xref: /qemu/hw/ppc/spapr.c (revision 6a84737c80b3febb093e066d451a44e61b54159a)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 
85 #include "monitor/monitor.h"
86 
87 #include <libfdt.h>
88 
89 /* SLOF memory layout:
90  *
91  * SLOF raw image loaded at 0, copies its romfs right below the flat
92  * device-tree, then position SLOF itself 31M below that
93  *
94  * So we set FW_OVERHEAD to 40MB which should account for all of that
95  * and more
96  *
97  * We load our kernel at 4M, leaving space for SLOF initial image
98  */
99 #define FDT_MAX_SIZE            0x100000
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            (128 * MiB)
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
206 {
207     int index = spapr_get_vcpu_id(cpu);
208     uint32_t associativity[] = {cpu_to_be32(0x5),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(0x0),
212                                 cpu_to_be32(cpu->node_id),
213                                 cpu_to_be32(index)};
214 
215     /* Advertise NUMA via ibm,associativity */
216     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217                           sizeof(associativity));
218 }
219 
220 /* Populate the "ibm,pa-features" property */
221 static void spapr_populate_pa_features(SpaprMachineState *spapr,
222                                        PowerPCCPU *cpu,
223                                        void *fdt, int offset)
224 {
225     uint8_t pa_features_206[] = { 6, 0,
226         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
227     uint8_t pa_features_207[] = { 24, 0,
228         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
229         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
230         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
231         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
232     uint8_t pa_features_300[] = { 66, 0,
233         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
234         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
235         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
236         /* 6: DS207 */
237         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
238         /* 16: Vector */
239         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
240         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
241         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
242         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
243         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
244         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
245         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
246         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
248         /* 42: PM, 44: PC RA, 46: SC vec'd */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
250         /* 48: SIMD, 50: QP BFP, 52: String */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
252         /* 54: DecFP, 56: DecI, 58: SHA */
253         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
254         /* 60: NM atomic, 62: RNG */
255         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
256     };
257     uint8_t *pa_features = NULL;
258     size_t pa_size;
259 
260     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
261         pa_features = pa_features_206;
262         pa_size = sizeof(pa_features_206);
263     }
264     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
265         pa_features = pa_features_207;
266         pa_size = sizeof(pa_features_207);
267     }
268     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
269         pa_features = pa_features_300;
270         pa_size = sizeof(pa_features_300);
271     }
272     if (!pa_features) {
273         return;
274     }
275 
276     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
277         /*
278          * Note: we keep CI large pages off by default because a 64K capable
279          * guest provisioned with large pages might otherwise try to map a qemu
280          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
281          * even if that qemu runs on a 4k host.
282          * We dd this bit back here if we are confident this is not an issue
283          */
284         pa_features[3] |= 0x20;
285     }
286     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
287         pa_features[24] |= 0x80;    /* Transactional memory support */
288     }
289     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
290         /* Workaround for broken kernels that attempt (guest) radix
291          * mode when they can't handle it, if they see the radix bit set
292          * in pa-features. So hide it from them. */
293         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
294     }
295 
296     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
297 }
298 
299 static hwaddr spapr_node0_size(MachineState *machine)
300 {
301     if (machine->numa_state->num_nodes) {
302         int i;
303         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
304             if (machine->numa_state->nodes[i].node_mem) {
305                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
306                            machine->ram_size);
307             }
308         }
309     }
310     return machine->ram_size;
311 }
312 
313 static void add_str(GString *s, const gchar *s1)
314 {
315     g_string_append_len(s, s1, strlen(s1) + 1);
316 }
317 
318 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
319                                        hwaddr size)
320 {
321     uint32_t associativity[] = {
322         cpu_to_be32(0x4), /* length */
323         cpu_to_be32(0x0), cpu_to_be32(0x0),
324         cpu_to_be32(0x0), cpu_to_be32(nodeid)
325     };
326     char mem_name[32];
327     uint64_t mem_reg_property[2];
328     int off;
329 
330     mem_reg_property[0] = cpu_to_be64(start);
331     mem_reg_property[1] = cpu_to_be64(size);
332 
333     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
334     off = fdt_add_subnode(fdt, 0, mem_name);
335     _FDT(off);
336     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
337     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
338                       sizeof(mem_reg_property))));
339     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
340                       sizeof(associativity))));
341     return off;
342 }
343 
344 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
345 {
346     MachineState *machine = MACHINE(spapr);
347     hwaddr mem_start, node_size;
348     int i, nb_nodes = machine->numa_state->num_nodes;
349     NodeInfo *nodes = machine->numa_state->nodes;
350 
351     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
352         if (!nodes[i].node_mem) {
353             continue;
354         }
355         if (mem_start >= machine->ram_size) {
356             node_size = 0;
357         } else {
358             node_size = nodes[i].node_mem;
359             if (node_size > machine->ram_size - mem_start) {
360                 node_size = machine->ram_size - mem_start;
361             }
362         }
363         if (!mem_start) {
364             /* spapr_machine_init() checks for rma_size <= node0_size
365              * already */
366             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
367             mem_start += spapr->rma_size;
368             node_size -= spapr->rma_size;
369         }
370         for ( ; node_size; ) {
371             hwaddr sizetmp = pow2floor(node_size);
372 
373             /* mem_start != 0 here */
374             if (ctzl(mem_start) < ctzl(sizetmp)) {
375                 sizetmp = 1ULL << ctzl(mem_start);
376             }
377 
378             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
379             node_size -= sizetmp;
380             mem_start += sizetmp;
381         }
382     }
383 
384     return 0;
385 }
386 
387 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
388                                   SpaprMachineState *spapr)
389 {
390     MachineState *ms = MACHINE(spapr);
391     PowerPCCPU *cpu = POWERPC_CPU(cs);
392     CPUPPCState *env = &cpu->env;
393     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
394     int index = spapr_get_vcpu_id(cpu);
395     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
396                        0xffffffff, 0xffffffff};
397     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
398         : SPAPR_TIMEBASE_FREQ;
399     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
400     uint32_t page_sizes_prop[64];
401     size_t page_sizes_prop_size;
402     unsigned int smp_threads = ms->smp.threads;
403     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
404     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
405     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
406     SpaprDrc *drc;
407     int drc_index;
408     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
409     int i;
410 
411     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
412     if (drc) {
413         drc_index = spapr_drc_index(drc);
414         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
415     }
416 
417     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
418     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
419 
420     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
421     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
422                            env->dcache_line_size)));
423     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
424                            env->dcache_line_size)));
425     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
426                            env->icache_line_size)));
427     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
428                            env->icache_line_size)));
429 
430     if (pcc->l1_dcache_size) {
431         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
432                                pcc->l1_dcache_size)));
433     } else {
434         warn_report("Unknown L1 dcache size for cpu");
435     }
436     if (pcc->l1_icache_size) {
437         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
438                                pcc->l1_icache_size)));
439     } else {
440         warn_report("Unknown L1 icache size for cpu");
441     }
442 
443     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
444     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
445     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
446     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
447     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
448     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
449 
450     if (env->spr_cb[SPR_PURR].oea_read) {
451         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
452     }
453     if (env->spr_cb[SPR_SPURR].oea_read) {
454         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
455     }
456 
457     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
458         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
459                           segs, sizeof(segs))));
460     }
461 
462     /* Advertise VSX (vector extensions) if available
463      *   1               == VMX / Altivec available
464      *   2               == VSX available
465      *
466      * Only CPUs for which we create core types in spapr_cpu_core.c
467      * are possible, and all of those have VMX */
468     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
469         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
470     } else {
471         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
472     }
473 
474     /* Advertise DFP (Decimal Floating Point) if available
475      *   0 / no property == no DFP
476      *   1               == DFP available */
477     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
478         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
479     }
480 
481     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
482                                                       sizeof(page_sizes_prop));
483     if (page_sizes_prop_size) {
484         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
485                           page_sizes_prop, page_sizes_prop_size)));
486     }
487 
488     spapr_populate_pa_features(spapr, cpu, fdt, offset);
489 
490     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
491                            cs->cpu_index / vcpus_per_socket)));
492 
493     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
494                       pft_size_prop, sizeof(pft_size_prop))));
495 
496     if (ms->numa_state->num_nodes > 1) {
497         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
498     }
499 
500     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
501 
502     if (pcc->radix_page_info) {
503         for (i = 0; i < pcc->radix_page_info->count; i++) {
504             radix_AP_encodings[i] =
505                 cpu_to_be32(pcc->radix_page_info->entries[i]);
506         }
507         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
508                           radix_AP_encodings,
509                           pcc->radix_page_info->count *
510                           sizeof(radix_AP_encodings[0]))));
511     }
512 
513     /*
514      * We set this property to let the guest know that it can use the large
515      * decrementer and its width in bits.
516      */
517     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
518         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
519                               pcc->lrg_decr_bits)));
520 }
521 
522 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
523 {
524     CPUState **rev;
525     CPUState *cs;
526     int n_cpus;
527     int cpus_offset;
528     char *nodename;
529     int i;
530 
531     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
532     _FDT(cpus_offset);
533     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
534     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
535 
536     /*
537      * We walk the CPUs in reverse order to ensure that CPU DT nodes
538      * created by fdt_add_subnode() end up in the right order in FDT
539      * for the guest kernel the enumerate the CPUs correctly.
540      *
541      * The CPU list cannot be traversed in reverse order, so we need
542      * to do extra work.
543      */
544     n_cpus = 0;
545     rev = NULL;
546     CPU_FOREACH(cs) {
547         rev = g_renew(CPUState *, rev, n_cpus + 1);
548         rev[n_cpus++] = cs;
549     }
550 
551     for (i = n_cpus - 1; i >= 0; i--) {
552         CPUState *cs = rev[i];
553         PowerPCCPU *cpu = POWERPC_CPU(cs);
554         int index = spapr_get_vcpu_id(cpu);
555         DeviceClass *dc = DEVICE_GET_CLASS(cs);
556         int offset;
557 
558         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
559             continue;
560         }
561 
562         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
563         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
564         g_free(nodename);
565         _FDT(offset);
566         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
567     }
568 
569     g_free(rev);
570 }
571 
572 static int spapr_rng_populate_dt(void *fdt)
573 {
574     int node;
575     int ret;
576 
577     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
578     if (node <= 0) {
579         return -1;
580     }
581     ret = fdt_setprop_string(fdt, node, "device_type",
582                              "ibm,platform-facilities");
583     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
584     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
585 
586     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
587     if (node <= 0) {
588         return -1;
589     }
590     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
591 
592     return ret ? -1 : 0;
593 }
594 
595 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
596 {
597     MemoryDeviceInfoList *info;
598 
599     for (info = list; info; info = info->next) {
600         MemoryDeviceInfo *value = info->value;
601 
602         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
603             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
604 
605             if (addr >= pcdimm_info->addr &&
606                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
607                 return pcdimm_info->node;
608             }
609         }
610     }
611 
612     return -1;
613 }
614 
615 struct sPAPRDrconfCellV2 {
616      uint32_t seq_lmbs;
617      uint64_t base_addr;
618      uint32_t drc_index;
619      uint32_t aa_index;
620      uint32_t flags;
621 } QEMU_PACKED;
622 
623 typedef struct DrconfCellQueue {
624     struct sPAPRDrconfCellV2 cell;
625     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
626 } DrconfCellQueue;
627 
628 static DrconfCellQueue *
629 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
630                       uint32_t drc_index, uint32_t aa_index,
631                       uint32_t flags)
632 {
633     DrconfCellQueue *elem;
634 
635     elem = g_malloc0(sizeof(*elem));
636     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
637     elem->cell.base_addr = cpu_to_be64(base_addr);
638     elem->cell.drc_index = cpu_to_be32(drc_index);
639     elem->cell.aa_index = cpu_to_be32(aa_index);
640     elem->cell.flags = cpu_to_be32(flags);
641 
642     return elem;
643 }
644 
645 /* ibm,dynamic-memory-v2 */
646 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
647                                    int offset, MemoryDeviceInfoList *dimms)
648 {
649     MachineState *machine = MACHINE(spapr);
650     uint8_t *int_buf, *cur_index;
651     int ret;
652     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
653     uint64_t addr, cur_addr, size;
654     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
655     uint64_t mem_end = machine->device_memory->base +
656                        memory_region_size(&machine->device_memory->mr);
657     uint32_t node, buf_len, nr_entries = 0;
658     SpaprDrc *drc;
659     DrconfCellQueue *elem, *next;
660     MemoryDeviceInfoList *info;
661     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
662         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
663 
664     /* Entry to cover RAM and the gap area */
665     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
666                                  SPAPR_LMB_FLAGS_RESERVED |
667                                  SPAPR_LMB_FLAGS_DRC_INVALID);
668     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
669     nr_entries++;
670 
671     cur_addr = machine->device_memory->base;
672     for (info = dimms; info; info = info->next) {
673         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
674 
675         addr = di->addr;
676         size = di->size;
677         node = di->node;
678 
679         /*
680          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
681          * area is marked hotpluggable in the next iteration for the bigger
682          * chunk including the NVDIMM occupied area.
