_device_addr & 0xffffffff),
975         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
976         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
977     };
978     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
979     uint32_t maxdomains[] = {
980         cpu_to_be32(4),
981         maxdomain,
982         maxdomain,
983         maxdomain,
984         cpu_to_be32(spapr->gpu_numa_id),
985     };
986 
987     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
988 
989     /* hypertas */
990     add_str(hypertas, "hcall-pft");
991     add_str(hypertas, "hcall-term");
992     add_str(hypertas, "hcall-dabr");
993     add_str(hypertas, "hcall-interrupt");
994     add_str(hypertas, "hcall-tce");
995     add_str(hypertas, "hcall-vio");
996     add_str(hypertas, "hcall-splpar");
997     add_str(hypertas, "hcall-join");
998     add_str(hypertas, "hcall-bulk");
999     add_str(hypertas, "hcall-set-mode");
1000     add_str(hypertas, "hcall-sprg0");
1001     add_str(hypertas, "hcall-copy");
1002     add_str(hypertas, "hcall-debug");
1003     add_str(hypertas, "hcall-vphn");
1004     add_str(qemu_hypertas, "hcall-memop1");
1005 
1006     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1007         add_str(hypertas, "hcall-multi-tce");
1008     }
1009 
1010     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1011         add_str(hypertas, "hcall-hpt-resize");
1012     }
1013 
1014     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1015                      hypertas->str, hypertas->len));
1016     g_string_free(hypertas, TRUE);
1017     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1018                      qemu_hypertas->str, qemu_hypertas->len));
1019     g_string_free(qemu_hypertas, TRUE);
1020 
1021     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1022                      refpoints, sizeof(refpoints)));
1023 
1024     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1025                      maxdomains, sizeof(maxdomains)));
1026 
1027     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1028                           RTAS_ERROR_LOG_MAX));
1029     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1030                           RTAS_EVENT_SCAN_RATE));
1031 
1032     g_assert(msi_nonbroken);
1033     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1034 
1035     /*
1036      * According to PAPR, rtas ibm,os-term does not guarantee a return
1037      * back to the guest cpu.
1038      *
1039      * While an additional ibm,extended-os-term property indicates
1040      * that rtas call return will always occur. Set this property.
1041      */
1042     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1043 
1044     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1045                      lrdr_capacity, sizeof(lrdr_capacity)));
1046 
1047     spapr_dt_rtas_tokens(fdt, rtas);
1048 }
1049 
1050 /*
1051  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1052  * and the XIVE features that the guest may request and thus the valid
1053  * values for bytes 23..26 of option vector 5:
1054  */
1055 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1056                                           int chosen)
1057 {
1058     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1059 
1060     char val[2 * 4] = {
1061         23, 0x00, /* XICS / XIVE mode */
1062         24, 0x00, /* Hash/Radix, filled in below. */
1063         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1064         26, 0x40, /* Radix options: GTSE == yes. */
1065     };
1066 
1067     if (spapr->irq->xics && spapr->irq->xive) {
1068         val[1] = SPAPR_OV5_XIVE_BOTH;
1069     } else if (spapr->irq->xive) {
1070         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1071     } else {
1072         assert(spapr->irq->xics);
1073         val[1] = SPAPR_OV5_XIVE_LEGACY;
1074     }
1075 
1076     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1077                           first_ppc_cpu->compat_pvr)) {
1078         /*
1079          * If we're in a pre POWER9 compat mode then the guest should
1080          * do hash and use the legacy interrupt mode
1081          */
1082         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1083         val[3] = 0x00; /* Hash */
1084     } else if (kvm_enabled()) {
1085         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1086             val[3] = 0x80; /* OV5_MMU_BOTH */
1087         } else if (kvmppc_has_cap_mmu_radix()) {
1088             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1089         } else {
1090             val[3] = 0x00; /* Hash */
1091         }
1092     } else {
1093         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1094         val[3] = 0xC0;
1095     }
1096     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1097                      val, sizeof(val)));
1098 }
1099 
1100 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1101 {
1102     MachineState *machine = MACHINE(spapr);
1103     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1104     int chosen;
1105     const char *boot_device = machine->boot_order;
1106     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1107     size_t cb = 0;
1108     char *bootlist = get_boot_devices_list(&cb);
1109 
1110     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1111 
1112     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1113         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1114                                 machine->kernel_cmdline));
1115     }
1116     if (spapr->initrd_size) {
1117         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1118                               spapr->initrd_base));
1119         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1120                               spapr->initrd_base + spapr->initrd_size));
1121     }
1122 
1123     if (spapr->kernel_size) {
1124         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1125                               cpu_to_be64(spapr->kernel_size) };
1126 
1127         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1128                          &kprop, sizeof(kprop)));
1129         if (spapr->kernel_le) {
1130             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1131         }
1132     }
1133     if (boot_menu) {
1134         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1135     }
1136     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1137     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1138     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1139 
1140     if (cb && bootlist) {
1141         int i;
1142 
1143         for (i = 0; i < cb; i++) {
1144             if (bootlist[i] == '\n') {
1145                 bootlist[i] = ' ';
1146             }
1147         }
1148         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1149     }
1150 
1151     if (boot_device && strlen(boot_device)) {
1152         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1153     }
1154 
1155     if (!spapr->has_graphics && stdout_path) {
1156         /*
1157          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1158          * kernel. New platforms should only use the "stdout-path" property. Set
1159          * the new property and continue using older property to remain
1160          * compatible with the existing firmware.
1161          */
1162         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1163         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1164     }
1165 
1166     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1167     if (smc->linux_pci_probe) {
1168         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1169     }
1170 
1171     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1172 
1173     g_free(stdout_path);
1174     g_free(bootlist);
1175 }
1176 
1177 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1178 {
1179     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1180      * KVM to work under pHyp with some guest co-operation */
1181     int hypervisor;
1182     uint8_t hypercall[16];
1183 
1184     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1185     /* indicate KVM hypercall interface */
1186     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1187     if (kvmppc_has_cap_fixup_hcalls()) {
1188         /*
1189          * Older KVM versions with older guest kernels were broken
1190          * with the magic page, don't allow the guest to map it.
1191          */
1192         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1193                                   sizeof(hypercall))) {
1194             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1195                              hypercall, sizeof(hypercall)));
1196         }
1197     }
1198 }
1199 
1200 static void *spapr_build_fdt(SpaprMachineState *spapr, bool reset)
1201 {
1202     MachineState *machine = MACHINE(spapr);
1203     MachineClass *mc = MACHINE_GET_CLASS(machine);
1204     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1205     int ret;
1206     void *fdt;
1207     SpaprPhbState *phb;
1208     char *buf;
1209 
1210     fdt = g_malloc0(FDT_MAX_SIZE);
1211     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1212 
1213     /* Root node */
1214     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1215     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1216     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1217 
1218     /* Guest UUID & Name*/
1219     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1220     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1221     if (qemu_uuid_set) {
1222         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1223     }
1224     g_free(buf);
1225 
1226     if (qemu_get_vm_name()) {
1227         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1228                                 qemu_get_vm_name()));
1229     }
1230 
1231     /* Host Model & Serial Number */
1232     if (spapr->host_model) {
1233         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1234     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1235         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1236         g_free(buf);
1237     }
1238 
1239     if (spapr->host_serial) {
1240         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1241     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1242         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1243         g_free(buf);
1244     }
1245 
1246     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1247     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1248 
1249     /* /interrupt controller */
1250     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1251 
1252     ret = spapr_populate_memory(spapr, fdt);
1253     if (ret < 0) {
1254         error_report("couldn't setup memory nodes in fdt");
1255         exit(1);
1256     }
1257 
1258     /* /vdevice */
1259     spapr_dt_vdevice(spapr->vio_bus, fdt);
1260 
1261     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1262         ret = spapr_rng_populate_dt(fdt);
1263         if (ret < 0) {
1264             error_report("could not set up rng device in the fdt");
1265             exit(1);
1266         }
1267     }
1268 
1269     QLIST_FOREACH(phb, &spapr->phbs, list) {
1270         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1271         if (ret < 0) {
1272             error_report("couldn't setup PCI devices in fdt");
1273             exit(1);
1274         }
1275     }
1276 
1277     /* cpus */
1278     spapr_populate_cpus_dt_node(fdt, spapr);
1279 
1280     if (smc->dr_lmb_enabled) {
1281         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1282     }
1283 
1284     if (mc->has_hotpluggable_cpus) {
1285         int offset = fdt_path_offset(fdt, "/cpus");
1286         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1287         if (ret < 0) {
1288             error_report("Couldn't set up CPU DR device tree properties");
1289             exit(1);
1290         }
1291     }
1292 
1293     /* /event-sources */
1294     spapr_dt_events(spapr, fdt);
1295 
1296     /* /rtas */
1297     spapr_dt_rtas(spapr, fdt);
1298 
1299     /* /chosen */
1300     if (reset) {
1301         spapr_dt_chosen(spapr, fdt);
1302     }
1303 
1304     /* /hypervisor */
1305     if (kvm_enabled()) {
1306         spapr_dt_hypervisor(spapr, fdt);
1307     }
1308 
1309     /* Build memory reserve map */
1310     if (reset) {
1311         if (spapr->kernel_size) {
1312             _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1313         }
1314         if (spapr->initrd_size) {
1315             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1316                                   spapr->initrd_size)));
1317         }
1318     }
1319 
1320     /* ibm,client-architecture-support updates */
1321     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1322     if (ret < 0) {
1323         error_report("couldn't setup CAS properties fdt");
1324         exit(1);
1325     }
1326 
1327     if (smc->dr_phb_enabled) {
1328         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1329         if (ret < 0) {
1330             error_report("Couldn't set up PHB DR device tree properties");
1331             exit(1);
1332         }
1333     }
1334 
1335     return fdt;
1336 }
1337 
1338 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1339 {
1340     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1341 }
1342 
1343 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1344                                     PowerPCCPU *cpu)
1345 {
1346     CPUPPCState *env = &cpu->env;
1347 
1348     /* The TCG path should also be holding the BQL at this point */
1349     g_assert(qemu_mutex_iothread_locked());
1350 
1351     if (msr_pr) {
1352         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1353         env->gpr[3] = H_PRIVILEGE;
1354     } else {
1355         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1356     }
1357 }
1358 
1359 struct LPCRSyncState {
1360     target_ulong value;
1361     target_ulong mask;
1362 };
1363 
1364 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1365 {
1366     struct LPCRSyncState *s = arg.host_ptr;
1367     PowerPCCPU *cpu = POWERPC_CPU(cs);
1368     CPUPPCState *env = &cpu->env;
1369     target_ulong lpcr;
1370 
1371     cpu_synchronize_state(cs);
1372     lpcr = env->spr[SPR_LPCR];
1373     lpcr &= ~s->mask;
1374     lpcr |= s->value;
1375     ppc_store_lpcr(cpu, lpcr);
1376 }
1377 
1378 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1379 {
1380     CPUState *cs;
1381     struct LPCRSyncState s = {
1382         .value = value,
1383         .mask = mask
1384     };
1385     CPU_FOREACH(cs) {
1386         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1387     }
1388 }
1389 
1390 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1391 {
1392     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1393 
1394     /* Copy PATE1:GR into PATE0:HR */
1395     entry->dw0 = spapr->patb_entry & PATE0_HR;
1396     entry->dw1 = spapr->patb_entry;
1397 }
1398 
1399 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1400 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1401 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1402 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1403 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1404 
1405 /*
1406  * Get the fd to access the kernel htab, re-opening it if necessary
1407  */
1408 static int get_htab_fd(SpaprMachineState *spapr)
1409 {
1410     Error *local_err = NULL;
1411 
1412     if (spapr->htab_fd >= 0) {
1413         return spapr->htab_fd;
1414     }
1415 
1416     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1417     if (spapr->htab_fd < 0) {
1418         error_report_err(local_err);
1419     }
1420 
1421     return spapr->htab_fd;
1422 }
1423 
1424 void close_htab_fd(SpaprMachineState *spapr)
1425 {
1426     if (spapr->htab_fd >= 0) {
1427         close(spapr->htab_fd);
1428     }
1429     spapr->htab_fd = -1;
1430 }
1431 
1432 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1433 {
1434     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1435 
1436     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1437 }
1438 
1439 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1440 {
1441     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1442 
1443     assert(kvm_enabled());
1444 
1445     if (!spapr->htab) {
1446         return 0;
1447     }
1448 
1449     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1450 }
1451 
1452 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1453                                                 hwaddr ptex, int n)
1454 {
1455     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1456     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1457 
1458     if (!spapr->htab) {
1459         /*
1460          * HTAB is controlled by KVM. Fetch into temporary buffer
1461          */
1462         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1463         kvmppc_read_hptes(hptes, ptex, n);
1464         return hptes;
1465     }
1466 
1467     /*
1468      * HTAB is controlled by QEMU. Just point to the internally
1469      * accessible PTEG.
