xref: /qemu/hw/ppc/spapr.c (revision 97b32a6afa78ae68fb16344b9a144b6f433f42a2)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "mmu-hash64.h"
50 #include "mmu-book3s-v3.h"
51 #include "cpu-models.h"
52 #include "hw/core/cpu.h"
53 
54 #include "hw/boards.h"
55 #include "hw/ppc/ppc.h"
56 #include "hw/loader.h"
57 
58 #include "hw/ppc/fdt.h"
59 #include "hw/ppc/spapr.h"
60 #include "hw/ppc/spapr_vio.h"
61 #include "hw/qdev-properties.h"
62 #include "hw/pci-host/spapr.h"
63 #include "hw/pci/msi.h"
64 
65 #include "hw/pci/pci.h"
66 #include "hw/scsi/scsi.h"
67 #include "hw/virtio/virtio-scsi.h"
68 #include "hw/virtio/vhost-scsi-common.h"
69 
70 #include "exec/address-spaces.h"
71 #include "exec/ram_addr.h"
72 #include "hw/usb.h"
73 #include "qemu/config-file.h"
74 #include "qemu/error-report.h"
75 #include "trace.h"
76 #include "hw/nmi.h"
77 #include "hw/intc/intc.h"
78 
79 #include "qemu/cutils.h"
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 
84 #include "monitor/monitor.h"
85 
86 #include <libfdt.h>
87 
88 /* SLOF memory layout:
89  *
90  * SLOF raw image loaded at 0, copies its romfs right below the flat
91  * device-tree, then position SLOF itself 31M below that
92  *
93  * So we set FW_OVERHEAD to 40MB which should account for all of that
94  * and more
95  *
96  * We load our kernel at 4M, leaving space for SLOF initial image
97  */
98 #define FDT_MAX_SIZE            0x100000
99 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
100 #define FW_MAX_SIZE             0x400000
101 #define FW_FILE_NAME            "slof.bin"
102 #define FW_OVERHEAD             0x2800000
103 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
104 
105 #define MIN_RMA_SLOF            128UL
106 
107 #define PHANDLE_INTC            0x00001111
108 
109 /* These two functions implement the VCPU id numbering: one to compute them
110  * all and one to identify thread 0 of a VCORE. Any change to the first one
111  * is likely to have an impact on the second one, so let's keep them close.
112  */
113 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
114 {
115     MachineState *ms = MACHINE(spapr);
116     unsigned int smp_threads = ms->smp.threads;
117 
118     assert(spapr->vsmt);
119     return
120         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
121 }
122 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
123                                       PowerPCCPU *cpu)
124 {
125     assert(spapr->vsmt);
126     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
127 }
128 
129 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
130 {
131     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
132      * and newer QEMUs don't even have them. In both cases, we don't want
133      * to send anything on the wire.
134      */
135     return false;
136 }
137 
138 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
139     .name = "icp/server",
140     .version_id = 1,
141     .minimum_version_id = 1,
142     .needed = pre_2_10_vmstate_dummy_icp_needed,
143     .fields = (VMStateField[]) {
144         VMSTATE_UNUSED(4), /* uint32_t xirr */
145         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
146         VMSTATE_UNUSED(1), /* uint8_t mfrr */
147         VMSTATE_END_OF_LIST()
148     },
149 };
150 
151 static void pre_2_10_vmstate_register_dummy_icp(int i)
152 {
153     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
154                      (void *)(uintptr_t) i);
155 }
156 
157 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
158 {
159     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
160                        (void *)(uintptr_t) i);
161 }
162 
163 int spapr_max_server_number(SpaprMachineState *spapr)
164 {
165     MachineState *ms = MACHINE(spapr);
166 
167     assert(spapr->vsmt);
168     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
169 }
170 
171 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
172                                   int smt_threads)
173 {
174     int i, ret = 0;
175     uint32_t servers_prop[smt_threads];
176     uint32_t gservers_prop[smt_threads * 2];
177     int index = spapr_get_vcpu_id(cpu);
178 
179     if (cpu->compat_pvr) {
180         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
181         if (ret < 0) {
182             return ret;
183         }
184     }
185 
186     /* Build interrupt servers and gservers properties */
187     for (i = 0; i < smt_threads; i++) {
188         servers_prop[i] = cpu_to_be32(index + i);
189         /* Hack, direct the group queues back to cpu 0 */
190         gservers_prop[i*2] = cpu_to_be32(index + i);
191         gservers_prop[i*2 + 1] = 0;
192     }
193     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
194                       servers_prop, sizeof(servers_prop));
195     if (ret < 0) {
196         return ret;
197     }
198     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
199                       gservers_prop, sizeof(gservers_prop));
200 
201     return ret;
202 }
203 
204 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
205 {
206     int index = spapr_get_vcpu_id(cpu);
207     uint32_t associativity[] = {cpu_to_be32(0x5),
208                                 cpu_to_be32(0x0),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(cpu->node_id),
212                                 cpu_to_be32(index)};
213 
214     /* Advertise NUMA via ibm,associativity */
215     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
216                           sizeof(associativity));
217 }
218 
219 /* Populate the "ibm,pa-features" property */
220 static void spapr_populate_pa_features(SpaprMachineState *spapr,
221                                        PowerPCCPU *cpu,
222                                        void *fdt, int offset)
223 {
224     uint8_t pa_features_206[] = { 6, 0,
225         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
226     uint8_t pa_features_207[] = { 24, 0,
227         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
228         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
229         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
230         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
231     uint8_t pa_features_300[] = { 66, 0,
232         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
233         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
234         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
235         /* 6: DS207 */
236         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
237         /* 16: Vector */
238         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
239         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
241         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
242         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
243         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
244         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
245         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
246         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
247         /* 42: PM, 44: PC RA, 46: SC vec'd */
248         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
249         /* 48: SIMD, 50: QP BFP, 52: String */
250         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
251         /* 54: DecFP, 56: DecI, 58: SHA */
252         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
253         /* 60: NM atomic, 62: RNG */
254         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
255     };
256     uint8_t *pa_features = NULL;
257     size_t pa_size;
258 
259     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
260         pa_features = pa_features_206;
261         pa_size = sizeof(pa_features_206);
262     }
263     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
264         pa_features = pa_features_207;
265         pa_size = sizeof(pa_features_207);
266     }
267     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
268         pa_features = pa_features_300;
269         pa_size = sizeof(pa_features_300);
270     }
271     if (!pa_features) {
272         return;
273     }
274 
275     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
276         /*
277          * Note: we keep CI large pages off by default because a 64K capable
278          * guest provisioned with large pages might otherwise try to map a qemu
279          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
280          * even if that qemu runs on a 4k host.
281          * We dd this bit back here if we are confident this is not an issue
282          */
283         pa_features[3] |= 0x20;
284     }
285     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
286         pa_features[24] |= 0x80;    /* Transactional memory support */
287     }
288     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
289         /* Workaround for broken kernels that attempt (guest) radix
290          * mode when they can't handle it, if they see the radix bit set
291          * in pa-features. So hide it from them. */
292         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
293     }
294 
295     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
296 }
297 
298 static hwaddr spapr_node0_size(MachineState *machine)
299 {
300     if (machine->numa_state->num_nodes) {
301         int i;
302         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
303             if (machine->numa_state->nodes[i].node_mem) {
304                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
305                            machine->ram_size);
306             }
307         }
308     }
309     return machine->ram_size;
310 }
311 
312 static void add_str(GString *s, const gchar *s1)
313 {
314     g_string_append_len(s, s1, strlen(s1) + 1);
315 }
316 
317 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
318                                        hwaddr size)
319 {
320     uint32_t associativity[] = {
321         cpu_to_be32(0x4), /* length */
322         cpu_to_be32(0x0), cpu_to_be32(0x0),
323         cpu_to_be32(0x0), cpu_to_be32(nodeid)
324     };
325     char mem_name[32];
326     uint64_t mem_reg_property[2];
327     int off;
328 
329     mem_reg_property[0] = cpu_to_be64(start);
330     mem_reg_property[1] = cpu_to_be64(size);
331 
332     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
333     off = fdt_add_subnode(fdt, 0, mem_name);
334     _FDT(off);
335     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
336     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
337                       sizeof(mem_reg_property))));
338     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
339                       sizeof(associativity))));
340     return off;
341 }
342 
343 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
344 {
345     MachineState *machine = MACHINE(spapr);
346     hwaddr mem_start, node_size;
347     int i, nb_nodes = machine->numa_state->num_nodes;
348     NodeInfo *nodes = machine->numa_state->nodes;
349 
350     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
351         if (!nodes[i].node_mem) {
352             continue;
353         }
354         if (mem_start >= machine->ram_size) {
355             node_size = 0;
356         } else {
357             node_size = nodes[i].node_mem;
358             if (node_size > machine->ram_size - mem_start) {
359                 node_size = machine->ram_size - mem_start;
360             }
361         }
362         if (!mem_start) {
363             /* spapr_machine_init() checks for rma_size <= node0_size
364              * already */
365             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
366             mem_start += spapr->rma_size;
367             node_size -= spapr->rma_size;
368         }
369         for ( ; node_size; ) {
370             hwaddr sizetmp = pow2floor(node_size);
371 
372             /* mem_start != 0 here */
373             if (ctzl(mem_start) < ctzl(sizetmp)) {
374                 sizetmp = 1ULL << ctzl(mem_start);
375             }
376 
377             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
378             node_size -= sizetmp;
379             mem_start += sizetmp;
380         }
381     }
382 
383     return 0;
384 }
385 
386 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
387                                   SpaprMachineState *spapr)
388 {
389     MachineState *ms = MACHINE(spapr);
390     PowerPCCPU *cpu = POWERPC_CPU(cs);
391     CPUPPCState *env = &cpu->env;
392     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
393     int index = spapr_get_vcpu_id(cpu);
394     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
395                        0xffffffff, 0xffffffff};
396     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
397         : SPAPR_TIMEBASE_FREQ;
398     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
399     uint32_t page_sizes_prop[64];
400     size_t page_sizes_prop_size;
401     unsigned int smp_threads = ms->smp.threads;
402     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
403     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
404     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
405     SpaprDrc *drc;
406     int drc_index;
407     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
408     int i;
409 
410     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
411     if (drc) {
412         drc_index = spapr_drc_index(drc);
413         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
414     }
415 
416     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
417     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
418 
419     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
420     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
421                            env->dcache_line_size)));
422     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
423                            env->dcache_line_size)));
424     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
425                            env->icache_line_size)));
426     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
427                            env->icache_line_size)));
428 
429     if (pcc->l1_dcache_size) {
430         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
431                                pcc->l1_dcache_size)));
432     } else {
433         warn_report("Unknown L1 dcache size for cpu");
434     }
435     if (pcc->l1_icache_size) {
436         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
437                                pcc->l1_icache_size)));
438     } else {
439         warn_report("Unknown L1 icache size for cpu");
440     }
441 
442     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
443     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
444     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
445     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
446     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
447     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
448 
449     if (env->spr_cb[SPR_PURR].oea_read) {
450         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
451     }
452     if (env->spr_cb[SPR_SPURR].oea_read) {
453         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
454     }
455 
456     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
457         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
458                           segs, sizeof(segs))));
459     }
460 
461     /* Advertise VSX (vector extensions) if available
462      *   1               == VMX / Altivec available
463      *   2               == VSX available
464      *
465      * Only CPUs for which we create core types in spapr_cpu_core.c
466      * are possible, and all of those have VMX */
467     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
468         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
469     } else {
470         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
471     }
472 
473     /* Advertise DFP (Decimal Floating Point) if available
474      *   0 / no property == no DFP
475      *   1               == DFP available */
476     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
477         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
478     }
479 
480     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
481                                                       sizeof(page_sizes_prop));
482     if (page_sizes_prop_size) {
483         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
484                           page_sizes_prop, page_sizes_prop_size)));
485     }
486 
487     spapr_populate_pa_features(spapr, cpu, fdt, offset);
488 
489     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
490                            cs->cpu_index / vcpus_per_socket)));
491 
492     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
493                       pft_size_prop, sizeof(pft_size_prop))));
494 
495     if (ms->numa_state->num_nodes > 1) {
496         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
497     }
498 
499     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
500 
501     if (pcc->radix_page_info) {
502         for (i = 0; i < pcc->radix_page_info->count; i++) {
503             radix_AP_encodings[i] =
504                 cpu_to_be32(pcc->radix_page_info->entries[i]);
505         }
506         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
507                           radix_AP_encodings,
508                           pcc->radix_page_info->count *
509                           sizeof(radix_AP_encodings[0]))));
510     }
511 
512     /*
513      * We set this property to let the guest know that it can use the large
514      * decrementer and its width in bits.
515      */
516     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
517         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
518                               pcc->lrg_decr_bits)));
519 }
520 
521 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
522 {
523     CPUState **rev;
524     CPUState *cs;
525     int n_cpus;
526     int cpus_offset;
527     char *nodename;
528     int i;
529 
530     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
531     _FDT(cpus_offset);
532     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
533     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
534 
535     /*
536      * We walk the CPUs in reverse order to ensure that CPU DT nodes
537      * created by fdt_add_subnode() end up in the right order in FDT
538      * for the guest kernel the enumerate the CPUs correctly.
539      *
540      * The CPU list cannot be traversed in reverse order, so we need
541      * to do extra work.
542      */
543     n_cpus = 0;
544     rev = NULL;
545     CPU_FOREACH(cs) {
546         rev = g_renew(CPUState *, rev, n_cpus + 1);
547         rev[n_cpus++] = cs;
548     }
549 
550     for (i = n_cpus - 1; i >= 0; i--) {
551         CPUState *cs = rev[i];
552         PowerPCCPU *cpu = POWERPC_CPU(cs);
553         int index = spapr_get_vcpu_id(cpu);
554         DeviceClass *dc = DEVICE_GET_CLASS(cs);
555         int offset;
556 
557         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
558             continue;
559         }
560 
561         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
562         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
563         g_free(nodename);
564         _FDT(offset);
565         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
566     }
567 
568     g_free(rev);
569 }
570 
571 static int spapr_rng_populate_dt(void *fdt)
572 {
573     int node;
574     int ret;
575 
576     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
577     if (node <= 0) {
578         return -1;
579     }
580     ret = fdt_setprop_string(fdt, node, "device_type",
581                              "ibm,platform-facilities");
582     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
583     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
584 
585     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
586     if (node <= 0) {
587         return -1;
588     }
589     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
590 
591     return ret ? -1 : 0;
592 }
593 
594 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
595 {
596     MemoryDeviceInfoList *info;
597 
598     for (info = list; info; info = info->next) {
599         MemoryDeviceInfo *value = info->value;
600 
601         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
602             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
603 
604             if (addr >= pcdimm_info->addr &&
605                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
606                 return pcdimm_info->node;
607             }
608         }
609     }
610 
611     return -1;
612 }
613 
614 struct sPAPRDrconfCellV2 {
615      uint32_t seq_lmbs;
616      uint64_t base_addr;
617      uint32_t drc_index;
618      uint32_t aa_index;
619      uint32_t flags;
620 } QEMU_PACKED;
621 
622 typedef struct DrconfCellQueue {
623     struct sPAPRDrconfCellV2 cell;
624     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
625 } DrconfCellQueue;
626 
627 static DrconfCellQueue *
628 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
629                       uint32_t drc_index, uint32_t aa_index,
630                       uint32_t flags)
631 {
632     DrconfCellQueue *elem;
633 
634     elem = g_malloc0(sizeof(*elem));
635     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
636     elem->cell.base_addr = cpu_to_be64(base_addr);
637     elem->cell.drc_index = cpu_to_be32(drc_index);
638     elem->cell.aa_index = cpu_to_be32(aa_index);
639     elem->cell.flags = cpu_to_be32(flags);
640 
641     return elem;
642 }
643 
644 /* ibm,dynamic-memory-v2 */
645 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
646                                    int offset, MemoryDeviceInfoList *dimms)
647 {
648     MachineState *machine = MACHINE(spapr);
649     uint8_t *int_buf, *cur_index;
650     int ret;
651     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
652     uint64_t addr, cur_addr, size;
653     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
654     uint64_t mem_end = machine->device_memory->base +
655                        memory_region_size(&machine->device_memory->mr);
656     uint32_t node, buf_len, nr_entries = 0;
657     SpaprDrc *drc;
658     DrconfCellQueue *elem, *next;
659     MemoryDeviceInfoList *info;
660     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
661         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
662 
663     /* Entry to cover RAM and the gap area */
664     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
665                                  SPAPR_LMB_FLAGS_RESERVED |
666                                  SPAPR_LMB_FLAGS_DRC_INVALID);
667     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
668     nr_entries++;
669 
670     cur_addr = machine->device_memory->base;
671     for (info = dimms; info; info = info->next) {
672         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
673 
674         addr = di->addr;
675         size = di->size;
676         node = di->node;
677 
678         /* Entry for hot-pluggable area */
679         if (cur_addr < addr) {
680             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
681             g_assert(drc);
682             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
683                                          cur_addr, spapr_drc_index(drc), -1, 0);
684             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
685             nr_entries++;
686         }
687 
688         /* Entry for DIMM */
689         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
690         g_assert(drc);
691         elem = spapr_get_drconf_cell(size / lmb_size, addr,
692                                      spapr_drc_index(drc), node,
693                                      SPAPR_LMB_FLAGS_ASSIGNED);
694         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
695         nr_entries++;
696         cur_addr = addr + size;
697     }
698 
699     /* Entry for remaining hotpluggable area */
700     if (cur_addr < mem_end) {
701         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
702         g_assert(drc);
703         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
704                                      cur_addr, spapr_drc_index(drc), -1, 0);
705         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
706         nr_entries++;
707     }
708 
709     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
710     int_buf = cur_index = g_malloc0(buf_len);
711     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
712     cur_index += sizeof(nr_entries);
713 
714     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
715         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
716         cur_index += sizeof(elem->cell);
717         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
718         g_free(elem);
719     }
720 
721     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
722     g_free(int_buf);
723     if (ret < 0) {
724         return -1;
725     }
726     return 0;
727 }
728 
729 /* ibm,dynamic-memory */
730 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
731                                    int offset, MemoryDeviceInfoList *dimms)
732 {
733     MachineState *machine = MACHINE(spapr);
734     int i, ret;
735     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
736     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
737     uint32_t nr_lmbs = (machine->device_memory->base +
738                        memory_region_size(&machine->device_memory->mr)) /
739                        lmb_size;
740     uint32_t *int_buf, *cur_index, buf_len;
741 
742     /*
743      * Allocate enough buffer size to fit in ibm,dynamic-memory
744      */
745     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
746     cur_index = int_buf = g_malloc0(buf_len);
747     int_buf[0] = cpu_to_be32(nr_lmbs);
748     cur_index++;
749     for (i = 0; i < nr_lmbs; i++) {
750         uint64_t addr = i * lmb_size;
751         uint32_t *dynamic_memory = cur_index;
752 
753         if (i >= device_lmb_start) {
754             SpaprDrc *drc;
755 
756             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
757             g_assert(drc);
758 
759             dynamic_memory[0] = cpu_to_be32(addr >> 32);
760             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
761             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
762             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
763             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
764             if (memory_region_present(get_system_memory(), addr)) {
765                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
766             } else {
767                 dynamic_memory[5] = cpu_to_be32(0);
768             }
769         } else {
770             /*
771              * LMB information for RMA, boot time RAM and gap b/n RAM and
772              * device memory region -- all these are marked as reserved
773              * and as having no valid DRC.
