xref: /openbmc/qemu/hw/ppc/spapr.c (revision 19f70347)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 
84 #include "monitor/monitor.h"
85 
86 #include <libfdt.h>
87 
88 /* SLOF memory layout:
89  *
90  * SLOF raw image loaded at 0, copies its romfs right below the flat
91  * device-tree, then position SLOF itself 31M below that
92  *
93  * So we set FW_OVERHEAD to 40MB which should account for all of that
94  * and more
95  *
96  * We load our kernel at 4M, leaving space for SLOF initial image
97  */
98 #define FDT_MAX_SIZE            0x100000
99 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
100 #define FW_MAX_SIZE             0x400000
101 #define FW_FILE_NAME            "slof.bin"
102 #define FW_OVERHEAD             0x2800000
103 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
104 
105 #define MIN_RMA_SLOF            128UL
106 
107 #define PHANDLE_INTC            0x00001111
108 
109 /* These two functions implement the VCPU id numbering: one to compute them
110  * all and one to identify thread 0 of a VCORE. Any change to the first one
111  * is likely to have an impact on the second one, so let's keep them close.
112  */
113 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
114 {
115     MachineState *ms = MACHINE(spapr);
116     unsigned int smp_threads = ms->smp.threads;
117 
118     assert(spapr->vsmt);
119     return
120         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
121 }
122 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
123                                       PowerPCCPU *cpu)
124 {
125     assert(spapr->vsmt);
126     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 int spapr_max_server_number(SpaprMachineState *spapr)
164 {
165     MachineState *ms = MACHINE(spapr);
166 
167     assert(spapr->vsmt);
168     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
169 }
170 
171 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
172                                   int smt_threads)
173 {
174     int i, ret = 0;
175     uint32_t servers_prop[smt_threads];
176     uint32_t gservers_prop[smt_threads * 2];
177     int index = spapr_get_vcpu_id(cpu);
178 
179     if (cpu->compat_pvr) {
180         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
181         if (ret < 0) {
182             return ret;
183         }
184     }
185 
186     /* Build interrupt servers and gservers properties */
187     for (i = 0; i < smt_threads; i++) {
188         servers_prop[i] = cpu_to_be32(index + i);
189         /* Hack, direct the group queues back to cpu 0 */
190         gservers_prop[i*2] = cpu_to_be32(index + i);
191         gservers_prop[i*2 + 1] = 0;
192     }
193     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
194                       servers_prop, sizeof(servers_prop));
195     if (ret < 0) {
196         return ret;
197     }
198     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
199                       gservers_prop, sizeof(gservers_prop));
200 
201     return ret;
202 }
203 
204 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
205 {
206     int index = spapr_get_vcpu_id(cpu);
207     uint32_t associativity[] = {cpu_to_be32(0x5),
208                                 cpu_to_be32(0x0),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(cpu->node_id),
212                                 cpu_to_be32(index)};
213 
214     /* Advertise NUMA via ibm,associativity */
215     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
216                           sizeof(associativity));
217 }
218 
219 /* Populate the "ibm,pa-features" property */
220 static void spapr_populate_pa_features(SpaprMachineState *spapr,
221                                        PowerPCCPU *cpu,
222                                        void *fdt, int offset)
223 {
224     uint8_t pa_features_206[] = { 6, 0,
225         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
226     uint8_t pa_features_207[] = { 24, 0,
227         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
228         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
229         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
230         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
231     uint8_t pa_features_300[] = { 66, 0,
232         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
233         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
234         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
235         /* 6: DS207 */
236         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
237         /* 16: Vector */
238         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
239         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
241         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
242         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
243         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
244         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
245         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
246         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
247         /* 42: PM, 44: PC RA, 46: SC vec'd */
248         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
249         /* 48: SIMD, 50: QP BFP, 52: String */
250         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
251         /* 54: DecFP, 56: DecI, 58: SHA */
252         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
253         /* 60: NM atomic, 62: RNG */
254         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
255     };
256     uint8_t *pa_features = NULL;
257     size_t pa_size;
258 
259     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
260         pa_features = pa_features_206;
261         pa_size = sizeof(pa_features_206);
262     }
263     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
264         pa_features = pa_features_207;
265         pa_size = sizeof(pa_features_207);
266     }
267     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
268         pa_features = pa_features_300;
269         pa_size = sizeof(pa_features_300);
270     }
271     if (!pa_features) {
272         return;
273     }
274 
275     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
276         /*
277          * Note: we keep CI large pages off by default because a 64K capable
278          * guest provisioned with large pages might otherwise try to map a qemu
279          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
280          * even if that qemu runs on a 4k host.
281          * We dd this bit back here if we are confident this is not an issue
282          */
283         pa_features[3] |= 0x20;
284     }
285     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
286         pa_features[24] |= 0x80;    /* Transactional memory support */
287     }
288     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
289         /* Workaround for broken kernels that attempt (guest) radix
290          * mode when they can't handle it, if they see the radix bit set
291          * in pa-features. So hide it from them. */
292         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
293     }
294 
295     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
296 }
297 
298 static hwaddr spapr_node0_size(MachineState *machine)
299 {
300     if (machine->numa_state->num_nodes) {
301         int i;
302         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
303             if (machine->numa_state->nodes[i].node_mem) {
304                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
305                            machine->ram_size);
306             }
307         }
308     }
309     return machine->ram_size;
310 }
311 
312 static void add_str(GString *s, const gchar *s1)
313 {
314     g_string_append_len(s, s1, strlen(s1) + 1);
315 }
316 
317 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
318                                        hwaddr size)
319 {
320     uint32_t associativity[] = {
321         cpu_to_be32(0x4), /* length */
322         cpu_to_be32(0x0), cpu_to_be32(0x0),
323         cpu_to_be32(0x0), cpu_to_be32(nodeid)
324     };
325     char mem_name[32];
326     uint64_t mem_reg_property[2];
327     int off;
328 
329     mem_reg_property[0] = cpu_to_be64(start);
330     mem_reg_property[1] = cpu_to_be64(size);
331 
332     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
333     off = fdt_add_subnode(fdt, 0, mem_name);
334     _FDT(off);
335     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
336     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
337                       sizeof(mem_reg_property))));
338     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
339                       sizeof(associativity))));
340     return off;
341 }
342 
343 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
344 {
345     MachineState *machine = MACHINE(spapr);
346     hwaddr mem_start, node_size;
347     int i, nb_nodes = machine->numa_state->num_nodes;
348     NodeInfo *nodes = machine->numa_state->nodes;
349 
350     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
351         if (!nodes[i].node_mem) {
352             continue;
353         }
354         if (mem_start >= machine->ram_size) {
355             node_size = 0;
356         } else {
357             node_size = nodes[i].node_mem;
358             if (node_size > machine->ram_size - mem_start) {
359                 node_size = machine->ram_size - mem_start;
360             }
361         }
362         if (!mem_start) {
363             /* spapr_machine_init() checks for rma_size <= node0_size
364              * already */
365             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
366             mem_start += spapr->rma_size;
367             node_size -= spapr->rma_size;
368         }
369         for ( ; node_size; ) {
370             hwaddr sizetmp = pow2floor(node_size);
371 
372             /* mem_start != 0 here */
373             if (ctzl(mem_start) < ctzl(sizetmp)) {
374                 sizetmp = 1ULL << ctzl(mem_start);
375             }
376 
377             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
378             node_size -= sizetmp;
379             mem_start += sizetmp;
380         }
381     }
382 
383     return 0;
384 }
385 
386 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
387                                   SpaprMachineState *spapr)
388 {
389     MachineState *ms = MACHINE(spapr);
390     PowerPCCPU *cpu = POWERPC_CPU(cs);
391     CPUPPCState *env = &cpu->env;
392     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
393     int index = spapr_get_vcpu_id(cpu);
394     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
395                        0xffffffff, 0xffffffff};
396     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
397         : SPAPR_TIMEBASE_FREQ;
398     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
399     uint32_t page_sizes_prop[64];
400     size_t page_sizes_prop_size;
401     unsigned int smp_threads = ms->smp.threads;
402     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
403     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
404     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
405     SpaprDrc *drc;
406     int drc_index;
407     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
408     int i;
409 
410     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
411     if (drc) {
412         drc_index = spapr_drc_index(drc);
413         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
414     }
415 
416     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
417     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
418 
419     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
420     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
421                            env->dcache_line_size)));
422     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
423                            env->dcache_line_size)));
424     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
425                            env->icache_line_size)));
426     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
427                            env->icache_line_size)));
428 
429     if (pcc->l1_dcache_size) {
430         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
431                                pcc->l1_dcache_size)));
432     } else {
433         warn_report("Unknown L1 dcache size for cpu");
434     }
435     if (pcc->l1_icache_size) {
436         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
437                                pcc->l1_icache_size)));
438     } else {
439         warn_report("Unknown L1 icache size for cpu");
440     }
441 
442     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
443     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
444     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
445     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
446     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
447     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
448 
449     if (env->spr_cb[SPR_PURR].oea_read) {
450         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
451     }
452     if (env->spr_cb[SPR_SPURR].oea_read) {
453         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
454     }
455 
456     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
457         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
458                           segs, sizeof(segs))));
459     }
460 
461     /* Advertise VSX (vector extensions) if available
462      *   1               == VMX / Altivec available
463      *   2               == VSX available
464      *
465      * Only CPUs for which we create core types in spapr_cpu_core.c
466      * are possible, and all of those have VMX */
467     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
468         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
469     } else {
470         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
471     }
472 
473     /* Advertise DFP (Decimal Floating Point) if available
474      *   0 / no property == no DFP
475      *   1               == DFP available */
476     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
477         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
478     }
479 
480     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
481                                                       sizeof(page_sizes_prop));
482     if (page_sizes_prop_size) {
483         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
484                           page_sizes_prop, page_sizes_prop_size)));
485     }
486 
487     spapr_populate_pa_features(spapr, cpu, fdt, offset);
488 
489     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
490                            cs->cpu_index / vcpus_per_socket)));
491 
492     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
493                       pft_size_prop, sizeof(pft_size_prop))));
494 
495     if (ms->numa_state->num_nodes > 1) {
496         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
497     }
498 
499     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
500 
501     if (pcc->radix_page_info) {
502         for (i = 0; i < pcc->radix_page_info->count; i++) {
503             radix_AP_encodings[i] =
504                 cpu_to_be32(pcc->radix_page_info->entries[i]);
505         }
506         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
507                           radix_AP_encodings,
508                           pcc->radix_page_info->count *
509                           sizeof(radix_AP_encodings[0]))));
510     }
511 
512     /*
513      * We set this property to let the guest know that it can use the large
514      * decrementer and its width in bits.
515      */
516     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
517         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
518                               pcc->lrg_decr_bits)));
519 }
520 
521 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
522 {
523     CPUState **rev;
524     CPUState *cs;
525     int n_cpus;
526     int cpus_offset;
527     char *nodename;
528     int i;
529 
530     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
531     _FDT(cpus_offset);
532     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
533     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
534 
535     /*
536      * We walk the CPUs in reverse order to ensure that CPU DT nodes
537      * created by fdt_add_subnode() end up in the right order in FDT
538      * for the guest kernel the enumerate the CPUs correctly.
539      *
540      * The CPU list cannot be traversed in reverse order, so we need
541      * to do extra work.
542      */
543     n_cpus = 0;
544     rev = NULL;
545     CPU_FOREACH(cs) {
546         rev = g_renew(CPUState *, rev, n_cpus + 1);
547         rev[n_cpus++] = cs;
548     }
549 
550     for (i = n_cpus - 1; i >= 0; i--) {
551         CPUState *cs = rev[i];
552         PowerPCCPU *cpu = POWERPC_CPU(cs);
553         int index = spapr_get_vcpu_id(cpu);
554         DeviceClass *dc = DEVICE_GET_CLASS(cs);
555         int offset;
556 
557         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
558             continue;
559         }
560 
561         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
562         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
563         g_free(nodename);
564         _FDT(offset);
565         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
566     }
567 
568     g_free(rev);
569 }
570 
571 static int spapr_rng_populate_dt(void *fdt)
572 {
573     int node;
574     int ret;
575 
576     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
577     if (node <= 0) {
578         return -1;
579     }
580     ret = fdt_setprop_string(fdt, node, "device_type",
581                              "ibm,platform-facilities");
582     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
583     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
584 
585     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
586     if (node <= 0) {
587         return -1;
588     }
589     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
590 
591     return ret ? -1 : 0;
592 }
593 
594 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
595 {
596     MemoryDeviceInfoList *info;
597 
598     for (info = list; info; info = info->next) {
599         MemoryDeviceInfo *value = info->value;
600 
601         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
602             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
603 
604             if (addr >= pcdimm_info->addr &&
605                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
606                 return pcdimm_info->node;
607             }
608         }
609     }
610 
611     return -1;
612 }
613 
614 struct sPAPRDrconfCellV2 {
615      uint32_t seq_lmbs;
616      uint64_t base_addr;
617      uint32_t drc_index;
618      uint32_t aa_index;
619      uint32_t flags;
620 } QEMU_PACKED;
621 
622 typedef struct DrconfCellQueue {
623     struct sPAPRDrconfCellV2 cell;
624     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
625 } DrconfCellQueue;
626 
627 static DrconfCellQueue *
628 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
629                       uint32_t drc_index, uint32_t aa_index,
630                       uint32_t flags)
631 {
632     DrconfCellQueue *elem;
633 
634     elem = g_malloc0(sizeof(*elem));
635     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
636     elem->cell.base_addr = cpu_to_be64(base_addr);
637     elem->cell.drc_index = cpu_to_be32(drc_index);
638     elem->cell.aa_index = cpu_to_be32(aa_index);
639     elem->cell.flags = cpu_to_be32(flags);
640 
641     return elem;
642 }
643 
644 /* ibm,dynamic-memory-v2 */
645 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
646                                    int offset, MemoryDeviceInfoList *dimms)
647 {
648     MachineState *machine = MACHINE(spapr);
649     uint8_t *int_buf, *cur_index;
650     int ret;
651     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
652     uint64_t addr, cur_addr, size;
653     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
654     uint64_t mem_end = machine->device_memory->base +
655                        memory_region_size(&machine->device_memory->mr);
656     uint32_t node, buf_len, nr_entries = 0;
657     SpaprDrc *drc;
658     DrconfCellQueue *elem, *next;
659     MemoryDeviceInfoList *info;
660     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
661         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
662 
663     /* Entry to cover RAM and the gap area */
664     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
665                                  SPAPR_LMB_FLAGS_RESERVED |
666                                  SPAPR_LMB_FLAGS_DRC_INVALID);
667     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
668     nr_entries++;
669 
670     cur_addr = machine->device_memory->base;
671     for (info = dimms; info; info = info->next) {
672         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
673 
674         addr = di->addr;
675         size = di->size;
676         node = di->node;
677 
678         /* Entry for hot-pluggable area */
679         if (cur_addr < addr) {
680             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
681             g_assert(drc);
682             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
683                                          cur_addr, spapr_drc_index(drc), -1, 0);
684             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
685             nr_entries++;
686         }
687 
688         /* Entry for DIMM */
689         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
690         g_assert(drc);
691         elem = spapr_get_drconf_cell(size / lmb_size, addr,
692                                      spapr_drc_index(drc), node,
693                                      SPAPR_LMB_FLAGS_ASSIGNED);
694         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
695         nr_entries++;
696         cur_addr = addr + size;
697     }
698 
699     /* Entry for remaining hotpluggable area */
700     if (cur_addr < mem_end) {
701         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
702         g_assert(drc);
703         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
704                                      cur_addr, spapr_drc_index(drc), -1, 0);
705         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
706         nr_entries++;
707     }
708 
709     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
710     int_buf = cur_index = g_malloc0(buf_len);
711     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
712     cur_index += sizeof(nr_entries);
713 
714     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
715         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
716         cur_index += sizeof(elem->cell);
717         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
718         g_free(elem);
719     }
720 
721     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
722     g_free(int_buf);
723     if (ret < 0) {
724         return -1;
725     }
726     return 0;
727 }
728 
729 /* ibm,dynamic-memory */
730 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
731                                    int offset, MemoryDeviceInfoList *dimms)
732 {
733     MachineState *machine = MACHINE(spapr);
734     int i, ret;
735     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
736     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
737     uint32_t nr_lmbs = (machine->device_memory->base +
738                        memory_region_size(&machine->device_memory->mr)) /
739                        lmb_size;
740     uint32_t *int_buf, *cur_index, buf_len;
741 
742     /*
743      * Allocate enough buffer size to fit in ibm,dynamic-memory
744      */
745     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
746     cur_index = int_buf = g_malloc0(buf_len);
747     int_buf[0] = cpu_to_be32(nr_lmbs);
748     cur_index++;
749     for (i = 0; i < nr_lmbs; i++) {
750         uint64_t addr = i * lmb_size;
751         uint32_t *dynamic_memory = cur_index;
752 
753         if (i >= device_lmb_start) {
754             SpaprDrc *drc;
755 
756             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
757             g_assert(drc);
758 
759             dynamic_memory[0] = cpu_to_be32(addr >> 32);
760             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
761             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
762             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
763             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
764             if (memory_region_present(get_system_memory(), addr)) {
765                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
766             } else {
767                 dynamic_memory[5] = cpu_to_be32(0);
768             }
769         } else {
770             /*
771              * LMB information for RMA, boot time RAM and gap b/n RAM and
772              * device memory region -- all these are marked as reserved
773              * and as having no valid DRC.