1470      */
1471     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1472 }
1473 
1474 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1475                               const ppc_hash_pte64_t *hptes,
1476                               hwaddr ptex, int n)
1477 {
1478     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1479 
1480     if (!spapr->htab) {
1481         g_free((void *)hptes);
1482     }
1483 
1484     /* Nothing to do for qemu managed HPT */
1485 }
1486 
1487 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1488                       uint64_t pte0, uint64_t pte1)
1489 {
1490     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1491     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1492 
1493     if (!spapr->htab) {
1494         kvmppc_write_hpte(ptex, pte0, pte1);
1495     } else {
1496         if (pte0 & HPTE64_V_VALID) {
1497             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1498             /*
1499              * When setting valid, we write PTE1 first. This ensures
1500              * proper synchronization with the reading code in
1501              * ppc_hash64_pteg_search()
1502              */
1503             smp_wmb();
1504             stq_p(spapr->htab + offset, pte0);
1505         } else {
1506             stq_p(spapr->htab + offset, pte0);
1507             /*
1508              * When clearing it we set PTE0 first. This ensures proper
1509              * synchronization with the reading code in
1510              * ppc_hash64_pteg_search()
1511              */
1512             smp_wmb();
1513             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1514         }
1515     }
1516 }
1517 
1518 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1519                              uint64_t pte1)
1520 {
1521     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1522     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1523 
1524     if (!spapr->htab) {
1525         /* There should always be a hash table when this is called */
1526         error_report("spapr_hpte_set_c called with no hash table !");
1527         return;
1528     }
1529 
1530     /* The HW performs a non-atomic byte update */
1531     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1532 }
1533 
1534 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1535                              uint64_t pte1)
1536 {
1537     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1538     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1539 
1540     if (!spapr->htab) {
1541         /* There should always be a hash table when this is called */
1542         error_report("spapr_hpte_set_r called with no hash table !");
1543         return;
1544     }
1545 
1546     /* The HW performs a non-atomic byte update */
1547     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1548 }
1549 
1550 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1551 {
1552     int shift;
1553 
1554     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1555      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1556      * that's much more than is needed for Linux guests */
1557     shift = ctz64(pow2ceil(ramsize)) - 7;
1558     shift = MAX(shift, 18); /* Minimum architected size */
1559     shift = MIN(shift, 46); /* Maximum architected size */
1560     return shift;
1561 }
1562 
1563 void spapr_free_hpt(SpaprMachineState *spapr)
1564 {
1565     g_free(spapr->htab);
1566     spapr->htab = NULL;
1567     spapr->htab_shift = 0;
1568     close_htab_fd(spapr);
1569 }
1570 
1571 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1572                           Error **errp)
1573 {
1574     long rc;
1575 
1576     /* Clean up any HPT info from a previous boot */
1577     spapr_free_hpt(spapr);
1578 
1579     rc = kvmppc_reset_htab(shift);
1580     if (rc < 0) {
1581         /* kernel-side HPT needed, but couldn't allocate one */
1582         error_setg_errno(errp, errno,
1583                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1584                          shift);
1585         /* This is almost certainly fatal, but if the caller really
1586          * wants to carry on with shift == 0, it's welcome to try */
1587     } else if (rc > 0) {
1588         /* kernel-side HPT allocated */
1589         if (rc != shift) {
1590             error_setg(errp,
1591                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1592                        shift, rc);
1593         }
1594 
1595         spapr->htab_shift = shift;
1596         spapr->htab = NULL;
1597     } else {
1598         /* kernel-side HPT not needed, allocate in userspace instead */
1599         size_t size = 1ULL << shift;
1600         int i;
1601 
1602         spapr->htab = qemu_memalign(size, size);
1603         if (!spapr->htab) {
1604             error_setg_errno(errp, errno,
1605                              "Could not allocate HPT of order %d", shift);
1606             return;
1607         }
1608 
1609         memset(spapr->htab, 0, size);
1610         spapr->htab_shift = shift;
1611 
1612         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1613             DIRTY_HPTE(HPTE(spapr->htab, i));
1614         }
1615     }
1616     /* We're setting up a hash table, so that means we're not radix */
1617     spapr->patb_entry = 0;
1618     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1619 }
1620 
1621 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1622 {
1623     int hpt_shift;
1624 
1625     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1626         || (spapr->cas_reboot
1627             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1628         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1629     } else {
1630         uint64_t current_ram_size;
1631 
1632         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1633         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1634     }
1635     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1636 
1637     if (spapr->vrma_adjust) {
1638         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1639                                           spapr->htab_shift);
1640     }
1641 }
1642 
1643 static int spapr_reset_drcs(Object *child, void *opaque)
1644 {
1645     SpaprDrc *drc =
1646         (SpaprDrc *) object_dynamic_cast(child,
1647                                                  TYPE_SPAPR_DR_CONNECTOR);
1648 
1649     if (drc) {
1650         spapr_drc_reset(drc);
1651     }
1652 
1653     return 0;
1654 }
1655 
1656 static void spapr_machine_reset(MachineState *machine)
1657 {
1658     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1659     PowerPCCPU *first_ppc_cpu;
1660     hwaddr fdt_addr;
1661     void *fdt;
1662     int rc;
1663 
1664     spapr_caps_apply(spapr);
1665 
1666     first_ppc_cpu = POWERPC_CPU(first_cpu);
1667     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1668         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1669                               spapr->max_compat_pvr)) {
1670         /*
1671          * If using KVM with radix mode available, VCPUs can be started
1672          * without a HPT because KVM will start them in radix mode.
1673          * Set the GR bit in PATE so that we know there is no HPT.
1674          */
1675         spapr->patb_entry = PATE1_GR;
1676         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1677     } else {
1678         spapr_setup_hpt_and_vrma(spapr);
1679     }
1680 
1681     qemu_devices_reset();
1682 
1683     /*
1684      * If this reset wasn't generated by CAS, we should reset our
1685      * negotiated options and start from scratch
1686      */
1687     if (!spapr->cas_reboot) {
1688         spapr_ovec_cleanup(spapr->ov5_cas);
1689         spapr->ov5_cas = spapr_ovec_new();
1690 
1691         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1692     }
1693 
1694     /*
1695      * This is fixing some of the default configuration of the XIVE
1696      * devices. To be called after the reset of the machine devices.
1697      */
1698     spapr_irq_reset(spapr, &error_fatal);
1699 
1700     /*
1701      * There is no CAS under qtest. Simulate one to please the code that
1702      * depends on spapr->ov5_cas. This is especially needed to test device
1703      * unplug, so we do that before resetting the DRCs.
1704      */
1705     if (qtest_enabled()) {
1706         spapr_ovec_cleanup(spapr->ov5_cas);
1707         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1708     }
1709 
1710     /* DRC reset may cause a device to be unplugged. This will cause troubles
1711      * if this device is used by another device (eg, a running vhost backend
1712      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1713      * situations, we reset DRCs after all devices have been reset.
1714      */
1715     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1716 
1717     spapr_clear_pending_events(spapr);
1718 
1719     /*
1720      * We place the device tree and RTAS just below either the top of the RMA,
1721      * or just below 2GB, whichever is lower, so that it can be
1722      * processed with 32-bit real mode code if necessary
1723      */
1724     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1725 
1726     fdt = spapr_build_fdt(spapr, true);
1727 
1728     rc = fdt_pack(fdt);
1729 
1730     /* Should only fail if we've built a corrupted tree */
1731     assert(rc == 0);
1732 
1733     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1734         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1735                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1736         exit(1);
1737     }
1738 
1739     /* Load the fdt */
1740     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1741     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1742     g_free(spapr->fdt_blob);
1743     spapr->fdt_size = fdt_totalsize(fdt);
1744     spapr->fdt_initial_size = spapr->fdt_size;
1745     spapr->fdt_blob = fdt;
1746 
1747     /* Set up the entry state */
1748     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1749     first_ppc_cpu->env.gpr[5] = 0;
1750 
1751     spapr->cas_reboot = false;
1752 }
1753 
1754 static void spapr_create_nvram(SpaprMachineState *spapr)
1755 {
1756     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1757     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1758 
1759     if (dinfo) {
1760         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1761                             &error_fatal);
1762     }
1763 
1764     qdev_init_nofail(dev);
1765 
1766     spapr->nvram = (struct SpaprNvram *)dev;
1767 }
1768 
1769 static void spapr_rtc_create(SpaprMachineState *spapr)
1770 {
1771     object_initialize_child(OBJECT(spapr), "rtc",
1772                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1773                             &error_fatal, NULL);
1774     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1775                               &error_fatal);
1776     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1777                               "date", &error_fatal);
1778 }
1779 
1780 /* Returns whether we want to use VGA or not */
1781 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1782 {
1783     switch (vga_interface_type) {
1784     case VGA_NONE:
1785         return false;
1786     case VGA_DEVICE:
1787         return true;
1788     case VGA_STD:
1789     case VGA_VIRTIO:
1790     case VGA_CIRRUS:
1791         return pci_vga_init(pci_bus) != NULL;
1792     default:
1793         error_setg(errp,
1794                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1795         return false;
1796     }
1797 }
1798 
1799 static int spapr_pre_load(void *opaque)
1800 {
1801     int rc;
1802 
1803     rc = spapr_caps_pre_load(opaque);
1804     if (rc) {
1805         return rc;
1806     }
1807 
1808     return 0;
1809 }
1810 
1811 static int spapr_post_load(void *opaque, int version_id)
1812 {
1813     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1814     int err = 0;
1815 
1816     err = spapr_caps_post_migration(spapr);
1817     if (err) {
1818         return err;
1819     }
1820 
1821     /*
1822      * In earlier versions, there was no separate qdev for the PAPR
1823      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1824      * So when migrating from those versions, poke the incoming offset
1825      * value into the RTC device
1826      */
1827     if (version_id < 3) {
1828         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1829         if (err) {
1830             return err;
1831         }
1832     }
1833 
1834     if (kvm_enabled() && spapr->patb_entry) {
1835         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1836         bool radix = !!(spapr->patb_entry & PATE1_GR);
1837         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1838 
1839         /*
1840          * Update LPCR:HR and UPRT as they may not be set properly in
1841          * the stream
1842          */
1843         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1844                             LPCR_HR | LPCR_UPRT);
1845 
1846         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1847         if (err) {
1848             error_report("Process table config unsupported by the host");
1849             return -EINVAL;
1850         }
1851     }
1852 
1853     err = spapr_irq_post_load(spapr, version_id);
1854     if (err) {
1855         return err;
1856     }
1857 
1858     return err;
1859 }
1860 
1861 static int spapr_pre_save(void *opaque)
1862 {
1863     int rc;
1864 
1865     rc = spapr_caps_pre_save(opaque);
1866     if (rc) {
1867         return rc;
1868     }
1869 
1870     return 0;
1871 }
1872 
1873 static bool version_before_3(void *opaque, int version_id)
1874 {
1875     return version_id < 3;
1876 }
1877 
1878 static bool spapr_pending_events_needed(void *opaque)
1879 {
1880     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1881     return !QTAILQ_EMPTY(&spapr->pending_events);
1882 }
1883 
1884 static const VMStateDescription vmstate_spapr_event_entry = {
1885     .name = "spapr_event_log_entry",
1886     .version_id = 1,
1887     .minimum_version_id = 1,
1888     .fields = (VMStateField[]) {
1889         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1890         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1891         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1892                                      NULL, extended_length),
1893         VMSTATE_END_OF_LIST()
1894     },
1895 };
1896 
1897 static const VMStateDescription vmstate_spapr_pending_events = {
1898     .name = "spapr_pending_events",
1899     .version_id = 1,
1900     .minimum_version_id = 1,
1901     .needed = spapr_pending_events_needed,
1902     .fields = (VMStateField[]) {
1903         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1904                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1905         VMSTATE_END_OF_LIST()
1906     },
1907 };
1908 
1909 static bool spapr_ov5_cas_needed(void *opaque)
1910 {
1911     SpaprMachineState *spapr = opaque;
1912     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1913     SpaprOptionVector *ov5_legacy = spapr_ovec_new();
1914     SpaprOptionVector *ov5_removed = spapr_ovec_new();
1915     bool cas_needed;
1916 
1917     /* Prior to the introduction of SpaprOptionVector, we had two option
1918      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1919      * Both of these options encode machine topology into the device-tree
1920      * in such a way that the now-booted OS should still be able to interact
1921      * appropriately with QEMU regardless of what options were actually
1922      * negotiatied on the source side.
1923      *
1924      * As such, we can avoid migrating the CAS-negotiated options if these
1925      * are the only options available on the current machine/platform.
1926      * Since these are the only options available for pseries-2.7 and
1927      * earlier, this allows us to maintain old->new/new->old migration
1928      * compatibility.
1929      *
1930      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1931      * via default pseries-2.8 machines and explicit command-line parameters.
1932      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1933      * of the actual CAS-negotiated values to continue working properly. For
1934      * example, availability of memory unplug depends on knowing whether
1935      * OV5_HP_EVT was negotiated via CAS.
1936      *
1937      * Thus, for any cases where the set of available CAS-negotiatable
1938      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1939      * include the CAS-negotiated options in the migration stream, unless
1940      * if they affect boot time behaviour only.
1941      */
1942     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1943     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1944     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1945 
1946     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1947      * the mask itself since in the future it's possible "legacy" bits may be
1948      * removed via machine options, which could generate a false positive
1949      * that breaks migration.
