xref: /openbmc/qemu/hw/ppc/spapr.c (revision fd990e86)
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 #include "hw/ppc/spapr_nvdimm.h"
84 
85 #include "monitor/monitor.h"
86 
87 #include <libfdt.h>
88 
89 /* SLOF memory layout:
90  *
91  * SLOF raw image loaded at 0, copies its romfs right below the flat
92  * device-tree, then position SLOF itself 31M below that
93  *
94  * So we set FW_OVERHEAD to 40MB which should account for all of that
95  * and more
96  *
97  * We load our kernel at 4M, leaving space for SLOF initial image
98  */
99 #define FDT_MAX_SIZE            0x100000
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            128UL
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
206 {
207     int index = spapr_get_vcpu_id(cpu);
208     uint32_t associativity[] = {cpu_to_be32(0x5),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(0x0),
212                                 cpu_to_be32(cpu->node_id),
213                                 cpu_to_be32(index)};
214 
215     /* Advertise NUMA via ibm,associativity */
216     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217                           sizeof(associativity));
218 }
219 
220 /* Populate the "ibm,pa-features" property */
221 static void spapr_populate_pa_features(SpaprMachineState *spapr,
222                                        PowerPCCPU *cpu,
223                                        void *fdt, int offset)
224 {
225     uint8_t pa_features_206[] = { 6, 0,
226         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
227     uint8_t pa_features_207[] = { 24, 0,
228         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
229         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
230         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
231         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
232     uint8_t pa_features_300[] = { 66, 0,
233         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
234         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
235         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
236         /* 6: DS207 */
237         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
238         /* 16: Vector */
239         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
240         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
241         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
242         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
243         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
244         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
245         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
246         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
248         /* 42: PM, 44: PC RA, 46: SC vec'd */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
250         /* 48: SIMD, 50: QP BFP, 52: String */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
252         /* 54: DecFP, 56: DecI, 58: SHA */
253         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
254         /* 60: NM atomic, 62: RNG */
255         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
256     };
257     uint8_t *pa_features = NULL;
258     size_t pa_size;
259 
260     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
261         pa_features = pa_features_206;
262         pa_size = sizeof(pa_features_206);
263     }
264     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
265         pa_features = pa_features_207;
266         pa_size = sizeof(pa_features_207);
267     }
268     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
269         pa_features = pa_features_300;
270         pa_size = sizeof(pa_features_300);
271     }
272     if (!pa_features) {
273         return;
274     }
275 
276     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
277         /*
278          * Note: we keep CI large pages off by default because a 64K capable
279          * guest provisioned with large pages might otherwise try to map a qemu
280          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
281          * even if that qemu runs on a 4k host.
282          * We dd this bit back here if we are confident this is not an issue
283          */
284         pa_features[3] |= 0x20;
285     }
286     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
287         pa_features[24] |= 0x80;    /* Transactional memory support */
288     }
289     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
290         /* Workaround for broken kernels that attempt (guest) radix
291          * mode when they can't handle it, if they see the radix bit set
292          * in pa-features. So hide it from them. */
293         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
294     }
295 
296     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
297 }
298 
299 static hwaddr spapr_node0_size(MachineState *machine)
300 {
301     if (machine->numa_state->num_nodes) {
302         int i;
303         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
304             if (machine->numa_state->nodes[i].node_mem) {
305                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
306                            machine->ram_size);
307             }
308         }
309     }
310     return machine->ram_size;
311 }
312 
313 static void add_str(GString *s, const gchar *s1)
314 {
315     g_string_append_len(s, s1, strlen(s1) + 1);
316 }
317 
318 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
319                                        hwaddr size)
320 {
321     uint32_t associativity[] = {
322         cpu_to_be32(0x4), /* length */
323         cpu_to_be32(0x0), cpu_to_be32(0x0),
324         cpu_to_be32(0x0), cpu_to_be32(nodeid)
325     };
326     char mem_name[32];
327     uint64_t mem_reg_property[2];
328     int off;
329 
330     mem_reg_property[0] = cpu_to_be64(start);
331     mem_reg_property[1] = cpu_to_be64(size);
332 
333     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
334     off = fdt_add_subnode(fdt, 0, mem_name);
335     _FDT(off);
336     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
337     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
338                       sizeof(mem_reg_property))));
339     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
340                       sizeof(associativity))));
341     return off;
342 }
343 
344 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
345 {
346     MachineState *machine = MACHINE(spapr);
347     hwaddr mem_start, node_size;
348     int i, nb_nodes = machine->numa_state->num_nodes;
349     NodeInfo *nodes = machine->numa_state->nodes;
350 
351     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
352         if (!nodes[i].node_mem) {
353             continue;
354         }
355         if (mem_start >= machine->ram_size) {
356             node_size = 0;
357         } else {
358             node_size = nodes[i].node_mem;
359             if (node_size > machine->ram_size - mem_start) {
360                 node_size = machine->ram_size - mem_start;
361             }
362         }
363         if (!mem_start) {
364             /* spapr_machine_init() checks for rma_size <= node0_size
365              * already */
366             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
367             mem_start += spapr->rma_size;
368             node_size -= spapr->rma_size;
369         }
370         for ( ; node_size; ) {
371             hwaddr sizetmp = pow2floor(node_size);
372 
373             /* mem_start != 0 here */
374             if (ctzl(mem_start) < ctzl(sizetmp)) {
375                 sizetmp = 1ULL << ctzl(mem_start);
376             }
377 
378             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
379             node_size -= sizetmp;
380             mem_start += sizetmp;
381         }
382     }
383 
384     return 0;
385 }
386 
387 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
388                                   SpaprMachineState *spapr)
389 {
390     MachineState *ms = MACHINE(spapr);
391     PowerPCCPU *cpu = POWERPC_CPU(cs);
392     CPUPPCState *env = &cpu->env;
393     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
394     int index = spapr_get_vcpu_id(cpu);
395     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
396                        0xffffffff, 0xffffffff};
397     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
398         : SPAPR_TIMEBASE_FREQ;
399     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
400     uint32_t page_sizes_prop[64];
401     size_t page_sizes_prop_size;
402     unsigned int smp_threads = ms->smp.threads;
403     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
404     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
405     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
406     SpaprDrc *drc;
407     int drc_index;
408     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
409     int i;
410 
411     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
412     if (drc) {
413         drc_index = spapr_drc_index(drc);
414         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
415     }
416 
417     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
418     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
419 
420     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
421     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
422                            env->dcache_line_size)));
423     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
424                            env->dcache_line_size)));
425     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
426                            env->icache_line_size)));
427     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
428                            env->icache_line_size)));
429 
430     if (pcc->l1_dcache_size) {
431         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
432                                pcc->l1_dcache_size)));
433     } else {
434         warn_report("Unknown L1 dcache size for cpu");
435     }
436     if (pcc->l1_icache_size) {
437         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
438                                pcc->l1_icache_size)));
439     } else {
440         warn_report("Unknown L1 icache size for cpu");
441     }
442 
443     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
444     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
445     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
446     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
447     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
448     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
449 
450     if (env->spr_cb[SPR_PURR].oea_read) {
451         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
452     }
453     if (env->spr_cb[SPR_SPURR].oea_read) {
454         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
455     }
456 
457     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
458         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
459                           segs, sizeof(segs))));
460     }
461 
462     /* Advertise VSX (vector extensions) if available
463      *   1               == VMX / Altivec available
464      *   2               == VSX available
465      *
466      * Only CPUs for which we create core types in spapr_cpu_core.c
467      * are possible, and all of those have VMX */
468     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
469         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
470     } else {
471         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
472     }
473 
474     /* Advertise DFP (Decimal Floating Point) if available
475      *   0 / no property == no DFP
476      *   1               == DFP available */
477     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
478         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
479     }
480 
481     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
482                                                       sizeof(page_sizes_prop));
483     if (page_sizes_prop_size) {
484         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
485                           page_sizes_prop, page_sizes_prop_size)));
486     }
487 
488     spapr_populate_pa_features(spapr, cpu, fdt, offset);
489 
490     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
491                            cs->cpu_index / vcpus_per_socket)));
492 
493     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
494                       pft_size_prop, sizeof(pft_size_prop))));
495 
496     if (ms->numa_state->num_nodes > 1) {
497         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
498     }
499 
500     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
501 
502     if (pcc->radix_page_info) {
503         for (i = 0; i < pcc->radix_page_info->count; i++) {
504             radix_AP_encodings[i] =
505                 cpu_to_be32(pcc->radix_page_info->entries[i]);
506         }
507         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
508                           radix_AP_encodings,
509                           pcc->radix_page_info->count *
510                           sizeof(radix_AP_encodings[0]))));
511     }
512 
513     /*
514      * We set this property to let the guest know that it can use the large
515      * decrementer and its width in bits.
516      */
517     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
518         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
519                               pcc->lrg_decr_bits)));
520 }
521 
522 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
523 {
524     CPUState **rev;
525     CPUState *cs;
526     int n_cpus;
527     int cpus_offset;
528     char *nodename;
529     int i;
530 
531     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
532     _FDT(cpus_offset);
533     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
534     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
535 
536     /*
537      * We walk the CPUs in reverse order to ensure that CPU DT nodes
538      * created by fdt_add_subnode() end up in the right order in FDT
539      * for the guest kernel the enumerate the CPUs correctly.
540      *
541      * The CPU list cannot be traversed in reverse order, so we need
542      * to do extra work.
543      */
544     n_cpus = 0;
545     rev = NULL;
546     CPU_FOREACH(cs) {
547         rev = g_renew(CPUState *, rev, n_cpus + 1);
548         rev[n_cpus++] = cs;
549     }
550 
551     for (i = n_cpus - 1; i >= 0; i--) {
552         CPUState *cs = rev[i];
553         PowerPCCPU *cpu = POWERPC_CPU(cs);
554         int index = spapr_get_vcpu_id(cpu);
555         DeviceClass *dc = DEVICE_GET_CLASS(cs);
556         int offset;
557 
558         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
559             continue;
560         }
561 
562         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
563         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
564         g_free(nodename);
565         _FDT(offset);
566         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
567     }
568 
569     g_free(rev);
570 }
571 
572 static int spapr_rng_populate_dt(void *fdt)
573 {
574     int node;
575     int ret;
576 
577     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
578     if (node <= 0) {
579         return -1;
580     }
581     ret = fdt_setprop_string(fdt, node, "device_type",
582                              "ibm,platform-facilities");
583     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
584     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
585 
586     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
587     if (node <= 0) {
588         return -1;
589     }
590     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
591 
592     return ret ? -1 : 0;
593 }
594 
595 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
596 {
597     MemoryDeviceInfoList *info;
598 
599     for (info = list; info; info = info->next) {
600         MemoryDeviceInfo *value = info->value;
601 
602         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
603             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
604 
605             if (addr >= pcdimm_info->addr &&
606                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
607                 return pcdimm_info->node;
608             }
609         }
610     }
611 
612     return -1;
613 }
614 
615 struct sPAPRDrconfCellV2 {
616      uint32_t seq_lmbs;
617      uint64_t base_addr;
618      uint32_t drc_index;
619      uint32_t aa_index;
620      uint32_t flags;
621 } QEMU_PACKED;
622 
623 typedef struct DrconfCellQueue {
624     struct sPAPRDrconfCellV2 cell;
625     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
626 } DrconfCellQueue;
627 
628 static DrconfCellQueue *
629 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
630                       uint32_t drc_index, uint32_t aa_index,
631                       uint32_t flags)
632 {
633     DrconfCellQueue *elem;
634 
635     elem = g_malloc0(sizeof(*elem));
636     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
637     elem->cell.base_addr = cpu_to_be64(base_addr);
638     elem->cell.drc_index = cpu_to_be32(drc_index);
639     elem->cell.aa_index = cpu_to_be32(aa_index);
640     elem->cell.flags = cpu_to_be32(flags);
641 
642     return elem;
643 }
644 
645 /* ibm,dynamic-memory-v2 */
646 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
647                                    int offset, MemoryDeviceInfoList *dimms)
648 {
649     MachineState *machine = MACHINE(spapr);
650     uint8_t *int_buf, *cur_index;
651     int ret;
652     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
653     uint64_t addr, cur_addr, size;
654     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
655     uint64_t mem_end = machine->device_memory->base +
656                        memory_region_size(&machine->device_memory->mr);
657     uint32_t node, buf_len, nr_entries = 0;
658     SpaprDrc *drc;
659     DrconfCellQueue *elem, *next;
660     MemoryDeviceInfoList *info;
661     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
662         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
663 
664     /* Entry to cover RAM and the gap area */
665     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
666                                  SPAPR_LMB_FLAGS_RESERVED |
667                                  SPAPR_LMB_FLAGS_DRC_INVALID);
668     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
669     nr_entries++;
670 
671     cur_addr = machine->device_memory->base;
672     for (info = dimms; info; info = info->next) {
673         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
674 
675         addr = di->addr;
676         size = di->size;
677         node = di->node;
678 
679         /*
680          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
681          * area is marked hotpluggable in the next iteration for the bigger
682          * chunk including the NVDIMM occupied area.
683          */
684         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
685             continue;
686 
687         /* Entry for hot-pluggable area */
688         if (cur_addr < addr) {
689             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
690             g_assert(drc);
691             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
692                                          cur_addr, spapr_drc_index(drc), -1, 0);
693             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
694             nr_entries++;
695         }
696 
697         /* Entry for DIMM */
698         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
699         g_assert(drc);
700         elem = spapr_get_drconf_cell(size / lmb_size, addr,
701                                      spapr_drc_index(drc), node,
702                                      SPAPR_LMB_FLAGS_ASSIGNED);
703         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
704         nr_entries++;
705         cur_addr = addr + size;
706     }
707 
708     /* Entry for remaining hotpluggable area */
709     if (cur_addr < mem_end) {
710         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
711         g_assert(drc);
712         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
713                                      cur_addr, spapr_drc_index(drc), -1, 0);
714         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
715         nr_entries++;
716     }
717 
718     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
719     int_buf = cur_index = g_malloc0(buf_len);
720     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
721     cur_index += sizeof(nr_entries);
722 
723     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
724         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
725         cur_index += sizeof(elem->cell);
726         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
727         g_free(elem);
728     }
729 
730     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
731     g_free(int_buf);
732     if (ret < 0) {
733         return -1;
734     }
735     return 0;
736 }
737 
738 /* ibm,dynamic-memory */
739 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
740                                    int offset, MemoryDeviceInfoList *dimms)
741 {
742     MachineState *machine = MACHINE(spapr);
743     int i, ret;
744     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
745     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
746     uint32_t nr_lmbs = (machine->device_memory->base +
747                        memory_region_size(&machine->device_memory->mr)) /
748                        lmb_size;
749     uint32_t *int_buf, *cur_index, buf_len;
750 
751     /*
752      * Allocate enough buffer size to fit in ibm,dynamic-memory
753      */
754     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
755     cur_index = int_buf = g_malloc0(buf_len);
756     int_buf[0] = cpu_to_be32(nr_lmbs);
757     cur_index++;
758     for (i = 0; i < nr_lmbs; i++) {
759         uint64_t addr = i * lmb_size;
760         uint32_t *dynamic_memory = cur_index;
761 
762         if (i >= device_lmb_start) {
763             SpaprDrc *drc;
764 
765             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
766             g_assert(drc);
767 
768             dynamic_memory[0] = cpu_to_be32(addr >> 32);
769             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
770             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
771             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
772             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
773             if (memory_region_present(get_system_memory(), addr)) {
774                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
775             } else {
776                 dynamic_memory[5] = cpu_to_be32(0);
777             }
778         } else {
779             /*
780              * LMB information for RMA, boot time RAM and gap b/n RAM and
781              * device memory region -- all these are marked as reserved
782              * and as having no valid DRC.
783              */
784             dynamic_memory[0] = cpu_to_be32(addr >> 32);
785             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
786             dynamic_memory[2] = cpu_to_be32(0);
787             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
788             dynamic_memory[4] = cpu_to_be32(-1);
789             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
790                                             SPAPR_LMB_FLAGS_DRC_INVALID);
791         }
792 
793         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
794     }
795     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
796     g_free(int_buf);
797     if (ret < 0) {
798         return -1;
799     }
800     return 0;
801 }
802 
803 /*
804  * Adds ibm,dynamic-reconfiguration-memory node.