774              */
775             dynamic_memory[0] = cpu_to_be32(addr >> 32);
776             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
777             dynamic_memory[2] = cpu_to_be32(0);
778             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
779             dynamic_memory[4] = cpu_to_be32(-1);
780             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
781                                             SPAPR_LMB_FLAGS_DRC_INVALID);
782         }
783 
784         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
785     }
786     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
787     g_free(int_buf);
788     if (ret < 0) {
789         return -1;
790     }
791     return 0;
792 }
793 
794 /*
795  * Adds ibm,dynamic-reconfiguration-memory node.
796  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
797  * of this device tree node.
798  */
799 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
800 {
801     MachineState *machine = MACHINE(spapr);
802     int nb_numa_nodes = machine->numa_state->num_nodes;
803     int ret, i, offset;
804     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
805     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
806     uint32_t *int_buf, *cur_index, buf_len;
807     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
808     MemoryDeviceInfoList *dimms = NULL;
809 
810     /*
811      * Don't create the node if there is no device memory
812      */
813     if (machine->ram_size == machine->maxram_size) {
814         return 0;
815     }
816 
817     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
818 
819     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
820                     sizeof(prop_lmb_size));
821     if (ret < 0) {
822         return ret;
823     }
824 
825     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
826     if (ret < 0) {
827         return ret;
828     }
829 
830     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
831     if (ret < 0) {
832         return ret;
833     }
834 
835     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
836     dimms = qmp_memory_device_list();
837     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
838         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
839     } else {
840         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
841     }
842     qapi_free_MemoryDeviceInfoList(dimms);
843 
844     if (ret < 0) {
845         return ret;
846     }
847 
848     /* ibm,associativity-lookup-arrays */
849     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
850     cur_index = int_buf = g_malloc0(buf_len);
851     int_buf[0] = cpu_to_be32(nr_nodes);
852     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
853     cur_index += 2;
854     for (i = 0; i < nr_nodes; i++) {
855         uint32_t associativity[] = {
856             cpu_to_be32(0x0),
857             cpu_to_be32(0x0),
858             cpu_to_be32(0x0),
859             cpu_to_be32(i)
860         };
861         memcpy(cur_index, associativity, sizeof(associativity));
862         cur_index += 4;
863     }
864     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
865             (cur_index - int_buf) * sizeof(uint32_t));
866     g_free(int_buf);
867 
868     return ret;
869 }
870 
871 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
872                                 SpaprOptionVector *ov5_updates)
873 {
874     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
875     int ret = 0, offset;
876 
877     /* Generate ibm,dynamic-reconfiguration-memory node if required */
878     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
879         g_assert(smc->dr_lmb_enabled);
880         ret = spapr_populate_drconf_memory(spapr, fdt);
881         if (ret) {
882             return ret;
883         }
884     }
885 
886     offset = fdt_path_offset(fdt, "/chosen");
887     if (offset < 0) {
888         offset = fdt_add_subnode(fdt, 0, "chosen");
889         if (offset < 0) {
890             return offset;
891         }
892     }
893     return spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
894                                   "ibm,architecture-vec-5");
895 }
896 
897 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
898 {
899     MachineState *ms = MACHINE(spapr);
900     int rtas;
901     GString *hypertas = g_string_sized_new(256);
902     GString *qemu_hypertas = g_string_sized_new(256);
903     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
904     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
905         memory_region_size(&MACHINE(spapr)->device_memory->mr);
906     uint32_t lrdr_capacity[] = {
907         cpu_to_be32(max_device_addr >> 32),
908         cpu_to_be32(max_device_addr & 0xffffffff),
909         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
910         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
911     };
912     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
913     uint32_t maxdomains[] = {
914         cpu_to_be32(4),
915         maxdomain,
916         maxdomain,
917         maxdomain,
918         cpu_to_be32(spapr->gpu_numa_id),
919     };
920 
921     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
922 
923     /* hypertas */
924     add_str(hypertas, "hcall-pft");
925     add_str(hypertas, "hcall-term");
926     add_str(hypertas, "hcall-dabr");
927     add_str(hypertas, "hcall-interrupt");
928     add_str(hypertas, "hcall-tce");
929     add_str(hypertas, "hcall-vio");
930     add_str(hypertas, "hcall-splpar");
931     add_str(hypertas, "hcall-join");
932     add_str(hypertas, "hcall-bulk");
933     add_str(hypertas, "hcall-set-mode");
934     add_str(hypertas, "hcall-sprg0");
935     add_str(hypertas, "hcall-copy");
936     add_str(hypertas, "hcall-debug");
937     add_str(hypertas, "hcall-vphn");
938     add_str(qemu_hypertas, "hcall-memop1");
939 
940     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
941         add_str(hypertas, "hcall-multi-tce");
942     }
943 
944     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
945         add_str(hypertas, "hcall-hpt-resize");
946     }
947 
948     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
949                      hypertas->str, hypertas->len));
950     g_string_free(hypertas, TRUE);
951     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
952                      qemu_hypertas->str, qemu_hypertas->len));
953     g_string_free(qemu_hypertas, TRUE);
954 
955     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
956                      refpoints, sizeof(refpoints)));
957 
958     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
959                      maxdomains, sizeof(maxdomains)));
960 
961     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
962                           RTAS_ERROR_LOG_MAX));
963     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
964                           RTAS_EVENT_SCAN_RATE));
965 
966     g_assert(msi_nonbroken);
967     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
968 
969     /*
970      * According to PAPR, rtas ibm,os-term does not guarantee a return
971      * back to the guest cpu.
972      *
973      * While an additional ibm,extended-os-term property indicates
974      * that rtas call return will always occur. Set this property.
975      */
976     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
977 
978     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
979                      lrdr_capacity, sizeof(lrdr_capacity)));
980 
981     spapr_dt_rtas_tokens(fdt, rtas);
982 }
983 
984 /*
985  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
986  * and the XIVE features that the guest may request and thus the valid
987  * values for bytes 23..26 of option vector 5:
988  */
989 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
990                                           int chosen)
991 {
992     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
993 
994     char val[2 * 4] = {
995         23, 0x00, /* XICS / XIVE mode */
996         24, 0x00, /* Hash/Radix, filled in below. */
997         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
998         26, 0x40, /* Radix options: GTSE == yes. */
999     };
1000 
1001     if (spapr->irq->xics && spapr->irq->xive) {
1002         val[1] = SPAPR_OV5_XIVE_BOTH;
1003     } else if (spapr->irq->xive) {
1004         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1005     } else {
1006         assert(spapr->irq->xics);
1007         val[1] = SPAPR_OV5_XIVE_LEGACY;
1008     }
1009 
1010     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1011                           first_ppc_cpu->compat_pvr)) {
1012         /*
1013          * If we're in a pre POWER9 compat mode then the guest should
1014          * do hash and use the legacy interrupt mode
1015          */
1016         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1017         val[3] = 0x00; /* Hash */
1018     } else if (kvm_enabled()) {
1019         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1020             val[3] = 0x80; /* OV5_MMU_BOTH */
1021         } else if (kvmppc_has_cap_mmu_radix()) {
1022             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1023         } else {
1024             val[3] = 0x00; /* Hash */
1025         }
1026     } else {
1027         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1028         val[3] = 0xC0;
1029     }
1030     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1031                      val, sizeof(val)));
1032 }
1033 
1034 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1035 {
1036     MachineState *machine = MACHINE(spapr);
1037     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1038     int chosen;
1039     const char *boot_device = machine->boot_order;
1040     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1041     size_t cb = 0;
1042     char *bootlist = get_boot_devices_list(&cb);
1043 
1044     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1045 
1046     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1047         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1048                                 machine->kernel_cmdline));
1049     }
1050     if (spapr->initrd_size) {
1051         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1052                               spapr->initrd_base));
1053         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1054                               spapr->initrd_base + spapr->initrd_size));
1055     }
1056 
1057     if (spapr->kernel_size) {
1058         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1059                               cpu_to_be64(spapr->kernel_size) };
1060 
1061         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1062                          &kprop, sizeof(kprop)));
1063         if (spapr->kernel_le) {
1064             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1065         }
1066     }
1067     if (boot_menu) {
1068         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1069     }
1070     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1071     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1072     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1073 
1074     if (cb && bootlist) {
1075         int i;
1076 
1077         for (i = 0; i < cb; i++) {
1078             if (bootlist[i] == '\n') {
1079                 bootlist[i] = ' ';
1080             }
1081         }
1082         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1083     }
1084 
1085     if (boot_device && strlen(boot_device)) {
1086         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1087     }
1088 
1089     if (!spapr->has_graphics && stdout_path) {
1090         /*
1091          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1092          * kernel. New platforms should only use the "stdout-path" property. Set
1093          * the new property and continue using older property to remain
1094          * compatible with the existing firmware.
1095          */
1096         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1097         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1098     }
1099 
1100     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1101     if (smc->linux_pci_probe) {
1102         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1103     }
1104 
1105     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1106 
1107     g_free(stdout_path);
1108     g_free(bootlist);
1109 }
1110 
1111 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1112 {
1113     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1114      * KVM to work under pHyp with some guest co-operation */
1115     int hypervisor;
1116     uint8_t hypercall[16];
1117 
1118     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1119     /* indicate KVM hypercall interface */
1120     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1121     if (kvmppc_has_cap_fixup_hcalls()) {
1122         /*
1123          * Older KVM versions with older guest kernels were broken
1124          * with the magic page, don't allow the guest to map it.
1125          */
1126         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1127                                   sizeof(hypercall))) {
1128             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1129                              hypercall, sizeof(hypercall)));
1130         }
1131     }
1132 }
1133 
1134 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1135 {
1136     MachineState *machine = MACHINE(spapr);
1137     MachineClass *mc = MACHINE_GET_CLASS(machine);
1138     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1139     int ret;
1140     void *fdt;
1141     SpaprPhbState *phb;
1142     char *buf;
1143 
1144     fdt = g_malloc0(space);
1145     _FDT((fdt_create_empty_tree(fdt, space)));
1146 
1147     /* Root node */
1148     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1149     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1150     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1151 
1152     /* Guest UUID & Name*/
1153     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1154     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1155     if (qemu_uuid_set) {
1156         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1157     }
1158     g_free(buf);
1159 
1160     if (qemu_get_vm_name()) {
1161         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1162                                 qemu_get_vm_name()));
1163     }
1164 
1165     /* Host Model & Serial Number */
1166     if (spapr->host_model) {
1167         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1168     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1169         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1170         g_free(buf);
1171     }
1172 
1173     if (spapr->host_serial) {
1174         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1175     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1176         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1177         g_free(buf);
1178     }
1179 
1180     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1181     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1182 
1183     /* /interrupt controller */
1184     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1185 
1186     ret = spapr_populate_memory(spapr, fdt);
1187     if (ret < 0) {
1188         error_report("couldn't setup memory nodes in fdt");
1189         exit(1);
1190     }
1191 
1192     /* /vdevice */
1193     spapr_dt_vdevice(spapr->vio_bus, fdt);
1194 
1195     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1196         ret = spapr_rng_populate_dt(fdt);
1197         if (ret < 0) {
1198             error_report("could not set up rng device in the fdt");
1199             exit(1);
1200         }
1201     }
1202 
1203     QLIST_FOREACH(phb, &spapr->phbs, list) {
1204         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1205         if (ret < 0) {
1206             error_report("couldn't setup PCI devices in fdt");
1207             exit(1);
1208         }
1209     }
1210 
1211     /* cpus */
1212     spapr_populate_cpus_dt_node(fdt, spapr);
1213 
1214     if (smc->dr_lmb_enabled) {
1215         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1216     }
1217 
1218     if (mc->has_hotpluggable_cpus) {
1219         int offset = fdt_path_offset(fdt, "/cpus");
1220         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1221         if (ret < 0) {
1222             error_report("Couldn't set up CPU DR device tree properties");
1223             exit(1);
1224         }
1225     }
1226 
1227     /* /event-sources */
1228     spapr_dt_events(spapr, fdt);
1229 
1230     /* /rtas */
1231     spapr_dt_rtas(spapr, fdt);
1232 
1233     /* /chosen */
1234     if (reset) {
1235         spapr_dt_chosen(spapr, fdt);
1236     }
1237 
1238     /* /hypervisor */
1239     if (kvm_enabled()) {
1240         spapr_dt_hypervisor(spapr, fdt);
1241     }
1242 
1243     /* Build memory reserve map */
1244     if (reset) {
1245         if (spapr->kernel_size) {
1246             _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1247         }
1248         if (spapr->initrd_size) {
1249             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1250                                   spapr->initrd_size)));
1251         }
1252     }
1253 
1254     /* ibm,client-architecture-support updates */
1255     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1256     if (ret < 0) {
1257         error_report("couldn't setup CAS properties fdt");
1258         exit(1);
1259     }
1260 
1261     if (smc->dr_phb_enabled) {
1262         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1263         if (ret < 0) {
1264             error_report("Couldn't set up PHB DR device tree properties");
1265             exit(1);
1266         }
1267     }
1268 
1269     return fdt;
1270 }
1271 
1272 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1273 {
1274     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1275 }
1276 
1277 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1278                                     PowerPCCPU *cpu)
1279 {
1280     CPUPPCState *env = &cpu->env;
1281 
1282     /* The TCG path should also be holding the BQL at this point */
1283     g_assert(qemu_mutex_iothread_locked());
1284 
1285     if (msr_pr) {
1286         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1287         env->gpr[3] = H_PRIVILEGE;
1288     } else {
1289         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1290     }
1291 }
1292 
1293 struct LPCRSyncState {
1294     target_ulong value;
1295     target_ulong mask;
1296 };
1297 
1298 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1299 {
1300     struct LPCRSyncState *s = arg.host_ptr;
1301     PowerPCCPU *cpu = POWERPC_CPU(cs);
1302     CPUPPCState *env = &cpu->env;
1303     target_ulong lpcr;
1304 
1305     cpu_synchronize_state(cs);
1306     lpcr = env->spr[SPR_LPCR];
1307     lpcr &= ~s->mask;
1308     lpcr |= s->value;
1309     ppc_store_lpcr(cpu, lpcr);
1310 }
1311 
1312 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1313 {
1314     CPUState *cs;
1315     struct LPCRSyncState s = {
1316         .value = value,
1317         .mask = mask
1318     };
1319     CPU_FOREACH(cs) {
1320         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1321     }
1322 }
1323 
1324 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1325 {
1326     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1327 
1328     /* Copy PATE1:GR into PATE0:HR */
1329     entry->dw0 = spapr->patb_entry & PATE0_HR;
1330     entry->dw1 = spapr->patb_entry;
1331 }
1332 
1333 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1334 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1335 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1336 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1337 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1338 
1339 /*
1340  * Get the fd to access the kernel htab, re-opening it if necessary
1341  */
1342 static int get_htab_fd(SpaprMachineState *spapr)
1343 {
1344     Error *local_err = NULL;
1345 
1346     if (spapr->htab_fd >= 0) {
1347         return spapr->htab_fd;
1348     }
1349 
1350     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1351     if (spapr->htab_fd < 0) {
1352         error_report_err(local_err);
1353     }
1354 
1355     return spapr->htab_fd;
1356 }
1357 
1358 void close_htab_fd(SpaprMachineState *spapr)
1359 {
1360     if (spapr->htab_fd >= 0) {
1361         close(spapr->htab_fd);
1362     }
1363     spapr->htab_fd = -1;
1364 }
1365 
1366 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1367 {
1368     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1369 
1370     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1371 }
1372 
1373 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1374 {
1375     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1376 
1377     assert(kvm_enabled());
1378 
1379     if (!spapr->htab) {
1380         return 0;
1381     }
1382 
1383     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1384 }
1385 
1386 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1387                                                 hwaddr ptex, int n)
1388 {
1389     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1390     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1391 
1392     if (!spapr->htab) {
1393         /*
1394          * HTAB is controlled by KVM. Fetch into temporary buffer
1395          */
1396         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1397         kvmppc_read_hptes(hptes, ptex, n);
1398         return hptes;
1399     }
1400 
1401     /*
1402      * HTAB is controlled by QEMU. Just point to the internally
1403      * accessible PTEG.