774              */
775             dynamic_memory[0] = cpu_to_be32(addr >> 32);
776             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
777             dynamic_memory[2] = cpu_to_be32(0);
778             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
779             dynamic_memory[4] = cpu_to_be32(-1);
780             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
781                                             SPAPR_LMB_FLAGS_DRC_INVALID);
782         }
783 
784         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
785     }
786     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
787     g_free(int_buf);
788     if (ret < 0) {
789         return -1;
790     }
791     return 0;
792 }
793 
794 /*
795  * Adds ibm,dynamic-reconfiguration-memory node.
796  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
797  * of this device tree node.
798  */
799 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
800 {
801     MachineState *machine = MACHINE(spapr);
802     int nb_numa_nodes = machine->numa_state->num_nodes;
803     int ret, i, offset;
804     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
805     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
806     uint32_t *int_buf, *cur_index, buf_len;
807     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
808     MemoryDeviceInfoList *dimms = NULL;
809 
810     /*
811      * Don't create the node if there is no device memory
812      */
813     if (machine->ram_size == machine->maxram_size) {
814         return 0;
815     }
816 
817     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
818 
819     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
820                     sizeof(prop_lmb_size));
821     if (ret < 0) {
822         return ret;
823     }
824 
825     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
826     if (ret < 0) {
827         return ret;
828     }
829 
830     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
831     if (ret < 0) {
832         return ret;
833     }
834 
835     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
836     dimms = qmp_memory_device_list();
837     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
838         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
839     } else {
840         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
841     }
842     qapi_free_MemoryDeviceInfoList(dimms);
843 
844     if (ret < 0) {
845         return ret;
846     }
847 
848     /* ibm,associativity-lookup-arrays */
849     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
850     cur_index = int_buf = g_malloc0(buf_len);
851     int_buf[0] = cpu_to_be32(nr_nodes);
852     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
853     cur_index += 2;
854     for (i = 0; i < nr_nodes; i++) {
855         uint32_t associativity[] = {
856             cpu_to_be32(0x0),
857             cpu_to_be32(0x0),
858             cpu_to_be32(0x0),
859             cpu_to_be32(i)
860         };
861         memcpy(cur_index, associativity, sizeof(associativity));
862         cur_index += 4;
863     }
864     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
865             (cur_index - int_buf) * sizeof(uint32_t));
866     g_free(int_buf);
867 
868     return ret;
869 }
870 
871 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
872                                 SpaprOptionVector *ov5_updates)
873 {
874     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
875     int ret = 0, offset;
876 
877     /* Generate ibm,dynamic-reconfiguration-memory node if required */
878     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
879         g_assert(smc->dr_lmb_enabled);
880         ret = spapr_populate_drconf_memory(spapr, fdt);
881         if (ret) {
882             goto out;
883         }
884     }
885 
886     offset = fdt_path_offset(fdt, "/chosen");
887     if (offset < 0) {
888         offset = fdt_add_subnode(fdt, 0, "chosen");
889         if (offset < 0) {
890             return offset;
891         }
892     }
893     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
894                                  "ibm,architecture-vec-5");
895 
896 out:
897     return ret;
898 }
899 
900 static bool spapr_hotplugged_dev_before_cas(void)
901 {
902     Object *drc_container, *obj;
903     ObjectProperty *prop;
904     ObjectPropertyIterator iter;
905 
906     drc_container = container_get(object_get_root(), "/dr-connector");
907     object_property_iter_init(&iter, drc_container);
908     while ((prop = object_property_iter_next(&iter))) {
909         if (!strstart(prop->type, "link<", NULL)) {
910             continue;
911         }
912         obj = object_property_get_link(drc_container, prop->name, NULL);
913         if (spapr_drc_needed(obj)) {
914             return true;
915         }
916     }
917     return false;
918 }
919 
920 static void *spapr_build_fdt(SpaprMachineState *spapr, bool reset,
921                              size_t space);
922 
923 int spapr_h_cas_compose_response(SpaprMachineState *spapr,
924                                  target_ulong addr, target_ulong size,
925                                  SpaprOptionVector *ov5_updates)
926 {
927     void *fdt;
928     SpaprDeviceTreeUpdateHeader hdr = { .version_id = 1 };
929 
930     if (spapr_hotplugged_dev_before_cas()) {
931         return 1;
932     }
933 
934     if (size < sizeof(hdr)) {
935         error_report("SLOF provided insufficient CAS buffer "
936                      TARGET_FMT_lu " (min: %zu)", size, sizeof(hdr));
937         exit(EXIT_FAILURE);
938     }
939 
940     size -= sizeof(hdr);
941 
942     fdt = spapr_build_fdt(spapr, false, size);
943     _FDT((fdt_pack(fdt)));
944 
945     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
946     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
947     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
948 
949     g_free(spapr->fdt_blob);
950     spapr->fdt_size = fdt_totalsize(fdt);
951     spapr->fdt_initial_size = spapr->fdt_size;
952     spapr->fdt_blob = fdt;
953 
954     return 0;
955 }
956 
957 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
958 {
959     MachineState *ms = MACHINE(spapr);
960     int rtas;
961     GString *hypertas = g_string_sized_new(256);
962     GString *qemu_hypertas = g_string_sized_new(256);
963     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
964     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
965         memory_region_size(&MACHINE(spapr)->device_memory->mr);
966     uint32_t lrdr_capacity[] = {
967         cpu_to_be32(max_device_addr >> 32),
968         cpu_to_be32(max_device_addr & 0xffffffff),
969         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
970         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
971     };
972     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
973     uint32_t maxdomains[] = {
974         cpu_to_be32(4),
975         maxdomain,
976         maxdomain,
977         maxdomain,
978         cpu_to_be32(spapr->gpu_numa_id),
979     };
980 
981     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
982 
983     /* hypertas */
984     add_str(hypertas, "hcall-pft");
985     add_str(hypertas, "hcall-term");
986     add_str(hypertas, "hcall-dabr");
987     add_str(hypertas, "hcall-interrupt");
988     add_str(hypertas, "hcall-tce");
989     add_str(hypertas, "hcall-vio");
990     add_str(hypertas, "hcall-splpar");
991     add_str(hypertas, "hcall-join");
992     add_str(hypertas, "hcall-bulk");
993     add_str(hypertas, "hcall-set-mode");
994     add_str(hypertas, "hcall-sprg0");
995     add_str(hypertas, "hcall-copy");
996     add_str(hypertas, "hcall-debug");
997     add_str(hypertas, "hcall-vphn");
998     add_str(qemu_hypertas, "hcall-memop1");
999 
1000     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1001         add_str(hypertas, "hcall-multi-tce");
1002     }
1003 
1004     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1005         add_str(hypertas, "hcall-hpt-resize");
1006     }
1007 
1008     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1009                      hypertas->str, hypertas->len));
1010     g_string_free(hypertas, TRUE);
1011     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1012                      qemu_hypertas->str, qemu_hypertas->len));
1013     g_string_free(qemu_hypertas, TRUE);
1014 
1015     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1016                      refpoints, sizeof(refpoints)));
1017 
1018     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1019                      maxdomains, sizeof(maxdomains)));
1020 
1021     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1022                           RTAS_ERROR_LOG_MAX));
1023     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1024                           RTAS_EVENT_SCAN_RATE));
1025 
1026     g_assert(msi_nonbroken);
1027     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1028 
1029     /*
1030      * According to PAPR, rtas ibm,os-term does not guarantee a return
1031      * back to the guest cpu.
1032      *
1033      * While an additional ibm,extended-os-term property indicates
1034      * that rtas call return will always occur. Set this property.
1035      */
1036     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1037 
1038     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1039                      lrdr_capacity, sizeof(lrdr_capacity)));
1040 
1041     spapr_dt_rtas_tokens(fdt, rtas);
1042 }
1043 
1044 /*
1045  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1046  * and the XIVE features that the guest may request and thus the valid
1047  * values for bytes 23..26 of option vector 5:
1048  */
1049 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1050                                           int chosen)
1051 {
1052     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1053 
1054     char val[2 * 4] = {
1055         23, 0x00, /* XICS / XIVE mode */
1056         24, 0x00, /* Hash/Radix, filled in below. */
1057         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1058         26, 0x40, /* Radix options: GTSE == yes. */
1059     };
1060 
1061     if (spapr->irq->xics && spapr->irq->xive) {
1062         val[1] = SPAPR_OV5_XIVE_BOTH;
1063     } else if (spapr->irq->xive) {
1064         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1065     } else {
1066         assert(spapr->irq->xics);
1067         val[1] = SPAPR_OV5_XIVE_LEGACY;
1068     }
1069 
1070     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1071                           first_ppc_cpu->compat_pvr)) {
1072         /*
1073          * If we're in a pre POWER9 compat mode then the guest should
1074          * do hash and use the legacy interrupt mode
1075          */
1076         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1077         val[3] = 0x00; /* Hash */
1078     } else if (kvm_enabled()) {
1079         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1080             val[3] = 0x80; /* OV5_MMU_BOTH */
1081         } else if (kvmppc_has_cap_mmu_radix()) {
1082             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1083         } else {
1084             val[3] = 0x00; /* Hash */
1085         }
1086     } else {
1087         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1088         val[3] = 0xC0;
1089     }
1090     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1091                      val, sizeof(val)));
1092 }
1093 
1094 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1095 {
1096     MachineState *machine = MACHINE(spapr);
1097     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1098     int chosen;
1099     const char *boot_device = machine->boot_order;
1100     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1101     size_t cb = 0;
1102     char *bootlist = get_boot_devices_list(&cb);
1103 
1104     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1105 
1106     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1107         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1108                                 machine->kernel_cmdline));
1109     }
1110     if (spapr->initrd_size) {
1111         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1112                               spapr->initrd_base));
1113         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1114                               spapr->initrd_base + spapr->initrd_size));
1115     }
1116 
1117     if (spapr->kernel_size) {
1118         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1119                               cpu_to_be64(spapr->kernel_size) };
1120 
1121         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1122                          &kprop, sizeof(kprop)));
1123         if (spapr->kernel_le) {
1124             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1125         }
1126     }
1127     if (boot_menu) {
1128         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1129     }
1130     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1131     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1132     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1133 
1134     if (cb && bootlist) {
1135         int i;
1136 
1137         for (i = 0; i < cb; i++) {
1138             if (bootlist[i] == '\n') {
1139                 bootlist[i] = ' ';
1140             }
1141         }
1142         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1143     }
1144 
1145     if (boot_device && strlen(boot_device)) {
1146         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1147     }
1148 
1149     if (!spapr->has_graphics && stdout_path) {
1150         /*
1151          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1152          * kernel. New platforms should only use the "stdout-path" property. Set
1153          * the new property and continue using older property to remain
1154          * compatible with the existing firmware.
1155          */
1156         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1157         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1158     }
1159 
1160     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1161     if (smc->linux_pci_probe) {
1162         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1163     }
1164 
1165     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1166 
1167     g_free(stdout_path);
1168     g_free(bootlist);
1169 }
1170 
1171 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1172 {
1173     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1174      * KVM to work under pHyp with some guest co-operation */
1175     int hypervisor;
1176     uint8_t hypercall[16];
1177 
1178     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1179     /* indicate KVM hypercall interface */
1180     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1181     if (kvmppc_has_cap_fixup_hcalls()) {
1182         /*
1183          * Older KVM versions with older guest kernels were broken
1184          * with the magic page, don't allow the guest to map it.