805  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
806  * of this device tree node.
807  */
808 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
809 {
810     MachineState *machine = MACHINE(spapr);
811     int nb_numa_nodes = machine->numa_state->num_nodes;
812     int ret, i, offset;
813     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
814     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
815     uint32_t *int_buf, *cur_index, buf_len;
816     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
817     MemoryDeviceInfoList *dimms = NULL;
818 
819     /*
820      * Don't create the node if there is no device memory
821      */
822     if (machine->ram_size == machine->maxram_size) {
823         return 0;
824     }
825 
826     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
827 
828     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
829                     sizeof(prop_lmb_size));
830     if (ret < 0) {
831         return ret;
832     }
833 
834     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
835     if (ret < 0) {
836         return ret;
837     }
838 
839     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
840     if (ret < 0) {
841         return ret;
842     }
843 
844     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
845     dimms = qmp_memory_device_list();
846     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
847         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
848     } else {
849         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
850     }
851     qapi_free_MemoryDeviceInfoList(dimms);
852 
853     if (ret < 0) {
854         return ret;
855     }
856 
857     /* ibm,associativity-lookup-arrays */
858     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
859     cur_index = int_buf = g_malloc0(buf_len);
860     int_buf[0] = cpu_to_be32(nr_nodes);
861     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
862     cur_index += 2;
863     for (i = 0; i < nr_nodes; i++) {
864         uint32_t associativity[] = {
865             cpu_to_be32(0x0),
866             cpu_to_be32(0x0),
867             cpu_to_be32(0x0),
868             cpu_to_be32(i)
869         };
870         memcpy(cur_index, associativity, sizeof(associativity));
871         cur_index += 4;
872     }
873     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
874             (cur_index - int_buf) * sizeof(uint32_t));
875     g_free(int_buf);
876 
877     return ret;
878 }
879 
880 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
881                                 SpaprOptionVector *ov5_updates)
882 {
883     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
884     int ret = 0, offset;
885 
886     /* Generate ibm,dynamic-reconfiguration-memory node if required */
887     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
888         g_assert(smc->dr_lmb_enabled);
889         ret = spapr_populate_drconf_memory(spapr, fdt);
890         if (ret) {
891             return ret;
892         }
893     }
894 
895     offset = fdt_path_offset(fdt, "/chosen");
896     if (offset < 0) {
897         offset = fdt_add_subnode(fdt, 0, "chosen");
898         if (offset < 0) {
899             return offset;
900         }
901     }
902     return spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
903                                   "ibm,architecture-vec-5");
904 }
905 
906 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
907 {
908     MachineState *ms = MACHINE(spapr);
909     int rtas;
910     GString *hypertas = g_string_sized_new(256);
911     GString *qemu_hypertas = g_string_sized_new(256);
912     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
913     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
914         memory_region_size(&MACHINE(spapr)->device_memory->mr);
915     uint32_t lrdr_capacity[] = {
916         cpu_to_be32(max_device_addr >> 32),
917         cpu_to_be32(max_device_addr & 0xffffffff),
918         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
919         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
920     };
921     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
922     uint32_t maxdomains[] = {
923         cpu_to_be32(4),
924         maxdomain,
925         maxdomain,
926         maxdomain,
927         cpu_to_be32(spapr->gpu_numa_id),
928     };
929 
930     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
931 
932     /* hypertas */
933     add_str(hypertas, "hcall-pft");
934     add_str(hypertas, "hcall-term");
935     add_str(hypertas, "hcall-dabr");
936     add_str(hypertas, "hcall-interrupt");
937     add_str(hypertas, "hcall-tce");
938     add_str(hypertas, "hcall-vio");
939     add_str(hypertas, "hcall-splpar");
940     add_str(hypertas, "hcall-join");
941     add_str(hypertas, "hcall-bulk");
942     add_str(hypertas, "hcall-set-mode");
943     add_str(hypertas, "hcall-sprg0");
944     add_str(hypertas, "hcall-copy");
945     add_str(hypertas, "hcall-debug");
946     add_str(hypertas, "hcall-vphn");
947     add_str(qemu_hypertas, "hcall-memop1");
948 
949     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
950         add_str(hypertas, "hcall-multi-tce");
951     }
952 
953     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
954         add_str(hypertas, "hcall-hpt-resize");
955     }
956 
957     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
958                      hypertas->str, hypertas->len));
959     g_string_free(hypertas, TRUE);
960     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
961                      qemu_hypertas->str, qemu_hypertas->len));
962     g_string_free(qemu_hypertas, TRUE);
963 
964     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
965                      refpoints, sizeof(refpoints)));
966 
967     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
968                      maxdomains, sizeof(maxdomains)));
969 
970     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
971                           RTAS_ERROR_LOG_MAX));
972     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
973                           RTAS_EVENT_SCAN_RATE));
974 
975     g_assert(msi_nonbroken);
976     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
977 
978     /*
979      * According to PAPR, rtas ibm,os-term does not guarantee a return
980      * back to the guest cpu.
981      *
982      * While an additional ibm,extended-os-term property indicates
983      * that rtas call return will always occur. Set this property.
984      */
985     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
986 
987     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
988                      lrdr_capacity, sizeof(lrdr_capacity)));
989 
990     spapr_dt_rtas_tokens(fdt, rtas);
991 }
992 
993 /*
994  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
995  * and the XIVE features that the guest may request and thus the valid
996  * values for bytes 23..26 of option vector 5:
997  */
998 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
999                                           int chosen)
1000 {
1001     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1002 
1003     char val[2 * 4] = {
1004         23, 0x00, /* XICS / XIVE mode */
1005         24, 0x00, /* Hash/Radix, filled in below. */
1006         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1007         26, 0x40, /* Radix options: GTSE == yes. */
1008     };
1009 
1010     if (spapr->irq->xics && spapr->irq->xive) {
1011         val[1] = SPAPR_OV5_XIVE_BOTH;
1012     } else if (spapr->irq->xive) {
1013         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1014     } else {
1015         assert(spapr->irq->xics);
1016         val[1] = SPAPR_OV5_XIVE_LEGACY;
1017     }
1018 
1019     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1020                           first_ppc_cpu->compat_pvr)) {
1021         /*
1022          * If we're in a pre POWER9 compat mode then the guest should
1023          * do hash and use the legacy interrupt mode
1024          */
1025         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1026         val[3] = 0x00; /* Hash */
1027     } else if (kvm_enabled()) {
1028         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1029             val[3] = 0x80; /* OV5_MMU_BOTH */
1030         } else if (kvmppc_has_cap_mmu_radix()) {
1031             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1032         } else {
1033             val[3] = 0x00; /* Hash */
1034         }
1035     } else {
1036         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1037         val[3] = 0xC0;
1038     }
1039     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1040                      val, sizeof(val)));
1041 }
1042 
1043 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1044 {
1045     MachineState *machine = MACHINE(spapr);
1046     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1047     int chosen;
1048     const char *boot_device = machine->boot_order;
1049     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1050     size_t cb = 0;
1051     char *bootlist = get_boot_devices_list(&cb);
1052 
1053     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1054 
1055     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1056         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1057                                 machine->kernel_cmdline));
1058     }
1059     if (spapr->initrd_size) {
1060         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1061                               spapr->initrd_base));
1062         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1063                               spapr->initrd_base + spapr->initrd_size));
1064     }
1065 
1066     if (spapr->kernel_size) {
1067         uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1068                               cpu_to_be64(spapr->kernel_size) };
1069 
1070         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1071                          &kprop, sizeof(kprop)));
1072         if (spapr->kernel_le) {
1073             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1074         }
1075     }
1076     if (boot_menu) {
1077         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1078     }
1079     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1080     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1081     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1082 
1083     if (cb && bootlist) {
1084         int i;
1085 
1086         for (i = 0; i < cb; i++) {
1087             if (bootlist[i] == '\n') {
1088                 bootlist[i] = ' ';
1089             }
1090         }
1091         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1092     }
1093 
1094     if (boot_device && strlen(boot_device)) {
1095         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1096     }
1097 
1098     if (!spapr->has_graphics && stdout_path) {
1099         /*
1100          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1101          * kernel. New platforms should only use the "stdout-path" property. Set
1102          * the new property and continue using older property to remain
1103          * compatible with the existing firmware.
1104          */
1105         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1106         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1107     }
1108 
1109     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1110     if (smc->linux_pci_probe) {
1111         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1112     }
1113 
1114     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1115 
1116     g_free(stdout_path);
1117     g_free(bootlist);
1118 }
1119 
1120 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1121 {
1122     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1123      * KVM to work under pHyp with some guest co-operation */
1124     int hypervisor;
1125     uint8_t hypercall[16];
1126 
1127     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1128     /* indicate KVM hypercall interface */
1129     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1130     if (kvmppc_has_cap_fixup_hcalls()) {
1131         /*
1132          * Older KVM versions with older guest kernels were broken
1133          * with the magic page, don't allow the guest to map it.
1134          */
1135         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1136                                   sizeof(hypercall))) {
1137             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1138                              hypercall, sizeof(hypercall)));
1139         }
1140     }
1141 }
1142 
1143 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1144 {
1145     MachineState *machine = MACHINE(spapr);
1146     MachineClass *mc = MACHINE_GET_CLASS(machine);
1147     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1148     int ret;
1149     void *fdt;
1150     SpaprPhbState *phb;
1151     char *buf;
1152 
1153     fdt = g_malloc0(space);
1154     _FDT((fdt_create_empty_tree(fdt, space)));
1155 
1156     /* Root node */
1157     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1158     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1159     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1160 
1161     /* Guest UUID & Name*/
1162     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1163     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1164     if (qemu_uuid_set) {
1165         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1166     }
1167     g_free(buf);
1168 
1169     if (qemu_get_vm_name()) {
1170         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1171                                 qemu_get_vm_name()));
1172     }
1173 
1174     /* Host Model & Serial Number */
1175     if (spapr->host_model) {
1176         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1177     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1178         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1179         g_free(buf);
1180     }
1181 
1182     if (spapr->host_serial) {
1183         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1184     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1185         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1186         g_free(buf);
1187     }
1188 
1189     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1190     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1191 
1192     /* /interrupt controller */
1193     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1194 
1195     ret = spapr_populate_memory(spapr, fdt);
1196     if (ret < 0) {
1197         error_report("couldn't setup memory nodes in fdt");
1198         exit(1);
1199     }
1200 
1201     /* /vdevice */
1202     spapr_dt_vdevice(spapr->vio_bus, fdt);
1203 
1204     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1205         ret = spapr_rng_populate_dt(fdt);
1206         if (ret < 0) {
1207             error_report("could not set up rng device in the fdt");
1208             exit(1);
1209         }
1210     }
1211 
1212     QLIST_FOREACH(phb, &spapr->phbs, list) {
1213         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1214         if (ret < 0) {
1215             error_report("couldn't setup PCI devices in fdt");
1216             exit(1);
1217         }
1218     }
1219 
1220     /* cpus */
1221     spapr_populate_cpus_dt_node(fdt, spapr);
1222 
1223     if (smc->dr_lmb_enabled) {
1224         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1225     }
1226 
1227     if (mc->has_hotpluggable_cpus) {
1228         int offset = fdt_path_offset(fdt, "/cpus");
1229         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1230         if (ret < 0) {
1231             error_report("Couldn't set up CPU DR device tree properties");
1232             exit(1);
1233         }
1234     }
1235 
1236     /* /event-sources */
1237     spapr_dt_events(spapr, fdt);
1238 
1239     /* /rtas */
1240     spapr_dt_rtas(spapr, fdt);
1241 
1242     /* /chosen */
1243     if (reset) {
1244         spapr_dt_chosen(spapr, fdt);
1245     }
1246 
1247     /* /hypervisor */
1248     if (kvm_enabled()) {
1249         spapr_dt_hypervisor(spapr, fdt);
1250     }
1251 
1252     /* Build memory reserve map */
1253     if (reset) {
1254         if (spapr->kernel_size) {
1255             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1256                                   spapr->kernel_size)));
1257         }
1258         if (spapr->initrd_size) {
1259             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1260                                   spapr->initrd_size)));
1261         }
1262     }
1263 
1264     /* ibm,client-architecture-support updates */
1265     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1266     if (ret < 0) {
1267         error_report("couldn't setup CAS properties fdt");
1268         exit(1);
1269     }
1270 
1271     if (smc->dr_phb_enabled) {
1272         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1273         if (ret < 0) {
1274             error_report("Couldn't set up PHB DR device tree properties");
1275             exit(1);
1276         }
1277     }
1278 
1279     /* NVDIMM devices */
1280     if (mc->nvdimm_supported) {
1281         spapr_dt_persistent_memory(fdt);
1282     }
1283 
1284     return fdt;
1285 }
1286 
1287 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1288 {
1289     SpaprMachineState *spapr = opaque;
1290 
1291     return (addr & 0x0fffffff) + spapr->kernel_addr;
1292 }
1293 
1294 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1295                                     PowerPCCPU *cpu)
1296 {
1297     CPUPPCState *env = &cpu->env;
1298 
1299     /* The TCG path should also be holding the BQL at this point */
1300     g_assert(qemu_mutex_iothread_locked());
1301 
1302     if (msr_pr) {
1303         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1304         env->gpr[3] = H_PRIVILEGE;
1305     } else {
1306         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1307     }
1308 }
1309 
1310 struct LPCRSyncState {
1311     target_ulong value;
1312     target_ulong mask;
1313 };
1314 
1315 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1316 {
1317     struct LPCRSyncState *s = arg.host_ptr;
1318     PowerPCCPU *cpu = POWERPC_CPU(cs);
1319     CPUPPCState *env = &cpu->env;
1320     target_ulong lpcr;
1321 
1322     cpu_synchronize_state(cs);
1323     lpcr = env->spr[SPR_LPCR];
1324     lpcr &= ~s->mask;
1325     lpcr |= s->value;
1326     ppc_store_lpcr(cpu, lpcr);
1327 }
1328 
1329 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1330 {
1331     CPUState *cs;
1332     struct LPCRSyncState s = {
1333         .value = value,
1334         .mask = mask
1335     };
1336     CPU_FOREACH(cs) {
1337         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1338     }
1339 }
1340 
1341 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1342 {
1343     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1344 
1345     /* Copy PATE1:GR into PATE0:HR */
1346     entry->dw0 = spapr->patb_entry & PATE0_HR;
1347     entry->dw1 = spapr->patb_entry;
1348 }
1349 
1350 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1351 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1352 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1353 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1354 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1355 
1356 /*
1357  * Get the fd to access the kernel htab, re-opening it if necessary
1358  */
1359 static int get_htab_fd(SpaprMachineState *spapr)
1360 {
1361     Error *local_err = NULL;
1362 
1363     if (spapr->htab_fd >= 0) {
1364         return spapr->htab_fd;
1365     }
1366 
1367     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1368     if (spapr->htab_fd < 0) {
1369         error_report_err(local_err);
1370     }
1371 
1372     return spapr->htab_fd;
1373 }
1374 
1375 void close_htab_fd(SpaprMachineState *spapr)
1376 {
1377     if (spapr->htab_fd >= 0) {
1378         close(spapr->htab_fd);
1379     }
1380     spapr->htab_fd = -1;
1381 }
1382 
1383 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1384 {
1385     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1386 
1387     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1388 }
1389 
1390 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1391 {
1392     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1393 
1394     assert(kvm_enabled());
1395 
1396     if (!spapr->htab) {
1397         return 0;
1398     }
1399 
1400     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1401 }
1402 
1403 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1404                                                 hwaddr ptex, int n)
1405 {
1406     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1407     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1408 
1409     if (!spapr->htab) {
1410         /*
1411          * HTAB is controlled by KVM. Fetch into temporary buffer
1412          */
1413         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1414         kvmppc_read_hptes(hptes, ptex, n);
1415         return hptes;
1416     }
1417 
1418     /*
1419      * HTAB is controlled by QEMU. Just point to the internally
1420      * accessible PTEG.