1404      */
1405     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1406 }
1407 
1408 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1409                               const ppc_hash_pte64_t *hptes,
1410                               hwaddr ptex, int n)
1411 {
1412     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1413 
1414     if (!spapr->htab) {
1415         g_free((void *)hptes);
1416     }
1417 
1418     /* Nothing to do for qemu managed HPT */
1419 }
1420 
1421 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1422                       uint64_t pte0, uint64_t pte1)
1423 {
1424     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1425     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1426 
1427     if (!spapr->htab) {
1428         kvmppc_write_hpte(ptex, pte0, pte1);
1429     } else {
1430         if (pte0 & HPTE64_V_VALID) {
1431             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1432             /*
1433              * When setting valid, we write PTE1 first. This ensures
1434              * proper synchronization with the reading code in
1435              * ppc_hash64_pteg_search()
1436              */
1437             smp_wmb();
1438             stq_p(spapr->htab + offset, pte0);
1439         } else {
1440             stq_p(spapr->htab + offset, pte0);
1441             /*
1442              * When clearing it we set PTE0 first. This ensures proper
1443              * synchronization with the reading code in
1444              * ppc_hash64_pteg_search()
1445              */
1446             smp_wmb();
1447             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1448         }
1449     }
1450 }
1451 
1452 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1453                              uint64_t pte1)
1454 {
1455     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1456     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1457 
1458     if (!spapr->htab) {
1459         /* There should always be a hash table when this is called */
1460         error_report("spapr_hpte_set_c called with no hash table !");
1461         return;
1462     }
1463 
1464     /* The HW performs a non-atomic byte update */
1465     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1466 }
1467 
1468 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1469                              uint64_t pte1)
1470 {
1471     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1472     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1473 
1474     if (!spapr->htab) {
1475         /* There should always be a hash table when this is called */
1476         error_report("spapr_hpte_set_r called with no hash table !");
1477         return;
1478     }
1479 
1480     /* The HW performs a non-atomic byte update */
1481     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1482 }
1483 
1484 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1485 {
1486     int shift;
1487 
1488     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1489      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1490      * that's much more than is needed for Linux guests */
1491     shift = ctz64(pow2ceil(ramsize)) - 7;
1492     shift = MAX(shift, 18); /* Minimum architected size */
1493     shift = MIN(shift, 46); /* Maximum architected size */
1494     return shift;
1495 }
1496 
1497 void spapr_free_hpt(SpaprMachineState *spapr)
1498 {
1499     g_free(spapr->htab);
1500     spapr->htab = NULL;
1501     spapr->htab_shift = 0;
1502     close_htab_fd(spapr);
1503 }
1504 
1505 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1506                           Error **errp)
1507 {
1508     long rc;
1509 
1510     /* Clean up any HPT info from a previous boot */
1511     spapr_free_hpt(spapr);
1512 
1513     rc = kvmppc_reset_htab(shift);
1514     if (rc < 0) {
1515         /* kernel-side HPT needed, but couldn't allocate one */
1516         error_setg_errno(errp, errno,
1517                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1518                          shift);
1519         /* This is almost certainly fatal, but if the caller really
1520          * wants to carry on with shift == 0, it's welcome to try */
1521     } else if (rc > 0) {
1522         /* kernel-side HPT allocated */
1523         if (rc != shift) {
1524             error_setg(errp,
1525                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1526                        shift, rc);
1527         }
1528 
1529         spapr->htab_shift = shift;
1530         spapr->htab = NULL;
1531     } else {
1532         /* kernel-side HPT not needed, allocate in userspace instead */
1533         size_t size = 1ULL << shift;
1534         int i;
1535 
1536         spapr->htab = qemu_memalign(size, size);
1537         if (!spapr->htab) {
1538             error_setg_errno(errp, errno,
1539                              "Could not allocate HPT of order %d", shift);
1540             return;
1541         }
1542 
1543         memset(spapr->htab, 0, size);
1544         spapr->htab_shift = shift;
1545 
1546         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1547             DIRTY_HPTE(HPTE(spapr->htab, i));
1548         }
1549     }
1550     /* We're setting up a hash table, so that means we're not radix */
1551     spapr->patb_entry = 0;
1552     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1553 }
1554 
1555 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1556 {
1557     int hpt_shift;
1558 
1559     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1560         || (spapr->cas_reboot
1561             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1562         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1563     } else {
1564         uint64_t current_ram_size;
1565 
1566         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1567         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1568     }
1569     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1570 
1571     if (spapr->vrma_adjust) {
1572         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1573                                           spapr->htab_shift);
1574     }
1575 }
1576 
1577 static int spapr_reset_drcs(Object *child, void *opaque)
1578 {
1579     SpaprDrc *drc =
1580         (SpaprDrc *) object_dynamic_cast(child,
1581                                                  TYPE_SPAPR_DR_CONNECTOR);
1582 
1583     if (drc) {
1584         spapr_drc_reset(drc);
1585     }
1586 
1587     return 0;
1588 }
1589 
1590 static void spapr_machine_reset(MachineState *machine)
1591 {
1592     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1593     PowerPCCPU *first_ppc_cpu;
1594     hwaddr fdt_addr;
1595     void *fdt;
1596     int rc;
1597 
1598     kvmppc_svm_off(&error_fatal);
1599     spapr_caps_apply(spapr);
1600 
1601     first_ppc_cpu = POWERPC_CPU(first_cpu);
1602     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1603         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1604                               spapr->max_compat_pvr)) {
1605         /*
1606          * If using KVM with radix mode available, VCPUs can be started
1607          * without a HPT because KVM will start them in radix mode.
1608          * Set the GR bit in PATE so that we know there is no HPT.
1609          */
1610         spapr->patb_entry = PATE1_GR;
1611         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1612     } else {
1613         spapr_setup_hpt_and_vrma(spapr);
1614     }
1615 
1616     qemu_devices_reset();
1617 
1618     /*
1619      * If this reset wasn't generated by CAS, we should reset our
1620      * negotiated options and start from scratch
1621      */
1622     if (!spapr->cas_reboot) {
1623         spapr_ovec_cleanup(spapr->ov5_cas);
1624         spapr->ov5_cas = spapr_ovec_new();
1625 
1626         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1627     }
1628 
1629     /*
1630      * This is fixing some of the default configuration of the XIVE
1631      * devices. To be called after the reset of the machine devices.
1632      */
1633     spapr_irq_reset(spapr, &error_fatal);
1634 
1635     /*
1636      * There is no CAS under qtest. Simulate one to please the code that
1637      * depends on spapr->ov5_cas. This is especially needed to test device
1638      * unplug, so we do that before resetting the DRCs.
1639      */
1640     if (qtest_enabled()) {
1641         spapr_ovec_cleanup(spapr->ov5_cas);
1642         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1643     }
1644 
1645     /* DRC reset may cause a device to be unplugged. This will cause troubles
1646      * if this device is used by another device (eg, a running vhost backend
1647      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1648      * situations, we reset DRCs after all devices have been reset.
1649      */
1650     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1651 
1652     spapr_clear_pending_events(spapr);
1653 
1654     /*
1655      * We place the device tree and RTAS just below either the top of the RMA,
1656      * or just below 2GB, whichever is lower, so that it can be
1657      * processed with 32-bit real mode code if necessary
1658      */
1659     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1660 
1661     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1662 
1663     rc = fdt_pack(fdt);
1664 
1665     /* Should only fail if we've built a corrupted tree */
1666     assert(rc == 0);
1667 
1668     /* Load the fdt */
1669     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1670     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1671     g_free(spapr->fdt_blob);
1672     spapr->fdt_size = fdt_totalsize(fdt);
1673     spapr->fdt_initial_size = spapr->fdt_size;
1674     spapr->fdt_blob = fdt;
1675 
1676     /* Set up the entry state */
1677     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1678     first_ppc_cpu->env.gpr[5] = 0;
1679 
1680     spapr->cas_reboot = false;
1681 
1682     spapr->mc_status = -1;
1683     spapr->guest_machine_check_addr = -1;
1684 
1685     /* Signal all vCPUs waiting on this condition */
1686     qemu_cond_broadcast(&spapr->mc_delivery_cond);
1687 
1688     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1689 }
1690 
1691 static void spapr_create_nvram(SpaprMachineState *spapr)
1692 {
1693     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1694     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1695 
1696     if (dinfo) {
1697         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1698                             &error_fatal);
1699     }
1700 
1701     qdev_init_nofail(dev);
1702 
1703     spapr->nvram = (struct SpaprNvram *)dev;
1704 }
1705 
1706 static void spapr_rtc_create(SpaprMachineState *spapr)
1707 {
1708     object_initialize_child(OBJECT(spapr), "rtc",
1709                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1710                             &error_fatal, NULL);
1711     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1712                               &error_fatal);
1713     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1714                               "date", &error_fatal);
1715 }
1716 
1717 /* Returns whether we want to use VGA or not */
1718 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1719 {
1720     switch (vga_interface_type) {
1721     case VGA_NONE:
1722         return false;
1723     case VGA_DEVICE:
1724         return true;
1725     case VGA_STD:
1726     case VGA_VIRTIO:
1727     case VGA_CIRRUS:
1728         return pci_vga_init(pci_bus) != NULL;
1729     default:
1730         error_setg(errp,
1731                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1732         return false;
1733     }
1734 }
1735 
1736 static int spapr_pre_load(void *opaque)
1737 {
1738     int rc;
1739 
1740     rc = spapr_caps_pre_load(opaque);
1741     if (rc) {
1742         return rc;
1743     }
1744 
1745     return 0;
1746 }
1747 
1748 static int spapr_post_load(void *opaque, int version_id)
1749 {
1750     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1751     int err = 0;
1752 
1753     err = spapr_caps_post_migration(spapr);
1754     if (err) {
1755         return err;
1756     }
1757 
1758     /*
1759      * In earlier versions, there was no separate qdev for the PAPR
1760      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1761      * So when migrating from those versions, poke the incoming offset
1762      * value into the RTC device
1763      */
1764     if (version_id < 3) {
1765         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1766         if (err) {
1767             return err;
1768         }
1769     }
1770 
1771     if (kvm_enabled() && spapr->patb_entry) {
1772         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1773         bool radix = !!(spapr->patb_entry & PATE1_GR);
1774         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1775 
1776         /*
1777          * Update LPCR:HR and UPRT as they may not be set properly in
1778          * the stream
1779          */
1780         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1781                             LPCR_HR | LPCR_UPRT);
1782 
1783         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1784         if (err) {
1785             error_report("Process table config unsupported by the host");
1786             return -EINVAL;
1787         }
1788     }
1789 
1790     err = spapr_irq_post_load(spapr, version_id);
1791     if (err) {
1792         return err;
1793     }
1794 
1795     return err;
1796 }
1797 
1798 static int spapr_pre_save(void *opaque)
1799 {
1800     int rc;
1801 
1802     rc = spapr_caps_pre_save(opaque);
1803     if (rc) {
1804         return rc;
1805     }
1806 
1807     return 0;
1808 }
1809 
1810 static bool version_before_3(void *opaque, int version_id)
1811 {
1812     return version_id < 3;
1813 }
1814 
1815 static bool spapr_pending_events_needed(void *opaque)
1816 {
1817     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1818     return !QTAILQ_EMPTY(&spapr->pending_events);
1819 }
1820 
1821 static const VMStateDescription vmstate_spapr_event_entry = {
1822     .name = "spapr_event_log_entry",
1823     .version_id = 1,
1824     .minimum_version_id = 1,
1825     .fields = (VMStateField[]) {
1826         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1827         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1828         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1829                                      NULL, extended_length),
1830         VMSTATE_END_OF_LIST()
1831     },
1832 };
1833 
1834 static const VMStateDescription vmstate_spapr_pending_events = {
1835     .name = "spapr_pending_events",
1836     .version_id = 1,
1837     .minimum_version_id = 1,
1838     .needed = spapr_pending_events_needed,
1839     .fields = (VMStateField[]) {
1840         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1841                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1842         VMSTATE_END_OF_LIST()
1843     },
1844 };
1845 
1846 static bool spapr_ov5_cas_needed(void *opaque)
1847 {
1848     SpaprMachineState *spapr = opaque;
1849     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1850     bool cas_needed;
1851 
1852     /* Prior to the introduction of SpaprOptionVector, we had two option
1853      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1854      * Both of these options encode machine topology into the device-tree
1855      * in such a way that the now-booted OS should still be able to interact
1856      * appropriately with QEMU regardless of what options were actually
1857      * negotiatied on the source side.
1858      *
1859      * As such, we can avoid migrating the CAS-negotiated options if these
1860      * are the only options available on the current machine/platform.
1861      * Since these are the only options available for pseries-2.7 and
1862      * earlier, this allows us to maintain old->new/new->old migration
1863      * compatibility.
1864      *
1865      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1866      * via default pseries-2.8 machines and explicit command-line parameters.
1867      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1868      * of the actual CAS-negotiated values to continue working properly. For
1869      * example, availability of memory unplug depends on knowing whether
1870      * OV5_HP_EVT was negotiated via CAS.
1871      *
1872      * Thus, for any cases where the set of available CAS-negotiatable
1873      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1874      * include the CAS-negotiated options in the migration stream, unless
1875      * if they affect boot time behaviour only.
1876      */
1877     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1878     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1879     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1880 
1881     /* We need extra information if we have any bits outside the mask
1882      * defined above */
1883     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1884 
1885     spapr_ovec_cleanup(ov5_mask);
1886 
1887     return cas_needed;
1888 }
1889 
1890 static const VMStateDescription vmstate_spapr_ov5_cas = {
1891     .name = "spapr_option_vector_ov5_cas",
1892     .version_id = 1,
1893     .minimum_version_id = 1,
1894     .needed = spapr_ov5_cas_needed,
1895     .fields = (VMStateField[]) {
1896         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1897                                  vmstate_spapr_ovec, SpaprOptionVector),
1898         VMSTATE_END_OF_LIST()
1899     },
1900 };
1901 
1902 static bool spapr_patb_entry_needed(void *opaque)
1903 {
1904     SpaprMachineState *spapr = opaque;
1905 
1906     return !!spapr->patb_entry;
1907 }
1908 
1909 static const VMStateDescription vmstate_spapr_patb_entry = {
1910     .name = "spapr_patb_entry",
1911     .version_id = 1,
1912     .minimum_version_id = 1,
1913     .needed = spapr_patb_entry_needed,
1914     .fields = (VMStateField[]) {
1915         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1916         VMSTATE_END_OF_LIST()
1917     },
1918 };
1919 
1920 static bool spapr_irq_map_needed(void *opaque)
1921 {
1922     SpaprMachineState *spapr = opaque;
1923 
1924     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1925 }
1926 
1927 static const VMStateDescription vmstate_spapr_irq_map = {
1928     .name = "spapr_irq_map",
1929     .version_id = 1,
1930     .minimum_version_id = 1,
1931     .needed = spapr_irq_map_needed,
1932     .fields = (VMStateField[]) {
1933         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1934         VMSTATE_END_OF_LIST()
1935     },
1936 };
1937 
1938 static bool spapr_dtb_needed(void *opaque)
1939 {
1940     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1941 
1942     return smc->update_dt_enabled;
1943 }
1944 
1945 static int spapr_dtb_pre_load(void *opaque)
1946 {
1947     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1948 
1949     g_free(spapr->fdt_blob);
1950     spapr->fdt_blob = NULL;
1951     spapr->fdt_size = 0;
1952 
1953     return 0;
1954 }
1955 
1956 static const VMStateDescription vmstate_spapr_dtb = {
1957     .name = "spapr_dtb",
1958     .version_id = 1,
1959     .minimum_version_id = 1,
1960     .needed = spapr_dtb_needed,
1961     .pre_load = spapr_dtb_pre_load,
1962     .fields = (VMStateField[]) {
1963         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1964         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1965         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1966                                      fdt_size),
1967         VMSTATE_END_OF_LIST()
1968     },
1969 };
1970 
1971 static bool spapr_fwnmi_needed(void *opaque)
1972 {
1973     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1974 
1975     return spapr->guest_machine_check_addr != -1;
1976 }
1977 
1978 static int spapr_fwnmi_pre_save(void *opaque)
1979 {
1980     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1981 
1982     /*
1983      * Check if machine check handling is in progress and print a
1984      * warning message.