683          */
684         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
685             continue;
686 
687         /* Entry for hot-pluggable area */
688         if (cur_addr < addr) {
689             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
690             g_assert(drc);
691             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
692                                          cur_addr, spapr_drc_index(drc), -1, 0);
693             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
694             nr_entries++;
695         }
696 
697         /* Entry for DIMM */
698         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
699         g_assert(drc);
700         elem = spapr_get_drconf_cell(size / lmb_size, addr,
701                                      spapr_drc_index(drc), node,
702                                      SPAPR_LMB_FLAGS_ASSIGNED);
703         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
704         nr_entries++;
705         cur_addr = addr + size;
706     }
707 
708     /* Entry for remaining hotpluggable area */
709     if (cur_addr < mem_end) {
710         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
711         g_assert(drc);
712         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
713                                      cur_addr, spapr_drc_index(drc), -1, 0);
714         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
715         nr_entries++;
716     }
717 
718     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
719     int_buf = cur_index = g_malloc0(buf_len);
720     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
721     cur_index += sizeof(nr_entries);
722 
723     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
724         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
725         cur_index += sizeof(elem->cell);
726         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
727         g_free(elem);
728     }
729 
730     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
731     g_free(int_buf);
732     if (ret < 0) {
733         return -1;
734     }
735     return 0;
736 }
737 
738 /* ibm,dynamic-memory */
739 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
740                                    int offset, MemoryDeviceInfoList *dimms)
741 {
742     MachineState *machine = MACHINE(spapr);
743     int i, ret;
744     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
745     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
746     uint32_t nr_lmbs = (machine->device_memory->base +
747                        memory_region_size(&machine->device_memory->mr)) /
748                        lmb_size;
749     uint32_t *int_buf, *cur_index, buf_len;
750 
751     /*
752      * Allocate enough buffer size to fit in ibm,dynamic-memory
753      */
754     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
755     cur_index = int_buf = g_malloc0(buf_len);
756     int_buf[0] = cpu_to_be32(nr_lmbs);
757     cur_index++;
758     for (i = 0; i < nr_lmbs; i++) {
759         uint64_t addr = i * lmb_size;
760         uint32_t *dynamic_memory = cur_index;
761 
762         if (i >= device_lmb_start) {
763             SpaprDrc *drc;
764 
765             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
766             g_assert(drc);
767 
768             dynamic_memory[0] = cpu_to_be32(addr >> 32);
769             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
770             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
771             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
772             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
773             if (memory_region_present(get_system_memory(), addr)) {
774                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
775             } else {
776                 dynamic_memory[5] = cpu_to_be32(0);
777             }
778         } else {
779             /*
780              * LMB information for RMA, boot time RAM and gap b/n RAM and
781              * device memory region -- all these are marked as reserved
782              * and as having no valid DRC.
783              */
784             dynamic_memory[0] = cpu_to_be32(addr >> 32);
785             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
786             dynamic_memory[2] = cpu_to_be32(0);
787             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
788             dynamic_memory[4] = cpu_to_be32(-1);
789             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
790                                             SPAPR_LMB_FLAGS_DRC_INVALID);
791         }
792 
793         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
794     }
795     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
796     g_free(int_buf);
797     if (ret < 0) {
798         return -1;
799     }
800     return 0;
801 }
802 
803 /*
804  * Adds ibm,dynamic-reconfiguration-memory node.
805  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
806  * of this device tree node.
807  */
808 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
809 {
810     MachineState *machine = MACHINE(spapr);
811     int nb_numa_nodes = machine->numa_state->num_nodes;
812     int ret, i, offset;
813     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
814     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
815     uint32_t *int_buf, *cur_index, buf_len;
816     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
817     MemoryDeviceInfoList *dimms = NULL;
818 
819     /*
820      * Don't create the node if there is no device memory
821      */
822     if (machine->ram_size == machine->maxram_size) {
823         return 0;
824     }
825 
826     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
827 
828     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
829                     sizeof(prop_lmb_size));
830     if (ret < 0) {
831         return ret;
832     }
833 
834     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
835     if (ret < 0) {
836         return ret;
837     }
838 
839     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
840     if (ret < 0) {
841         return ret;
842     }
843 
844     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
845     dimms = qmp_memory_device_list();
846     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
847         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
848     } else {
849         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
850     }
851     qapi_free_MemoryDeviceInfoList(dimms);
852 
853     if (ret < 0) {
854         return ret;
855     }
856 
857     /* ibm,associativity-lookup-arrays */
858     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
859     cur_index = int_buf = g_malloc0(buf_len);
860     int_buf[0] = cpu_to_be32(nr_nodes);
861     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
862     cur_index += 2;
863     for (i = 0; i < nr_nodes; i++) {
864         uint32_t associativity[] = {
865             cpu_to_be32(0x0),
866             cpu_to_be32(0x0),
867             cpu_to_be32(0x0),
868             cpu_to_be32(i)
869         };
870         memcpy(cur_index, associativity, sizeof(associativity));
871         cur_index += 4;
872     }
873     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
874             (cur_index - int_buf) * sizeof(uint32_t));
875     g_free(int_buf);
876 
877     return ret;
878 }
879 
880 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
881                                 SpaprOptionVector *ov5_updates)
882 {
883     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
884     int ret = 0, offset;
885 
886     /* Generate ibm,dynamic-reconfiguration-memory node if required */
887     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
888         g_assert(smc->dr_lmb_enabled);
889         ret = spapr_populate_drconf_memory(spapr, fdt);
890         if (ret) {
891             return ret;
892         }
893     }
894 
895     offset = fdt_path_offset(fdt, "/chosen");
896     if (offset < 0) {
897         offset = fdt_add_subnode(fdt, 0, "chosen");
898         if (offset < 0) {
899             return offset;
900         }
901     }
902     return spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
903                                   "ibm,architecture-vec-5");
904 }
905 
906 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
907 {
908     MachineState *ms = MACHINE(spapr);
909     int rtas;
910     GString *hypertas = g_string_sized_new(256);
911     GString *qemu_hypertas = g_string_sized_new(256);
912     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
913     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
914         memory_region_size(&MACHINE(spapr)->device_memory->mr);
915     uint32_t lrdr_capacity[] = {
916         cpu_to_be32(max_device_addr >> 32),
917         cpu_to_be32(max_device_addr & 0xffffffff),
918         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
919         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
920     };
921     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
922     uint32_t maxdomains[] = {
923         cpu_to_be32(4),
924         maxdomain,
925         maxdomain,
926         maxdomain,
927         cpu_to_be32(spapr->gpu_numa_id),
928     };
929 
930     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
931 
932     /* hypertas */
933     add_str(hypertas, "hcall-pft");
934     add_str(hypertas, "hcall-term");
935     add_str(hypertas, "hcall-dabr");
936     add_str(hypertas, "hcall-interrupt");
937     add_str(hypertas, "hcall-tce");
938     add_str(hypertas, "hcall-vio");
939     add_str(hypertas, "hcall-splpar");
940     add_str(hypertas, "hcall-join");
941     add_str(hypertas, "hcall-bulk");
942     add_str(hypertas, "hcall-set-mode");
943     add_str(hypertas, "hcall-sprg0");
944     add_str(hypertas, "hcall-copy");
945     add_str(hypertas, "hcall-debug");
946     add_str(hypertas, "hcall-vphn");
947     add_str(qemu_hypertas, "hcall-memop1");
948 
949     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
950         add_str(hypertas, "hcall-multi-tce");
951     }
952 
953     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
954         add_str(hypertas, "hcall-hpt-resize");
955     }
956 
957     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
958                      hypertas->str, hypertas->len));
959     g_string_free(hypertas, TRUE);
960     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
961                      qemu_hypertas->str, qemu_hypertas->len));
962     g_string_free(qemu_hypertas, TRUE);
963 
964     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
965                      refpoints, sizeof(refpoints)));
966 
967     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
968                      maxdomains, sizeof(maxdomains)));
969 
970     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
971                           RTAS_ERROR_LOG_MAX));
972     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
973                           RTAS_EVENT_SCAN_RATE));
974 
975     g_assert(msi_nonbroken);
976     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
977 
978     /*
979      * According to PAPR, rtas ibm,os-term does not guarantee a return
980      * back to the guest cpu.
981      *
982      * While an additional ibm,extended-os-term property indicates
983      * that rtas call return will always occur. Set this property.
984      */
985     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
986 
987     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
988                      lrdr_capacity, sizeof(lrdr_capacity)));
989 
990     spapr_dt_rtas_tokens(fdt, rtas);
991 }
992 
993 /*
994  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
995  * and the XIVE features that the guest may request and thus the valid
996  * values for bytes 23..26 of option vector 5:
997  */
998 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
999                                           int chosen)
1000 {
1001     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1002 
1003     char val[2 * 4] = {
1004         23, 0x00, /* XICS / XIVE mode */
1005         24, 0x00, /* Hash/Radix, filled in below. */
1006         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1007         26, 0x40, /* Radix options: GTSE == yes. */
1008     };
1009 
1010     if (spapr->irq->xics && spapr->irq->xive) {
1011         val[1] = SPAPR_OV5_XIVE_BOTH;
1012     } else if (spapr->irq->xive) {
1013         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1014     } else {
1015         assert(spapr->irq->xics);
1016         val[1] = SPAPR_OV5_XIVE_LEGACY;
1017     }
1018 
1019     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1020                           first_ppc_cpu->compat_pvr)) {
1021         /*
1022          * If we're in a pre POWER9 compat mode then the guest should
1023          * do hash and use the legacy interrupt mode
1024          */
1025         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1026         val[3] = 0x00; /* Hash */
1027     } else if (kvm_enabled()) {
1028         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1029             val[3] = 0x80; /* OV5_MMU_BOTH */
1030         } else if (kvmppc_has_cap_mmu_radix()) {
1031             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1032         } else {
1033             val[3] = 0x00; /* Hash */
1034         }
1035     } else {
1036         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1037         val[3] = 0xC0;
1038     }
1039     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1040                      val, sizeof(val)));
1041 }
1042 
1043 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1044 {
1045     MachineState *machine = MACHINE(spapr);
1046     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1047     int chosen;
1048     const char *boot_device = machine->boot_order;
1049     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1050     size_t cb = 0;
1051     char *bootlist = get_boot_devices_list(&cb);
1052 
1053     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1054 
1055     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1056         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1057                                 machine->kernel_cmdline));
1058     }
1059     if (spapr->initrd_size) {
1060         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1061                               spapr->initrd_base));
1062         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1063                               spapr->initrd_base + spapr->initrd_size));
1064     }
1065 
1066     if (spapr->kernel_size) {
1067         uint64_t kprop[2] = { cpu_to_be64(spapr->kernel_addr),
1068                               cpu_to_be64(spapr->kernel_size) };
1069 
1070         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1071                          &kprop, sizeof(kprop)));
1072         if (spapr->kernel_le) {
1073             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1074         }
1075     }
1076     if (boot_menu) {
1077         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1078     }
1079     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1080     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1081     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1082 
1083     if (cb && bootlist) {
1084         int i;
1085 
1086         for (i = 0; i < cb; i++) {
1087             if (bootlist[i] == '\n') {
1088                 bootlist[i] = ' ';
1089             }
1090         }
1091         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1092     }
1093 
1094     if (boot_device && strlen(boot_device)) {
1095         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1096     }
1097 
1098     if (!spapr->has_graphics && stdout_path) {
1099         /*
1100          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1101          * kernel. New platforms should only use the "stdout-path" property. Set
1102          * the new property and continue using older property to remain
1103          * compatible with the existing firmware.