1421      */
1422     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1423 }
1424 
1425 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1426                               const ppc_hash_pte64_t *hptes,
1427                               hwaddr ptex, int n)
1428 {
1429     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1430 
1431     if (!spapr->htab) {
1432         g_free((void *)hptes);
1433     }
1434 
1435     /* Nothing to do for qemu managed HPT */
1436 }
1437 
1438 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1439                       uint64_t pte0, uint64_t pte1)
1440 {
1441     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1442     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1443 
1444     if (!spapr->htab) {
1445         kvmppc_write_hpte(ptex, pte0, pte1);
1446     } else {
1447         if (pte0 & HPTE64_V_VALID) {
1448             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1449             /*
1450              * When setting valid, we write PTE1 first. This ensures
1451              * proper synchronization with the reading code in
1452              * ppc_hash64_pteg_search()
1453              */
1454             smp_wmb();
1455             stq_p(spapr->htab + offset, pte0);
1456         } else {
1457             stq_p(spapr->htab + offset, pte0);
1458             /*
1459              * When clearing it we set PTE0 first. This ensures proper
1460              * synchronization with the reading code in
1461              * ppc_hash64_pteg_search()
1462              */
1463             smp_wmb();
1464             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1465         }
1466     }
1467 }
1468 
1469 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1470                              uint64_t pte1)
1471 {
1472     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1473     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1474 
1475     if (!spapr->htab) {
1476         /* There should always be a hash table when this is called */
1477         error_report("spapr_hpte_set_c called with no hash table !");
1478         return;
1479     }
1480 
1481     /* The HW performs a non-atomic byte update */
1482     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1483 }
1484 
1485 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1486                              uint64_t pte1)
1487 {
1488     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1489     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1490 
1491     if (!spapr->htab) {
1492         /* There should always be a hash table when this is called */
1493         error_report("spapr_hpte_set_r called with no hash table !");
1494         return;
1495     }
1496 
1497     /* The HW performs a non-atomic byte update */
1498     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1499 }
1500 
1501 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1502 {
1503     int shift;
1504 
1505     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1506      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1507      * that's much more than is needed for Linux guests */
1508     shift = ctz64(pow2ceil(ramsize)) - 7;
1509     shift = MAX(shift, 18); /* Minimum architected size */
1510     shift = MIN(shift, 46); /* Maximum architected size */
1511     return shift;
1512 }
1513 
1514 void spapr_free_hpt(SpaprMachineState *spapr)
1515 {
1516     g_free(spapr->htab);
1517     spapr->htab = NULL;
1518     spapr->htab_shift = 0;
1519     close_htab_fd(spapr);
1520 }
1521 
1522 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1523                           Error **errp)
1524 {
1525     long rc;
1526 
1527     /* Clean up any HPT info from a previous boot */
1528     spapr_free_hpt(spapr);
1529 
1530     rc = kvmppc_reset_htab(shift);
1531     if (rc < 0) {
1532         /* kernel-side HPT needed, but couldn't allocate one */
1533         error_setg_errno(errp, errno,
1534                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1535                          shift);
1536         /* This is almost certainly fatal, but if the caller really
1537          * wants to carry on with shift == 0, it's welcome to try */
1538     } else if (rc > 0) {
1539         /* kernel-side HPT allocated */
1540         if (rc != shift) {
1541             error_setg(errp,
1542                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1543                        shift, rc);
1544         }
1545 
1546         spapr->htab_shift = shift;
1547         spapr->htab = NULL;
1548     } else {
1549         /* kernel-side HPT not needed, allocate in userspace instead */
1550         size_t size = 1ULL << shift;
1551         int i;
1552 
1553         spapr->htab = qemu_memalign(size, size);
1554         if (!spapr->htab) {
1555             error_setg_errno(errp, errno,
1556                              "Could not allocate HPT of order %d", shift);
1557             return;
1558         }
1559 
1560         memset(spapr->htab, 0, size);
1561         spapr->htab_shift = shift;
1562 
1563         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1564             DIRTY_HPTE(HPTE(spapr->htab, i));
1565         }
1566     }
1567     /* We're setting up a hash table, so that means we're not radix */
1568     spapr->patb_entry = 0;
1569     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1570 }
1571 
1572 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1573 {
1574     int hpt_shift;
1575 
1576     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1577         || (spapr->cas_reboot
1578             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1579         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1580     } else {
1581         uint64_t current_ram_size;
1582 
1583         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1584         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1585     }
1586     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1587 
1588     if (spapr->vrma_adjust) {
1589         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1590                                           spapr->htab_shift);
1591     }
1592 }
1593 
1594 static int spapr_reset_drcs(Object *child, void *opaque)
1595 {
1596     SpaprDrc *drc =
1597         (SpaprDrc *) object_dynamic_cast(child,
1598                                                  TYPE_SPAPR_DR_CONNECTOR);
1599 
1600     if (drc) {
1601         spapr_drc_reset(drc);
1602     }
1603 
1604     return 0;
1605 }
1606 
1607 static void spapr_machine_reset(MachineState *machine)
1608 {
1609     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1610     PowerPCCPU *first_ppc_cpu;
1611     hwaddr fdt_addr;
1612     void *fdt;
1613     int rc;
1614 
1615     kvmppc_svm_off(&error_fatal);
1616     spapr_caps_apply(spapr);
1617 
1618     first_ppc_cpu = POWERPC_CPU(first_cpu);
1619     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1620         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1621                               spapr->max_compat_pvr)) {
1622         /*
1623          * If using KVM with radix mode available, VCPUs can be started
1624          * without a HPT because KVM will start them in radix mode.
1625          * Set the GR bit in PATE so that we know there is no HPT.
1626          */
1627         spapr->patb_entry = PATE1_GR;
1628         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1629     } else {
1630         spapr_setup_hpt_and_vrma(spapr);
1631     }
1632 
1633     qemu_devices_reset();
1634 
1635     /*
1636      * If this reset wasn't generated by CAS, we should reset our
1637      * negotiated options and start from scratch
1638      */
1639     if (!spapr->cas_reboot) {
1640         spapr_ovec_cleanup(spapr->ov5_cas);
1641         spapr->ov5_cas = spapr_ovec_new();
1642 
1643         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1644     }
1645 
1646     /*
1647      * This is fixing some of the default configuration of the XIVE
1648      * devices. To be called after the reset of the machine devices.
1649      */
1650     spapr_irq_reset(spapr, &error_fatal);
1651 
1652     /*
1653      * There is no CAS under qtest. Simulate one to please the code that
1654      * depends on spapr->ov5_cas. This is especially needed to test device
1655      * unplug, so we do that before resetting the DRCs.
1656      */
1657     if (qtest_enabled()) {
1658         spapr_ovec_cleanup(spapr->ov5_cas);
1659         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1660     }
1661 
1662     /* DRC reset may cause a device to be unplugged. This will cause troubles
1663      * if this device is used by another device (eg, a running vhost backend
1664      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1665      * situations, we reset DRCs after all devices have been reset.
1666      */
1667     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1668 
1669     spapr_clear_pending_events(spapr);
1670 
1671     /*
1672      * We place the device tree and RTAS just below either the top of the RMA,
1673      * or just below 2GB, whichever is lower, so that it can be
1674      * processed with 32-bit real mode code if necessary
1675      */
1676     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1677 
1678     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1679 
1680     rc = fdt_pack(fdt);
1681 
1682     /* Should only fail if we've built a corrupted tree */
1683     assert(rc == 0);
1684 
1685     /* Load the fdt */
1686     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1687     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1688     g_free(spapr->fdt_blob);
1689     spapr->fdt_size = fdt_totalsize(fdt);
1690     spapr->fdt_initial_size = spapr->fdt_size;
1691     spapr->fdt_blob = fdt;
1692 
1693     /* Set up the entry state */
1694     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1695     first_ppc_cpu->env.gpr[5] = 0;
1696 
1697     spapr->cas_reboot = false;
1698 
1699     spapr->mc_status = -1;
1700     spapr->guest_machine_check_addr = -1;
1701 
1702     /* Signal all vCPUs waiting on this condition */
1703     qemu_cond_broadcast(&spapr->mc_delivery_cond);
1704 
1705     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1706 }
1707 
1708 static void spapr_create_nvram(SpaprMachineState *spapr)
1709 {
1710     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1711     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1712 
1713     if (dinfo) {
1714         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1715                             &error_fatal);
1716     }
1717 
1718     qdev_init_nofail(dev);
1719 
1720     spapr->nvram = (struct SpaprNvram *)dev;
1721 }
1722 
1723 static void spapr_rtc_create(SpaprMachineState *spapr)
1724 {
1725     object_initialize_child(OBJECT(spapr), "rtc",
1726                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1727                             &error_fatal, NULL);
1728     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1729                               &error_fatal);
1730     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1731                               "date", &error_fatal);
1732 }
1733 
1734 /* Returns whether we want to use VGA or not */
1735 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1736 {
1737     switch (vga_interface_type) {
1738     case VGA_NONE:
1739         return false;
1740     case VGA_DEVICE:
1741         return true;
1742     case VGA_STD:
1743     case VGA_VIRTIO:
1744     case VGA_CIRRUS:
1745         return pci_vga_init(pci_bus) != NULL;
1746     default:
1747         error_setg(errp,
1748                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1749         return false;
1750     }
1751 }
1752 
1753 static int spapr_pre_load(void *opaque)
1754 {
1755     int rc;
1756 
1757     rc = spapr_caps_pre_load(opaque);
1758     if (rc) {
1759         return rc;
1760     }
1761 
1762     return 0;
1763 }
1764 
1765 static int spapr_post_load(void *opaque, int version_id)
1766 {
1767     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1768     int err = 0;
1769 
1770     err = spapr_caps_post_migration(spapr);
1771     if (err) {
1772         return err;
1773     }
1774 
1775     /*
1776      * In earlier versions, there was no separate qdev for the PAPR
1777      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1778      * So when migrating from those versions, poke the incoming offset
1779      * value into the RTC device
1780      */
1781     if (version_id < 3) {
1782         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1783         if (err) {
1784             return err;
1785         }
1786     }
1787 
1788     if (kvm_enabled() && spapr->patb_entry) {
1789         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1790         bool radix = !!(spapr->patb_entry & PATE1_GR);
1791         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1792 
1793         /*
1794          * Update LPCR:HR and UPRT as they may not be set properly in
1795          * the stream
1796          */
1797         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1798                             LPCR_HR | LPCR_UPRT);
1799 
1800         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1801         if (err) {
1802             error_report("Process table config unsupported by the host");
1803             return -EINVAL;
1804         }
1805     }
1806 
1807     err = spapr_irq_post_load(spapr, version_id);
1808     if (err) {
1809         return err;
1810     }
1811 
1812     return err;
1813 }
1814 
1815 static int spapr_pre_save(void *opaque)
1816 {
1817     int rc;
1818 
1819     rc = spapr_caps_pre_save(opaque);
1820     if (rc) {
1821         return rc;
1822     }
1823 
1824     return 0;
1825 }
1826 
1827 static bool version_before_3(void *opaque, int version_id)
1828 {
1829     return version_id < 3;
1830 }
1831 
1832 static bool spapr_pending_events_needed(void *opaque)
1833 {
1834     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1835     return !QTAILQ_EMPTY(&spapr->pending_events);
1836 }
1837 
1838 static const VMStateDescription vmstate_spapr_event_entry = {
1839     .name = "spapr_event_log_entry",
1840     .version_id = 1,
1841     .minimum_version_id = 1,
1842     .fields = (VMStateField[]) {
1843         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1844         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1845         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1846                                      NULL, extended_length),
1847         VMSTATE_END_OF_LIST()
1848     },
1849 };
1850 
1851 static const VMStateDescription vmstate_spapr_pending_events = {
1852     .name = "spapr_pending_events",
1853     .version_id = 1,
1854     .minimum_version_id = 1,
1855     .needed = spapr_pending_events_needed,
1856     .fields = (VMStateField[]) {
1857         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1858                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1859         VMSTATE_END_OF_LIST()
1860     },
1861 };
1862 
1863 static bool spapr_ov5_cas_needed(void *opaque)
1864 {
1865     SpaprMachineState *spapr = opaque;
1866     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1867     bool cas_needed;
1868 
1869     /* Prior to the introduction of SpaprOptionVector, we had two option
1870      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1871      * Both of these options encode machine topology into the device-tree
1872      * in such a way that the now-booted OS should still be able to interact
1873      * appropriately with QEMU regardless of what options were actually
1874      * negotiatied on the source side.
1875      *
1876      * As such, we can avoid migrating the CAS-negotiated options if these
1877      * are the only options available on the current machine/platform.
1878      * Since these are the only options available for pseries-2.7 and
1879      * earlier, this allows us to maintain old->new/new->old migration
1880      * compatibility.
1881      *
1882      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1883      * via default pseries-2.8 machines and explicit command-line parameters.
1884      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1885      * of the actual CAS-negotiated values to continue working properly. For
1886      * example, availability of memory unplug depends on knowing whether
1887      * OV5_HP_EVT was negotiated via CAS.
1888      *
1889      * Thus, for any cases where the set of available CAS-negotiatable
1890      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1891      * include the CAS-negotiated options in the migration stream, unless
1892      * if they affect boot time behaviour only.
1893      */
1894     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1895     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1896     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1897 
1898     /* We need extra information if we have any bits outside the mask
1899      * defined above */
1900     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1901 
1902     spapr_ovec_cleanup(ov5_mask);
1903 
1904     return cas_needed;
1905 }
1906 
1907 static const VMStateDescription vmstate_spapr_ov5_cas = {
1908     .name = "spapr_option_vector_ov5_cas",
1909     .version_id = 1,
1910     .minimum_version_id = 1,
1911     .needed = spapr_ov5_cas_needed,
1912     .fields = (VMStateField[]) {
1913         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1914                                  vmstate_spapr_ovec, SpaprOptionVector),
1915         VMSTATE_END_OF_LIST()
1916     },
1917 };
1918 
1919 static bool spapr_patb_entry_needed(void *opaque)
1920 {
1921     SpaprMachineState *spapr = opaque;
1922 
1923     return !!spapr->patb_entry;
1924 }
1925 
1926 static const VMStateDescription vmstate_spapr_patb_entry = {
1927     .name = "spapr_patb_entry",
1928     .version_id = 1,
1929     .minimum_version_id = 1,
1930     .needed = spapr_patb_entry_needed,
1931     .fields = (VMStateField[]) {
1932         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1933         VMSTATE_END_OF_LIST()
1934     },
1935 };
1936 
1937 static bool spapr_irq_map_needed(void *opaque)
1938 {
1939     SpaprMachineState *spapr = opaque;
1940 
1941     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1942 }
1943 
1944 static const VMStateDescription vmstate_spapr_irq_map = {
1945     .name = "spapr_irq_map",
1946     .version_id = 1,
1947     .minimum_version_id = 1,
1948     .needed = spapr_irq_map_needed,
1949     .fields = (VMStateField[]) {
1950         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1951         VMSTATE_END_OF_LIST()
1952     },
1953 };
1954 
1955 static bool spapr_dtb_needed(void *opaque)
1956 {
1957     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1958 
1959     return smc->update_dt_enabled;
1960 }
1961 
1962 static int spapr_dtb_pre_load(void *opaque)
1963 {
1964     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1965 
1966     g_free(spapr->fdt_blob);
1967     spapr->fdt_blob = NULL;
1968     spapr->fdt_size = 0;
1969 
1970     return 0;
1971 }
1972 
1973 static const VMStateDescription vmstate_spapr_dtb = {
1974     .name = "spapr_dtb",
1975     .version_id = 1,
1976     .minimum_version_id = 1,
1977     .needed = spapr_dtb_needed,
1978     .pre_load = spapr_dtb_pre_load,
1979     .fields = (VMStateField[]) {
1980         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1981         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1982         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1983                                      fdt_size),
1984         VMSTATE_END_OF_LIST()
1985     },
1986 };
1987 
1988 static bool spapr_fwnmi_needed(void *opaque)
1989 {
1990     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1991 
1992     return spapr->guest_machine_check_addr != -1;
1993 }
1994 
1995 static int spapr_fwnmi_pre_save(void *opaque)
1996 {
1997     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1998 
1999     /*
2000      * Check if machine check handling is in progress and print a
2001      * warning message.
