xref: /openbmc/qemu/hw/ppc/spapr.c (revision 6cc64796f2b031186b6ae9c6c5932ea972a8c6cd)
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 "hw/ppc/spapr_cpu_core.h"
80 #include "hw/mem/memory-device.h"
81 #include "hw/ppc/spapr_tpm_proxy.h"
82 
83 #include "monitor/monitor.h"
84 
85 #include <libfdt.h>
86 
87 /* SLOF memory layout:
88  *
89  * SLOF raw image loaded at 0, copies its romfs right below the flat
90  * device-tree, then position SLOF itself 31M below that
91  *
92  * So we set FW_OVERHEAD to 40MB which should account for all of that
93  * and more
94  *
95  * We load our kernel at 4M, leaving space for SLOF initial image
96  */
97 #define FDT_MAX_SIZE            0x100000
98 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
99 #define FW_MAX_SIZE             0x400000
100 #define FW_FILE_NAME            "slof.bin"
101 #define FW_OVERHEAD             0x2800000
102 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
103 
104 #define MIN_RMA_SLOF            128UL
105 
106 #define PHANDLE_INTC            0x00001111
107 
108 /* These two functions implement the VCPU id numbering: one to compute them
109  * all and one to identify thread 0 of a VCORE. Any change to the first one
110  * is likely to have an impact on the second one, so let's keep them close.
111  */
112 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
113 {
114     MachineState *ms = MACHINE(spapr);
115     unsigned int smp_threads = ms->smp.threads;
116 
117     assert(spapr->vsmt);
118     return
119         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
120 }
121 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
122                                       PowerPCCPU *cpu)
123 {
124     assert(spapr->vsmt);
125     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
126 }
127 
128 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
129 {
130     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
131      * and newer QEMUs don't even have them. In both cases, we don't want
132      * to send anything on the wire.
133      */
134     return false;
135 }
136 
137 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
138     .name = "icp/server",
139     .version_id = 1,
140     .minimum_version_id = 1,
141     .needed = pre_2_10_vmstate_dummy_icp_needed,
142     .fields = (VMStateField[]) {
143         VMSTATE_UNUSED(4), /* uint32_t xirr */
144         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
145         VMSTATE_UNUSED(1), /* uint8_t mfrr */
146         VMSTATE_END_OF_LIST()
147     },
148 };
149 
150 static void pre_2_10_vmstate_register_dummy_icp(int i)
151 {
152     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
153                      (void *)(uintptr_t) i);
154 }
155 
156 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
157 {
158     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
159                        (void *)(uintptr_t) i);
160 }
161 
162 int spapr_max_server_number(SpaprMachineState *spapr)
163 {
164     MachineState *ms = MACHINE(spapr);
165 
166     assert(spapr->vsmt);
167     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
168 }
169 
170 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
171                                   int smt_threads)
172 {
173     int i, ret = 0;
174     uint32_t servers_prop[smt_threads];
175     uint32_t gservers_prop[smt_threads * 2];
176     int index = spapr_get_vcpu_id(cpu);
177 
178     if (cpu->compat_pvr) {
179         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
180         if (ret < 0) {
181             return ret;
182         }
183     }
184 
185     /* Build interrupt servers and gservers properties */
186     for (i = 0; i < smt_threads; i++) {
187         servers_prop[i] = cpu_to_be32(index + i);
188         /* Hack, direct the group queues back to cpu 0 */
189         gservers_prop[i*2] = cpu_to_be32(index + i);
190         gservers_prop[i*2 + 1] = 0;
191     }
192     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
193                       servers_prop, sizeof(servers_prop));
194     if (ret < 0) {
195         return ret;
196     }
197     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
198                       gservers_prop, sizeof(gservers_prop));
199 
200     return ret;
201 }
202 
203 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
204 {
205     int index = spapr_get_vcpu_id(cpu);
206     uint32_t associativity[] = {cpu_to_be32(0x5),
207                                 cpu_to_be32(0x0),
208                                 cpu_to_be32(0x0),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(cpu->node_id),
211                                 cpu_to_be32(index)};
212 
213     /* Advertise NUMA via ibm,associativity */
214     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
215                           sizeof(associativity));
216 }
217 
218 /* Populate the "ibm,pa-features" property */
219 static void spapr_populate_pa_features(SpaprMachineState *spapr,
220                                        PowerPCCPU *cpu,
221                                        void *fdt, int offset)
222 {
223     uint8_t pa_features_206[] = { 6, 0,
224         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
225     uint8_t pa_features_207[] = { 24, 0,
226         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
227         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
228         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
229         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
230     uint8_t pa_features_300[] = { 66, 0,
231         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
232         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
233         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
234         /* 6: DS207 */
235         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
236         /* 16: Vector */
237         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
238         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
239         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
240         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
242         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
243         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
244         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
245         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
246         /* 42: PM, 44: PC RA, 46: SC vec'd */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
248         /* 48: SIMD, 50: QP BFP, 52: String */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
250         /* 54: DecFP, 56: DecI, 58: SHA */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
252         /* 60: NM atomic, 62: RNG */
253         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
254     };
255     uint8_t *pa_features = NULL;
256     size_t pa_size;
257 
258     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
259         pa_features = pa_features_206;
260         pa_size = sizeof(pa_features_206);
261     }
262     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
263         pa_features = pa_features_207;
264         pa_size = sizeof(pa_features_207);
265     }
266     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
267         pa_features = pa_features_300;
268         pa_size = sizeof(pa_features_300);
269     }
270     if (!pa_features) {
271         return;
272     }
273 
274     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
275         /*
276          * Note: we keep CI large pages off by default because a 64K capable
277          * guest provisioned with large pages might otherwise try to map a qemu
278          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
279          * even if that qemu runs on a 4k host.
280          * We dd this bit back here if we are confident this is not an issue
281          */
282         pa_features[3] |= 0x20;
283     }
284     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
285         pa_features[24] |= 0x80;    /* Transactional memory support */
286     }
287     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
288         /* Workaround for broken kernels that attempt (guest) radix
289          * mode when they can't handle it, if they see the radix bit set
290          * in pa-features. So hide it from them. */
291         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
292     }
293 
294     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
295 }
296 
297 static hwaddr spapr_node0_size(MachineState *machine)
298 {
299     if (machine->numa_state->num_nodes) {
300         int i;
301         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
302             if (machine->numa_state->nodes[i].node_mem) {
303                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
304                            machine->ram_size);
305             }
306         }
307     }
308     return machine->ram_size;
309 }
310 
311 static void add_str(GString *s, const gchar *s1)
312 {
313     g_string_append_len(s, s1, strlen(s1) + 1);
314 }
315 
316 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
317                                        hwaddr size)
318 {
319     uint32_t associativity[] = {
320         cpu_to_be32(0x4), /* length */
321         cpu_to_be32(0x0), cpu_to_be32(0x0),
322         cpu_to_be32(0x0), cpu_to_be32(nodeid)
323     };
324     char mem_name[32];
325     uint64_t mem_reg_property[2];
326     int off;
327 
328     mem_reg_property[0] = cpu_to_be64(start);
329     mem_reg_property[1] = cpu_to_be64(size);
330 
331     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
332     off = fdt_add_subnode(fdt, 0, mem_name);
333     _FDT(off);
334     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
335     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
336                       sizeof(mem_reg_property))));
337     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
338                       sizeof(associativity))));
339     return off;
340 }
341 
342 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
343 {
344     MachineState *machine = MACHINE(spapr);
345     hwaddr mem_start, node_size;
346     int i, nb_nodes = machine->numa_state->num_nodes;
347     NodeInfo *nodes = machine->numa_state->nodes;
348 
349     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
350         if (!nodes[i].node_mem) {
351             continue;
352         }
353         if (mem_start >= machine->ram_size) {
354             node_size = 0;
355         } else {
356             node_size = nodes[i].node_mem;
357             if (node_size > machine->ram_size - mem_start) {
358                 node_size = machine->ram_size - mem_start;
359             }
360         }
361         if (!mem_start) {
362             /* spapr_machine_init() checks for rma_size <= node0_size
363              * already */
364             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
365             mem_start += spapr->rma_size;
366             node_size -= spapr->rma_size;
367         }
368         for ( ; node_size; ) {
369             hwaddr sizetmp = pow2floor(node_size);
370 
371             /* mem_start != 0 here */
372             if (ctzl(mem_start) < ctzl(sizetmp)) {
373                 sizetmp = 1ULL << ctzl(mem_start);
374             }
375 
376             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
377             node_size -= sizetmp;
378             mem_start += sizetmp;
379         }
380     }
381 
382     return 0;
383 }
384 
385 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
386                                   SpaprMachineState *spapr)
387 {
388     MachineState *ms = MACHINE(spapr);
389     PowerPCCPU *cpu = POWERPC_CPU(cs);
390     CPUPPCState *env = &cpu->env;
391     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
392     int index = spapr_get_vcpu_id(cpu);
393     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
394                        0xffffffff, 0xffffffff};
395     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
396         : SPAPR_TIMEBASE_FREQ;
397     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
398     uint32_t page_sizes_prop[64];
399     size_t page_sizes_prop_size;
400     unsigned int smp_threads = ms->smp.threads;
401     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
402     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
403     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
404     SpaprDrc *drc;
405     int drc_index;
406     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
407     int i;
408 
409     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
410     if (drc) {
411         drc_index = spapr_drc_index(drc);
412         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
413     }
414 
415     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
416     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
417 
418     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
419     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
420                            env->dcache_line_size)));
421     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
422                            env->dcache_line_size)));
423     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
424                            env->icache_line_size)));
425     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
426                            env->icache_line_size)));
427 
428     if (pcc->l1_dcache_size) {
429         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
430                                pcc->l1_dcache_size)));
431     } else {
432         warn_report("Unknown L1 dcache size for cpu");
433     }
434     if (pcc->l1_icache_size) {
435         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
436                                pcc->l1_icache_size)));
437     } else {
438         warn_report("Unknown L1 icache size for cpu");
439     }
440 
441     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
442     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
443     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
444     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
445     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
446     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
447 
448     if (env->spr_cb[SPR_PURR].oea_read) {
449         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
450     }
451     if (env->spr_cb[SPR_SPURR].oea_read) {
452         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
453     }
454 
455     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
456         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
457                           segs, sizeof(segs))));
458     }
459 
460     /* Advertise VSX (vector extensions) if available
461      *   1               == VMX / Altivec available
462      *   2               == VSX available
463      *
464      * Only CPUs for which we create core types in spapr_cpu_core.c
465      * are possible, and all of those have VMX */
466     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
467         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
468     } else {
469         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
470     }
471 
472     /* Advertise DFP (Decimal Floating Point) if available
473      *   0 / no property == no DFP
474      *   1               == DFP available */
475     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
476         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
477     }
478 
479     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
480                                                       sizeof(page_sizes_prop));
481     if (page_sizes_prop_size) {
482         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
483                           page_sizes_prop, page_sizes_prop_size)));
484     }
485 
486     spapr_populate_pa_features(spapr, cpu, fdt, offset);
487 
488     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
489                            cs->cpu_index / vcpus_per_socket)));
490 
491     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
492                       pft_size_prop, sizeof(pft_size_prop))));
493 
494     if (ms->numa_state->num_nodes > 1) {
495         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
496     }
497 
498     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
499 
500     if (pcc->radix_page_info) {
501         for (i = 0; i < pcc->radix_page_info->count; i++) {
502             radix_AP_encodings[i] =
503                 cpu_to_be32(pcc->radix_page_info->entries[i]);
504         }
505         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
506                           radix_AP_encodings,
507                           pcc->radix_page_info->count *
508                           sizeof(radix_AP_encodings[0]))));
509     }
510 
511     /*
512      * We set this property to let the guest know that it can use the large
513      * decrementer and its width in bits.
514      */
515     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
516         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
517                               pcc->lrg_decr_bits)));
518 }
519 
520 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
521 {
522     CPUState **rev;
523     CPUState *cs;
524     int n_cpus;
525     int cpus_offset;
526     char *nodename;
527     int i;
528 
529     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
530     _FDT(cpus_offset);
531     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
532     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
533 
534     /*
535      * We walk the CPUs in reverse order to ensure that CPU DT nodes
536      * created by fdt_add_subnode() end up in the right order in FDT
537      * for the guest kernel the enumerate the CPUs correctly.
538      *
539      * The CPU list cannot be traversed in reverse order, so we need
540      * to do extra work.
541      */
542     n_cpus = 0;
543     rev = NULL;
544     CPU_FOREACH(cs) {
545         rev = g_renew(CPUState *, rev, n_cpus + 1);
546         rev[n_cpus++] = cs;
547     }
548 
549     for (i = n_cpus - 1; i >= 0; i--) {
550         CPUState *cs = rev[i];
551         PowerPCCPU *cpu = POWERPC_CPU(cs);
552         int index = spapr_get_vcpu_id(cpu);
553         DeviceClass *dc = DEVICE_GET_CLASS(cs);
554         int offset;
555 
556         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
557             continue;
558         }
559 
560         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
561         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
562         g_free(nodename);
563         _FDT(offset);
564         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
565     }
566 
567     g_free(rev);
568 }
569 
570 static int spapr_rng_populate_dt(void *fdt)
571 {
572     int node;
573     int ret;
574 
575     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
576     if (node <= 0) {
577         return -1;
578     }
579     ret = fdt_setprop_string(fdt, node, "device_type",
580                              "ibm,platform-facilities");
581     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
582     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
583 
584     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
585     if (node <= 0) {
586         return -1;
587     }
588     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
589 
590     return ret ? -1 : 0;
591 }
592 
593 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
594 {
595     MemoryDeviceInfoList *info;
596 
597     for (info = list; info; info = info->next) {
598         MemoryDeviceInfo *value = info->value;
599 
600         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
601             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
602 
603             if (addr >= pcdimm_info->addr &&
604                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
605                 return pcdimm_info->node;
606             }
607         }
608     }
609 
610     return -1;
611 }
612 
613 struct sPAPRDrconfCellV2 {
614      uint32_t seq_lmbs;
615      uint64_t base_addr;
616      uint32_t drc_index;
617      uint32_t aa_index;
618      uint32_t flags;
619 } QEMU_PACKED;
620 
621 typedef struct DrconfCellQueue {
622     struct sPAPRDrconfCellV2 cell;
623     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
624 } DrconfCellQueue;
625 
626 static DrconfCellQueue *
627 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
628                       uint32_t drc_index, uint32_t aa_index,
629                       uint32_t flags)
630 {
631     DrconfCellQueue *elem;
632 
633     elem = g_malloc0(sizeof(*elem));
634     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
635     elem->cell.base_addr = cpu_to_be64(base_addr);
636     elem->cell.drc_index = cpu_to_be32(drc_index);
637     elem->cell.aa_index = cpu_to_be32(aa_index);
638     elem->cell.flags = cpu_to_be32(flags);
639 
640     return elem;
641 }
642 
643 /* ibm,dynamic-memory-v2 */
644 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
645                                    int offset, MemoryDeviceInfoList *dimms)
646 {
647     MachineState *machine = MACHINE(spapr);
648     uint8_t *int_buf, *cur_index;
649     int ret;
650     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
651     uint64_t addr, cur_addr, size;
652     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
653     uint64_t mem_end = machine->device_memory->base +
654                        memory_region_size(&machine->device_memory->mr);
655     uint32_t node, buf_len, nr_entries = 0;
656     SpaprDrc *drc;
657     DrconfCellQueue *elem, *next;
658     MemoryDeviceInfoList *info;
659     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
660         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
661 
662     /* Entry to cover RAM and the gap area */
663     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
664                                  SPAPR_LMB_FLAGS_RESERVED |
665                                  SPAPR_LMB_FLAGS_DRC_INVALID);
666     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
667     nr_entries++;
668 
669     cur_addr = machine->device_memory->base;
670     for (info = dimms; info; info = info->next) {
671         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
672 
673         addr = di->addr;
674         size = di->size;
675         node = di->node;
676 
677         /* Entry for hot-pluggable area */
678         if (cur_addr < addr) {
679             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
680             g_assert(drc);
681             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
682                                          cur_addr, spapr_drc_index(drc), -1, 0);
683             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
684             nr_entries++;
685         }
686 
687         /* Entry for DIMM */
688         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
689         g_assert(drc);
690         elem = spapr_get_drconf_cell(size / lmb_size, addr,
691                                      spapr_drc_index(drc), node,
692                                      SPAPR_LMB_FLAGS_ASSIGNED);
693         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
694         nr_entries++;
695         cur_addr = addr + size;
696     }
697 
698     /* Entry for remaining hotpluggable area */
699     if (cur_addr < mem_end) {
700         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
701         g_assert(drc);
702         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
703                                      cur_addr, spapr_drc_index(drc), -1, 0);
704         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
705         nr_entries++;
706     }
707 
708     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
709     int_buf = cur_index = g_malloc0(buf_len);
710     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
711     cur_index += sizeof(nr_entries);
712 
713     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
714         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
715         cur_index += sizeof(elem->cell);
716         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
717         g_free(elem);
718     }
719 
720     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
721     g_free(int_buf);
722     if (ret < 0) {
723         return -1;
724     }
725     return 0;
726 }
727 
728 /* ibm,dynamic-memory */
729 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
730                                    int offset, MemoryDeviceInfoList *dimms)
731 {
732     MachineState *machine = MACHINE(spapr);
733     int i, ret;
734     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
735     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
736     uint32_t nr_lmbs = (machine->device_memory->base +
737                        memory_region_size(&machine->device_memory->mr)) /
738                        lmb_size;
739     uint32_t *int_buf, *cur_index, buf_len;
740 
741     /*
742      * Allocate enough buffer size to fit in ibm,dynamic-memory
743      */
744     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
745     cur_index = int_buf = g_malloc0(buf_len);
746     int_buf[0] = cpu_to_be32(nr_lmbs);
747     cur_index++;
748     for (i = 0; i < nr_lmbs; i++) {
749         uint64_t addr = i * lmb_size;
750         uint32_t *dynamic_memory = cur_index;
751 
752         if (i >= device_lmb_start) {
753             SpaprDrc *drc;
754 
755             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
756             g_assert(drc);
757 
758             dynamic_memory[0] = cpu_to_be32(addr >> 32);
759             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
760             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
761             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
762             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
763             if (memory_region_present(get_system_memory(), addr)) {
764                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
765             } else {
766                 dynamic_memory[5] = cpu_to_be32(0);
767             }
768         } else {
769             /*
770              * LMB information for RMA, boot time RAM and gap b/n RAM and
771              * device memory region -- all these are marked as reserved
772              * and as having no valid DRC.