1104          */
1105         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1106         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1107     }
1108 
1109     /* We can deal with BAR reallocation just fine, advertise it to the guest */
1110     if (smc->linux_pci_probe) {
1111         _FDT(fdt_setprop_cell(fdt, chosen, "linux,pci-probe-only", 0));
1112     }
1113 
1114     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1115 
1116     g_free(stdout_path);
1117     g_free(bootlist);
1118 }
1119 
1120 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1121 {
1122     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1123      * KVM to work under pHyp with some guest co-operation */
1124     int hypervisor;
1125     uint8_t hypercall[16];
1126 
1127     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1128     /* indicate KVM hypercall interface */
1129     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1130     if (kvmppc_has_cap_fixup_hcalls()) {
1131         /*
1132          * Older KVM versions with older guest kernels were broken
1133          * with the magic page, don't allow the guest to map it.
1134          */
1135         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1136                                   sizeof(hypercall))) {
1137             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1138                              hypercall, sizeof(hypercall)));
1139         }
1140     }
1141 }
1142 
1143 void *spapr_build_fdt(SpaprMachineState *spapr, bool reset, size_t space)
1144 {
1145     MachineState *machine = MACHINE(spapr);
1146     MachineClass *mc = MACHINE_GET_CLASS(machine);
1147     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1148     int ret;
1149     void *fdt;
1150     SpaprPhbState *phb;
1151     char *buf;
1152 
1153     fdt = g_malloc0(space);
1154     _FDT((fdt_create_empty_tree(fdt, space)));
1155 
1156     /* Root node */
1157     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1158     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1159     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1160 
1161     /* Guest UUID & Name*/
1162     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1163     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1164     if (qemu_uuid_set) {
1165         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1166     }
1167     g_free(buf);
1168 
1169     if (qemu_get_vm_name()) {
1170         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1171                                 qemu_get_vm_name()));
1172     }
1173 
1174     /* Host Model & Serial Number */
1175     if (spapr->host_model) {
1176         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1177     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1178         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1179         g_free(buf);
1180     }
1181 
1182     if (spapr->host_serial) {
1183         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1184     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1185         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1186         g_free(buf);
1187     }
1188 
1189     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1190     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1191 
1192     /* /interrupt controller */
1193     spapr_irq_dt(spapr, spapr_max_server_number(spapr), fdt, PHANDLE_INTC);
1194 
1195     ret = spapr_populate_memory(spapr, fdt);
1196     if (ret < 0) {
1197         error_report("couldn't setup memory nodes in fdt");
1198         exit(1);
1199     }
1200 
1201     /* /vdevice */
1202     spapr_dt_vdevice(spapr->vio_bus, fdt);
1203 
1204     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1205         ret = spapr_rng_populate_dt(fdt);
1206         if (ret < 0) {
1207             error_report("could not set up rng device in the fdt");
1208             exit(1);
1209         }
1210     }
1211 
1212     QLIST_FOREACH(phb, &spapr->phbs, list) {
1213         ret = spapr_dt_phb(spapr, phb, PHANDLE_INTC, fdt, NULL);
1214         if (ret < 0) {
1215             error_report("couldn't setup PCI devices in fdt");
1216             exit(1);
1217         }
1218     }
1219 
1220     /* cpus */
1221     spapr_populate_cpus_dt_node(fdt, spapr);
1222 
1223     if (smc->dr_lmb_enabled) {
1224         _FDT(spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1225     }
1226 
1227     if (mc->has_hotpluggable_cpus) {
1228         int offset = fdt_path_offset(fdt, "/cpus");
1229         ret = spapr_dt_drc(fdt, offset, NULL, SPAPR_DR_CONNECTOR_TYPE_CPU);
1230         if (ret < 0) {
1231             error_report("Couldn't set up CPU DR device tree properties");
1232             exit(1);
1233         }
1234     }
1235 
1236     /* /event-sources */
1237     spapr_dt_events(spapr, fdt);
1238 
1239     /* /rtas */
1240     spapr_dt_rtas(spapr, fdt);
1241 
1242     /* /chosen */
1243     if (reset) {
1244         spapr_dt_chosen(spapr, fdt);
1245     }
1246 
1247     /* /hypervisor */
1248     if (kvm_enabled()) {
1249         spapr_dt_hypervisor(spapr, fdt);
1250     }
1251 
1252     /* Build memory reserve map */
1253     if (reset) {
1254         if (spapr->kernel_size) {
1255             _FDT((fdt_add_mem_rsv(fdt, spapr->kernel_addr,
1256                                   spapr->kernel_size)));
1257         }
1258         if (spapr->initrd_size) {
1259             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1260                                   spapr->initrd_size)));
1261         }
1262     }
1263 
1264     /* ibm,client-architecture-support updates */
1265     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1266     if (ret < 0) {
1267         error_report("couldn't setup CAS properties fdt");
1268         exit(1);
1269     }
1270 
1271     if (smc->dr_phb_enabled) {
1272         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1273         if (ret < 0) {
1274             error_report("Couldn't set up PHB DR device tree properties");
1275             exit(1);
1276         }
1277     }
1278 
1279     /* NVDIMM devices */
1280     if (mc->nvdimm_supported) {
1281         spapr_dt_persistent_memory(fdt);
1282     }
1283 
1284     return fdt;
1285 }
1286 
1287 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1288 {
1289     SpaprMachineState *spapr = opaque;
1290 
1291     return (addr & 0x0fffffff) + spapr->kernel_addr;
1292 }
1293 
1294 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1295                                     PowerPCCPU *cpu)
1296 {
1297     CPUPPCState *env = &cpu->env;
1298 
1299     /* The TCG path should also be holding the BQL at this point */
1300     g_assert(qemu_mutex_iothread_locked());
1301 
1302     if (msr_pr) {
1303         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1304         env->gpr[3] = H_PRIVILEGE;
1305     } else {
1306         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1307     }
1308 }
1309 
1310 struct LPCRSyncState {
1311     target_ulong value;
1312     target_ulong mask;
1313 };
1314 
1315 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1316 {
1317     struct LPCRSyncState *s = arg.host_ptr;
1318     PowerPCCPU *cpu = POWERPC_CPU(cs);
1319     CPUPPCState *env = &cpu->env;
1320     target_ulong lpcr;
1321 
1322     cpu_synchronize_state(cs);
1323     lpcr = env->spr[SPR_LPCR];
1324     lpcr &= ~s->mask;
1325     lpcr |= s->value;
1326     ppc_store_lpcr(cpu, lpcr);
1327 }
1328 
1329 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1330 {
1331     CPUState *cs;
1332     struct LPCRSyncState s = {
1333         .value = value,
1334         .mask = mask
1335     };
1336     CPU_FOREACH(cs) {
1337         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1338     }
1339 }
1340 
1341 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1342 {
1343     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1344 
1345     /* Copy PATE1:GR into PATE0:HR */
1346     entry->dw0 = spapr->patb_entry & PATE0_HR;
1347     entry->dw1 = spapr->patb_entry;
1348 }
1349 
1350 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1351 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1352 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1353 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1354 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1355 
1356 /*
1357  * Get the fd to access the kernel htab, re-opening it if necessary
1358  */
1359 static int get_htab_fd(SpaprMachineState *spapr)
1360 {
1361     Error *local_err = NULL;
1362 
1363     if (spapr->htab_fd >= 0) {
1364         return spapr->htab_fd;
1365     }
1366 
1367     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1368     if (spapr->htab_fd < 0) {
1369         error_report_err(local_err);
1370     }
1371 
1372     return spapr->htab_fd;
1373 }
1374 
1375 void close_htab_fd(SpaprMachineState *spapr)
1376 {
1377     if (spapr->htab_fd >= 0) {
1378         close(spapr->htab_fd);
1379     }
1380     spapr->htab_fd = -1;
1381 }
1382 
1383 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1384 {
1385     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1386 
1387     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1388 }
1389 
1390 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1391 {
1392     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1393 
1394     assert(kvm_enabled());
1395 
1396     if (!spapr->htab) {
1397         return 0;
1398     }
1399 
1400     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1401 }
1402 
1403 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1404                                                 hwaddr ptex, int n)
1405 {
1406     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1407     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1408 
1409     if (!spapr->htab) {
1410         /*
1411          * HTAB is controlled by KVM. Fetch into temporary buffer
1412          */
1413         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1414         kvmppc_read_hptes(hptes, ptex, n);
1415         return hptes;
1416     }
1417 
1418     /*
1419      * HTAB is controlled by QEMU. Just point to the internally
1420      * accessible PTEG.