1185          */
1186         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1187                                   sizeof(hypercall))) {
1188             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1189                              hypercall, sizeof(hypercall)));
1190         }
1191     }
1192 }
1193 
1194 static void *spapr_build_fdt(SpaprMachineState *spapr, bool reset,
1195                              size_t space)
1196 {
1197     MachineState *machine = MACHINE(spapr);
1198     MachineClass *mc = MACHINE_GET_CLASS(machine);
1199     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1200     int ret;
1201     void *fdt;
1202     SpaprPhbState *phb;
1203     char *buf;
1204 
1205     fdt = g_malloc0(space);
1206     _FDT((fdt_create_empty_tree(fdt, space)));
1207 
1208     /* Root node */
1209     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1210     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1211     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1212 
1213     /* Guest UUID & Name*/
1214     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1215     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1216     if (qemu_uuid_set) {
1217         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1218     }
1219     g_free(buf);
1220 
1221     if (qemu_get_vm_name()) {
1222         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1223                                 qemu_get_vm_name()));
1224     }
1225 
1226     /* Host Model & Serial Number */
1227     if (spapr->host_model) {
1228         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1229     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1230         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1231         g_free(buf);
1232     }
1233 
1234     if (spapr->host_serial) {
1235         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1236     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1237         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1238         g_free(buf);
1239     }
1240 
1241     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1242     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1243 
1244     /* /interrupt controller */
1245     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1246 
1247     ret = spapr_populate_memory(spapr, fdt);
1248     if (ret < 0) {
1249         error_report("couldn't setup memory nodes in fdt");
1250         exit(1);
1251     }
1252 
1253     /* /vdevice */
1254     spapr_dt_vdevice(spapr->vio_bus, fdt);
1255 
1256     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1257         ret = spapr_rng_populate_dt(fdt);
1258         if (ret < 0) {
1259             error_report("could not set up rng device in the fdt");
1260             exit(1);
1261         }
1262     }
1263 
1264     QLIST_FOREACH(phb, &spapr->phbs, list) {
1265         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1266         if (ret < 0) {
1267             error_report("couldn't setup PCI devices in fdt");
1268             exit(1);
1269         }
1270     }
1271 
1272     /* cpus */
1273     spapr_populate_cpus_dt_node(fdt, spapr);
1274 
1275     if (smc->dr_lmb_enabled) {
1276         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1277     }
1278 
1279     if (mc->has_hotpluggable_cpus) {
1280         int offset = fdt_path_offset(fdt, "/cpus");
1281         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1282         if (ret < 0) {
1283             error_report("Couldn't set up CPU DR device tree properties");
1284             exit(1);
1285         }
1286     }
1287 
1288     /* /event-sources */
1289     spapr_dt_events(spapr, fdt);
1290 
1291     /* /rtas */
1292     spapr_dt_rtas(spapr, fdt);
1293 
1294     /* /chosen */
1295     if (reset) {
1296         spapr_dt_chosen(spapr, fdt);
1297     }
1298 
1299     /* /hypervisor */
1300     if (kvm_enabled()) {
1301         spapr_dt_hypervisor(spapr, fdt);
1302     }
1303 
1304     /* Build memory reserve map */
1305     if (reset) {
1306         if (spapr->kernel_size) {
1307             _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1308         }
1309         if (spapr->initrd_size) {
1310             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1311                                   spapr->initrd_size)));
1312         }
1313     }
1314 
1315     /* ibm,client-architecture-support updates */
1316     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1317     if (ret < 0) {
1318         error_report("couldn't setup CAS properties fdt");
1319         exit(1);
1320     }
1321 
1322     if (smc->dr_phb_enabled) {
1323         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1324         if (ret < 0) {
1325             error_report("Couldn't set up PHB DR device tree properties");
1326             exit(1);
1327         }
1328     }
1329 
1330     return fdt;
1331 }
1332 
1333 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1334 {
1335     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1336 }
1337 
1338 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1339                                     PowerPCCPU *cpu)
1340 {
1341     CPUPPCState *env = &cpu->env;
1342 
1343     /* The TCG path should also be holding the BQL at this point */
1344     g_assert(qemu_mutex_iothread_locked());
1345 
1346     if (msr_pr) {
1347         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1348         env->gpr[3] = H_PRIVILEGE;
1349     } else {
1350         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1351     }
1352 }
1353 
1354 struct LPCRSyncState {
1355     target_ulong value;
1356     target_ulong mask;
1357 };
1358 
1359 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1360 {
1361     struct LPCRSyncState *s = arg.host_ptr;
1362     PowerPCCPU *cpu = POWERPC_CPU(cs);
1363     CPUPPCState *env = &cpu->env;
1364     target_ulong lpcr;
1365 
1366     cpu_synchronize_state(cs);
1367     lpcr = env->spr[SPR_LPCR];
1368     lpcr &= ~s->mask;
1369     lpcr |= s->value;
1370     ppc_store_lpcr(cpu, lpcr);
1371 }
1372 
1373 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1374 {
1375     CPUState *cs;
1376     struct LPCRSyncState s = {
1377         .value = value,
1378         .mask = mask
1379     };
1380     CPU_FOREACH(cs) {
1381         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1382     }
1383 }
1384 
1385 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1386 {
1387     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1388 
1389     /* Copy PATE1:GR into PATE0:HR */
1390     entry->dw0 = spapr->patb_entry & PATE0_HR;
1391     entry->dw1 = spapr->patb_entry;
1392 }
1393 
1394 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1395 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1396 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1397 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1398 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1399 
1400 /*
1401  * Get the fd to access the kernel htab, re-opening it if necessary
1402  */
1403 static int get_htab_fd(SpaprMachineState *spapr)
1404 {
1405     Error *local_err = NULL;
1406 
1407     if (spapr->htab_fd >= 0) {
1408         return spapr->htab_fd;
1409     }
1410 
1411     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1412     if (spapr->htab_fd < 0) {
1413         error_report_err(local_err);
1414     }
1415 
1416     return spapr->htab_fd;
1417 }
1418 
1419 void close_htab_fd(SpaprMachineState *spapr)
1420 {
1421     if (spapr->htab_fd >= 0) {
1422         close(spapr->htab_fd);
1423     }
1424     spapr->htab_fd = -1;
1425 }
1426 
1427 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1428 {
1429     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1430 
1431     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1432 }
1433 
1434 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1435 {
1436     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1437 
1438     assert(kvm_enabled());
1439 
1440     if (!spapr->htab) {
1441         return 0;
1442     }
1443 
1444     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1445 }
1446 
1447 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1448                                                 hwaddr ptex, int n)
1449 {
1450     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1451     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1452 
1453     if (!spapr->htab) {
1454         /*
1455          * HTAB is controlled by KVM. Fetch into temporary buffer
1456          */
1457         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1458         kvmppc_read_hptes(hptes, ptex, n);
1459         return hptes;
1460     }
1461 
1462     /*
1463      * HTAB is controlled by QEMU. Just point to the internally
1464      * accessible PTEG.
1465      */
1466     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1467 }
1468 
1469 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1470                               const ppc_hash_pte64_t *hptes,
1471                               hwaddr ptex, int n)
1472 {
1473     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1474 
1475     if (!spapr->htab) {
1476         g_free((void *)hptes);
1477     }
1478 
1479     /* Nothing to do for qemu managed HPT */
1480 }
1481 
1482 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1483                       uint64_t pte0, uint64_t pte1)
1484 {
1485     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1486     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1487 
1488     if (!spapr->htab) {
1489         kvmppc_write_hpte(ptex, pte0, pte1);
1490     } else {
1491         if (pte0 & HPTE64_V_VALID) {
1492             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1493             /*
1494              * When setting valid, we write PTE1 first. This ensures
1495              * proper synchronization with the reading code in
1496              * ppc_hash64_pteg_search()
1497              */
1498             smp_wmb();
1499             stq_p(spapr->htab + offset, pte0);
1500         } else {
1501             stq_p(spapr->htab + offset, pte0);
1502             /*
1503              * When clearing it we set PTE0 first. This ensures proper
1504              * synchronization with the reading code in
1505              * ppc_hash64_pteg_search()
1506              */
1507             smp_wmb();
1508             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1509         }
1510     }
1511 }
1512 
1513 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1514                              uint64_t pte1)
1515 {
1516     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1517     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1518 
1519     if (!spapr->htab) {
1520         /* There should always be a hash table when this is called */
1521         error_report("spapr_hpte_set_c called with no hash table !");
1522         return;
1523     }
1524 
1525     /* The HW performs a non-atomic byte update */
1526     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1527 }
1528 
1529 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1530                              uint64_t pte1)
1531 {
1532     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1533     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1534 
1535     if (!spapr->htab) {
1536         /* There should always be a hash table when this is called */
1537         error_report("spapr_hpte_set_r called with no hash table !");
1538         return;
1539     }
1540 
1541     /* The HW performs a non-atomic byte update */
1542     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1543 }
1544 
1545 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1546 {
1547     int shift;
1548 
1549     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1550      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1551      * that's much more than is needed for Linux guests */
1552     shift = ctz64(pow2ceil(ramsize)) - 7;
1553     shift = MAX(shift, 18); /* Minimum architected size */
1554     shift = MIN(shift, 46); /* Maximum architected size */
1555     return shift;
1556 }
1557 
1558 void spapr_free_hpt(SpaprMachineState *spapr)
1559 {
1560     g_free(spapr->htab);
1561     spapr->htab = NULL;
1562     spapr->htab_shift = 0;
1563     close_htab_fd(spapr);
1564 }
1565 
1566 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1567                           Error **errp)
1568 {
1569     long rc;
1570 
1571     /* Clean up any HPT info from a previous boot */
1572     spapr_free_hpt(spapr);
1573 
1574     rc = kvmppc_reset_htab(shift);
1575     if (rc < 0) {
1576         /* kernel-side HPT needed, but couldn't allocate one */
1577         error_setg_errno(errp, errno,
1578                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1579                          shift);
1580         /* This is almost certainly fatal, but if the caller really
1581          * wants to carry on with shift == 0, it's welcome to try */
1582     } else if (rc > 0) {
1583         /* kernel-side HPT allocated */
1584         if (rc != shift) {
1585             error_setg(errp,
1586                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1587                        shift, rc);
1588         }
1589 
1590         spapr->htab_shift = shift;
1591         spapr->htab = NULL;
1592     } else {
1593         /* kernel-side HPT not needed, allocate in userspace instead */
1594         size_t size = 1ULL << shift;
1595         int i;
1596 
1597         spapr->htab = qemu_memalign(size, size);
1598         if (!spapr->htab) {
1599             error_setg_errno(errp, errno,
1600                              "Could not allocate HPT of order %d", shift);
1601             return;
1602         }
1603 
1604         memset(spapr->htab, 0, size);
1605         spapr->htab_shift = shift;
1606 
1607         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1608             DIRTY_HPTE(HPTE(spapr->htab, i));
1609         }
1610     }
1611     /* We're setting up a hash table, so that means we're not radix */
1612     spapr->patb_entry = 0;
1613     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1614 }
1615 
1616 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1617 {
1618     int hpt_shift;
1619 
1620     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1621         || (spapr->cas_reboot
1622             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1623         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1624     } else {
1625         uint64_t current_ram_size;
1626 
1627         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1628         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1629     }
1630     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1631 
1632     if (spapr->vrma_adjust) {
1633         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1634                                           spapr->htab_shift);
1635     }
1636 }
1637 
1638 static int spapr_reset_drcs(Object *child, void *opaque)
1639 {
1640     SpaprDrc *drc =
1641         (SpaprDrc *) object_dynamic_cast(child,
1642                                                  TYPE_SPAPR_DR_CONNECTOR);
1643 
1644     if (drc) {
1645         spapr_drc_reset(drc);
1646     }
1647 
1648     return 0;
1649 }
1650 
1651 static void spapr_machine_reset(MachineState *machine)
1652 {
1653     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1654     PowerPCCPU *first_ppc_cpu;
1655     hwaddr fdt_addr;
1656     void *fdt;
1657     int rc;
1658 
1659     spapr_caps_apply(spapr);
1660 
1661     first_ppc_cpu = POWERPC_CPU(first_cpu);
1662     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1663         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1664                               spapr->max_compat_pvr)) {
1665         /*
1666          * If using KVM with radix mode available, VCPUs can be started
1667          * without a HPT because KVM will start them in radix mode.
1668          * Set the GR bit in PATE so that we know there is no HPT.
1669          */
1670         spapr->patb_entry = PATE1_GR;
1671         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1672     } else {
1673         spapr_setup_hpt_and_vrma(spapr);
1674     }
1675 
1676     qemu_devices_reset();
1677 
1678     /*
1679      * If this reset wasn't generated by CAS, we should reset our
1680      * negotiated options and start from scratch
1681      */
1682     if (!spapr->cas_reboot) {
1683         spapr_ovec_cleanup(spapr->ov5_cas);
1684         spapr->ov5_cas = spapr_ovec_new();
1685 
1686         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1687     }
1688 
1689     /*
1690      * This is fixing some of the default configuration of the XIVE
1691      * devices. To be called after the reset of the machine devices.
1692      */
1693     spapr_irq_reset(spapr, &error_fatal);
1694 
1695     /*
1696      * There is no CAS under qtest. Simulate one to please the code that
1697      * depends on spapr->ov5_cas. This is especially needed to test device
1698      * unplug, so we do that before resetting the DRCs.
1699      */
1700     if (qtest_enabled()) {
1701         spapr_ovec_cleanup(spapr->ov5_cas);
1702         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1703     }
1704 
1705     /* DRC reset may cause a device to be unplugged. This will cause troubles
1706      * if this device is used by another device (eg, a running vhost backend
1707      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1708      * situations, we reset DRCs after all devices have been reset.
1709      */
1710     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1711 
1712     spapr_clear_pending_events(spapr);
1713 
1714     /*
1715      * We place the device tree and RTAS just below either the top of the RMA,
1716      * or just below 2GB, whichever is lower, so that it can be
1717      * processed with 32-bit real mode code if necessary
1718      */
1719     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1720 
1721     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1722 
1723     rc = fdt_pack(fdt);
1724 
1725     /* Should only fail if we've built a corrupted tree */
1726     assert(rc == 0);
1727 
1728     /* Load the fdt */
1729     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1730     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1731     g_free(spapr->fdt_blob);
1732     spapr->fdt_size = fdt_totalsize(fdt);
1733     spapr->fdt_initial_size = spapr->fdt_size;
1734     spapr->fdt_blob = fdt;
1735 
1736     /* Set up the entry state */
1737     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1738     first_ppc_cpu->env.gpr[5] = 0;
1739 
1740     spapr->cas_reboot = false;
1741 }
1742 
1743 static void spapr_create_nvram(SpaprMachineState *spapr)
1744 {
1745     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1746     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1747 
1748     if (dinfo) {
1749         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1750                             &error_fatal);
1751     }
1752 
1753     qdev_init_nofail(dev);
1754 
1755     spapr->nvram = (struct SpaprNvram *)dev;
1756 }
1757 
1758 static void spapr_rtc_create(SpaprMachineState *spapr)
1759 {
1760     object_initialize_child(OBJECT(spapr), "rtc",
1761                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1762                             &error_fatal, NULL);
1763     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1764                               &error_fatal);
1765     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1766                               "date", &error_fatal);
1767 }
1768 
1769 /* Returns whether we want to use VGA or not */
1770 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1771 {
1772     switch (vga_interface_type) {
1773     case VGA_NONE:
1774         return false;
1775     case VGA_DEVICE:
1776         return true;
1777     case VGA_STD:
1778     case VGA_VIRTIO:
1779     case VGA_CIRRUS:
1780         return pci_vga_init(pci_bus) != NULL;
1781     default:
1782         error_setg(errp,
1783                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1784         return false;
1785     }
1786 }
1787 
1788 static int spapr_pre_load(void *opaque)
1789 {
1790     int rc;
1791 
1792     rc = spapr_caps_pre_load(opaque);
1793     if (rc) {
1794         return rc;
1795     }
1796 
1797     return 0;
1798 }
1799 
1800 static int spapr_post_load(void *opaque, int version_id)
1801 {
1802     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1803     int err = 0;
1804 
1805     err = spapr_caps_post_migration(spapr);
1806     if (err) {
1807         return err;
1808     }
1809 
1810     /*
1811      * In earlier versions, there was no separate qdev for the PAPR
1812      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1813      * So when migrating from those versions, poke the incoming offset
1814      * value into the RTC device
1815      */
1816     if (version_id < 3) {
1817         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1818         if (err) {
1819             return err;
1820         }
1821     }
1822 
1823     if (kvm_enabled() && spapr->patb_entry) {
1824         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1825         bool radix = !!(spapr->patb_entry & PATE1_GR);
1826         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1827 
1828         /*
1829          * Update LPCR:HR and UPRT as they may not be set properly in
1830          * the stream
1831          */
1832         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1833                             LPCR_HR | LPCR_UPRT);
1834 
1835         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1836         if (err) {
1837             error_report("Process table config unsupported by the host");
1838             return -EINVAL;
1839         }
1840     }
1841 
1842     err = spapr_irq_post_load(spapr, version_id);
1843     if (err) {
1844         return err;
1845     }
1846 
1847     return err;
1848 }
1849 
1850 static int spapr_pre_save(void *opaque)
1851 {
1852     int rc;
1853 
1854     rc = spapr_caps_pre_save(opaque);
1855     if (rc) {
1856         return rc;
1857     }
1858 
1859     return 0;
1860 }
1861 
1862 static bool version_before_3(void *opaque, int version_id)
1863 {
1864     return version_id < 3;
1865 }
1866 
1867 static bool spapr_pending_events_needed(void *opaque)
1868 {
1869     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1870     return !QTAILQ_EMPTY(&spapr->pending_events);
1871 }
1872 
1873 static const VMStateDescription vmstate_spapr_event_entry = {
1874     .name = "spapr_event_log_entry",
1875     .version_id = 1,
1876     .minimum_version_id = 1,
1877     .fields = (VMStateField[]) {
1878         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1879         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1880         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1881                                      NULL, extended_length),
1882         VMSTATE_END_OF_LIST()
1883     },
1884 };
1885 
1886 static const VMStateDescription vmstate_spapr_pending_events = {
1887     .name = "spapr_pending_events",
1888     .version_id = 1,
1889     .minimum_version_id = 1,
1890     .needed = spapr_pending_events_needed,
1891     .fields = (VMStateField[]) {
1892         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1893                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1894         VMSTATE_END_OF_LIST()
1895     },
1896 };
1897 
1898 static bool spapr_ov5_cas_needed(void *opaque)
1899 {
1900     SpaprMachineState *spapr = opaque;
1901     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1902     SpaprOptionVector *ov5_legacy = spapr_ovec_new();
1903     SpaprOptionVector *ov5_removed = spapr_ovec_new();
1904     bool cas_needed;
1905 
1906     /* Prior to the introduction of SpaprOptionVector, we had two option
1907      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1908      * Both of these options encode machine topology into the device-tree
1909      * in such a way that the now-booted OS should still be able to interact
1910      * appropriately with QEMU regardless of what options were actually
1911      * negotiatied on the source side.