773              */
774             dynamic_memory[0] = cpu_to_be32(addr >> 32);
775             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
776             dynamic_memory[2] = cpu_to_be32(0);
777             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
778             dynamic_memory[4] = cpu_to_be32(-1);
779             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
780                                             SPAPR_LMB_FLAGS_DRC_INVALID);
781         }
782 
783         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
784     }
785     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
786     g_free(int_buf);
787     if (ret < 0) {
788         return -1;
789     }
790     return 0;
791 }
792 
793 /*
794  * Adds ibm,dynamic-reconfiguration-memory node.
795  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
796  * of this device tree node.
797  */
798 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
799 {
800     MachineState *machine = MACHINE(spapr);
801     int nb_numa_nodes = machine->numa_state->num_nodes;
802     int ret, i, offset;
803     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
804     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
805     uint32_t *int_buf, *cur_index, buf_len;
806     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
807     MemoryDeviceInfoList *dimms = NULL;
808 
809     /*
810      * Don't create the node if there is no device memory
811      */
812     if (machine->ram_size == machine->maxram_size) {
813         return 0;
814     }
815 
816     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
817 
818     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
819                     sizeof(prop_lmb_size));
820     if (ret < 0) {
821         return ret;
822     }
823 
824     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
825     if (ret < 0) {
826         return ret;
827     }
828 
829     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
830     if (ret < 0) {
831         return ret;
832     }
833 
834     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
835     dimms = qmp_memory_device_list();
836     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
837         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
838     } else {
839         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
840     }
841     qapi_free_MemoryDeviceInfoList(dimms);
842 
843     if (ret < 0) {
844         return ret;
845     }
846 
847     /* ibm,associativity-lookup-arrays */
848     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
849     cur_index = int_buf = g_malloc0(buf_len);
850     int_buf[0] = cpu_to_be32(nr_nodes);
851     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
852     cur_index += 2;
853     for (i = 0; i < nr_nodes; i++) {
854         uint32_t associativity[] = {
855             cpu_to_be32(0x0),
856             cpu_to_be32(0x0),
857             cpu_to_be32(0x0),
858             cpu_to_be32(i)
859         };
860         memcpy(cur_index, associativity, sizeof(associativity));
861         cur_index += 4;
862     }
863     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
864             (cur_index - int_buf) * sizeof(uint32_t));
865     g_free(int_buf);
866 
867     return ret;
868 }
869 
870 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
871                                 SpaprOptionVector *ov5_updates)
872 {
873     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
874     int ret = 0, offset;
875 
876     /* Generate ibm,dynamic-reconfiguration-memory node if required */
877     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
878         g_assert(smc->dr_lmb_enabled);
879         ret = spapr_populate_drconf_memory(spapr, fdt);
880         if (ret) {
881             goto out;
882         }
883     }
884 
885     offset = fdt_path_offset(fdt, "/chosen");
886     if (offset < 0) {
887         offset = fdt_add_subnode(fdt, 0, "chosen");
888         if (offset < 0) {
889             return offset;
890         }
891     }
892     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
893                                  "ibm,architecture-vec-5");
894 
895 out:
896     return ret;
897 }
898 
899 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
900 {
901     MachineState *ms = MACHINE(spapr);
902     int rtas;
903     GString *hypertas = g_string_sized_new(256);
904     GString *qemu_hypertas = g_string_sized_new(256);
905     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
906     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
907         memory_region_size(&MACHINE(spapr)->device_memory->mr);
908     uint32_t lrdr_capacity[] = {
909         cpu_to_be32(max_device_addr >> 32),
910         cpu_to_be32(max_device_addr & 0xffffffff),
911         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
912         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
913     };
914     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
915     uint32_t maxdomains[] = {
916         cpu_to_be32(4),
917         maxdomain,
918         maxdomain,
919         maxdomain,
920         cpu_to_be32(spapr->gpu_numa_id),
921     };
922 
923     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
924 
925     /* hypertas */
926     add_str(hypertas, "hcall-pft");
927     add_str(hypertas, "hcall-term");
928     add_str(hypertas, "hcall-dabr");
929     add_str(hypertas, "hcall-interrupt");
930     add_str(hypertas, "hcall-tce");
931     add_str(hypertas, "hcall-vio");
932     add_str(hypertas, "hcall-splpar");
933     add_str(hypertas, "hcall-join");
934     add_str(hypertas, "hcall-bulk");
935     add_str(hypertas, "hcall-set-mode");
936     add_str(hypertas, "hcall-sprg0");
937     add_str(hypertas, "hcall-copy");
938     add_str(hypertas, "hcall-debug");
939     add_str(hypertas, "hcall-vphn");
940     add_str(qemu_hypertas, "hcall-memop1");
941 
942     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
943         add_str(hypertas, "hcall-multi-tce");
944     }
945 
946     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
947         add_str(hypertas, "hcall-hpt-resize");
948     }
949 
950     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
951                      hypertas->str, hypertas->len));
952     g_string_free(hypertas, TRUE);
953     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
954                      qemu_hypertas->str, qemu_hypertas->len));
955     g_string_free(qemu_hypertas, TRUE);
956 
957     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
958                      refpoints, sizeof(refpoints)));
959 
960     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
961                      maxdomains, sizeof(maxdomains)));
962 
963     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
964                           RTAS_ERROR_LOG_MAX));
965     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
966                           RTAS_EVENT_SCAN_RATE));
967 
968     g_assert(msi_nonbroken);
969     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
970 
971     /*
972      * According to PAPR, rtas ibm,os-term does not guarantee a return
973      * back to the guest cpu.
974      *
975      * While an additional ibm,extended-os-term property indicates
976      * that rtas call return will always occur. Set this property.
977      */
978     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
979 
980     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
981                      lrdr_capacity, sizeof(lrdr_capacity)));
982 
983     spapr_dt_rtas_tokens(fdt, rtas);
984 }
985 
986 /*
987  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
988  * and the XIVE features that the guest may request and thus the valid
989  * values for bytes 23..26 of option vector 5:
990  */
991 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
992                                           int chosen)
993 {
994     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
995 
996     char val[2 * 4] = {
997         23, 0x00, /* XICS / XIVE mode */
998         24, 0x00, /* Hash/Radix, filled in below. */
999         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1000         26, 0x40, /* Radix options: GTSE == yes. */
1001     };
1002 
1003     if (spapr->irq->xics && spapr->irq->xive) {
1004         val[1] = SPAPR_OV5_XIVE_BOTH;
1005     } else if (spapr->irq->xive) {
1006         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1007     } else {
1008         assert(spapr->irq->xics);
1009         val[1] = SPAPR_OV5_XIVE_LEGACY;
1010     }
1011 
1012     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1013                           first_ppc_cpu->compat_pvr)) {
1014         /*
1015          * If we're in a pre POWER9 compat mode then the guest should
1016          * do hash and use the legacy interrupt mode
1017          */
1018         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1019         val[3] = 0x00; /* Hash */
1020     } else if (kvm_enabled()) {
1021         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1022             val[3] = 0x80; /* OV5_MMU_BOTH */
1023         } else if (kvmppc_has_cap_mmu_radix()) {
1024             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1025         } else {
1026             val[3] = 0x00; /* Hash */
1027         }
1028     } else {
1029         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1030         val[3] = 0xC0;
1031     }
1032     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1033                      val, sizeof(val)));
1034 }
1035 
1036 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1037 {
1038     MachineState *machine = MACHINE(spapr);
1039     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1040     int chosen;
1041     const char *boot_device = machine->boot_order;
1042     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1043     size_t cb = 0;
1044     char *bootlist = get_boot_devices_list(&cb);
1045 
1046     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1047 
1048     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1049         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1050                                 machine->kernel_cmdline));
1051     }
1052     if (spapr->initrd_size) {
1053         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1054                               spapr->initrd_base));
1055         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1056                               spapr->initrd_base + spapr->initrd_size));
1057     }
1058 
1059     if (spapr->kernel_size) {
1060         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1061                               cpu_to_be64(spapr->kernel_size) };
1062 
1063         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1064                          &kprop, sizeof(kprop)));
1065         if (spapr->kernel_le) {
1066             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1067         }
1068     }
1069     if (boot_menu) {
1070         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1071     }
1072     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1073     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1074     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1075 
1076     if (cb && bootlist) {
1077         int i;
1078 
1079         for (i = 0; i < cb; i++) {
1080             if (bootlist[i] == '\n') {
1081                 bootlist[i] = ' ';
1082             }
1083         }
1084         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1085     }
1086 
1087     if (boot_device && strlen(boot_device)) {
1088         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1089     }
1090 
1091     if (!spapr->has_graphics && stdout_path) {
1092         /*
1093          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1094          * kernel. New platforms should only use the "stdout-path" property. Set
1095          * the new property and continue using older property to remain
1096          * compatible with the existing firmware.
1097          */
1098         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1099         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1100     }
1101 
1102     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1103     if (smc->linux_pci_probe) {
1104         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1105     }
1106 
1107     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1108 
1109     g_free(stdout_path);
1110     g_free(bootlist);
1111 }
1112 
1113 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1114 {
1115     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1116      * KVM to work under pHyp with some guest co-operation */
1117     int hypervisor;
1118     uint8_t hypercall[16];
1119 
1120     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1121     /* indicate KVM hypercall interface */
1122     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1123     if (kvmppc_has_cap_fixup_hcalls()) {
1124         /*
1125          * Older KVM versions with older guest kernels were broken
1126          * with the magic page, don't allow the guest to map it.
1127          */
1128         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1129                                   sizeof(hypercall))) {
1130             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1131                              hypercall, sizeof(hypercall)));
1132         }
1133     }
1134 }
1135 
1136 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1137 {
1138     MachineState *machine = MACHINE(spapr);
1139     MachineClass *mc = MACHINE_GET_CLASS(machine);
1140     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1141     int ret;
1142     void *fdt;
1143     SpaprPhbState *phb;
1144     char *buf;
1145 
1146     fdt = g_malloc0(space);
1147     _FDT((fdt_create_empty_tree(fdt, space)));
1148 
1149     /* Root node */
1150     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1151     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1152     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1153 
1154     /* Guest UUID & Name*/
1155     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1156     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1157     if (qemu_uuid_set) {
1158         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1159     }
1160     g_free(buf);
1161 
1162     if (qemu_get_vm_name()) {
1163         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1164                                 qemu_get_vm_name()));
1165     }
1166 
1167     /* Host Model & Serial Number */
1168     if (spapr->host_model) {
1169         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1170     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1171         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1172         g_free(buf);
1173     }
1174 
1175     if (spapr->host_serial) {
1176         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1177     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1178         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1179         g_free(buf);
1180     }
1181 
1182     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1183     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1184 
1185     /* /interrupt controller */
1186     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1187 
1188     ret = spapr_populate_memory(spapr, fdt);
1189     if (ret < 0) {
1190         error_report("couldn't setup memory nodes in fdt");
1191         exit(1);
1192     }
1193 
1194     /* /vdevice */
1195     spapr_dt_vdevice(spapr->vio_bus, fdt);
1196 
1197     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1198         ret = spapr_rng_populate_dt(fdt);
1199         if (ret < 0) {
1200             error_report("could not set up rng device in the fdt");
1201             exit(1);
1202         }
1203     }
1204 
1205     QLIST_FOREACH(phb, &spapr->phbs, list) {
1206         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1207         if (ret < 0) {
1208             error_report("couldn't setup PCI devices in fdt");
1209             exit(1);
1210         }
1211     }
1212 
1213     /* cpus */
1214     spapr_populate_cpus_dt_node(fdt, spapr);
1215 
1216     if (smc->dr_lmb_enabled) {
1217         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1218     }
1219 
1220     if (mc->has_hotpluggable_cpus) {
1221         int offset = fdt_path_offset(fdt, "/cpus");
1222         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1223         if (ret < 0) {
1224             error_report("Couldn't set up CPU DR device tree properties");
1225             exit(1);
1226         }
1227     }
1228 
1229     /* /event-sources */
1230     spapr_dt_events(spapr, fdt);
1231 
1232     /* /rtas */
1233     spapr_dt_rtas(spapr, fdt);
1234 
1235     /* /chosen */
1236     if (reset) {
1237         spapr_dt_chosen(spapr, fdt);
1238     }
1239 
1240     /* /hypervisor */
1241     if (kvm_enabled()) {
1242         spapr_dt_hypervisor(spapr, fdt);
1243     }
1244 
1245     /* Build memory reserve map */
1246     if (reset) {
1247         if (spapr->kernel_size) {
1248             _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1249         }
1250         if (spapr->initrd_size) {
1251             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1252                                   spapr->initrd_size)));
1253         }
1254     }
1255 
1256     /* ibm,client-architecture-support updates */
1257     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1258     if (ret < 0) {
1259         error_report("couldn't setup CAS properties fdt");
1260         exit(1);
1261     }
1262 
1263     if (smc->dr_phb_enabled) {
1264         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1265         if (ret < 0) {
1266             error_report("Couldn't set up PHB DR device tree properties");
1267             exit(1);
1268         }
1269     }
1270 
1271     return fdt;
1272 }
1273 
1274 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1275 {
1276     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1277 }
1278 
1279 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1280                                     PowerPCCPU *cpu)
1281 {
1282     CPUPPCState *env = &cpu->env;
1283 
1284     /* The TCG path should also be holding the BQL at this point */
1285     g_assert(qemu_mutex_iothread_locked());
1286 
1287     if (msr_pr) {
1288         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1289         env->gpr[3] = H_PRIVILEGE;
1290     } else {
1291         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1292     }
1293 }
1294 
1295 struct LPCRSyncState {
1296     target_ulong value;
1297     target_ulong mask;
1298 };
1299 
1300 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1301 {
1302     struct LPCRSyncState *s = arg.host_ptr;
1303     PowerPCCPU *cpu = POWERPC_CPU(cs);
1304     CPUPPCState *env = &cpu->env;
1305     target_ulong lpcr;
1306 
1307     cpu_synchronize_state(cs);
1308     lpcr = env->spr[SPR_LPCR];
1309     lpcr &= ~s->mask;
1310     lpcr |= s->value;
1311     ppc_store_lpcr(cpu, lpcr);
1312 }
1313 
1314 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1315 {
1316     CPUState *cs;
1317     struct LPCRSyncState s = {
1318         .value = value,
1319         .mask = mask
1320     };
1321     CPU_FOREACH(cs) {
1322         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1323     }
1324 }
1325 
1326 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1327 {
1328     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1329 
1330     /* Copy PATE1:GR into PATE0:HR */
1331     entry->dw0 = spapr->patb_entry & PATE0_HR;
1332     entry->dw1 = spapr->patb_entry;
1333 }
1334 
1335 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1336 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1337 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1338 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1339 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1340 
1341 /*
1342  * Get the fd to access the kernel htab, re-opening it if necessary
1343  */
1344 static int get_htab_fd(SpaprMachineState *spapr)
1345 {
1346     Error *local_err = NULL;
1347 
1348     if (spapr->htab_fd >= 0) {
1349         return spapr->htab_fd;
1350     }
1351 
1352     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1353     if (spapr->htab_fd < 0) {
1354         error_report_err(local_err);
1355     }
1356 
1357     return spapr->htab_fd;
1358 }
1359 
1360 void close_htab_fd(SpaprMachineState *spapr)
1361 {
1362     if (spapr->htab_fd >= 0) {
1363         close(spapr->htab_fd);
1364     }
1365     spapr->htab_fd = -1;
1366 }
1367 
1368 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1369 {
1370     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1371 
1372     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1373 }
1374 
1375 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1376 {
1377     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1378 
1379     assert(kvm_enabled());
1380 
1381     if (!spapr->htab) {
1382         return 0;
1383     }
1384 
1385     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1386 }
1387 
1388 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1389                                                 hwaddr ptex, int n)
1390 {
1391     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1392     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1393 
1394     if (!spapr->htab) {
1395         /*
1396          * HTAB is controlled by KVM. Fetch into temporary buffer
1397          */
1398         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1399         kvmppc_read_hptes(hptes, ptex, n);
1400         return hptes;
1401     }
1402 
1403     /*
1404      * HTAB is controlled by QEMU. Just point to the internally
1405      * accessible PTEG.