1421      */
1422     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1423 }
1424 
1425 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1426                               const ppc_hash_pte64_t *hptes,
1427                               hwaddr ptex, int n)
1428 {
1429     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1430 
1431     if (!spapr->htab) {
1432         g_free((void *)hptes);
1433     }
1434 
1435     /* Nothing to do for qemu managed HPT */
1436 }
1437 
1438 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1439                       uint64_t pte0, uint64_t pte1)
1440 {
1441     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1442     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1443 
1444     if (!spapr->htab) {
1445         kvmppc_write_hpte(ptex, pte0, pte1);
1446     } else {
1447         if (pte0 & HPTE64_V_VALID) {
1448             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1449             /*
1450              * When setting valid, we write PTE1 first. This ensures
1451              * proper synchronization with the reading code in
1452              * ppc_hash64_pteg_search()
1453              */
1454             smp_wmb();
1455             stq_p(spapr->htab + offset, pte0);
1456         } else {
1457             stq_p(spapr->htab + offset, pte0);
1458             /*
1459              * When clearing it we set PTE0 first. This ensures proper
1460              * synchronization with the reading code in
1461              * ppc_hash64_pteg_search()
1462              */
1463             smp_wmb();
1464             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1465         }
1466     }
1467 }
1468 
1469 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1470                              uint64_t pte1)
1471 {
1472     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1473     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1474 
1475     if (!spapr->htab) {
1476         /* There should always be a hash table when this is called */
1477         error_report("spapr_hpte_set_c called with no hash table !");
1478         return;
1479     }
1480 
1481     /* The HW performs a non-atomic byte update */
1482     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1483 }
1484 
1485 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1486                              uint64_t pte1)
1487 {
1488     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1489     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1490 
1491     if (!spapr->htab) {
1492         /* There should always be a hash table when this is called */
1493         error_report("spapr_hpte_set_r called with no hash table !");
1494         return;
1495     }
1496 
1497     /* The HW performs a non-atomic byte update */
1498     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1499 }
1500 
1501 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1502 {
1503     int shift;
1504 
1505     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1506      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1507      * that's much more than is needed for Linux guests */
1508     shift = ctz64(pow2ceil(ramsize)) - 7;
1509     shift = MAX(shift, 18); /* Minimum architected size */
1510     shift = MIN(shift, 46); /* Maximum architected size */
1511     return shift;
1512 }
1513 
1514 void spapr_free_hpt(SpaprMachineState *spapr)
1515 {
1516     g_free(spapr->htab);
1517     spapr->htab = NULL;
1518     spapr->htab_shift = 0;
1519     close_htab_fd(spapr);
1520 }
1521 
1522 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1523                           Error **errp)
1524 {
1525     long rc;
1526 
1527     /* Clean up any HPT info from a previous boot */
1528     spapr_free_hpt(spapr);
1529 
1530     rc = kvmppc_reset_htab(shift);
1531     if (rc < 0) {
1532         /* kernel-side HPT needed, but couldn't allocate one */
1533         error_setg_errno(errp, errno,
1534                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1535                          shift);
1536         /* This is almost certainly fatal, but if the caller really
1537          * wants to carry on with shift == 0, it's welcome to try */
1538     } else if (rc > 0) {
1539         /* kernel-side HPT allocated */
1540         if (rc != shift) {
1541             error_setg(errp,
1542                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1543                        shift, rc);
1544         }
1545 
1546         spapr->htab_shift = shift;
1547         spapr->htab = NULL;
1548     } else {
1549         /* kernel-side HPT not needed, allocate in userspace instead */
1550         size_t size = 1ULL << shift;
1551         int i;
1552 
1553         spapr->htab = qemu_memalign(size, size);
1554         if (!spapr->htab) {
1555             error_setg_errno(errp, errno,
1556                              "Could not allocate HPT of order %d", shift);
1557             return;
1558         }
1559 
1560         memset(spapr->htab, 0, size);
1561         spapr->htab_shift = shift;
1562 
1563         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1564             DIRTY_HPTE(HPTE(spapr->htab, i));
1565         }
1566     }
1567     /* We're setting up a hash table, so that means we're not radix */
1568     spapr->patb_entry = 0;
1569     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1570 }
1571 
1572 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1573 {
1574     int hpt_shift;
1575 
1576     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1577         || (spapr->cas_reboot
1578             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1579         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1580     } else {
1581         uint64_t current_ram_size;
1582 
1583         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1584         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1585     }
1586     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1587 
1588     if (spapr->vrma_adjust) {
1589         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1590 
1591         spapr->rma_size = MIN(spapr_node0_size(MACHINE(spapr)), vrma_limit);
1592     }
1593 }
1594 
1595 static int spapr_reset_drcs(Object *child, void *opaque)
1596 {
1597     SpaprDrc *drc =
1598         (SpaprDrc *) object_dynamic_cast(child,
1599                                                  TYPE_SPAPR_DR_CONNECTOR);
1600 
1601     if (drc) {
1602         spapr_drc_reset(drc);
1603     }
1604 
1605     return 0;
1606 }
1607 
1608 static void spapr_machine_reset(MachineState *machine)
1609 {
1610     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1611     PowerPCCPU *first_ppc_cpu;
1612     hwaddr fdt_addr;
1613     void *fdt;
1614     int rc;
1615 
1616     kvmppc_svm_off(&error_fatal);
1617     spapr_caps_apply(spapr);
1618 
1619     first_ppc_cpu = POWERPC_CPU(first_cpu);
1620     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1621         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1622                               spapr->max_compat_pvr)) {
1623         /*
1624          * If using KVM with radix mode available, VCPUs can be started
1625          * without a HPT because KVM will start them in radix mode.
1626          * Set the GR bit in PATE so that we know there is no HPT.
1627          */
1628         spapr->patb_entry = PATE1_GR;
1629         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1630     } else {
1631         spapr_setup_hpt_and_vrma(spapr);
1632     }
1633 
1634     qemu_devices_reset();
1635 
1636     /*
1637      * If this reset wasn't generated by CAS, we should reset our
1638      * negotiated options and start from scratch
1639      */
1640     if (!spapr->cas_reboot) {
1641         spapr_ovec_cleanup(spapr->ov5_cas);
1642         spapr->ov5_cas = spapr_ovec_new();
1643 
1644         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1645     }
1646 
1647     /*
1648      * This is fixing some of the default configuration of the XIVE
1649      * devices. To be called after the reset of the machine devices.
1650      */
1651     spapr_irq_reset(spapr, &error_fatal);
1652 
1653     /*
1654      * There is no CAS under qtest. Simulate one to please the code that
1655      * depends on spapr->ov5_cas. This is especially needed to test device
1656      * unplug, so we do that before resetting the DRCs.
1657      */
1658     if (qtest_enabled()) {
1659         spapr_ovec_cleanup(spapr->ov5_cas);
1660         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1661     }
1662 
1663     /* DRC reset may cause a device to be unplugged. This will cause troubles
1664      * if this device is used by another device (eg, a running vhost backend
1665      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1666      * situations, we reset DRCs after all devices have been reset.
1667      */
1668     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1669 
1670     spapr_clear_pending_events(spapr);
1671 
1672     /*
1673      * We place the device tree and RTAS just below either the top of the RMA,
1674      * or just below 2GB, whichever is lower, so that it can be
1675      * processed with 32-bit real mode code if necessary
1676      */
1677     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1678 
1679     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1680 
1681     rc = fdt_pack(fdt);
1682 
1683     /* Should only fail if we've built a corrupted tree */
1684     assert(rc == 0);
1685 
1686     /* Load the fdt */
1687     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1688     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1689     g_free(spapr->fdt_blob);
1690     spapr->fdt_size = fdt_totalsize(fdt);
1691     spapr->fdt_initial_size = spapr->fdt_size;
1692     spapr->fdt_blob = fdt;
1693 
1694     /* Set up the entry state */
1695     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1696     first_ppc_cpu->env.gpr[5] = 0;
1697 
1698     spapr->cas_reboot = false;
1699 
1700     spapr->mc_status = -1;
1701     spapr->guest_machine_check_addr = -1;
1702 
1703     /* Signal all vCPUs waiting on this condition */
1704     qemu_cond_broadcast(&spapr->mc_delivery_cond);
1705 
1706     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1707 }
1708 
1709 static void spapr_create_nvram(SpaprMachineState *spapr)
1710 {
1711     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1712     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1713 
1714     if (dinfo) {
1715         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1716                             &error_fatal);
1717     }
1718 
1719     qdev_init_nofail(dev);
1720 
1721     spapr->nvram = (struct SpaprNvram *)dev;
1722 }
1723 
1724 static void spapr_rtc_create(SpaprMachineState *spapr)
1725 {
1726     object_initialize_child(OBJECT(spapr), "rtc",
1727                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1728                             &error_fatal, NULL);
1729     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1730                               &error_fatal);
1731     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1732                               "date", &error_fatal);
1733 }
1734 
1735 /* Returns whether we want to use VGA or not */
1736 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1737 {
1738     switch (vga_interface_type) {
1739     case VGA_NONE:
1740         return false;
1741     case VGA_DEVICE:
1742         return true;
1743     case VGA_STD:
1744     case VGA_VIRTIO:
1745     case VGA_CIRRUS:
1746         return pci_vga_init(pci_bus) != NULL;
1747     default:
1748         error_setg(errp,
1749                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1750         return false;
1751     }
1752 }
1753 
1754 static int spapr_pre_load(void *opaque)
1755 {
1756     int rc;
1757 
1758     rc = spapr_caps_pre_load(opaque);
1759     if (rc) {
1760         return rc;
1761     }
1762 
1763     return 0;
1764 }
1765 
1766 static int spapr_post_load(void *opaque, int version_id)
1767 {
1768     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1769     int err = 0;
1770 
1771     err = spapr_caps_post_migration(spapr);
1772     if (err) {
1773         return err;
1774     }
1775 
1776     /*
1777      * In earlier versions, there was no separate qdev for the PAPR
1778      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1779      * So when migrating from those versions, poke the incoming offset
1780      * value into the RTC device
1781      */
1782     if (version_id < 3) {
1783         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1784         if (err) {
1785             return err;
1786         }
1787     }
1788 
1789     if (kvm_enabled() && spapr->patb_entry) {
1790         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1791         bool radix = !!(spapr->patb_entry & PATE1_GR);
1792         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1793 
1794         /*
1795          * Update LPCR:HR and UPRT as they may not be set properly in
1796          * the stream
1797          */
1798         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1799                             LPCR_HR | LPCR_UPRT);
1800 
1801         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1802         if (err) {
1803             error_report("Process table config unsupported by the host");
1804             return -EINVAL;
1805         }
1806     }
1807 
1808     err = spapr_irq_post_load(spapr, version_id);
1809     if (err) {
1810         return err;
1811     }
1812 
1813     return err;
1814 }
1815 
1816 static int spapr_pre_save(void *opaque)
1817 {
1818     int rc;
1819 
1820     rc = spapr_caps_pre_save(opaque);
1821     if (rc) {
1822         return rc;
1823     }
1824 
1825     return 0;
1826 }
1827 
1828 static bool version_before_3(void *opaque, int version_id)
1829 {
1830     return version_id < 3;
1831 }
1832 
1833 static bool spapr_pending_events_needed(void *opaque)
1834 {
1835     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1836     return !QTAILQ_EMPTY(&spapr->pending_events);
1837 }
1838 
1839 static const VMStateDescription vmstate_spapr_event_entry = {
1840     .name = "spapr_event_log_entry",
1841     .version_id = 1,
1842     .minimum_version_id = 1,
1843     .fields = (VMStateField[]) {
1844         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1845         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1846         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1847                                      NULL, extended_length),
1848         VMSTATE_END_OF_LIST()
1849     },
1850 };
1851 
1852 static const VMStateDescription vmstate_spapr_pending_events = {
1853     .name = "spapr_pending_events",
1854     .version_id = 1,
1855     .minimum_version_id = 1,
1856     .needed = spapr_pending_events_needed,
1857     .fields = (VMStateField[]) {
1858         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1859                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1860         VMSTATE_END_OF_LIST()
1861     },
1862 };
1863 
1864 static bool spapr_ov5_cas_needed(void *opaque)
1865 {
1866     SpaprMachineState *spapr = opaque;
1867     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1868     bool cas_needed;
1869 
1870     /* Prior to the introduction of SpaprOptionVector, we had two option
1871      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1872      * Both of these options encode machine topology into the device-tree
1873      * in such a way that the now-booted OS should still be able to interact
1874      * appropriately with QEMU regardless of what options were actually
1875      * negotiatied on the source side.
1876      *
1877      * As such, we can avoid migrating the CAS-negotiated options if these
1878      * are the only options available on the current machine/platform.
1879      * Since these are the only options available for pseries-2.7 and
1880      * earlier, this allows us to maintain old->new/new->old migration
1881      * compatibility.
1882      *
1883      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1884      * via default pseries-2.8 machines and explicit command-line parameters.
1885      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1886      * of the actual CAS-negotiated values to continue working properly. For
1887      * example, availability of memory unplug depends on knowing whether
1888      * OV5_HP_EVT was negotiated via CAS.