1912      *
1913      * As such, we can avoid migrating the CAS-negotiated options if these
1914      * are the only options available on the current machine/platform.
1915      * Since these are the only options available for pseries-2.7 and
1916      * earlier, this allows us to maintain old->new/new->old migration
1917      * compatibility.
1918      *
1919      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1920      * via default pseries-2.8 machines and explicit command-line parameters.
1921      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1922      * of the actual CAS-negotiated values to continue working properly. For
1923      * example, availability of memory unplug depends on knowing whether
1924      * OV5_HP_EVT was negotiated via CAS.
1925      *
1926      * Thus, for any cases where the set of available CAS-negotiatable
1927      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1928      * include the CAS-negotiated options in the migration stream, unless
1929      * if they affect boot time behaviour only.
1930      */
1931     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1932     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1933     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1934 
1935     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1936      * the mask itself since in the future it's possible "legacy" bits may be
1937      * removed via machine options, which could generate a false positive
1938      * that breaks migration.
1939      */
1940     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1941     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1942 
1943     spapr_ovec_cleanup(ov5_mask);
1944     spapr_ovec_cleanup(ov5_legacy);
1945     spapr_ovec_cleanup(ov5_removed);
1946 
1947     return cas_needed;
1948 }
1949 
1950 static const VMStateDescription vmstate_spapr_ov5_cas = {
1951     .name = "spapr_option_vector_ov5_cas",
1952     .version_id = 1,
1953     .minimum_version_id = 1,
1954     .needed = spapr_ov5_cas_needed,
1955     .fields = (VMStateField[]) {
1956         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1957                                  vmstate_spapr_ovec, SpaprOptionVector),
1958         VMSTATE_END_OF_LIST()
1959     },
1960 };
1961 
1962 static bool spapr_patb_entry_needed(void *opaque)
1963 {
1964     SpaprMachineState *spapr = opaque;
1965 
1966     return !!spapr->patb_entry;
1967 }
1968 
1969 static const VMStateDescription vmstate_spapr_patb_entry = {
1970     .name = "spapr_patb_entry",
1971     .version_id = 1,
1972     .minimum_version_id = 1,
1973     .needed = spapr_patb_entry_needed,
1974     .fields = (VMStateField[]) {
1975         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1976         VMSTATE_END_OF_LIST()
1977     },
1978 };
1979 
1980 static bool spapr_irq_map_needed(void *opaque)
1981 {
1982     SpaprMachineState *spapr = opaque;
1983 
1984     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1985 }
1986 
1987 static const VMStateDescription vmstate_spapr_irq_map = {
1988     .name = "spapr_irq_map",
1989     .version_id = 1,
1990     .minimum_version_id = 1,
1991     .needed = spapr_irq_map_needed,
1992     .fields = (VMStateField[]) {
1993         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1994         VMSTATE_END_OF_LIST()
1995     },
1996 };
1997 
1998 static bool spapr_dtb_needed(void *opaque)
1999 {
2000     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2001 
2002     return smc->update_dt_enabled;
2003 }
2004 
2005 static int spapr_dtb_pre_load(void *opaque)
2006 {
2007     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2008 
2009     g_free(spapr->fdt_blob);
2010     spapr->fdt_blob = NULL;
2011     spapr->fdt_size = 0;
2012 
2013     return 0;
2014 }
2015 
2016 static const VMStateDescription vmstate_spapr_dtb = {
2017     .name = "spapr_dtb",
2018     .version_id = 1,
2019     .minimum_version_id = 1,
2020     .needed = spapr_dtb_needed,
2021     .pre_load = spapr_dtb_pre_load,
2022     .fields = (VMStateField[]) {
2023         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2024         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2025         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2026                                      fdt_size),
2027         VMSTATE_END_OF_LIST()
2028     },
2029 };
2030 
2031 static const VMStateDescription vmstate_spapr = {
2032     .name = "spapr",
2033     .version_id = 3,
2034     .minimum_version_id = 1,
2035     .pre_load = spapr_pre_load,
2036     .post_load = spapr_post_load,
2037     .pre_save = spapr_pre_save,
2038     .fields = (VMStateField[]) {
2039         /* used to be @next_irq */
2040         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2041 
2042         /* RTC offset */
2043         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2044 
2045         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2046         VMSTATE_END_OF_LIST()
2047     },
2048     .subsections = (const VMStateDescription*[]) {
2049         &vmstate_spapr_ov5_cas,
2050         &vmstate_spapr_patb_entry,
2051         &vmstate_spapr_pending_events,
2052         &vmstate_spapr_cap_htm,
2053         &vmstate_spapr_cap_vsx,
2054         &vmstate_spapr_cap_dfp,
2055         &vmstate_spapr_cap_cfpc,
2056         &vmstate_spapr_cap_sbbc,
2057         &vmstate_spapr_cap_ibs,
2058         &vmstate_spapr_cap_hpt_maxpagesize,
2059         &vmstate_spapr_irq_map,
2060         &vmstate_spapr_cap_nested_kvm_hv,
2061         &vmstate_spapr_dtb,
2062         &vmstate_spapr_cap_large_decr,
2063         &vmstate_spapr_cap_ccf_assist,
2064         NULL
2065     }
2066 };
2067 
2068 static int htab_save_setup(QEMUFile *f, void *opaque)
2069 {
2070     SpaprMachineState *spapr = opaque;
2071 
2072     /* "Iteration" header */
2073     if (!spapr->htab_shift) {
2074         qemu_put_be32(f, -1);
2075     } else {
2076         qemu_put_be32(f, spapr->htab_shift);
2077     }
2078 
2079     if (spapr->htab) {
2080         spapr->htab_save_index = 0;
2081         spapr->htab_first_pass = true;
2082     } else {
2083         if (spapr->htab_shift) {
2084             assert(kvm_enabled());
2085         }
2086     }
2087 
2088 
2089     return 0;
2090 }
2091 
2092 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2093                             int chunkstart, int n_valid, int n_invalid)
2094 {
2095     qemu_put_be32(f, chunkstart);
2096     qemu_put_be16(f, n_valid);
2097     qemu_put_be16(f, n_invalid);
2098     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2099                     HASH_PTE_SIZE_64 * n_valid);
2100 }
2101 
2102 static void htab_save_end_marker(QEMUFile *f)
2103 {
2104     qemu_put_be32(f, 0);
2105     qemu_put_be16(f, 0);
2106     qemu_put_be16(f, 0);
2107 }
2108 
2109 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2110                                  int64_t max_ns)
2111 {
2112     bool has_timeout = max_ns != -1;
2113     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2114     int index = spapr->htab_save_index;
2115     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2116 
2117     assert(spapr->htab_first_pass);
2118 
2119     do {
2120         int chunkstart;
2121 
2122         /* Consume invalid HPTEs */
2123         while ((index < htabslots)
2124                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2125             CLEAN_HPTE(HPTE(spapr->htab, index));
2126             index++;
2127         }
2128 
2129         /* Consume valid HPTEs */
2130         chunkstart = index;
2131         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2132                && HPTE_VALID(HPTE(spapr->htab, index))) {
2133             CLEAN_HPTE(HPTE(spapr->htab, index));
2134             index++;
2135         }
2136 
2137         if (index > chunkstart) {
2138             int n_valid = index - chunkstart;
2139 
2140             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2141 
2142             if (has_timeout &&
2143                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2144                 break;
2145             }
2146         }
2147     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2148 
2149     if (index >= htabslots) {
2150         assert(index == htabslots);
2151         index = 0;
2152         spapr->htab_first_pass = false;
2153     }
2154     spapr->htab_save_index = index;
2155 }
2156 
2157 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2158                                 int64_t max_ns)
2159 {
2160     bool final = max_ns < 0;
2161     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2162     int examined = 0, sent = 0;
2163     int index = spapr->htab_save_index;
2164     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2165 
2166     assert(!spapr->htab_first_pass);
2167 
2168     do {
2169         int chunkstart, invalidstart;
2170 
2171         /* Consume non-dirty HPTEs */
2172         while ((index < htabslots)
2173                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2174             index++;
2175             examined++;
2176         }
2177 
2178         chunkstart = index;
2179         /* Consume valid dirty HPTEs */
2180         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2181                && HPTE_DIRTY(HPTE(spapr->htab, index))
2182                && HPTE_VALID(HPTE(spapr->htab, index))) {
2183             CLEAN_HPTE(HPTE(spapr->htab, index));
2184             index++;
2185             examined++;
2186         }
2187 
2188         invalidstart = index;
2189         /* Consume invalid dirty HPTEs */
2190         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2191                && HPTE_DIRTY(HPTE(spapr->htab, index))
2192                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2193             CLEAN_HPTE(HPTE(spapr->htab, index));
2194             index++;
2195             examined++;
2196         }
2197 
2198         if (index > chunkstart) {
2199             int n_valid = invalidstart - chunkstart;
2200             int n_invalid = index - invalidstart;
2201 
2202             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2203             sent += index - chunkstart;
2204 
2205             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2206                 break;
2207             }
2208         }
2209 
2210         if (examined >= htabslots) {
2211             break;
2212         }
2213 
2214         if (index >= htabslots) {
2215             assert(index == htabslots);
2216             index = 0;
2217         }
2218     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2219 
2220     if (index >= htabslots) {
2221         assert(index == htabslots);
2222         index = 0;
2223     }
2224 
2225     spapr->htab_save_index = index;
2226 
2227     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2228 }
2229 
2230 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2231 #define MAX_KVM_BUF_SIZE    2048
2232 
2233 static int htab_save_iterate(QEMUFile *f, void *opaque)
2234 {
2235     SpaprMachineState *spapr = opaque;
2236     int fd;
2237     int rc = 0;
2238 
2239     /* Iteration header */
2240     if (!spapr->htab_shift) {
2241         qemu_put_be32(f, -1);
2242         return 1;
2243     } else {
2244         qemu_put_be32(f, 0);
2245     }
2246 
2247     if (!spapr->htab) {
2248         assert(kvm_enabled());
2249 
2250         fd = get_htab_fd(spapr);
2251         if (fd < 0) {
2252             return fd;
2253         }
2254 
2255         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2256         if (rc < 0) {
2257             return rc;
2258         }
2259     } else  if (spapr->htab_first_pass) {
2260         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2261     } else {
2262         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2263     }
2264 
2265     htab_save_end_marker(f);
2266 
2267     return rc;
2268 }
2269 
2270 static int htab_save_complete(QEMUFile *f, void *opaque)
2271 {
2272     SpaprMachineState *spapr = opaque;
2273     int fd;
2274 
2275     /* Iteration header */
2276     if (!spapr->htab_shift) {
2277         qemu_put_be32(f, -1);
2278         return 0;
2279     } else {
2280         qemu_put_be32(f, 0);
2281     }
2282 
2283     if (!spapr->htab) {
2284         int rc;
2285 
2286         assert(kvm_enabled());
2287 
2288         fd = get_htab_fd(spapr);
2289         if (fd < 0) {
2290             return fd;
2291         }
2292 
2293         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2294         if (rc < 0) {
2295             return rc;
2296         }
2297     } else {
2298         if (spapr->htab_first_pass) {
2299             htab_save_first_pass(f, spapr, -1);
2300         }
2301         htab_save_later_pass(f, spapr, -1);
2302     }
2303 
2304     /* End marker */
2305     htab_save_end_marker(f);
2306 
2307     return 0;
2308 }
2309 
2310 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2311 {
2312     SpaprMachineState *spapr = opaque;
2313     uint32_t section_hdr;
2314     int fd = -1;
2315     Error *local_err = NULL;
2316 
2317     if (version_id < 1 || version_id > 1) {
2318         error_report("htab_load() bad version");
2319         return -EINVAL;
2320     }
2321 
2322     section_hdr = qemu_get_be32(f);
2323 
2324     if (section_hdr == -1) {
2325         spapr_free_hpt(spapr);
2326         return 0;
2327     }
2328 
2329     if (section_hdr) {
2330         /* First section gives the htab size */
2331         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2332         if (local_err) {
2333             error_report_err(local_err);
2334             return -EINVAL;
2335         }
2336         return 0;
2337     }
2338 
2339     if (!spapr->htab) {
2340         assert(kvm_enabled());
2341 
2342         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2343         if (fd < 0) {
2344             error_report_err(local_err);
2345             return fd;
2346         }
2347     }
2348 
2349     while (true) {
2350         uint32_t index;
2351         uint16_t n_valid, n_invalid;
2352 
2353         index = qemu_get_be32(f);
2354         n_valid = qemu_get_be16(f);
2355         n_invalid = qemu_get_be16(f);
2356 
2357         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2358             /* End of Stream */
2359             break;
2360         }
2361 
2362         if ((index + n_valid + n_invalid) >
2363             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2364             /* Bad index in stream */
2365             error_report(
2366                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2367                 index, n_valid, n_invalid, spapr->htab_shift);
2368             return -EINVAL;
2369         }
2370 
2371         if (spapr->htab) {
2372             if (n_valid) {
2373                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2374                                 HASH_PTE_SIZE_64 * n_valid);
2375             }
2376             if (n_invalid) {
2377                 memset(HPTE(spapr->htab, index + n_valid), 0,
2378                        HASH_PTE_SIZE_64 * n_invalid);
2379             }
2380         } else {
2381             int rc;
2382 
2383             assert(fd >= 0);
2384 
2385             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2386             if (rc < 0) {
2387                 return rc;
2388             }
2389         }
2390     }
2391 
2392     if (!spapr->htab) {
2393         assert(fd >= 0);
2394         close(fd);
2395     }
2396 
2397     return 0;
2398 }
2399 
2400 static void htab_save_cleanup(void *opaque)
2401 {
2402     SpaprMachineState *spapr = opaque;
2403 
2404     close_htab_fd(spapr);
2405 }
2406 
2407 static SaveVMHandlers savevm_htab_handlers = {
2408     .save_setup = htab_save_setup,
2409     .save_live_iterate = htab_save_iterate,
2410     .save_live_complete_precopy = htab_save_complete,
2411     .save_cleanup = htab_save_cleanup,
2412     .load_state = htab_load,
2413 };
2414 
2415 static void spapr_boot_set(void *opaque, const char *boot_device,
2416                            Error **errp)
2417 {
2418     MachineState *machine = MACHINE(opaque);
2419     machine->boot_order = g_strdup(boot_device);
2420 }
2421 
2422 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2423 {
2424     MachineState *machine = MACHINE(spapr);
2425     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2426     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2427     int i;
2428 
2429     for (i = 0; i < nr_lmbs; i++) {
2430         uint64_t addr;
2431 
2432         addr = i * lmb_size + machine->device_memory->base;
2433         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2434                                addr / lmb_size);
2435     }
2436 }
2437 
2438 /*
2439  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2440  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2441  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2442  */
2443 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2444 {
2445     int i;
2446 
2447     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2448         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2449                    " is not aligned to %" PRIu64 " MiB",
2450                    machine->ram_size,
2451                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2452         return;
2453     }
2454 
2455     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2456         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2457                    " is not aligned to %" PRIu64 " MiB",
2458                    machine->ram_size,
2459                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2460         return;
2461     }
2462 
2463     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2464         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2465             error_setg(errp,
2466                        "Node %d memory size 0x%" PRIx64
2467                        " is not aligned to %" PRIu64 " MiB",
2468                        i, machine->numa_state->nodes[i].node_mem,
2469                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2470             return;
2471         }
2472     }
2473 }
2474 
2475 /* find cpu slot in machine->possible_cpus by core_id */
2476 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2477 {
2478     int index = id / ms->smp.threads;
2479 
2480     if (index >= ms->possible_cpus->len) {
2481         return NULL;
2482     }
2483     if (idx) {
2484         *idx = index;
2485     }
2486     return &ms->possible_cpus->cpus[index];
2487 }
2488 
2489 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2490 {
2491     MachineState *ms = MACHINE(spapr);
2492     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2493     Error *local_err = NULL;
2494     bool vsmt_user = !!spapr->vsmt;
2495     int kvm_smt = kvmppc_smt_threads();
2496     int ret;
2497     unsigned int smp_threads = ms->smp.threads;
2498 
2499     if (!kvm_enabled() && (smp_threads > 1)) {
2500         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2501                      "on a pseries machine");
2502         goto out;
2503     }
2504     if (!is_power_of_2(smp_threads)) {
2505         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2506                      "machine because it must be a power of 2", smp_threads);
2507         goto out;
2508     }
2509 
2510     /* Detemine the VSMT mode to use: */
2511     if (vsmt_user) {
2512         if (spapr->vsmt < smp_threads) {
2513             error_setg(&local_err, "Cannot support VSMT mode %d"
2514                          " because it must be >= threads/core (%d)",
2515                          spapr->vsmt, smp_threads);
2516             goto out;
2517         }
2518         /* In this case, spapr->vsmt has been set by the command line */
2519     } else if (!smc->smp_threads_vsmt) {
2520         /*
2521          * Default VSMT value is tricky, because we need it to be as
2522          * consistent as possible (for migration), but this requires
2523          * changing it for at least some existing cases.  We pick 8 as
2524          * the value that we'd get with KVM on POWER8, the
2525          * overwhelmingly common case in production systems.