1950      */
1951     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1952     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1953 
1954     spapr_ovec_cleanup(ov5_mask);
1955     spapr_ovec_cleanup(ov5_legacy);
1956     spapr_ovec_cleanup(ov5_removed);
1957 
1958     return cas_needed;
1959 }
1960 
1961 static const VMStateDescription vmstate_spapr_ov5_cas = {
1962     .name = "spapr_option_vector_ov5_cas",
1963     .version_id = 1,
1964     .minimum_version_id = 1,
1965     .needed = spapr_ov5_cas_needed,
1966     .fields = (VMStateField[]) {
1967         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1968                                  vmstate_spapr_ovec, SpaprOptionVector),
1969         VMSTATE_END_OF_LIST()
1970     },
1971 };
1972 
1973 static bool spapr_patb_entry_needed(void *opaque)
1974 {
1975     SpaprMachineState *spapr = opaque;
1976 
1977     return !!spapr->patb_entry;
1978 }
1979 
1980 static const VMStateDescription vmstate_spapr_patb_entry = {
1981     .name = "spapr_patb_entry",
1982     .version_id = 1,
1983     .minimum_version_id = 1,
1984     .needed = spapr_patb_entry_needed,
1985     .fields = (VMStateField[]) {
1986         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1987         VMSTATE_END_OF_LIST()
1988     },
1989 };
1990 
1991 static bool spapr_irq_map_needed(void *opaque)
1992 {
1993     SpaprMachineState *spapr = opaque;
1994 
1995     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1996 }
1997 
1998 static const VMStateDescription vmstate_spapr_irq_map = {
1999     .name = "spapr_irq_map",
2000     .version_id = 1,
2001     .minimum_version_id = 1,
2002     .needed = spapr_irq_map_needed,
2003     .fields = (VMStateField[]) {
2004         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2005         VMSTATE_END_OF_LIST()
2006     },
2007 };
2008 
2009 static bool spapr_dtb_needed(void *opaque)
2010 {
2011     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2012 
2013     return smc->update_dt_enabled;
2014 }
2015 
2016 static int spapr_dtb_pre_load(void *opaque)
2017 {
2018     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2019 
2020     g_free(spapr->fdt_blob);
2021     spapr->fdt_blob = NULL;
2022     spapr->fdt_size = 0;
2023 
2024     return 0;
2025 }
2026 
2027 static const VMStateDescription vmstate_spapr_dtb = {
2028     .name = "spapr_dtb",
2029     .version_id = 1,
2030     .minimum_version_id = 1,
2031     .needed = spapr_dtb_needed,
2032     .pre_load = spapr_dtb_pre_load,
2033     .fields = (VMStateField[]) {
2034         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2035         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2036         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2037                                      fdt_size),
2038         VMSTATE_END_OF_LIST()
2039     },
2040 };
2041 
2042 static const VMStateDescription vmstate_spapr = {
2043     .name = "spapr",
2044     .version_id = 3,
2045     .minimum_version_id = 1,
2046     .pre_load = spapr_pre_load,
2047     .post_load = spapr_post_load,
2048     .pre_save = spapr_pre_save,
2049     .fields = (VMStateField[]) {
2050         /* used to be @next_irq */
2051         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2052 
2053         /* RTC offset */
2054         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2055 
2056         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2057         VMSTATE_END_OF_LIST()
2058     },
2059     .subsections = (const VMStateDescription*[]) {
2060         &vmstate_spapr_ov5_cas,
2061         &vmstate_spapr_patb_entry,
2062         &vmstate_spapr_pending_events,
2063         &vmstate_spapr_cap_htm,
2064         &vmstate_spapr_cap_vsx,
2065         &vmstate_spapr_cap_dfp,
2066         &vmstate_spapr_cap_cfpc,
2067         &vmstate_spapr_cap_sbbc,
2068         &vmstate_spapr_cap_ibs,
2069         &vmstate_spapr_cap_hpt_maxpagesize,
2070         &vmstate_spapr_irq_map,
2071         &vmstate_spapr_cap_nested_kvm_hv,
2072         &vmstate_spapr_dtb,
2073         &vmstate_spapr_cap_large_decr,
2074         &vmstate_spapr_cap_ccf_assist,
2075         NULL
2076     }
2077 };
2078 
2079 static int htab_save_setup(QEMUFile *f, void *opaque)
2080 {
2081     SpaprMachineState *spapr = opaque;
2082 
2083     /* "Iteration" header */
2084     if (!spapr->htab_shift) {
2085         qemu_put_be32(f, -1);
2086     } else {
2087         qemu_put_be32(f, spapr->htab_shift);
2088     }
2089 
2090     if (spapr->htab) {
2091         spapr->htab_save_index = 0;
2092         spapr->htab_first_pass = true;
2093     } else {
2094         if (spapr->htab_shift) {
2095             assert(kvm_enabled());
2096         }
2097     }
2098 
2099 
2100     return 0;
2101 }
2102 
2103 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2104                             int chunkstart, int n_valid, int n_invalid)
2105 {
2106     qemu_put_be32(f, chunkstart);
2107     qemu_put_be16(f, n_valid);
2108     qemu_put_be16(f, n_invalid);
2109     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2110                     HASH_PTE_SIZE_64 * n_valid);
2111 }
2112 
2113 static void htab_save_end_marker(QEMUFile *f)
2114 {
2115     qemu_put_be32(f, 0);
2116     qemu_put_be16(f, 0);
2117     qemu_put_be16(f, 0);
2118 }
2119 
2120 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2121                                  int64_t max_ns)
2122 {
2123     bool has_timeout = max_ns != -1;
2124     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2125     int index = spapr->htab_save_index;
2126     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2127 
2128     assert(spapr->htab_first_pass);
2129 
2130     do {
2131         int chunkstart;
2132 
2133         /* Consume invalid HPTEs */
2134         while ((index < htabslots)
2135                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2136             CLEAN_HPTE(HPTE(spapr->htab, index));
2137             index++;
2138         }
2139 
2140         /* Consume valid HPTEs */
2141         chunkstart = index;
2142         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2143                && HPTE_VALID(HPTE(spapr->htab, index))) {
2144             CLEAN_HPTE(HPTE(spapr->htab, index));
2145             index++;
2146         }
2147 
2148         if (index > chunkstart) {
2149             int n_valid = index - chunkstart;
2150 
2151             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2152 
2153             if (has_timeout &&
2154                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2155                 break;
2156             }
2157         }
2158     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2159 
2160     if (index >= htabslots) {
2161         assert(index == htabslots);
2162         index = 0;
2163         spapr->htab_first_pass = false;
2164     }
2165     spapr->htab_save_index = index;
2166 }
2167 
2168 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2169                                 int64_t max_ns)
2170 {
2171     bool final = max_ns < 0;
2172     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2173     int examined = 0, sent = 0;
2174     int index = spapr->htab_save_index;
2175     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2176 
2177     assert(!spapr->htab_first_pass);
2178 
2179     do {
2180         int chunkstart, invalidstart;
2181 
2182         /* Consume non-dirty HPTEs */
2183         while ((index < htabslots)
2184                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2185             index++;
2186             examined++;
2187         }
2188 
2189         chunkstart = index;
2190         /* Consume valid dirty HPTEs */
2191         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2192                && HPTE_DIRTY(HPTE(spapr->htab, index))
2193                && HPTE_VALID(HPTE(spapr->htab, index))) {
2194             CLEAN_HPTE(HPTE(spapr->htab, index));
2195             index++;
2196             examined++;
2197         }
2198 
2199         invalidstart = index;
2200         /* Consume invalid dirty HPTEs */
2201         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2202                && HPTE_DIRTY(HPTE(spapr->htab, index))
2203                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2204             CLEAN_HPTE(HPTE(spapr->htab, index));
2205             index++;
2206             examined++;
2207         }
2208 
2209         if (index > chunkstart) {
2210             int n_valid = invalidstart - chunkstart;
2211             int n_invalid = index - invalidstart;
2212 
2213             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2214             sent += index - chunkstart;
2215 
2216             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2217                 break;
2218             }
2219         }
2220 
2221         if (examined >= htabslots) {
2222             break;
2223         }
2224 
2225         if (index >= htabslots) {
2226             assert(index == htabslots);
2227             index = 0;
2228         }
2229     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2230 
2231     if (index >= htabslots) {
2232         assert(index == htabslots);
2233         index = 0;
2234     }
2235 
2236     spapr->htab_save_index = index;
2237 
2238     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2239 }
2240 
2241 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2242 #define MAX_KVM_BUF_SIZE    2048
2243 
2244 static int htab_save_iterate(QEMUFile *f, void *opaque)
2245 {
2246     SpaprMachineState *spapr = opaque;
2247     int fd;
2248     int rc = 0;
2249 
2250     /* Iteration header */
2251     if (!spapr->htab_shift) {
2252         qemu_put_be32(f, -1);
2253         return 1;
2254     } else {
2255         qemu_put_be32(f, 0);
2256     }
2257 
2258     if (!spapr->htab) {
2259         assert(kvm_enabled());
2260 
2261         fd = get_htab_fd(spapr);
2262         if (fd < 0) {
2263             return fd;
2264         }
2265 
2266         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2267         if (rc < 0) {
2268             return rc;
2269         }
2270     } else  if (spapr->htab_first_pass) {
2271         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2272     } else {
2273         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2274     }
2275 
2276     htab_save_end_marker(f);
2277 
2278     return rc;
2279 }
2280 
2281 static int htab_save_complete(QEMUFile *f, void *opaque)
2282 {
2283     SpaprMachineState *spapr = opaque;
2284     int fd;
2285 
2286     /* Iteration header */
2287     if (!spapr->htab_shift) {
2288         qemu_put_be32(f, -1);
2289         return 0;
2290     } else {
2291         qemu_put_be32(f, 0);
2292     }
2293 
2294     if (!spapr->htab) {
2295         int rc;
2296 
2297         assert(kvm_enabled());
2298 
2299         fd = get_htab_fd(spapr);
2300         if (fd < 0) {
2301             return fd;
2302         }
2303 
2304         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2305         if (rc < 0) {
2306             return rc;
2307         }
2308     } else {
2309         if (spapr->htab_first_pass) {
2310             htab_save_first_pass(f, spapr, -1);
2311         }
2312         htab_save_later_pass(f, spapr, -1);
2313     }
2314 
2315     /* End marker */
2316     htab_save_end_marker(f);
2317 
2318     return 0;
2319 }
2320 
2321 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2322 {
2323     SpaprMachineState *spapr = opaque;
2324     uint32_t section_hdr;
2325     int fd = -1;
2326     Error *local_err = NULL;
2327 
2328     if (version_id < 1 || version_id > 1) {
2329         error_report("htab_load() bad version");
2330         return -EINVAL;
2331     }
2332 
2333     section_hdr = qemu_get_be32(f);
2334 
2335     if (section_hdr == -1) {
2336         spapr_free_hpt(spapr);
2337         return 0;
2338     }
2339 
2340     if (section_hdr) {
2341         /* First section gives the htab size */
2342         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2343         if (local_err) {
2344             error_report_err(local_err);
2345             return -EINVAL;
2346         }
2347         return 0;
2348     }
2349 
2350     if (!spapr->htab) {
2351         assert(kvm_enabled());
2352 
2353         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2354         if (fd < 0) {
2355             error_report_err(local_err);
2356             return fd;
2357         }
2358     }
2359 
2360     while (true) {
2361         uint32_t index;
2362         uint16_t n_valid, n_invalid;
2363 
2364         index = qemu_get_be32(f);
2365         n_valid = qemu_get_be16(f);
2366         n_invalid = qemu_get_be16(f);
2367 
2368         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2369             /* End of Stream */
2370             break;
2371         }
2372 
2373         if ((index + n_valid + n_invalid) >
2374             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2375             /* Bad index in stream */
2376             error_report(
2377                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2378                 index, n_valid, n_invalid, spapr->htab_shift);
2379             return -EINVAL;
2380         }
2381 
2382         if (spapr->htab) {
2383             if (n_valid) {
2384                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2385                                 HASH_PTE_SIZE_64 * n_valid);
2386             }
2387             if (n_invalid) {
2388                 memset(HPTE(spapr->htab, index + n_valid), 0,
2389                        HASH_PTE_SIZE_64 * n_invalid);
2390             }
2391         } else {
2392             int rc;
2393 
2394             assert(fd >= 0);
2395 
2396             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2397             if (rc < 0) {
2398                 return rc;
2399             }
2400         }
2401     }
2402 
2403     if (!spapr->htab) {
2404         assert(fd >= 0);
2405         close(fd);
2406     }
2407 
2408     return 0;
2409 }
2410 
2411 static void htab_save_cleanup(void *opaque)
2412 {
2413     SpaprMachineState *spapr = opaque;
2414 
2415     close_htab_fd(spapr);
2416 }
2417 
2418 static SaveVMHandlers savevm_htab_handlers = {
2419     .save_setup = htab_save_setup,
2420     .save_live_iterate = htab_save_iterate,
2421     .save_live_complete_precopy = htab_save_complete,
2422     .save_cleanup = htab_save_cleanup,
2423     .load_state = htab_load,
2424 };
2425 
2426 static void spapr_boot_set(void *opaque, const char *boot_device,
2427                            Error **errp)
2428 {
2429     MachineState *machine = MACHINE(opaque);
2430     machine->boot_order = g_strdup(boot_device);
2431 }
2432 
2433 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2434 {
2435     MachineState *machine = MACHINE(spapr);
2436     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2437     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2438     int i;
2439 
2440     for (i = 0; i < nr_lmbs; i++) {
2441         uint64_t addr;
2442 
2443         addr = i * lmb_size + machine->device_memory->base;
2444         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2445                                addr / lmb_size);
2446     }
2447 }
2448 
2449 /*
2450  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2451  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2452  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2453  */
2454 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2455 {
2456     int i;
2457 
2458     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2459         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2460                    " is not aligned to %" PRIu64 " MiB",
2461                    machine->ram_size,
2462                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2463         return;
2464     }
2465 
2466     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2467         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2468                    " is not aligned to %" PRIu64 " MiB",
2469                    machine->ram_size,
2470                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2471         return;
2472     }
2473 
2474     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2475         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2476             error_setg(errp,
2477                        "Node %d memory size 0x%" PRIx64
2478                        " is not aligned to %" PRIu64 " MiB",
2479                        i, machine->numa_state->nodes[i].node_mem,
2480                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2481             return;
2482         }
2483     }
2484 }
2485 
2486 /* find cpu slot in machine->possible_cpus by core_id */
2487 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2488 {
2489     int index = id / ms->smp.threads;
2490 
2491     if (index >= ms->possible_cpus->len) {
2492         return NULL;
2493     }
2494     if (idx) {
2495         *idx = index;
2496     }
2497     return &ms->possible_cpus->cpus[index];
2498 }
2499 
2500 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2501 {
2502     MachineState *ms = MACHINE(spapr);
2503     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2504     Error *local_err = NULL;
2505     bool vsmt_user = !!spapr->vsmt;
2506     int kvm_smt = kvmppc_smt_threads();
2507     int ret;
2508     unsigned int smp_threads = ms->smp.threads;
2509 
2510     if (!kvm_enabled() && (smp_threads > 1)) {
2511         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2512                      "on a pseries machine");
2513         goto out;
2514     }
2515     if (!is_power_of_2(smp_threads)) {
2516         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2517                      "machine because it must be a power of 2", smp_threads);
2518         goto out;
2519     }
2520 
2521     /* Detemine the VSMT mode to use: */
2522     if (vsmt_user) {
2523         if (spapr->vsmt < smp_threads) {
2524             error_setg(&local_err, "Cannot support VSMT mode %d"
2525                          " because it must be >= threads/core (%d)",
2526                          spapr->vsmt, smp_threads);
2527             goto out;
2528         }
2529         /* In this case, spapr->vsmt has been set by the command line */
2530     } else if (!smc->smp_threads_vsmt) {
2531         /*
2532          * Default VSMT value is tricky, because we need it to be as
2533          * consistent as possible (for migration), but this requires
2534          * changing it for at least some existing cases.  We pick 8 as
2535          * the value that we'd get with KVM on POWER8, the
2536          * overwhelmingly common case in production systems.