1406      */
1407     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1408 }
1409 
1410 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1411                               const ppc_hash_pte64_t *hptes,
1412                               hwaddr ptex, int n)
1413 {
1414     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1415 
1416     if (!spapr->htab) {
1417         g_free((void *)hptes);
1418     }
1419 
1420     /* Nothing to do for qemu managed HPT */
1421 }
1422 
1423 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1424                       uint64_t pte0, uint64_t pte1)
1425 {
1426     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1427     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1428 
1429     if (!spapr->htab) {
1430         kvmppc_write_hpte(ptex, pte0, pte1);
1431     } else {
1432         if (pte0 & HPTE64_V_VALID) {
1433             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1434             /*
1435              * When setting valid, we write PTE1 first. This ensures
1436              * proper synchronization with the reading code in
1437              * ppc_hash64_pteg_search()
1438              */
1439             smp_wmb();
1440             stq_p(spapr->htab + offset, pte0);
1441         } else {
1442             stq_p(spapr->htab + offset, pte0);
1443             /*
1444              * When clearing it we set PTE0 first. This ensures proper
1445              * synchronization with the reading code in
1446              * ppc_hash64_pteg_search()
1447              */
1448             smp_wmb();
1449             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1450         }
1451     }
1452 }
1453 
1454 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1455                              uint64_t pte1)
1456 {
1457     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1458     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1459 
1460     if (!spapr->htab) {
1461         /* There should always be a hash table when this is called */
1462         error_report("spapr_hpte_set_c called with no hash table !");
1463         return;
1464     }
1465 
1466     /* The HW performs a non-atomic byte update */
1467     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1468 }
1469 
1470 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1471                              uint64_t pte1)
1472 {
1473     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1474     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1475 
1476     if (!spapr->htab) {
1477         /* There should always be a hash table when this is called */
1478         error_report("spapr_hpte_set_r called with no hash table !");
1479         return;
1480     }
1481 
1482     /* The HW performs a non-atomic byte update */
1483     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1484 }
1485 
1486 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1487 {
1488     int shift;
1489 
1490     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1491      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1492      * that's much more than is needed for Linux guests */
1493     shift = ctz64(pow2ceil(ramsize)) - 7;
1494     shift = MAX(shift, 18); /* Minimum architected size */
1495     shift = MIN(shift, 46); /* Maximum architected size */
1496     return shift;
1497 }
1498 
1499 void spapr_free_hpt(SpaprMachineState *spapr)
1500 {
1501     g_free(spapr->htab);
1502     spapr->htab = NULL;
1503     spapr->htab_shift = 0;
1504     close_htab_fd(spapr);
1505 }
1506 
1507 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1508                           Error **errp)
1509 {
1510     long rc;
1511 
1512     /* Clean up any HPT info from a previous boot */
1513     spapr_free_hpt(spapr);
1514 
1515     rc = kvmppc_reset_htab(shift);
1516     if (rc < 0) {
1517         /* kernel-side HPT needed, but couldn't allocate one */
1518         error_setg_errno(errp, errno,
1519                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1520                          shift);
1521         /* This is almost certainly fatal, but if the caller really
1522          * wants to carry on with shift == 0, it's welcome to try */
1523     } else if (rc > 0) {
1524         /* kernel-side HPT allocated */
1525         if (rc != shift) {
1526             error_setg(errp,
1527                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1528                        shift, rc);
1529         }
1530 
1531         spapr->htab_shift = shift;
1532         spapr->htab = NULL;
1533     } else {
1534         /* kernel-side HPT not needed, allocate in userspace instead */
1535         size_t size = 1ULL << shift;
1536         int i;
1537 
1538         spapr->htab = qemu_memalign(size, size);
1539         if (!spapr->htab) {
1540             error_setg_errno(errp, errno,
1541                              "Could not allocate HPT of order %d", shift);
1542             return;
1543         }
1544 
1545         memset(spapr->htab, 0, size);
1546         spapr->htab_shift = shift;
1547 
1548         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1549             DIRTY_HPTE(HPTE(spapr->htab, i));
1550         }
1551     }
1552     /* We're setting up a hash table, so that means we're not radix */
1553     spapr->patb_entry = 0;
1554     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1555 }
1556 
1557 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1558 {
1559     int hpt_shift;
1560 
1561     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1562         || (spapr->cas_reboot
1563             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1564         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1565     } else {
1566         uint64_t current_ram_size;
1567 
1568         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1569         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1570     }
1571     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1572 
1573     if (spapr->vrma_adjust) {
1574         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1575                                           spapr->htab_shift);
1576     }
1577 }
1578 
1579 static int spapr_reset_drcs(Object *child, void *opaque)
1580 {
1581     SpaprDrc *drc =
1582         (SpaprDrc *) object_dynamic_cast(child,
1583                                                  TYPE_SPAPR_DR_CONNECTOR);
1584 
1585     if (drc) {
1586         spapr_drc_reset(drc);
1587     }
1588 
1589     return 0;
1590 }
1591 
1592 static void spapr_machine_reset(MachineState *machine)
1593 {
1594     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1595     PowerPCCPU *first_ppc_cpu;
1596     hwaddr fdt_addr;
1597     void *fdt;
1598     int rc;
1599 
1600     kvmppc_svm_off(&error_fatal);
1601     spapr_caps_apply(spapr);
1602 
1603     first_ppc_cpu = POWERPC_CPU(first_cpu);
1604     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1605         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1606                               spapr->max_compat_pvr)) {
1607         /*
1608          * If using KVM with radix mode available, VCPUs can be started
1609          * without a HPT because KVM will start them in radix mode.
1610          * Set the GR bit in PATE so that we know there is no HPT.
1611          */
1612         spapr->patb_entry = PATE1_GR;
1613         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1614     } else {
1615         spapr_setup_hpt_and_vrma(spapr);
1616     }
1617 
1618     qemu_devices_reset();
1619 
1620     /*
1621      * If this reset wasn't generated by CAS, we should reset our
1622      * negotiated options and start from scratch
1623      */
1624     if (!spapr->cas_reboot) {
1625         spapr_ovec_cleanup(spapr->ov5_cas);
1626         spapr->ov5_cas = spapr_ovec_new();
1627 
1628         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1629     }
1630 
1631     /*
1632      * This is fixing some of the default configuration of the XIVE
1633      * devices. To be called after the reset of the machine devices.
1634      */
1635     spapr_irq_reset(spapr, &error_fatal);
1636 
1637     /*
1638      * There is no CAS under qtest. Simulate one to please the code that
1639      * depends on spapr->ov5_cas. This is especially needed to test device
1640      * unplug, so we do that before resetting the DRCs.
1641      */
1642     if (qtest_enabled()) {
1643         spapr_ovec_cleanup(spapr->ov5_cas);
1644         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1645     }
1646 
1647     /* DRC reset may cause a device to be unplugged. This will cause troubles
1648      * if this device is used by another device (eg, a running vhost backend
1649      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1650      * situations, we reset DRCs after all devices have been reset.
1651      */
1652     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1653 
1654     spapr_clear_pending_events(spapr);
1655 
1656     /*
1657      * We place the device tree and RTAS just below either the top of the RMA,
1658      * or just below 2GB, whichever is lower, so that it can be
1659      * processed with 32-bit real mode code if necessary
1660      */
1661     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1662 
1663     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1664 
1665     rc = fdt_pack(fdt);
1666 
1667     /* Should only fail if we've built a corrupted tree */
1668     assert(rc == 0);
1669 
1670     /* Load the fdt */
1671     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1672     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1673     g_free(spapr->fdt_blob);
1674     spapr->fdt_size = fdt_totalsize(fdt);
1675     spapr->fdt_initial_size = spapr->fdt_size;
1676     spapr->fdt_blob = fdt;
1677 
1678     /* Set up the entry state */
1679     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1680     first_ppc_cpu->env.gpr[5] = 0;
1681 
1682     spapr->cas_reboot = false;
1683 }
1684 
1685 static void spapr_create_nvram(SpaprMachineState *spapr)
1686 {
1687     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1688     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1689 
1690     if (dinfo) {
1691         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1692                             &error_fatal);
1693     }
1694 
1695     qdev_init_nofail(dev);
1696 
1697     spapr->nvram = (struct SpaprNvram *)dev;
1698 }
1699 
1700 static void spapr_rtc_create(SpaprMachineState *spapr)
1701 {
1702     object_initialize_child(OBJECT(spapr), "rtc",
1703                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1704                             &error_fatal, NULL);
1705     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1706                               &error_fatal);
1707     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1708                               "date", &error_fatal);
1709 }
1710 
1711 /* Returns whether we want to use VGA or not */
1712 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1713 {
1714     switch (vga_interface_type) {
1715     case VGA_NONE:
1716         return false;
1717     case VGA_DEVICE:
1718         return true;
1719     case VGA_STD:
1720     case VGA_VIRTIO:
1721     case VGA_CIRRUS:
1722         return pci_vga_init(pci_bus) != NULL;
1723     default:
1724         error_setg(errp,
1725                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1726         return false;
1727     }
1728 }
1729 
1730 static int spapr_pre_load(void *opaque)
1731 {
1732     int rc;
1733 
1734     rc = spapr_caps_pre_load(opaque);
1735     if (rc) {
1736         return rc;
1737     }
1738 
1739     return 0;
1740 }
1741 
1742 static int spapr_post_load(void *opaque, int version_id)
1743 {
1744     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1745     int err = 0;
1746 
1747     err = spapr_caps_post_migration(spapr);
1748     if (err) {
1749         return err;
1750     }
1751 
1752     /*
1753      * In earlier versions, there was no separate qdev for the PAPR
1754      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1755      * So when migrating from those versions, poke the incoming offset
1756      * value into the RTC device
1757      */
1758     if (version_id < 3) {
1759         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1760         if (err) {
1761             return err;
1762         }
1763     }
1764 
1765     if (kvm_enabled() && spapr->patb_entry) {
1766         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1767         bool radix = !!(spapr->patb_entry & PATE1_GR);
1768         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1769 
1770         /*
1771          * Update LPCR:HR and UPRT as they may not be set properly in
1772          * the stream
1773          */
1774         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1775                             LPCR_HR | LPCR_UPRT);
1776 
1777         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1778         if (err) {
1779             error_report("Process table config unsupported by the host");
1780             return -EINVAL;
1781         }
1782     }
1783 
1784     err = spapr_irq_post_load(spapr, version_id);
1785     if (err) {
1786         return err;
1787     }
1788 
1789     return err;
1790 }
1791 
1792 static int spapr_pre_save(void *opaque)
1793 {
1794     int rc;
1795 
1796     rc = spapr_caps_pre_save(opaque);
1797     if (rc) {
1798         return rc;
1799     }
1800 
1801     return 0;
1802 }
1803 
1804 static bool version_before_3(void *opaque, int version_id)
1805 {
1806     return version_id < 3;
1807 }
1808 
1809 static bool spapr_pending_events_needed(void *opaque)
1810 {
1811     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1812     return !QTAILQ_EMPTY(&spapr->pending_events);
1813 }
1814 
1815 static const VMStateDescription vmstate_spapr_event_entry = {
1816     .name = "spapr_event_log_entry",
1817     .version_id = 1,
1818     .minimum_version_id = 1,
1819     .fields = (VMStateField[]) {
1820         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1821         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1822         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1823                                      NULL, extended_length),
1824         VMSTATE_END_OF_LIST()
1825     },
1826 };
1827 
1828 static const VMStateDescription vmstate_spapr_pending_events = {
1829     .name = "spapr_pending_events",
1830     .version_id = 1,
1831     .minimum_version_id = 1,
1832     .needed = spapr_pending_events_needed,
1833     .fields = (VMStateField[]) {
1834         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1835                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1836         VMSTATE_END_OF_LIST()
1837     },
1838 };
1839 
1840 static bool spapr_ov5_cas_needed(void *opaque)
1841 {
1842     SpaprMachineState *spapr = opaque;
1843     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1844     bool cas_needed;
1845 
1846     /* Prior to the introduction of SpaprOptionVector, we had two option
1847      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1848      * Both of these options encode machine topology into the device-tree
1849      * in such a way that the now-booted OS should still be able to interact
1850      * appropriately with QEMU regardless of what options were actually
1851      * negotiatied on the source side.
1852      *
1853      * As such, we can avoid migrating the CAS-negotiated options if these
1854      * are the only options available on the current machine/platform.
1855      * Since these are the only options available for pseries-2.7 and
1856      * earlier, this allows us to maintain old->new/new->old migration
1857      * compatibility.
1858      *
1859      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1860      * via default pseries-2.8 machines and explicit command-line parameters.
1861      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1862      * of the actual CAS-negotiated values to continue working properly. For
1863      * example, availability of memory unplug depends on knowing whether
1864      * OV5_HP_EVT was negotiated via CAS.
1865      *
1866      * Thus, for any cases where the set of available CAS-negotiatable
1867      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1868      * include the CAS-negotiated options in the migration stream, unless
1869      * if they affect boot time behaviour only.