1985      */
1986     if (spapr->mc_status != -1) {
1987         warn_report("A machine check is being handled during migration. The"
1988                 "handler may run and log hardware error on the destination");
1989     }
1990 
1991     return 0;
1992 }
1993 
1994 static const VMStateDescription vmstate_spapr_machine_check = {
1995     .name = "spapr_machine_check",
1996     .version_id = 1,
1997     .minimum_version_id = 1,
1998     .needed = spapr_fwnmi_needed,
1999     .pre_save = spapr_fwnmi_pre_save,
2000     .fields = (VMStateField[]) {
2001         VMSTATE_UINT64(guest_machine_check_addr, SpaprMachineState),
2002         VMSTATE_INT32(mc_status, SpaprMachineState),
2003         VMSTATE_END_OF_LIST()
2004     },
2005 };
2006 
2007 static const VMStateDescription vmstate_spapr = {
2008     .name = "spapr",
2009     .version_id = 3,
2010     .minimum_version_id = 1,
2011     .pre_load = spapr_pre_load,
2012     .post_load = spapr_post_load,
2013     .pre_save = spapr_pre_save,
2014     .fields = (VMStateField[]) {
2015         /* used to be @next_irq */
2016         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2017 
2018         /* RTC offset */
2019         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2020 
2021         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2022         VMSTATE_END_OF_LIST()
2023     },
2024     .subsections = (const VMStateDescription*[]) {
2025         &vmstate_spapr_ov5_cas,
2026         &vmstate_spapr_patb_entry,
2027         &vmstate_spapr_pending_events,
2028         &vmstate_spapr_cap_htm,
2029         &vmstate_spapr_cap_vsx,
2030         &vmstate_spapr_cap_dfp,
2031         &vmstate_spapr_cap_cfpc,
2032         &vmstate_spapr_cap_sbbc,
2033         &vmstate_spapr_cap_ibs,
2034         &vmstate_spapr_cap_hpt_maxpagesize,
2035         &vmstate_spapr_irq_map,
2036         &vmstate_spapr_cap_nested_kvm_hv,
2037         &vmstate_spapr_dtb,
2038         &vmstate_spapr_cap_large_decr,
2039         &vmstate_spapr_cap_ccf_assist,
2040         &vmstate_spapr_cap_fwnmi,
2041         &vmstate_spapr_machine_check,
2042         NULL
2043     }
2044 };
2045 
2046 static int htab_save_setup(QEMUFile *f, void *opaque)
2047 {
2048     SpaprMachineState *spapr = opaque;
2049 
2050     /* "Iteration" header */
2051     if (!spapr->htab_shift) {
2052         qemu_put_be32(f, -1);
2053     } else {
2054         qemu_put_be32(f, spapr->htab_shift);
2055     }
2056 
2057     if (spapr->htab) {
2058         spapr->htab_save_index = 0;
2059         spapr->htab_first_pass = true;
2060     } else {
2061         if (spapr->htab_shift) {
2062             assert(kvm_enabled());
2063         }
2064     }
2065 
2066 
2067     return 0;
2068 }
2069 
2070 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2071                             int chunkstart, int n_valid, int n_invalid)
2072 {
2073     qemu_put_be32(f, chunkstart);
2074     qemu_put_be16(f, n_valid);
2075     qemu_put_be16(f, n_invalid);
2076     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2077                     HASH_PTE_SIZE_64 * n_valid);
2078 }
2079 
2080 static void htab_save_end_marker(QEMUFile *f)
2081 {
2082     qemu_put_be32(f, 0);
2083     qemu_put_be16(f, 0);
2084     qemu_put_be16(f, 0);
2085 }
2086 
2087 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2088                                  int64_t max_ns)
2089 {
2090     bool has_timeout = max_ns != -1;
2091     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2092     int index = spapr->htab_save_index;
2093     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2094 
2095     assert(spapr->htab_first_pass);
2096 
2097     do {
2098         int chunkstart;
2099 
2100         /* Consume invalid HPTEs */
2101         while ((index < htabslots)
2102                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2103             CLEAN_HPTE(HPTE(spapr->htab, index));
2104             index++;
2105         }
2106 
2107         /* Consume valid HPTEs */
2108         chunkstart = index;
2109         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2110                && HPTE_VALID(HPTE(spapr->htab, index))) {
2111             CLEAN_HPTE(HPTE(spapr->htab, index));
2112             index++;
2113         }
2114 
2115         if (index > chunkstart) {
2116             int n_valid = index - chunkstart;
2117 
2118             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2119 
2120             if (has_timeout &&
2121                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2122                 break;
2123             }
2124         }
2125     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2126 
2127     if (index >= htabslots) {
2128         assert(index == htabslots);
2129         index = 0;
2130         spapr->htab_first_pass = false;
2131     }
2132     spapr->htab_save_index = index;
2133 }
2134 
2135 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2136                                 int64_t max_ns)
2137 {
2138     bool final = max_ns < 0;
2139     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2140     int examined = 0, sent = 0;
2141     int index = spapr->htab_save_index;
2142     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2143 
2144     assert(!spapr->htab_first_pass);
2145 
2146     do {
2147         int chunkstart, invalidstart;
2148 
2149         /* Consume non-dirty HPTEs */
2150         while ((index < htabslots)
2151                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2152             index++;
2153             examined++;
2154         }
2155 
2156         chunkstart = index;
2157         /* Consume valid dirty HPTEs */
2158         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2159                && HPTE_DIRTY(HPTE(spapr->htab, index))
2160                && HPTE_VALID(HPTE(spapr->htab, index))) {
2161             CLEAN_HPTE(HPTE(spapr->htab, index));
2162             index++;
2163             examined++;
2164         }
2165 
2166         invalidstart = index;
2167         /* Consume invalid dirty HPTEs */
2168         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2169                && HPTE_DIRTY(HPTE(spapr->htab, index))
2170                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2171             CLEAN_HPTE(HPTE(spapr->htab, index));
2172             index++;
2173             examined++;
2174         }
2175 
2176         if (index > chunkstart) {
2177             int n_valid = invalidstart - chunkstart;
2178             int n_invalid = index - invalidstart;
2179 
2180             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2181             sent += index - chunkstart;
2182 
2183             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2184                 break;
2185             }
2186         }
2187 
2188         if (examined >= htabslots) {
2189             break;
2190         }
2191 
2192         if (index >= htabslots) {
2193             assert(index == htabslots);
2194             index = 0;
2195         }
2196     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2197 
2198     if (index >= htabslots) {
2199         assert(index == htabslots);
2200         index = 0;
2201     }
2202 
2203     spapr->htab_save_index = index;
2204 
2205     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2206 }
2207 
2208 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2209 #define MAX_KVM_BUF_SIZE    2048
2210 
2211 static int htab_save_iterate(QEMUFile *f, void *opaque)
2212 {
2213     SpaprMachineState *spapr = opaque;
2214     int fd;
2215     int rc = 0;
2216 
2217     /* Iteration header */
2218     if (!spapr->htab_shift) {
2219         qemu_put_be32(f, -1);
2220         return 1;
2221     } else {
2222         qemu_put_be32(f, 0);
2223     }
2224 
2225     if (!spapr->htab) {
2226         assert(kvm_enabled());
2227 
2228         fd = get_htab_fd(spapr);
2229         if (fd < 0) {
2230             return fd;
2231         }
2232 
2233         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2234         if (rc < 0) {
2235             return rc;
2236         }
2237     } else  if (spapr->htab_first_pass) {
2238         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2239     } else {
2240         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2241     }
2242 
2243     htab_save_end_marker(f);
2244 
2245     return rc;
2246 }
2247 
2248 static int htab_save_complete(QEMUFile *f, void *opaque)
2249 {
2250     SpaprMachineState *spapr = opaque;
2251     int fd;
2252 
2253     /* Iteration header */
2254     if (!spapr->htab_shift) {
2255         qemu_put_be32(f, -1);
2256         return 0;
2257     } else {
2258         qemu_put_be32(f, 0);
2259     }
2260 
2261     if (!spapr->htab) {
2262         int rc;
2263 
2264         assert(kvm_enabled());
2265 
2266         fd = get_htab_fd(spapr);
2267         if (fd < 0) {
2268             return fd;
2269         }
2270 
2271         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2272         if (rc < 0) {
2273             return rc;
2274         }
2275     } else {
2276         if (spapr->htab_first_pass) {
2277             htab_save_first_pass(f, spapr, -1);
2278         }
2279         htab_save_later_pass(f, spapr, -1);
2280     }
2281 
2282     /* End marker */
2283     htab_save_end_marker(f);
2284 
2285     return 0;
2286 }
2287 
2288 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2289 {
2290     SpaprMachineState *spapr = opaque;
2291     uint32_t section_hdr;
2292     int fd = -1;
2293     Error *local_err = NULL;
2294 
2295     if (version_id < 1 || version_id > 1) {
2296         error_report("htab_load() bad version");
2297         return -EINVAL;
2298     }
2299 
2300     section_hdr = qemu_get_be32(f);
2301 
2302     if (section_hdr == -1) {
2303         spapr_free_hpt(spapr);
2304         return 0;
2305     }
2306 
2307     if (section_hdr) {
2308         /* First section gives the htab size */
2309         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2310         if (local_err) {
2311             error_report_err(local_err);
2312             return -EINVAL;
2313         }
2314         return 0;
2315     }
2316 
2317     if (!spapr->htab) {
2318         assert(kvm_enabled());
2319 
2320         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2321         if (fd < 0) {
2322             error_report_err(local_err);
2323             return fd;
2324         }
2325     }
2326 
2327     while (true) {
2328         uint32_t index;
2329         uint16_t n_valid, n_invalid;
2330 
2331         index = qemu_get_be32(f);
2332         n_valid = qemu_get_be16(f);
2333         n_invalid = qemu_get_be16(f);
2334 
2335         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2336             /* End of Stream */
2337             break;
2338         }
2339 
2340         if ((index + n_valid + n_invalid) >
2341             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2342             /* Bad index in stream */
2343             error_report(
2344                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2345                 index, n_valid, n_invalid, spapr->htab_shift);
2346             return -EINVAL;
2347         }
2348 
2349         if (spapr->htab) {
2350             if (n_valid) {
2351                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2352                                 HASH_PTE_SIZE_64 * n_valid);
2353             }
2354             if (n_invalid) {
2355                 memset(HPTE(spapr->htab, index + n_valid), 0,
2356                        HASH_PTE_SIZE_64 * n_invalid);
2357             }
2358         } else {
2359             int rc;
2360 
2361             assert(fd >= 0);
2362 
2363             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2364             if (rc < 0) {
2365                 return rc;
2366             }
2367         }
2368     }
2369 
2370     if (!spapr->htab) {
2371         assert(fd >= 0);
2372         close(fd);
2373     }
2374 
2375     return 0;
2376 }
2377 
2378 static void htab_save_cleanup(void *opaque)
2379 {
2380     SpaprMachineState *spapr = opaque;
2381 
2382     close_htab_fd(spapr);
2383 }
2384 
2385 static SaveVMHandlers savevm_htab_handlers = {
2386     .save_setup = htab_save_setup,
2387     .save_live_iterate = htab_save_iterate,
2388     .save_live_complete_precopy = htab_save_complete,
2389     .save_cleanup = htab_save_cleanup,
2390     .load_state = htab_load,
2391 };
2392 
2393 static void spapr_boot_set(void *opaque, const char *boot_device,
2394                            Error **errp)
2395 {
2396     MachineState *machine = MACHINE(opaque);
2397     machine->boot_order = g_strdup(boot_device);
2398 }
2399 
2400 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2401 {
2402     MachineState *machine = MACHINE(spapr);
2403     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2404     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2405     int i;
2406 
2407     for (i = 0; i < nr_lmbs; i++) {
2408         uint64_t addr;
2409 
2410         addr = i * lmb_size + machine->device_memory->base;
2411         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2412                                addr / lmb_size);
2413     }
2414 }
2415 
2416 /*
2417  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2418  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2419  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2420  */
2421 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2422 {
2423     int i;
2424 
2425     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2426         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2427                    " is not aligned to %" PRIu64 " MiB",
2428                    machine->ram_size,
2429                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2430         return;
2431     }
2432 
2433     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2434         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2435                    " is not aligned to %" PRIu64 " MiB",
2436                    machine->ram_size,
2437                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2438         return;
2439     }
2440 
2441     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2442         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2443             error_setg(errp,
2444                        "Node %d memory size 0x%" PRIx64
2445                        " is not aligned to %" PRIu64 " MiB",
2446                        i, machine->numa_state->nodes[i].node_mem,
2447                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2448             return;
2449         }
2450     }
2451 }
2452 
2453 /* find cpu slot in machine->possible_cpus by core_id */
2454 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2455 {
2456     int index = id / ms->smp.threads;
2457 
2458     if (index >= ms->possible_cpus->len) {
2459         return NULL;
2460     }
2461     if (idx) {
2462         *idx = index;
2463     }
2464     return &ms->possible_cpus->cpus[index];
2465 }
2466 
2467 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2468 {
2469     MachineState *ms = MACHINE(spapr);
2470     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2471     Error *local_err = NULL;
2472     bool vsmt_user = !!spapr->vsmt;
2473     int kvm_smt = kvmppc_smt_threads();
2474     int ret;
2475     unsigned int smp_threads = ms->smp.threads;
2476 
2477     if (!kvm_enabled() && (smp_threads > 1)) {
2478         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2479                      "on a pseries machine");
2480         goto out;
2481     }
2482     if (!is_power_of_2(smp_threads)) {
2483         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2484                      "machine because it must be a power of 2", smp_threads);
2485         goto out;
2486     }
2487 
2488     /* Detemine the VSMT mode to use: */
2489     if (vsmt_user) {
2490         if (spapr->vsmt < smp_threads) {
2491             error_setg(&local_err, "Cannot support VSMT mode %d"
2492                          " because it must be >= threads/core (%d)",
2493                          spapr->vsmt, smp_threads);
2494             goto out;
2495         }
2496         /* In this case, spapr->vsmt has been set by the command line */
2497     } else if (!smc->smp_threads_vsmt) {
2498         /*
2499          * Default VSMT value is tricky, because we need it to be as
2500          * consistent as possible (for migration), but this requires
2501          * changing it for at least some existing cases.  We pick 8 as
2502          * the value that we'd get with KVM on POWER8, the
2503          * overwhelmingly common case in production systems.