2537          */
2538         spapr->vsmt = MAX(8, smp_threads);
2539     } else {
2540         spapr->vsmt = smp_threads;
2541     }
2542 
2543     /* KVM: If necessary, set the SMT mode: */
2544     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2545         ret = kvmppc_set_smt_threads(spapr->vsmt);
2546         if (ret) {
2547             /* Looks like KVM isn't able to change VSMT mode */
2548             error_setg(&local_err,
2549                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2550                        spapr->vsmt, ret);
2551             /* We can live with that if the default one is big enough
2552              * for the number of threads, and a submultiple of the one
2553              * we want.  In this case we'll waste some vcpu ids, but
2554              * behaviour will be correct */
2555             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2556                 warn_report_err(local_err);
2557                 local_err = NULL;
2558                 goto out;
2559             } else {
2560                 if (!vsmt_user) {
2561                     error_append_hint(&local_err,
2562                                       "On PPC, a VM with %d threads/core"
2563                                       " on a host with %d threads/core"
2564                                       " requires the use of VSMT mode %d.\n",
2565                                       smp_threads, kvm_smt, spapr->vsmt);
2566                 }
2567                 kvmppc_hint_smt_possible(&local_err);
2568                 goto out;
2569             }
2570         }
2571     }
2572     /* else TCG: nothing to do currently */
2573 out:
2574     error_propagate(errp, local_err);
2575 }
2576 
2577 static void spapr_init_cpus(SpaprMachineState *spapr)
2578 {
2579     MachineState *machine = MACHINE(spapr);
2580     MachineClass *mc = MACHINE_GET_CLASS(machine);
2581     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2582     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2583     const CPUArchIdList *possible_cpus;
2584     unsigned int smp_cpus = machine->smp.cpus;
2585     unsigned int smp_threads = machine->smp.threads;
2586     unsigned int max_cpus = machine->smp.max_cpus;
2587     int boot_cores_nr = smp_cpus / smp_threads;
2588     int i;
2589 
2590     possible_cpus = mc->possible_cpu_arch_ids(machine);
2591     if (mc->has_hotpluggable_cpus) {
2592         if (smp_cpus % smp_threads) {
2593             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2594                          smp_cpus, smp_threads);
2595             exit(1);
2596         }
2597         if (max_cpus % smp_threads) {
2598             error_report("max_cpus (%u) must be multiple of threads (%u)",
2599                          max_cpus, smp_threads);
2600             exit(1);
2601         }
2602     } else {
2603         if (max_cpus != smp_cpus) {
2604             error_report("This machine version does not support CPU hotplug");
2605             exit(1);
2606         }
2607         boot_cores_nr = possible_cpus->len;
2608     }
2609 
2610     if (smc->pre_2_10_has_unused_icps) {
2611         int i;
2612 
2613         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2614             /* Dummy entries get deregistered when real ICPState objects
2615              * are registered during CPU core hotplug.
2616              */
2617             pre_2_10_vmstate_register_dummy_icp(i);
2618         }
2619     }
2620 
2621     for (i = 0; i < possible_cpus->len; i++) {
2622         int core_id = i * smp_threads;
2623 
2624         if (mc->has_hotpluggable_cpus) {
2625             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2626                                    spapr_vcpu_id(spapr, core_id));
2627         }
2628 
2629         if (i < boot_cores_nr) {
2630             Object *core  = object_new(type);
2631             int nr_threads = smp_threads;
2632 
2633             /* Handle the partially filled core for older machine types */
2634             if ((i + 1) * smp_threads >= smp_cpus) {
2635                 nr_threads = smp_cpus - i * smp_threads;
2636             }
2637 
2638             object_property_set_int(core, nr_threads, "nr-threads",
2639                                     &error_fatal);
2640             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2641                                     &error_fatal);
2642             object_property_set_bool(core, true, "realized", &error_fatal);
2643 
2644             object_unref(core);
2645         }
2646     }
2647 }
2648 
2649 static PCIHostState *spapr_create_default_phb(void)
2650 {
2651     DeviceState *dev;
2652 
2653     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2654     qdev_prop_set_uint32(dev, "index", 0);
2655     qdev_init_nofail(dev);
2656 
2657     return PCI_HOST_BRIDGE(dev);
2658 }
2659 
2660 /* pSeries LPAR / sPAPR hardware init */
2661 static void spapr_machine_init(MachineState *machine)
2662 {
2663     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2664     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2665     const char *kernel_filename = machine->kernel_filename;
2666     const char *initrd_filename = machine->initrd_filename;
2667     PCIHostState *phb;
2668     int i;
2669     MemoryRegion *sysmem = get_system_memory();
2670     MemoryRegion *ram = g_new(MemoryRegion, 1);
2671     hwaddr node0_size = spapr_node0_size(machine);
2672     long load_limit, fw_size;
2673     char *filename;
2674     Error *resize_hpt_err = NULL;
2675 
2676     msi_nonbroken = true;
2677 
2678     QLIST_INIT(&spapr->phbs);
2679     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2680 
2681     /* Determine capabilities to run with */
2682     spapr_caps_init(spapr);
2683 
2684     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2685     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2686         /*
2687          * If the user explicitly requested a mode we should either
2688          * supply it, or fail completely (which we do below).  But if
2689          * it's not set explicitly, we reset our mode to something
2690          * that works
2691          */
2692         if (resize_hpt_err) {
2693             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2694             error_free(resize_hpt_err);
2695             resize_hpt_err = NULL;
2696         } else {
2697             spapr->resize_hpt = smc->resize_hpt_default;
2698         }
2699     }
2700 
2701     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2702 
2703     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2704         /*
2705          * User requested HPT resize, but this host can't supply it.  Bail out
2706          */
2707         error_report_err(resize_hpt_err);
2708         exit(1);
2709     }
2710 
2711     spapr->rma_size = node0_size;
2712 
2713     /* With KVM, we don't actually know whether KVM supports an
2714      * unbounded RMA (PR KVM) or is limited by the hash table size
2715      * (HV KVM using VRMA), so we always assume the latter
2716      *
2717      * In that case, we also limit the initial allocations for RTAS
2718      * etc... to 256M since we have no way to know what the VRMA size
2719      * is going to be as it depends on the size of the hash table
2720      * which isn't determined yet.
2721      */
2722     if (kvm_enabled()) {
2723         spapr->vrma_adjust = 1;
2724         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2725     }
2726 
2727     /* Actually we don't support unbounded RMA anymore since we added
2728      * proper emulation of HV mode. The max we can get is 16G which
2729      * also happens to be what we configure for PAPR mode so make sure
2730      * we don't do anything bigger than that
2731      */
2732     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2733 
2734     if (spapr->rma_size > node0_size) {
2735         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2736                      spapr->rma_size);
2737         exit(1);
2738     }
2739 
2740     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2741     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2742 
2743     /*
2744      * VSMT must be set in order to be able to compute VCPU ids, ie to
2745      * call spapr_max_server_number() or spapr_vcpu_id().
2746      */
2747     spapr_set_vsmt_mode(spapr, &error_fatal);
2748 
2749     /* Set up Interrupt Controller before we create the VCPUs */
2750     spapr_irq_init(spapr, &error_fatal);
2751 
2752     /* Set up containers for ibm,client-architecture-support negotiated options
2753      */
2754     spapr->ov5 = spapr_ovec_new();
2755     spapr->ov5_cas = spapr_ovec_new();
2756 
2757     if (smc->dr_lmb_enabled) {
2758         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2759         spapr_validate_node_memory(machine, &error_fatal);
2760     }
2761 
2762     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2763 
2764     /* advertise support for dedicated HP event source to guests */
2765     if (spapr->use_hotplug_event_source) {
2766         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2767     }
2768 
2769     /* advertise support for HPT resizing */
2770     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2771         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2772     }
2773 
2774     /* advertise support for ibm,dyamic-memory-v2 */
2775     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2776 
2777     /* advertise XIVE on POWER9 machines */
2778     if (spapr->irq->xive) {
2779         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2780     }
2781 
2782     /* init CPUs */
2783     spapr_init_cpus(spapr);
2784 
2785     /*
2786      * check we don't have a memory-less/cpu-less NUMA node
2787      * Firmware relies on the existing memory/cpu topology to provide the
2788      * NUMA topology to the kernel.
2789      * And the linux kernel needs to know the NUMA topology at start
2790      * to be able to hotplug CPUs later.
2791      */
2792     if (machine->numa_state->num_nodes) {
2793         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2794             /* check for memory-less node */
2795             if (machine->numa_state->nodes[i].node_mem == 0) {
2796                 CPUState *cs;
2797                 int found = 0;
2798                 /* check for cpu-less node */
2799                 CPU_FOREACH(cs) {
2800                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2801                     if (cpu->node_id == i) {
2802                         found = 1;
2803                         break;
2804                     }
2805                 }
2806                 /* memory-less and cpu-less node */
2807                 if (!found) {
2808                     error_report(
2809                        "Memory-less/cpu-less nodes are not supported (node %d)",
2810                                  i);
2811                     exit(1);
2812                 }
2813             }
2814         }
2815 
2816     }
2817 
2818     /*
2819      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2820      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2821      * called from vPHB reset handler so we initialize the counter here.
2822      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2823      * must be equally distant from any other node.
2824      * The final value of spapr->gpu_numa_id is going to be written to
2825      * max-associativity-domains in spapr_build_fdt().
2826      */
2827     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2828 
2829     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2830         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2831                               spapr->max_compat_pvr)) {
2832         /* KVM and TCG always allow GTSE with radix... */
2833         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2834     }
2835     /* ... but not with hash (currently). */
2836 
2837     if (kvm_enabled()) {
2838         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2839         kvmppc_enable_logical_ci_hcalls();
2840         kvmppc_enable_set_mode_hcall();
2841 
2842         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2843         kvmppc_enable_clear_ref_mod_hcalls();
2844 
2845         /* Enable H_PAGE_INIT */
2846         kvmppc_enable_h_page_init();
2847     }
2848 
2849     /* allocate RAM */
2850     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2851                                          machine->ram_size);
2852     memory_region_add_subregion(sysmem, 0, ram);
2853 
2854     /* always allocate the device memory information */
2855     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2856 
2857     /* initialize hotplug memory address space */
2858     if (machine->ram_size < machine->maxram_size) {
2859         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2860         /*
2861          * Limit the number of hotpluggable memory slots to half the number
2862          * slots that KVM supports, leaving the other half for PCI and other
2863          * devices. However ensure that number of slots doesn't drop below 32.