1889      *
1890      * Thus, for any cases where the set of available CAS-negotiatable
1891      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1892      * include the CAS-negotiated options in the migration stream, unless
1893      * if they affect boot time behaviour only.
1894      */
1895     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1896     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1897     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1898 
1899     /* We need extra information if we have any bits outside the mask
1900      * defined above */
1901     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1902 
1903     spapr_ovec_cleanup(ov5_mask);
1904 
1905     return cas_needed;
1906 }
1907 
1908 static const VMStateDescription vmstate_spapr_ov5_cas = {
1909     .name = "spapr_option_vector_ov5_cas",
1910     .version_id = 1,
1911     .minimum_version_id = 1,
1912     .needed = spapr_ov5_cas_needed,
1913     .fields = (VMStateField[]) {
1914         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1915                                  vmstate_spapr_ovec, SpaprOptionVector),
1916         VMSTATE_END_OF_LIST()
1917     },
1918 };
1919 
1920 static bool spapr_patb_entry_needed(void *opaque)
1921 {
1922     SpaprMachineState *spapr = opaque;
1923 
1924     return !!spapr->patb_entry;
1925 }
1926 
1927 static const VMStateDescription vmstate_spapr_patb_entry = {
1928     .name = "spapr_patb_entry",
1929     .version_id = 1,
1930     .minimum_version_id = 1,
1931     .needed = spapr_patb_entry_needed,
1932     .fields = (VMStateField[]) {
1933         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1934         VMSTATE_END_OF_LIST()
1935     },
1936 };
1937 
1938 static bool spapr_irq_map_needed(void *opaque)
1939 {
1940     SpaprMachineState *spapr = opaque;
1941 
1942     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1943 }
1944 
1945 static const VMStateDescription vmstate_spapr_irq_map = {
1946     .name = "spapr_irq_map",
1947     .version_id = 1,
1948     .minimum_version_id = 1,
1949     .needed = spapr_irq_map_needed,
1950     .fields = (VMStateField[]) {
1951         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1952         VMSTATE_END_OF_LIST()
1953     },
1954 };
1955 
1956 static bool spapr_dtb_needed(void *opaque)
1957 {
1958     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1959 
1960     return smc->update_dt_enabled;
1961 }
1962 
1963 static int spapr_dtb_pre_load(void *opaque)
1964 {
1965     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1966 
1967     g_free(spapr->fdt_blob);
1968     spapr->fdt_blob = NULL;
1969     spapr->fdt_size = 0;
1970 
1971     return 0;
1972 }
1973 
1974 static const VMStateDescription vmstate_spapr_dtb = {
1975     .name = "spapr_dtb",
1976     .version_id = 1,
1977     .minimum_version_id = 1,
1978     .needed = spapr_dtb_needed,
1979     .pre_load = spapr_dtb_pre_load,
1980     .fields = (VMStateField[]) {
1981         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1982         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1983         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1984                                      fdt_size),
1985         VMSTATE_END_OF_LIST()
1986     },
1987 };
1988 
1989 static bool spapr_fwnmi_needed(void *opaque)
1990 {
1991     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1992 
1993     return spapr->guest_machine_check_addr != -1;
1994 }
1995 
1996 static int spapr_fwnmi_pre_save(void *opaque)
1997 {
1998     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1999 
2000     /*
2001      * Check if machine check handling is in progress and print a
2002      * warning message.
2003      */
2004     if (spapr->mc_status != -1) {
2005         warn_report("A machine check is being handled during migration. The"
2006                 "handler may run and log hardware error on the destination");
2007     }
2008 
2009     return 0;
2010 }
2011 
2012 static const VMStateDescription vmstate_spapr_machine_check = {
2013     .name = "spapr_machine_check",
2014     .version_id = 1,
2015     .minimum_version_id = 1,
2016     .needed = spapr_fwnmi_needed,
2017     .pre_save = spapr_fwnmi_pre_save,
2018     .fields = (VMStateField[]) {
2019         VMSTATE_UINT64(guest_machine_check_addr, SpaprMachineState),
2020         VMSTATE_INT32(mc_status, SpaprMachineState),
2021         VMSTATE_END_OF_LIST()
2022     },
2023 };
2024 
2025 static const VMStateDescription vmstate_spapr = {
2026     .name = "spapr",
2027     .version_id = 3,
2028     .minimum_version_id = 1,
2029     .pre_load = spapr_pre_load,
2030     .post_load = spapr_post_load,
2031     .pre_save = spapr_pre_save,
2032     .fields = (VMStateField[]) {
2033         /* used to be @next_irq */
2034         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2035 
2036         /* RTC offset */
2037         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2038 
2039         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2040         VMSTATE_END_OF_LIST()
2041     },
2042     .subsections = (const VMStateDescription*[]) {
2043         &vmstate_spapr_ov5_cas,
2044         &vmstate_spapr_patb_entry,
2045         &vmstate_spapr_pending_events,
2046         &vmstate_spapr_cap_htm,
2047         &vmstate_spapr_cap_vsx,
2048         &vmstate_spapr_cap_dfp,
2049         &vmstate_spapr_cap_cfpc,
2050         &vmstate_spapr_cap_sbbc,
2051         &vmstate_spapr_cap_ibs,
2052         &vmstate_spapr_cap_hpt_maxpagesize,
2053         &vmstate_spapr_irq_map,
2054         &vmstate_spapr_cap_nested_kvm_hv,
2055         &vmstate_spapr_dtb,
2056         &vmstate_spapr_cap_large_decr,
2057         &vmstate_spapr_cap_ccf_assist,
2058         &vmstate_spapr_cap_fwnmi,
2059         &vmstate_spapr_machine_check,
2060         NULL
2061     }
2062 };
2063 
2064 static int htab_save_setup(QEMUFile *f, void *opaque)
2065 {
2066     SpaprMachineState *spapr = opaque;
2067 
2068     /* "Iteration" header */
2069     if (!spapr->htab_shift) {
2070         qemu_put_be32(f, -1);
2071     } else {
2072         qemu_put_be32(f, spapr->htab_shift);
2073     }
2074 
2075     if (spapr->htab) {
2076         spapr->htab_save_index = 0;
2077         spapr->htab_first_pass = true;
2078     } else {
2079         if (spapr->htab_shift) {
2080             assert(kvm_enabled());
2081         }
2082     }
2083 
2084 
2085     return 0;
2086 }
2087 
2088 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2089                             int chunkstart, int n_valid, int n_invalid)
2090 {
2091     qemu_put_be32(f, chunkstart);
2092     qemu_put_be16(f, n_valid);
2093     qemu_put_be16(f, n_invalid);
2094     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2095                     HASH_PTE_SIZE_64 * n_valid);
2096 }
2097 
2098 static void htab_save_end_marker(QEMUFile *f)
2099 {
2100     qemu_put_be32(f, 0);
2101     qemu_put_be16(f, 0);
2102     qemu_put_be16(f, 0);
2103 }
2104 
2105 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2106                                  int64_t max_ns)
2107 {
2108     bool has_timeout = max_ns != -1;
2109     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2110     int index = spapr->htab_save_index;
2111     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2112 
2113     assert(spapr->htab_first_pass);
2114 
2115     do {
2116         int chunkstart;
2117 
2118         /* Consume invalid HPTEs */
2119         while ((index < htabslots)
2120                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2121             CLEAN_HPTE(HPTE(spapr->htab, index));
2122             index++;
2123         }
2124 
2125         /* Consume valid HPTEs */
2126         chunkstart = index;
2127         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2128                && HPTE_VALID(HPTE(spapr->htab, index))) {
2129             CLEAN_HPTE(HPTE(spapr->htab, index));
2130             index++;
2131         }
2132 
2133         if (index > chunkstart) {
2134             int n_valid = index - chunkstart;
2135 
2136             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2137 
2138             if (has_timeout &&
2139                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2140                 break;
2141             }
2142         }
2143     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2144 
2145     if (index >= htabslots) {
2146         assert(index == htabslots);
2147         index = 0;
2148         spapr->htab_first_pass = false;
2149     }
2150     spapr->htab_save_index = index;
2151 }
2152 
2153 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2154                                 int64_t max_ns)
2155 {
2156     bool final = max_ns < 0;
2157     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2158     int examined = 0, sent = 0;
2159     int index = spapr->htab_save_index;
2160     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2161 
2162     assert(!spapr->htab_first_pass);
2163 
2164     do {
2165         int chunkstart, invalidstart;
2166 
2167         /* Consume non-dirty HPTEs */
2168         while ((index < htabslots)
2169                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2170             index++;
2171             examined++;
2172         }
2173 
2174         chunkstart = index;
2175         /* Consume valid dirty HPTEs */
2176         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2177                && HPTE_DIRTY(HPTE(spapr->htab, index))
2178                && HPTE_VALID(HPTE(spapr->htab, index))) {
2179             CLEAN_HPTE(HPTE(spapr->htab, index));
2180             index++;
2181             examined++;
2182         }
2183 
2184         invalidstart = index;
2185         /* Consume invalid dirty HPTEs */
2186         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2187                && HPTE_DIRTY(HPTE(spapr->htab, index))
2188                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2189             CLEAN_HPTE(HPTE(spapr->htab, index));
2190             index++;
2191             examined++;
2192         }
2193 
2194         if (index > chunkstart) {
2195             int n_valid = invalidstart - chunkstart;
2196             int n_invalid = index - invalidstart;
2197 
2198             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2199             sent += index - chunkstart;
2200 
2201             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2202                 break;
2203             }
2204         }
2205 
2206         if (examined >= htabslots) {
2207             break;
2208         }
2209 
2210         if (index >= htabslots) {
2211             assert(index == htabslots);
2212             index = 0;
2213         }
2214     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2215 
2216     if (index >= htabslots) {
2217         assert(index == htabslots);
2218         index = 0;
2219     }
2220 
2221     spapr->htab_save_index = index;
2222 
2223     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2224 }
2225 
2226 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2227 #define MAX_KVM_BUF_SIZE    2048
2228 
2229 static int htab_save_iterate(QEMUFile *f, void *opaque)
2230 {
2231     SpaprMachineState *spapr = opaque;
2232     int fd;
2233     int rc = 0;
2234 
2235     /* Iteration header */
2236     if (!spapr->htab_shift) {
2237         qemu_put_be32(f, -1);
2238         return 1;
2239     } else {
2240         qemu_put_be32(f, 0);
2241     }
2242 
2243     if (!spapr->htab) {
2244         assert(kvm_enabled());
2245 
2246         fd = get_htab_fd(spapr);
2247         if (fd < 0) {
2248             return fd;
2249         }
2250 
2251         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2252         if (rc < 0) {
2253             return rc;
2254         }
2255     } else  if (spapr->htab_first_pass) {
2256         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2257     } else {
2258         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2259     }
2260 
2261     htab_save_end_marker(f);
2262 
2263     return rc;
2264 }
2265 
2266 static int htab_save_complete(QEMUFile *f, void *opaque)
2267 {
2268     SpaprMachineState *spapr = opaque;
2269     int fd;
2270 
2271     /* Iteration header */
2272     if (!spapr->htab_shift) {
2273         qemu_put_be32(f, -1);
2274         return 0;
2275     } else {
2276         qemu_put_be32(f, 0);
2277     }
2278 
2279     if (!spapr->htab) {
2280         int rc;
2281 
2282         assert(kvm_enabled());
2283 
2284         fd = get_htab_fd(spapr);
2285         if (fd < 0) {
2286             return fd;
2287         }
2288 
2289         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2290         if (rc < 0) {
2291             return rc;
2292         }
2293     } else {
2294         if (spapr->htab_first_pass) {
2295             htab_save_first_pass(f, spapr, -1);
2296         }
2297         htab_save_later_pass(f, spapr, -1);
2298     }
2299 
2300     /* End marker */
2301     htab_save_end_marker(f);
2302 
2303     return 0;
2304 }
2305 
2306 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2307 {
2308     SpaprMachineState *spapr = opaque;
2309     uint32_t section_hdr;
2310     int fd = -1;
2311     Error *local_err = NULL;
2312 
2313     if (version_id < 1 || version_id > 1) {
2314         error_report("htab_load() bad version");
2315         return -EINVAL;
2316     }
2317 
2318     section_hdr = qemu_get_be32(f);
2319 
2320     if (section_hdr == -1) {
2321         spapr_free_hpt(spapr);
2322         return 0;
2323     }
2324 
2325     if (section_hdr) {
2326         /* First section gives the htab size */
2327         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2328         if (local_err) {
2329             error_report_err(local_err);
2330             return -EINVAL;
2331         }
2332         return 0;
2333     }
2334 