1870      */
1871     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1872     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1873     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1874 
1875     /* We need extra information if we have any bits outside the mask
1876      * defined above */
1877     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1878 
1879     spapr_ovec_cleanup(ov5_mask);
1880 
1881     return cas_needed;
1882 }
1883 
1884 static const VMStateDescription vmstate_spapr_ov5_cas = {
1885     .name = "spapr_option_vector_ov5_cas",
1886     .version_id = 1,
1887     .minimum_version_id = 1,
1888     .needed = spapr_ov5_cas_needed,
1889     .fields = (VMStateField[]) {
1890         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1891                                  vmstate_spapr_ovec, SpaprOptionVector),
1892         VMSTATE_END_OF_LIST()
1893     },
1894 };
1895 
1896 static bool spapr_patb_entry_needed(void *opaque)
1897 {
1898     SpaprMachineState *spapr = opaque;
1899 
1900     return !!spapr->patb_entry;
1901 }
1902 
1903 static const VMStateDescription vmstate_spapr_patb_entry = {
1904     .name = "spapr_patb_entry",
1905     .version_id = 1,
1906     .minimum_version_id = 1,
1907     .needed = spapr_patb_entry_needed,
1908     .fields = (VMStateField[]) {
1909         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1910         VMSTATE_END_OF_LIST()
1911     },
1912 };
1913 
1914 static bool spapr_irq_map_needed(void *opaque)
1915 {
1916     SpaprMachineState *spapr = opaque;
1917 
1918     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1919 }
1920 
1921 static const VMStateDescription vmstate_spapr_irq_map = {
1922     .name = "spapr_irq_map",
1923     .version_id = 1,
1924     .minimum_version_id = 1,
1925     .needed = spapr_irq_map_needed,
1926     .fields = (VMStateField[]) {
1927         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1928         VMSTATE_END_OF_LIST()
1929     },
1930 };
1931 
1932 static bool spapr_dtb_needed(void *opaque)
1933 {
1934     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1935 
1936     return smc->update_dt_enabled;
1937 }
1938 
1939 static int spapr_dtb_pre_load(void *opaque)
1940 {
1941     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1942 
1943     g_free(spapr->fdt_blob);
1944     spapr->fdt_blob = NULL;
1945     spapr->fdt_size = 0;
1946 
1947     return 0;
1948 }
1949 
1950 static const VMStateDescription vmstate_spapr_dtb = {
1951     .name = "spapr_dtb",
1952     .version_id = 1,
1953     .minimum_version_id = 1,
1954     .needed = spapr_dtb_needed,
1955     .pre_load = spapr_dtb_pre_load,
1956     .fields = (VMStateField[]) {
1957         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1958         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1959         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1960                                      fdt_size),
1961         VMSTATE_END_OF_LIST()
1962     },
1963 };
1964 
1965 static const VMStateDescription vmstate_spapr = {
1966     .name = "spapr",
1967     .version_id = 3,
1968     .minimum_version_id = 1,
1969     .pre_load = spapr_pre_load,
1970     .post_load = spapr_post_load,
1971     .pre_save = spapr_pre_save,
1972     .fields = (VMStateField[]) {
1973         /* used to be @next_irq */
1974         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1975 
1976         /* RTC offset */
1977         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
1978 
1979         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
1980         VMSTATE_END_OF_LIST()
1981     },
1982     .subsections = (const VMStateDescription*[]) {
1983         &vmstate_spapr_ov5_cas,
1984         &vmstate_spapr_patb_entry,
1985         &vmstate_spapr_pending_events,
1986         &vmstate_spapr_cap_htm,
1987         &vmstate_spapr_cap_vsx,
1988         &vmstate_spapr_cap_dfp,
1989         &vmstate_spapr_cap_cfpc,
1990         &vmstate_spapr_cap_sbbc,
1991         &vmstate_spapr_cap_ibs,
1992         &vmstate_spapr_cap_hpt_maxpagesize,
1993         &vmstate_spapr_irq_map,
1994         &vmstate_spapr_cap_nested_kvm_hv,
1995         &vmstate_spapr_dtb,
1996         &vmstate_spapr_cap_large_decr,
1997         &vmstate_spapr_cap_ccf_assist,
1998         NULL
1999     }
2000 };
2001 
2002 static int htab_save_setup(QEMUFile *f, void *opaque)
2003 {
2004     SpaprMachineState *spapr = opaque;
2005 
2006     /* "Iteration" header */
2007     if (!spapr->htab_shift) {
2008         qemu_put_be32(f, -1);
2009     } else {
2010         qemu_put_be32(f, spapr->htab_shift);
2011     }
2012 
2013     if (spapr->htab) {
2014         spapr->htab_save_index = 0;
2015         spapr->htab_first_pass = true;
2016     } else {
2017         if (spapr->htab_shift) {
2018             assert(kvm_enabled());
2019         }
2020     }
2021 
2022 
2023     return 0;
2024 }
2025 
2026 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2027                             int chunkstart, int n_valid, int n_invalid)
2028 {
2029     qemu_put_be32(f, chunkstart);
2030     qemu_put_be16(f, n_valid);
2031     qemu_put_be16(f, n_invalid);
2032     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2033                     HASH_PTE_SIZE_64 * n_valid);
2034 }
2035 
2036 static void htab_save_end_marker(QEMUFile *f)
2037 {
2038     qemu_put_be32(f, 0);
2039     qemu_put_be16(f, 0);
2040     qemu_put_be16(f, 0);
2041 }
2042 
2043 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2044                                  int64_t max_ns)
2045 {
2046     bool has_timeout = max_ns != -1;
2047     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2048     int index = spapr->htab_save_index;
2049     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2050 
2051     assert(spapr->htab_first_pass);
2052 
2053     do {
2054         int chunkstart;
2055 
2056         /* Consume invalid HPTEs */
2057         while ((index < htabslots)
2058                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2059             CLEAN_HPTE(HPTE(spapr->htab, index));
2060             index++;
2061         }
2062 
2063         /* Consume valid HPTEs */
2064         chunkstart = index;
2065         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2066                && HPTE_VALID(HPTE(spapr->htab, index))) {
2067             CLEAN_HPTE(HPTE(spapr->htab, index));
2068             index++;
2069         }
2070 
2071         if (index > chunkstart) {
2072             int n_valid = index - chunkstart;
2073 
2074             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2075 
2076             if (has_timeout &&
2077                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2078                 break;
2079             }
2080         }
2081     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2082 
2083     if (index >= htabslots) {
2084         assert(index == htabslots);
2085         index = 0;
2086         spapr->htab_first_pass = false;
2087     }
2088     spapr->htab_save_index = index;
2089 }
2090 
2091 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2092                                 int64_t max_ns)
2093 {
2094     bool final = max_ns < 0;
2095     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2096     int examined = 0, sent = 0;
2097     int index = spapr->htab_save_index;
2098     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2099 
2100     assert(!spapr->htab_first_pass);
2101 
2102     do {
2103         int chunkstart, invalidstart;
2104 
2105         /* Consume non-dirty HPTEs */
2106         while ((index < htabslots)
2107                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2108             index++;
2109             examined++;
2110         }
2111 
2112         chunkstart = index;
2113         /* Consume valid dirty HPTEs */
2114         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2115                && HPTE_DIRTY(HPTE(spapr->htab, index))
2116                && HPTE_VALID(HPTE(spapr->htab, index))) {
2117             CLEAN_HPTE(HPTE(spapr->htab, index));
2118             index++;
2119             examined++;
2120         }
2121 
2122         invalidstart = index;
2123         /* Consume invalid dirty HPTEs */
2124         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2125                && HPTE_DIRTY(HPTE(spapr->htab, index))
2126                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2127             CLEAN_HPTE(HPTE(spapr->htab, index));
2128             index++;
2129             examined++;
2130         }
2131 
2132         if (index > chunkstart) {
2133             int n_valid = invalidstart - chunkstart;
2134             int n_invalid = index - invalidstart;
2135 
2136             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2137             sent += index - chunkstart;
2138 
2139             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2140                 break;
2141             }
2142         }
2143 
2144         if (examined >= htabslots) {
2145             break;
2146         }
2147 
2148         if (index >= htabslots) {
2149             assert(index == htabslots);
2150             index = 0;
2151         }
2152     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2153 
2154     if (index >= htabslots) {
2155         assert(index == htabslots);
2156         index = 0;
2157     }
2158 
2159     spapr->htab_save_index = index;
2160 
2161     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2162 }
2163 
2164 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2165 #define MAX_KVM_BUF_SIZE    2048
2166 
2167 static int htab_save_iterate(QEMUFile *f, void *opaque)
2168 {
2169     SpaprMachineState *spapr = opaque;
2170     int fd;
2171     int rc = 0;
2172 
2173     /* Iteration header */
2174     if (!spapr->htab_shift) {
2175         qemu_put_be32(f, -1);
2176         return 1;
2177     } else {
2178         qemu_put_be32(f, 0);
2179     }
2180 
2181     if (!spapr->htab) {
2182         assert(kvm_enabled());
2183 
2184         fd = get_htab_fd(spapr);
2185         if (fd < 0) {
2186             return fd;
2187         }
2188 
2189         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2190         if (rc < 0) {
2191             return rc;
2192         }
2193     } else  if (spapr->htab_first_pass) {
2194         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2195     } else {
2196         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2197     }
2198 
2199     htab_save_end_marker(f);
2200 
2201     return rc;
2202 }
2203 
2204 static int htab_save_complete(QEMUFile *f, void *opaque)
2205 {
2206     SpaprMachineState *spapr = opaque;
2207     int fd;
2208 
2209     /* Iteration header */
2210     if (!spapr->htab_shift) {
2211         qemu_put_be32(f, -1);
2212         return 0;
2213     } else {
2214         qemu_put_be32(f, 0);
2215     }
2216 
2217     if (!spapr->htab) {
2218         int rc;
2219 
2220         assert(kvm_enabled());
2221 
2222         fd = get_htab_fd(spapr);
2223         if (fd < 0) {
2224             return fd;
2225         }
2226 
2227         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2228         if (rc < 0) {
2229             return rc;
2230         }
2231     } else {
2232         if (spapr->htab_first_pass) {
2233             htab_save_first_pass(f, spapr, -1);
2234         }
2235         htab_save_later_pass(f, spapr, -1);
2236     }
2237 
2238     /* End marker */
2239     htab_save_end_marker(f);
2240 
2241     return 0;
2242 }
2243 
2244 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2245 {
2246     SpaprMachineState *spapr = opaque;
2247     uint32_t section_hdr;
2248     int fd = -1;
2249     Error *local_err = NULL;
2250 
2251     if (version_id < 1 || version_id > 1) {
2252         error_report("htab_load() bad version");
2253         return -EINVAL;
2254     }
2255 
2256     section_hdr = qemu_get_be32(f);
2257 
2258     if (section_hdr == -1) {
2259         spapr_free_hpt(spapr);
2260         return 0;
2261     }
2262 
2263     if (section_hdr) {
2264         /* First section gives the htab size */
2265         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2266         if (local_err) {
2267             error_report_err(local_err);
2268             return -EINVAL;
2269         }
2270         return 0;
2271     }
2272 
2273     if (!spapr->htab) {
2274         assert(kvm_enabled());
2275 
2276         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2277         if (fd < 0) {
2278             error_report_err(local_err);
2279             return fd;
2280         }
2281     }
2282 
2283     while (true) {
2284         uint32_t index;
2285         uint16_t n_valid, n_invalid;
2286 
2287         index = qemu_get_be32(f);
2288         n_valid = qemu_get_be16(f);
2289         n_invalid = qemu_get_be16(f);
2290 
2291         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2292             /* End of Stream */
2293             break;
2294         }
2295 
2296         if ((index + n_valid + n_invalid) >
2297             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2298             /* Bad index in stream */
2299             error_report(
2300                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2301                 index, n_valid, n_invalid, spapr->htab_shift);
2302             return -EINVAL;
2303         }
2304 
2305         if (spapr->htab) {
2306             if (n_valid) {
2307                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2308                                 HASH_PTE_SIZE_64 * n_valid);
2309             }
2310             if (n_invalid) {
2311                 memset(HPTE(spapr->htab, index + n_valid), 0,
2312                        HASH_PTE_SIZE_64 * n_invalid);
2313             }
2314         } else {
2315             int rc;
2316 
2317             assert(fd >= 0);
2318 
2319             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2320             if (rc < 0) {
2321                 return rc;
2322             }
2323         }
2324     }
2325 
2326     if (!spapr->htab) {
2327         assert(fd >= 0);
2328         close(fd);
2329     }
2330 
2331     return 0;
2332 }
2333 
2334 static void htab_save_cleanup(void *opaque)
2335 {
2336     SpaprMachineState *spapr = opaque;
2337 
2338     close_htab_fd(spapr);
2339 }
2340 
2341 static SaveVMHandlers savevm_htab_handlers = {
2342     .save_setup = htab_save_setup,
2343     .save_live_iterate = htab_save_iterate,
2344     .save_live_complete_precopy = htab_save_complete,
2345     .save_cleanup = htab_save_cleanup,
2346     .load_state = htab_load,
2347 };
2348 
2349 static void spapr_boot_set(void *opaque, const char *boot_device,
2350                            Error **errp)
2351 {
2352     MachineState *machine = MACHINE(opaque);
2353     machine->boot_order = g_strdup(boot_device);
2354 }
2355 
2356 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2357 {
2358     MachineState *machine = MACHINE(spapr);
2359     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2360     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2361     int i;
2362 
2363     for (i = 0; i < nr_lmbs; i++) {
2364         uint64_t addr;
2365 
2366         addr = i * lmb_size + machine->device_memory->base;
2367         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2368                                addr / lmb_size);
2369     }
2370 }
2371 
2372 /*
2373  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2374  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2375  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2376  */
2377 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2378 {
2379     int i;
2380 
2381     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2382         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2383                    " is not aligned to %" PRIu64 " MiB",
2384                    machine->ram_size,
2385                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2386         return;
2387     }
2388 
2389     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2390         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2391                    " is not aligned to %" PRIu64 " MiB",
2392                    machine->ram_size,
2393                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2394         return;
2395     }
2396 
2397     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2398         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2399             error_setg(errp,
2400                        "Node %d memory size 0x%" PRIx64
2401                        " is not aligned to %" PRIu64 " MiB",
2402                        i, machine->numa_state->nodes[i].node_mem,
2403                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2404             return;
2405         }
2406     }
2407 }
2408 
2409 /* find cpu slot in machine->possible_cpus by core_id */
2410 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2411 {
2412     int index = id / ms->smp.threads;
2413 
2414     if (index >= ms->possible_cpus->len) {
2415         return NULL;
2416     }
2417     if (idx) {
2418         *idx = index;
2419     }
2420     return &ms->possible_cpus->cpus[index];
2421 }
2422 
2423 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2424 {
2425     MachineState *ms = MACHINE(spapr);
2426     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2427     Error *local_err = NULL;
2428     bool vsmt_user = !!spapr->vsmt;
2429     int kvm_smt = kvmppc_smt_threads();
2430     int ret;
2431     unsigned int smp_threads = ms->smp.threads;
2432 
2433     if (!kvm_enabled() && (smp_threads > 1)) {
2434         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2435                      "on a pseries machine");
2436         goto out;
2437     }
2438     if (!is_power_of_2(smp_threads)) {
2439         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2440                      "machine because it must be a power of 2", smp_threads);
2441         goto out;
2442     }
2443 
2444     /* Detemine the VSMT mode to use: */
2445     if (vsmt_user) {
2446         if (spapr->vsmt < smp_threads) {
2447             error_setg(&local_err, "Cannot support VSMT mode %d"
2448                          " because it must be >= threads/core (%d)",
2449                          spapr->vsmt, smp_threads);
2450             goto out;
2451         }
2452         /* In this case, spapr->vsmt has been set by the command line */
2453     } else if (!smc->smp_threads_vsmt) {
2454         /*
2455          * Default VSMT value is tricky, because we need it to be as
2456          * consistent as possible (for migration), but this requires
2457          * changing it for at least some existing cases.  We pick 8 as
2458          * the value that we'd get with KVM on POWER8, the
2459          * overwhelmingly common case in production systems.