2002      */
2003     if (spapr->mc_status != -1) {
2004         warn_report("A machine check is being handled during migration. The"
2005                 "handler may run and log hardware error on the destination");
2006     }
2007 
2008     return 0;
2009 }
2010 
2011 static const VMStateDescription vmstate_spapr_machine_check = {
2012     .name = "spapr_machine_check",
2013     .version_id = 1,
2014     .minimum_version_id = 1,
2015     .needed = spapr_fwnmi_needed,
2016     .pre_save = spapr_fwnmi_pre_save,
2017     .fields = (VMStateField[]) {
2018         VMSTATE_UINT64(guest_machine_check_addr, SpaprMachineState),
2019         VMSTATE_INT32(mc_status, SpaprMachineState),
2020         VMSTATE_END_OF_LIST()
2021     },
2022 };
2023 
2024 static const VMStateDescription vmstate_spapr = {
2025     .name = "spapr",
2026     .version_id = 3,
2027     .minimum_version_id = 1,
2028     .pre_load = spapr_pre_load,
2029     .post_load = spapr_post_load,
2030     .pre_save = spapr_pre_save,
2031     .fields = (VMStateField[]) {
2032         /* used to be @next_irq */
2033         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2034 
2035         /* RTC offset */
2036         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2037 
2038         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2039         VMSTATE_END_OF_LIST()
2040     },
2041     .subsections = (const VMStateDescription*[]) {
2042         &vmstate_spapr_ov5_cas,
2043         &vmstate_spapr_patb_entry,
2044         &vmstate_spapr_pending_events,
2045         &vmstate_spapr_cap_htm,
2046         &vmstate_spapr_cap_vsx,
2047         &vmstate_spapr_cap_dfp,
2048         &vmstate_spapr_cap_cfpc,
2049         &vmstate_spapr_cap_sbbc,
2050         &vmstate_spapr_cap_ibs,
2051         &vmstate_spapr_cap_hpt_maxpagesize,
2052         &vmstate_spapr_irq_map,
2053         &vmstate_spapr_cap_nested_kvm_hv,
2054         &vmstate_spapr_dtb,
2055         &vmstate_spapr_cap_large_decr,
2056         &vmstate_spapr_cap_ccf_assist,
2057         &vmstate_spapr_cap_fwnmi,
2058         &vmstate_spapr_machine_check,
2059         NULL
2060     }
2061 };
2062 
2063 static int htab_save_setup(QEMUFile *f, void *opaque)
2064 {
2065     SpaprMachineState *spapr = opaque;
2066 
2067     /* "Iteration" header */
2068     if (!spapr->htab_shift) {
2069         qemu_put_be32(f, -1);
2070     } else {
2071         qemu_put_be32(f, spapr->htab_shift);
2072     }
2073 
2074     if (spapr->htab) {
2075         spapr->htab_save_index = 0;
2076         spapr->htab_first_pass = true;
2077     } else {
2078         if (spapr->htab_shift) {
2079             assert(kvm_enabled());
2080         }
2081     }
2082 
2083 
2084     return 0;
2085 }
2086 
2087 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2088                             int chunkstart, int n_valid, int n_invalid)
2089 {
2090     qemu_put_be32(f, chunkstart);
2091     qemu_put_be16(f, n_valid);
2092     qemu_put_be16(f, n_invalid);
2093     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2094                     HASH_PTE_SIZE_64 * n_valid);
2095 }
2096 
2097 static void htab_save_end_marker(QEMUFile *f)
2098 {
2099     qemu_put_be32(f, 0);
2100     qemu_put_be16(f, 0);
2101     qemu_put_be16(f, 0);
2102 }
2103 
2104 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2105                                  int64_t max_ns)
2106 {
2107     bool has_timeout = max_ns != -1;
2108     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2109     int index = spapr->htab_save_index;
2110     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2111 
2112     assert(spapr->htab_first_pass);
2113 
2114     do {
2115         int chunkstart;
2116 
2117         /* Consume invalid HPTEs */
2118         while ((index < htabslots)
2119                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2120             CLEAN_HPTE(HPTE(spapr->htab, index));
2121             index++;
2122         }
2123 
2124         /* Consume valid HPTEs */
2125         chunkstart = index;
2126         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2127                && HPTE_VALID(HPTE(spapr->htab, index))) {
2128             CLEAN_HPTE(HPTE(spapr->htab, index));
2129             index++;
2130         }
2131 
2132         if (index > chunkstart) {
2133             int n_valid = index - chunkstart;
2134 
2135             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2136 
2137             if (has_timeout &&
2138                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2139                 break;
2140             }
2141         }
2142     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2143 
2144     if (index >= htabslots) {
2145         assert(index == htabslots);
2146         index = 0;
2147         spapr->htab_first_pass = false;
2148     }
2149     spapr->htab_save_index = index;
2150 }
2151 
2152 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2153                                 int64_t max_ns)
2154 {
2155     bool final = max_ns < 0;
2156     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2157     int examined = 0, sent = 0;
2158     int index = spapr->htab_save_index;
2159     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2160 
2161     assert(!spapr->htab_first_pass);
2162 
2163     do {
2164         int chunkstart, invalidstart;
2165 
2166         /* Consume non-dirty HPTEs */
2167         while ((index < htabslots)
2168                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2169             index++;
2170             examined++;
2171         }
2172 
2173         chunkstart = index;
2174         /* Consume valid dirty HPTEs */
2175         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2176                && HPTE_DIRTY(HPTE(spapr->htab, index))
2177                && HPTE_VALID(HPTE(spapr->htab, index))) {
2178             CLEAN_HPTE(HPTE(spapr->htab, index));
2179             index++;
2180             examined++;
2181         }
2182 
2183         invalidstart = index;
2184         /* Consume invalid dirty HPTEs */
2185         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2186                && HPTE_DIRTY(HPTE(spapr->htab, index))
2187                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2188             CLEAN_HPTE(HPTE(spapr->htab, index));
2189             index++;
2190             examined++;
2191         }
2192 
2193         if (index > chunkstart) {
2194             int n_valid = invalidstart - chunkstart;
2195             int n_invalid = index - invalidstart;
2196 
2197             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2198             sent += index - chunkstart;
2199 
2200             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2201                 break;
2202             }
2203         }
2204 
2205         if (examined >= htabslots) {
2206             break;
2207         }
2208 
2209         if (index >= htabslots) {
2210             assert(index == htabslots);
2211             index = 0;
2212         }
2213     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2214 
2215     if (index >= htabslots) {
2216         assert(index == htabslots);
2217         index = 0;
2218     }
2219 
2220     spapr->htab_save_index = index;
2221 
2222     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2223 }
2224 
2225 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2226 #define MAX_KVM_BUF_SIZE    2048
2227 
2228 static int htab_save_iterate(QEMUFile *f, void *opaque)
2229 {
2230     SpaprMachineState *spapr = opaque;
2231     int fd;
2232     int rc = 0;
2233 
2234     /* Iteration header */
2235     if (!spapr->htab_shift) {
2236         qemu_put_be32(f, -1);
2237         return 1;
2238     } else {
2239         qemu_put_be32(f, 0);
2240     }
2241 
2242     if (!spapr->htab) {
2243         assert(kvm_enabled());
2244 
2245         fd = get_htab_fd(spapr);
2246         if (fd < 0) {
2247             return fd;
2248         }
2249 
2250         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2251         if (rc < 0) {
2252             return rc;
2253         }
2254     } else  if (spapr->htab_first_pass) {
2255         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2256     } else {
2257         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2258     }
2259 
2260     htab_save_end_marker(f);
2261 
2262     return rc;
2263 }
2264 
2265 static int htab_save_complete(QEMUFile *f, void *opaque)
2266 {
2267     SpaprMachineState *spapr = opaque;
2268     int fd;
2269 
2270     /* Iteration header */
2271     if (!spapr->htab_shift) {
2272         qemu_put_be32(f, -1);
2273         return 0;
2274     } else {
2275         qemu_put_be32(f, 0);
2276     }
2277 
2278     if (!spapr->htab) {
2279         int rc;
2280 
2281         assert(kvm_enabled());
2282 
2283         fd = get_htab_fd(spapr);
2284         if (fd < 0) {
2285             return fd;
2286         }
2287 
2288         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2289         if (rc < 0) {
2290             return rc;
2291         }
2292     } else {
2293         if (spapr->htab_first_pass) {
2294             htab_save_first_pass(f, spapr, -1);
2295         }
2296         htab_save_later_pass(f, spapr, -1);
2297     }
2298 
2299     /* End marker */
2300     htab_save_end_marker(f);
2301 
2302     return 0;
2303 }
2304 
2305 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2306 {
2307     SpaprMachineState *spapr = opaque;
2308     uint32_t section_hdr;
2309     int fd = -1;
2310     Error *local_err = NULL;
2311 
2312     if (version_id < 1 || version_id > 1) {
2313         error_report("htab_load() bad version");
2314         return -EINVAL;
2315     }
2316 
2317     section_hdr = qemu_get_be32(f);
2318 
2319     if (section_hdr == -1) {
2320         spapr_free_hpt(spapr);
2321         return 0;
2322     }
2323 
2324     if (section_hdr) {
2325         /* First section gives the htab size */
2326         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2327         if (local_err) {
2328             error_report_err(local_err);
2329             return -EINVAL;
2330         }
2331         return 0;
2332     }
2333 
2334     if (!spapr->htab) {
2335         assert(kvm_enabled());
2336 
2337         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2338         if (fd < 0) {
2339             error_report_err(local_err);
2340             return fd;
2341         }
2342     }
2343 
2344     while (true) {
2345         uint32_t index;
2346         uint16_t n_valid, n_invalid;
2347 
2348         index = qemu_get_be32(f);
2349         n_valid = qemu_get_be16(f);
2350         n_invalid = qemu_get_be16(f);
2351 
2352         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2353             /* End of Stream */
2354             break;
2355         }
2356 
2357         if ((index + n_valid + n_invalid) >
2358             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2359             /* Bad index in stream */
2360             error_report(
2361                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2362                 index, n_valid, n_invalid, spapr->htab_shift);
2363             return -EINVAL;
2364         }
2365 
2366         if (spapr->htab) {
2367             if (n_valid) {
2368                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2369                                 HASH_PTE_SIZE_64 * n_valid);
2370             }
2371             if (n_invalid) {
2372                 memset(HPTE(spapr->htab, index + n_valid), 0,
2373                        HASH_PTE_SIZE_64 * n_invalid);
2374             }
2375         } else {
2376             int rc;
2377 
2378             assert(fd >= 0);
2379 
2380             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2381             if (rc < 0) {
2382                 return rc;
2383             }
2384         }
2385     }
2386 
2387     if (!spapr->htab) {
2388         assert(fd >= 0);
2389         close(fd);
2390     }
2391 
2392     return 0;
2393 }
2394 
2395 static void htab_save_cleanup(void *opaque)
2396 {
2397     SpaprMachineState *spapr = opaque;
2398 
2399     close_htab_fd(spapr);
2400 }
2401 
2402 static SaveVMHandlers savevm_htab_handlers = {
2403     .save_setup = htab_save_setup,
2404     .save_live_iterate = htab_save_iterate,
2405     .save_live_complete_precopy = htab_save_complete,
2406     .save_cleanup = htab_save_cleanup,
2407     .load_state = htab_load,
2408 };
2409 
2410 static void spapr_boot_set(void *opaque, const char *boot_device,
2411                            Error **errp)
2412 {
2413     MachineState *machine = MACHINE(opaque);
2414     machine->boot_order = g_strdup(boot_device);
2415 }
2416 
2417 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2418 {
2419     MachineState *machine = MACHINE(spapr);
2420     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2421     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2422     int i;
2423 
2424     for (i = 0; i < nr_lmbs; i++) {
2425         uint64_t addr;
2426 
2427         addr = i * lmb_size + machine->device_memory->base;
2428         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2429                                addr / lmb_size);
2430     }
2431 }
2432 
2433 /*
2434  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2435  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2436  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2437  */
2438 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2439 {
2440     int i;
2441 
2442     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2443         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2444                    " is not aligned to %" PRIu64 " MiB",
2445                    machine->ram_size,
2446                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2447         return;
2448     }
2449 
2450     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2451         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2452                    " is not aligned to %" PRIu64 " MiB",
2453                    machine->ram_size,
2454                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2455         return;
2456     }
2457 
2458     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2459         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2460             error_setg(errp,
2461                        "Node %d memory size 0x%" PRIx64
2462                        " is not aligned to %" PRIu64 " MiB",
2463                        i, machine->numa_state->nodes[i].node_mem,
2464                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2465             return;
2466         }
2467     }
2468 }
2469 
2470 /* find cpu slot in machine->possible_cpus by core_id */
2471 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2472 {
2473     int index = id / ms->smp.threads;
2474 
2475     if (index >= ms->possible_cpus->len) {
2476         return NULL;
2477     }
2478     if (idx) {
2479         *idx = index;
2480     }
2481     return &ms->possible_cpus->cpus[index];
2482 }
2483 
2484 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2485 {
2486     MachineState *ms = MACHINE(spapr);
2487     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2488     Error *local_err = NULL;
2489     bool vsmt_user = !!spapr->vsmt;
2490     int kvm_smt = kvmppc_smt_threads();
2491     int ret;
2492     unsigned int smp_threads = ms->smp.threads;
2493 
2494     if (!kvm_enabled() && (smp_threads > 1)) {
2495         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2496                      "on a pseries machine");
2497         goto out;
2498     }
2499     if (!is_power_of_2(smp_threads)) {
2500         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2501                      "machine because it must be a power of 2", smp_threads);
2502         goto out;
2503     }
2504 
2505     /* Detemine the VSMT mode to use: */
2506     if (vsmt_user) {
2507         if (spapr->vsmt < smp_threads) {
2508             error_setg(&local_err, "Cannot support VSMT mode %d"
2509                          " because it must be >= threads/core (%d)",
2510                          spapr->vsmt, smp_threads);
2511             goto out;
2512         }
2513         /* In this case, spapr->vsmt has been set by the command line */
2514     } else if (!smc->smp_threads_vsmt) {
2515         /*
2516          * Default VSMT value is tricky, because we need it to be as
2517          * consistent as possible (for migration), but this requires
2518          * changing it for at least some existing cases.  We pick 8 as
2519          * the value that we'd get with KVM on POWER8, the
2520          * overwhelmingly common case in production systems.
2521          */
2522         spapr->vsmt = MAX(8, smp_threads);
2523     } else {
2524         spapr->vsmt = smp_threads;
2525     }
2526 
2527     /* KVM: If necessary, set the SMT mode: */
2528     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2529         ret = kvmppc_set_smt_threads(spapr->vsmt);
2530         if (ret) {
2531             /* Looks like KVM isn't able to change VSMT mode */
2532             error_setg(&local_err,
2533                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2534                        spapr->vsmt, ret);
2535             /* We can live with that if the default one is big enough
2536              * for the number of threads, and a submultiple of the one
2537              * we want.  In this case we'll waste some vcpu ids, but
2538              * behaviour will be correct */
2539             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2540                 warn_report_err(local_err);
2541                 local_err = NULL;
2542                 goto out;
2543             } else {
2544                 if (!vsmt_user) {
2545                     error_append_hint(&local_err,
2546                                       "On PPC, a VM with %d threads/core"
2547                                       " on a host with %d threads/core"
2548                                       " requires the use of VSMT mode %d.\n",
2549                                       smp_threads, kvm_smt, spapr->vsmt);
2550                 }
2551                 kvmppc_error_append_smt_possible_hint(&local_err);
2552                 goto out;
2553             }
2554         }
2555     }
2556     /* else TCG: nothing to do currently */
2557 out:
2558     error_propagate(errp, local_err);
2559 }
2560 
2561 static void spapr_init_cpus(SpaprMachineState *spapr)
2562 {
2563     MachineState *machine = MACHINE(spapr);
2564     MachineClass *mc = MACHINE_GET_CLASS(machine);
2565     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2566     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2567     const CPUArchIdList *possible_cpus;
2568     unsigned int smp_cpus = machine->smp.cpus;
2569     unsigned int smp_threads = machine->smp.threads;
2570     unsigned int max_cpus = machine->smp.max_cpus;
2571     int boot_cores_nr = smp_cpus / smp_threads;
2572     int i;
2573 
2574     possible_cpus = mc->possible_cpu_arch_ids(machine);
2575     if (mc->has_hotpluggable_cpus) {
2576         if (smp_cpus % smp_threads) {
2577             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2578                          smp_cpus, smp_threads);
2579             exit(1);
2580         }
2581         if (max_cpus % smp_threads) {
2582             error_report("max_cpus (%u) must be multiple of threads (%u)",
2583                          max_cpus, smp_threads);
2584             exit(1);
2585         }
2586     } else {
2587         if (max_cpus != smp_cpus) {
2588             error_report("This machine version does not support CPU hotplug");
2589             exit(1);
2590         }
2591         boot_cores_nr = possible_cpus->len;
2592     }
2593 
2594     if (smc->pre_2_10_has_unused_icps) {
2595         int i;
2596 
2597         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2598             /* Dummy entries get deregistered when real ICPState objects
2599              * are registered during CPU core hotplug.