2504          */
2505         spapr->vsmt = MAX(8, smp_threads);
2506     } else {
2507         spapr->vsmt = smp_threads;
2508     }
2509 
2510     /* KVM: If necessary, set the SMT mode: */
2511     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2512         ret = kvmppc_set_smt_threads(spapr->vsmt);
2513         if (ret) {
2514             /* Looks like KVM isn't able to change VSMT mode */
2515             error_setg(&local_err,
2516                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2517                        spapr->vsmt, ret);
2518             /* We can live with that if the default one is big enough
2519              * for the number of threads, and a submultiple of the one
2520              * we want.  In this case we'll waste some vcpu ids, but
2521              * behaviour will be correct */
2522             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2523                 warn_report_err(local_err);
2524                 local_err = NULL;
2525                 goto out;
2526             } else {
2527                 if (!vsmt_user) {
2528                     error_append_hint(&local_err,
2529                                       "On PPC, a VM with %d threads/core"
2530                                       " on a host with %d threads/core"
2531                                       " requires the use of VSMT mode %d.\n",
2532                                       smp_threads, kvm_smt, spapr->vsmt);
2533                 }
2534                 kvmppc_error_append_smt_possible_hint(&local_err);
2535                 goto out;
2536             }
2537         }
2538     }
2539     /* else TCG: nothing to do currently */
2540 out:
2541     error_propagate(errp, local_err);
2542 }
2543 
2544 static void spapr_init_cpus(SpaprMachineState *spapr)
2545 {
2546     MachineState *machine = MACHINE(spapr);
2547     MachineClass *mc = MACHINE_GET_CLASS(machine);
2548     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2549     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2550     const CPUArchIdList *possible_cpus;
2551     unsigned int smp_cpus = machine->smp.cpus;
2552     unsigned int smp_threads = machine->smp.threads;
2553     unsigned int max_cpus = machine->smp.max_cpus;
2554     int boot_cores_nr = smp_cpus / smp_threads;
2555     int i;
2556 
2557     possible_cpus = mc->possible_cpu_arch_ids(machine);
2558     if (mc->has_hotpluggable_cpus) {
2559         if (smp_cpus % smp_threads) {
2560             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2561                          smp_cpus, smp_threads);
2562             exit(1);
2563         }
2564         if (max_cpus % smp_threads) {
2565             error_report("max_cpus (%u) must be multiple of threads (%u)",
2566                          max_cpus, smp_threads);
2567             exit(1);
2568         }
2569     } else {
2570         if (max_cpus != smp_cpus) {
2571             error_report("This machine version does not support CPU hotplug");
2572             exit(1);
2573         }
2574         boot_cores_nr = possible_cpus->len;
2575     }
2576 
2577     if (smc->pre_2_10_has_unused_icps) {
2578         int i;
2579 
2580         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2581             /* Dummy entries get deregistered when real ICPState objects
2582              * are registered during CPU core hotplug.
2583              */
2584             pre_2_10_vmstate_register_dummy_icp(i);
2585         }
2586     }
2587 
2588     for (i = 0; i < possible_cpus->len; i++) {
2589         int core_id = i * smp_threads;
2590 
2591         if (mc->has_hotpluggable_cpus) {
2592             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2593                                    spapr_vcpu_id(spapr, core_id));
2594         }
2595 
2596         if (i < boot_cores_nr) {
2597             Object *core  = object_new(type);
2598             int nr_threads = smp_threads;
2599 
2600             /* Handle the partially filled core for older machine types */
2601             if ((i + 1) * smp_threads >= smp_cpus) {
2602                 nr_threads = smp_cpus - i * smp_threads;
2603             }
2604 
2605             object_property_set_int(core, nr_threads, "nr-threads",
2606                                     &error_fatal);
2607             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2608                                     &error_fatal);
2609             object_property_set_bool(core, true, "realized", &error_fatal);
2610 
2611             object_unref(core);
2612         }
2613     }
2614 }
2615 
2616 static PCIHostState *spapr_create_default_phb(void)
2617 {
2618     DeviceState *dev;
2619 
2620     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2621     qdev_prop_set_uint32(dev, "index", 0);
2622     qdev_init_nofail(dev);
2623 
2624     return PCI_HOST_BRIDGE(dev);
2625 }
2626 
2627 /* pSeries LPAR / sPAPR hardware init */
2628 static void spapr_machine_init(MachineState *machine)
2629 {
2630     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2631     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2632     const char *kernel_filename = machine->kernel_filename;
2633     const char *initrd_filename = machine->initrd_filename;
2634     PCIHostState *phb;
2635     int i;
2636     MemoryRegion *sysmem = get_system_memory();
2637     MemoryRegion *ram = g_new(MemoryRegion, 1);
2638     hwaddr node0_size = spapr_node0_size(machine);
2639     long load_limit, fw_size;
2640     char *filename;
2641     Error *resize_hpt_err = NULL;
2642 
2643     msi_nonbroken = true;
2644 
2645     QLIST_INIT(&spapr->phbs);
2646     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2647 
2648     /* Determine capabilities to run with */
2649     spapr_caps_init(spapr);
2650 
2651     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2652     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2653         /*
2654          * If the user explicitly requested a mode we should either
2655          * supply it, or fail completely (which we do below).  But if
2656          * it's not set explicitly, we reset our mode to something
2657          * that works
2658          */
2659         if (resize_hpt_err) {
2660             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2661             error_free(resize_hpt_err);
2662             resize_hpt_err = NULL;
2663         } else {
2664             spapr->resize_hpt = smc->resize_hpt_default;
2665         }
2666     }
2667 
2668     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2669 
2670     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2671         /*
2672          * User requested HPT resize, but this host can't supply it.  Bail out
2673          */
2674         error_report_err(resize_hpt_err);
2675         exit(1);
2676     }
2677 
2678     spapr->rma_size = node0_size;
2679 
2680     /* With KVM, we don't actually know whether KVM supports an
2681      * unbounded RMA (PR KVM) or is limited by the hash table size
2682      * (HV KVM using VRMA), so we always assume the latter
2683      *
2684      * In that case, we also limit the initial allocations for RTAS
2685      * etc... to 256M since we have no way to know what the VRMA size
2686      * is going to be as it depends on the size of the hash table
2687      * which isn't determined yet.
2688      */
2689     if (kvm_enabled()) {
2690         spapr->vrma_adjust = 1;
2691         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2692     }
2693 
2694     /* Actually we don't support unbounded RMA anymore since we added
2695      * proper emulation of HV mode. The max we can get is 16G which
2696      * also happens to be what we configure for PAPR mode so make sure
2697      * we don't do anything bigger than that
2698      */
2699     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2700 
2701     if (spapr->rma_size > node0_size) {
2702         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2703                      spapr->rma_size);
2704         exit(1);
2705     }
2706 
2707     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2708     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2709 
2710     /*
2711      * VSMT must be set in order to be able to compute VCPU ids, ie to
2712      * call spapr_max_server_number() or spapr_vcpu_id().
2713      */
2714     spapr_set_vsmt_mode(spapr, &error_fatal);
2715 
2716     /* Set up Interrupt Controller before we create the VCPUs */
2717     spapr_irq_init(spapr, &error_fatal);
2718 
2719     /* Set up containers for ibm,client-architecture-support negotiated options
2720      */
2721     spapr->ov5 = spapr_ovec_new();
2722     spapr->ov5_cas = spapr_ovec_new();
2723 
2724     if (smc->dr_lmb_enabled) {
2725         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2726         spapr_validate_node_memory(machine, &error_fatal);
2727     }
2728 
2729     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2730 
2731     /* advertise support for dedicated HP event source to guests */
2732     if (spapr->use_hotplug_event_source) {
2733         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2734     }
2735 
2736     /* advertise support for HPT resizing */
2737     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2738         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2739     }
2740 
2741     /* advertise support for ibm,dyamic-memory-v2 */
2742     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2743 
2744     /* advertise XIVE on POWER9 machines */
2745     if (spapr->irq->xive) {
2746         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2747     }
2748 
2749     /* init CPUs */
2750     spapr_init_cpus(spapr);
2751 
2752     /*
2753      * check we don't have a memory-less/cpu-less NUMA node
2754      * Firmware relies on the existing memory/cpu topology to provide the
2755      * NUMA topology to the kernel.
2756      * And the linux kernel needs to know the NUMA topology at start
2757      * to be able to hotplug CPUs later.
2758      */
2759     if (machine->numa_state->num_nodes) {
2760         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2761             /* check for memory-less node */
2762             if (machine->numa_state->nodes[i].node_mem == 0) {
2763                 CPUState *cs;
2764                 int found = 0;
2765                 /* check for cpu-less node */
2766                 CPU_FOREACH(cs) {
2767                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2768                     if (cpu->node_id == i) {
2769                         found = 1;
2770                         break;
2771                     }
2772                 }
2773                 /* memory-less and cpu-less node */
2774                 if (!found) {
2775                     error_report(
2776                        "Memory-less/cpu-less nodes are not supported (node %d)",
2777                                  i);
2778                     exit(1);
2779                 }
2780             }
2781         }
2782 
2783     }
2784 
2785     /*
2786      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2787      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2788      * called from vPHB reset handler so we initialize the counter here.
2789      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2790      * must be equally distant from any other node.
2791      * The final value of spapr->gpu_numa_id is going to be written to
2792      * max-associativity-domains in spapr_build_fdt().
2793      */
2794     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2795 
2796     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2797         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2798                               spapr->max_compat_pvr)) {
2799         /* KVM and TCG always allow GTSE with radix... */
2800         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2801     }
2802     /* ... but not with hash (currently). */
2803 
2804     if (kvm_enabled()) {
2805         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2806         kvmppc_enable_logical_ci_hcalls();
2807         kvmppc_enable_set_mode_hcall();
2808 
2809         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2810         kvmppc_enable_clear_ref_mod_hcalls();
2811 
2812         /* Enable H_PAGE_INIT */
2813         kvmppc_enable_h_page_init();
2814     }
2815 
2816     /* allocate RAM */
2817     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2818                                          machine->ram_size);
2819     memory_region_add_subregion(sysmem, 0, ram);
2820 
2821     /* always allocate the device memory information */
2822     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2823 
2824     /* initialize hotplug memory address space */
2825     if (machine->ram_size < machine->maxram_size) {
2826         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2827         /*
2828          * Limit the number of hotpluggable memory slots to half the number
2829          * slots that KVM supports, leaving the other half for PCI and other
2830          * devices. However ensure that number of slots doesn't drop below 32.
2831          */
2832         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2833                            SPAPR_MAX_RAM_SLOTS;
2834 
2835         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2836             max_memslots = SPAPR_MAX_RAM_SLOTS;
2837         }
2838         if (machine->ram_slots > max_memslots) {
2839             error_report("Specified number of memory slots %"
2840                          PRIu64" exceeds max supported %d",
2841                          machine->ram_slots, max_memslots);
2842             exit(1);
2843         }
2844 
2845         machine->device_memory->base = ROUND_UP(machine->ram_size,
2846                                                 SPAPR_DEVICE_MEM_ALIGN);
2847         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2848                            "device-memory", device_mem_size);
2849         memory_region_add_subregion(sysmem, machine->device_memory->base,
2850                                     &machine->device_memory->mr);
2851     }
2852 
2853     if (smc->dr_lmb_enabled) {
2854         spapr_create_lmb_dr_connectors(spapr);
2855     }
2856 
2857     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI_MCE) == SPAPR_CAP_ON) {
2858         /* Create the error string for live migration blocker */
2859         error_setg(&spapr->fwnmi_migration_blocker,
2860             "A machine check is being handled during migration. The handler"
2861             "may run and log hardware error on the destination");
2862     }
2863 
2864     /* Set up RTAS event infrastructure */
2865     spapr_events_init(spapr);
2866 
2867     /* Set up the RTC RTAS interfaces */
2868     spapr_rtc_create(spapr);
2869 
2870     /* Set up VIO bus */
2871     spapr->vio_bus = spapr_vio_bus_init();
2872 
2873     for (i = 0; i < serial_max_hds(); i++) {
2874         if (serial_hd(i)) {
2875             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2876         }
2877     }
2878 
2879     /* We always have at least the nvram device on VIO */
2880     spapr_create_nvram(spapr);
2881 
2882     /*
2883      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2884      * connectors (described in root DT node's "ibm,drc-types" property)
2885      * are pre-initialized here. additional child connectors (such as
2886      * connectors for a PHBs PCI slots) are added as needed during their
2887      * parent's realization.
2888      */
2889     if (smc->dr_phb_enabled) {
2890         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2891             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2892         }
2893     }
2894 
2895     /* Set up PCI */
2896     spapr_pci_rtas_init();
2897 
2898     phb = spapr_create_default_phb();
2899 
2900     for (i = 0; i < nb_nics; i++) {
2901         NICInfo *nd = &nd_table[i];
2902 
2903         if (!nd->model) {
2904             nd->model = g_strdup("spapr-vlan");
2905         }
2906 
2907         if (g_str_equal(nd->model, "spapr-vlan") ||
2908             g_str_equal(nd->model, "ibmveth")) {
2909             spapr_vlan_create(spapr->vio_bus, nd);
2910         } else {
2911             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2912         }
2913     }
2914 
2915     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2916         spapr_vscsi_create(spapr->vio_bus);
2917     }
2918 
2919     /* Graphics */
2920     if (spapr_vga_init(phb->bus, &error_fatal)) {
2921         spapr->has_graphics = true;
2922         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2923     }
2924 
2925     if (machine->usb) {
2926         if (smc->use_ohci_by_default) {
2927             pci_create_simple(phb->bus, -1, "pci-ohci");
2928         } else {
2929             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2930         }
2931 
2932         if (spapr->has_graphics) {
2933             USBBus *usb_bus = usb_bus_find(-1);
2934 
2935             usb_create_simple(usb_bus, "usb-kbd");
2936             usb_create_simple(usb_bus, "usb-mouse");
2937         }
2938     }
2939 
2940     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2941         error_report(
2942             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2943             MIN_RMA_SLOF);
2944         exit(1);
2945     }
2946 
2947     if (kernel_filename) {
2948         uint64_t lowaddr = 0;
2949 
2950         spapr->kernel_size = load_elf(kernel_filename, NULL,
2951                                       translate_kernel_address, NULL,
2952                                       NULL, &lowaddr, NULL, NULL, 1,
2953                                       PPC_ELF_MACHINE, 0, 0);
2954         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2955             spapr->kernel_size = load_elf(kernel_filename, NULL,
2956                                           translate_kernel_address, NULL, NULL,
2957                                           &lowaddr, NULL, NULL, 0,
2958                                           PPC_ELF_MACHINE, 0, 0);
2959             spapr->kernel_le = spapr->kernel_size > 0;
2960         }
2961         if (spapr->kernel_size < 0) {
2962             error_report("error loading %s: %s", kernel_filename,
2963                          load_elf_strerror(spapr->kernel_size));
2964             exit(1);
2965         }
2966 
2967         /* load initrd */
2968         if (initrd_filename) {
2969             /* Try to locate the initrd in the gap between the kernel
2970              * and the firmware. Add a bit of space just in case
2971              */
2972             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2973                                   + 0x1ffff) & ~0xffff;
2974             spapr->initrd_size = load_image_targphys(initrd_filename,
2975                                                      spapr->initrd_base,
2976                                                      load_limit
2977                                                      - spapr->initrd_base);
2978             if (spapr->initrd_size < 0) {
2979                 error_report("could not load initial ram disk '%s'",
2980                              initrd_filename);
2981                 exit(1);
2982             }
2983         }
2984     }
2985 
2986     if (bios_name == NULL) {
2987         bios_name = FW_FILE_NAME;
2988     }
2989     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2990     if (!filename) {
2991         error_report("Could not find LPAR firmware '%s'", bios_name);
2992         exit(1);
2993     }
2994     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2995     if (fw_size <= 0) {
2996         error_report("Could not load LPAR firmware '%s'", filename);
2997         exit(1);
2998     }
2999     g_free(filename);
3000 
3001     /* FIXME: Should register things through the MachineState's qdev
3002      * interface, this is a legacy from the sPAPREnvironment structure
3003      * which predated MachineState but had a similar function */
3004     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3005     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3006                          &savevm_htab_handlers, spapr);
3007 
3008     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3009                              &error_fatal);
3010 
3011     qemu_register_boot_set(spapr_boot_set, spapr);
3012 
3013     /*
3014      * Nothing needs to be done to resume a suspended guest because
3015      * suspending does not change the machine state, so no need for
3016      * a ->wakeup method.