2460          */
2461         spapr->vsmt = MAX(8, smp_threads);
2462     } else {
2463         spapr->vsmt = smp_threads;
2464     }
2465 
2466     /* KVM: If necessary, set the SMT mode: */
2467     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2468         ret = kvmppc_set_smt_threads(spapr->vsmt);
2469         if (ret) {
2470             /* Looks like KVM isn't able to change VSMT mode */
2471             error_setg(&local_err,
2472                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2473                        spapr->vsmt, ret);
2474             /* We can live with that if the default one is big enough
2475              * for the number of threads, and a submultiple of the one
2476              * we want.  In this case we'll waste some vcpu ids, but
2477              * behaviour will be correct */
2478             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2479                 warn_report_err(local_err);
2480                 local_err = NULL;
2481                 goto out;
2482             } else {
2483                 if (!vsmt_user) {
2484                     error_append_hint(&local_err,
2485                                       "On PPC, a VM with %d threads/core"
2486                                       " on a host with %d threads/core"
2487                                       " requires the use of VSMT mode %d.\n",
2488                                       smp_threads, kvm_smt, spapr->vsmt);
2489                 }
2490                 kvmppc_error_append_smt_possible_hint(&local_err);
2491                 goto out;
2492             }
2493         }
2494     }
2495     /* else TCG: nothing to do currently */
2496 out:
2497     error_propagate(errp, local_err);
2498 }
2499 
2500 static void spapr_init_cpus(SpaprMachineState *spapr)
2501 {
2502     MachineState *machine = MACHINE(spapr);
2503     MachineClass *mc = MACHINE_GET_CLASS(machine);
2504     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2505     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2506     const CPUArchIdList *possible_cpus;
2507     unsigned int smp_cpus = machine->smp.cpus;
2508     unsigned int smp_threads = machine->smp.threads;
2509     unsigned int max_cpus = machine->smp.max_cpus;
2510     int boot_cores_nr = smp_cpus / smp_threads;
2511     int i;
2512 
2513     possible_cpus = mc->possible_cpu_arch_ids(machine);
2514     if (mc->has_hotpluggable_cpus) {
2515         if (smp_cpus % smp_threads) {
2516             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2517                          smp_cpus, smp_threads);
2518             exit(1);
2519         }
2520         if (max_cpus % smp_threads) {
2521             error_report("max_cpus (%u) must be multiple of threads (%u)",
2522                          max_cpus, smp_threads);
2523             exit(1);
2524         }
2525     } else {
2526         if (max_cpus != smp_cpus) {
2527             error_report("This machine version does not support CPU hotplug");
2528             exit(1);
2529         }
2530         boot_cores_nr = possible_cpus->len;
2531     }
2532 
2533     if (smc->pre_2_10_has_unused_icps) {
2534         int i;
2535 
2536         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2537             /* Dummy entries get deregistered when real ICPState objects
2538              * are registered during CPU core hotplug.
2539              */
2540             pre_2_10_vmstate_register_dummy_icp(i);
2541         }
2542     }
2543 
2544     for (i = 0; i < possible_cpus->len; i++) {
2545         int core_id = i * smp_threads;
2546 
2547         if (mc->has_hotpluggable_cpus) {
2548             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2549                                    spapr_vcpu_id(spapr, core_id));
2550         }
2551 
2552         if (i < boot_cores_nr) {
2553             Object *core  = object_new(type);
2554             int nr_threads = smp_threads;
2555 
2556             /* Handle the partially filled core for older machine types */
2557             if ((i + 1) * smp_threads >= smp_cpus) {
2558                 nr_threads = smp_cpus - i * smp_threads;
2559             }
2560 
2561             object_property_set_int(core, nr_threads, "nr-threads",
2562                                     &error_fatal);
2563             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2564                                     &error_fatal);
2565             object_property_set_bool(core, true, "realized", &error_fatal);
2566 
2567             object_unref(core);
2568         }
2569     }
2570 }
2571 
2572 static PCIHostState *spapr_create_default_phb(void)
2573 {
2574     DeviceState *dev;
2575 
2576     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2577     qdev_prop_set_uint32(dev, "index", 0);
2578     qdev_init_nofail(dev);
2579 
2580     return PCI_HOST_BRIDGE(dev);
2581 }
2582 
2583 /* pSeries LPAR / sPAPR hardware init */
2584 static void spapr_machine_init(MachineState *machine)
2585 {
2586     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2587     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2588     const char *kernel_filename = machine->kernel_filename;
2589     const char *initrd_filename = machine->initrd_filename;
2590     PCIHostState *phb;
2591     int i;
2592     MemoryRegion *sysmem = get_system_memory();
2593     MemoryRegion *ram = g_new(MemoryRegion, 1);
2594     hwaddr node0_size = spapr_node0_size(machine);
2595     long load_limit, fw_size;
2596     char *filename;
2597     Error *resize_hpt_err = NULL;
2598 
2599     msi_nonbroken = true;
2600 
2601     QLIST_INIT(&spapr->phbs);
2602     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2603 
2604     /* Determine capabilities to run with */
2605     spapr_caps_init(spapr);
2606 
2607     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2608     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2609         /*
2610          * If the user explicitly requested a mode we should either
2611          * supply it, or fail completely (which we do below).  But if
2612          * it's not set explicitly, we reset our mode to something
2613          * that works
2614          */
2615         if (resize_hpt_err) {
2616             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2617             error_free(resize_hpt_err);
2618             resize_hpt_err = NULL;
2619         } else {
2620             spapr->resize_hpt = smc->resize_hpt_default;
2621         }
2622     }
2623 
2624     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2625 
2626     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2627         /*
2628          * User requested HPT resize, but this host can't supply it.  Bail out
2629          */
2630         error_report_err(resize_hpt_err);
2631         exit(1);
2632     }
2633 
2634     spapr->rma_size = node0_size;
2635 
2636     /* With KVM, we don't actually know whether KVM supports an
2637      * unbounded RMA (PR KVM) or is limited by the hash table size
2638      * (HV KVM using VRMA), so we always assume the latter
2639      *
2640      * In that case, we also limit the initial allocations for RTAS
2641      * etc... to 256M since we have no way to know what the VRMA size
2642      * is going to be as it depends on the size of the hash table
2643      * which isn't determined yet.
2644      */
2645     if (kvm_enabled()) {
2646         spapr->vrma_adjust = 1;
2647         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2648     }
2649 
2650     /* Actually we don't support unbounded RMA anymore since we added
2651      * proper emulation of HV mode. The max we can get is 16G which
2652      * also happens to be what we configure for PAPR mode so make sure
2653      * we don't do anything bigger than that
2654      */
2655     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2656 
2657     if (spapr->rma_size > node0_size) {
2658         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2659                      spapr->rma_size);
2660         exit(1);
2661     }
2662 
2663     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2664     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2665 
2666     /*
2667      * VSMT must be set in order to be able to compute VCPU ids, ie to
2668      * call spapr_max_server_number() or spapr_vcpu_id().
2669      */
2670     spapr_set_vsmt_mode(spapr, &error_fatal);
2671 
2672     /* Set up Interrupt Controller before we create the VCPUs */
2673     spapr_irq_init(spapr, &error_fatal);
2674 
2675     /* Set up containers for ibm,client-architecture-support negotiated options
2676      */
2677     spapr->ov5 = spapr_ovec_new();
2678     spapr->ov5_cas = spapr_ovec_new();
2679 
2680     if (smc->dr_lmb_enabled) {
2681         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2682         spapr_validate_node_memory(machine, &error_fatal);
2683     }
2684 
2685     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2686 
2687     /* advertise support for dedicated HP event source to guests */
2688     if (spapr->use_hotplug_event_source) {
2689         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2690     }
2691 
2692     /* advertise support for HPT resizing */
2693     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2694         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2695     }
2696 
2697     /* advertise support for ibm,dyamic-memory-v2 */
2698     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2699 
2700     /* advertise XIVE on POWER9 machines */
2701     if (spapr->irq->xive) {
2702         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2703     }
2704 
2705     /* init CPUs */
2706     spapr_init_cpus(spapr);
2707 
2708     /*
2709      * check we don't have a memory-less/cpu-less NUMA node
2710      * Firmware relies on the existing memory/cpu topology to provide the
2711      * NUMA topology to the kernel.
2712      * And the linux kernel needs to know the NUMA topology at start
2713      * to be able to hotplug CPUs later.
2714      */
2715     if (machine->numa_state->num_nodes) {
2716         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2717             /* check for memory-less node */
2718             if (machine->numa_state->nodes[i].node_mem == 0) {
2719                 CPUState *cs;
2720                 int found = 0;
2721                 /* check for cpu-less node */
2722                 CPU_FOREACH(cs) {
2723                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2724                     if (cpu->node_id == i) {
2725                         found = 1;
2726                         break;
2727                     }
2728                 }
2729                 /* memory-less and cpu-less node */
2730                 if (!found) {
2731                     error_report(
2732                        "Memory-less/cpu-less nodes are not supported (node %d)",
2733                                  i);
2734                     exit(1);
2735                 }
2736             }
2737         }
2738 
2739     }
2740 
2741     /*
2742      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2743      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2744      * called from vPHB reset handler so we initialize the counter here.
2745      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2746      * must be equally distant from any other node.
2747      * The final value of spapr->gpu_numa_id is going to be written to
2748      * max-associativity-domains in spapr_build_fdt().
2749      */
2750     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2751 
2752     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2753         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2754                               spapr->max_compat_pvr)) {
2755         /* KVM and TCG always allow GTSE with radix... */
2756         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2757     }
2758     /* ... but not with hash (currently). */
2759 
2760     if (kvm_enabled()) {
2761         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2762         kvmppc_enable_logical_ci_hcalls();
2763         kvmppc_enable_set_mode_hcall();
2764 
2765         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2766         kvmppc_enable_clear_ref_mod_hcalls();
2767 
2768         /* Enable H_PAGE_INIT */
2769         kvmppc_enable_h_page_init();
2770     }
2771 
2772     /* allocate RAM */
2773     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2774                                          machine->ram_size);
2775     memory_region_add_subregion(sysmem, 0, ram);
2776 
2777     /* always allocate the device memory information */
2778     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2779 
2780     /* initialize hotplug memory address space */
2781     if (machine->ram_size < machine->maxram_size) {
2782         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2783         /*
2784          * Limit the number of hotpluggable memory slots to half the number
2785          * slots that KVM supports, leaving the other half for PCI and other
2786          * devices. However ensure that number of slots doesn't drop below 32.
2787          */
2788         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2789                            SPAPR_MAX_RAM_SLOTS;
2790 
2791         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2792             max_memslots = SPAPR_MAX_RAM_SLOTS;
2793         }
2794         if (machine->ram_slots > max_memslots) {
2795             error_report("Specified number of memory slots %"
2796                          PRIu64" exceeds max supported %d",
2797                          machine->ram_slots, max_memslots);
2798             exit(1);
2799         }
2800 
2801         machine->device_memory->base = ROUND_UP(machine->ram_size,
2802                                                 SPAPR_DEVICE_MEM_ALIGN);
2803         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2804                            "device-memory", device_mem_size);
2805         memory_region_add_subregion(sysmem, machine->device_memory->base,
2806                                     &machine->device_memory->mr);
2807     }
2808 
2809     if (smc->dr_lmb_enabled) {
2810         spapr_create_lmb_dr_connectors(spapr);
2811     }
2812 
2813     /* Set up RTAS event infrastructure */
2814     spapr_events_init(spapr);
2815 
2816     /* Set up the RTC RTAS interfaces */
2817     spapr_rtc_create(spapr);
2818 
2819     /* Set up VIO bus */
2820     spapr->vio_bus = spapr_vio_bus_init();
2821 
2822     for (i = 0; i < serial_max_hds(); i++) {
2823         if (serial_hd(i)) {
2824             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2825         }
2826     }
2827 
2828     /* We always have at least the nvram device on VIO */
2829     spapr_create_nvram(spapr);
2830 
2831     /*
2832      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2833      * connectors (described in root DT node's "ibm,drc-types" property)
2834      * are pre-initialized here. additional child connectors (such as
2835      * connectors for a PHBs PCI slots) are added as needed during their
2836      * parent's realization.
2837      */
2838     if (smc->dr_phb_enabled) {
2839         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2840             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2841         }
2842     }
2843 
2844     /* Set up PCI */
2845     spapr_pci_rtas_init();
2846 
2847     phb = spapr_create_default_phb();
2848 
2849     for (i = 0; i < nb_nics; i++) {
2850         NICInfo *nd = &nd_table[i];
2851 
2852         if (!nd->model) {
2853             nd->model = g_strdup("spapr-vlan");
2854         }
2855 
2856         if (g_str_equal(nd->model, "spapr-vlan") ||
2857             g_str_equal(nd->model, "ibmveth")) {
2858             spapr_vlan_create(spapr->vio_bus, nd);
2859         } else {
2860             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2861         }
2862     }
2863 
2864     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2865         spapr_vscsi_create(spapr->vio_bus);
2866     }
2867 
2868     /* Graphics */
2869     if (spapr_vga_init(phb->bus, &error_fatal)) {
2870         spapr->has_graphics = true;
2871         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2872     }
2873 
2874     if (machine->usb) {
2875         if (smc->use_ohci_by_default) {
2876             pci_create_simple(phb->bus, -1, "pci-ohci");
2877         } else {
2878             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2879         }
2880 
2881         if (spapr->has_graphics) {
2882             USBBus *usb_bus = usb_bus_find(-1);
2883 
2884             usb_create_simple(usb_bus, "usb-kbd");
2885             usb_create_simple(usb_bus, "usb-mouse");
2886         }
2887     }
2888 
2889     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2890         error_report(
2891             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2892             MIN_RMA_SLOF);
2893         exit(1);
2894     }
2895 
2896     if (kernel_filename) {
2897         uint64_t lowaddr = 0;
2898 
2899         spapr->kernel_size = load_elf(kernel_filename, NULL,
2900                                       translate_kernel_address, NULL,
2901                                       NULL, &lowaddr, NULL, 1,
2902                                       PPC_ELF_MACHINE, 0, 0);
2903         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2904             spapr->kernel_size = load_elf(kernel_filename, NULL,
2905                                           translate_kernel_address, NULL, NULL,
2906                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2907                                           0, 0);
2908             spapr->kernel_le = spapr->kernel_size > 0;
2909         }
2910         if (spapr->kernel_size < 0) {
2911             error_report("error loading %s: %s", kernel_filename,
2912                          load_elf_strerror(spapr->kernel_size));
2913             exit(1);
2914         }
2915 
2916         /* load initrd */
2917         if (initrd_filename) {
2918             /* Try to locate the initrd in the gap between the kernel
2919              * and the firmware. Add a bit of space just in case
2920              */
2921             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2922                                   + 0x1ffff) & ~0xffff;
2923             spapr->initrd_size = load_image_targphys(initrd_filename,
2924                                                      spapr->initrd_base,
2925                                                      load_limit
2926                                                      - spapr->initrd_base);
2927             if (spapr->initrd_size < 0) {
2928                 error_report("could not load initial ram disk '%s'",
2929                              initrd_filename);
2930                 exit(1);
2931             }
2932         }
2933     }
2934 
2935     if (bios_name == NULL) {
2936         bios_name = FW_FILE_NAME;
2937     }
2938     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2939     if (!filename) {
2940         error_report("Could not find LPAR firmware '%s'", bios_name);
2941         exit(1);
2942     }
2943     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2944     if (fw_size <= 0) {
2945         error_report("Could not load LPAR firmware '%s'", filename);
2946         exit(1);
2947     }
2948     g_free(filename);
2949 
2950     /* FIXME: Should register things through the MachineState's qdev
2951      * interface, this is a legacy from the sPAPREnvironment structure
2952      * which predated MachineState but had a similar function */
2953     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2954     register_savevm_live("spapr/htab", -1, 1,
2955                          &savevm_htab_handlers, spapr);
2956 
2957     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
2958                              &error_fatal);
2959 
2960     qemu_register_boot_set(spapr_boot_set, spapr);
2961 
2962     /*
2963      * Nothing needs to be done to resume a suspended guest because
2964      * suspending does not change the machine state, so no need for
2965      * a ->wakeup method.
2966      */
2967     qemu_register_wakeup_support();
2968 
2969     if (kvm_enabled()) {
2970         /* to stop and start vmclock */
2971         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2972                                          &spapr->tb);
2973 
2974         kvmppc_spapr_enable_inkernel_multitce();
2975     }
2976 }
2977 
2978 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
2979 {
2980     if (!vm_type) {
2981         return 0;
2982     }
2983 
2984     if (!strcmp(vm_type, "HV")) {
2985         return 1;
2986     }
2987 
2988     if (!strcmp(vm_type, "PR")) {
2989         return 2;
2990     }
2991 
2992     error_report("Unknown kvm-type specified '%s'", vm_type);
2993     exit(1);
2994 }
2995 
2996 /*
2997  * Implementation of an interface to adjust firmware path
2998  * for the bootindex property handling.