2864          */
2865         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2866                            SPAPR_MAX_RAM_SLOTS;
2867 
2868         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2869             max_memslots = SPAPR_MAX_RAM_SLOTS;
2870         }
2871         if (machine->ram_slots > max_memslots) {
2872             error_report("Specified number of memory slots %"
2873                          PRIu64" exceeds max supported %d",
2874                          machine->ram_slots, max_memslots);
2875             exit(1);
2876         }
2877 
2878         machine->device_memory->base = ROUND_UP(machine->ram_size,
2879                                                 SPAPR_DEVICE_MEM_ALIGN);
2880         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2881                            "device-memory", device_mem_size);
2882         memory_region_add_subregion(sysmem, machine->device_memory->base,
2883                                     &machine->device_memory->mr);
2884     }
2885 
2886     if (smc->dr_lmb_enabled) {
2887         spapr_create_lmb_dr_connectors(spapr);
2888     }
2889 
2890     /* Set up RTAS event infrastructure */
2891     spapr_events_init(spapr);
2892 
2893     /* Set up the RTC RTAS interfaces */
2894     spapr_rtc_create(spapr);
2895 
2896     /* Set up VIO bus */
2897     spapr->vio_bus = spapr_vio_bus_init();
2898 
2899     for (i = 0; i < serial_max_hds(); i++) {
2900         if (serial_hd(i)) {
2901             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2902         }
2903     }
2904 
2905     /* We always have at least the nvram device on VIO */
2906     spapr_create_nvram(spapr);
2907 
2908     /*
2909      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2910      * connectors (described in root DT node's "ibm,drc-types" property)
2911      * are pre-initialized here. additional child connectors (such as
2912      * connectors for a PHBs PCI slots) are added as needed during their
2913      * parent's realization.
2914      */
2915     if (smc->dr_phb_enabled) {
2916         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2917             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2918         }
2919     }
2920 
2921     /* Set up PCI */
2922     spapr_pci_rtas_init();
2923 
2924     phb = spapr_create_default_phb();
2925 
2926     for (i = 0; i < nb_nics; i++) {
2927         NICInfo *nd = &nd_table[i];
2928 
2929         if (!nd->model) {
2930             nd->model = g_strdup("spapr-vlan");
2931         }
2932 
2933         if (g_str_equal(nd->model, "spapr-vlan") ||
2934             g_str_equal(nd->model, "ibmveth")) {
2935             spapr_vlan_create(spapr->vio_bus, nd);
2936         } else {
2937             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2938         }
2939     }
2940 
2941     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2942         spapr_vscsi_create(spapr->vio_bus);
2943     }
2944 
2945     /* Graphics */
2946     if (spapr_vga_init(phb->bus, &error_fatal)) {
2947         spapr->has_graphics = true;
2948         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2949     }
2950 
2951     if (machine->usb) {
2952         if (smc->use_ohci_by_default) {
2953             pci_create_simple(phb->bus, -1, "pci-ohci");
2954         } else {
2955             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2956         }
2957 
2958         if (spapr->has_graphics) {
2959             USBBus *usb_bus = usb_bus_find(-1);
2960 
2961             usb_create_simple(usb_bus, "usb-kbd");
2962             usb_create_simple(usb_bus, "usb-mouse");
2963         }
2964     }
2965 
2966     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2967         error_report(
2968             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2969             MIN_RMA_SLOF);
2970         exit(1);
2971     }
2972 
2973     if (kernel_filename) {
2974         uint64_t lowaddr = 0;
2975 
2976         spapr->kernel_size = load_elf(kernel_filename, NULL,
2977                                       translate_kernel_address, NULL,
2978                                       NULL, &lowaddr, NULL, 1,
2979                                       PPC_ELF_MACHINE, 0, 0);
2980         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2981             spapr->kernel_size = load_elf(kernel_filename, NULL,
2982                                           translate_kernel_address, NULL, NULL,
2983                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2984                                           0, 0);
2985             spapr->kernel_le = spapr->kernel_size > 0;
2986         }
2987         if (spapr->kernel_size < 0) {
2988             error_report("error loading %s: %s", kernel_filename,
2989                          load_elf_strerror(spapr->kernel_size));
2990             exit(1);
2991         }
2992 
2993         /* load initrd */
2994         if (initrd_filename) {
2995             /* Try to locate the initrd in the gap between the kernel
2996              * and the firmware. Add a bit of space just in case
2997              */
2998             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2999                                   + 0x1ffff) & ~0xffff;
3000             spapr->initrd_size = load_image_targphys(initrd_filename,
3001                                                      spapr->initrd_base,
3002                                                      load_limit
3003                                                      - spapr->initrd_base);
3004             if (spapr->initrd_size < 0) {
3005                 error_report("could not load initial ram disk '%s'",
3006                              initrd_filename);
3007                 exit(1);
3008             }
3009         }
3010     }
3011 
3012     if (bios_name == NULL) {
3013         bios_name = FW_FILE_NAME;
3014     }
3015     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3016     if (!filename) {
3017         error_report("Could not find LPAR firmware '%s'", bios_name);
3018         exit(1);
3019     }
3020     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3021     if (fw_size <= 0) {
3022         error_report("Could not load LPAR firmware '%s'", filename);
3023         exit(1);
3024     }
3025     g_free(filename);
3026 
3027     /* FIXME: Should register things through the MachineState's qdev
3028      * interface, this is a legacy from the sPAPREnvironment structure
3029      * which predated MachineState but had a similar function */
3030     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3031     register_savevm_live("spapr/htab", -1, 1,
3032                          &savevm_htab_handlers, spapr);
3033 
3034     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3035                              &error_fatal);
3036 
3037     qemu_register_boot_set(spapr_boot_set, spapr);
3038 
3039     /*
3040      * Nothing needs to be done to resume a suspended guest because
3041      * suspending does not change the machine state, so no need for
3042      * a ->wakeup method.
3043      */
3044     qemu_register_wakeup_support();
3045 
3046     if (kvm_enabled()) {
3047         /* to stop and start vmclock */
3048         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3049                                          &spapr->tb);
3050 
3051         kvmppc_spapr_enable_inkernel_multitce();
3052     }
3053 }
3054 
3055 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3056 {
3057     if (!vm_type) {
3058         return 0;
3059     }
3060 
3061     if (!strcmp(vm_type, "HV")) {
3062         return 1;
3063     }
3064 
3065     if (!strcmp(vm_type, "PR")) {
3066         return 2;
3067     }
3068 
3069     error_report("Unknown kvm-type specified '%s'", vm_type);
3070     exit(1);
3071 }
3072 
3073 /*
3074  * Implementation of an interface to adjust firmware path
3075  * for the bootindex property handling.
3076  */
3077 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3078                                    DeviceState *dev)
3079 {
3080 #define CAST(type, obj, name) \
3081     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3082     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3083     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3084     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3085 
3086     if (d) {
3087         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3088         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3089         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3090 
3091         if (spapr) {
3092             /*
3093              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3094              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3095              * 0x8000 | (target << 8) | (bus << 5) | lun
3096              * (see the "Logical unit addressing format" table in SAM5)
3097              */
3098             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3099             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3100                                    (uint64_t)id << 48);
3101         } else if (virtio) {
3102             /*
3103              * We use SRP luns of the form 01000000 | (target << 8) | lun
3104              * in the top 32 bits of the 64-bit LUN
3105              * Note: the quote above is from SLOF and it is wrong,
3106              * the actual binding is:
3107              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3108              */
3109             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3110             if (d->lun >= 256) {
3111                 /* Use the LUN "flat space addressing method" */
3112                 id |= 0x4000;
3113             }
3114             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3115                                    (uint64_t)id << 32);
3116         } else if (usb) {
3117             /*
3118              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3119              * in the top 32 bits of the 64-bit LUN
3120              */
3121             unsigned usb_port = atoi(usb->port->path);
3122             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3123             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3124                                    (uint64_t)id << 32);
3125         }
3126     }
3127 
3128     /*
3129      * SLOF probes the USB devices, and if it recognizes that the device is a
3130      * storage device, it changes its name to "storage" instead of "usb-host",
3131      * and additionally adds a child node for the SCSI LUN, so the correct
3132      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3133      */
3134     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3135         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3136         if (usb_host_dev_is_scsi_storage(usbdev)) {
3137             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3138         }
3139     }
3140 
3141     if (phb) {
3142         /* Replace "pci" with "pci@800000020000000" */
3143         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3144     }
3145 
3146     if (vsc) {
3147         /* Same logic as virtio above */
3148         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3149         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3150     }
3151 
3152     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3153         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3154         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3155         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3156     }
3157 
3158     return NULL;
3159 }
3160 
3161 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3162 {
3163     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3164 
3165     return g_strdup(spapr->kvm_type);
3166 }
3167 
3168 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3169 {
3170     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3171 
3172     g_free(spapr->kvm_type);
3173     spapr->kvm_type = g_strdup(value);
3174 }
3175 
3176 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3177 {
3178     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3179 
3180     return spapr->use_hotplug_event_source;
3181 }
3182 
3183 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3184                                             Error **errp)
3185 {
3186     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3187 
3188     spapr->use_hotplug_event_source = value;
3189 }
3190 
3191 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3192 {
3193     return true;
3194 }
3195 
3196 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3197 {
3198     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3199 
3200     switch (spapr->resize_hpt) {
3201     case SPAPR_RESIZE_HPT_DEFAULT:
3202         return g_strdup("default");
3203     case SPAPR_RESIZE_HPT_DISABLED:
3204         return g_strdup("disabled");
3205     case SPAPR_RESIZE_HPT_ENABLED:
3206         return g_strdup("enabled");
3207     case SPAPR_RESIZE_HPT_REQUIRED:
3208         return g_strdup("required");
3209     }
3210     g_assert_not_reached();
3211 }
3212 
3213 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3214 {
3215     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3216 
3217     if (strcmp(value, "default") == 0) {
3218         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3219     } else if (strcmp(value, "disabled") == 0) {
3220         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3221     } else if (strcmp(value, "enabled") == 0) {
3222         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3223     } else if (strcmp(value, "required") == 0) {
3224         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3225     } else {
3226         error_setg(errp, "Bad value for \"resize-hpt\" property");
3227     }
3228 }
3229 
3230 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3231                                    void *opaque, Error **errp)
3232 {
3233     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3234 }
3235 
3236 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3237                                    void *opaque, Error **errp)
3238 {
3239     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3240 }
3241 
3242 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3243 {
3244     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3245 
3246     if (spapr->irq == &spapr_irq_xics_legacy) {
3247         return g_strdup("legacy");
3248     } else if (spapr->irq == &spapr_irq_xics) {
3249         return g_strdup("xics");
3250     } else if (spapr->irq == &spapr_irq_xive) {
3251         return g_strdup("xive");
3252     } else if (spapr->irq == &spapr_irq_dual) {
3253         return g_strdup("dual");
3254     }
3255     g_assert_not_reached();
3256 }
3257 
3258 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3259 {
3260     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3261 
3262     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3263         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3264         return;
3265     }
3266 
3267     /* The legacy IRQ backend can not be set */
3268     if (strcmp(value, "xics") == 0) {
3269         spapr->irq = &spapr_irq_xics;
3270     } else if (strcmp(value, "xive") == 0) {
3271         spapr->irq = &spapr_irq_xive;
3272     } else if (strcmp(value, "dual") == 0) {
3273         spapr->irq = &spapr_irq_dual;
3274     } else {
3275         error_setg(errp, "Bad value for \"ic-mode\" property");
3276     }
3277 }
3278 
3279 static char *spapr_get_host_model(Object *obj, Error **errp)
3280 {
3281     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3282 
3283     return g_strdup(spapr->host_model);
3284 }
3285 
3286 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3287 {
3288     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3289 
3290     g_free(spapr->host_model);
3291     spapr->host_model = g_strdup(value);
3292 }
3293 
3294 static char *spapr_get_host_serial(Object *obj, Error **errp)
3295 {
3296     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3297 
3298     return g_strdup(spapr->host_serial);
3299 }
3300 
3301 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3302 {
3303     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3304 
3305     g_free(spapr->host_serial);
3306     spapr->host_serial = g_strdup(value);
3307 }
3308 
3309 static void spapr_instance_init(Object *obj)
3310 {
3311     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3312     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3313 
3314     spapr->htab_fd = -1;
3315     spapr->use_hotplug_event_source = true;
3316     object_property_add_str(obj, "kvm-type",
3317                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3318     object_property_set_description(obj, "kvm-type",
3319                                     "Specifies the KVM virtualization mode (HV, PR)",
3320                                     NULL);
3321     object_property_add_bool(obj, "modern-hotplug-events",
3322                             spapr_get_modern_hotplug_events,
3323                             spapr_set_modern_hotplug_events,
3324                             NULL);
3325     object_property_set_description(obj, "modern-hotplug-events",
3326                                     "Use dedicated hotplug event mechanism in"
3327                                     " place of standard EPOW events when possible"
3328                                     " (required for memory hot-unplug support)",
3329                                     NULL);
3330     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3331                             "Maximum permitted CPU compatibility mode",
3332                             &error_fatal);
3333 
3334     object_property_add_str(obj, "resize-hpt",
3335                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3336     object_property_set_description(obj, "resize-hpt",