3017      */
3018     qemu_register_wakeup_support();
3019 
3020     if (kvm_enabled()) {
3021         /* to stop and start vmclock */
3022         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3023                                          &spapr->tb);
3024 
3025         kvmppc_spapr_enable_inkernel_multitce();
3026     }
3027 
3028     qemu_cond_init(&spapr->mc_delivery_cond);
3029 }
3030 
3031 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3032 {
3033     if (!vm_type) {
3034         return 0;
3035     }
3036 
3037     if (!strcmp(vm_type, "HV")) {
3038         return 1;
3039     }
3040 
3041     if (!strcmp(vm_type, "PR")) {
3042         return 2;
3043     }
3044 
3045     error_report("Unknown kvm-type specified '%s'", vm_type);
3046     exit(1);
3047 }
3048 
3049 /*
3050  * Implementation of an interface to adjust firmware path
3051  * for the bootindex property handling.
3052  */
3053 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3054                                    DeviceState *dev)
3055 {
3056 #define CAST(type, obj, name) \
3057     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3058     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3059     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3060     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3061 
3062     if (d) {
3063         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3064         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3065         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3066 
3067         if (spapr) {
3068             /*
3069              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3070              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3071              * 0x8000 | (target << 8) | (bus << 5) | lun
3072              * (see the "Logical unit addressing format" table in SAM5)
3073              */
3074             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3075             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3076                                    (uint64_t)id << 48);
3077         } else if (virtio) {
3078             /*
3079              * We use SRP luns of the form 01000000 | (target << 8) | lun
3080              * in the top 32 bits of the 64-bit LUN
3081              * Note: the quote above is from SLOF and it is wrong,
3082              * the actual binding is:
3083              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3084              */
3085             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3086             if (d->lun >= 256) {
3087                 /* Use the LUN "flat space addressing method" */
3088                 id |= 0x4000;
3089             }
3090             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3091                                    (uint64_t)id << 32);
3092         } else if (usb) {
3093             /*
3094              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3095              * in the top 32 bits of the 64-bit LUN
3096              */
3097             unsigned usb_port = atoi(usb->port->path);
3098             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3099             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3100                                    (uint64_t)id << 32);
3101         }
3102     }
3103 
3104     /*
3105      * SLOF probes the USB devices, and if it recognizes that the device is a
3106      * storage device, it changes its name to "storage" instead of "usb-host",
3107      * and additionally adds a child node for the SCSI LUN, so the correct
3108      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3109      */
3110     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3111         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3112         if (usb_host_dev_is_scsi_storage(usbdev)) {
3113             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3114         }
3115     }
3116 
3117     if (phb) {
3118         /* Replace "pci" with "pci@800000020000000" */
3119         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3120     }
3121 
3122     if (vsc) {
3123         /* Same logic as virtio above */
3124         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3125         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3126     }
3127 
3128     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3129         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3130         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3131         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3132     }
3133 
3134     return NULL;
3135 }
3136 
3137 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3138 {
3139     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3140 
3141     return g_strdup(spapr->kvm_type);
3142 }
3143 
3144 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3145 {
3146     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3147 
3148     g_free(spapr->kvm_type);
3149     spapr->kvm_type = g_strdup(value);
3150 }
3151 
3152 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3153 {
3154     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3155 
3156     return spapr->use_hotplug_event_source;
3157 }
3158 
3159 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3160                                             Error **errp)
3161 {
3162     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3163 
3164     spapr->use_hotplug_event_source = value;
3165 }
3166 
3167 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3168 {
3169     return true;
3170 }
3171 
3172 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3173 {
3174     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3175 
3176     switch (spapr->resize_hpt) {
3177     case SPAPR_RESIZE_HPT_DEFAULT:
3178         return g_strdup("default");
3179     case SPAPR_RESIZE_HPT_DISABLED:
3180         return g_strdup("disabled");
3181     case SPAPR_RESIZE_HPT_ENABLED:
3182         return g_strdup("enabled");
3183     case SPAPR_RESIZE_HPT_REQUIRED:
3184         return g_strdup("required");
3185     }
3186     g_assert_not_reached();
3187 }
3188 
3189 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3190 {
3191     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3192 
3193     if (strcmp(value, "default") == 0) {
3194         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3195     } else if (strcmp(value, "disabled") == 0) {
3196         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3197     } else if (strcmp(value, "enabled") == 0) {
3198         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3199     } else if (strcmp(value, "required") == 0) {
3200         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3201     } else {
3202         error_setg(errp, "Bad value for \"resize-hpt\" property");
3203     }
3204 }
3205 
3206 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3207                                    void *opaque, Error **errp)
3208 {
3209     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3210 }
3211 
3212 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3213                                    void *opaque, Error **errp)
3214 {
3215     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3216 }
3217 
3218 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3219 {
3220     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3221 
3222     if (spapr->irq == &spapr_irq_xics_legacy) {
3223         return g_strdup("legacy");
3224     } else if (spapr->irq == &spapr_irq_xics) {
3225         return g_strdup("xics");
3226     } else if (spapr->irq == &spapr_irq_xive) {
3227         return g_strdup("xive");
3228     } else if (spapr->irq == &spapr_irq_dual) {
3229         return g_strdup("dual");
3230     }
3231     g_assert_not_reached();
3232 }
3233 
3234 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3235 {
3236     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3237 
3238     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3239         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3240         return;
3241     }
3242 
3243     /* The legacy IRQ backend can not be set */
3244     if (strcmp(value, "xics") == 0) {
3245         spapr->irq = &spapr_irq_xics;
3246     } else if (strcmp(value, "xive") == 0) {
3247         spapr->irq = &spapr_irq_xive;
3248     } else if (strcmp(value, "dual") == 0) {
3249         spapr->irq = &spapr_irq_dual;
3250     } else {
3251         error_setg(errp, "Bad value for \"ic-mode\" property");
3252     }
3253 }
3254 
3255 static char *spapr_get_host_model(Object *obj, Error **errp)
3256 {
3257     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3258 
3259     return g_strdup(spapr->host_model);
3260 }
3261 
3262 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3263 {
3264     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3265 
3266     g_free(spapr->host_model);
3267     spapr->host_model = g_strdup(value);
3268 }
3269 
3270 static char *spapr_get_host_serial(Object *obj, Error **errp)
3271 {
3272     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3273 
3274     return g_strdup(spapr->host_serial);
3275 }
3276 
3277 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3278 {
3279     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3280 
3281     g_free(spapr->host_serial);
3282     spapr->host_serial = g_strdup(value);
3283 }
3284 
3285 static void spapr_instance_init(Object *obj)
3286 {
3287     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3288     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3289 
3290     spapr->htab_fd = -1;
3291     spapr->use_hotplug_event_source = true;
3292     object_property_add_str(obj, "kvm-type",
3293                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3294     object_property_set_description(obj, "kvm-type",
3295                                     "Specifies the KVM virtualization mode (HV, PR)",
3296                                     NULL);
3297     object_property_add_bool(obj, "modern-hotplug-events",
3298                             spapr_get_modern_hotplug_events,
3299                             spapr_set_modern_hotplug_events,
3300                             NULL);
3301     object_property_set_description(obj, "modern-hotplug-events",
3302                                     "Use dedicated hotplug event mechanism in"
3303                                     " place of standard EPOW events when possible"
3304                                     " (required for memory hot-unplug support)",
3305                                     NULL);
3306     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3307                             "Maximum permitted CPU compatibility mode",
3308                             &error_fatal);
3309 
3310     object_property_add_str(obj, "resize-hpt",
3311                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3312     object_property_set_description(obj, "resize-hpt",
3313                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3314                                     NULL);
3315     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3316                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3317     object_property_set_description(obj, "vsmt",
3318                                     "Virtual SMT: KVM behaves as if this were"
3319                                     " the host's SMT mode", &error_abort);
3320     object_property_add_bool(obj, "vfio-no-msix-emulation",
3321                              spapr_get_msix_emulation, NULL, NULL);
3322 
3323     /* The machine class defines the default interrupt controller mode */
3324     spapr->irq = smc->irq;
3325     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3326                             spapr_set_ic_mode, NULL);
3327     object_property_set_description(obj, "ic-mode",
3328                  "Specifies the interrupt controller mode (xics, xive, dual)",
3329                  NULL);
3330 
3331     object_property_add_str(obj, "host-model",
3332         spapr_get_host_model, spapr_set_host_model,
3333         &error_abort);
3334     object_property_set_description(obj, "host-model",
3335         "Host model to advertise in guest device tree", &error_abort);
3336     object_property_add_str(obj, "host-serial",
3337         spapr_get_host_serial, spapr_set_host_serial,
3338         &error_abort);
3339     object_property_set_description(obj, "host-serial",
3340         "Host serial number to advertise in guest device tree", &error_abort);
3341 }
3342 
3343 static void spapr_machine_finalizefn(Object *obj)
3344 {
3345     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3346 
3347     g_free(spapr->kvm_type);
3348 }
3349 
3350 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3351 {
3352     cpu_synchronize_state(cs);
3353     ppc_cpu_do_system_reset(cs);
3354 }
3355 
3356 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3357 {
3358     CPUState *cs;
3359 
3360     CPU_FOREACH(cs) {
3361         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3362     }
3363 }
3364 
3365 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3366                           void *fdt, int *fdt_start_offset, Error **errp)
3367 {
3368     uint64_t addr;
3369     uint32_t node;
3370 
3371     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3372     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3373                                     &error_abort);
3374     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3375                                                    SPAPR_MEMORY_BLOCK_SIZE);
3376     return 0;
3377 }
3378 
3379 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3380                            bool dedicated_hp_event_source, Error **errp)
3381 {
3382     SpaprDrc *drc;
3383     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3384     int i;
3385     uint64_t addr = addr_start;
3386     bool hotplugged = spapr_drc_hotplugged(dev);
3387     Error *local_err = NULL;
3388 
3389     for (i = 0; i < nr_lmbs; i++) {
3390         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3391                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3392         g_assert(drc);
3393 
3394         spapr_drc_attach(drc, dev, &local_err);
3395         if (local_err) {
3396             while (addr > addr_start) {
3397                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3398                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3399                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3400                 spapr_drc_detach(drc);
3401             }
3402             error_propagate(errp, local_err);
3403             return;
3404         }
3405         if (!hotplugged) {
3406             spapr_drc_reset(drc);
3407         }
3408         addr += SPAPR_MEMORY_BLOCK_SIZE;
3409     }
3410     /* send hotplug notification to the
3411      * guest only in case of hotplugged memory
3412      */
3413     if (hotplugged) {
3414         if (dedicated_hp_event_source) {
3415             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3416                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3417             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3418                                                    nr_lmbs,
3419                                                    spapr_drc_index(drc));
3420         } else {
3421             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3422                                            nr_lmbs);
3423         }
3424     }
3425 }
3426 
3427 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3428                               Error **errp)
3429 {
3430     Error *local_err = NULL;
3431     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3432     PCDIMMDevice *dimm = PC_DIMM(dev);
3433     uint64_t size, addr;
3434 
3435     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3436 
3437     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3438     if (local_err) {
3439         goto out;
3440     }
3441 
3442     addr = object_property_get_uint(OBJECT(dimm),
3443                                     PC_DIMM_ADDR_PROP, &local_err);
3444     if (local_err) {
3445         goto out_unplug;
3446     }
3447 
3448     spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3449                    &local_err);
3450     if (local_err) {
3451         goto out_unplug;
3452     }
3453 
3454     return;
3455 
3456 out_unplug:
3457     pc_dimm_unplug(dimm, MACHINE(ms));
3458 out:
3459     error_propagate(errp, local_err);
3460 }
3461 
3462 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3463                                   Error **errp)
3464 {
3465     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3466     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3467     PCDIMMDevice *dimm = PC_DIMM(dev);
3468     Error *local_err = NULL;
3469     uint64_t size;
3470     Object *memdev;
3471     hwaddr pagesize;
3472 
3473     if (!smc->dr_lmb_enabled) {
3474         error_setg(errp, "Memory hotplug not supported for this machine");
3475         return;
3476     }
3477 
3478     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3479     if (local_err) {
3480         error_propagate(errp, local_err);
3481         return;
3482     }
3483 
3484     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3485         error_setg(errp, "Hotplugged memory size must be a multiple of "
3486                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3487         return;
3488     }
3489 
3490     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3491                                       &error_abort);
3492     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3493     spapr_check_pagesize(spapr, pagesize, &local_err);
3494     if (local_err) {
3495         error_propagate(errp, local_err);
3496         return;
3497     }
3498 
3499     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3500 }
3501 
3502 struct SpaprDimmState {
3503     PCDIMMDevice *dimm;
3504     uint32_t nr_lmbs;
3505     QTAILQ_ENTRY(SpaprDimmState) next;
3506 };
3507 
3508 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3509                                                        PCDIMMDevice *dimm)
3510 {
3511     SpaprDimmState *dimm_state = NULL;
3512 
3513     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3514         if (dimm_state->dimm == dimm) {
3515             break;
3516         }
3517     }
3518     return dimm_state;
3519 }
3520 
3521 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3522                                                       uint32_t nr_lmbs,
3523                                                       PCDIMMDevice *dimm)
3524 {
3525     SpaprDimmState *ds = NULL;
3526 
3527     /*
3528      * If this request is for a DIMM whose removal had failed earlier
3529      * (due to guest's refusal to remove the LMBs), we would have this
3530      * dimm already in the pending_dimm_unplugs list. In that
3531      * case don't add again.
3532      */
3533     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3534     if (!ds) {
3535         ds = g_malloc0(sizeof(SpaprDimmState));
3536         ds->nr_lmbs = nr_lmbs;
3537         ds->dimm = dimm;
3538         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3539     }
3540     return ds;
3541 }
3542 
3543 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3544                                               SpaprDimmState *dimm_state)
3545 {
3546     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3547     g_free(dimm_state);
3548 }
3549 
3550 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3551                                                         PCDIMMDevice *dimm)
3552 {
3553     SpaprDrc *drc;
3554     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3555                                                   &error_abort);
3556     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3557     uint32_t avail_lmbs = 0;
3558     uint64_t addr_start, addr;
3559     int i;
3560 
3561     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3562                                          &error_abort);
3563 
3564     addr = addr_start;
3565     for (i = 0; i < nr_lmbs; i++) {
3566         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3567                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3568         g_assert(drc);
3569         if (drc->dev) {
3570             avail_lmbs++;
3571         }
3572         addr += SPAPR_MEMORY_BLOCK_SIZE;
3573     }
3574 
3575     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3576 }
3577 
3578 /* Callback to be called during DRC release. */
3579 void spapr_lmb_release(DeviceState *dev)
3580 {
3581     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3582     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3583     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3584 
3585     /* This information will get lost if a migration occurs
3586      * during the unplug process. In this case recover it. */
3587     if (ds == NULL) {
3588         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3589         g_assert(ds);
3590         /* The DRC being examined by the caller at least must be counted */
3591         g_assert(ds->nr_lmbs);
3592     }
3593 
3594     if (--ds->nr_lmbs) {
3595         return;
3596     }
3597 
3598     /*
3599      * Now that all the LMBs have been removed by the guest, call the
3600      * unplug handler chain. This can never fail.
3601      */
3602     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3603     object_unparent(OBJECT(dev));
3604 }
3605 
3606 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3607 {
3608     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3609     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3610 
3611     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3612     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3613     spapr_pending_dimm_unplugs_remove(spapr, ds);
3614 }
3615 
3616 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3617                                         DeviceState *dev, Error **errp)
3618 {
3619     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3620     Error *local_err = NULL;
3621     PCDIMMDevice *dimm = PC_DIMM(dev);
3622     uint32_t nr_lmbs;
3623     uint64_t size, addr_start, addr;
3624     int i;
3625     SpaprDrc *drc;
3626 
3627     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3628     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3629 
3630     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3631                                          &local_err);
3632     if (local_err) {
3633         goto out;
3634     }
3635 
3636     /*
3637      * An existing pending dimm state for this DIMM means that there is an
3638      * unplug operation in progress, waiting for the spapr_lmb_release
3639      * callback to complete the job (BQL can't cover that far). In this case,
3640      * bail out to avoid detaching DRCs that were already released.