2999  */
3000 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3001                                    DeviceState *dev)
3002 {
3003 #define CAST(type, obj, name) \
3004     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3005     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3006     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3007     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3008 
3009     if (d) {
3010         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3011         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3012         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3013 
3014         if (spapr) {
3015             /*
3016              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3017              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3018              * 0x8000 | (target << 8) | (bus << 5) | lun
3019              * (see the "Logical unit addressing format" table in SAM5)
3020              */
3021             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3022             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3023                                    (uint64_t)id << 48);
3024         } else if (virtio) {
3025             /*
3026              * We use SRP luns of the form 01000000 | (target << 8) | lun
3027              * in the top 32 bits of the 64-bit LUN
3028              * Note: the quote above is from SLOF and it is wrong,
3029              * the actual binding is:
3030              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3031              */
3032             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3033             if (d->lun >= 256) {
3034                 /* Use the LUN "flat space addressing method" */
3035                 id |= 0x4000;
3036             }
3037             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3038                                    (uint64_t)id << 32);
3039         } else if (usb) {
3040             /*
3041              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3042              * in the top 32 bits of the 64-bit LUN
3043              */
3044             unsigned usb_port = atoi(usb->port->path);
3045             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3046             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3047                                    (uint64_t)id << 32);
3048         }
3049     }
3050 
3051     /*
3052      * SLOF probes the USB devices, and if it recognizes that the device is a
3053      * storage device, it changes its name to "storage" instead of "usb-host",
3054      * and additionally adds a child node for the SCSI LUN, so the correct
3055      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3056      */
3057     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3058         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3059         if (usb_host_dev_is_scsi_storage(usbdev)) {
3060             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3061         }
3062     }
3063 
3064     if (phb) {
3065         /* Replace "pci" with "pci@800000020000000" */
3066         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3067     }
3068 
3069     if (vsc) {
3070         /* Same logic as virtio above */
3071         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3072         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3073     }
3074 
3075     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3076         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3077         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3078         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3079     }
3080 
3081     return NULL;
3082 }
3083 
3084 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3085 {
3086     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3087 
3088     return g_strdup(spapr->kvm_type);
3089 }
3090 
3091 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3092 {
3093     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3094 
3095     g_free(spapr->kvm_type);
3096     spapr->kvm_type = g_strdup(value);
3097 }
3098 
3099 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3100 {
3101     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3102 
3103     return spapr->use_hotplug_event_source;
3104 }
3105 
3106 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3107                                             Error **errp)
3108 {
3109     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3110 
3111     spapr->use_hotplug_event_source = value;
3112 }
3113 
3114 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3115 {
3116     return true;
3117 }
3118 
3119 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3120 {
3121     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3122 
3123     switch (spapr->resize_hpt) {
3124     case SPAPR_RESIZE_HPT_DEFAULT:
3125         return g_strdup("default");
3126     case SPAPR_RESIZE_HPT_DISABLED:
3127         return g_strdup("disabled");
3128     case SPAPR_RESIZE_HPT_ENABLED:
3129         return g_strdup("enabled");
3130     case SPAPR_RESIZE_HPT_REQUIRED:
3131         return g_strdup("required");
3132     }
3133     g_assert_not_reached();
3134 }
3135 
3136 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3137 {
3138     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3139 
3140     if (strcmp(value, "default") == 0) {
3141         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3142     } else if (strcmp(value, "disabled") == 0) {
3143         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3144     } else if (strcmp(value, "enabled") == 0) {
3145         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3146     } else if (strcmp(value, "required") == 0) {
3147         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3148     } else {
3149         error_setg(errp, "Bad value for \"resize-hpt\" property");
3150     }
3151 }
3152 
3153 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3154                                    void *opaque, Error **errp)
3155 {
3156     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3157 }
3158 
3159 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3160                                    void *opaque, Error **errp)
3161 {
3162     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3163 }
3164 
3165 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3166 {
3167     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3168 
3169     if (spapr->irq == &spapr_irq_xics_legacy) {
3170         return g_strdup("legacy");
3171     } else if (spapr->irq == &spapr_irq_xics) {
3172         return g_strdup("xics");
3173     } else if (spapr->irq == &spapr_irq_xive) {
3174         return g_strdup("xive");
3175     } else if (spapr->irq == &spapr_irq_dual) {
3176         return g_strdup("dual");
3177     }
3178     g_assert_not_reached();
3179 }
3180 
3181 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3182 {
3183     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3184 
3185     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3186         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3187         return;
3188     }
3189 
3190     /* The legacy IRQ backend can not be set */
3191     if (strcmp(value, "xics") == 0) {
3192         spapr->irq = &spapr_irq_xics;
3193     } else if (strcmp(value, "xive") == 0) {
3194         spapr->irq = &spapr_irq_xive;
3195     } else if (strcmp(value, "dual") == 0) {
3196         spapr->irq = &spapr_irq_dual;
3197     } else {
3198         error_setg(errp, "Bad value for \"ic-mode\" property");
3199     }
3200 }
3201 
3202 static char *spapr_get_host_model(Object *obj, Error **errp)
3203 {
3204     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3205 
3206     return g_strdup(spapr->host_model);
3207 }
3208 
3209 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3210 {
3211     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3212 
3213     g_free(spapr->host_model);
3214     spapr->host_model = g_strdup(value);
3215 }
3216 
3217 static char *spapr_get_host_serial(Object *obj, Error **errp)
3218 {
3219     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3220 
3221     return g_strdup(spapr->host_serial);
3222 }
3223 
3224 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3225 {
3226     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3227 
3228     g_free(spapr->host_serial);
3229     spapr->host_serial = g_strdup(value);
3230 }
3231 
3232 static void spapr_instance_init(Object *obj)
3233 {
3234     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3235     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3236 
3237     spapr->htab_fd = -1;
3238     spapr->use_hotplug_event_source = true;
3239     object_property_add_str(obj, "kvm-type",
3240                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3241     object_property_set_description(obj, "kvm-type",
3242                                     "Specifies the KVM virtualization mode (HV, PR)",
3243                                     NULL);
3244     object_property_add_bool(obj, "modern-hotplug-events",
3245                             spapr_get_modern_hotplug_events,
3246                             spapr_set_modern_hotplug_events,
3247                             NULL);
3248     object_property_set_description(obj, "modern-hotplug-events",
3249                                     "Use dedicated hotplug event mechanism in"
3250                                     " place of standard EPOW events when possible"
3251                                     " (required for memory hot-unplug support)",
3252                                     NULL);
3253     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3254                             "Maximum permitted CPU compatibility mode",
3255                             &error_fatal);
3256 
3257     object_property_add_str(obj, "resize-hpt",
3258                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3259     object_property_set_description(obj, "resize-hpt",
3260                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3261                                     NULL);
3262     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3263                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3264     object_property_set_description(obj, "vsmt",
3265                                     "Virtual SMT: KVM behaves as if this were"
3266                                     " the host's SMT mode", &error_abort);
3267     object_property_add_bool(obj, "vfio-no-msix-emulation",
3268                              spapr_get_msix_emulation, NULL, NULL);
3269 
3270     /* The machine class defines the default interrupt controller mode */
3271     spapr->irq = smc->irq;
3272     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3273                             spapr_set_ic_mode, NULL);
3274     object_property_set_description(obj, "ic-mode",
3275                  "Specifies the interrupt controller mode (xics, xive, dual)",
3276                  NULL);
3277 
3278     object_property_add_str(obj, "host-model",
3279         spapr_get_host_model, spapr_set_host_model,
3280         &error_abort);
3281     object_property_set_description(obj, "host-model",
3282         "Host model to advertise in guest device tree", &error_abort);
3283     object_property_add_str(obj, "host-serial",
3284         spapr_get_host_serial, spapr_set_host_serial,
3285         &error_abort);
3286     object_property_set_description(obj, "host-serial",
3287         "Host serial number to advertise in guest device tree", &error_abort);
3288 }
3289 
3290 static void spapr_machine_finalizefn(Object *obj)
3291 {
3292     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3293 
3294     g_free(spapr->kvm_type);
3295 }
3296 
3297 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3298 {
3299     cpu_synchronize_state(cs);
3300     ppc_cpu_do_system_reset(cs);
3301 }
3302 
3303 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3304 {
3305     CPUState *cs;
3306 
3307     CPU_FOREACH(cs) {
3308         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3309     }
3310 }
3311 
3312 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3313                           void *fdt, int *fdt_start_offset, Error **errp)
3314 {
3315     uint64_t addr;
3316     uint32_t node;
3317 
3318     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3319     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3320                                     &error_abort);
3321     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3322                                                    SPAPR_MEMORY_BLOCK_SIZE);
3323     return 0;
3324 }
3325 
3326 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3327                            bool dedicated_hp_event_source, Error **errp)
3328 {
3329     SpaprDrc *drc;
3330     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3331     int i;
3332     uint64_t addr = addr_start;
3333     bool hotplugged = spapr_drc_hotplugged(dev);
3334     Error *local_err = NULL;
3335 
3336     for (i = 0; i < nr_lmbs; i++) {
3337         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3338                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3339         g_assert(drc);
3340 
3341         spapr_drc_attach(drc, dev, &local_err);
3342         if (local_err) {
3343             while (addr > addr_start) {
3344                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3345                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3346                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3347                 spapr_drc_detach(drc);
3348             }
3349             error_propagate(errp, local_err);
3350             return;
3351         }
3352         if (!hotplugged) {
3353             spapr_drc_reset(drc);
3354         }
3355         addr += SPAPR_MEMORY_BLOCK_SIZE;
3356     }
3357     /* send hotplug notification to the
3358      * guest only in case of hotplugged memory
3359      */
3360     if (hotplugged) {
3361         if (dedicated_hp_event_source) {
3362             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3363                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3364             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3365                                                    nr_lmbs,
3366                                                    spapr_drc_index(drc));
3367         } else {
3368             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3369                                            nr_lmbs);
3370         }
3371     }
3372 }
3373 
3374 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3375                               Error **errp)
3376 {
3377     Error *local_err = NULL;
3378     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3379     PCDIMMDevice *dimm = PC_DIMM(dev);
3380     uint64_t size, addr;
3381 
3382     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3383 
3384     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3385     if (local_err) {
3386         goto out;
3387     }
3388 
3389     addr = object_property_get_uint(OBJECT(dimm),
3390                                     PC_DIMM_ADDR_PROP, &local_err);
3391     if (local_err) {
3392         goto out_unplug;
3393     }
3394 
3395     spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3396                    &local_err);
3397     if (local_err) {
3398         goto out_unplug;
3399     }
3400 
3401     return;
3402 
3403 out_unplug:
3404     pc_dimm_unplug(dimm, MACHINE(ms));
3405 out:
3406     error_propagate(errp, local_err);
3407 }
3408 
3409 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3410                                   Error **errp)
3411 {
3412     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3413     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3414     PCDIMMDevice *dimm = PC_DIMM(dev);
3415     Error *local_err = NULL;
3416     uint64_t size;
3417     Object *memdev;
3418     hwaddr pagesize;
3419 
3420     if (!smc->dr_lmb_enabled) {
3421         error_setg(errp, "Memory hotplug not supported for this machine");
3422         return;
3423     }
3424 
3425     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3426     if (local_err) {
3427         error_propagate(errp, local_err);
3428         return;
3429     }
3430 
3431     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3432         error_setg(errp, "Hotplugged memory size must be a multiple of "
3433                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3434         return;
3435     }
3436 
3437     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3438                                       &error_abort);
3439     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3440     spapr_check_pagesize(spapr, pagesize, &local_err);
3441     if (local_err) {
3442         error_propagate(errp, local_err);
3443         return;
3444     }
3445 
3446     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3447 }
3448 
3449 struct SpaprDimmState {
3450     PCDIMMDevice *dimm;
3451     uint32_t nr_lmbs;
3452     QTAILQ_ENTRY(SpaprDimmState) next;
3453 };
3454 
3455 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3456                                                        PCDIMMDevice *dimm)
3457 {
3458     SpaprDimmState *dimm_state = NULL;
3459 
3460     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3461         if (dimm_state->dimm == dimm) {
3462             break;
3463         }
3464     }
3465     return dimm_state;
3466 }
3467 
3468 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3469                                                       uint32_t nr_lmbs,
3470                                                       PCDIMMDevice *dimm)
3471 {
3472     SpaprDimmState *ds = NULL;
3473 
3474     /*
3475      * If this request is for a DIMM whose removal had failed earlier
3476      * (due to guest's refusal to remove the LMBs), we would have this
3477      * dimm already in the pending_dimm_unplugs list. In that
3478      * case don't add again.
3479      */
3480     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3481     if (!ds) {
3482         ds = g_malloc0(sizeof(SpaprDimmState));
3483         ds->nr_lmbs = nr_lmbs;
3484         ds->dimm = dimm;
3485         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3486     }
3487     return ds;
3488 }
3489 
3490 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3491                                               SpaprDimmState *dimm_state)
3492 {
3493     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3494     g_free(dimm_state);
3495 }
3496 
3497 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3498                                                         PCDIMMDevice *dimm)
3499 {
3500     SpaprDrc *drc;
3501     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3502                                                   &error_abort);
3503     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3504     uint32_t avail_lmbs = 0;
3505     uint64_t addr_start, addr;
3506     int i;
3507 
3508     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3509                                          &error_abort);
3510 
3511     addr = addr_start;
3512     for (i = 0; i < nr_lmbs; i++) {
3513         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3514                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3515         g_assert(drc);
3516         if (drc->dev) {
3517             avail_lmbs++;
3518         }
3519         addr += SPAPR_MEMORY_BLOCK_SIZE;
3520     }
3521 
3522     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3523 }
3524 
3525 /* Callback to be called during DRC release. */
3526 void spapr_lmb_release(DeviceState *dev)
3527 {
3528     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3529     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3530     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3531 
3532     /* This information will get lost if a migration occurs
3533      * during the unplug process. In this case recover it. */
3534     if (ds == NULL) {
3535         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3536         g_assert(ds);
3537         /* The DRC being examined by the caller at least must be counted */
3538         g_assert(ds->nr_lmbs);
3539     }
3540 
3541     if (--ds->nr_lmbs) {
3542         return;
3543     }
3544 
3545     /*
3546      * Now that all the LMBs have been removed by the guest, call the
3547      * unplug handler chain. This can never fail.
3548      */
3549     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3550     object_unparent(OBJECT(dev));
3551 }
3552 
3553 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3554 {
3555     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3556     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3557 
3558     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3559     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3560     spapr_pending_dimm_unplugs_remove(spapr, ds);
3561 }
3562 
3563 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3564                                         DeviceState *dev, Error **errp)
3565 {
3566     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3567     Error *local_err = NULL;
3568     PCDIMMDevice *dimm = PC_DIMM(dev);
3569     uint32_t nr_lmbs;
3570     uint64_t size, addr_start, addr;
3571     int i;
3572     SpaprDrc *drc;
3573 
3574     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3575     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3576 
3577     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3578                                          &local_err);
3579     if (local_err) {
3580         goto out;
3581     }
3582 
3583     /*
3584      * An existing pending dimm state for this DIMM means that there is an
3585      * unplug operation in progress, waiting for the spapr_lmb_release
3586      * callback to complete the job (BQL can't cover that far). In this case,
3587      * bail out to avoid detaching DRCs that were already released.