2526          */
2527         spapr->vsmt = MAX(8, smp_threads);
2528     } else {
2529         spapr->vsmt = smp_threads;
2530     }
2531 
2532     /* KVM: If necessary, set the SMT mode: */
2533     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2534         ret = kvmppc_set_smt_threads(spapr->vsmt);
2535         if (ret) {
2536             /* Looks like KVM isn't able to change VSMT mode */
2537             error_setg(&local_err,
2538                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2539                        spapr->vsmt, ret);
2540             /* We can live with that if the default one is big enough
2541              * for the number of threads, and a submultiple of the one
2542              * we want.  In this case we'll waste some vcpu ids, but
2543              * behaviour will be correct */
2544             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2545                 warn_report_err(local_err);
2546                 local_err = NULL;
2547                 goto out;
2548             } else {
2549                 if (!vsmt_user) {
2550                     error_append_hint(&local_err,
2551                                       "On PPC, a VM with %d threads/core"
2552                                       " on a host with %d threads/core"
2553                                       " requires the use of VSMT mode %d.\n",
2554                                       smp_threads, kvm_smt, spapr->vsmt);
2555                 }
2556                 kvmppc_error_append_smt_possible_hint(&local_err);
2557                 goto out;
2558             }
2559         }
2560     }
2561     /* else TCG: nothing to do currently */
2562 out:
2563     error_propagate(errp, local_err);
2564 }
2565 
2566 static void spapr_init_cpus(SpaprMachineState *spapr)
2567 {
2568     MachineState *machine = MACHINE(spapr);
2569     MachineClass *mc = MACHINE_GET_CLASS(machine);
2570     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2571     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2572     const CPUArchIdList *possible_cpus;
2573     unsigned int smp_cpus = machine->smp.cpus;
2574     unsigned int smp_threads = machine->smp.threads;
2575     unsigned int max_cpus = machine->smp.max_cpus;
2576     int boot_cores_nr = smp_cpus / smp_threads;
2577     int i;
2578 
2579     possible_cpus = mc->possible_cpu_arch_ids(machine);
2580     if (mc->has_hotpluggable_cpus) {
2581         if (smp_cpus % smp_threads) {
2582             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2583                          smp_cpus, smp_threads);
2584             exit(1);
2585         }
2586         if (max_cpus % smp_threads) {
2587             error_report("max_cpus (%u) must be multiple of threads (%u)",
2588                          max_cpus, smp_threads);
2589             exit(1);
2590         }
2591     } else {
2592         if (max_cpus != smp_cpus) {
2593             error_report("This machine version does not support CPU hotplug");
2594             exit(1);
2595         }
2596         boot_cores_nr = possible_cpus->len;
2597     }
2598 
2599     if (smc->pre_2_10_has_unused_icps) {
2600         int i;
2601 
2602         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2603             /* Dummy entries get deregistered when real ICPState objects
2604              * are registered during CPU core hotplug.
2605              */
2606             pre_2_10_vmstate_register_dummy_icp(i);
2607         }
2608     }
2609 
2610     for (i = 0; i < possible_cpus->len; i++) {
2611         int core_id = i * smp_threads;
2612 
2613         if (mc->has_hotpluggable_cpus) {
2614             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2615                                    spapr_vcpu_id(spapr, core_id));
2616         }
2617 
2618         if (i < boot_cores_nr) {
2619             Object *core  = object_new(type);
2620             int nr_threads = smp_threads;
2621 
2622             /* Handle the partially filled core for older machine types */
2623             if ((i + 1) * smp_threads >= smp_cpus) {
2624                 nr_threads = smp_cpus - i * smp_threads;
2625             }
2626 
2627             object_property_set_int(core, nr_threads, "nr-threads",
2628                                     &error_fatal);
2629             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2630                                     &error_fatal);
2631             object_property_set_bool(core, true, "realized", &error_fatal);
2632 
2633             object_unref(core);
2634         }
2635     }
2636 }
2637 
2638 static PCIHostState *spapr_create_default_phb(void)
2639 {
2640     DeviceState *dev;
2641 
2642     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2643     qdev_prop_set_uint32(dev, "index", 0);
2644     qdev_init_nofail(dev);
2645 
2646     return PCI_HOST_BRIDGE(dev);
2647 }
2648 
2649 /* pSeries LPAR / sPAPR hardware init */
2650 static void spapr_machine_init(MachineState *machine)
2651 {
2652     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2653     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2654     const char *kernel_filename = machine->kernel_filename;
2655     const char *initrd_filename = machine->initrd_filename;
2656     PCIHostState *phb;
2657     int i;
2658     MemoryRegion *sysmem = get_system_memory();
2659     MemoryRegion *ram = g_new(MemoryRegion, 1);
2660     hwaddr node0_size = spapr_node0_size(machine);
2661     long load_limit, fw_size;
2662     char *filename;
2663     Error *resize_hpt_err = NULL;
2664 
2665     msi_nonbroken = true;
2666 
2667     QLIST_INIT(&spapr->phbs);
2668     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2669 
2670     /* Determine capabilities to run with */
2671     spapr_caps_init(spapr);
2672 
2673     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2674     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2675         /*
2676          * If the user explicitly requested a mode we should either
2677          * supply it, or fail completely (which we do below).  But if
2678          * it's not set explicitly, we reset our mode to something
2679          * that works
2680          */
2681         if (resize_hpt_err) {
2682             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2683             error_free(resize_hpt_err);
2684             resize_hpt_err = NULL;
2685         } else {
2686             spapr->resize_hpt = smc->resize_hpt_default;
2687         }
2688     }
2689 
2690     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2691 
2692     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2693         /*
2694          * User requested HPT resize, but this host can't supply it.  Bail out
2695          */
2696         error_report_err(resize_hpt_err);
2697         exit(1);
2698     }
2699 
2700     spapr->rma_size = node0_size;
2701 
2702     /* With KVM, we don't actually know whether KVM supports an
2703      * unbounded RMA (PR KVM) or is limited by the hash table size
2704      * (HV KVM using VRMA), so we always assume the latter
2705      *
2706      * In that case, we also limit the initial allocations for RTAS
2707      * etc... to 256M since we have no way to know what the VRMA size
2708      * is going to be as it depends on the size of the hash table
2709      * which isn't determined yet.
2710      */
2711     if (kvm_enabled()) {
2712         spapr->vrma_adjust = 1;
2713         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2714     }
2715 
2716     /* Actually we don't support unbounded RMA anymore since we added
2717      * proper emulation of HV mode. The max we can get is 16G which
2718      * also happens to be what we configure for PAPR mode so make sure
2719      * we don't do anything bigger than that
2720      */
2721     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2722 
2723     if (spapr->rma_size > node0_size) {
2724         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2725                      spapr->rma_size);
2726         exit(1);
2727     }
2728 
2729     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2730     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2731 
2732     /*
2733      * VSMT must be set in order to be able to compute VCPU ids, ie to
2734      * call spapr_max_server_number() or spapr_vcpu_id().
2735      */
2736     spapr_set_vsmt_mode(spapr, &error_fatal);
2737 
2738     /* Set up Interrupt Controller before we create the VCPUs */
2739     spapr_irq_init(spapr, &error_fatal);
2740 
2741     /* Set up containers for ibm,client-architecture-support negotiated options
2742      */
2743     spapr->ov5 = spapr_ovec_new();
2744     spapr->ov5_cas = spapr_ovec_new();
2745 
2746     if (smc->dr_lmb_enabled) {
2747         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2748         spapr_validate_node_memory(machine, &error_fatal);
2749     }
2750 
2751     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2752 
2753     /* advertise support for dedicated HP event source to guests */
2754     if (spapr->use_hotplug_event_source) {
2755         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2756     }
2757 
2758     /* advertise support for HPT resizing */
2759     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2760         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2761     }
2762 
2763     /* advertise support for ibm,dyamic-memory-v2 */
2764     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2765 
2766     /* advertise XIVE on POWER9 machines */
2767     if (spapr->irq->xive) {
2768         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2769     }
2770 
2771     /* init CPUs */
2772     spapr_init_cpus(spapr);
2773 
2774     /*
2775      * check we don't have a memory-less/cpu-less NUMA node
2776      * Firmware relies on the existing memory/cpu topology to provide the
2777      * NUMA topology to the kernel.
2778      * And the linux kernel needs to know the NUMA topology at start
2779      * to be able to hotplug CPUs later.
2780      */
2781     if (machine->numa_state->num_nodes) {
2782         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2783             /* check for memory-less node */
2784             if (machine->numa_state->nodes[i].node_mem == 0) {
2785                 CPUState *cs;
2786                 int found = 0;
2787                 /* check for cpu-less node */
2788                 CPU_FOREACH(cs) {
2789                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2790                     if (cpu->node_id == i) {
2791                         found = 1;
2792                         break;
2793                     }
2794                 }
2795                 /* memory-less and cpu-less node */
2796                 if (!found) {
2797                     error_report(
2798                        "Memory-less/cpu-less nodes are not supported (node %d)",
2799                                  i);
2800                     exit(1);
2801                 }
2802             }
2803         }
2804 
2805     }
2806 
2807     /*
2808      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2809      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2810      * called from vPHB reset handler so we initialize the counter here.
2811      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2812      * must be equally distant from any other node.
2813      * The final value of spapr->gpu_numa_id is going to be written to
2814      * max-associativity-domains in spapr_build_fdt().
2815      */
2816     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2817 
2818     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2819         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2820                               spapr->max_compat_pvr)) {
2821         /* KVM and TCG always allow GTSE with radix... */
2822         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2823     }
2824     /* ... but not with hash (currently). */
2825 
2826     if (kvm_enabled()) {
2827         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2828         kvmppc_enable_logical_ci_hcalls();
2829         kvmppc_enable_set_mode_hcall();
2830 
2831         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2832         kvmppc_enable_clear_ref_mod_hcalls();
2833 
2834         /* Enable H_PAGE_INIT */
2835         kvmppc_enable_h_page_init();
2836     }
2837 
2838     /* allocate RAM */
2839     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2840                                          machine->ram_size);
2841     memory_region_add_subregion(sysmem, 0, ram);
2842 
2843     /* always allocate the device memory information */
2844     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2845 
2846     /* initialize hotplug memory address space */
2847     if (machine->ram_size < machine->maxram_size) {
2848         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2849         /*
2850          * Limit the number of hotpluggable memory slots to half the number
2851          * slots that KVM supports, leaving the other half for PCI and other
2852          * devices. However ensure that number of slots doesn't drop below 32.
2853          */
2854         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2855                            SPAPR_MAX_RAM_SLOTS;
2856 
2857         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2858             max_memslots = SPAPR_MAX_RAM_SLOTS;
2859         }
2860         if (machine->ram_slots > max_memslots) {
2861             error_report("Specified number of memory slots %"
2862                          PRIu64" exceeds max supported %d",
2863                          machine->ram_slots, max_memslots);
2864             exit(1);
2865         }
2866 
2867         machine->device_memory->base = ROUND_UP(machine->ram_size,
2868                                                 SPAPR_DEVICE_MEM_ALIGN);
2869         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2870                            "device-memory", device_mem_size);
2871         memory_region_add_subregion(sysmem, machine->device_memory->base,
2872                                     &machine->device_memory->mr);
2873     }
2874 
2875     if (smc->dr_lmb_enabled) {
2876         spapr_create_lmb_dr_connectors(spapr);
2877     }
2878 
2879     /* Set up RTAS event infrastructure */
2880     spapr_events_init(spapr);
2881 
2882     /* Set up the RTC RTAS interfaces */
2883     spapr_rtc_create(spapr);
2884 
2885     /* Set up VIO bus */
2886     spapr->vio_bus = spapr_vio_bus_init();
2887 
2888     for (i = 0; i < serial_max_hds(); i++) {
2889         if (serial_hd(i)) {
2890             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2891         }
2892     }
2893 
2894     /* We always have at least the nvram device on VIO */
2895     spapr_create_nvram(spapr);
2896 
2897     /*
2898      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2899      * connectors (described in root DT node's "ibm,drc-types" property)
2900      * are pre-initialized here. additional child connectors (such as
2901      * connectors for a PHBs PCI slots) are added as needed during their
2902      * parent's realization.
2903      */
2904     if (smc->dr_phb_enabled) {
2905         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2906             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2907         }
2908     }
2909 
2910     /* Set up PCI */
2911     spapr_pci_rtas_init();
2912 
2913     phb = spapr_create_default_phb();
2914 
2915     for (i = 0; i < nb_nics; i++) {
2916         NICInfo *nd = &nd_table[i];
2917 
2918         if (!nd->model) {
2919             nd->model = g_strdup("spapr-vlan");
2920         }
2921 
2922         if (g_str_equal(nd->model, "spapr-vlan") ||
2923             g_str_equal(nd->model, "ibmveth")) {
2924             spapr_vlan_create(spapr->vio_bus, nd);
2925         } else {
2926             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2927         }
2928     }
2929 
2930     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2931         spapr_vscsi_create(spapr->vio_bus);
2932     }
2933 
2934     /* Graphics */
2935     if (spapr_vga_init(phb->bus, &error_fatal)) {
2936         spapr->has_graphics = true;
2937         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2938     }
2939 
2940     if (machine->usb) {
2941         if (smc->use_ohci_by_default) {
2942             pci_create_simple(phb->bus, -1, "pci-ohci");
2943         } else {
2944             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2945         }
2946 
2947         if (spapr->has_graphics) {
2948             USBBus *usb_bus = usb_bus_find(-1);
2949 
2950             usb_create_simple(usb_bus, "usb-kbd");
2951             usb_create_simple(usb_bus, "usb-mouse");
2952         }
2953     }
2954 
2955     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2956         error_report(
2957             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2958             MIN_RMA_SLOF);
2959         exit(1);
2960     }
2961 
2962     if (kernel_filename) {
2963         uint64_t lowaddr = 0;
2964 
2965         spapr->kernel_size = load_elf(kernel_filename, NULL,
2966                                       translate_kernel_address, NULL,
2967                                       NULL, &lowaddr, NULL, 1,
2968                                       PPC_ELF_MACHINE, 0, 0);
2969         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2970             spapr->kernel_size = load_elf(kernel_filename, NULL,
2971                                           translate_kernel_address, NULL, NULL,
2972                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2973                                           0, 0);
2974             spapr->kernel_le = spapr->kernel_size > 0;
2975         }
2976         if (spapr->kernel_size < 0) {
2977             error_report("error loading %s: %s", kernel_filename,
2978                          load_elf_strerror(spapr->kernel_size));
2979             exit(1);
2980         }
2981 
2982         /* load initrd */
2983         if (initrd_filename) {
2984             /* Try to locate the initrd in the gap between the kernel
2985              * and the firmware. Add a bit of space just in case
2986              */
2987             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2988                                   + 0x1ffff) & ~0xffff;
2989             spapr->initrd_size = load_image_targphys(initrd_filename,
2990                                                      spapr->initrd_base,
2991                                                      load_limit
2992                                                      - spapr->initrd_base);
2993             if (spapr->initrd_size < 0) {
2994                 error_report("could not load initial ram disk '%s'",
2995                              initrd_filename);
2996                 exit(1);
2997             }
2998         }
2999     }
3000 
3001     if (bios_name == NULL) {
3002         bios_name = FW_FILE_NAME;
3003     }
3004     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3005     if (!filename) {
3006         error_report("Could not find LPAR firmware '%s'", bios_name);
3007         exit(1);
3008     }
3009     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3010     if (fw_size <= 0) {
3011         error_report("Could not load LPAR firmware '%s'", filename);
3012         exit(1);
3013     }
3014     g_free(filename);
3015 
3016     /* FIXME: Should register things through the MachineState's qdev
3017      * interface, this is a legacy from the sPAPREnvironment structure
3018      * which predated MachineState but had a similar function */
3019     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3020     register_savevm_live("spapr/htab", -1, 1,
3021                          &savevm_htab_handlers, spapr);
3022 
3023     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3024                              &error_fatal);
3025 
3026     qemu_register_boot_set(spapr_boot_set, spapr);
3027 
3028     /*
3029      * Nothing needs to be done to resume a suspended guest because
3030      * suspending does not change the machine state, so no need for
3031      * a ->wakeup method.