2335     if (!spapr->htab) {
2336         assert(kvm_enabled());
2337 
2338         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2339         if (fd < 0) {
2340             error_report_err(local_err);
2341             return fd;
2342         }
2343     }
2344 
2345     while (true) {
2346         uint32_t index;
2347         uint16_t n_valid, n_invalid;
2348 
2349         index = qemu_get_be32(f);
2350         n_valid = qemu_get_be16(f);
2351         n_invalid = qemu_get_be16(f);
2352 
2353         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2354             /* End of Stream */
2355             break;
2356         }
2357 
2358         if ((index + n_valid + n_invalid) >
2359             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2360             /* Bad index in stream */
2361             error_report(
2362                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2363                 index, n_valid, n_invalid, spapr->htab_shift);
2364             return -EINVAL;
2365         }
2366 
2367         if (spapr->htab) {
2368             if (n_valid) {
2369                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2370                                 HASH_PTE_SIZE_64 * n_valid);
2371             }
2372             if (n_invalid) {
2373                 memset(HPTE(spapr->htab, index + n_valid), 0,
2374                        HASH_PTE_SIZE_64 * n_invalid);
2375             }
2376         } else {
2377             int rc;
2378 
2379             assert(fd >= 0);
2380 
2381             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2382             if (rc < 0) {
2383                 return rc;
2384             }
2385         }
2386     }
2387 
2388     if (!spapr->htab) {
2389         assert(fd >= 0);
2390         close(fd);
2391     }
2392 
2393     return 0;
2394 }
2395 
2396 static void htab_save_cleanup(void *opaque)
2397 {
2398     SpaprMachineState *spapr = opaque;
2399 
2400     close_htab_fd(spapr);
2401 }
2402 
2403 static SaveVMHandlers savevm_htab_handlers = {
2404     .save_setup = htab_save_setup,
2405     .save_live_iterate = htab_save_iterate,
2406     .save_live_complete_precopy = htab_save_complete,
2407     .save_cleanup = htab_save_cleanup,
2408     .load_state = htab_load,
2409 };
2410 
2411 static void spapr_boot_set(void *opaque, const char *boot_device,
2412                            Error **errp)
2413 {
2414     MachineState *machine = MACHINE(opaque);
2415     machine->boot_order = g_strdup(boot_device);
2416 }
2417 
2418 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2419 {
2420     MachineState *machine = MACHINE(spapr);
2421     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2422     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2423     int i;
2424 
2425     for (i = 0; i < nr_lmbs; i++) {
2426         uint64_t addr;
2427 
2428         addr = i * lmb_size + machine->device_memory->base;
2429         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2430                                addr / lmb_size);
2431     }
2432 }
2433 
2434 /*
2435  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2436  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2437  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2438  */
2439 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2440 {
2441     int i;
2442 
2443     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2444         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2445                    " is not aligned to %" PRIu64 " MiB",
2446                    machine->ram_size,
2447                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2448         return;
2449     }
2450 
2451     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2452         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2453                    " is not aligned to %" PRIu64 " MiB",
2454                    machine->ram_size,
2455                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2456         return;
2457     }
2458 
2459     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2460         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2461             error_setg(errp,
2462                        "Node %d memory size 0x%" PRIx64
2463                        " is not aligned to %" PRIu64 " MiB",
2464                        i, machine->numa_state->nodes[i].node_mem,
2465                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2466             return;
2467         }
2468     }
2469 }
2470 
2471 /* find cpu slot in machine->possible_cpus by core_id */
2472 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2473 {
2474     int index = id / ms->smp.threads;
2475 
2476     if (index >= ms->possible_cpus->len) {
2477         return NULL;
2478     }
2479     if (idx) {
2480         *idx = index;
2481     }
2482     return &ms->possible_cpus->cpus[index];
2483 }
2484 
2485 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2486 {
2487     MachineState *ms = MACHINE(spapr);
2488     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2489     Error *local_err = NULL;
2490     bool vsmt_user = !!spapr->vsmt;
2491     int kvm_smt = kvmppc_smt_threads();
2492     int ret;
2493     unsigned int smp_threads = ms->smp.threads;
2494 
2495     if (!kvm_enabled() && (smp_threads > 1)) {
2496         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2497                      "on a pseries machine");
2498         goto out;
2499     }
2500     if (!is_power_of_2(smp_threads)) {
2501         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2502                      "machine because it must be a power of 2", smp_threads);
2503         goto out;
2504     }
2505 
2506     /* Detemine the VSMT mode to use: */
2507     if (vsmt_user) {
2508         if (spapr->vsmt < smp_threads) {
2509             error_setg(&local_err, "Cannot support VSMT mode %d"
2510                          " because it must be >= threads/core (%d)",
2511                          spapr->vsmt, smp_threads);
2512             goto out;
2513         }
2514         /* In this case, spapr->vsmt has been set by the command line */
2515     } else if (!smc->smp_threads_vsmt) {
2516         /*
2517          * Default VSMT value is tricky, because we need it to be as
2518          * consistent as possible (for migration), but this requires
2519          * changing it for at least some existing cases.  We pick 8 as
2520          * the value that we'd get with KVM on POWER8, the
2521          * overwhelmingly common case in production systems.
2522          */
2523         spapr->vsmt = MAX(8, smp_threads);
2524     } else {
2525         spapr->vsmt = smp_threads;
2526     }
2527 
2528     /* KVM: If necessary, set the SMT mode: */
2529     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2530         ret = kvmppc_set_smt_threads(spapr->vsmt);
2531         if (ret) {
2532             /* Looks like KVM isn't able to change VSMT mode */
2533             error_setg(&local_err,
2534                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2535                        spapr->vsmt, ret);
2536             /* We can live with that if the default one is big enough
2537              * for the number of threads, and a submultiple of the one
2538              * we want.  In this case we'll waste some vcpu ids, but
2539              * behaviour will be correct */
2540             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2541                 warn_report_err(local_err);
2542                 local_err = NULL;
2543                 goto out;
2544             } else {
2545                 if (!vsmt_user) {
2546                     error_append_hint(&local_err,
2547                                       "On PPC, a VM with %d threads/core"
2548                                       " on a host with %d threads/core"
2549                                       " requires the use of VSMT mode %d.\n",
2550                                       smp_threads, kvm_smt, spapr->vsmt);
2551                 }
2552                 kvmppc_error_append_smt_possible_hint(&local_err);
2553                 goto out;
2554             }
2555         }
2556     }
2557     /* else TCG: nothing to do currently */
2558 out:
2559     error_propagate(errp, local_err);
2560 }
2561 
2562 static void spapr_init_cpus(SpaprMachineState *spapr)
2563 {
2564     MachineState *machine = MACHINE(spapr);
2565     MachineClass *mc = MACHINE_GET_CLASS(machine);
2566     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2567     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2568     const CPUArchIdList *possible_cpus;
2569     unsigned int smp_cpus = machine->smp.cpus;
2570     unsigned int smp_threads = machine->smp.threads;
2571     unsigned int max_cpus = machine->smp.max_cpus;
2572     int boot_cores_nr = smp_cpus / smp_threads;
2573     int i;
2574 
2575     possible_cpus = mc->possible_cpu_arch_ids(machine);
2576     if (mc->has_hotpluggable_cpus) {
2577         if (smp_cpus % smp_threads) {
2578             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2579                          smp_cpus, smp_threads);
2580             exit(1);
2581         }
2582         if (max_cpus % smp_threads) {
2583             error_report("max_cpus (%u) must be multiple of threads (%u)",
2584                          max_cpus, smp_threads);
2585             exit(1);
2586         }
2587     } else {
2588         if (max_cpus != smp_cpus) {
2589             error_report("This machine version does not support CPU hotplug");
2590             exit(1);
2591         }
2592         boot_cores_nr = possible_cpus->len;
2593     }
2594 
2595     if (smc->pre_2_10_has_unused_icps) {
2596         int i;
2597 
2598         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2599             /* Dummy entries get deregistered when real ICPState objects
2600              * are registered during CPU core hotplug.
2601              */
2602             pre_2_10_vmstate_register_dummy_icp(i);
2603         }
2604     }
2605 
2606     for (i = 0; i < possible_cpus->len; i++) {
2607         int core_id = i * smp_threads;
2608 
2609         if (mc->has_hotpluggable_cpus) {
2610             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2611                                    spapr_vcpu_id(spapr, core_id));
2612         }
2613 
2614         if (i < boot_cores_nr) {
2615             Object *core  = object_new(type);
2616             int nr_threads = smp_threads;
2617 
2618             /* Handle the partially filled core for older machine types */
2619             if ((i + 1) * smp_threads >= smp_cpus) {
2620                 nr_threads = smp_cpus - i * smp_threads;
2621             }
2622 
2623             object_property_set_int(core, nr_threads, "nr-threads",
2624                                     &error_fatal);
2625             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2626                                     &error_fatal);
2627             object_property_set_bool(core, true, "realized", &error_fatal);
2628 
2629             object_unref(core);
2630         }
2631     }
2632 }
2633 
2634 static PCIHostState *spapr_create_default_phb(void)
2635 {
2636     DeviceState *dev;
2637 
2638     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2639     qdev_prop_set_uint32(dev, "index", 0);
2640     qdev_init_nofail(dev);
2641 
2642     return PCI_HOST_BRIDGE(dev);
2643 }
2644 
2645 /* pSeries LPAR / sPAPR hardware init */
2646 static void spapr_machine_init(MachineState *machine)
2647 {
2648     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2649     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2650     MachineClass *mc = MACHINE_GET_CLASS(machine);
2651     const char *kernel_filename = machine->kernel_filename;
2652     const char *initrd_filename = machine->initrd_filename;
2653     PCIHostState *phb;
2654     int i;
2655     MemoryRegion *sysmem = get_system_memory();
2656     hwaddr node0_size = spapr_node0_size(machine);
2657     long load_limit, fw_size;
2658     char *filename;
2659     Error *resize_hpt_err = NULL;
2660 
2661     msi_nonbroken = true;
2662 
2663     QLIST_INIT(&spapr->phbs);
2664     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2665 
2666     /* Determine capabilities to run with */
2667     spapr_caps_init(spapr);
2668 
2669     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2670     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2671         /*
2672          * If the user explicitly requested a mode we should either
2673          * supply it, or fail completely (which we do below).  But if
2674          * it's not set explicitly, we reset our mode to something
2675          * that works
2676          */
2677         if (resize_hpt_err) {
2678             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2679             error_free(resize_hpt_err);
2680             resize_hpt_err = NULL;
2681         } else {
2682             spapr->resize_hpt = smc->resize_hpt_default;
2683         }
2684     }
2685 
2686     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2687 
2688     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2689         /*
2690          * User requested HPT resize, but this host can't supply it.  Bail out
2691          */
2692         error_report_err(resize_hpt_err);
2693         exit(1);
2694     }
2695 
2696     spapr->rma_size = node0_size;
2697 
2698     /* With KVM, we don't actually know whether KVM supports an
2699      * unbounded RMA (PR KVM) or is limited by the hash table size
2700      * (HV KVM using VRMA), so we always assume the latter
2701      *
2702      * In that case, we also limit the initial allocations for RTAS
2703      * etc... to 256M since we have no way to know what the VRMA size
2704      * is going to be as it depends on the size of the hash table
2705      * which isn't determined yet.