3337                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3338                                     NULL);
3339     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3340                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3341     object_property_set_description(obj, "vsmt",
3342                                     "Virtual SMT: KVM behaves as if this were"
3343                                     " the host's SMT mode", &error_abort);
3344     object_property_add_bool(obj, "vfio-no-msix-emulation",
3345                              spapr_get_msix_emulation, NULL, NULL);
3346 
3347     /* The machine class defines the default interrupt controller mode */
3348     spapr->irq = smc->irq;
3349     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3350                             spapr_set_ic_mode, NULL);
3351     object_property_set_description(obj, "ic-mode",
3352                  "Specifies the interrupt controller mode (xics, xive, dual)",
3353                  NULL);
3354 
3355     object_property_add_str(obj, "host-model",
3356         spapr_get_host_model, spapr_set_host_model,
3357         &error_abort);
3358     object_property_set_description(obj, "host-model",
3359         "Host model to advertise in guest device tree", &error_abort);
3360     object_property_add_str(obj, "host-serial",
3361         spapr_get_host_serial, spapr_set_host_serial,
3362         &error_abort);
3363     object_property_set_description(obj, "host-serial",
3364         "Host serial number to advertise in guest device tree", &error_abort);
3365 }
3366 
3367 static void spapr_machine_finalizefn(Object *obj)
3368 {
3369     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3370 
3371     g_free(spapr->kvm_type);
3372 }
3373 
3374 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3375 {
3376     cpu_synchronize_state(cs);
3377     ppc_cpu_do_system_reset(cs);
3378 }
3379 
3380 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3381 {
3382     CPUState *cs;
3383 
3384     CPU_FOREACH(cs) {
3385         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3386     }
3387 }
3388 
3389 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3390                           void *fdt, int *fdt_start_offset, Error **errp)
3391 {
3392     uint64_t addr;
3393     uint32_t node;
3394 
3395     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3396     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3397                                     &error_abort);
3398     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3399                                                    SPAPR_MEMORY_BLOCK_SIZE);
3400     return 0;
3401 }
3402 
3403 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3404                            bool dedicated_hp_event_source, Error **errp)
3405 {
3406     SpaprDrc *drc;
3407     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3408     int i;
3409     uint64_t addr = addr_start;
3410     bool hotplugged = spapr_drc_hotplugged(dev);
3411     Error *local_err = NULL;
3412 
3413     for (i = 0; i < nr_lmbs; i++) {
3414         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3415                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3416         g_assert(drc);
3417 
3418         spapr_drc_attach(drc, dev, &local_err);
3419         if (local_err) {
3420             while (addr > addr_start) {
3421                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3422                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3423                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3424                 spapr_drc_detach(drc);
3425             }
3426             error_propagate(errp, local_err);
3427             return;
3428         }
3429         if (!hotplugged) {
3430             spapr_drc_reset(drc);
3431         }
3432         addr += SPAPR_MEMORY_BLOCK_SIZE;
3433     }
3434     /* send hotplug notification to the
3435      * guest only in case of hotplugged memory
3436      */
3437     if (hotplugged) {
3438         if (dedicated_hp_event_source) {
3439             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3440                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3441             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3442                                                    nr_lmbs,
3443                                                    spapr_drc_index(drc));
3444         } else {
3445             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3446                                            nr_lmbs);
3447         }
3448     }
3449 }
3450 
3451 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3452                               Error **errp)
3453 {
3454     Error *local_err = NULL;
3455     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3456     PCDIMMDevice *dimm = PC_DIMM(dev);
3457     uint64_t size, addr;
3458 
3459     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3460 
3461     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3462     if (local_err) {
3463         goto out;
3464     }
3465 
3466     addr = object_property_get_uint(OBJECT(dimm),
3467                                     PC_DIMM_ADDR_PROP, &local_err);
3468     if (local_err) {
3469         goto out_unplug;
3470     }
3471 
3472     spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3473                    &local_err);
3474     if (local_err) {
3475         goto out_unplug;
3476     }
3477 
3478     return;
3479 
3480 out_unplug:
3481     pc_dimm_unplug(dimm, MACHINE(ms));
3482 out:
3483     error_propagate(errp, local_err);
3484 }
3485 
3486 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3487                                   Error **errp)
3488 {
3489     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3490     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3491     PCDIMMDevice *dimm = PC_DIMM(dev);
3492     Error *local_err = NULL;
3493     uint64_t size;
3494     Object *memdev;
3495     hwaddr pagesize;
3496 
3497     if (!smc->dr_lmb_enabled) {
3498         error_setg(errp, "Memory hotplug not supported for this machine");
3499         return;
3500     }
3501 
3502     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3503     if (local_err) {
3504         error_propagate(errp, local_err);
3505         return;
3506     }
3507 
3508     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3509         error_setg(errp, "Hotplugged memory size must be a multiple of "
3510                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3511         return;
3512     }
3513 
3514     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3515                                       &error_abort);
3516     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3517     spapr_check_pagesize(spapr, pagesize, &local_err);
3518     if (local_err) {
3519         error_propagate(errp, local_err);
3520         return;
3521     }
3522 
3523     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3524 }
3525 
3526 struct SpaprDimmState {
3527     PCDIMMDevice *dimm;
3528     uint32_t nr_lmbs;
3529     QTAILQ_ENTRY(SpaprDimmState) next;
3530 };
3531 
3532 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3533                                                        PCDIMMDevice *dimm)
3534 {
3535     SpaprDimmState *dimm_state = NULL;
3536 
3537     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3538         if (dimm_state->dimm == dimm) {
3539             break;
3540         }
3541     }
3542     return dimm_state;
3543 }
3544 
3545 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3546                                                       uint32_t nr_lmbs,
3547                                                       PCDIMMDevice *dimm)
3548 {
3549     SpaprDimmState *ds = NULL;
3550 
3551     /*
3552      * If this request is for a DIMM whose removal had failed earlier
3553      * (due to guest's refusal to remove the LMBs), we would have this
3554      * dimm already in the pending_dimm_unplugs list. In that
3555      * case don't add again.
3556      */
3557     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3558     if (!ds) {
3559         ds = g_malloc0(sizeof(SpaprDimmState));
3560         ds->nr_lmbs = nr_lmbs;
3561         ds->dimm = dimm;
3562         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3563     }
3564     return ds;
3565 }
3566 
3567 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3568                                               SpaprDimmState *dimm_state)
3569 {
3570     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3571     g_free(dimm_state);
3572 }
3573 
3574 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3575                                                         PCDIMMDevice *dimm)
3576 {
3577     SpaprDrc *drc;
3578     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3579                                                   &error_abort);
3580     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3581     uint32_t avail_lmbs = 0;
3582     uint64_t addr_start, addr;
3583     int i;
3584 
3585     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3586                                          &error_abort);
3587 
3588     addr = addr_start;
3589     for (i = 0; i < nr_lmbs; i++) {
3590         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3591                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3592         g_assert(drc);
3593         if (drc->dev) {
3594             avail_lmbs++;
3595         }
3596         addr += SPAPR_MEMORY_BLOCK_SIZE;
3597     }
3598 
3599     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3600 }
3601 
3602 /* Callback to be called during DRC release. */
3603 void spapr_lmb_release(DeviceState *dev)
3604 {
3605     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3606     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3607     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3608 
3609     /* This information will get lost if a migration occurs
3610      * during the unplug process. In this case recover it. */
3611     if (ds == NULL) {
3612         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3613         g_assert(ds);
3614         /* The DRC being examined by the caller at least must be counted */
3615         g_assert(ds->nr_lmbs);
3616     }
3617 
3618     if (--ds->nr_lmbs) {
3619         return;
3620     }
3621 
3622     /*
3623      * Now that all the LMBs have been removed by the guest, call the
3624      * unplug handler chain. This can never fail.
3625      */
3626     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3627     object_unparent(OBJECT(dev));
3628 }
3629 
3630 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3631 {
3632     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3633     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3634 
3635     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3636     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3637     spapr_pending_dimm_unplugs_remove(spapr, ds);
3638 }
3639 
3640 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3641                                         DeviceState *dev, Error **errp)
3642 {
3643     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3644     Error *local_err = NULL;
3645     PCDIMMDevice *dimm = PC_DIMM(dev);
3646     uint32_t nr_lmbs;
3647     uint64_t size, addr_start, addr;
3648     int i;
3649     SpaprDrc *drc;
3650 
3651     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3652     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3653 
3654     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3655                                          &local_err);
3656     if (local_err) {
3657         goto out;
3658     }
3659 
3660     /*
3661      * An existing pending dimm state for this DIMM means that there is an
3662      * unplug operation in progress, waiting for the spapr_lmb_release
3663      * callback to complete the job (BQL can't cover that far). In this case,
3664      * bail out to avoid detaching DRCs that were already released.
3665      */
3666     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3667         error_setg(&local_err,
3668                    "Memory unplug already in progress for device %s",
3669                    dev->id);
3670         goto out;
3671     }
3672 
3673     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3674 
3675     addr = addr_start;
3676     for (i = 0; i < nr_lmbs; i++) {
3677         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3678                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3679         g_assert(drc);
3680 
3681         spapr_drc_detach(drc);
3682         addr += SPAPR_MEMORY_BLOCK_SIZE;
3683     }
3684 
3685     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3686                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3687     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3688                                               nr_lmbs, spapr_drc_index(drc));
3689 out:
3690     error_propagate(errp, local_err);
3691 }
3692 
3693 /* Callback to be called during DRC release. */
3694 void spapr_core_release(DeviceState *dev)
3695 {
3696     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3697 
3698     /* Call the unplug handler chain. This can never fail. */
3699     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3700     object_unparent(OBJECT(dev));
3701 }
3702 
3703 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3704 {
3705     MachineState *ms = MACHINE(hotplug_dev);
3706     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3707     CPUCore *cc = CPU_CORE(dev);
3708     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3709 
3710     if (smc->pre_2_10_has_unused_icps) {
3711         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3712         int i;
3713 
3714         for (i = 0; i < cc->nr_threads; i++) {
3715             CPUState *cs = CPU(sc->threads[i]);
3716 
3717             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3718         }
3719     }
3720 
3721     assert(core_slot);
3722     core_slot->cpu = NULL;
3723     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3724 }
3725 
3726 static
3727 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3728                                Error **errp)
3729 {
3730     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3731     int index;
3732     SpaprDrc *drc;
3733     CPUCore *cc = CPU_CORE(dev);
3734 
3735     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3736         error_setg(errp, "Unable to find CPU core with core-id: %d",
3737                    cc->core_id);
3738         return;
3739     }
3740     if (index == 0) {
3741         error_setg(errp, "Boot CPU core may not be unplugged");
3742         return;
3743     }
3744 
3745     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3746                           spapr_vcpu_id(spapr, cc->core_id));
3747     g_assert(drc);
3748 
3749     if (!spapr_drc_unplug_requested(drc)) {
3750         spapr_drc_detach(drc);
3751         spapr_hotplug_req_remove_by_index(drc);
3752     }
3753 }
3754 
3755 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3756                            void *fdt, int *fdt_start_offset, Error **errp)
3757 {
3758     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3759     CPUState *cs = CPU(core->threads[0]);
3760     PowerPCCPU *cpu = POWERPC_CPU(cs);
3761     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3762     int id = spapr_get_vcpu_id(cpu);
3763     char *nodename;
3764     int offset;
3765 
3766     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3767     offset = fdt_add_subnode(fdt, 0, nodename);
3768     g_free(nodename);
3769 
3770     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3771 
3772     *fdt_start_offset = offset;
3773     return 0;
3774 }
3775 
3776 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3777                             Error **errp)
3778 {
3779     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3780     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3781     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3782     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3783     CPUCore *cc = CPU_CORE(dev);
3784     CPUState *cs;
3785     SpaprDrc *drc;
3786     Error *local_err = NULL;
3787     CPUArchId *core_slot;
3788     int index;
3789     bool hotplugged = spapr_drc_hotplugged(dev);
3790     int i;
3791 
3792     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3793     if (!core_slot) {
3794         error_setg(errp, "Unable to find CPU core with core-id: %d",
3795                    cc->core_id);
3796         return;
3797     }
3798     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3799                           spapr_vcpu_id(spapr, cc->core_id));
3800 
3801     g_assert(drc || !mc->has_hotpluggable_cpus);
3802 
3803     if (drc) {
3804         spapr_drc_attach(drc, dev, &local_err);
3805         if (local_err) {
3806             error_propagate(errp, local_err);
3807             return;
3808         }
3809 
3810         if (hotplugged) {
3811             /*
3812              * Send hotplug notification interrupt to the guest only
3813              * in case of hotplugged CPUs.
3814              */
3815             spapr_hotplug_req_add_by_index(drc);
3816         } else {
3817             spapr_drc_reset(drc);
3818         }
3819     }
3820 
3821     core_slot->cpu = OBJECT(dev);
3822 
3823     if (smc->pre_2_10_has_unused_icps) {
3824         for (i = 0; i < cc->nr_threads; i++) {
3825             cs = CPU(core->threads[i]);
3826             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3827         }
3828     }
3829 
3830     /*
3831      * Set compatibility mode to match the boot CPU, which was either set
3832      * by the machine reset code or by CAS.