3588      */
3589     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3590         error_setg(&local_err,
3591                    "Memory unplug already in progress for device %s",
3592                    dev->id);
3593         goto out;
3594     }
3595 
3596     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3597 
3598     addr = addr_start;
3599     for (i = 0; i < nr_lmbs; i++) {
3600         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3601                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3602         g_assert(drc);
3603 
3604         spapr_drc_detach(drc);
3605         addr += SPAPR_MEMORY_BLOCK_SIZE;
3606     }
3607 
3608     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3609                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3610     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3611                                               nr_lmbs, spapr_drc_index(drc));
3612 out:
3613     error_propagate(errp, local_err);
3614 }
3615 
3616 /* Callback to be called during DRC release. */
3617 void spapr_core_release(DeviceState *dev)
3618 {
3619     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3620 
3621     /* Call the unplug handler chain. This can never fail. */
3622     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3623     object_unparent(OBJECT(dev));
3624 }
3625 
3626 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3627 {
3628     MachineState *ms = MACHINE(hotplug_dev);
3629     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3630     CPUCore *cc = CPU_CORE(dev);
3631     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3632 
3633     if (smc->pre_2_10_has_unused_icps) {
3634         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3635         int i;
3636 
3637         for (i = 0; i < cc->nr_threads; i++) {
3638             CPUState *cs = CPU(sc->threads[i]);
3639 
3640             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3641         }
3642     }
3643 
3644     assert(core_slot);
3645     core_slot->cpu = NULL;
3646     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3647 }
3648 
3649 static
3650 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3651                                Error **errp)
3652 {
3653     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3654     int index;
3655     SpaprDrc *drc;
3656     CPUCore *cc = CPU_CORE(dev);
3657 
3658     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3659         error_setg(errp, "Unable to find CPU core with core-id: %d",
3660                    cc->core_id);
3661         return;
3662     }
3663     if (index == 0) {
3664         error_setg(errp, "Boot CPU core may not be unplugged");
3665         return;
3666     }
3667 
3668     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3669                           spapr_vcpu_id(spapr, cc->core_id));
3670     g_assert(drc);
3671 
3672     if (!spapr_drc_unplug_requested(drc)) {
3673         spapr_drc_detach(drc);
3674         spapr_hotplug_req_remove_by_index(drc);
3675     }
3676 }
3677 
3678 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3679                            void *fdt, int *fdt_start_offset, Error **errp)
3680 {
3681     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3682     CPUState *cs = CPU(core->threads[0]);
3683     PowerPCCPU *cpu = POWERPC_CPU(cs);
3684     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3685     int id = spapr_get_vcpu_id(cpu);
3686     char *nodename;
3687     int offset;
3688 
3689     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3690     offset = fdt_add_subnode(fdt, 0, nodename);
3691     g_free(nodename);
3692 
3693     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3694 
3695     *fdt_start_offset = offset;
3696     return 0;
3697 }
3698 
3699 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3700                             Error **errp)
3701 {
3702     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3703     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3704     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3705     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3706     CPUCore *cc = CPU_CORE(dev);
3707     CPUState *cs;
3708     SpaprDrc *drc;
3709     Error *local_err = NULL;
3710     CPUArchId *core_slot;
3711     int index;
3712     bool hotplugged = spapr_drc_hotplugged(dev);
3713     int i;
3714 
3715     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3716     if (!core_slot) {
3717         error_setg(errp, "Unable to find CPU core with core-id: %d",
3718                    cc->core_id);
3719         return;
3720     }
3721     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3722                           spapr_vcpu_id(spapr, cc->core_id));
3723 
3724     g_assert(drc || !mc->has_hotpluggable_cpus);
3725 
3726     if (drc) {
3727         spapr_drc_attach(drc, dev, &local_err);
3728         if (local_err) {
3729             error_propagate(errp, local_err);
3730             return;
3731         }
3732 
3733         if (hotplugged) {
3734             /*
3735              * Send hotplug notification interrupt to the guest only
3736              * in case of hotplugged CPUs.
3737              */
3738             spapr_hotplug_req_add_by_index(drc);
3739         } else {
3740             spapr_drc_reset(drc);
3741         }
3742     }
3743 
3744     core_slot->cpu = OBJECT(dev);
3745 
3746     if (smc->pre_2_10_has_unused_icps) {
3747         for (i = 0; i < cc->nr_threads; i++) {
3748             cs = CPU(core->threads[i]);
3749             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3750         }
3751     }
3752 
3753     /*
3754      * Set compatibility mode to match the boot CPU, which was either set
3755      * by the machine reset code or by CAS.
3756      */
3757     if (hotplugged) {
3758         for (i = 0; i < cc->nr_threads; i++) {
3759             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3760                            &local_err);
3761             if (local_err) {
3762                 error_propagate(errp, local_err);
3763                 return;
3764             }
3765         }
3766     }
3767 }
3768 
3769 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3770                                 Error **errp)
3771 {
3772     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3773     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3774     Error *local_err = NULL;
3775     CPUCore *cc = CPU_CORE(dev);
3776     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3777     const char *type = object_get_typename(OBJECT(dev));
3778     CPUArchId *core_slot;
3779     int index;
3780     unsigned int smp_threads = machine->smp.threads;
3781 
3782     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3783         error_setg(&local_err, "CPU hotplug not supported for this machine");
3784         goto out;
3785     }
3786 
3787     if (strcmp(base_core_type, type)) {
3788         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3789         goto out;
3790     }
3791 
3792     if (cc->core_id % smp_threads) {
3793         error_setg(&local_err, "invalid core id %d", cc->core_id);
3794         goto out;
3795     }
3796 
3797     /*
3798      * In general we should have homogeneous threads-per-core, but old
3799      * (pre hotplug support) machine types allow the last core to have
3800      * reduced threads as a compatibility hack for when we allowed
3801      * total vcpus not a multiple of threads-per-core.
3802      */
3803     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3804         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3805                    cc->nr_threads, smp_threads);
3806         goto out;
3807     }
3808 
3809     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3810     if (!core_slot) {
3811         error_setg(&local_err, "core id %d out of range", cc->core_id);
3812         goto out;
3813     }
3814 
3815     if (core_slot->cpu) {
3816         error_setg(&local_err, "core %d already populated", cc->core_id);
3817         goto out;
3818     }
3819 
3820     numa_cpu_pre_plug(core_slot, dev, &local_err);
3821 
3822 out:
3823     error_propagate(errp, local_err);
3824 }
3825 
3826 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3827                           void *fdt, int *fdt_start_offset, Error **errp)
3828 {
3829     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3830     int intc_phandle;
3831 
3832     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3833     if (intc_phandle <= 0) {
3834         return -1;
3835     }
3836 
3837     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3838         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3839         return -1;
3840     }
3841 
3842     /* generally SLOF creates these, for hotplug it's up to QEMU */
3843     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3844 
3845     return 0;
3846 }
3847 
3848 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3849                                Error **errp)
3850 {
3851     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3852     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3853     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3854     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3855 
3856     if (dev->hotplugged && !smc->dr_phb_enabled) {
3857         error_setg(errp, "PHB hotplug not supported for this machine");
3858         return;
3859     }
3860 
3861     if (sphb->index == (uint32_t)-1) {
3862         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3863         return;
3864     }
3865 
3866     /*
3867      * This will check that sphb->index doesn't exceed the maximum number of
3868      * PHBs for the current machine type.
3869      */
3870     smc->phb_placement(spapr, sphb->index,
3871                        &sphb->buid, &sphb->io_win_addr,
3872                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3873                        windows_supported, sphb->dma_liobn,
3874                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3875                        errp);
3876 }
3877 
3878 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3879                            Error **errp)
3880 {
3881     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3882     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3883     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3884     SpaprDrc *drc;
3885     bool hotplugged = spapr_drc_hotplugged(dev);
3886     Error *local_err = NULL;
3887 
3888     if (!smc->dr_phb_enabled) {
3889         return;
3890     }
3891 
3892     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3893     /* hotplug hooks should check it's enabled before getting this far */
3894     assert(drc);
3895 
3896     spapr_drc_attach(drc, DEVICE(dev), &local_err);
3897     if (local_err) {
3898         error_propagate(errp, local_err);
3899         return;
3900     }
3901 
3902     if (hotplugged) {
3903         spapr_hotplug_req_add_by_index(drc);
3904     } else {
3905         spapr_drc_reset(drc);
3906     }
3907 }
3908 
3909 void spapr_phb_release(DeviceState *dev)
3910 {
3911     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3912 
3913     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3914     object_unparent(OBJECT(dev));
3915 }
3916 
3917 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3918 {
3919     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3920 }
3921 
3922 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3923                                      DeviceState *dev, Error **errp)
3924 {
3925     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3926     SpaprDrc *drc;
3927 
3928     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3929     assert(drc);
3930 
3931     if (!spapr_drc_unplug_requested(drc)) {
3932         spapr_drc_detach(drc);
3933         spapr_hotplug_req_remove_by_index(drc);
3934     }
3935 }
3936 
3937 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3938                                  Error **errp)
3939 {
3940     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3941     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
3942 
3943     if (spapr->tpm_proxy != NULL) {
3944         error_setg(errp, "Only one TPM proxy can be specified for this machine");
3945         return;
3946     }
3947 
3948     spapr->tpm_proxy = tpm_proxy;
3949 }
3950 
3951 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3952 {
3953     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3954 
3955     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3956     object_unparent(OBJECT(dev));
3957     spapr->tpm_proxy = NULL;
3958 }
3959 
3960 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3961                                       DeviceState *dev, Error **errp)
3962 {
3963     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3964         spapr_memory_plug(hotplug_dev, dev, errp);
3965     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3966         spapr_core_plug(hotplug_dev, dev, errp);
3967     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
3968         spapr_phb_plug(hotplug_dev, dev, errp);
3969     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
3970         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
3971     }
3972 }
3973 
3974 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3975                                         DeviceState *dev, Error **errp)
3976 {
3977     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3978         spapr_memory_unplug(hotplug_dev, dev);
3979     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3980         spapr_core_unplug(hotplug_dev, dev);
3981     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
3982         spapr_phb_unplug(hotplug_dev, dev);
3983     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
3984         spapr_tpm_proxy_unplug(hotplug_dev, dev);
3985     }
3986 }
3987 
3988 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3989                                                 DeviceState *dev, Error **errp)
3990 {
3991     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3992     MachineClass *mc = MACHINE_GET_CLASS(sms);
3993     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3994 
3995     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3996         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3997             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3998         } else {
3999             /* NOTE: this means there is a window after guest reset, prior to
4000              * CAS negotiation, where unplug requests will fail due to the
4001              * capability not being detected yet. This is a bit different than
4002              * the case with PCI unplug, where the events will be queued and
4003              * eventually handled by the guest after boot
4004              */
4005             error_setg(errp, "Memory hot unplug not supported for this guest");
4006         }
4007     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4008         if (!mc->has_hotpluggable_cpus) {
4009             error_setg(errp, "CPU hot unplug not supported on this machine");
4010             return;
4011         }
4012         spapr_core_unplug_request(hotplug_dev, dev, errp);
4013     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4014         if (!smc->dr_phb_enabled) {
4015             error_setg(errp, "PHB hot unplug not supported on this machine");
4016             return;
4017         }
4018         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4019     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4020         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4021     }
4022 }
4023 
4024 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4025                                           DeviceState *dev, Error **errp)
4026 {
4027     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4028         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4029     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4030         spapr_core_pre_plug(hotplug_dev, dev, errp);
4031     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4032         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4033     }
4034 }
4035 
4036 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4037                                                  DeviceState *dev)
4038 {
4039     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4040         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4041         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4042         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4043         return HOTPLUG_HANDLER(machine);
4044     }
4045     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4046         PCIDevice *pcidev = PCI_DEVICE(dev);
4047         PCIBus *root = pci_device_root_bus(pcidev);
4048         SpaprPhbState *phb =
4049             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4050                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4051 
4052         if (phb) {
4053             return HOTPLUG_HANDLER(phb);
4054         }
4055     }
4056     return NULL;
4057 }
4058 
4059 static CpuInstanceProperties
4060 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4061 {
4062     CPUArchId *core_slot;
4063     MachineClass *mc = MACHINE_GET_CLASS(machine);
4064 
4065     /* make sure possible_cpu are intialized */
4066     mc->possible_cpu_arch_ids(machine);
4067     /* get CPU core slot containing thread that matches cpu_index */
4068     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4069     assert(core_slot);
4070     return core_slot->props;
4071 }
4072 
4073 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4074 {
4075     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4076 }
4077 
4078 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4079 {
4080     int i;
4081     unsigned int smp_threads = machine->smp.threads;
4082     unsigned int smp_cpus = machine->smp.cpus;
4083     const char *core_type;
4084     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4085     MachineClass *mc = MACHINE_GET_CLASS(machine);
4086 
4087     if (!mc->has_hotpluggable_cpus) {
4088         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4089     }
4090     if (machine->possible_cpus) {
4091         assert(machine->possible_cpus->len == spapr_max_cores);
4092         return machine->possible_cpus;
4093     }
4094 
4095     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4096     if (!core_type) {
4097         error_report("Unable to find sPAPR CPU Core definition");
4098         exit(1);
4099     }
4100 
4101     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4102                              sizeof(CPUArchId) * spapr_max_cores);
4103     machine->possible_cpus->len = spapr_max_cores;
4104     for (i = 0; i < machine->possible_cpus->len; i++) {
4105         int core_id = i * smp_threads;
4106 
4107         machine->possible_cpus->cpus[i].type = core_type;
4108         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4109         machine->possible_cpus->cpus[i].arch_id = core_id;
4110         machine->possible_cpus->cpus[i].props.has_core_id = true;
4111         machine->possible_cpus->cpus[i].props.core_id = core_id;
4112     }
4113     return machine->possible_cpus;
4114 }
4115 
4116 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4117                                 uint64_t *buid, hwaddr *pio,
4118                                 hwaddr *mmio32, hwaddr *mmio64,
4119                                 unsigned n_dma, uint32_t *liobns,
4120                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4121 {
4122     /*
4123      * New-style PHB window placement.
4124      *
4125      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4126      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4127      * windows.
4128      *
4129      * Some guest kernels can't work with MMIO windows above 1<<46
4130      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4131      *
4132      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4133      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4134      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4135      * 1TiB 64-bit MMIO windows for each PHB.
4136      */
4137     const uint64_t base_buid = 0x800000020000000ULL;
4138     int i;
4139 
4140     /* Sanity check natural alignments */
4141     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4142     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4143     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4144     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4145     /* Sanity check bounds */
4146     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4147                       SPAPR_PCI_MEM32_WIN_SIZE);
4148     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4149                       SPAPR_PCI_MEM64_WIN_SIZE);
4150 
4151     if (index >= SPAPR_MAX_PHBS) {
4152         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4153                    SPAPR_MAX_PHBS - 1);
4154         return;
4155     }
4156 
4157     *buid = base_buid + index;
4158     for (i = 0; i < n_dma; ++i) {
4159         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4160     }
4161 
4162     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4163     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4164     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4165 
4166     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4167     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4168 }
4169 
4170 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4171 {
4172     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4173 
4174     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4175 }
4176 
4177 static void spapr_ics_resend(XICSFabric *dev)
4178 {
4179     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4180 
4181     ics_resend(spapr->ics);
4182 }
4183 
4184 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4185 {
4186     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4187 
4188     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4189 }
4190 
4191 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4192                                  Monitor *mon)
4193 {
4194     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4195 
4196     spapr_irq_print_info(spapr, mon);
4197     monitor_printf(mon, "irqchip: %s\n",
4198                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4199 }
4200 
4201 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4202                            uint8_t nvt_blk, uint32_t nvt_idx,
4203                            bool cam_ignore, uint8_t priority,
4204                            uint32_t logic_serv, XiveTCTXMatch *match)
4205 {
4206     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4207     XivePresenter *xptr = XIVE_PRESENTER(spapr->xive);
4208     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4209     int count;
4210 
4211     /* This is a XIVE only operation */
4212     assert(spapr->active_intc == SPAPR_INTC(spapr->xive));
4213 
4214     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4215                            priority, logic_serv, match);
4216     if (count < 0) {
4217         return count;
4218     }
4219 
4220     /*
4221      * When we implement the save and restore of the thread interrupt
4222      * contexts in the enter/exit CPU handlers of the machine and the
4223      * escalations in QEMU, we should be able to handle non dispatched
4224      * vCPUs.
4225      *
4226      * Until this is done, the sPAPR machine should find at least one
4227      * matching context always.