3032      */
3033     qemu_register_wakeup_support();
3034 
3035     if (kvm_enabled()) {
3036         /* to stop and start vmclock */
3037         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3038                                          &spapr->tb);
3039 
3040         kvmppc_spapr_enable_inkernel_multitce();
3041     }
3042 }
3043 
3044 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3045 {
3046     if (!vm_type) {
3047         return 0;
3048     }
3049 
3050     if (!strcmp(vm_type, "HV")) {
3051         return 1;
3052     }
3053 
3054     if (!strcmp(vm_type, "PR")) {
3055         return 2;
3056     }
3057 
3058     error_report("Unknown kvm-type specified '%s'", vm_type);
3059     exit(1);
3060 }
3061 
3062 /*
3063  * Implementation of an interface to adjust firmware path
3064  * for the bootindex property handling.
3065  */
3066 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3067                                    DeviceState *dev)
3068 {
3069 #define CAST(type, obj, name) \
3070     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3071     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3072     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3073     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3074 
3075     if (d) {
3076         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3077         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3078         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3079 
3080         if (spapr) {
3081             /*
3082              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3083              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3084              * 0x8000 | (target << 8) | (bus << 5) | lun
3085              * (see the "Logical unit addressing format" table in SAM5)
3086              */
3087             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3088             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3089                                    (uint64_t)id << 48);
3090         } else if (virtio) {
3091             /*
3092              * We use SRP luns of the form 01000000 | (target << 8) | lun
3093              * in the top 32 bits of the 64-bit LUN
3094              * Note: the quote above is from SLOF and it is wrong,
3095              * the actual binding is:
3096              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3097              */
3098             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3099             if (d->lun >= 256) {
3100                 /* Use the LUN "flat space addressing method" */
3101                 id |= 0x4000;
3102             }
3103             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3104                                    (uint64_t)id << 32);
3105         } else if (usb) {
3106             /*
3107              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3108              * in the top 32 bits of the 64-bit LUN
3109              */
3110             unsigned usb_port = atoi(usb->port->path);
3111             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3112             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3113                                    (uint64_t)id << 32);
3114         }
3115     }
3116 
3117     /*
3118      * SLOF probes the USB devices, and if it recognizes that the device is a
3119      * storage device, it changes its name to "storage" instead of "usb-host",
3120      * and additionally adds a child node for the SCSI LUN, so the correct
3121      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3122      */
3123     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3124         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3125         if (usb_host_dev_is_scsi_storage(usbdev)) {
3126             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3127         }
3128     }
3129 
3130     if (phb) {
3131         /* Replace "pci" with "pci@800000020000000" */
3132         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3133     }
3134 
3135     if (vsc) {
3136         /* Same logic as virtio above */
3137         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3138         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3139     }
3140 
3141     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3142         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3143         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3144         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3145     }
3146 
3147     return NULL;
3148 }
3149 
3150 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3151 {
3152     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3153 
3154     return g_strdup(spapr->kvm_type);
3155 }
3156 
3157 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     g_free(spapr->kvm_type);
3162     spapr->kvm_type = g_strdup(value);
3163 }
3164 
3165 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3166 {
3167     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3168 
3169     return spapr->use_hotplug_event_source;
3170 }
3171 
3172 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3173                                             Error **errp)
3174 {
3175     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3176 
3177     spapr->use_hotplug_event_source = value;
3178 }
3179 
3180 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3181 {
3182     return true;
3183 }
3184 
3185 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3186 {
3187     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3188 
3189     switch (spapr->resize_hpt) {
3190     case SPAPR_RESIZE_HPT_DEFAULT:
3191         return g_strdup("default");
3192     case SPAPR_RESIZE_HPT_DISABLED:
3193         return g_strdup("disabled");
3194     case SPAPR_RESIZE_HPT_ENABLED:
3195         return g_strdup("enabled");
3196     case SPAPR_RESIZE_HPT_REQUIRED:
3197         return g_strdup("required");
3198     }
3199     g_assert_not_reached();
3200 }
3201 
3202 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3203 {
3204     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3205 
3206     if (strcmp(value, "default") == 0) {
3207         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3208     } else if (strcmp(value, "disabled") == 0) {
3209         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3210     } else if (strcmp(value, "enabled") == 0) {
3211         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3212     } else if (strcmp(value, "required") == 0) {
3213         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3214     } else {
3215         error_setg(errp, "Bad value for \"resize-hpt\" property");
3216     }
3217 }
3218 
3219 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3220                                    void *opaque, Error **errp)
3221 {
3222     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3223 }
3224 
3225 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3226                                    void *opaque, Error **errp)
3227 {
3228     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3229 }
3230 
3231 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3232 {
3233     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3234 
3235     if (spapr->irq == &spapr_irq_xics_legacy) {
3236         return g_strdup("legacy");
3237     } else if (spapr->irq == &spapr_irq_xics) {
3238         return g_strdup("xics");
3239     } else if (spapr->irq == &spapr_irq_xive) {
3240         return g_strdup("xive");
3241     } else if (spapr->irq == &spapr_irq_dual) {
3242         return g_strdup("dual");
3243     }
3244     g_assert_not_reached();
3245 }
3246 
3247 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3248 {
3249     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3250 
3251     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3252         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3253         return;
3254     }
3255 
3256     /* The legacy IRQ backend can not be set */
3257     if (strcmp(value, "xics") == 0) {
3258         spapr->irq = &spapr_irq_xics;
3259     } else if (strcmp(value, "xive") == 0) {
3260         spapr->irq = &spapr_irq_xive;
3261     } else if (strcmp(value, "dual") == 0) {
3262         spapr->irq = &spapr_irq_dual;
3263     } else {
3264         error_setg(errp, "Bad value for \"ic-mode\" property");
3265     }
3266 }
3267 
3268 static char *spapr_get_host_model(Object *obj, Error **errp)
3269 {
3270     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3271 
3272     return g_strdup(spapr->host_model);
3273 }
3274 
3275 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3276 {
3277     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3278 
3279     g_free(spapr->host_model);
3280     spapr->host_model = g_strdup(value);
3281 }
3282 
3283 static char *spapr_get_host_serial(Object *obj, Error **errp)
3284 {
3285     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3286 
3287     return g_strdup(spapr->host_serial);
3288 }
3289 
3290 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3291 {
3292     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3293 
3294     g_free(spapr->host_serial);
3295     spapr->host_serial = g_strdup(value);
3296 }
3297 
3298 static void spapr_instance_init(Object *obj)
3299 {
3300     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3301     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3302 
3303     spapr->htab_fd = -1;
3304     spapr->use_hotplug_event_source = true;
3305     object_property_add_str(obj, "kvm-type",
3306                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3307     object_property_set_description(obj, "kvm-type",
3308                                     "Specifies the KVM virtualization mode (HV, PR)",
3309                                     NULL);
3310     object_property_add_bool(obj, "modern-hotplug-events",
3311                             spapr_get_modern_hotplug_events,
3312                             spapr_set_modern_hotplug_events,
3313                             NULL);
3314     object_property_set_description(obj, "modern-hotplug-events",
3315                                     "Use dedicated hotplug event mechanism in"
3316                                     " place of standard EPOW events when possible"
3317                                     " (required for memory hot-unplug support)",
3318                                     NULL);
3319     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3320                             "Maximum permitted CPU compatibility mode",
3321                             &error_fatal);
3322 
3323     object_property_add_str(obj, "resize-hpt",
3324                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3325     object_property_set_description(obj, "resize-hpt",
3326                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3327                                     NULL);
3328     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3329                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3330     object_property_set_description(obj, "vsmt",
3331                                     "Virtual SMT: KVM behaves as if this were"
3332                                     " the host's SMT mode", &error_abort);
3333     object_property_add_bool(obj, "vfio-no-msix-emulation",
3334                              spapr_get_msix_emulation, NULL, NULL);
3335 
3336     /* The machine class defines the default interrupt controller mode */
3337     spapr->irq = smc->irq;
3338     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3339                             spapr_set_ic_mode, NULL);
3340     object_property_set_description(obj, "ic-mode",
3341                  "Specifies the interrupt controller mode (xics, xive, dual)",
3342                  NULL);
3343 
3344     object_property_add_str(obj, "host-model",
3345         spapr_get_host_model, spapr_set_host_model,
3346         &error_abort);
3347     object_property_set_description(obj, "host-model",
3348         "Host model to advertise in guest device tree", &error_abort);
3349     object_property_add_str(obj, "host-serial",
3350         spapr_get_host_serial, spapr_set_host_serial,
3351         &error_abort);
3352     object_property_set_description(obj, "host-serial",
3353         "Host serial number to advertise in guest device tree", &error_abort);
3354 }
3355 
3356 static void spapr_machine_finalizefn(Object *obj)
3357 {
3358     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3359 
3360     g_free(spapr->kvm_type);
3361 }
3362 
3363 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3364 {
3365     cpu_synchronize_state(cs);
3366     ppc_cpu_do_system_reset(cs);
3367 }
3368 
3369 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3370 {
3371     CPUState *cs;
3372 
3373     CPU_FOREACH(cs) {
3374         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3375     }
3376 }
3377 
3378 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3379                           void *fdt, int *fdt_start_offset, Error **errp)
3380 {
3381     uint64_t addr;
3382     uint32_t node;
3383 
3384     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3385     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3386                                     &error_abort);
3387     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3388                                                    SPAPR_MEMORY_BLOCK_SIZE);
3389     return 0;
3390 }
3391 
3392 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3393                            bool dedicated_hp_event_source, Error **errp)
3394 {
3395     SpaprDrc *drc;
3396     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3397     int i;
3398     uint64_t addr = addr_start;
3399     bool hotplugged = spapr_drc_hotplugged(dev);
3400     Error *local_err = NULL;
3401 
3402     for (i = 0; i < nr_lmbs; i++) {
3403         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3404                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3405         g_assert(drc);
3406 
3407         spapr_drc_attach(drc, dev, &local_err);
3408         if (local_err) {
3409             while (addr > addr_start) {
3410                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3411                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3412                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3413                 spapr_drc_detach(drc);
3414             }
3415             error_propagate(errp, local_err);
3416             return;
3417         }
3418         if (!hotplugged) {
3419             spapr_drc_reset(drc);
3420         }
3421         addr += SPAPR_MEMORY_BLOCK_SIZE;
3422     }
3423     /* send hotplug notification to the
3424      * guest only in case of hotplugged memory
3425      */
3426     if (hotplugged) {
3427         if (dedicated_hp_event_source) {
3428             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3429                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3430             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3431                                                    nr_lmbs,
3432                                                    spapr_drc_index(drc));
3433         } else {
3434             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3435                                            nr_lmbs);
3436         }
3437     }
3438 }
3439 
3440 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3441                               Error **errp)
3442 {
3443     Error *local_err = NULL;
3444     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3445     PCDIMMDevice *dimm = PC_DIMM(dev);
3446     uint64_t size, addr;
3447 
3448     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3449 
3450     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3451     if (local_err) {
3452         goto out;
3453     }
3454 
3455     addr = object_property_get_uint(OBJECT(dimm),
3456                                     PC_DIMM_ADDR_PROP, &local_err);
3457     if (local_err) {
3458         goto out_unplug;
3459     }
3460 
3461     spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3462                    &local_err);
3463     if (local_err) {
3464         goto out_unplug;
3465     }
3466 
3467     return;
3468 
3469 out_unplug:
3470     pc_dimm_unplug(dimm, MACHINE(ms));
3471 out:
3472     error_propagate(errp, local_err);
3473 }
3474 
3475 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3476                                   Error **errp)
3477 {
3478     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3479     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3480     PCDIMMDevice *dimm = PC_DIMM(dev);
3481     Error *local_err = NULL;
3482     uint64_t size;
3483     Object *memdev;
3484     hwaddr pagesize;
3485 
3486     if (!smc->dr_lmb_enabled) {
3487         error_setg(errp, "Memory hotplug not supported for this machine");
3488         return;
3489     }
3490 
3491     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3492     if (local_err) {
3493         error_propagate(errp, local_err);
3494         return;
3495     }
3496 
3497     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3498         error_setg(errp, "Hotplugged memory size must be a multiple of "
3499                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3500         return;
3501     }
3502 
3503     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3504                                       &error_abort);
3505     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3506     spapr_check_pagesize(spapr, pagesize, &local_err);
3507     if (local_err) {
3508         error_propagate(errp, local_err);
3509         return;
3510     }
3511 
3512     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3513 }
3514 
3515 struct SpaprDimmState {
3516     PCDIMMDevice *dimm;
3517     uint32_t nr_lmbs;
3518     QTAILQ_ENTRY(SpaprDimmState) next;
3519 };
3520 
3521 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3522                                                        PCDIMMDevice *dimm)
3523 {
3524     SpaprDimmState *dimm_state = NULL;
3525 
3526     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3527         if (dimm_state->dimm == dimm) {
3528             break;
3529         }
3530     }
3531     return dimm_state;
3532 }
3533 
3534 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3535                                                       uint32_t nr_lmbs,
3536                                                       PCDIMMDevice *dimm)
3537 {
3538     SpaprDimmState *ds = NULL;
3539 
3540     /*
3541      * If this request is for a DIMM whose removal had failed earlier
3542      * (due to guest's refusal to remove the LMBs), we would have this
3543      * dimm already in the pending_dimm_unplugs list. In that
3544      * case don't add again.
3545      */
3546     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3547     if (!ds) {
3548         ds = g_malloc0(sizeof(SpaprDimmState));
3549         ds->nr_lmbs = nr_lmbs;
3550         ds->dimm = dimm;
3551         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3552     }
3553     return ds;
3554 }
3555 
3556 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3557                                               SpaprDimmState *dimm_state)
3558 {
3559     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3560     g_free(dimm_state);
3561 }
3562 
3563 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3564                                                         PCDIMMDevice *dimm)
3565 {
3566     SpaprDrc *drc;
3567     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3568                                                   &error_abort);
3569     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3570     uint32_t avail_lmbs = 0;
3571     uint64_t addr_start, addr;
3572     int i;
3573 
3574     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3575                                          &error_abort);
3576 
3577     addr = addr_start;
3578     for (i = 0; i < nr_lmbs; i++) {
3579         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3580                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3581         g_assert(drc);
3582         if (drc->dev) {
3583             avail_lmbs++;
3584         }
3585         addr += SPAPR_MEMORY_BLOCK_SIZE;
3586     }
3587 
3588     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3589 }
3590 
3591 /* Callback to be called during DRC release. */
3592 void spapr_lmb_release(DeviceState *dev)
3593 {
3594     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3595     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3596     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3597 
3598     /* This information will get lost if a migration occurs
3599      * during the unplug process. In this case recover it. */
3600     if (ds == NULL) {
3601         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3602         g_assert(ds);
3603         /* The DRC being examined by the caller at least must be counted */
3604         g_assert(ds->nr_lmbs);
3605     }
3606 
3607     if (--ds->nr_lmbs) {
3608         return;
3609     }
3610 
3611     /*
3612      * Now that all the LMBs have been removed by the guest, call the
3613      * unplug handler chain. This can never fail.
3614      */
3615     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3616     object_unparent(OBJECT(dev));
3617 }
3618 
3619 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3620 {
3621     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3622     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3623 
3624     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3625     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3626     spapr_pending_dimm_unplugs_remove(spapr, ds);
3627 }
3628 
3629 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3630                                         DeviceState *dev, Error **errp)
3631 {
3632     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3633     Error *local_err = NULL;
3634     PCDIMMDevice *dimm = PC_DIMM(dev);
3635     uint32_t nr_lmbs;
3636     uint64_t size, addr_start, addr;
3637     int i;
3638     SpaprDrc *drc;
3639 
3640     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3641     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3642 
3643     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3644                                          &local_err);
3645     if (local_err) {
3646         goto out;
3647     }
3648 
3649     /*
3650      * An existing pending dimm state for this DIMM means that there is an
3651      * unplug operation in progress, waiting for the spapr_lmb_release
3652      * callback to complete the job (BQL can't cover that far). In this case,
3653      * bail out to avoid detaching DRCs that were already released.