2600              */
2601             pre_2_10_vmstate_register_dummy_icp(i);
2602         }
2603     }
2604 
2605     for (i = 0; i < possible_cpus->len; i++) {
2606         int core_id = i * smp_threads;
2607 
2608         if (mc->has_hotpluggable_cpus) {
2609             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2610                                    spapr_vcpu_id(spapr, core_id));
2611         }
2612 
2613         if (i < boot_cores_nr) {
2614             Object *core  = object_new(type);
2615             int nr_threads = smp_threads;
2616 
2617             /* Handle the partially filled core for older machine types */
2618             if ((i + 1) * smp_threads >= smp_cpus) {
2619                 nr_threads = smp_cpus - i * smp_threads;
2620             }
2621 
2622             object_property_set_int(core, nr_threads, "nr-threads",
2623                                     &error_fatal);
2624             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2625                                     &error_fatal);
2626             object_property_set_bool(core, true, "realized", &error_fatal);
2627 
2628             object_unref(core);
2629         }
2630     }
2631 }
2632 
2633 static PCIHostState *spapr_create_default_phb(void)
2634 {
2635     DeviceState *dev;
2636 
2637     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2638     qdev_prop_set_uint32(dev, "index", 0);
2639     qdev_init_nofail(dev);
2640 
2641     return PCI_HOST_BRIDGE(dev);
2642 }
2643 
2644 /* pSeries LPAR / sPAPR hardware init */
2645 static void spapr_machine_init(MachineState *machine)
2646 {
2647     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2648     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2649     MachineClass *mc = MACHINE_GET_CLASS(machine);
2650     const char *kernel_filename = machine->kernel_filename;
2651     const char *initrd_filename = machine->initrd_filename;
2652     PCIHostState *phb;
2653     int i;
2654     MemoryRegion *sysmem = get_system_memory();
2655     hwaddr node0_size = spapr_node0_size(machine);
2656     long load_limit, fw_size;
2657     char *filename;
2658     Error *resize_hpt_err = NULL;
2659 
2660     msi_nonbroken = true;
2661 
2662     QLIST_INIT(&spapr->phbs);
2663     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2664 
2665     /* Determine capabilities to run with */
2666     spapr_caps_init(spapr);
2667 
2668     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2669     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2670         /*
2671          * If the user explicitly requested a mode we should either
2672          * supply it, or fail completely (which we do below).  But if
2673          * it's not set explicitly, we reset our mode to something
2674          * that works
2675          */
2676         if (resize_hpt_err) {
2677             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2678             error_free(resize_hpt_err);
2679             resize_hpt_err = NULL;
2680         } else {
2681             spapr->resize_hpt = smc->resize_hpt_default;
2682         }
2683     }
2684 
2685     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2686 
2687     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2688         /*
2689          * User requested HPT resize, but this host can't supply it.  Bail out
2690          */
2691         error_report_err(resize_hpt_err);
2692         exit(1);
2693     }
2694 
2695     spapr->rma_size = node0_size;
2696 
2697     /* With KVM, we don't actually know whether KVM supports an
2698      * unbounded RMA (PR KVM) or is limited by the hash table size
2699      * (HV KVM using VRMA), so we always assume the latter
2700      *
2701      * In that case, we also limit the initial allocations for RTAS
2702      * etc... to 256M since we have no way to know what the VRMA size
2703      * is going to be as it depends on the size of the hash table
2704      * which isn't determined yet.
2705      */
2706     if (kvm_enabled()) {
2707         spapr->vrma_adjust = 1;
2708         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2709     }
2710 
2711     /* Actually we don't support unbounded RMA anymore since we added
2712      * proper emulation of HV mode. The max we can get is 16G which
2713      * also happens to be what we configure for PAPR mode so make sure
2714      * we don't do anything bigger than that
2715      */
2716     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2717 
2718     if (spapr->rma_size > node0_size) {
2719         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2720                      spapr->rma_size);
2721         exit(1);
2722     }
2723 
2724     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2725     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2726 
2727     /*
2728      * VSMT must be set in order to be able to compute VCPU ids, ie to
2729      * call spapr_max_server_number() or spapr_vcpu_id().
2730      */
2731     spapr_set_vsmt_mode(spapr, &error_fatal);
2732 
2733     /* Set up Interrupt Controller before we create the VCPUs */
2734     spapr_irq_init(spapr, &error_fatal);
2735 
2736     /* Set up containers for ibm,client-architecture-support negotiated options
2737      */
2738     spapr->ov5 = spapr_ovec_new();
2739     spapr->ov5_cas = spapr_ovec_new();
2740 
2741     if (smc->dr_lmb_enabled) {
2742         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2743         spapr_validate_node_memory(machine, &error_fatal);
2744     }
2745 
2746     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2747 
2748     /* advertise support for dedicated HP event source to guests */
2749     if (spapr->use_hotplug_event_source) {
2750         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2751     }
2752 
2753     /* advertise support for HPT resizing */
2754     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2755         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2756     }
2757 
2758     /* advertise support for ibm,dyamic-memory-v2 */
2759     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2760 
2761     /* advertise XIVE on POWER9 machines */
2762     if (spapr->irq->xive) {
2763         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2764     }
2765 
2766     /* init CPUs */
2767     spapr_init_cpus(spapr);
2768 
2769     /*
2770      * check we don't have a memory-less/cpu-less NUMA node
2771      * Firmware relies on the existing memory/cpu topology to provide the
2772      * NUMA topology to the kernel.
2773      * And the linux kernel needs to know the NUMA topology at start
2774      * to be able to hotplug CPUs later.
2775      */
2776     if (machine->numa_state->num_nodes) {
2777         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2778             /* check for memory-less node */
2779             if (machine->numa_state->nodes[i].node_mem == 0) {
2780                 CPUState *cs;
2781                 int found = 0;
2782                 /* check for cpu-less node */
2783                 CPU_FOREACH(cs) {
2784                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2785                     if (cpu->node_id == i) {
2786                         found = 1;
2787                         break;
2788                     }
2789                 }
2790                 /* memory-less and cpu-less node */
2791                 if (!found) {
2792                     error_report(
2793                        "Memory-less/cpu-less nodes are not supported (node %d)",
2794                                  i);
2795                     exit(1);
2796                 }
2797             }
2798         }
2799 
2800     }
2801 
2802     /*
2803      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2804      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2805      * called from vPHB reset handler so we initialize the counter here.
2806      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2807      * must be equally distant from any other node.
2808      * The final value of spapr->gpu_numa_id is going to be written to
2809      * max-associativity-domains in spapr_build_fdt().
2810      */
2811     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2812 
2813     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2814         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2815                               spapr->max_compat_pvr)) {
2816         /* KVM and TCG always allow GTSE with radix... */
2817         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2818     }
2819     /* ... but not with hash (currently). */
2820 
2821     if (kvm_enabled()) {
2822         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2823         kvmppc_enable_logical_ci_hcalls();
2824         kvmppc_enable_set_mode_hcall();
2825 
2826         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2827         kvmppc_enable_clear_ref_mod_hcalls();
2828 
2829         /* Enable H_PAGE_INIT */
2830         kvmppc_enable_h_page_init();
2831     }
2832 
2833     /* map RAM */
2834     memory_region_add_subregion(sysmem, 0, machine->ram);
2835 
2836     /* always allocate the device memory information */
2837     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2838 
2839     /* initialize hotplug memory address space */
2840     if (machine->ram_size < machine->maxram_size) {
2841         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2842         /*
2843          * Limit the number of hotpluggable memory slots to half the number
2844          * slots that KVM supports, leaving the other half for PCI and other
2845          * devices. However ensure that number of slots doesn't drop below 32.
2846          */
2847         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2848                            SPAPR_MAX_RAM_SLOTS;
2849 
2850         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2851             max_memslots = SPAPR_MAX_RAM_SLOTS;
2852         }
2853         if (machine->ram_slots > max_memslots) {
2854             error_report("Specified number of memory slots %"
2855                          PRIu64" exceeds max supported %d",
2856                          machine->ram_slots, max_memslots);
2857             exit(1);
2858         }
2859 
2860         machine->device_memory->base = ROUND_UP(machine->ram_size,
2861                                                 SPAPR_DEVICE_MEM_ALIGN);
2862         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2863                            "device-memory", device_mem_size);
2864         memory_region_add_subregion(sysmem, machine->device_memory->base,
2865                                     &machine->device_memory->mr);
2866     }
2867 
2868     if (smc->dr_lmb_enabled) {
2869         spapr_create_lmb_dr_connectors(spapr);
2870     }
2871 
2872     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI_MCE) == SPAPR_CAP_ON) {
2873         /* Create the error string for live migration blocker */
2874         error_setg(&spapr->fwnmi_migration_blocker,
2875             "A machine check is being handled during migration. The handler"
2876             "may run and log hardware error on the destination");
2877     }
2878 
2879     if (mc->nvdimm_supported) {
2880         spapr_create_nvdimm_dr_connectors(spapr);
2881     }
2882 
2883     /* Set up RTAS event infrastructure */
2884     spapr_events_init(spapr);
2885 
2886     /* Set up the RTC RTAS interfaces */
2887     spapr_rtc_create(spapr);
2888 
2889     /* Set up VIO bus */
2890     spapr->vio_bus = spapr_vio_bus_init();
2891 
2892     for (i = 0; i < serial_max_hds(); i++) {
2893         if (serial_hd(i)) {
2894             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2895         }
2896     }
2897 
2898     /* We always have at least the nvram device on VIO */
2899     spapr_create_nvram(spapr);
2900 
2901     /*
2902      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2903      * connectors (described in root DT node's "ibm,drc-types" property)
2904      * are pre-initialized here. additional child connectors (such as
2905      * connectors for a PHBs PCI slots) are added as needed during their
2906      * parent's realization.
2907      */
2908     if (smc->dr_phb_enabled) {
2909         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2910             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2911         }
2912     }
2913 
2914     /* Set up PCI */
2915     spapr_pci_rtas_init();
2916 
2917     phb = spapr_create_default_phb();
2918 
2919     for (i = 0; i < nb_nics; i++) {
2920         NICInfo *nd = &nd_table[i];
2921 
2922         if (!nd->model) {
2923             nd->model = g_strdup("spapr-vlan");
2924         }
2925 
2926         if (g_str_equal(nd->model, "spapr-vlan") ||
2927             g_str_equal(nd->model, "ibmveth")) {
2928             spapr_vlan_create(spapr->vio_bus, nd);
2929         } else {
2930             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2931         }
2932     }
2933 
2934     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2935         spapr_vscsi_create(spapr->vio_bus);
2936     }
2937 
2938     /* Graphics */
2939     if (spapr_vga_init(phb->bus, &error_fatal)) {
2940         spapr->has_graphics = true;
2941         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2942     }
2943 
2944     if (machine->usb) {
2945         if (smc->use_ohci_by_default) {
2946             pci_create_simple(phb->bus, -1, "pci-ohci");
2947         } else {
2948             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2949         }
2950 
2951         if (spapr->has_graphics) {
2952             USBBus *usb_bus = usb_bus_find(-1);
2953 
2954             usb_create_simple(usb_bus, "usb-kbd");
2955             usb_create_simple(usb_bus, "usb-mouse");
2956         }
2957     }
2958 
2959     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2960         error_report(
2961             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2962             MIN_RMA_SLOF);
2963         exit(1);
2964     }
2965 
2966     if (kernel_filename) {
2967         uint64_t lowaddr = 0;
2968 
2969         spapr->kernel_size = load_elf(kernel_filename, NULL,
2970                                       translate_kernel_address, spapr,
2971                                       NULL, &lowaddr, NULL, NULL, 1,
2972                                       PPC_ELF_MACHINE, 0, 0);
2973         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2974             spapr->kernel_size = load_elf(kernel_filename, NULL,
2975                                           translate_kernel_address, spapr, NULL,
2976                                           &lowaddr, NULL, NULL, 0,
2977                                           PPC_ELF_MACHINE,
2978                                           0, 0);
2979             spapr->kernel_le = spapr->kernel_size > 0;
2980         }
2981         if (spapr->kernel_size < 0) {
2982             error_report("error loading %s: %s", kernel_filename,
2983                          load_elf_strerror(spapr->kernel_size));
2984             exit(1);
2985         }
2986 
2987         /* load initrd */
2988         if (initrd_filename) {
2989             /* Try to locate the initrd in the gap between the kernel
2990              * and the firmware. Add a bit of space just in case
2991              */
2992             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2993                                   + 0x1ffff) & ~0xffff;
2994             spapr->initrd_size = load_image_targphys(initrd_filename,
2995                                                      spapr->initrd_base,
2996                                                      load_limit
2997                                                      - spapr->initrd_base);
2998             if (spapr->initrd_size < 0) {
2999                 error_report("could not load initial ram disk '%s'",
3000                              initrd_filename);
3001                 exit(1);
3002             }
3003         }
3004     }
3005 
3006     if (bios_name == NULL) {
3007         bios_name = FW_FILE_NAME;
3008     }
3009     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3010     if (!filename) {
3011         error_report("Could not find LPAR firmware '%s'", bios_name);
3012         exit(1);
3013     }
3014     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3015     if (fw_size <= 0) {
3016         error_report("Could not load LPAR firmware '%s'", filename);
3017         exit(1);
3018     }
3019     g_free(filename);
3020 
3021     /* FIXME: Should register things through the MachineState's qdev
3022      * interface, this is a legacy from the sPAPREnvironment structure
3023      * which predated MachineState but had a similar function */
3024     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3025     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3026                          &savevm_htab_handlers, spapr);
3027 
3028     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3029                              &error_fatal);
3030 
3031     qemu_register_boot_set(spapr_boot_set, spapr);
3032 
3033     /*
3034      * Nothing needs to be done to resume a suspended guest because
3035      * suspending does not change the machine state, so no need for
3036      * a ->wakeup method.
3037      */
3038     qemu_register_wakeup_support();
3039 
3040     if (kvm_enabled()) {
3041         /* to stop and start vmclock */
3042         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3043                                          &spapr->tb);
3044 
3045         kvmppc_spapr_enable_inkernel_multitce();
3046     }
3047 
3048     qemu_cond_init(&spapr->mc_delivery_cond);
3049 }
3050 
3051 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3052 {
3053     if (!vm_type) {
3054         return 0;
3055     }
3056 
3057     if (!strcmp(vm_type, "HV")) {
3058         return 1;
3059     }
3060 
3061     if (!strcmp(vm_type, "PR")) {
3062         return 2;
3063     }
3064 
3065     error_report("Unknown kvm-type specified '%s'", vm_type);
3066     exit(1);
3067 }
3068 
3069 /*
3070  * Implementation of an interface to adjust firmware path
3071  * for the bootindex property handling.