4228      */
4229     if (count == 0) {
4230         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4231                       nvt_blk, nvt_idx);
4232     }
4233 
4234     return count;
4235 }
4236 
4237 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4238 {
4239     return cpu->vcpu_id;
4240 }
4241 
4242 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4243 {
4244     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4245     MachineState *ms = MACHINE(spapr);
4246     int vcpu_id;
4247 
4248     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4249 
4250     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4251         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4252         error_append_hint(errp, "Adjust the number of cpus to %d "
4253                           "or try to raise the number of threads per core\n",
4254                           vcpu_id * ms->smp.threads / spapr->vsmt);
4255         return;
4256     }
4257 
4258     cpu->vcpu_id = vcpu_id;
4259 }
4260 
4261 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4262 {
4263     CPUState *cs;
4264 
4265     CPU_FOREACH(cs) {
4266         PowerPCCPU *cpu = POWERPC_CPU(cs);
4267 
4268         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4269             return cpu;
4270         }
4271     }
4272 
4273     return NULL;
4274 }
4275 
4276 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4277 {
4278     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4279 
4280     /* These are only called by TCG, KVM maintains dispatch state */
4281 
4282     spapr_cpu->prod = false;
4283     if (spapr_cpu->vpa_addr) {
4284         CPUState *cs = CPU(cpu);
4285         uint32_t dispatch;
4286 
4287         dispatch = ldl_be_phys(cs->as,
4288                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4289         dispatch++;
4290         if ((dispatch & 1) != 0) {
4291             qemu_log_mask(LOG_GUEST_ERROR,
4292                           "VPA: incorrect dispatch counter value for "
4293                           "dispatched partition %u, correcting.\n", dispatch);
4294             dispatch++;
4295         }
4296         stl_be_phys(cs->as,
4297                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4298     }
4299 }
4300 
4301 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4302 {
4303     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4304 
4305     if (spapr_cpu->vpa_addr) {
4306         CPUState *cs = CPU(cpu);
4307         uint32_t dispatch;
4308 
4309         dispatch = ldl_be_phys(cs->as,
4310                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4311         dispatch++;
4312         if ((dispatch & 1) != 1) {
4313             qemu_log_mask(LOG_GUEST_ERROR,
4314                           "VPA: incorrect dispatch counter value for "
4315                           "preempted partition %u, correcting.\n", dispatch);
4316             dispatch++;
4317         }
4318         stl_be_phys(cs->as,
4319                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4320     }
4321 }
4322 
4323 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4324 {
4325     MachineClass *mc = MACHINE_CLASS(oc);
4326     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4327     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4328     NMIClass *nc = NMI_CLASS(oc);
4329     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4330     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4331     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4332     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4333     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4334 
4335     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4336     mc->ignore_boot_device_suffixes = true;
4337 
4338     /*
4339      * We set up the default / latest behaviour here.  The class_init
4340      * functions for the specific versioned machine types can override
4341      * these details for backwards compatibility
4342      */
4343     mc->init = spapr_machine_init;
4344     mc->reset = spapr_machine_reset;
4345     mc->block_default_type = IF_SCSI;
4346     mc->max_cpus = 1024;
4347     mc->no_parallel = 1;
4348     mc->default_boot_order = "";
4349     mc->default_ram_size = 512 * MiB;
4350     mc->default_display = "std";
4351     mc->kvm_type = spapr_kvm_type;
4352     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4353     mc->pci_allow_0_address = true;
4354     assert(!mc->get_hotplug_handler);
4355     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4356     hc->pre_plug = spapr_machine_device_pre_plug;
4357     hc->plug = spapr_machine_device_plug;
4358     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4359     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4360     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4361     hc->unplug_request = spapr_machine_device_unplug_request;
4362     hc->unplug = spapr_machine_device_unplug;
4363 
4364     smc->dr_lmb_enabled = true;
4365     smc->update_dt_enabled = true;
4366     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4367     mc->has_hotpluggable_cpus = true;
4368     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4369     fwc->get_dev_path = spapr_get_fw_dev_path;
4370     nc->nmi_monitor_handler = spapr_nmi;
4371     smc->phb_placement = spapr_phb_placement;
4372     vhc->hypercall = emulate_spapr_hypercall;
4373     vhc->hpt_mask = spapr_hpt_mask;
4374     vhc->map_hptes = spapr_map_hptes;
4375     vhc->unmap_hptes = spapr_unmap_hptes;
4376     vhc->hpte_set_c = spapr_hpte_set_c;
4377     vhc->hpte_set_r = spapr_hpte_set_r;
4378     vhc->get_pate = spapr_get_pate;
4379     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4380     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4381     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4382     xic->ics_get = spapr_ics_get;
4383     xic->ics_resend = spapr_ics_resend;
4384     xic->icp_get = spapr_icp_get;
4385     ispc->print_info = spapr_pic_print_info;
4386     /* Force NUMA node memory size to be a multiple of
4387      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4388      * in which LMBs are represented and hot-added
4389      */
4390     mc->numa_mem_align_shift = 28;
4391     mc->numa_mem_supported = true;
4392     mc->auto_enable_numa = true;
4393 
4394     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4395     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4396     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4397     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4398     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4399     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4400     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4401     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4402     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4403     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4404     spapr_caps_add_properties(smc, &error_abort);
4405     smc->irq = &spapr_irq_dual;
4406     smc->dr_phb_enabled = true;
4407     smc->linux_pci_probe = true;
4408     smc->smp_threads_vsmt = true;
4409     smc->nr_xirqs = SPAPR_NR_XIRQS;
4410     xfc->match_nvt = spapr_match_nvt;
4411 }
4412 
4413 static const TypeInfo spapr_machine_info = {
4414     .name          = TYPE_SPAPR_MACHINE,
4415     .parent        = TYPE_MACHINE,
4416     .abstract      = true,
4417     .instance_size = sizeof(SpaprMachineState),
4418     .instance_init = spapr_instance_init,
4419     .instance_finalize = spapr_machine_finalizefn,
4420     .class_size    = sizeof(SpaprMachineClass),
4421     .class_init    = spapr_machine_class_init,
4422     .interfaces = (InterfaceInfo[]) {
4423         { TYPE_FW_PATH_PROVIDER },
4424         { TYPE_NMI },
4425         { TYPE_HOTPLUG_HANDLER },
4426         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4427         { TYPE_XICS_FABRIC },
4428         { TYPE_INTERRUPT_STATS_PROVIDER },
4429         { TYPE_XIVE_FABRIC },
4430         { }
4431     },
4432 };
4433 
4434 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4435     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4436                                                     void *data)      \
4437     {                                                                \
4438         MachineClass *mc = MACHINE_CLASS(oc);                        \
4439         spapr_machine_##suffix##_class_options(mc);                  \
4440         if (latest) {                                                \
4441             mc->alias = "pseries";                                   \
4442             mc->is_default = 1;                                      \
4443         }                                                            \
4444     }                                                                \
4445     static const TypeInfo spapr_machine_##suffix##_info = {          \
4446         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4447         .parent = TYPE_SPAPR_MACHINE,                                \
4448         .class_init = spapr_machine_##suffix##_class_init,           \
4449     };                                                               \
4450     static void spapr_machine_register_##suffix(void)                \
4451     {                                                                \
4452         type_register(&spapr_machine_##suffix##_info);               \
4453     }                                                                \
4454     type_init(spapr_machine_register_##suffix)
4455 
4456 /*
4457  * pseries-5.0
4458  */
4459 static void spapr_machine_5_0_class_options(MachineClass *mc)
4460 {
4461     /* Defaults for the latest behaviour inherited from the base class */
4462 }
4463 
4464 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4465 
4466 /*
4467  * pseries-4.2
4468  */
4469 static void spapr_machine_4_2_class_options(MachineClass *mc)
4470 {
4471     spapr_machine_5_0_class_options(mc);
4472     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4473 }
4474 
4475 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4476 
4477 /*
4478  * pseries-4.1
4479  */
4480 static void spapr_machine_4_1_class_options(MachineClass *mc)
4481 {
4482     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4483     static GlobalProperty compat[] = {
4484         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4485         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4486     };
4487 
4488     spapr_machine_4_2_class_options(mc);
4489     smc->linux_pci_probe = false;
4490     smc->smp_threads_vsmt = false;
4491     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4492     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4493 }
4494 
4495 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4496 
4497 /*
4498  * pseries-4.0
4499  */
4500 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4501                               uint64_t *buid, hwaddr *pio,
4502                               hwaddr *mmio32, hwaddr *mmio64,
4503                               unsigned n_dma, uint32_t *liobns,
4504                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4505 {
4506     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4507                         nv2gpa, nv2atsd, errp);
4508     *nv2gpa = 0;
4509     *nv2atsd = 0;
4510 }
4511 
4512 static void spapr_machine_4_0_class_options(MachineClass *mc)
4513 {
4514     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4515 
4516     spapr_machine_4_1_class_options(mc);
4517     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4518     smc->phb_placement = phb_placement_4_0;
4519     smc->irq = &spapr_irq_xics;
4520     smc->pre_4_1_migration = true;
4521 }
4522 
4523 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4524 
4525 /*
4526  * pseries-3.1
4527  */
4528 static void spapr_machine_3_1_class_options(MachineClass *mc)
4529 {
4530     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4531 
4532     spapr_machine_4_0_class_options(mc);
4533     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4534 
4535     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4536     smc->update_dt_enabled = false;
4537     smc->dr_phb_enabled = false;
4538     smc->broken_host_serial_model = true;
4539     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4540     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4541     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4542     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4543 }
4544 
4545 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4546 
4547 /*
4548  * pseries-3.0
4549  */
4550 
4551 static void spapr_machine_3_0_class_options(MachineClass *mc)
4552 {
4553     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4554 
4555     spapr_machine_3_1_class_options(mc);
4556     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4557 
4558     smc->legacy_irq_allocation = true;
4559     smc->nr_xirqs = 0x400;
4560     smc->irq = &spapr_irq_xics_legacy;
4561 }
4562 
4563 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4564 
4565 /*
4566  * pseries-2.12
4567  */
4568 static void spapr_machine_2_12_class_options(MachineClass *mc)
4569 {
4570     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4571     static GlobalProperty compat[] = {
4572         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4573         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4574     };
4575 
4576     spapr_machine_3_0_class_options(mc);
4577     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4578     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4579 
4580     /* We depend on kvm_enabled() to choose a default value for the
4581      * hpt-max-page-size capability. Of course we can't do it here
4582      * because this is too early and the HW accelerator isn't initialzed
4583      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4584      */
4585     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4586 }
4587 
4588 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4589 
4590 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4591 {
4592     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4593 
4594     spapr_machine_2_12_class_options(mc);
4595     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4596     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4597     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4598 }
4599 
4600 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4601 
4602 /*
4603  * pseries-2.11
4604  */
4605 
4606 static void spapr_machine_2_11_class_options(MachineClass *mc)
4607 {
4608     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4609 
4610     spapr_machine_2_12_class_options(mc);
4611     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4612     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4613 }
4614 
4615 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4616 
4617 /*
4618  * pseries-2.10
4619  */
4620 
4621 static void spapr_machine_2_10_class_options(MachineClass *mc)
4622 {
4623     spapr_machine_2_11_class_options(mc);
4624     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4625 }
4626 
4627 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4628 
4629 /*
4630  * pseries-2.9
4631  */
4632 
4633 static void spapr_machine_2_9_class_options(MachineClass *mc)
4634 {
4635     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4636     static GlobalProperty compat[] = {
4637         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4638     };
4639 
4640     spapr_machine_2_10_class_options(mc);
4641     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4642     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4643     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4644     smc->pre_2_10_has_unused_icps = true;
4645     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4646 }
4647 
4648 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4649 
4650 /*
4651  * pseries-2.8
4652  */
4653 
4654 static void spapr_machine_2_8_class_options(MachineClass *mc)
4655 {
4656     static GlobalProperty compat[] = {
4657         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4658     };
4659 
4660     spapr_machine_2_9_class_options(mc);
4661     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4662     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4663     mc->numa_mem_align_shift = 23;
4664 }
4665 
4666 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4667 
4668 /*
4669  * pseries-2.7
4670  */
4671 
4672 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4673                               uint64_t *buid, hwaddr *pio,
4674                               hwaddr *mmio32, hwaddr *mmio64,
4675                               unsigned n_dma, uint32_t *liobns,
4676                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4677 {
4678     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4679     const uint64_t base_buid = 0x800000020000000ULL;
4680     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4681     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4682     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4683     const uint32_t max_index = 255;
4684     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4685 
4686     uint64_t ram_top = MACHINE(spapr)->ram_size;
4687     hwaddr phb0_base, phb_base;
4688     int i;
4689 
4690     /* Do we have device memory? */
4691     if (MACHINE(spapr)->maxram_size > ram_top) {
4692         /* Can't just use maxram_size, because there may be an
4693          * alignment gap between normal and device memory regions
4694          */
4695         ram_top = MACHINE(spapr)->device_memory->base +
4696             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4697     }
4698 
4699     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4700 
4701     if (index > max_index) {
4702         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4703                    max_index);
4704         return;
4705     }
4706 
4707     *buid = base_buid + index;
4708     for (i = 0; i < n_dma; ++i) {
4709         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4710     }
4711 
4712     phb_base = phb0_base + index * phb_spacing;
4713     *pio = phb_base + pio_offset;
4714     *mmio32 = phb_base + mmio_offset;
4715     /*
4716      * We don't set the 64-bit MMIO window, relying on the PHB's
4717      * fallback behaviour of automatically splitting a large "32-bit"
4718      * window into contiguous 32-bit and 64-bit windows
4719      */
4720 
4721     *nv2gpa = 0;
4722     *nv2atsd = 0;
4723 }
4724 
4725 static void spapr_machine_2_7_class_options(MachineClass *mc)
4726 {
4727     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4728     static GlobalProperty compat[] = {
4729         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4730         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4731         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4732         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4733     };
4734 
4735     spapr_machine_2_8_class_options(mc);
4736     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4737     mc->default_machine_opts = "modern-hotplug-events=off";
4738     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4739     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4740     smc->phb_placement = phb_placement_2_7;
4741 }
4742 
4743 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4744 
4745 /*
4746  * pseries-2.6
4747  */
4748 
4749 static void spapr_machine_2_6_class_options(MachineClass *mc)
4750 {
4751     static GlobalProperty compat[] = {
4752         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4753     };
4754 
4755     spapr_machine_2_7_class_options(mc);
4756     mc->has_hotpluggable_cpus = false;
4757     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4758     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4759 }
4760 
4761 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4762 
4763 /*
4764  * pseries-2.5
4765  */
4766 
4767 static void spapr_machine_2_5_class_options(MachineClass *mc)
4768 {
4769     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4770     static GlobalProperty compat[] = {
4771         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4772     };
4773 
4774     spapr_machine_2_6_class_options(mc);
4775     smc->use_ohci_by_default = true;
4776     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4777     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4778 }
4779 
4780 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4781 
4782 /*
4783  * pseries-2.4
4784  */
4785 
4786 static void spapr_machine_2_4_class_options(MachineClass *mc)
4787 {
4788     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4789 
4790     spapr_machine_2_5_class_options(mc);
4791     smc->dr_lmb_enabled = false;
4792     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4793 }
4794 
4795 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4796 
4797 /*
4798  * pseries-2.3
4799  */
4800 
4801 static void spapr_machine_2_3_class_options(MachineClass *mc)
4802 {
4803     static GlobalProperty compat[] = {
4804         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4805     };
4806     spapr_machine_2_4_class_options(mc);
4807     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4808     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4809 }
4810 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4811 
4812 /*
4813  * pseries-2.2
4814  */
4815 
4816 static void spapr_machine_2_2_class_options(MachineClass *mc)
4817 {
4818     static GlobalProperty compat[] = {
4819         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4820     };
4821 
4822     spapr_machine_2_3_class_options(mc);
4823     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4824     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4825     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4826 }
4827 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4828 
4829 /*
4830  * pseries-2.1
4831  */
4832 
4833 static void spapr_machine_2_1_class_options(MachineClass *mc)
4834 {
4835     spapr_machine_2_2_class_options(mc);
4836     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4837 }
4838 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4839 
4840 static void spapr_machine_register_types(void)
4841 {
4842     type_register_static(&spapr_machine_info);
4843 }
4844 
4845 type_init(spapr_machine_register_types)
4846