3641      */
3642     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3643         error_setg(&local_err,
3644                    "Memory unplug already in progress for device %s",
3645                    dev->id);
3646         goto out;
3647     }
3648 
3649     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3650 
3651     addr = addr_start;
3652     for (i = 0; i < nr_lmbs; i++) {
3653         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3654                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3655         g_assert(drc);
3656 
3657         spapr_drc_detach(drc);
3658         addr += SPAPR_MEMORY_BLOCK_SIZE;
3659     }
3660 
3661     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3662                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3663     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3664                                               nr_lmbs, spapr_drc_index(drc));
3665 out:
3666     error_propagate(errp, local_err);
3667 }
3668 
3669 /* Callback to be called during DRC release. */
3670 void spapr_core_release(DeviceState *dev)
3671 {
3672     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3673 
3674     /* Call the unplug handler chain. This can never fail. */
3675     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3676     object_unparent(OBJECT(dev));
3677 }
3678 
3679 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3680 {
3681     MachineState *ms = MACHINE(hotplug_dev);
3682     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3683     CPUCore *cc = CPU_CORE(dev);
3684     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3685 
3686     if (smc->pre_2_10_has_unused_icps) {
3687         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3688         int i;
3689 
3690         for (i = 0; i < cc->nr_threads; i++) {
3691             CPUState *cs = CPU(sc->threads[i]);
3692 
3693             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3694         }
3695     }
3696 
3697     assert(core_slot);
3698     core_slot->cpu = NULL;
3699     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3700 }
3701 
3702 static
3703 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3704                                Error **errp)
3705 {
3706     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3707     int index;
3708     SpaprDrc *drc;
3709     CPUCore *cc = CPU_CORE(dev);
3710 
3711     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3712         error_setg(errp, "Unable to find CPU core with core-id: %d",
3713                    cc->core_id);
3714         return;
3715     }
3716     if (index == 0) {
3717         error_setg(errp, "Boot CPU core may not be unplugged");
3718         return;
3719     }
3720 
3721     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3722                           spapr_vcpu_id(spapr, cc->core_id));
3723     g_assert(drc);
3724 
3725     if (!spapr_drc_unplug_requested(drc)) {
3726         spapr_drc_detach(drc);
3727         spapr_hotplug_req_remove_by_index(drc);
3728     }
3729 }
3730 
3731 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3732                            void *fdt, int *fdt_start_offset, Error **errp)
3733 {
3734     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3735     CPUState *cs = CPU(core->threads[0]);
3736     PowerPCCPU *cpu = POWERPC_CPU(cs);
3737     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3738     int id = spapr_get_vcpu_id(cpu);
3739     char *nodename;
3740     int offset;
3741 
3742     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3743     offset = fdt_add_subnode(fdt, 0, nodename);
3744     g_free(nodename);
3745 
3746     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3747 
3748     *fdt_start_offset = offset;
3749     return 0;
3750 }
3751 
3752 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3753                             Error **errp)
3754 {
3755     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3756     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3757     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3758     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3759     CPUCore *cc = CPU_CORE(dev);
3760     CPUState *cs;
3761     SpaprDrc *drc;
3762     Error *local_err = NULL;
3763     CPUArchId *core_slot;
3764     int index;
3765     bool hotplugged = spapr_drc_hotplugged(dev);
3766     int i;
3767 
3768     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3769     if (!core_slot) {
3770         error_setg(errp, "Unable to find CPU core with core-id: %d",
3771                    cc->core_id);
3772         return;
3773     }
3774     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3775                           spapr_vcpu_id(spapr, cc->core_id));
3776 
3777     g_assert(drc || !mc->has_hotpluggable_cpus);
3778 
3779     if (drc) {
3780         spapr_drc_attach(drc, dev, &local_err);
3781         if (local_err) {
3782             error_propagate(errp, local_err);
3783             return;
3784         }
3785 
3786         if (hotplugged) {
3787             /*
3788              * Send hotplug notification interrupt to the guest only
3789              * in case of hotplugged CPUs.
3790              */
3791             spapr_hotplug_req_add_by_index(drc);
3792         } else {
3793             spapr_drc_reset(drc);
3794         }
3795     }
3796 
3797     core_slot->cpu = OBJECT(dev);
3798 
3799     if (smc->pre_2_10_has_unused_icps) {
3800         for (i = 0; i < cc->nr_threads; i++) {
3801             cs = CPU(core->threads[i]);
3802             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3803         }
3804     }
3805 
3806     /*
3807      * Set compatibility mode to match the boot CPU, which was either set
3808      * by the machine reset code or by CAS.
3809      */
3810     if (hotplugged) {
3811         for (i = 0; i < cc->nr_threads; i++) {
3812             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3813                            &local_err);
3814             if (local_err) {
3815                 error_propagate(errp, local_err);
3816                 return;
3817             }
3818         }
3819     }
3820 }
3821 
3822 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3823                                 Error **errp)
3824 {
3825     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3826     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3827     Error *local_err = NULL;
3828     CPUCore *cc = CPU_CORE(dev);
3829     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3830     const char *type = object_get_typename(OBJECT(dev));
3831     CPUArchId *core_slot;
3832     int index;
3833     unsigned int smp_threads = machine->smp.threads;
3834 
3835     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3836         error_setg(&local_err, "CPU hotplug not supported for this machine");
3837         goto out;
3838     }
3839 
3840     if (strcmp(base_core_type, type)) {
3841         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3842         goto out;
3843     }
3844 
3845     if (cc->core_id % smp_threads) {
3846         error_setg(&local_err, "invalid core id %d", cc->core_id);
3847         goto out;
3848     }
3849 
3850     /*
3851      * In general we should have homogeneous threads-per-core, but old
3852      * (pre hotplug support) machine types allow the last core to have
3853      * reduced threads as a compatibility hack for when we allowed
3854      * total vcpus not a multiple of threads-per-core.
3855      */
3856     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3857         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3858                    cc->nr_threads, smp_threads);
3859         goto out;
3860     }
3861 
3862     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3863     if (!core_slot) {
3864         error_setg(&local_err, "core id %d out of range", cc->core_id);
3865         goto out;
3866     }
3867 
3868     if (core_slot->cpu) {
3869         error_setg(&local_err, "core %d already populated", cc->core_id);
3870         goto out;
3871     }
3872 
3873     numa_cpu_pre_plug(core_slot, dev, &local_err);
3874 
3875 out:
3876     error_propagate(errp, local_err);
3877 }
3878 
3879 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3880                           void *fdt, int *fdt_start_offset, Error **errp)
3881 {
3882     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3883     int intc_phandle;
3884 
3885     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3886     if (intc_phandle <= 0) {
3887         return -1;
3888     }
3889 
3890     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3891         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3892         return -1;
3893     }
3894 
3895     /* generally SLOF creates these, for hotplug it's up to QEMU */
3896     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3897 
3898     return 0;
3899 }
3900 
3901 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3902                                Error **errp)
3903 {
3904     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3905     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3906     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3907     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3908 
3909     if (dev->hotplugged && !smc->dr_phb_enabled) {
3910         error_setg(errp, "PHB hotplug not supported for this machine");
3911         return;
3912     }
3913 
3914     if (sphb->index == (uint32_t)-1) {
3915         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3916         return;
3917     }
3918 
3919     /*
3920      * This will check that sphb->index doesn't exceed the maximum number of
3921      * PHBs for the current machine type.
3922      */
3923     smc->phb_placement(spapr, sphb->index,
3924                        &sphb->buid, &sphb->io_win_addr,
3925                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3926                        windows_supported, sphb->dma_liobn,
3927                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3928                        errp);
3929 }
3930 
3931 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3932                            Error **errp)
3933 {
3934     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3935     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3936     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3937     SpaprDrc *drc;
3938     bool hotplugged = spapr_drc_hotplugged(dev);
3939     Error *local_err = NULL;
3940 
3941     if (!smc->dr_phb_enabled) {
3942         return;
3943     }
3944 
3945     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3946     /* hotplug hooks should check it's enabled before getting this far */
3947     assert(drc);
3948 
3949     spapr_drc_attach(drc, DEVICE(dev), &local_err);
3950     if (local_err) {
3951         error_propagate(errp, local_err);
3952         return;
3953     }
3954 
3955     if (hotplugged) {
3956         spapr_hotplug_req_add_by_index(drc);
3957     } else {
3958         spapr_drc_reset(drc);
3959     }
3960 }
3961 
3962 void spapr_phb_release(DeviceState *dev)
3963 {
3964     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3965 
3966     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3967     object_unparent(OBJECT(dev));
3968 }
3969 
3970 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3971 {
3972     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3973 }
3974 
3975 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3976                                      DeviceState *dev, Error **errp)
3977 {
3978     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3979     SpaprDrc *drc;
3980 
3981     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3982     assert(drc);
3983 
3984     if (!spapr_drc_unplug_requested(drc)) {
3985         spapr_drc_detach(drc);
3986         spapr_hotplug_req_remove_by_index(drc);
3987     }
3988 }
3989 
3990 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3991                                  Error **errp)
3992 {
3993     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3994     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
3995 
3996     if (spapr->tpm_proxy != NULL) {
3997         error_setg(errp, "Only one TPM proxy can be specified for this machine");
3998         return;
3999     }
4000 
4001     spapr->tpm_proxy = tpm_proxy;
4002 }
4003 
4004 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4005 {
4006     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4007 
4008     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4009     object_unparent(OBJECT(dev));
4010     spapr->tpm_proxy = NULL;
4011 }
4012 
4013 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4014                                       DeviceState *dev, Error **errp)
4015 {
4016     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4017         spapr_memory_plug(hotplug_dev, dev, errp);
4018     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4019         spapr_core_plug(hotplug_dev, dev, errp);
4020     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4021         spapr_phb_plug(hotplug_dev, dev, errp);
4022     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4023         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4024     }
4025 }
4026 
4027 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4028                                         DeviceState *dev, Error **errp)
4029 {
4030     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4031         spapr_memory_unplug(hotplug_dev, dev);
4032     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4033         spapr_core_unplug(hotplug_dev, dev);
4034     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4035         spapr_phb_unplug(hotplug_dev, dev);
4036     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4037         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4038     }
4039 }
4040 
4041 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4042                                                 DeviceState *dev, Error **errp)
4043 {
4044     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4045     MachineClass *mc = MACHINE_GET_CLASS(sms);
4046     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4047 
4048     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4049         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4050             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4051         } else {
4052             /* NOTE: this means there is a window after guest reset, prior to
4053              * CAS negotiation, where unplug requests will fail due to the
4054              * capability not being detected yet. This is a bit different than
4055              * the case with PCI unplug, where the events will be queued and
4056              * eventually handled by the guest after boot
4057              */
4058             error_setg(errp, "Memory hot unplug not supported for this guest");
4059         }
4060     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4061         if (!mc->has_hotpluggable_cpus) {
4062             error_setg(errp, "CPU hot unplug not supported on this machine");
4063             return;
4064         }
4065         spapr_core_unplug_request(hotplug_dev, dev, errp);
4066     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4067         if (!smc->dr_phb_enabled) {
4068             error_setg(errp, "PHB hot unplug not supported on this machine");
4069             return;
4070         }
4071         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4072     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4073         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4074     }
4075 }
4076 
4077 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4078                                           DeviceState *dev, Error **errp)
4079 {
4080     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4081         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4082     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4083         spapr_core_pre_plug(hotplug_dev, dev, errp);
4084     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4085         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4086     }
4087 }
4088 
4089 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4090                                                  DeviceState *dev)
4091 {
4092     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4093         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4094         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4095         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4096         return HOTPLUG_HANDLER(machine);
4097     }
4098     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4099         PCIDevice *pcidev = PCI_DEVICE(dev);
4100         PCIBus *root = pci_device_root_bus(pcidev);
4101         SpaprPhbState *phb =
4102             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4103                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4104 
4105         if (phb) {
4106             return HOTPLUG_HANDLER(phb);
4107         }
4108     }
4109     return NULL;
4110 }
4111 
4112 static CpuInstanceProperties
4113 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4114 {
4115     CPUArchId *core_slot;
4116     MachineClass *mc = MACHINE_GET_CLASS(machine);
4117 
4118     /* make sure possible_cpu are intialized */
4119     mc->possible_cpu_arch_ids(machine);
4120     /* get CPU core slot containing thread that matches cpu_index */
4121     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4122     assert(core_slot);
4123     return core_slot->props;
4124 }
4125 
4126 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4127 {
4128     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4129 }
4130 
4131 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4132 {
4133     int i;
4134     unsigned int smp_threads = machine->smp.threads;
4135     unsigned int smp_cpus = machine->smp.cpus;
4136     const char *core_type;
4137     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4138     MachineClass *mc = MACHINE_GET_CLASS(machine);
4139 
4140     if (!mc->has_hotpluggable_cpus) {
4141         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4142     }
4143     if (machine->possible_cpus) {
4144         assert(machine->possible_cpus->len == spapr_max_cores);
4145         return machine->possible_cpus;
4146     }
4147 
4148     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4149     if (!core_type) {
4150         error_report("Unable to find sPAPR CPU Core definition");
4151         exit(1);
4152     }
4153 
4154     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4155                              sizeof(CPUArchId) * spapr_max_cores);
4156     machine->possible_cpus->len = spapr_max_cores;
4157     for (i = 0; i < machine->possible_cpus->len; i++) {
4158         int core_id = i * smp_threads;
4159 
4160         machine->possible_cpus->cpus[i].type = core_type;
4161         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4162         machine->possible_cpus->cpus[i].arch_id = core_id;
4163         machine->possible_cpus->cpus[i].props.has_core_id = true;
4164         machine->possible_cpus->cpus[i].props.core_id = core_id;
4165     }
4166     return machine->possible_cpus;
4167 }
4168 
4169 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4170                                 uint64_t *buid, hwaddr *pio,
4171                                 hwaddr *mmio32, hwaddr *mmio64,
4172                                 unsigned n_dma, uint32_t *liobns,
4173                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4174 {
4175     /*
4176      * New-style PHB window placement.
4177      *
4178      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4179      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4180      * windows.
4181      *
4182      * Some guest kernels can't work with MMIO windows above 1<<46
4183      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4184      *
4185      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4186      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4187      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4188      * 1TiB 64-bit MMIO windows for each PHB.
4189      */
4190     const uint64_t base_buid = 0x800000020000000ULL;
4191     int i;
4192 
4193     /* Sanity check natural alignments */
4194     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4195     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4196     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4197     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4198     /* Sanity check bounds */
4199     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4200                       SPAPR_PCI_MEM32_WIN_SIZE);
4201     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4202                       SPAPR_PCI_MEM64_WIN_SIZE);
4203 
4204     if (index >= SPAPR_MAX_PHBS) {
4205         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4206                    SPAPR_MAX_PHBS - 1);
4207         return;
4208     }
4209 
4210     *buid = base_buid + index;
4211     for (i = 0; i < n_dma; ++i) {
4212         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4213     }
4214 
4215     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4216     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4217     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4218 
4219     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4220     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4221 }
4222 
4223 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4224 {
4225     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4226 
4227     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4228 }
4229 
4230 static void spapr_ics_resend(XICSFabric *dev)
4231 {
4232     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4233 
4234     ics_resend(spapr->ics);
4235 }
4236 
4237 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4238 {
4239     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4240 
4241     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4242 }
4243 
4244 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4245                                  Monitor *mon)
4246 {
4247     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4248 
4249     spapr_irq_print_info(spapr, mon);
4250     monitor_printf(mon, "irqchip: %s\n",
4251                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4252 }
4253 
4254 /*
4255  * This is a XIVE only operation
4256  */
4257 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4258                            uint8_t nvt_blk, uint32_t nvt_idx,
4259                            bool cam_ignore, uint8_t priority,
4260                            uint32_t logic_serv, XiveTCTXMatch *match)
4261 {
4262     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4263     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4264     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4265     int count;
4266 
4267     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4268                            priority, logic_serv, match);
4269     if (count < 0) {
4270         return count;
4271     }
4272 
4273     /*
4274      * When we implement the save and restore of the thread interrupt
4275      * contexts in the enter/exit CPU handlers of the machine and the
4276      * escalations in QEMU, we should be able to handle non dispatched
4277      * vCPUs.
4278      *
4279      * Until this is done, the sPAPR machine should find at least one
4280      * matching context always.