3072  */
3073 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3074                                    DeviceState *dev)
3075 {
3076 #define CAST(type, obj, name) \
3077     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3078     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3079     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3080     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3081 
3082     if (d) {
3083         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3084         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3085         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3086 
3087         if (spapr) {
3088             /*
3089              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3090              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3091              * 0x8000 | (target << 8) | (bus << 5) | lun
3092              * (see the "Logical unit addressing format" table in SAM5)
3093              */
3094             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3095             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3096                                    (uint64_t)id << 48);
3097         } else if (virtio) {
3098             /*
3099              * We use SRP luns of the form 01000000 | (target << 8) | lun
3100              * in the top 32 bits of the 64-bit LUN
3101              * Note: the quote above is from SLOF and it is wrong,
3102              * the actual binding is:
3103              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3104              */
3105             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3106             if (d->lun >= 256) {
3107                 /* Use the LUN "flat space addressing method" */
3108                 id |= 0x4000;
3109             }
3110             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3111                                    (uint64_t)id << 32);
3112         } else if (usb) {
3113             /*
3114              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3115              * in the top 32 bits of the 64-bit LUN
3116              */
3117             unsigned usb_port = atoi(usb->port->path);
3118             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3119             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3120                                    (uint64_t)id << 32);
3121         }
3122     }
3123 
3124     /*
3125      * SLOF probes the USB devices, and if it recognizes that the device is a
3126      * storage device, it changes its name to "storage" instead of "usb-host",
3127      * and additionally adds a child node for the SCSI LUN, so the correct
3128      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3129      */
3130     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3131         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3132         if (usb_host_dev_is_scsi_storage(usbdev)) {
3133             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3134         }
3135     }
3136 
3137     if (phb) {
3138         /* Replace "pci" with "pci@800000020000000" */
3139         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3140     }
3141 
3142     if (vsc) {
3143         /* Same logic as virtio above */
3144         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3145         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3146     }
3147 
3148     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3149         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3150         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3151         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3152     }
3153 
3154     return NULL;
3155 }
3156 
3157 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     return g_strdup(spapr->kvm_type);
3162 }
3163 
3164 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3165 {
3166     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3167 
3168     g_free(spapr->kvm_type);
3169     spapr->kvm_type = g_strdup(value);
3170 }
3171 
3172 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3173 {
3174     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3175 
3176     return spapr->use_hotplug_event_source;
3177 }
3178 
3179 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3180                                             Error **errp)
3181 {
3182     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3183 
3184     spapr->use_hotplug_event_source = value;
3185 }
3186 
3187 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3188 {
3189     return true;
3190 }
3191 
3192 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3193 {
3194     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3195 
3196     switch (spapr->resize_hpt) {
3197     case SPAPR_RESIZE_HPT_DEFAULT:
3198         return g_strdup("default");
3199     case SPAPR_RESIZE_HPT_DISABLED:
3200         return g_strdup("disabled");
3201     case SPAPR_RESIZE_HPT_ENABLED:
3202         return g_strdup("enabled");
3203     case SPAPR_RESIZE_HPT_REQUIRED:
3204         return g_strdup("required");
3205     }
3206     g_assert_not_reached();
3207 }
3208 
3209 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3210 {
3211     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3212 
3213     if (strcmp(value, "default") == 0) {
3214         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3215     } else if (strcmp(value, "disabled") == 0) {
3216         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3217     } else if (strcmp(value, "enabled") == 0) {
3218         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3219     } else if (strcmp(value, "required") == 0) {
3220         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3221     } else {
3222         error_setg(errp, "Bad value for \"resize-hpt\" property");
3223     }
3224 }
3225 
3226 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3227                                    void *opaque, Error **errp)
3228 {
3229     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3230 }
3231 
3232 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3233                                    void *opaque, Error **errp)
3234 {
3235     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3236 }
3237 
3238 static void spapr_get_kernel_addr(Object *obj, Visitor *v, const char *name,
3239                                   void *opaque, Error **errp)
3240 {
3241     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3242 }
3243 
3244 static void spapr_set_kernel_addr(Object *obj, Visitor *v, const char *name,
3245                                   void *opaque, Error **errp)
3246 {
3247     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3248 }
3249 
3250 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3251 {
3252     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3253 
3254     if (spapr->irq == &spapr_irq_xics_legacy) {
3255         return g_strdup("legacy");
3256     } else if (spapr->irq == &spapr_irq_xics) {
3257         return g_strdup("xics");
3258     } else if (spapr->irq == &spapr_irq_xive) {
3259         return g_strdup("xive");
3260     } else if (spapr->irq == &spapr_irq_dual) {
3261         return g_strdup("dual");
3262     }
3263     g_assert_not_reached();
3264 }
3265 
3266 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3267 {
3268     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3269 
3270     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3271         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3272         return;
3273     }
3274 
3275     /* The legacy IRQ backend can not be set */
3276     if (strcmp(value, "xics") == 0) {
3277         spapr->irq = &spapr_irq_xics;
3278     } else if (strcmp(value, "xive") == 0) {
3279         spapr->irq = &spapr_irq_xive;
3280     } else if (strcmp(value, "dual") == 0) {
3281         spapr->irq = &spapr_irq_dual;
3282     } else {
3283         error_setg(errp, "Bad value for \"ic-mode\" property");
3284     }
3285 }
3286 
3287 static char *spapr_get_host_model(Object *obj, Error **errp)
3288 {
3289     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3290 
3291     return g_strdup(spapr->host_model);
3292 }
3293 
3294 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3295 {
3296     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3297 
3298     g_free(spapr->host_model);
3299     spapr->host_model = g_strdup(value);
3300 }
3301 
3302 static char *spapr_get_host_serial(Object *obj, Error **errp)
3303 {
3304     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3305 
3306     return g_strdup(spapr->host_serial);
3307 }
3308 
3309 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3310 {
3311     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3312 
3313     g_free(spapr->host_serial);
3314     spapr->host_serial = g_strdup(value);
3315 }
3316 
3317 static void spapr_instance_init(Object *obj)
3318 {
3319     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3320     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3321 
3322     spapr->htab_fd = -1;
3323     spapr->use_hotplug_event_source = true;
3324     object_property_add_str(obj, "kvm-type",
3325                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3326     object_property_set_description(obj, "kvm-type",
3327                                     "Specifies the KVM virtualization mode (HV, PR)",
3328                                     NULL);
3329     object_property_add_bool(obj, "modern-hotplug-events",
3330                             spapr_get_modern_hotplug_events,
3331                             spapr_set_modern_hotplug_events,
3332                             NULL);
3333     object_property_set_description(obj, "modern-hotplug-events",
3334                                     "Use dedicated hotplug event mechanism in"
3335                                     " place of standard EPOW events when possible"
3336                                     " (required for memory hot-unplug support)",
3337                                     NULL);
3338     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3339                             "Maximum permitted CPU compatibility mode",
3340                             &error_fatal);
3341 
3342     object_property_add_str(obj, "resize-hpt",
3343                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3344     object_property_set_description(obj, "resize-hpt",
3345                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3346                                     NULL);
3347     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3348                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3349     object_property_set_description(obj, "vsmt",
3350                                     "Virtual SMT: KVM behaves as if this were"
3351                                     " the host's SMT mode", &error_abort);
3352     object_property_add_bool(obj, "vfio-no-msix-emulation",
3353                              spapr_get_msix_emulation, NULL, NULL);
3354 
3355     object_property_add(obj, "kernel-addr", "uint64", spapr_get_kernel_addr,
3356                         spapr_set_kernel_addr, NULL, &spapr->kernel_addr,
3357                         &error_abort);
3358     object_property_set_description(obj, "kernel-addr",
3359                                     stringify(KERNEL_LOAD_ADDR)
3360                                     " for -kernel is the default",
3361                                     NULL);
3362     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3363     /* The machine class defines the default interrupt controller mode */
3364     spapr->irq = smc->irq;
3365     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3366                             spapr_set_ic_mode, NULL);
3367     object_property_set_description(obj, "ic-mode",
3368                  "Specifies the interrupt controller mode (xics, xive, dual)",
3369                  NULL);
3370 
3371     object_property_add_str(obj, "host-model",
3372         spapr_get_host_model, spapr_set_host_model,
3373         &error_abort);
3374     object_property_set_description(obj, "host-model",
3375         "Host model to advertise in guest device tree", &error_abort);
3376     object_property_add_str(obj, "host-serial",
3377         spapr_get_host_serial, spapr_set_host_serial,
3378         &error_abort);
3379     object_property_set_description(obj, "host-serial",
3380         "Host serial number to advertise in guest device tree", &error_abort);
3381 }
3382 
3383 static void spapr_machine_finalizefn(Object *obj)
3384 {
3385     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3386 
3387     g_free(spapr->kvm_type);
3388 }
3389 
3390 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3391 {
3392     cpu_synchronize_state(cs);
3393     ppc_cpu_do_system_reset(cs);
3394 }
3395 
3396 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3397 {
3398     CPUState *cs;
3399 
3400     CPU_FOREACH(cs) {
3401         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3402     }
3403 }
3404 
3405 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3406                           void *fdt, int *fdt_start_offset, Error **errp)
3407 {
3408     uint64_t addr;
3409     uint32_t node;
3410 
3411     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3412     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3413                                     &error_abort);
3414     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3415                                                    SPAPR_MEMORY_BLOCK_SIZE);
3416     return 0;
3417 }
3418 
3419 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3420                            bool dedicated_hp_event_source, Error **errp)
3421 {
3422     SpaprDrc *drc;
3423     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3424     int i;
3425     uint64_t addr = addr_start;
3426     bool hotplugged = spapr_drc_hotplugged(dev);
3427     Error *local_err = NULL;
3428 
3429     for (i = 0; i < nr_lmbs; i++) {
3430         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3431                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3432         g_assert(drc);
3433 
3434         spapr_drc_attach(drc, dev, &local_err);
3435         if (local_err) {
3436             while (addr > addr_start) {
3437                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3438                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3439                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3440                 spapr_drc_detach(drc);
3441             }
3442             error_propagate(errp, local_err);
3443             return;
3444         }
3445         if (!hotplugged) {
3446             spapr_drc_reset(drc);
3447         }
3448         addr += SPAPR_MEMORY_BLOCK_SIZE;
3449     }
3450     /* send hotplug notification to the
3451      * guest only in case of hotplugged memory
3452      */
3453     if (hotplugged) {
3454         if (dedicated_hp_event_source) {
3455             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3456                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3457             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3458                                                    nr_lmbs,
3459                                                    spapr_drc_index(drc));
3460         } else {
3461             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3462                                            nr_lmbs);
3463         }
3464     }
3465 }
3466 
3467 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3468                               Error **errp)
3469 {
3470     Error *local_err = NULL;
3471     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3472     PCDIMMDevice *dimm = PC_DIMM(dev);
3473     uint64_t size, addr, slot;
3474     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3475 
3476     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3477 
3478     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3479     if (local_err) {
3480         goto out;
3481     }
3482 
3483     if (!is_nvdimm) {
3484         addr = object_property_get_uint(OBJECT(dimm),
3485                                         PC_DIMM_ADDR_PROP, &local_err);
3486         if (local_err) {
3487             goto out_unplug;
3488         }
3489         spapr_add_lmbs(dev, addr, size,
3490                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3491                        &local_err);
3492     } else {
3493         slot = object_property_get_uint(OBJECT(dimm),
3494                                         PC_DIMM_SLOT_PROP, &local_err);
3495         if (local_err) {
3496             goto out_unplug;
3497         }
3498         spapr_add_nvdimm(dev, slot, &local_err);
3499     }
3500 
3501     if (local_err) {
3502         goto out_unplug;
3503     }
3504 
3505     return;
3506 
3507 out_unplug:
3508     pc_dimm_unplug(dimm, MACHINE(ms));
3509 out:
3510     error_propagate(errp, local_err);
3511 }
3512 
3513 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3514                                   Error **errp)
3515 {
3516     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3517     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3518     const MachineClass *mc = MACHINE_CLASS(smc);
3519     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3520     PCDIMMDevice *dimm = PC_DIMM(dev);
3521     Error *local_err = NULL;
3522     uint64_t size;
3523     Object *memdev;
3524     hwaddr pagesize;
3525 
3526     if (!smc->dr_lmb_enabled) {
3527         error_setg(errp, "Memory hotplug not supported for this machine");
3528         return;
3529     }
3530 
3531     if (is_nvdimm && !mc->nvdimm_supported) {
3532         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3533         return;
3534     }
3535 
3536     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3537     if (local_err) {
3538         error_propagate(errp, local_err);
3539         return;
3540     }
3541 
3542     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3543         error_setg(errp, "Hotplugged memory size must be a multiple of "
3544                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3545         return;
3546     } else if (is_nvdimm) {
3547         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3548         if (local_err) {
3549             error_propagate(errp, local_err);
3550             return;
3551         }
3552     }
3553 
3554     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3555                                       &error_abort);
3556     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3557     spapr_check_pagesize(spapr, pagesize, &local_err);
3558     if (local_err) {
3559         error_propagate(errp, local_err);
3560         return;
3561     }
3562 
3563     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3564 }
3565 
3566 struct SpaprDimmState {
3567     PCDIMMDevice *dimm;
3568     uint32_t nr_lmbs;
3569     QTAILQ_ENTRY(SpaprDimmState) next;
3570 };
3571 
3572 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3573                                                        PCDIMMDevice *dimm)
3574 {
3575     SpaprDimmState *dimm_state = NULL;
3576 
3577     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3578         if (dimm_state->dimm == dimm) {
3579             break;
3580         }
3581     }
3582     return dimm_state;
3583 }
3584 
3585 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3586                                                       uint32_t nr_lmbs,
3587                                                       PCDIMMDevice *dimm)
3588 {
3589     SpaprDimmState *ds = NULL;
3590 
3591     /*
3592      * If this request is for a DIMM whose removal had failed earlier
3593      * (due to guest's refusal to remove the LMBs), we would have this
3594      * dimm already in the pending_dimm_unplugs list. In that
3595      * case don't add again.
3596      */
3597     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3598     if (!ds) {
3599         ds = g_malloc0(sizeof(SpaprDimmState));
3600         ds->nr_lmbs = nr_lmbs;
3601         ds->dimm = dimm;
3602         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3603     }
3604     return ds;
3605 }
3606 
3607 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3608                                               SpaprDimmState *dimm_state)
3609 {
3610     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3611     g_free(dimm_state);
3612 }
3613 
3614 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3615                                                         PCDIMMDevice *dimm)
3616 {
3617     SpaprDrc *drc;
3618     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3619                                                   &error_abort);
3620     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3621     uint32_t avail_lmbs = 0;
3622     uint64_t addr_start, addr;
3623     int i;
3624 
3625     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3626                                          &error_abort);
3627 
3628     addr = addr_start;
3629     for (i = 0; i < nr_lmbs; i++) {
3630         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3631                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3632         g_assert(drc);
3633         if (drc->dev) {
3634             avail_lmbs++;
3635         }
3636         addr += SPAPR_MEMORY_BLOCK_SIZE;
3637     }
3638 
3639     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3640 }
3641 
3642 /* Callback to be called during DRC release. */
3643 void spapr_lmb_release(DeviceState *dev)
3644 {
3645     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3646     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3647     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3648 
3649     /* This information will get lost if a migration occurs
3650      * during the unplug process. In this case recover it. */
3651     if (ds == NULL) {
3652         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3653         g_assert(ds);
3654         /* The DRC being examined by the caller at least must be counted */
3655         g_assert(ds->nr_lmbs);
3656     }
3657 
3658     if (--ds->nr_lmbs) {
3659         return;
3660     }
3661 
3662     /*
3663      * Now that all the LMBs have been removed by the guest, call the
3664      * unplug handler chain. This can never fail.
3665      */
3666     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3667     object_unparent(OBJECT(dev));
3668 }
3669 
3670 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3671 {
3672     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3673     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3674 
3675     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3676     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3677     spapr_pending_dimm_unplugs_remove(spapr, ds);
3678 }
3679 
3680 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3681                                         DeviceState *dev, Error **errp)
3682 {
3683     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3684     Error *local_err = NULL;
3685     PCDIMMDevice *dimm = PC_DIMM(dev);
3686     uint32_t nr_lmbs;
3687     uint64_t size, addr_start, addr;
3688     int i;
3689     SpaprDrc *drc;
3690 
3691     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3692         error_setg(&local_err,
3693                    "nvdimm device hot unplug is not supported yet.");
3694         goto out;
3695     }
3696 
3697     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3698     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3699 
3700     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3701                                          &local_err);
3702     if (local_err) {
3703         goto out;
3704     }
3705 
3706     /*
3707      * An existing pending dimm state for this DIMM means that there is an
3708      * unplug operation in progress, waiting for the spapr_lmb_release
3709      * callback to complete the job (BQL can't cover that far). In this case,
3710      * bail out to avoid detaching DRCs that were already released.