3833      */
3834     if (hotplugged) {
3835         for (i = 0; i < cc->nr_threads; i++) {
3836             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3837                            &local_err);
3838             if (local_err) {
3839                 error_propagate(errp, local_err);
3840                 return;
3841             }
3842         }
3843     }
3844 }
3845 
3846 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3847                                 Error **errp)
3848 {
3849     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3850     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3851     Error *local_err = NULL;
3852     CPUCore *cc = CPU_CORE(dev);
3853     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3854     const char *type = object_get_typename(OBJECT(dev));
3855     CPUArchId *core_slot;
3856     int index;
3857     unsigned int smp_threads = machine->smp.threads;
3858 
3859     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3860         error_setg(&local_err, "CPU hotplug not supported for this machine");
3861         goto out;
3862     }
3863 
3864     if (strcmp(base_core_type, type)) {
3865         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3866         goto out;
3867     }
3868 
3869     if (cc->core_id % smp_threads) {
3870         error_setg(&local_err, "invalid core id %d", cc->core_id);
3871         goto out;
3872     }
3873 
3874     /*
3875      * In general we should have homogeneous threads-per-core, but old
3876      * (pre hotplug support) machine types allow the last core to have
3877      * reduced threads as a compatibility hack for when we allowed
3878      * total vcpus not a multiple of threads-per-core.
3879      */
3880     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3881         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3882                    cc->nr_threads, smp_threads);
3883         goto out;
3884     }
3885 
3886     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3887     if (!core_slot) {
3888         error_setg(&local_err, "core id %d out of range", cc->core_id);
3889         goto out;
3890     }
3891 
3892     if (core_slot->cpu) {
3893         error_setg(&local_err, "core %d already populated", cc->core_id);
3894         goto out;
3895     }
3896 
3897     numa_cpu_pre_plug(core_slot, dev, &local_err);
3898 
3899 out:
3900     error_propagate(errp, local_err);
3901 }
3902 
3903 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3904                           void *fdt, int *fdt_start_offset, Error **errp)
3905 {
3906     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3907     int intc_phandle;
3908 
3909     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3910     if (intc_phandle <= 0) {
3911         return -1;
3912     }
3913 
3914     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3915         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3916         return -1;
3917     }
3918 
3919     /* generally SLOF creates these, for hotplug it's up to QEMU */
3920     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3921 
3922     return 0;
3923 }
3924 
3925 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3926                                Error **errp)
3927 {
3928     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3929     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3930     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3931     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3932 
3933     if (dev->hotplugged && !smc->dr_phb_enabled) {
3934         error_setg(errp, "PHB hotplug not supported for this machine");
3935         return;
3936     }
3937 
3938     if (sphb->index == (uint32_t)-1) {
3939         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3940         return;
3941     }
3942 
3943     /*
3944      * This will check that sphb->index doesn't exceed the maximum number of
3945      * PHBs for the current machine type.
3946      */
3947     smc->phb_placement(spapr, sphb->index,
3948                        &sphb->buid, &sphb->io_win_addr,
3949                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3950                        windows_supported, sphb->dma_liobn,
3951                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3952                        errp);
3953 }
3954 
3955 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3956                            Error **errp)
3957 {
3958     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3959     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3960     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3961     SpaprDrc *drc;
3962     bool hotplugged = spapr_drc_hotplugged(dev);
3963     Error *local_err = NULL;
3964 
3965     if (!smc->dr_phb_enabled) {
3966         return;
3967     }
3968 
3969     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3970     /* hotplug hooks should check it's enabled before getting this far */
3971     assert(drc);
3972 
3973     spapr_drc_attach(drc, DEVICE(dev), &local_err);
3974     if (local_err) {
3975         error_propagate(errp, local_err);
3976         return;
3977     }
3978 
3979     if (hotplugged) {
3980         spapr_hotplug_req_add_by_index(drc);
3981     } else {
3982         spapr_drc_reset(drc);
3983     }
3984 }
3985 
3986 void spapr_phb_release(DeviceState *dev)
3987 {
3988     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3989 
3990     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3991     object_unparent(OBJECT(dev));
3992 }
3993 
3994 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3995 {
3996     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3997 }
3998 
3999 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4000                                      DeviceState *dev, Error **errp)
4001 {
4002     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4003     SpaprDrc *drc;
4004 
4005     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4006     assert(drc);
4007 
4008     if (!spapr_drc_unplug_requested(drc)) {
4009         spapr_drc_detach(drc);
4010         spapr_hotplug_req_remove_by_index(drc);
4011     }
4012 }
4013 
4014 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4015                                  Error **errp)
4016 {
4017     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4018     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4019 
4020     if (spapr->tpm_proxy != NULL) {
4021         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4022         return;
4023     }
4024 
4025     spapr->tpm_proxy = tpm_proxy;
4026 }
4027 
4028 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4029 {
4030     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4031 
4032     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4033     object_unparent(OBJECT(dev));
4034     spapr->tpm_proxy = NULL;
4035 }
4036 
4037 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4038                                       DeviceState *dev, Error **errp)
4039 {
4040     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4041         spapr_memory_plug(hotplug_dev, dev, errp);
4042     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4043         spapr_core_plug(hotplug_dev, dev, errp);
4044     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4045         spapr_phb_plug(hotplug_dev, dev, errp);
4046     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4047         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4048     }
4049 }
4050 
4051 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4052                                         DeviceState *dev, Error **errp)
4053 {
4054     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4055         spapr_memory_unplug(hotplug_dev, dev);
4056     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4057         spapr_core_unplug(hotplug_dev, dev);
4058     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4059         spapr_phb_unplug(hotplug_dev, dev);
4060     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4061         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4062     }
4063 }
4064 
4065 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4066                                                 DeviceState *dev, Error **errp)
4067 {
4068     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4069     MachineClass *mc = MACHINE_GET_CLASS(sms);
4070     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4071 
4072     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4073         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4074             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4075         } else {
4076             /* NOTE: this means there is a window after guest reset, prior to
4077              * CAS negotiation, where unplug requests will fail due to the
4078              * capability not being detected yet. This is a bit different than
4079              * the case with PCI unplug, where the events will be queued and
4080              * eventually handled by the guest after boot
4081              */
4082             error_setg(errp, "Memory hot unplug not supported for this guest");
4083         }
4084     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4085         if (!mc->has_hotpluggable_cpus) {
4086             error_setg(errp, "CPU hot unplug not supported on this machine");
4087             return;
4088         }
4089         spapr_core_unplug_request(hotplug_dev, dev, errp);
4090     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4091         if (!smc->dr_phb_enabled) {
4092             error_setg(errp, "PHB hot unplug not supported on this machine");
4093             return;
4094         }
4095         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4096     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4097         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4098     }
4099 }
4100 
4101 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4102                                           DeviceState *dev, Error **errp)
4103 {
4104     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4105         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4106     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4107         spapr_core_pre_plug(hotplug_dev, dev, errp);
4108     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4109         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4110     }
4111 }
4112 
4113 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4114                                                  DeviceState *dev)
4115 {
4116     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4117         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4118         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4119         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4120         return HOTPLUG_HANDLER(machine);
4121     }
4122     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4123         PCIDevice *pcidev = PCI_DEVICE(dev);
4124         PCIBus *root = pci_device_root_bus(pcidev);
4125         SpaprPhbState *phb =
4126             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4127                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4128 
4129         if (phb) {
4130             return HOTPLUG_HANDLER(phb);
4131         }
4132     }
4133     return NULL;
4134 }
4135 
4136 static CpuInstanceProperties
4137 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4138 {
4139     CPUArchId *core_slot;
4140     MachineClass *mc = MACHINE_GET_CLASS(machine);
4141 
4142     /* make sure possible_cpu are intialized */
4143     mc->possible_cpu_arch_ids(machine);
4144     /* get CPU core slot containing thread that matches cpu_index */
4145     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4146     assert(core_slot);
4147     return core_slot->props;
4148 }
4149 
4150 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4151 {
4152     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4153 }
4154 
4155 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4156 {
4157     int i;
4158     unsigned int smp_threads = machine->smp.threads;
4159     unsigned int smp_cpus = machine->smp.cpus;
4160     const char *core_type;
4161     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4162     MachineClass *mc = MACHINE_GET_CLASS(machine);
4163 
4164     if (!mc->has_hotpluggable_cpus) {
4165         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4166     }
4167     if (machine->possible_cpus) {
4168         assert(machine->possible_cpus->len == spapr_max_cores);
4169         return machine->possible_cpus;
4170     }
4171 
4172     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4173     if (!core_type) {
4174         error_report("Unable to find sPAPR CPU Core definition");
4175         exit(1);
4176     }
4177 
4178     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4179                              sizeof(CPUArchId) * spapr_max_cores);
4180     machine->possible_cpus->len = spapr_max_cores;
4181     for (i = 0; i < machine->possible_cpus->len; i++) {
4182         int core_id = i * smp_threads;
4183 
4184         machine->possible_cpus->cpus[i].type = core_type;
4185         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4186         machine->possible_cpus->cpus[i].arch_id = core_id;
4187         machine->possible_cpus->cpus[i].props.has_core_id = true;
4188         machine->possible_cpus->cpus[i].props.core_id = core_id;
4189     }
4190     return machine->possible_cpus;
4191 }
4192 
4193 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4194                                 uint64_t *buid, hwaddr *pio,
4195                                 hwaddr *mmio32, hwaddr *mmio64,
4196                                 unsigned n_dma, uint32_t *liobns,
4197                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4198 {
4199     /*
4200      * New-style PHB window placement.
4201      *
4202      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4203      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4204      * windows.
4205      *
4206      * Some guest kernels can't work with MMIO windows above 1<<46
4207      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4208      *
4209      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4210      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4211      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4212      * 1TiB 64-bit MMIO windows for each PHB.