4281      */
4282     if (count == 0) {
4283         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4284                       nvt_blk, nvt_idx);
4285     }
4286 
4287     return count;
4288 }
4289 
4290 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4291 {
4292     return cpu->vcpu_id;
4293 }
4294 
4295 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4296 {
4297     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4298     MachineState *ms = MACHINE(spapr);
4299     int vcpu_id;
4300 
4301     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4302 
4303     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4304         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4305         error_append_hint(errp, "Adjust the number of cpus to %d "
4306                           "or try to raise the number of threads per core\n",
4307                           vcpu_id * ms->smp.threads / spapr->vsmt);
4308         return;
4309     }
4310 
4311     cpu->vcpu_id = vcpu_id;
4312 }
4313 
4314 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4315 {
4316     CPUState *cs;
4317 
4318     CPU_FOREACH(cs) {
4319         PowerPCCPU *cpu = POWERPC_CPU(cs);
4320 
4321         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4322             return cpu;
4323         }
4324     }
4325 
4326     return NULL;
4327 }
4328 
4329 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4330 {
4331     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4332 
4333     /* These are only called by TCG, KVM maintains dispatch state */
4334 
4335     spapr_cpu->prod = false;
4336     if (spapr_cpu->vpa_addr) {
4337         CPUState *cs = CPU(cpu);
4338         uint32_t dispatch;
4339 
4340         dispatch = ldl_be_phys(cs->as,
4341                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4342         dispatch++;
4343         if ((dispatch & 1) != 0) {
4344             qemu_log_mask(LOG_GUEST_ERROR,
4345                           "VPA: incorrect dispatch counter value for "
4346                           "dispatched partition %u, correcting.\n", dispatch);
4347             dispatch++;
4348         }
4349         stl_be_phys(cs->as,
4350                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4351     }
4352 }
4353 
4354 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4355 {
4356     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4357 
4358     if (spapr_cpu->vpa_addr) {
4359         CPUState *cs = CPU(cpu);
4360         uint32_t dispatch;
4361 
4362         dispatch = ldl_be_phys(cs->as,
4363                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4364         dispatch++;
4365         if ((dispatch & 1) != 1) {
4366             qemu_log_mask(LOG_GUEST_ERROR,
4367                           "VPA: incorrect dispatch counter value for "
4368                           "preempted partition %u, correcting.\n", dispatch);
4369             dispatch++;
4370         }
4371         stl_be_phys(cs->as,
4372                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4373     }
4374 }
4375 
4376 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4377 {
4378     MachineClass *mc = MACHINE_CLASS(oc);
4379     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4380     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4381     NMIClass *nc = NMI_CLASS(oc);
4382     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4383     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4384     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4385     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4386     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4387 
4388     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4389     mc->ignore_boot_device_suffixes = true;
4390 
4391     /*
4392      * We set up the default / latest behaviour here.  The class_init
4393      * functions for the specific versioned machine types can override
4394      * these details for backwards compatibility
4395      */
4396     mc->init = spapr_machine_init;
4397     mc->reset = spapr_machine_reset;
4398     mc->block_default_type = IF_SCSI;
4399     mc->max_cpus = 1024;
4400     mc->no_parallel = 1;
4401     mc->default_boot_order = "";
4402     mc->default_ram_size = 512 * MiB;
4403     mc->default_display = "std";
4404     mc->kvm_type = spapr_kvm_type;
4405     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4406     mc->pci_allow_0_address = true;
4407     assert(!mc->get_hotplug_handler);
4408     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4409     hc->pre_plug = spapr_machine_device_pre_plug;
4410     hc->plug = spapr_machine_device_plug;
4411     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4412     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4413     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4414     hc->unplug_request = spapr_machine_device_unplug_request;
4415     hc->unplug = spapr_machine_device_unplug;
4416 
4417     smc->dr_lmb_enabled = true;
4418     smc->update_dt_enabled = true;
4419     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4420     mc->has_hotpluggable_cpus = true;
4421     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4422     fwc->get_dev_path = spapr_get_fw_dev_path;
4423     nc->nmi_monitor_handler = spapr_nmi;
4424     smc->phb_placement = spapr_phb_placement;
4425     vhc->hypercall = emulate_spapr_hypercall;
4426     vhc->hpt_mask = spapr_hpt_mask;
4427     vhc->map_hptes = spapr_map_hptes;
4428     vhc->unmap_hptes = spapr_unmap_hptes;
4429     vhc->hpte_set_c = spapr_hpte_set_c;
4430     vhc->hpte_set_r = spapr_hpte_set_r;
4431     vhc->get_pate = spapr_get_pate;
4432     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4433     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4434     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4435     xic->ics_get = spapr_ics_get;
4436     xic->ics_resend = spapr_ics_resend;
4437     xic->icp_get = spapr_icp_get;
4438     ispc->print_info = spapr_pic_print_info;
4439     /* Force NUMA node memory size to be a multiple of
4440      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4441      * in which LMBs are represented and hot-added
4442      */
4443     mc->numa_mem_align_shift = 28;
4444     mc->numa_mem_supported = true;
4445     mc->auto_enable_numa = true;
4446 
4447     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4448     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4449     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4450     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4451     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4452     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4453     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4454     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4455     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4456     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4457     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_ON;
4458     spapr_caps_add_properties(smc, &error_abort);
4459     smc->irq = &spapr_irq_dual;
4460     smc->dr_phb_enabled = true;
4461     smc->linux_pci_probe = true;
4462     smc->smp_threads_vsmt = true;
4463     smc->nr_xirqs = SPAPR_NR_XIRQS;
4464     xfc->match_nvt = spapr_match_nvt;
4465 }
4466 
4467 static const TypeInfo spapr_machine_info = {
4468     .name          = TYPE_SPAPR_MACHINE,
4469     .parent        = TYPE_MACHINE,
4470     .abstract      = true,
4471     .instance_size = sizeof(SpaprMachineState),
4472     .instance_init = spapr_instance_init,
4473     .instance_finalize = spapr_machine_finalizefn,
4474     .class_size    = sizeof(SpaprMachineClass),
4475     .class_init    = spapr_machine_class_init,
4476     .interfaces = (InterfaceInfo[]) {
4477         { TYPE_FW_PATH_PROVIDER },
4478         { TYPE_NMI },
4479         { TYPE_HOTPLUG_HANDLER },
4480         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4481         { TYPE_XICS_FABRIC },
4482         { TYPE_INTERRUPT_STATS_PROVIDER },
4483         { TYPE_XIVE_FABRIC },
4484         { }
4485     },
4486 };
4487 
4488 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4489     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4490                                                     void *data)      \
4491     {                                                                \
4492         MachineClass *mc = MACHINE_CLASS(oc);                        \
4493         spapr_machine_##suffix##_class_options(mc);                  \
4494         if (latest) {                                                \
4495             mc->alias = "pseries";                                   \
4496             mc->is_default = 1;                                      \
4497         }                                                            \
4498     }                                                                \
4499     static const TypeInfo spapr_machine_##suffix##_info = {          \
4500         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4501         .parent = TYPE_SPAPR_MACHINE,                                \
4502         .class_init = spapr_machine_##suffix##_class_init,           \
4503     };                                                               \
4504     static void spapr_machine_register_##suffix(void)                \
4505     {                                                                \
4506         type_register(&spapr_machine_##suffix##_info);               \
4507     }                                                                \
4508     type_init(spapr_machine_register_##suffix)
4509 
4510 /*
4511  * pseries-5.0
4512  */
4513 static void spapr_machine_5_0_class_options(MachineClass *mc)
4514 {
4515     /* Defaults for the latest behaviour inherited from the base class */
4516 }
4517 
4518 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4519 
4520 /*
4521  * pseries-4.2
4522  */
4523 static void spapr_machine_4_2_class_options(MachineClass *mc)
4524 {
4525     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4526 
4527     spapr_machine_5_0_class_options(mc);
4528     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4529     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4530     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_OFF;
4531 }
4532 
4533 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4534 
4535 /*
4536  * pseries-4.1
4537  */
4538 static void spapr_machine_4_1_class_options(MachineClass *mc)
4539 {
4540     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4541     static GlobalProperty compat[] = {
4542         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4543         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4544     };
4545 
4546     spapr_machine_4_2_class_options(mc);
4547     smc->linux_pci_probe = false;
4548     smc->smp_threads_vsmt = false;
4549     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4550     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4551 }
4552 
4553 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4554 
4555 /*
4556  * pseries-4.0
4557  */
4558 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4559                               uint64_t *buid, hwaddr *pio,
4560                               hwaddr *mmio32, hwaddr *mmio64,
4561                               unsigned n_dma, uint32_t *liobns,
4562                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4563 {
4564     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4565                         nv2gpa, nv2atsd, errp);
4566     *nv2gpa = 0;
4567     *nv2atsd = 0;
4568 }
4569 
4570 static void spapr_machine_4_0_class_options(MachineClass *mc)
4571 {
4572     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4573 
4574     spapr_machine_4_1_class_options(mc);
4575     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4576     smc->phb_placement = phb_placement_4_0;
4577     smc->irq = &spapr_irq_xics;
4578     smc->pre_4_1_migration = true;
4579 }
4580 
4581 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4582 
4583 /*
4584  * pseries-3.1
4585  */
4586 static void spapr_machine_3_1_class_options(MachineClass *mc)
4587 {
4588     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4589 
4590     spapr_machine_4_0_class_options(mc);
4591     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4592 
4593     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4594     smc->update_dt_enabled = false;
4595     smc->dr_phb_enabled = false;
4596     smc->broken_host_serial_model = true;
4597     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4598     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4599     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4600     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4601 }
4602 
4603 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4604 
4605 /*
4606  * pseries-3.0
4607  */
4608 
4609 static void spapr_machine_3_0_class_options(MachineClass *mc)
4610 {
4611     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4612 
4613     spapr_machine_3_1_class_options(mc);
4614     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4615 
4616     smc->legacy_irq_allocation = true;
4617     smc->nr_xirqs = 0x400;
4618     smc->irq = &spapr_irq_xics_legacy;
4619 }
4620 
4621 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4622 
4623 /*
4624  * pseries-2.12
4625  */
4626 static void spapr_machine_2_12_class_options(MachineClass *mc)
4627 {
4628     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4629     static GlobalProperty compat[] = {
4630         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4631         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4632     };
4633 
4634     spapr_machine_3_0_class_options(mc);
4635     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4636     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4637 
4638     /* We depend on kvm_enabled() to choose a default value for the
4639      * hpt-max-page-size capability. Of course we can't do it here
4640      * because this is too early and the HW accelerator isn't initialzed
4641      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4642      */
4643     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4644 }
4645 
4646 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4647 
4648 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4649 {
4650     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4651 
4652     spapr_machine_2_12_class_options(mc);
4653     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4654     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4655     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4656 }
4657 
4658 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4659 
4660 /*
4661  * pseries-2.11
4662  */
4663 
4664 static void spapr_machine_2_11_class_options(MachineClass *mc)
4665 {
4666     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4667 
4668     spapr_machine_2_12_class_options(mc);
4669     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4670     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4671 }
4672 
4673 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4674 
4675 /*
4676  * pseries-2.10
4677  */
4678 
4679 static void spapr_machine_2_10_class_options(MachineClass *mc)
4680 {
4681     spapr_machine_2_11_class_options(mc);
4682     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4683 }
4684 
4685 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4686 
4687 /*
4688  * pseries-2.9
4689  */
4690 
4691 static void spapr_machine_2_9_class_options(MachineClass *mc)
4692 {
4693     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4694     static GlobalProperty compat[] = {
4695         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4696     };
4697 
4698     spapr_machine_2_10_class_options(mc);
4699     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4700     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4701     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4702     smc->pre_2_10_has_unused_icps = true;
4703     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4704 }
4705 
4706 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4707 
4708 /*
4709  * pseries-2.8
4710  */
4711 
4712 static void spapr_machine_2_8_class_options(MachineClass *mc)
4713 {
4714     static GlobalProperty compat[] = {
4715         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4716     };
4717 
4718     spapr_machine_2_9_class_options(mc);
4719     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4720     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4721     mc->numa_mem_align_shift = 23;
4722 }
4723 
4724 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4725 
4726 /*
4727  * pseries-2.7
4728  */
4729 
4730 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4731                               uint64_t *buid, hwaddr *pio,
4732                               hwaddr *mmio32, hwaddr *mmio64,
4733                               unsigned n_dma, uint32_t *liobns,
4734                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4735 {
4736     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4737     const uint64_t base_buid = 0x800000020000000ULL;
4738     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4739     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4740     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4741     const uint32_t max_index = 255;
4742     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4743 
4744     uint64_t ram_top = MACHINE(spapr)->ram_size;
4745     hwaddr phb0_base, phb_base;
4746     int i;
4747 
4748     /* Do we have device memory? */
4749     if (MACHINE(spapr)->maxram_size > ram_top) {
4750         /* Can't just use maxram_size, because there may be an
4751          * alignment gap between normal and device memory regions
4752          */
4753         ram_top = MACHINE(spapr)->device_memory->base +
4754             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4755     }
4756 
4757     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4758 
4759     if (index > max_index) {
4760         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4761                    max_index);
4762         return;
4763     }
4764 
4765     *buid = base_buid + index;
4766     for (i = 0; i < n_dma; ++i) {
4767         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4768     }
4769 
4770     phb_base = phb0_base + index * phb_spacing;
4771     *pio = phb_base + pio_offset;
4772     *mmio32 = phb_base + mmio_offset;
4773     /*
4774      * We don't set the 64-bit MMIO window, relying on the PHB's
4775      * fallback behaviour of automatically splitting a large "32-bit"
4776      * window into contiguous 32-bit and 64-bit windows
4777      */
4778 
4779     *nv2gpa = 0;
4780     *nv2atsd = 0;
4781 }
4782 
4783 static void spapr_machine_2_7_class_options(MachineClass *mc)
4784 {
4785     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4786     static GlobalProperty compat[] = {
4787         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4788         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4789         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4790         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4791     };
4792 
4793     spapr_machine_2_8_class_options(mc);
4794     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4795     mc->default_machine_opts = "modern-hotplug-events=off";
4796     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4797     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4798     smc->phb_placement = phb_placement_2_7;
4799 }
4800 
4801 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4802 
4803 /*
4804  * pseries-2.6
4805  */
4806 
4807 static void spapr_machine_2_6_class_options(MachineClass *mc)
4808 {
4809     static GlobalProperty compat[] = {
4810         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4811     };
4812 
4813     spapr_machine_2_7_class_options(mc);
4814     mc->has_hotpluggable_cpus = false;
4815     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4816     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4817 }
4818 
4819 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4820 
4821 /*
4822  * pseries-2.5
4823  */
4824 
4825 static void spapr_machine_2_5_class_options(MachineClass *mc)
4826 {
4827     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4828     static GlobalProperty compat[] = {
4829         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4830     };
4831 
4832     spapr_machine_2_6_class_options(mc);
4833     smc->use_ohci_by_default = true;
4834     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4835     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4836 }
4837 
4838 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4839 
4840 /*
4841  * pseries-2.4
4842  */
4843 
4844 static void spapr_machine_2_4_class_options(MachineClass *mc)
4845 {
4846     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4847 
4848     spapr_machine_2_5_class_options(mc);
4849     smc->dr_lmb_enabled = false;
4850     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4851 }
4852 
4853 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4854 
4855 /*
4856  * pseries-2.3
4857  */
4858 
4859 static void spapr_machine_2_3_class_options(MachineClass *mc)
4860 {
4861     static GlobalProperty compat[] = {
4862         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4863     };
4864     spapr_machine_2_4_class_options(mc);
4865     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4866     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4867 }
4868 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4869 
4870 /*
4871  * pseries-2.2
4872  */
4873 
4874 static void spapr_machine_2_2_class_options(MachineClass *mc)
4875 {
4876     static GlobalProperty compat[] = {
4877         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4878     };
4879 
4880     spapr_machine_2_3_class_options(mc);
4881     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4882     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4883     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4884 }
4885 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4886 
4887 /*
4888  * pseries-2.1
4889  */
4890 
4891 static void spapr_machine_2_1_class_options(MachineClass *mc)
4892 {
4893     spapr_machine_2_2_class_options(mc);
4894     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4895 }
4896 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4897 
4898 static void spapr_machine_register_types(void)
4899 {
4900     type_register_static(&spapr_machine_info);
4901 }
4902 
4903 type_init(spapr_machine_register_types)
4904