3711      */
3712     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3713         error_setg(&local_err,
3714                    "Memory unplug already in progress for device %s",
3715                    dev->id);
3716         goto out;
3717     }
3718 
3719     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3720 
3721     addr = addr_start;
3722     for (i = 0; i < nr_lmbs; i++) {
3723         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3724                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3725         g_assert(drc);
3726 
3727         spapr_drc_detach(drc);
3728         addr += SPAPR_MEMORY_BLOCK_SIZE;
3729     }
3730 
3731     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3732                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3733     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3734                                               nr_lmbs, spapr_drc_index(drc));
3735 out:
3736     error_propagate(errp, local_err);
3737 }
3738 
3739 /* Callback to be called during DRC release. */
3740 void spapr_core_release(DeviceState *dev)
3741 {
3742     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3743 
3744     /* Call the unplug handler chain. This can never fail. */
3745     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3746     object_unparent(OBJECT(dev));
3747 }
3748 
3749 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3750 {
3751     MachineState *ms = MACHINE(hotplug_dev);
3752     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3753     CPUCore *cc = CPU_CORE(dev);
3754     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3755 
3756     if (smc->pre_2_10_has_unused_icps) {
3757         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3758         int i;
3759 
3760         for (i = 0; i < cc->nr_threads; i++) {
3761             CPUState *cs = CPU(sc->threads[i]);
3762 
3763             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3764         }
3765     }
3766 
3767     assert(core_slot);
3768     core_slot->cpu = NULL;
3769     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3770 }
3771 
3772 static
3773 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3774                                Error **errp)
3775 {
3776     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3777     int index;
3778     SpaprDrc *drc;
3779     CPUCore *cc = CPU_CORE(dev);
3780 
3781     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3782         error_setg(errp, "Unable to find CPU core with core-id: %d",
3783                    cc->core_id);
3784         return;
3785     }
3786     if (index == 0) {
3787         error_setg(errp, "Boot CPU core may not be unplugged");
3788         return;
3789     }
3790 
3791     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3792                           spapr_vcpu_id(spapr, cc->core_id));
3793     g_assert(drc);
3794 
3795     if (!spapr_drc_unplug_requested(drc)) {
3796         spapr_drc_detach(drc);
3797         spapr_hotplug_req_remove_by_index(drc);
3798     }
3799 }
3800 
3801 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3802                            void *fdt, int *fdt_start_offset, Error **errp)
3803 {
3804     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3805     CPUState *cs = CPU(core->threads[0]);
3806     PowerPCCPU *cpu = POWERPC_CPU(cs);
3807     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3808     int id = spapr_get_vcpu_id(cpu);
3809     char *nodename;
3810     int offset;
3811 
3812     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3813     offset = fdt_add_subnode(fdt, 0, nodename);
3814     g_free(nodename);
3815 
3816     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3817 
3818     *fdt_start_offset = offset;
3819     return 0;
3820 }
3821 
3822 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3823                             Error **errp)
3824 {
3825     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3826     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3827     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3828     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3829     CPUCore *cc = CPU_CORE(dev);
3830     CPUState *cs;
3831     SpaprDrc *drc;
3832     Error *local_err = NULL;
3833     CPUArchId *core_slot;
3834     int index;
3835     bool hotplugged = spapr_drc_hotplugged(dev);
3836     int i;
3837 
3838     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3839     if (!core_slot) {
3840         error_setg(errp, "Unable to find CPU core with core-id: %d",
3841                    cc->core_id);
3842         return;
3843     }
3844     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3845                           spapr_vcpu_id(spapr, cc->core_id));
3846 
3847     g_assert(drc || !mc->has_hotpluggable_cpus);
3848 
3849     if (drc) {
3850         spapr_drc_attach(drc, dev, &local_err);
3851         if (local_err) {
3852             error_propagate(errp, local_err);
3853             return;
3854         }
3855 
3856         if (hotplugged) {
3857             /*
3858              * Send hotplug notification interrupt to the guest only
3859              * in case of hotplugged CPUs.
3860              */
3861             spapr_hotplug_req_add_by_index(drc);
3862         } else {
3863             spapr_drc_reset(drc);
3864         }
3865     }
3866 
3867     core_slot->cpu = OBJECT(dev);
3868 
3869     if (smc->pre_2_10_has_unused_icps) {
3870         for (i = 0; i < cc->nr_threads; i++) {
3871             cs = CPU(core->threads[i]);
3872             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3873         }
3874     }
3875 
3876     /*
3877      * Set compatibility mode to match the boot CPU, which was either set
3878      * by the machine reset code or by CAS.
3879      */
3880     if (hotplugged) {
3881         for (i = 0; i < cc->nr_threads; i++) {
3882             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3883                            &local_err);
3884             if (local_err) {
3885                 error_propagate(errp, local_err);
3886                 return;
3887             }
3888         }
3889     }
3890 }
3891 
3892 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3893                                 Error **errp)
3894 {
3895     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3896     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3897     Error *local_err = NULL;
3898     CPUCore *cc = CPU_CORE(dev);
3899     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3900     const char *type = object_get_typename(OBJECT(dev));
3901     CPUArchId *core_slot;
3902     int index;
3903     unsigned int smp_threads = machine->smp.threads;
3904 
3905     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3906         error_setg(&local_err, "CPU hotplug not supported for this machine");
3907         goto out;
3908     }
3909 
3910     if (strcmp(base_core_type, type)) {
3911         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3912         goto out;
3913     }
3914 
3915     if (cc->core_id % smp_threads) {
3916         error_setg(&local_err, "invalid core id %d", cc->core_id);
3917         goto out;
3918     }
3919 
3920     /*
3921      * In general we should have homogeneous threads-per-core, but old
3922      * (pre hotplug support) machine types allow the last core to have
3923      * reduced threads as a compatibility hack for when we allowed
3924      * total vcpus not a multiple of threads-per-core.
3925      */
3926     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3927         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3928                    cc->nr_threads, smp_threads);
3929         goto out;
3930     }
3931 
3932     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3933     if (!core_slot) {
3934         error_setg(&local_err, "core id %d out of range", cc->core_id);
3935         goto out;
3936     }
3937 
3938     if (core_slot->cpu) {
3939         error_setg(&local_err, "core %d already populated", cc->core_id);
3940         goto out;
3941     }
3942 
3943     numa_cpu_pre_plug(core_slot, dev, &local_err);
3944 
3945 out:
3946     error_propagate(errp, local_err);
3947 }
3948 
3949 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3950                           void *fdt, int *fdt_start_offset, Error **errp)
3951 {
3952     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3953     int intc_phandle;
3954 
3955     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3956     if (intc_phandle <= 0) {
3957         return -1;
3958     }
3959 
3960     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3961         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3962         return -1;
3963     }
3964 
3965     /* generally SLOF creates these, for hotplug it's up to QEMU */
3966     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3967 
3968     return 0;
3969 }
3970 
3971 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3972                                Error **errp)
3973 {
3974     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3975     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3976     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3977     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3978 
3979     if (dev->hotplugged && !smc->dr_phb_enabled) {
3980         error_setg(errp, "PHB hotplug not supported for this machine");
3981         return;
3982     }
3983 
3984     if (sphb->index == (uint32_t)-1) {
3985         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3986         return;
3987     }
3988 
3989     /*
3990      * This will check that sphb->index doesn't exceed the maximum number of
3991      * PHBs for the current machine type.
3992      */
3993     smc->phb_placement(spapr, sphb->index,
3994                        &sphb->buid, &sphb->io_win_addr,
3995                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3996                        windows_supported, sphb->dma_liobn,
3997                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3998                        errp);
3999 }
4000 
4001 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4002                            Error **errp)
4003 {
4004     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4005     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4006     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4007     SpaprDrc *drc;
4008     bool hotplugged = spapr_drc_hotplugged(dev);
4009     Error *local_err = NULL;
4010 
4011     if (!smc->dr_phb_enabled) {
4012         return;
4013     }
4014 
4015     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4016     /* hotplug hooks should check it's enabled before getting this far */
4017     assert(drc);
4018 
4019     spapr_drc_attach(drc, DEVICE(dev), &local_err);
4020     if (local_err) {
4021         error_propagate(errp, local_err);
4022         return;
4023     }
4024 
4025     if (hotplugged) {
4026         spapr_hotplug_req_add_by_index(drc);
4027     } else {
4028         spapr_drc_reset(drc);
4029     }
4030 }
4031 
4032 void spapr_phb_release(DeviceState *dev)
4033 {
4034     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4035 
4036     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4037     object_unparent(OBJECT(dev));
4038 }
4039 
4040 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4041 {
4042     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4043 }
4044 
4045 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4046                                      DeviceState *dev, Error **errp)
4047 {
4048     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4049     SpaprDrc *drc;
4050 
4051     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4052     assert(drc);
4053 
4054     if (!spapr_drc_unplug_requested(drc)) {
4055         spapr_drc_detach(drc);
4056         spapr_hotplug_req_remove_by_index(drc);
4057     }
4058 }
4059 
4060 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4061                                  Error **errp)
4062 {
4063     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4064     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4065 
4066     if (spapr->tpm_proxy != NULL) {
4067         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4068         return;
4069     }
4070 
4071     spapr->tpm_proxy = tpm_proxy;
4072 }
4073 
4074 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4075 {
4076     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4077 
4078     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4079     object_unparent(OBJECT(dev));
4080     spapr->tpm_proxy = NULL;
4081 }
4082 
4083 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4084                                       DeviceState *dev, Error **errp)
4085 {
4086     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4087         spapr_memory_plug(hotplug_dev, dev, errp);
4088     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4089         spapr_core_plug(hotplug_dev, dev, errp);
4090     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4091         spapr_phb_plug(hotplug_dev, dev, errp);
4092     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4093         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4094     }
4095 }
4096 
4097 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4098                                         DeviceState *dev, Error **errp)
4099 {
4100     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4101         spapr_memory_unplug(hotplug_dev, dev);
4102     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4103         spapr_core_unplug(hotplug_dev, dev);
4104     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4105         spapr_phb_unplug(hotplug_dev, dev);
4106     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4107         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4108     }
4109 }
4110 
4111 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4112                                                 DeviceState *dev, Error **errp)
4113 {
4114     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4115     MachineClass *mc = MACHINE_GET_CLASS(sms);
4116     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4117 
4118     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4119         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4120             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4121         } else {
4122             /* NOTE: this means there is a window after guest reset, prior to
4123              * CAS negotiation, where unplug requests will fail due to the
4124              * capability not being detected yet. This is a bit different than
4125              * the case with PCI unplug, where the events will be queued and
4126              * eventually handled by the guest after boot
4127              */
4128             error_setg(errp, "Memory hot unplug not supported for this guest");
4129         }
4130     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4131         if (!mc->has_hotpluggable_cpus) {
4132             error_setg(errp, "CPU hot unplug not supported on this machine");
4133             return;
4134         }
4135         spapr_core_unplug_request(hotplug_dev, dev, errp);
4136     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4137         if (!smc->dr_phb_enabled) {
4138             error_setg(errp, "PHB hot unplug not supported on this machine");
4139             return;
4140         }
4141         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4142     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4143         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4144     }
4145 }
4146 
4147 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4148                                           DeviceState *dev, Error **errp)
4149 {
4150     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4151         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4152     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4153         spapr_core_pre_plug(hotplug_dev, dev, errp);
4154     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4155         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4156     }
4157 }
4158 
4159 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4160                                                  DeviceState *dev)
4161 {
4162     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4163         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4164         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4165         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4166         return HOTPLUG_HANDLER(machine);
4167     }
4168     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4169         PCIDevice *pcidev = PCI_DEVICE(dev);
4170         PCIBus *root = pci_device_root_bus(pcidev);
4171         SpaprPhbState *phb =
4172             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4173                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4174 
4175         if (phb) {
4176             return HOTPLUG_HANDLER(phb);
4177         }
4178     }
4179     return NULL;
4180 }
4181 
4182 static CpuInstanceProperties
4183 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4184 {
4185     CPUArchId *core_slot;
4186     MachineClass *mc = MACHINE_GET_CLASS(machine);
4187 
4188     /* make sure possible_cpu are intialized */
4189     mc->possible_cpu_arch_ids(machine);
4190     /* get CPU core slot containing thread that matches cpu_index */
4191     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4192     assert(core_slot);
4193     return core_slot->props;
4194 }
4195 
4196 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4197 {
4198     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4199 }
4200 
4201 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4202 {
4203     int i;
4204     unsigned int smp_threads = machine->smp.threads;
4205     unsigned int smp_cpus = machine->smp.cpus;
4206     const char *core_type;
4207     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4208     MachineClass *mc = MACHINE_GET_CLASS(machine);
4209 
4210     if (!mc->has_hotpluggable_cpus) {
4211         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4212     }
4213     if (machine->possible_cpus) {
4214         assert(machine->possible_cpus->len == spapr_max_cores);
4215         return machine->possible_cpus;
4216     }
4217 
4218     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4219     if (!core_type) {
4220         error_report("Unable to find sPAPR CPU Core definition");
4221         exit(1);
4222     }
4223 
4224     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4225                              sizeof(CPUArchId) * spapr_max_cores);
4226     machine->possible_cpus->len = spapr_max_cores;
4227     for (i = 0; i < machine->possible_cpus->len; i++) {
4228         int core_id = i * smp_threads;
4229 
4230         machine->possible_cpus->cpus[i].type = core_type;
4231         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4232         machine->possible_cpus->cpus[i].arch_id = core_id;
4233         machine->possible_cpus->cpus[i].props.has_core_id = true;
4234         machine->possible_cpus->cpus[i].props.core_id = core_id;
4235     }
4236     return machine->possible_cpus;
4237 }
4238 
4239 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4240                                 uint64_t *buid, hwaddr *pio,
4241                                 hwaddr *mmio32, hwaddr *mmio64,
4242                                 unsigned n_dma, uint32_t *liobns,
4243                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4244 {
4245     /*
4246      * New-style PHB window placement.
4247      *
4248      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4249      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4250      * windows.
4251      *
4252      * Some guest kernels can't work with MMIO windows above 1<<46
4253      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4254      *
4255      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4256      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4257      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4258      * 1TiB 64-bit MMIO windows for each PHB.
4259      */
4260     const uint64_t base_buid = 0x800000020000000ULL;
4261     int i;
4262 
4263     /* Sanity check natural alignments */
4264     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4265     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4266     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4267     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4268     /* Sanity check bounds */
4269     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4270                       SPAPR_PCI_MEM32_WIN_SIZE);
4271     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4272                       SPAPR_PCI_MEM64_WIN_SIZE);
4273 
4274     if (index >= SPAPR_MAX_PHBS) {
4275         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4276                    SPAPR_MAX_PHBS - 1);
4277         return;
4278     }
4279 
4280     *buid = base_buid + index;
4281     for (i = 0; i < n_dma; ++i) {
4282         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4283     }
4284 
4285     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4286     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4287     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4288 
4289     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4290     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4291 }
4292 
4293 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4294 {
4295     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4296 
4297     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4298 }
4299 
4300 static void spapr_ics_resend(XICSFabric *dev)
4301 {
4302     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4303 
4304     ics_resend(spapr->ics);
4305 }
4306 
4307 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4308 {
4309     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4310 
4311     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4312 }
4313 
4314 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4315                                  Monitor *mon)
4316 {
4317     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4318 
4319     spapr_irq_print_info(spapr, mon);
4320     monitor_printf(mon, "irqchip: %s\n",
4321                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4322 }
4323 
4324 /*
4325  * This is a XIVE only operation
4326  */
4327 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4328                            uint8_t nvt_blk, uint32_t nvt_idx,
4329                            bool cam_ignore, uint8_t priority,
4330                            uint32_t logic_serv, XiveTCTXMatch *match)
4331 {
4332     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4333     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4334     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4335     int count;
4336 
4337     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4338                            priority, logic_serv, match);
4339     if (count < 0) {
4340         return count;
4341     }
4342 
4343     /*
4344      * When we implement the save and restore of the thread interrupt
4345      * contexts in the enter/exit CPU handlers of the machine and the
4346      * escalations in QEMU, we should be able to handle non dispatched
4347      * vCPUs.
4348      *
4349      * Until this is done, the sPAPR machine should find at least one
4350      * matching context always.