4213      */
4214     const uint64_t base_buid = 0x800000020000000ULL;
4215     int i;
4216 
4217     /* Sanity check natural alignments */
4218     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4219     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4220     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4221     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4222     /* Sanity check bounds */
4223     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4224                       SPAPR_PCI_MEM32_WIN_SIZE);
4225     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4226                       SPAPR_PCI_MEM64_WIN_SIZE);
4227 
4228     if (index >= SPAPR_MAX_PHBS) {
4229         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4230                    SPAPR_MAX_PHBS - 1);
4231         return;
4232     }
4233 
4234     *buid = base_buid + index;
4235     for (i = 0; i < n_dma; ++i) {
4236         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4237     }
4238 
4239     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4240     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4241     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4242 
4243     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4244     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4245 }
4246 
4247 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4248 {
4249     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4250 
4251     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4252 }
4253 
4254 static void spapr_ics_resend(XICSFabric *dev)
4255 {
4256     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4257 
4258     ics_resend(spapr->ics);
4259 }
4260 
4261 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4262 {
4263     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4264 
4265     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4266 }
4267 
4268 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4269                                  Monitor *mon)
4270 {
4271     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4272 
4273     spapr_irq_print_info(spapr, mon);
4274     monitor_printf(mon, "irqchip: %s\n",
4275                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4276 }
4277 
4278 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4279 {
4280     return cpu->vcpu_id;
4281 }
4282 
4283 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4284 {
4285     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4286     MachineState *ms = MACHINE(spapr);
4287     int vcpu_id;
4288 
4289     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4290 
4291     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4292         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4293         error_append_hint(errp, "Adjust the number of cpus to %d "
4294                           "or try to raise the number of threads per core\n",
4295                           vcpu_id * ms->smp.threads / spapr->vsmt);
4296         return;
4297     }
4298 
4299     cpu->vcpu_id = vcpu_id;
4300 }
4301 
4302 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4303 {
4304     CPUState *cs;
4305 
4306     CPU_FOREACH(cs) {
4307         PowerPCCPU *cpu = POWERPC_CPU(cs);
4308 
4309         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4310             return cpu;
4311         }
4312     }
4313 
4314     return NULL;
4315 }
4316 
4317 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4318 {
4319     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4320 
4321     /* These are only called by TCG, KVM maintains dispatch state */
4322 
4323     spapr_cpu->prod = false;
4324     if (spapr_cpu->vpa_addr) {
4325         CPUState *cs = CPU(cpu);
4326         uint32_t dispatch;
4327 
4328         dispatch = ldl_be_phys(cs->as,
4329                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4330         dispatch++;
4331         if ((dispatch & 1) != 0) {
4332             qemu_log_mask(LOG_GUEST_ERROR,
4333                           "VPA: incorrect dispatch counter value for "
4334                           "dispatched partition %u, correcting.\n", dispatch);
4335             dispatch++;
4336         }
4337         stl_be_phys(cs->as,
4338                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4339     }
4340 }
4341 
4342 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4343 {
4344     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4345 
4346     if (spapr_cpu->vpa_addr) {
4347         CPUState *cs = CPU(cpu);
4348         uint32_t dispatch;
4349 
4350         dispatch = ldl_be_phys(cs->as,
4351                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4352         dispatch++;
4353         if ((dispatch & 1) != 1) {
4354             qemu_log_mask(LOG_GUEST_ERROR,
4355                           "VPA: incorrect dispatch counter value for "
4356                           "preempted partition %u, correcting.\n", dispatch);
4357             dispatch++;
4358         }
4359         stl_be_phys(cs->as,
4360                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4361     }
4362 }
4363 
4364 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4365 {
4366     MachineClass *mc = MACHINE_CLASS(oc);
4367     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4368     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4369     NMIClass *nc = NMI_CLASS(oc);
4370     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4371     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4372     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4373     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4374 
4375     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4376     mc->ignore_boot_device_suffixes = true;
4377 
4378     /*
4379      * We set up the default / latest behaviour here.  The class_init
4380      * functions for the specific versioned machine types can override
4381      * these details for backwards compatibility
4382      */
4383     mc->init = spapr_machine_init;
4384     mc->reset = spapr_machine_reset;
4385     mc->block_default_type = IF_SCSI;
4386     mc->max_cpus = 1024;
4387     mc->no_parallel = 1;
4388     mc->default_boot_order = "";
4389     mc->default_ram_size = 512 * MiB;
4390     mc->default_display = "std";
4391     mc->kvm_type = spapr_kvm_type;
4392     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4393     mc->pci_allow_0_address = true;
4394     assert(!mc->get_hotplug_handler);
4395     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4396     hc->pre_plug = spapr_machine_device_pre_plug;
4397     hc->plug = spapr_machine_device_plug;
4398     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4399     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4400     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4401     hc->unplug_request = spapr_machine_device_unplug_request;
4402     hc->unplug = spapr_machine_device_unplug;
4403 
4404     smc->dr_lmb_enabled = true;
4405     smc->update_dt_enabled = true;
4406     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4407     mc->has_hotpluggable_cpus = true;
4408     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4409     fwc->get_dev_path = spapr_get_fw_dev_path;
4410     nc->nmi_monitor_handler = spapr_nmi;
4411     smc->phb_placement = spapr_phb_placement;
4412     vhc->hypercall = emulate_spapr_hypercall;
4413     vhc->hpt_mask = spapr_hpt_mask;
4414     vhc->map_hptes = spapr_map_hptes;
4415     vhc->unmap_hptes = spapr_unmap_hptes;
4416     vhc->hpte_set_c = spapr_hpte_set_c;
4417     vhc->hpte_set_r = spapr_hpte_set_r;
4418     vhc->get_pate = spapr_get_pate;
4419     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4420     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4421     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4422     xic->ics_get = spapr_ics_get;
4423     xic->ics_resend = spapr_ics_resend;
4424     xic->icp_get = spapr_icp_get;
4425     ispc->print_info = spapr_pic_print_info;
4426     /* Force NUMA node memory size to be a multiple of
4427      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4428      * in which LMBs are represented and hot-added
4429      */
4430     mc->numa_mem_align_shift = 28;
4431     mc->numa_mem_supported = true;
4432     mc->auto_enable_numa = true;
4433 
4434     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4435     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4436     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4437     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4438     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4439     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4440     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4441     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4442     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4443     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4444     spapr_caps_add_properties(smc, &error_abort);
4445     smc->irq = &spapr_irq_dual;
4446     smc->dr_phb_enabled = true;
4447     smc->linux_pci_probe = true;
4448     smc->smp_threads_vsmt = true;
4449     smc->nr_xirqs = SPAPR_NR_XIRQS;
4450 }
4451 
4452 static const TypeInfo spapr_machine_info = {
4453     .name          = TYPE_SPAPR_MACHINE,
4454     .parent        = TYPE_MACHINE,
4455     .abstract      = true,
4456     .instance_size = sizeof(SpaprMachineState),
4457     .instance_init = spapr_instance_init,
4458     .instance_finalize = spapr_machine_finalizefn,
4459     .class_size    = sizeof(SpaprMachineClass),
4460     .class_init    = spapr_machine_class_init,
4461     .interfaces = (InterfaceInfo[]) {
4462         { TYPE_FW_PATH_PROVIDER },
4463         { TYPE_NMI },
4464         { TYPE_HOTPLUG_HANDLER },
4465         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4466         { TYPE_XICS_FABRIC },
4467         { TYPE_INTERRUPT_STATS_PROVIDER },
4468         { }
4469     },
4470 };
4471 
4472 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4473     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4474                                                     void *data)      \
4475     {                                                                \
4476         MachineClass *mc = MACHINE_CLASS(oc);                        \
4477         spapr_machine_##suffix##_class_options(mc);                  \
4478         if (latest) {                                                \
4479             mc->alias = "pseries";                                   \
4480             mc->is_default = 1;                                      \
4481         }                                                            \
4482     }                                                                \
4483     static const TypeInfo spapr_machine_##suffix##_info = {          \
4484         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4485         .parent = TYPE_SPAPR_MACHINE,                                \
4486         .class_init = spapr_machine_##suffix##_class_init,           \
4487     };                                                               \
4488     static void spapr_machine_register_##suffix(void)                \
4489     {                                                                \
4490         type_register(&spapr_machine_##suffix##_info);               \
4491     }                                                                \
4492     type_init(spapr_machine_register_##suffix)
4493 
4494 /*
4495  * pseries-4.2
4496  */
4497 static void spapr_machine_4_2_class_options(MachineClass *mc)
4498 {
4499     /* Defaults for the latest behaviour inherited from the base class */
4500 }
4501 
4502 DEFINE_SPAPR_MACHINE(4_2, "4.2", true);
4503 
4504 /*
4505  * pseries-4.1
4506  */
4507 static void spapr_machine_4_1_class_options(MachineClass *mc)
4508 {
4509     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4510     static GlobalProperty compat[] = {
4511         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4512         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4513     };
4514 
4515     spapr_machine_4_2_class_options(mc);
4516     smc->linux_pci_probe = false;
4517     smc->smp_threads_vsmt = false;
4518     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4519     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4520 }
4521 
4522 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4523 
4524 /*
4525  * pseries-4.0
4526  */
4527 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4528                               uint64_t *buid, hwaddr *pio,
4529                               hwaddr *mmio32, hwaddr *mmio64,
4530                               unsigned n_dma, uint32_t *liobns,
4531                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4532 {
4533     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4534                         nv2gpa, nv2atsd, errp);
4535     *nv2gpa = 0;
4536     *nv2atsd = 0;
4537 }
4538 
4539 static void spapr_machine_4_0_class_options(MachineClass *mc)
4540 {
4541     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4542 
4543     spapr_machine_4_1_class_options(mc);
4544     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4545     smc->phb_placement = phb_placement_4_0;
4546     smc->irq = &spapr_irq_xics;
4547     smc->pre_4_1_migration = true;
4548 }
4549 
4550 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4551 
4552 /*
4553  * pseries-3.1
4554  */
4555 static void spapr_machine_3_1_class_options(MachineClass *mc)
4556 {
4557     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4558 
4559     spapr_machine_4_0_class_options(mc);
4560     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4561 
4562     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4563     smc->update_dt_enabled = false;
4564     smc->dr_phb_enabled = false;
4565     smc->broken_host_serial_model = true;
4566     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4567     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4568     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4569     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4570 }
4571 
4572 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4573 
4574 /*
4575  * pseries-3.0
4576  */
4577 
4578 static void spapr_machine_3_0_class_options(MachineClass *mc)
4579 {
4580     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4581 
4582     spapr_machine_3_1_class_options(mc);
4583     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4584 
4585     smc->legacy_irq_allocation = true;
4586     smc->nr_xirqs = 0x400;
4587     smc->irq = &spapr_irq_xics_legacy;
4588 }
4589 
4590 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4591 
4592 /*
4593  * pseries-2.12
4594  */
4595 static void spapr_machine_2_12_class_options(MachineClass *mc)
4596 {
4597     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4598     static GlobalProperty compat[] = {
4599         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4600         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4601     };
4602 
4603     spapr_machine_3_0_class_options(mc);
4604     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4605     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4606 
4607     /* We depend on kvm_enabled() to choose a default value for the
4608      * hpt-max-page-size capability. Of course we can't do it here
4609      * because this is too early and the HW accelerator isn't initialzed
4610      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4611      */
4612     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4613 }
4614 
4615 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4616 
4617 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4618 {
4619     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4620 
4621     spapr_machine_2_12_class_options(mc);
4622     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4623     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4624     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4625 }
4626 
4627 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4628 
4629 /*
4630  * pseries-2.11
4631  */
4632 
4633 static void spapr_machine_2_11_class_options(MachineClass *mc)
4634 {
4635     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4636 
4637     spapr_machine_2_12_class_options(mc);
4638     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4639     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4640 }
4641 
4642 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4643 
4644 /*
4645  * pseries-2.10
4646  */
4647 
4648 static void spapr_machine_2_10_class_options(MachineClass *mc)
4649 {
4650     spapr_machine_2_11_class_options(mc);
4651     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4652 }
4653 
4654 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4655 
4656 /*
4657  * pseries-2.9
4658  */
4659 
4660 static void spapr_machine_2_9_class_options(MachineClass *mc)
4661 {
4662     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4663     static GlobalProperty compat[] = {
4664         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4665     };
4666 
4667     spapr_machine_2_10_class_options(mc);
4668     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4669     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4670     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4671     smc->pre_2_10_has_unused_icps = true;
4672     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4673 }
4674 
4675 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4676 
4677 /*
4678  * pseries-2.8
4679  */
4680 
4681 static void spapr_machine_2_8_class_options(MachineClass *mc)
4682 {
4683     static GlobalProperty compat[] = {
4684         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4685     };
4686 
4687     spapr_machine_2_9_class_options(mc);
4688     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4689     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4690     mc->numa_mem_align_shift = 23;
4691 }
4692 
4693 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4694 
4695 /*
4696  * pseries-2.7
4697  */
4698 
4699 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4700                               uint64_t *buid, hwaddr *pio,
4701                               hwaddr *mmio32, hwaddr *mmio64,
4702                               unsigned n_dma, uint32_t *liobns,
4703                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4704 {
4705     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4706     const uint64_t base_buid = 0x800000020000000ULL;
4707     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4708     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4709     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4710     const uint32_t max_index = 255;
4711     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4712 
4713     uint64_t ram_top = MACHINE(spapr)->ram_size;
4714     hwaddr phb0_base, phb_base;
4715     int i;
4716 
4717     /* Do we have device memory? */
4718     if (MACHINE(spapr)->maxram_size > ram_top) {
4719         /* Can't just use maxram_size, because there may be an
4720          * alignment gap between normal and device memory regions
4721          */
4722         ram_top = MACHINE(spapr)->device_memory->base +
4723             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4724     }
4725 
4726     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4727 
4728     if (index > max_index) {
4729         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4730                    max_index);
4731         return;
4732     }
4733 
4734     *buid = base_buid + index;
4735     for (i = 0; i < n_dma; ++i) {
4736         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4737     }
4738 
4739     phb_base = phb0_base + index * phb_spacing;
4740     *pio = phb_base + pio_offset;
4741     *mmio32 = phb_base + mmio_offset;
4742     /*
4743      * We don't set the 64-bit MMIO window, relying on the PHB's
4744      * fallback behaviour of automatically splitting a large "32-bit"
4745      * window into contiguous 32-bit and 64-bit windows
4746      */
4747 
4748     *nv2gpa = 0;
4749     *nv2atsd = 0;
4750 }
4751 
4752 static void spapr_machine_2_7_class_options(MachineClass *mc)
4753 {
4754     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4755     static GlobalProperty compat[] = {
4756         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4757         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4758         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4759         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4760     };
4761 
4762     spapr_machine_2_8_class_options(mc);
4763     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4764     mc->default_machine_opts = "modern-hotplug-events=off";
4765     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4766     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4767     smc->phb_placement = phb_placement_2_7;
4768 }
4769 
4770 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4771 
4772 /*
4773  * pseries-2.6
4774  */
4775 
4776 static void spapr_machine_2_6_class_options(MachineClass *mc)
4777 {
4778     static GlobalProperty compat[] = {
4779         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4780     };
4781 
4782     spapr_machine_2_7_class_options(mc);
4783     mc->has_hotpluggable_cpus = false;
4784     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4785     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4786 }
4787 
4788 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4789 
4790 /*
4791  * pseries-2.5
4792  */
4793 
4794 static void spapr_machine_2_5_class_options(MachineClass *mc)
4795 {
4796     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4797     static GlobalProperty compat[] = {
4798         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4799     };
4800 
4801     spapr_machine_2_6_class_options(mc);
4802     smc->use_ohci_by_default = true;
4803     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4804     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4805 }
4806 
4807 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4808 
4809 /*
4810  * pseries-2.4
4811  */
4812 
4813 static void spapr_machine_2_4_class_options(MachineClass *mc)
4814 {
4815     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4816 
4817     spapr_machine_2_5_class_options(mc);
4818     smc->dr_lmb_enabled = false;
4819     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4820 }
4821 
4822 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4823 
4824 /*
4825  * pseries-2.3
4826  */
4827 
4828 static void spapr_machine_2_3_class_options(MachineClass *mc)
4829 {
4830     static GlobalProperty compat[] = {
4831         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4832     };
4833     spapr_machine_2_4_class_options(mc);
4834     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4835     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4836 }
4837 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4838 
4839 /*
4840  * pseries-2.2
4841  */
4842 
4843 static void spapr_machine_2_2_class_options(MachineClass *mc)
4844 {
4845     static GlobalProperty compat[] = {
4846         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4847     };
4848 
4849     spapr_machine_2_3_class_options(mc);
4850     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4851     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4852     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4853 }
4854 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4855 
4856 /*
4857  * pseries-2.1
4858  */
4859 
4860 static void spapr_machine_2_1_class_options(MachineClass *mc)
4861 {
4862     spapr_machine_2_2_class_options(mc);
4863     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4864 }
4865 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4866 
4867 static void spapr_machine_register_types(void)
4868 {
4869     type_register_static(&spapr_machine_info);
4870 }
4871 
4872 type_init(spapr_machine_register_types)
4873