3654      */
3655     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3656         error_setg(&local_err,
3657                    "Memory unplug already in progress for device %s",
3658                    dev->id);
3659         goto out;
3660     }
3661 
3662     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3663 
3664     addr = addr_start;
3665     for (i = 0; i < nr_lmbs; i++) {
3666         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3667                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3668         g_assert(drc);
3669 
3670         spapr_drc_detach(drc);
3671         addr += SPAPR_MEMORY_BLOCK_SIZE;
3672     }
3673 
3674     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3675                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3676     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3677                                               nr_lmbs, spapr_drc_index(drc));
3678 out:
3679     error_propagate(errp, local_err);
3680 }
3681 
3682 /* Callback to be called during DRC release. */
3683 void spapr_core_release(DeviceState *dev)
3684 {
3685     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3686 
3687     /* Call the unplug handler chain. This can never fail. */
3688     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3689     object_unparent(OBJECT(dev));
3690 }
3691 
3692 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3693 {
3694     MachineState *ms = MACHINE(hotplug_dev);
3695     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3696     CPUCore *cc = CPU_CORE(dev);
3697     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3698 
3699     if (smc->pre_2_10_has_unused_icps) {
3700         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3701         int i;
3702 
3703         for (i = 0; i < cc->nr_threads; i++) {
3704             CPUState *cs = CPU(sc->threads[i]);
3705 
3706             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3707         }
3708     }
3709 
3710     assert(core_slot);
3711     core_slot->cpu = NULL;
3712     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3713 }
3714 
3715 static
3716 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3717                                Error **errp)
3718 {
3719     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3720     int index;
3721     SpaprDrc *drc;
3722     CPUCore *cc = CPU_CORE(dev);
3723 
3724     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3725         error_setg(errp, "Unable to find CPU core with core-id: %d",
3726                    cc->core_id);
3727         return;
3728     }
3729     if (index == 0) {
3730         error_setg(errp, "Boot CPU core may not be unplugged");
3731         return;
3732     }
3733 
3734     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3735                           spapr_vcpu_id(spapr, cc->core_id));
3736     g_assert(drc);
3737 
3738     if (!spapr_drc_unplug_requested(drc)) {
3739         spapr_drc_detach(drc);
3740         spapr_hotplug_req_remove_by_index(drc);
3741     }
3742 }
3743 
3744 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3745                            void *fdt, int *fdt_start_offset, Error **errp)
3746 {
3747     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3748     CPUState *cs = CPU(core->threads[0]);
3749     PowerPCCPU *cpu = POWERPC_CPU(cs);
3750     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3751     int id = spapr_get_vcpu_id(cpu);
3752     char *nodename;
3753     int offset;
3754 
3755     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3756     offset = fdt_add_subnode(fdt, 0, nodename);
3757     g_free(nodename);
3758 
3759     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3760 
3761     *fdt_start_offset = offset;
3762     return 0;
3763 }
3764 
3765 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3766                             Error **errp)
3767 {
3768     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3769     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3770     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3771     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3772     CPUCore *cc = CPU_CORE(dev);
3773     CPUState *cs;
3774     SpaprDrc *drc;
3775     Error *local_err = NULL;
3776     CPUArchId *core_slot;
3777     int index;
3778     bool hotplugged = spapr_drc_hotplugged(dev);
3779     int i;
3780 
3781     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3782     if (!core_slot) {
3783         error_setg(errp, "Unable to find CPU core with core-id: %d",
3784                    cc->core_id);
3785         return;
3786     }
3787     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3788                           spapr_vcpu_id(spapr, cc->core_id));
3789 
3790     g_assert(drc || !mc->has_hotpluggable_cpus);
3791 
3792     if (drc) {
3793         spapr_drc_attach(drc, dev, &local_err);
3794         if (local_err) {
3795             error_propagate(errp, local_err);
3796             return;
3797         }
3798 
3799         if (hotplugged) {
3800             /*
3801              * Send hotplug notification interrupt to the guest only
3802              * in case of hotplugged CPUs.
3803              */
3804             spapr_hotplug_req_add_by_index(drc);
3805         } else {
3806             spapr_drc_reset(drc);
3807         }
3808     }
3809 
3810     core_slot->cpu = OBJECT(dev);
3811 
3812     if (smc->pre_2_10_has_unused_icps) {
3813         for (i = 0; i < cc->nr_threads; i++) {
3814             cs = CPU(core->threads[i]);
3815             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3816         }
3817     }
3818 
3819     /*
3820      * Set compatibility mode to match the boot CPU, which was either set
3821      * by the machine reset code or by CAS.
3822      */
3823     if (hotplugged) {
3824         for (i = 0; i < cc->nr_threads; i++) {
3825             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3826                            &local_err);
3827             if (local_err) {
3828                 error_propagate(errp, local_err);
3829                 return;
3830             }
3831         }
3832     }
3833 }
3834 
3835 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3836                                 Error **errp)
3837 {
3838     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3839     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3840     Error *local_err = NULL;
3841     CPUCore *cc = CPU_CORE(dev);
3842     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3843     const char *type = object_get_typename(OBJECT(dev));
3844     CPUArchId *core_slot;
3845     int index;
3846     unsigned int smp_threads = machine->smp.threads;
3847 
3848     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3849         error_setg(&local_err, "CPU hotplug not supported for this machine");
3850         goto out;
3851     }
3852 
3853     if (strcmp(base_core_type, type)) {
3854         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3855         goto out;
3856     }
3857 
3858     if (cc->core_id % smp_threads) {
3859         error_setg(&local_err, "invalid core id %d", cc->core_id);
3860         goto out;
3861     }
3862 
3863     /*
3864      * In general we should have homogeneous threads-per-core, but old
3865      * (pre hotplug support) machine types allow the last core to have
3866      * reduced threads as a compatibility hack for when we allowed
3867      * total vcpus not a multiple of threads-per-core.
3868      */
3869     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3870         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3871                    cc->nr_threads, smp_threads);
3872         goto out;
3873     }
3874 
3875     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3876     if (!core_slot) {
3877         error_setg(&local_err, "core id %d out of range", cc->core_id);
3878         goto out;
3879     }
3880 
3881     if (core_slot->cpu) {
3882         error_setg(&local_err, "core %d already populated", cc->core_id);
3883         goto out;
3884     }
3885 
3886     numa_cpu_pre_plug(core_slot, dev, &local_err);
3887 
3888 out:
3889     error_propagate(errp, local_err);
3890 }
3891 
3892 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3893                           void *fdt, int *fdt_start_offset, Error **errp)
3894 {
3895     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3896     int intc_phandle;
3897 
3898     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3899     if (intc_phandle <= 0) {
3900         return -1;
3901     }
3902 
3903     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3904         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3905         return -1;
3906     }
3907 
3908     /* generally SLOF creates these, for hotplug it's up to QEMU */
3909     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3910 
3911     return 0;
3912 }
3913 
3914 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3915                                Error **errp)
3916 {
3917     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3918     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3919     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3920     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3921 
3922     if (dev->hotplugged && !smc->dr_phb_enabled) {
3923         error_setg(errp, "PHB hotplug not supported for this machine");
3924         return;
3925     }
3926 
3927     if (sphb->index == (uint32_t)-1) {
3928         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3929         return;
3930     }
3931 
3932     /*
3933      * This will check that sphb->index doesn't exceed the maximum number of
3934      * PHBs for the current machine type.
3935      */
3936     smc->phb_placement(spapr, sphb->index,
3937                        &sphb->buid, &sphb->io_win_addr,
3938                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3939                        windows_supported, sphb->dma_liobn,
3940                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3941                        errp);
3942 }
3943 
3944 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3945                            Error **errp)
3946 {
3947     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3948     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3949     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3950     SpaprDrc *drc;
3951     bool hotplugged = spapr_drc_hotplugged(dev);
3952     Error *local_err = NULL;
3953 
3954     if (!smc->dr_phb_enabled) {
3955         return;
3956     }
3957 
3958     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3959     /* hotplug hooks should check it's enabled before getting this far */
3960     assert(drc);
3961 
3962     spapr_drc_attach(drc, DEVICE(dev), &local_err);
3963     if (local_err) {
3964         error_propagate(errp, local_err);
3965         return;
3966     }
3967 
3968     if (hotplugged) {
3969         spapr_hotplug_req_add_by_index(drc);
3970     } else {
3971         spapr_drc_reset(drc);
3972     }
3973 }
3974 
3975 void spapr_phb_release(DeviceState *dev)
3976 {
3977     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3978 
3979     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3980     object_unparent(OBJECT(dev));
3981 }
3982 
3983 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3984 {
3985     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3986 }
3987 
3988 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3989                                      DeviceState *dev, Error **errp)
3990 {
3991     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3992     SpaprDrc *drc;
3993 
3994     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3995     assert(drc);
3996 
3997     if (!spapr_drc_unplug_requested(drc)) {
3998         spapr_drc_detach(drc);
3999         spapr_hotplug_req_remove_by_index(drc);
4000     }
4001 }
4002 
4003 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4004                                  Error **errp)
4005 {
4006     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4007     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4008 
4009     if (spapr->tpm_proxy != NULL) {
4010         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4011         return;
4012     }
4013 
4014     spapr->tpm_proxy = tpm_proxy;
4015 }
4016 
4017 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4018 {
4019     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4020 
4021     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4022     object_unparent(OBJECT(dev));
4023     spapr->tpm_proxy = NULL;
4024 }
4025 
4026 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4027                                       DeviceState *dev, Error **errp)
4028 {
4029     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4030         spapr_memory_plug(hotplug_dev, dev, errp);
4031     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4032         spapr_core_plug(hotplug_dev, dev, errp);
4033     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4034         spapr_phb_plug(hotplug_dev, dev, errp);
4035     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4036         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4037     }
4038 }
4039 
4040 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4041                                         DeviceState *dev, Error **errp)
4042 {
4043     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4044         spapr_memory_unplug(hotplug_dev, dev);
4045     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4046         spapr_core_unplug(hotplug_dev, dev);
4047     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4048         spapr_phb_unplug(hotplug_dev, dev);
4049     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4050         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4051     }
4052 }
4053 
4054 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4055                                                 DeviceState *dev, Error **errp)
4056 {
4057     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4058     MachineClass *mc = MACHINE_GET_CLASS(sms);
4059     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4060 
4061     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4062         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4063             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4064         } else {
4065             /* NOTE: this means there is a window after guest reset, prior to
4066              * CAS negotiation, where unplug requests will fail due to the
4067              * capability not being detected yet. This is a bit different than
4068              * the case with PCI unplug, where the events will be queued and
4069              * eventually handled by the guest after boot
4070              */
4071             error_setg(errp, "Memory hot unplug not supported for this guest");
4072         }
4073     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4074         if (!mc->has_hotpluggable_cpus) {
4075             error_setg(errp, "CPU hot unplug not supported on this machine");
4076             return;
4077         }
4078         spapr_core_unplug_request(hotplug_dev, dev, errp);
4079     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4080         if (!smc->dr_phb_enabled) {
4081             error_setg(errp, "PHB hot unplug not supported on this machine");
4082             return;
4083         }
4084         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4085     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4086         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4087     }
4088 }
4089 
4090 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4091                                           DeviceState *dev, Error **errp)
4092 {
4093     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4094         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4095     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4096         spapr_core_pre_plug(hotplug_dev, dev, errp);
4097     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4098         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4099     }
4100 }
4101 
4102 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4103                                                  DeviceState *dev)
4104 {
4105     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4106         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4107         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4108         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4109         return HOTPLUG_HANDLER(machine);
4110     }
4111     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4112         PCIDevice *pcidev = PCI_DEVICE(dev);
4113         PCIBus *root = pci_device_root_bus(pcidev);
4114         SpaprPhbState *phb =
4115             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4116                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4117 
4118         if (phb) {
4119             return HOTPLUG_HANDLER(phb);
4120         }
4121     }
4122     return NULL;
4123 }
4124 
4125 static CpuInstanceProperties
4126 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4127 {
4128     CPUArchId *core_slot;
4129     MachineClass *mc = MACHINE_GET_CLASS(machine);
4130 
4131     /* make sure possible_cpu are intialized */
4132     mc->possible_cpu_arch_ids(machine);
4133     /* get CPU core slot containing thread that matches cpu_index */
4134     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4135     assert(core_slot);
4136     return core_slot->props;
4137 }
4138 
4139 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4140 {
4141     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4142 }
4143 
4144 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4145 {
4146     int i;
4147     unsigned int smp_threads = machine->smp.threads;
4148     unsigned int smp_cpus = machine->smp.cpus;
4149     const char *core_type;
4150     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4151     MachineClass *mc = MACHINE_GET_CLASS(machine);
4152 
4153     if (!mc->has_hotpluggable_cpus) {
4154         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4155     }
4156     if (machine->possible_cpus) {
4157         assert(machine->possible_cpus->len == spapr_max_cores);
4158         return machine->possible_cpus;
4159     }
4160 
4161     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4162     if (!core_type) {
4163         error_report("Unable to find sPAPR CPU Core definition");
4164         exit(1);
4165     }
4166 
4167     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4168                              sizeof(CPUArchId) * spapr_max_cores);
4169     machine->possible_cpus->len = spapr_max_cores;
4170     for (i = 0; i < machine->possible_cpus->len; i++) {
4171         int core_id = i * smp_threads;
4172 
4173         machine->possible_cpus->cpus[i].type = core_type;
4174         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4175         machine->possible_cpus->cpus[i].arch_id = core_id;
4176         machine->possible_cpus->cpus[i].props.has_core_id = true;
4177         machine->possible_cpus->cpus[i].props.core_id = core_id;
4178     }
4179     return machine->possible_cpus;
4180 }
4181 
4182 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4183                                 uint64_t *buid, hwaddr *pio,
4184                                 hwaddr *mmio32, hwaddr *mmio64,
4185                                 unsigned n_dma, uint32_t *liobns,
4186                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4187 {
4188     /*
4189      * New-style PHB window placement.
4190      *
4191      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4192      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4193      * windows.
4194      *
4195      * Some guest kernels can't work with MMIO windows above 1<<46
4196      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4197      *
4198      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4199      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4200      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4201      * 1TiB 64-bit MMIO windows for each PHB.
4202      */
4203     const uint64_t base_buid = 0x800000020000000ULL;
4204     int i;
4205 
4206     /* Sanity check natural alignments */
4207     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4208     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4209     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4210     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4211     /* Sanity check bounds */
4212     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4213                       SPAPR_PCI_MEM32_WIN_SIZE);
4214     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4215                       SPAPR_PCI_MEM64_WIN_SIZE);
4216 
4217     if (index >= SPAPR_MAX_PHBS) {
4218         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4219                    SPAPR_MAX_PHBS - 1);
4220         return;
4221     }
4222 
4223     *buid = base_buid + index;
4224     for (i = 0; i < n_dma; ++i) {
4225         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4226     }
4227 
4228     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4229     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4230     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4231 
4232     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4233     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4234 }
4235 
4236 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4237 {
4238     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4239 
4240     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4241 }
4242 
4243 static void spapr_ics_resend(XICSFabric *dev)
4244 {
4245     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4246 
4247     ics_resend(spapr->ics);
4248 }
4249 
4250 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4251 {
4252     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4253 
4254     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4255 }
4256 
4257 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4258                                  Monitor *mon)
4259 {
4260     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4261 
4262     spapr_irq_print_info(spapr, mon);
4263     monitor_printf(mon, "irqchip: %s\n",
4264                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4265 }
4266 
4267 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4268                            uint8_t nvt_blk, uint32_t nvt_idx,
4269                            bool cam_ignore, uint8_t priority,
4270                            uint32_t logic_serv, XiveTCTXMatch *match)
4271 {
4272     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4273     XivePresenter *xptr = XIVE_PRESENTER(spapr->xive);
4274     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4275     int count;
4276 
4277     /* This is a XIVE only operation */
4278     assert(spapr->active_intc == SPAPR_INTC(spapr->xive));
4279 
4280     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4281                            priority, logic_serv, match);
4282     if (count < 0) {
4283         return count;
4284     }
4285 
4286     /*
4287      * When we implement the save and restore of the thread interrupt
4288      * contexts in the enter/exit CPU handlers of the machine and the
4289      * escalations in QEMU, we should be able to handle non dispatched
4290      * vCPUs.