2706      */
2707     if (kvm_enabled()) {
2708         spapr->vrma_adjust = 1;
2709         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2710     }
2711 
2712     /* Actually we don't support unbounded RMA anymore since we added
2713      * proper emulation of HV mode. The max we can get is 16G which
2714      * also happens to be what we configure for PAPR mode so make sure
2715      * we don't do anything bigger than that
2716      */
2717     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2718 
2719     if (spapr->rma_size > node0_size) {
2720         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2721                      spapr->rma_size);
2722         exit(1);
2723     }
2724 
2725     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2726     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2727 
2728     /*
2729      * VSMT must be set in order to be able to compute VCPU ids, ie to
2730      * call spapr_max_server_number() or spapr_vcpu_id().
2731      */
2732     spapr_set_vsmt_mode(spapr, &error_fatal);
2733 
2734     /* Set up Interrupt Controller before we create the VCPUs */
2735     spapr_irq_init(spapr, &error_fatal);
2736 
2737     /* Set up containers for ibm,client-architecture-support negotiated options
2738      */
2739     spapr->ov5 = spapr_ovec_new();
2740     spapr->ov5_cas = spapr_ovec_new();
2741 
2742     if (smc->dr_lmb_enabled) {
2743         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2744         spapr_validate_node_memory(machine, &error_fatal);
2745     }
2746 
2747     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2748 
2749     /* advertise support for dedicated HP event source to guests */
2750     if (spapr->use_hotplug_event_source) {
2751         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2752     }
2753 
2754     /* advertise support for HPT resizing */
2755     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2756         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2757     }
2758 
2759     /* advertise support for ibm,dyamic-memory-v2 */
2760     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2761 
2762     /* advertise XIVE on POWER9 machines */
2763     if (spapr->irq->xive) {
2764         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2765     }
2766 
2767     /* init CPUs */
2768     spapr_init_cpus(spapr);
2769 
2770     /*
2771      * check we don't have a memory-less/cpu-less NUMA node
2772      * Firmware relies on the existing memory/cpu topology to provide the
2773      * NUMA topology to the kernel.
2774      * And the linux kernel needs to know the NUMA topology at start
2775      * to be able to hotplug CPUs later.
2776      */
2777     if (machine->numa_state->num_nodes) {
2778         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2779             /* check for memory-less node */
2780             if (machine->numa_state->nodes[i].node_mem == 0) {
2781                 CPUState *cs;
2782                 int found = 0;
2783                 /* check for cpu-less node */
2784                 CPU_FOREACH(cs) {
2785                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2786                     if (cpu->node_id == i) {
2787                         found = 1;
2788                         break;
2789                     }
2790                 }
2791                 /* memory-less and cpu-less node */
2792                 if (!found) {
2793                     error_report(
2794                        "Memory-less/cpu-less nodes are not supported (node %d)",
2795                                  i);
2796                     exit(1);
2797                 }
2798             }
2799         }
2800 
2801     }
2802 
2803     /*
2804      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2805      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2806      * called from vPHB reset handler so we initialize the counter here.
2807      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2808      * must be equally distant from any other node.
2809      * The final value of spapr->gpu_numa_id is going to be written to
2810      * max-associativity-domains in spapr_build_fdt().
2811      */
2812     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2813 
2814     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2815         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2816                               spapr->max_compat_pvr)) {
2817         /* KVM and TCG always allow GTSE with radix... */
2818         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2819     }
2820     /* ... but not with hash (currently). */
2821 
2822     if (kvm_enabled()) {
2823         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2824         kvmppc_enable_logical_ci_hcalls();
2825         kvmppc_enable_set_mode_hcall();
2826 
2827         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2828         kvmppc_enable_clear_ref_mod_hcalls();
2829 
2830         /* Enable H_PAGE_INIT */
2831         kvmppc_enable_h_page_init();
2832     }
2833 
2834     /* map RAM */
2835     memory_region_add_subregion(sysmem, 0, machine->ram);
2836 
2837     /* always allocate the device memory information */
2838     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2839 
2840     /* initialize hotplug memory address space */
2841     if (machine->ram_size < machine->maxram_size) {
2842         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2843         /*
2844          * Limit the number of hotpluggable memory slots to half the number
2845          * slots that KVM supports, leaving the other half for PCI and other
2846          * devices. However ensure that number of slots doesn't drop below 32.
2847          */
2848         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2849                            SPAPR_MAX_RAM_SLOTS;
2850 
2851         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2852             max_memslots = SPAPR_MAX_RAM_SLOTS;
2853         }
2854         if (machine->ram_slots > max_memslots) {
2855             error_report("Specified number of memory slots %"
2856                          PRIu64" exceeds max supported %d",
2857                          machine->ram_slots, max_memslots);
2858             exit(1);
2859         }
2860 
2861         machine->device_memory->base = ROUND_UP(machine->ram_size,
2862                                                 SPAPR_DEVICE_MEM_ALIGN);
2863         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2864                            "device-memory", device_mem_size);
2865         memory_region_add_subregion(sysmem, machine->device_memory->base,
2866                                     &machine->device_memory->mr);
2867     }
2868 
2869     if (smc->dr_lmb_enabled) {
2870         spapr_create_lmb_dr_connectors(spapr);
2871     }
2872 
2873     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI_MCE) == SPAPR_CAP_ON) {
2874         /* Create the error string for live migration blocker */
2875         error_setg(&spapr->fwnmi_migration_blocker,
2876             "A machine check is being handled during migration. The handler"
2877             "may run and log hardware error on the destination");
2878     }
2879 
2880     if (mc->nvdimm_supported) {
2881         spapr_create_nvdimm_dr_connectors(spapr);
2882     }
2883 
2884     /* Set up RTAS event infrastructure */
2885     spapr_events_init(spapr);
2886 
2887     /* Set up the RTC RTAS interfaces */
2888     spapr_rtc_create(spapr);
2889 
2890     /* Set up VIO bus */
2891     spapr->vio_bus = spapr_vio_bus_init();
2892 
2893     for (i = 0; i < serial_max_hds(); i++) {
2894         if (serial_hd(i)) {
2895             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2896         }
2897     }
2898 
2899     /* We always have at least the nvram device on VIO */
2900     spapr_create_nvram(spapr);
2901 
2902     /*
2903      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2904      * connectors (described in root DT node's "ibm,drc-types" property)
2905      * are pre-initialized here. additional child connectors (such as
2906      * connectors for a PHBs PCI slots) are added as needed during their
2907      * parent's realization.
2908      */
2909     if (smc->dr_phb_enabled) {
2910         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2911             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2912         }
2913     }
2914 
2915     /* Set up PCI */
2916     spapr_pci_rtas_init();
2917 
2918     phb = spapr_create_default_phb();
2919 
2920     for (i = 0; i < nb_nics; i++) {
2921         NICInfo *nd = &nd_table[i];
2922 
2923         if (!nd->model) {
2924             nd->model = g_strdup("spapr-vlan");
2925         }
2926 
2927         if (g_str_equal(nd->model, "spapr-vlan") ||
2928             g_str_equal(nd->model, "ibmveth")) {
2929             spapr_vlan_create(spapr->vio_bus, nd);
2930         } else {
2931             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2932         }
2933     }
2934 
2935     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2936         spapr_vscsi_create(spapr->vio_bus);
2937     }
2938 
2939     /* Graphics */
2940     if (spapr_vga_init(phb->bus, &error_fatal)) {
2941         spapr->has_graphics = true;
2942         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2943     }
2944 
2945     if (machine->usb) {
2946         if (smc->use_ohci_by_default) {
2947             pci_create_simple(phb->bus, -1, "pci-ohci");
2948         } else {
2949             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2950         }
2951 
2952         if (spapr->has_graphics) {
2953             USBBus *usb_bus = usb_bus_find(-1);
2954 
2955             usb_create_simple(usb_bus, "usb-kbd");
2956             usb_create_simple(usb_bus, "usb-mouse");
2957         }
2958     }
2959 
2960     if (spapr->rma_size < MIN_RMA_SLOF) {
2961         error_report(
2962             "pSeries SLOF firmware requires >= %ldMiB guest RMA (Real Mode Area memory)",
2963             MIN_RMA_SLOF / MiB);
2964         exit(1);
2965     }
2966 
2967     if (kernel_filename) {
2968         uint64_t lowaddr = 0;
2969 
2970         spapr->kernel_size = load_elf(kernel_filename, NULL,
2971                                       translate_kernel_address, spapr,
2972                                       NULL, &lowaddr, NULL, NULL, 1,
2973                                       PPC_ELF_MACHINE, 0, 0);
2974         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2975             spapr->kernel_size = load_elf(kernel_filename, NULL,
2976                                           translate_kernel_address, spapr, NULL,
2977                                           &lowaddr, NULL, NULL, 0,
2978                                           PPC_ELF_MACHINE,
2979                                           0, 0);
2980             spapr->kernel_le = spapr->kernel_size > 0;
2981         }
2982         if (spapr->kernel_size < 0) {
2983             error_report("error loading %s: %s", kernel_filename,
2984                          load_elf_strerror(spapr->kernel_size));
2985             exit(1);
2986         }
2987 
2988         /* load initrd */
2989         if (initrd_filename) {
2990             /* Try to locate the initrd in the gap between the kernel
2991              * and the firmware. Add a bit of space just in case
2992              */
2993             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2994                                   + 0x1ffff) & ~0xffff;
2995             spapr->initrd_size = load_image_targphys(initrd_filename,
2996                                                      spapr->initrd_base,
2997                                                      load_limit
2998                                                      - spapr->initrd_base);
2999             if (spapr->initrd_size < 0) {
3000                 error_report("could not load initial ram disk '%s'",
3001                              initrd_filename);
3002                 exit(1);
3003             }
3004         }
3005     }
3006 
3007     if (bios_name == NULL) {
3008         bios_name = FW_FILE_NAME;
3009     }
3010     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3011     if (!filename) {
3012         error_report("Could not find LPAR firmware '%s'", bios_name);
3013         exit(1);
3014     }
3015     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3016     if (fw_size <= 0) {
3017         error_report("Could not load LPAR firmware '%s'", filename);
3018         exit(1);
3019     }
3020     g_free(filename);
3021 
3022     /* FIXME: Should register things through the MachineState's qdev
3023      * interface, this is a legacy from the sPAPREnvironment structure
3024      * which predated MachineState but had a similar function */
3025     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3026     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3027                          &savevm_htab_handlers, spapr);
3028 
3029     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3030                              &error_fatal);
3031 
3032     qemu_register_boot_set(spapr_boot_set, spapr);
3033 
3034     /*
3035      * Nothing needs to be done to resume a suspended guest because
3036      * suspending does not change the machine state, so no need for
3037      * a ->wakeup method.