4351      */
4352     if (count == 0) {
4353         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4354                       nvt_blk, nvt_idx);
4355     }
4356 
4357     return count;
4358 }
4359 
4360 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4361 {
4362     return cpu->vcpu_id;
4363 }
4364 
4365 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4366 {
4367     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4368     MachineState *ms = MACHINE(spapr);
4369     int vcpu_id;
4370 
4371     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4372 
4373     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4374         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4375         error_append_hint(errp, "Adjust the number of cpus to %d "
4376                           "or try to raise the number of threads per core\n",
4377                           vcpu_id * ms->smp.threads / spapr->vsmt);
4378         return;
4379     }
4380 
4381     cpu->vcpu_id = vcpu_id;
4382 }
4383 
4384 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4385 {
4386     CPUState *cs;
4387 
4388     CPU_FOREACH(cs) {
4389         PowerPCCPU *cpu = POWERPC_CPU(cs);
4390 
4391         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4392             return cpu;
4393         }
4394     }
4395 
4396     return NULL;
4397 }
4398 
4399 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4400 {
4401     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4402 
4403     /* These are only called by TCG, KVM maintains dispatch state */
4404 
4405     spapr_cpu->prod = false;
4406     if (spapr_cpu->vpa_addr) {
4407         CPUState *cs = CPU(cpu);
4408         uint32_t dispatch;
4409 
4410         dispatch = ldl_be_phys(cs->as,
4411                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4412         dispatch++;
4413         if ((dispatch & 1) != 0) {
4414             qemu_log_mask(LOG_GUEST_ERROR,
4415                           "VPA: incorrect dispatch counter value for "
4416                           "dispatched partition %u, correcting.\n", dispatch);
4417             dispatch++;
4418         }
4419         stl_be_phys(cs->as,
4420                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4421     }
4422 }
4423 
4424 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4425 {
4426     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4427 
4428     if (spapr_cpu->vpa_addr) {
4429         CPUState *cs = CPU(cpu);
4430         uint32_t dispatch;
4431 
4432         dispatch = ldl_be_phys(cs->as,
4433                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4434         dispatch++;
4435         if ((dispatch & 1) != 1) {
4436             qemu_log_mask(LOG_GUEST_ERROR,
4437                           "VPA: incorrect dispatch counter value for "
4438                           "preempted partition %u, correcting.\n", dispatch);
4439             dispatch++;
4440         }
4441         stl_be_phys(cs->as,
4442                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4443     }
4444 }
4445 
4446 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4447 {
4448     MachineClass *mc = MACHINE_CLASS(oc);
4449     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4450     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4451     NMIClass *nc = NMI_CLASS(oc);
4452     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4453     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4454     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4455     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4456     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4457 
4458     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4459     mc->ignore_boot_device_suffixes = true;
4460 
4461     /*
4462      * We set up the default / latest behaviour here.  The class_init
4463      * functions for the specific versioned machine types can override
4464      * these details for backwards compatibility
4465      */
4466     mc->init = spapr_machine_init;
4467     mc->reset = spapr_machine_reset;
4468     mc->block_default_type = IF_SCSI;
4469     mc->max_cpus = 1024;
4470     mc->no_parallel = 1;
4471     mc->default_boot_order = "";
4472     mc->default_ram_size = 512 * MiB;
4473     mc->default_ram_id = "ppc_spapr.ram";
4474     mc->default_display = "std";
4475     mc->kvm_type = spapr_kvm_type;
4476     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4477     mc->pci_allow_0_address = true;
4478     assert(!mc->get_hotplug_handler);
4479     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4480     hc->pre_plug = spapr_machine_device_pre_plug;
4481     hc->plug = spapr_machine_device_plug;
4482     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4483     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4484     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4485     hc->unplug_request = spapr_machine_device_unplug_request;
4486     hc->unplug = spapr_machine_device_unplug;
4487 
4488     smc->dr_lmb_enabled = true;
4489     smc->update_dt_enabled = true;
4490     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4491     mc->has_hotpluggable_cpus = true;
4492     mc->nvdimm_supported = true;
4493     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4494     fwc->get_dev_path = spapr_get_fw_dev_path;
4495     nc->nmi_monitor_handler = spapr_nmi;
4496     smc->phb_placement = spapr_phb_placement;
4497     vhc->hypercall = emulate_spapr_hypercall;
4498     vhc->hpt_mask = spapr_hpt_mask;
4499     vhc->map_hptes = spapr_map_hptes;
4500     vhc->unmap_hptes = spapr_unmap_hptes;
4501     vhc->hpte_set_c = spapr_hpte_set_c;
4502     vhc->hpte_set_r = spapr_hpte_set_r;
4503     vhc->get_pate = spapr_get_pate;
4504     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4505     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4506     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4507     xic->ics_get = spapr_ics_get;
4508     xic->ics_resend = spapr_ics_resend;
4509     xic->icp_get = spapr_icp_get;
4510     ispc->print_info = spapr_pic_print_info;
4511     /* Force NUMA node memory size to be a multiple of
4512      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4513      * in which LMBs are represented and hot-added
4514      */
4515     mc->numa_mem_align_shift = 28;
4516     mc->numa_mem_supported = true;
4517     mc->auto_enable_numa = true;
4518 
4519     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4520     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4521     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4522     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4523     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4524     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4525     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4526     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4527     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4528     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4529     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_ON;
4530     spapr_caps_add_properties(smc, &error_abort);
4531     smc->irq = &spapr_irq_dual;
4532     smc->dr_phb_enabled = true;
4533     smc->linux_pci_probe = true;
4534     smc->smp_threads_vsmt = true;
4535     smc->nr_xirqs = SPAPR_NR_XIRQS;
4536     xfc->match_nvt = spapr_match_nvt;
4537 }
4538 
4539 static const TypeInfo spapr_machine_info = {
4540     .name          = TYPE_SPAPR_MACHINE,
4541     .parent        = TYPE_MACHINE,
4542     .abstract      = true,
4543     .instance_size = sizeof(SpaprMachineState),
4544     .instance_init = spapr_instance_init,
4545     .instance_finalize = spapr_machine_finalizefn,
4546     .class_size    = sizeof(SpaprMachineClass),
4547     .class_init    = spapr_machine_class_init,
4548     .interfaces = (InterfaceInfo[]) {
4549         { TYPE_FW_PATH_PROVIDER },
4550         { TYPE_NMI },
4551         { TYPE_HOTPLUG_HANDLER },
4552         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4553         { TYPE_XICS_FABRIC },
4554         { TYPE_INTERRUPT_STATS_PROVIDER },
4555         { TYPE_XIVE_FABRIC },
4556         { }
4557     },
4558 };
4559 
4560 static void spapr_machine_latest_class_options(MachineClass *mc)
4561 {
4562     mc->alias = "pseries";
4563     mc->is_default = true;
4564 }
4565 
4566 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4567     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4568                                                     void *data)      \
4569     {                                                                \
4570         MachineClass *mc = MACHINE_CLASS(oc);                        \
4571         spapr_machine_##suffix##_class_options(mc);                  \
4572         if (latest) {                                                \
4573             spapr_machine_latest_class_options(mc);                  \
4574         }                                                            \
4575     }                                                                \
4576     static const TypeInfo spapr_machine_##suffix##_info = {          \
4577         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4578         .parent = TYPE_SPAPR_MACHINE,                                \
4579         .class_init = spapr_machine_##suffix##_class_init,           \
4580     };                                                               \
4581     static void spapr_machine_register_##suffix(void)                \
4582     {                                                                \
4583         type_register(&spapr_machine_##suffix##_info);               \
4584     }                                                                \
4585     type_init(spapr_machine_register_##suffix)
4586 
4587 /*
4588  * pseries-5.0
4589  */
4590 static void spapr_machine_5_0_class_options(MachineClass *mc)
4591 {
4592     /* Defaults for the latest behaviour inherited from the base class */
4593 }
4594 
4595 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4596 
4597 /*
4598  * pseries-4.2
4599  */
4600 static void spapr_machine_4_2_class_options(MachineClass *mc)
4601 {
4602     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4603 
4604     spapr_machine_5_0_class_options(mc);
4605     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4606     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4607     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_OFF;
4608     mc->nvdimm_supported = false;
4609 }
4610 
4611 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4612 
4613 /*
4614  * pseries-4.1
4615  */
4616 static void spapr_machine_4_1_class_options(MachineClass *mc)
4617 {
4618     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4619     static GlobalProperty compat[] = {
4620         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4621         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4622     };
4623 
4624     spapr_machine_4_2_class_options(mc);
4625     smc->linux_pci_probe = false;
4626     smc->smp_threads_vsmt = false;
4627     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4628     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4629 }
4630 
4631 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4632 
4633 /*
4634  * pseries-4.0
4635  */
4636 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4637                               uint64_t *buid, hwaddr *pio,
4638                               hwaddr *mmio32, hwaddr *mmio64,
4639                               unsigned n_dma, uint32_t *liobns,
4640                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4641 {
4642     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4643                         nv2gpa, nv2atsd, errp);
4644     *nv2gpa = 0;
4645     *nv2atsd = 0;
4646 }
4647 
4648 static void spapr_machine_4_0_class_options(MachineClass *mc)
4649 {
4650     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4651 
4652     spapr_machine_4_1_class_options(mc);
4653     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4654     smc->phb_placement = phb_placement_4_0;
4655     smc->irq = &spapr_irq_xics;
4656     smc->pre_4_1_migration = true;
4657 }
4658 
4659 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4660 
4661 /*
4662  * pseries-3.1
4663  */
4664 static void spapr_machine_3_1_class_options(MachineClass *mc)
4665 {
4666     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4667 
4668     spapr_machine_4_0_class_options(mc);
4669     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4670 
4671     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4672     smc->update_dt_enabled = false;
4673     smc->dr_phb_enabled = false;
4674     smc->broken_host_serial_model = true;
4675     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4676     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4677     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4678     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4679 }
4680 
4681 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4682 
4683 /*
4684  * pseries-3.0
4685  */
4686 
4687 static void spapr_machine_3_0_class_options(MachineClass *mc)
4688 {
4689     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4690 
4691     spapr_machine_3_1_class_options(mc);
4692     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4693 
4694     smc->legacy_irq_allocation = true;
4695     smc->nr_xirqs = 0x400;
4696     smc->irq = &spapr_irq_xics_legacy;
4697 }
4698 
4699 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4700 
4701 /*
4702  * pseries-2.12
4703  */
4704 static void spapr_machine_2_12_class_options(MachineClass *mc)
4705 {
4706     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4707     static GlobalProperty compat[] = {
4708         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4709         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4710     };
4711 
4712     spapr_machine_3_0_class_options(mc);
4713     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4714     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4715 
4716     /* We depend on kvm_enabled() to choose a default value for the
4717      * hpt-max-page-size capability. Of course we can't do it here
4718      * because this is too early and the HW accelerator isn't initialzed
4719      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4720      */
4721     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4722 }
4723 
4724 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4725 
4726 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4727 {
4728     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4729 
4730     spapr_machine_2_12_class_options(mc);
4731     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4732     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4733     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4734 }
4735 
4736 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4737 
4738 /*
4739  * pseries-2.11
4740  */
4741 
4742 static void spapr_machine_2_11_class_options(MachineClass *mc)
4743 {
4744     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4745 
4746     spapr_machine_2_12_class_options(mc);
4747     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4748     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4749 }
4750 
4751 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4752 
4753 /*
4754  * pseries-2.10
4755  */
4756 
4757 static void spapr_machine_2_10_class_options(MachineClass *mc)
4758 {
4759     spapr_machine_2_11_class_options(mc);
4760     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4761 }
4762 
4763 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4764 
4765 /*
4766  * pseries-2.9
4767  */
4768 
4769 static void spapr_machine_2_9_class_options(MachineClass *mc)
4770 {
4771     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4772     static GlobalProperty compat[] = {
4773         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4774     };
4775 
4776     spapr_machine_2_10_class_options(mc);
4777     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4778     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4779     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4780     smc->pre_2_10_has_unused_icps = true;
4781     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4782 }
4783 
4784 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4785 
4786 /*
4787  * pseries-2.8
4788  */
4789 
4790 static void spapr_machine_2_8_class_options(MachineClass *mc)
4791 {
4792     static GlobalProperty compat[] = {
4793         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4794     };
4795 
4796     spapr_machine_2_9_class_options(mc);
4797     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4798     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4799     mc->numa_mem_align_shift = 23;
4800 }
4801 
4802 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4803 
4804 /*
4805  * pseries-2.7
4806  */
4807 
4808 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4809                               uint64_t *buid, hwaddr *pio,
4810                               hwaddr *mmio32, hwaddr *mmio64,
4811                               unsigned n_dma, uint32_t *liobns,
4812                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4813 {
4814     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4815     const uint64_t base_buid = 0x800000020000000ULL;
4816     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4817     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4818     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4819     const uint32_t max_index = 255;
4820     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4821 
4822     uint64_t ram_top = MACHINE(spapr)->ram_size;
4823     hwaddr phb0_base, phb_base;
4824     int i;
4825 
4826     /* Do we have device memory? */
4827     if (MACHINE(spapr)->maxram_size > ram_top) {
4828         /* Can't just use maxram_size, because there may be an
4829          * alignment gap between normal and device memory regions
4830          */
4831         ram_top = MACHINE(spapr)->device_memory->base +
4832             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4833     }
4834 
4835     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4836 
4837     if (index > max_index) {
4838         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4839                    max_index);
4840         return;
4841     }
4842 
4843     *buid = base_buid + index;
4844     for (i = 0; i < n_dma; ++i) {
4845         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4846     }
4847 
4848     phb_base = phb0_base + index * phb_spacing;
4849     *pio = phb_base + pio_offset;
4850     *mmio32 = phb_base + mmio_offset;
4851     /*
4852      * We don't set the 64-bit MMIO window, relying on the PHB's
4853      * fallback behaviour of automatically splitting a large "32-bit"
4854      * window into contiguous 32-bit and 64-bit windows
4855      */
4856 
4857     *nv2gpa = 0;
4858     *nv2atsd = 0;
4859 }
4860 
4861 static void spapr_machine_2_7_class_options(MachineClass *mc)
4862 {
4863     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4864     static GlobalProperty compat[] = {
4865         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4866         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4867         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4868         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4869     };
4870 
4871     spapr_machine_2_8_class_options(mc);
4872     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4873     mc->default_machine_opts = "modern-hotplug-events=off";
4874     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4875     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4876     smc->phb_placement = phb_placement_2_7;
4877 }
4878 
4879 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4880 
4881 /*
4882  * pseries-2.6
4883  */
4884 
4885 static void spapr_machine_2_6_class_options(MachineClass *mc)
4886 {
4887     static GlobalProperty compat[] = {
4888         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4889     };
4890 
4891     spapr_machine_2_7_class_options(mc);
4892     mc->has_hotpluggable_cpus = false;
4893     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4894     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4895 }
4896 
4897 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4898 
4899 /*
4900  * pseries-2.5
4901  */
4902 
4903 static void spapr_machine_2_5_class_options(MachineClass *mc)
4904 {
4905     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4906     static GlobalProperty compat[] = {
4907         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4908     };
4909 
4910     spapr_machine_2_6_class_options(mc);
4911     smc->use_ohci_by_default = true;
4912     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4913     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4914 }
4915 
4916 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4917 
4918 /*
4919  * pseries-2.4
4920  */
4921 
4922 static void spapr_machine_2_4_class_options(MachineClass *mc)
4923 {
4924     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4925 
4926     spapr_machine_2_5_class_options(mc);
4927     smc->dr_lmb_enabled = false;
4928     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4929 }
4930 
4931 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4932 
4933 /*
4934  * pseries-2.3
4935  */
4936 
4937 static void spapr_machine_2_3_class_options(MachineClass *mc)
4938 {
4939     static GlobalProperty compat[] = {
4940         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4941     };
4942     spapr_machine_2_4_class_options(mc);
4943     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4944     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4945 }
4946 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4947 
4948 /*
4949  * pseries-2.2
4950  */
4951 
4952 static void spapr_machine_2_2_class_options(MachineClass *mc)
4953 {
4954     static GlobalProperty compat[] = {
4955         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4956     };
4957 
4958     spapr_machine_2_3_class_options(mc);
4959     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4960     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4961     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4962 }
4963 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4964 
4965 /*
4966  * pseries-2.1
4967  */
4968 
4969 static void spapr_machine_2_1_class_options(MachineClass *mc)
4970 {
4971     spapr_machine_2_2_class_options(mc);
4972     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4973 }
4974 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4975 
4976 static void spapr_machine_register_types(void)
4977 {
4978     type_register_static(&spapr_machine_info);
4979 }
4980 
4981 type_init(spapr_machine_register_types)
4982