4291      *
4292      * Until this is done, the sPAPR machine should find at least one
4293      * matching context always.
4294      */
4295     if (count == 0) {
4296         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4297                       nvt_blk, nvt_idx);
4298     }
4299 
4300     return count;
4301 }
4302 
4303 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4304 {
4305     return cpu->vcpu_id;
4306 }
4307 
4308 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4309 {
4310     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4311     MachineState *ms = MACHINE(spapr);
4312     int vcpu_id;
4313 
4314     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4315 
4316     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4317         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4318         error_append_hint(errp, "Adjust the number of cpus to %d "
4319                           "or try to raise the number of threads per core\n",
4320                           vcpu_id * ms->smp.threads / spapr->vsmt);
4321         return;
4322     }
4323 
4324     cpu->vcpu_id = vcpu_id;
4325 }
4326 
4327 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4328 {
4329     CPUState *cs;
4330 
4331     CPU_FOREACH(cs) {
4332         PowerPCCPU *cpu = POWERPC_CPU(cs);
4333 
4334         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4335             return cpu;
4336         }
4337     }
4338 
4339     return NULL;
4340 }
4341 
4342 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4343 {
4344     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4345 
4346     /* These are only called by TCG, KVM maintains dispatch state */
4347 
4348     spapr_cpu->prod = false;
4349     if (spapr_cpu->vpa_addr) {
4350         CPUState *cs = CPU(cpu);
4351         uint32_t dispatch;
4352 
4353         dispatch = ldl_be_phys(cs->as,
4354                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4355         dispatch++;
4356         if ((dispatch & 1) != 0) {
4357             qemu_log_mask(LOG_GUEST_ERROR,
4358                           "VPA: incorrect dispatch counter value for "
4359                           "dispatched partition %u, correcting.\n", dispatch);
4360             dispatch++;
4361         }
4362         stl_be_phys(cs->as,
4363                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4364     }
4365 }
4366 
4367 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4368 {
4369     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4370 
4371     if (spapr_cpu->vpa_addr) {
4372         CPUState *cs = CPU(cpu);
4373         uint32_t dispatch;
4374 
4375         dispatch = ldl_be_phys(cs->as,
4376                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4377         dispatch++;
4378         if ((dispatch & 1) != 1) {
4379             qemu_log_mask(LOG_GUEST_ERROR,
4380                           "VPA: incorrect dispatch counter value for "
4381                           "preempted partition %u, correcting.\n", dispatch);
4382             dispatch++;
4383         }
4384         stl_be_phys(cs->as,
4385                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4386     }
4387 }
4388 
4389 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4390 {
4391     MachineClass *mc = MACHINE_CLASS(oc);
4392     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4393     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4394     NMIClass *nc = NMI_CLASS(oc);
4395     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4396     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4397     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4398     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4399     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4400 
4401     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4402     mc->ignore_boot_device_suffixes = true;
4403 
4404     /*
4405      * We set up the default / latest behaviour here.  The class_init
4406      * functions for the specific versioned machine types can override
4407      * these details for backwards compatibility
4408      */
4409     mc->init = spapr_machine_init;
4410     mc->reset = spapr_machine_reset;
4411     mc->block_default_type = IF_SCSI;
4412     mc->max_cpus = 1024;
4413     mc->no_parallel = 1;
4414     mc->default_boot_order = "";
4415     mc->default_ram_size = 512 * MiB;
4416     mc->default_display = "std";
4417     mc->kvm_type = spapr_kvm_type;
4418     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4419     mc->pci_allow_0_address = true;
4420     assert(!mc->get_hotplug_handler);
4421     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4422     hc->pre_plug = spapr_machine_device_pre_plug;
4423     hc->plug = spapr_machine_device_plug;
4424     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4425     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4426     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4427     hc->unplug_request = spapr_machine_device_unplug_request;
4428     hc->unplug = spapr_machine_device_unplug;
4429 
4430     smc->dr_lmb_enabled = true;
4431     smc->update_dt_enabled = true;
4432     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4433     mc->has_hotpluggable_cpus = true;
4434     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4435     fwc->get_dev_path = spapr_get_fw_dev_path;
4436     nc->nmi_monitor_handler = spapr_nmi;
4437     smc->phb_placement = spapr_phb_placement;
4438     vhc->hypercall = emulate_spapr_hypercall;
4439     vhc->hpt_mask = spapr_hpt_mask;
4440     vhc->map_hptes = spapr_map_hptes;
4441     vhc->unmap_hptes = spapr_unmap_hptes;
4442     vhc->hpte_set_c = spapr_hpte_set_c;
4443     vhc->hpte_set_r = spapr_hpte_set_r;
4444     vhc->get_pate = spapr_get_pate;
4445     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4446     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4447     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4448     xic->ics_get = spapr_ics_get;
4449     xic->ics_resend = spapr_ics_resend;
4450     xic->icp_get = spapr_icp_get;
4451     ispc->print_info = spapr_pic_print_info;
4452     /* Force NUMA node memory size to be a multiple of
4453      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4454      * in which LMBs are represented and hot-added
4455      */
4456     mc->numa_mem_align_shift = 28;
4457     mc->numa_mem_supported = true;
4458     mc->auto_enable_numa = true;
4459 
4460     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4461     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4462     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4463     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4464     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4465     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4466     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4467     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4468     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4469     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4470     spapr_caps_add_properties(smc, &error_abort);
4471     smc->irq = &spapr_irq_dual;
4472     smc->dr_phb_enabled = true;
4473     smc->linux_pci_probe = true;
4474     smc->smp_threads_vsmt = true;
4475     smc->nr_xirqs = SPAPR_NR_XIRQS;
4476     xfc->match_nvt = spapr_match_nvt;
4477 }
4478 
4479 static const TypeInfo spapr_machine_info = {
4480     .name          = TYPE_SPAPR_MACHINE,
4481     .parent        = TYPE_MACHINE,
4482     .abstract      = true,
4483     .instance_size = sizeof(SpaprMachineState),
4484     .instance_init = spapr_instance_init,
4485     .instance_finalize = spapr_machine_finalizefn,
4486     .class_size    = sizeof(SpaprMachineClass),
4487     .class_init    = spapr_machine_class_init,
4488     .interfaces = (InterfaceInfo[]) {
4489         { TYPE_FW_PATH_PROVIDER },
4490         { TYPE_NMI },
4491         { TYPE_HOTPLUG_HANDLER },
4492         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4493         { TYPE_XICS_FABRIC },
4494         { TYPE_INTERRUPT_STATS_PROVIDER },
4495         { TYPE_XIVE_FABRIC },
4496         { }
4497     },
4498 };
4499 
4500 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4501     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4502                                                     void *data)      \
4503     {                                                                \
4504         MachineClass *mc = MACHINE_CLASS(oc);                        \
4505         spapr_machine_##suffix##_class_options(mc);                  \
4506         if (latest) {                                                \
4507             mc->alias = "pseries";                                   \
4508             mc->is_default = 1;                                      \
4509         }                                                            \
4510     }                                                                \
4511     static const TypeInfo spapr_machine_##suffix##_info = {          \
4512         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4513         .parent = TYPE_SPAPR_MACHINE,                                \
4514         .class_init = spapr_machine_##suffix##_class_init,           \
4515     };                                                               \
4516     static void spapr_machine_register_##suffix(void)                \
4517     {                                                                \
4518         type_register(&spapr_machine_##suffix##_info);               \
4519     }                                                                \
4520     type_init(spapr_machine_register_##suffix)
4521 
4522 /*
4523  * pseries-5.0
4524  */
4525 static void spapr_machine_5_0_class_options(MachineClass *mc)
4526 {
4527     /* Defaults for the latest behaviour inherited from the base class */
4528 }
4529 
4530 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4531 
4532 /*
4533  * pseries-4.2
4534  */
4535 static void spapr_machine_4_2_class_options(MachineClass *mc)
4536 {
4537     spapr_machine_5_0_class_options(mc);
4538     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4539 }
4540 
4541 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4542 
4543 /*
4544  * pseries-4.1
4545  */
4546 static void spapr_machine_4_1_class_options(MachineClass *mc)
4547 {
4548     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4549     static GlobalProperty compat[] = {
4550         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4551         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4552     };
4553 
4554     spapr_machine_4_2_class_options(mc);
4555     smc->linux_pci_probe = false;
4556     smc->smp_threads_vsmt = false;
4557     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4558     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4559 }
4560 
4561 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4562 
4563 /*
4564  * pseries-4.0
4565  */
4566 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4567                               uint64_t *buid, hwaddr *pio,
4568                               hwaddr *mmio32, hwaddr *mmio64,
4569                               unsigned n_dma, uint32_t *liobns,
4570                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4571 {
4572     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4573                         nv2gpa, nv2atsd, errp);
4574     *nv2gpa = 0;
4575     *nv2atsd = 0;
4576 }
4577 
4578 static void spapr_machine_4_0_class_options(MachineClass *mc)
4579 {
4580     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4581 
4582     spapr_machine_4_1_class_options(mc);
4583     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4584     smc->phb_placement = phb_placement_4_0;
4585     smc->irq = &spapr_irq_xics;
4586     smc->pre_4_1_migration = true;
4587 }
4588 
4589 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4590 
4591 /*
4592  * pseries-3.1
4593  */
4594 static void spapr_machine_3_1_class_options(MachineClass *mc)
4595 {
4596     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4597 
4598     spapr_machine_4_0_class_options(mc);
4599     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4600 
4601     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4602     smc->update_dt_enabled = false;
4603     smc->dr_phb_enabled = false;
4604     smc->broken_host_serial_model = true;
4605     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4606     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4607     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4608     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4609 }
4610 
4611 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4612 
4613 /*
4614  * pseries-3.0
4615  */
4616 
4617 static void spapr_machine_3_0_class_options(MachineClass *mc)
4618 {
4619     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4620 
4621     spapr_machine_3_1_class_options(mc);
4622     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4623 
4624     smc->legacy_irq_allocation = true;
4625     smc->nr_xirqs = 0x400;
4626     smc->irq = &spapr_irq_xics_legacy;
4627 }
4628 
4629 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4630 
4631 /*
4632  * pseries-2.12
4633  */
4634 static void spapr_machine_2_12_class_options(MachineClass *mc)
4635 {
4636     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4637     static GlobalProperty compat[] = {
4638         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4639         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4640     };
4641 
4642     spapr_machine_3_0_class_options(mc);
4643     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4644     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4645 
4646     /* We depend on kvm_enabled() to choose a default value for the
4647      * hpt-max-page-size capability. Of course we can't do it here
4648      * because this is too early and the HW accelerator isn't initialzed
4649      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4650      */
4651     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4652 }
4653 
4654 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4655 
4656 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4657 {
4658     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4659 
4660     spapr_machine_2_12_class_options(mc);
4661     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4662     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4663     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4664 }
4665 
4666 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4667 
4668 /*
4669  * pseries-2.11
4670  */
4671 
4672 static void spapr_machine_2_11_class_options(MachineClass *mc)
4673 {
4674     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4675 
4676     spapr_machine_2_12_class_options(mc);
4677     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4678     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4679 }
4680 
4681 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4682 
4683 /*
4684  * pseries-2.10
4685  */
4686 
4687 static void spapr_machine_2_10_class_options(MachineClass *mc)
4688 {
4689     spapr_machine_2_11_class_options(mc);
4690     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4691 }
4692 
4693 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4694 
4695 /*
4696  * pseries-2.9
4697  */
4698 
4699 static void spapr_machine_2_9_class_options(MachineClass *mc)
4700 {
4701     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4702     static GlobalProperty compat[] = {
4703         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4704     };
4705 
4706     spapr_machine_2_10_class_options(mc);
4707     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4708     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4709     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4710     smc->pre_2_10_has_unused_icps = true;
4711     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4712 }
4713 
4714 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4715 
4716 /*
4717  * pseries-2.8
4718  */
4719 
4720 static void spapr_machine_2_8_class_options(MachineClass *mc)
4721 {
4722     static GlobalProperty compat[] = {
4723         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4724     };
4725 
4726     spapr_machine_2_9_class_options(mc);
4727     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4728     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4729     mc->numa_mem_align_shift = 23;
4730 }
4731 
4732 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4733 
4734 /*
4735  * pseries-2.7
4736  */
4737 
4738 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4739                               uint64_t *buid, hwaddr *pio,
4740                               hwaddr *mmio32, hwaddr *mmio64,
4741                               unsigned n_dma, uint32_t *liobns,
4742                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4743 {
4744     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4745     const uint64_t base_buid = 0x800000020000000ULL;
4746     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4747     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4748     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4749     const uint32_t max_index = 255;
4750     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4751 
4752     uint64_t ram_top = MACHINE(spapr)->ram_size;
4753     hwaddr phb0_base, phb_base;
4754     int i;
4755 
4756     /* Do we have device memory? */
4757     if (MACHINE(spapr)->maxram_size > ram_top) {
4758         /* Can't just use maxram_size, because there may be an
4759          * alignment gap between normal and device memory regions
4760          */
4761         ram_top = MACHINE(spapr)->device_memory->base +
4762             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4763     }
4764 
4765     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4766 
4767     if (index > max_index) {
4768         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4769                    max_index);
4770         return;
4771     }
4772 
4773     *buid = base_buid + index;
4774     for (i = 0; i < n_dma; ++i) {
4775         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4776     }
4777 
4778     phb_base = phb0_base + index * phb_spacing;
4779     *pio = phb_base + pio_offset;
4780     *mmio32 = phb_base + mmio_offset;
4781     /*
4782      * We don't set the 64-bit MMIO window, relying on the PHB's
4783      * fallback behaviour of automatically splitting a large "32-bit"
4784      * window into contiguous 32-bit and 64-bit windows
4785      */
4786 
4787     *nv2gpa = 0;
4788     *nv2atsd = 0;
4789 }
4790 
4791 static void spapr_machine_2_7_class_options(MachineClass *mc)
4792 {
4793     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4794     static GlobalProperty compat[] = {
4795         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4796         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4797         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4798         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4799     };
4800 
4801     spapr_machine_2_8_class_options(mc);
4802     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4803     mc->default_machine_opts = "modern-hotplug-events=off";
4804     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4805     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4806     smc->phb_placement = phb_placement_2_7;
4807 }
4808 
4809 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4810 
4811 /*
4812  * pseries-2.6
4813  */
4814 
4815 static void spapr_machine_2_6_class_options(MachineClass *mc)
4816 {
4817     static GlobalProperty compat[] = {
4818         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4819     };
4820 
4821     spapr_machine_2_7_class_options(mc);
4822     mc->has_hotpluggable_cpus = false;
4823     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4824     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4825 }
4826 
4827 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4828 
4829 /*
4830  * pseries-2.5
4831  */
4832 
4833 static void spapr_machine_2_5_class_options(MachineClass *mc)
4834 {
4835     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4836     static GlobalProperty compat[] = {
4837         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4838     };
4839 
4840     spapr_machine_2_6_class_options(mc);
4841     smc->use_ohci_by_default = true;
4842     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4843     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4844 }
4845 
4846 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4847 
4848 /*
4849  * pseries-2.4
4850  */
4851 
4852 static void spapr_machine_2_4_class_options(MachineClass *mc)
4853 {
4854     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4855 
4856     spapr_machine_2_5_class_options(mc);
4857     smc->dr_lmb_enabled = false;
4858     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4859 }
4860 
4861 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4862 
4863 /*
4864  * pseries-2.3
4865  */
4866 
4867 static void spapr_machine_2_3_class_options(MachineClass *mc)
4868 {
4869     static GlobalProperty compat[] = {
4870         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4871     };
4872     spapr_machine_2_4_class_options(mc);
4873     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4874     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4875 }
4876 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4877 
4878 /*
4879  * pseries-2.2
4880  */
4881 
4882 static void spapr_machine_2_2_class_options(MachineClass *mc)
4883 {
4884     static GlobalProperty compat[] = {
4885         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4886     };
4887 
4888     spapr_machine_2_3_class_options(mc);
4889     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4890     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4891     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4892 }
4893 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4894 
4895 /*
4896  * pseries-2.1
4897  */
4898 
4899 static void spapr_machine_2_1_class_options(MachineClass *mc)
4900 {
4901     spapr_machine_2_2_class_options(mc);
4902     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4903 }
4904 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4905 
4906 static void spapr_machine_register_types(void)
4907 {
4908     type_register_static(&spapr_machine_info);
4909 }
4910 
4911 type_init(spapr_machine_register_types)
4912