3038      */
3039     qemu_register_wakeup_support();
3040 
3041     if (kvm_enabled()) {
3042         /* to stop and start vmclock */
3043         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3044                                          &spapr->tb);
3045 
3046         kvmppc_spapr_enable_inkernel_multitce();
3047     }
3048 
3049     qemu_cond_init(&spapr->mc_delivery_cond);
3050 }
3051 
3052 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3053 {
3054     if (!vm_type) {
3055         return 0;
3056     }
3057 
3058     if (!strcmp(vm_type, "HV")) {
3059         return 1;
3060     }
3061 
3062     if (!strcmp(vm_type, "PR")) {
3063         return 2;
3064     }
3065 
3066     error_report("Unknown kvm-type specified '%s'", vm_type);
3067     exit(1);
3068 }
3069 
3070 /*
3071  * Implementation of an interface to adjust firmware path
3072  * for the bootindex property handling.
3073  */
3074 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3075                                    DeviceState *dev)
3076 {
3077 #define CAST(type, obj, name) \
3078     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3079     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3080     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3081     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3082 
3083     if (d) {
3084         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3085         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3086         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3087 
3088         if (spapr) {
3089             /*
3090              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3091              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3092              * 0x8000 | (target << 8) | (bus << 5) | lun
3093              * (see the "Logical unit addressing format" table in SAM5)
3094              */
3095             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3096             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3097                                    (uint64_t)id << 48);
3098         } else if (virtio) {
3099             /*
3100              * We use SRP luns of the form 01000000 | (target << 8) | lun
3101              * in the top 32 bits of the 64-bit LUN
3102              * Note: the quote above is from SLOF and it is wrong,
3103              * the actual binding is:
3104              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3105              */
3106             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3107             if (d->lun >= 256) {
3108                 /* Use the LUN "flat space addressing method" */
3109                 id |= 0x4000;
3110             }
3111             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3112                                    (uint64_t)id << 32);
3113         } else if (usb) {
3114             /*
3115              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3116              * in the top 32 bits of the 64-bit LUN
3117              */
3118             unsigned usb_port = atoi(usb->port->path);
3119             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3120             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3121                                    (uint64_t)id << 32);
3122         }
3123     }
3124 
3125     /*
3126      * SLOF probes the USB devices, and if it recognizes that the device is a
3127      * storage device, it changes its name to "storage" instead of "usb-host",
3128      * and additionally adds a child node for the SCSI LUN, so the correct
3129      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3130      */
3131     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3132         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3133         if (usb_host_dev_is_scsi_storage(usbdev)) {
3134             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3135         }
3136     }
3137 
3138     if (phb) {
3139         /* Replace "pci" with "pci@800000020000000" */
3140         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3141     }
3142 
3143     if (vsc) {
3144         /* Same logic as virtio above */
3145         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3146         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3147     }
3148 
3149     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3150         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3151         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3152         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3153     }
3154 
3155     return NULL;
3156 }
3157 
3158 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3159 {
3160     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3161 
3162     return g_strdup(spapr->kvm_type);
3163 }
3164 
3165 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3166 {
3167     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3168 
3169     g_free(spapr->kvm_type);
3170     spapr->kvm_type = g_strdup(value);
3171 }
3172 
3173 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3174 {
3175     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3176 
3177     return spapr->use_hotplug_event_source;
3178 }
3179 
3180 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3181                                             Error **errp)
3182 {
3183     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3184 
3185     spapr->use_hotplug_event_source = value;
3186 }
3187 
3188 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3189 {
3190     return true;
3191 }
3192 
3193 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3194 {
3195     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3196 
3197     switch (spapr->resize_hpt) {
3198     case SPAPR_RESIZE_HPT_DEFAULT:
3199         return g_strdup("default");
3200     case SPAPR_RESIZE_HPT_DISABLED:
3201         return g_strdup("disabled");
3202     case SPAPR_RESIZE_HPT_ENABLED:
3203         return g_strdup("enabled");
3204     case SPAPR_RESIZE_HPT_REQUIRED:
3205         return g_strdup("required");
3206     }
3207     g_assert_not_reached();
3208 }
3209 
3210 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3211 {
3212     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3213 
3214     if (strcmp(value, "default") == 0) {
3215         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3216     } else if (strcmp(value, "disabled") == 0) {
3217         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3218     } else if (strcmp(value, "enabled") == 0) {
3219         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3220     } else if (strcmp(value, "required") == 0) {
3221         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3222     } else {
3223         error_setg(errp, "Bad value for \"resize-hpt\" property");
3224     }
3225 }
3226 
3227 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3228                                    void *opaque, Error **errp)
3229 {
3230     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3231 }
3232 
3233 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3234                                    void *opaque, Error **errp)
3235 {
3236     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3237 }
3238 
3239 static void spapr_get_kernel_addr(Object *obj, Visitor *v, const char *name,
3240                                   void *opaque, Error **errp)
3241 {
3242     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3243 }
3244 
3245 static void spapr_set_kernel_addr(Object *obj, Visitor *v, const char *name,
3246                                   void *opaque, Error **errp)
3247 {
3248     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3249 }
3250 
3251 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3252 {
3253     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3254 
3255     if (spapr->irq == &spapr_irq_xics_legacy) {
3256         return g_strdup("legacy");
3257     } else if (spapr->irq == &spapr_irq_xics) {
3258         return g_strdup("xics");
3259     } else if (spapr->irq == &spapr_irq_xive) {
3260         return g_strdup("xive");
3261     } else if (spapr->irq == &spapr_irq_dual) {
3262         return g_strdup("dual");
3263     }
3264     g_assert_not_reached();
3265 }
3266 
3267 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3268 {
3269     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3270 
3271     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3272         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3273         return;
3274     }
3275 
3276     /* The legacy IRQ backend can not be set */
3277     if (strcmp(value, "xics") == 0) {
3278         spapr->irq = &spapr_irq_xics;
3279     } else if (strcmp(value, "xive") == 0) {
3280         spapr->irq = &spapr_irq_xive;
3281     } else if (strcmp(value, "dual") == 0) {
3282         spapr->irq = &spapr_irq_dual;
3283     } else {
3284         error_setg(errp, "Bad value for \"ic-mode\" property");
3285     }
3286 }
3287 
3288 static char *spapr_get_host_model(Object *obj, Error **errp)
3289 {
3290     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3291 
3292     return g_strdup(spapr->host_model);
3293 }
3294 
3295 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3296 {
3297     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3298 
3299     g_free(spapr->host_model);
3300     spapr->host_model = g_strdup(value);
3301 }
3302 
3303 static char *spapr_get_host_serial(Object *obj, Error **errp)
3304 {
3305     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3306 
3307     return g_strdup(spapr->host_serial);
3308 }
3309 
3310 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3311 {
3312     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3313 
3314     g_free(spapr->host_serial);
3315     spapr->host_serial = g_strdup(value);
3316 }
3317 
3318 static void spapr_instance_init(Object *obj)
3319 {
3320     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3321     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3322 
3323     spapr->htab_fd = -1;
3324     spapr->use_hotplug_event_source = true;
3325     object_property_add_str(obj, "kvm-type",
3326                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3327     object_property_set_description(obj, "kvm-type",
3328                                     "Specifies the KVM virtualization mode (HV, PR)",
3329                                     NULL);
3330     object_property_add_bool(obj, "modern-hotplug-events",
3331                             spapr_get_modern_hotplug_events,
3332                             spapr_set_modern_hotplug_events,
3333                             NULL);
3334     object_property_set_description(obj, "modern-hotplug-events",
3335                                     "Use dedicated hotplug event mechanism in"
3336                                     " place of standard EPOW events when possible"
3337                                     " (required for memory hot-unplug support)",
3338                                     NULL);
3339     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3340                             "Maximum permitted CPU compatibility mode",
3341                             &error_fatal);
3342 
3343     object_property_add_str(obj, "resize-hpt",
3344                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3345     object_property_set_description(obj, "resize-hpt",
3346                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3347                                     NULL);
3348     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3349                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3350     object_property_set_description(obj, "vsmt",
3351                                     "Virtual SMT: KVM behaves as if this were"
3352                                     " the host's SMT mode", &error_abort);
3353     object_property_add_bool(obj, "vfio-no-msix-emulation",
3354                              spapr_get_msix_emulation, NULL, NULL);
3355 
3356     object_property_add(obj, "kernel-addr", "uint64", spapr_get_kernel_addr,
3357                         spapr_set_kernel_addr, NULL, &spapr->kernel_addr,
3358                         &error_abort);
3359     object_property_set_description(obj, "kernel-addr",
3360                                     stringify(KERNEL_LOAD_ADDR)
3361                                     " for -kernel is the default",
3362                                     NULL);
3363     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3364     /* The machine class defines the default interrupt controller mode */
3365     spapr->irq = smc->irq;
3366     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3367                             spapr_set_ic_mode, NULL);
3368     object_property_set_description(obj, "ic-mode",
3369                  "Specifies the interrupt controller mode (xics, xive, dual)",
3370                  NULL);
3371 
3372     object_property_add_str(obj, "host-model",
3373         spapr_get_host_model, spapr_set_host_model,
3374         &error_abort);
3375     object_property_set_description(obj, "host-model",
3376         "Host model to advertise in guest device tree", &error_abort);
3377     object_property_add_str(obj, "host-serial",
3378         spapr_get_host_serial, spapr_set_host_serial,
3379         &error_abort);
3380     object_property_set_description(obj, "host-serial",
3381         "Host serial number to advertise in guest device tree", &error_abort);
3382 }
3383 
3384 static void spapr_machine_finalizefn(Object *obj)
3385 {
3386     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3387 
3388     g_free(spapr->kvm_type);
3389 }
3390 
3391 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3392 {
3393     cpu_synchronize_state(cs);
3394     ppc_cpu_do_system_reset(cs);
3395 }
3396 
3397 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3398 {
3399     CPUState *cs;
3400 
3401     CPU_FOREACH(cs) {
3402         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3403     }
3404 }
3405 
3406 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3407                           void *fdt, int *fdt_start_offset, Error **errp)
3408 {
3409     uint64_t addr;
3410     uint32_t node;
3411 
3412     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3413     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3414                                     &error_abort);
3415     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3416                                                    SPAPR_MEMORY_BLOCK_SIZE);
3417     return 0;
3418 }
3419 
3420 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3421                            bool dedicated_hp_event_source, Error **errp)
3422 {
3423     SpaprDrc *drc;
3424     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3425     int i;
3426     uint64_t addr = addr_start;
3427     bool hotplugged = spapr_drc_hotplugged(dev);
3428     Error *local_err = NULL;
3429 
3430     for (i = 0; i < nr_lmbs; i++) {
3431         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3432                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3433         g_assert(drc);
3434 
3435         spapr_drc_attach(drc, dev, &local_err);
3436         if (local_err) {
3437             while (addr > addr_start) {
3438                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3439                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3440                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3441                 spapr_drc_detach(drc);
3442             }
3443             error_propagate(errp, local_err);
3444             return;
3445         }
3446         if (!hotplugged) {
3447             spapr_drc_reset(drc);
3448         }
3449         addr += SPAPR_MEMORY_BLOCK_SIZE;
3450     }
3451     /* send hotplug notification to the
3452      * guest only in case of hotplugged memory
3453      */
3454     if (hotplugged) {
3455         if (dedicated_hp_event_source) {
3456             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3457                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3458             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3459                                                    nr_lmbs,
3460                                                    spapr_drc_index(drc));
3461         } else {
3462             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3463                                            nr_lmbs);
3464         }
3465     }
3466 }
3467 
3468 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3469                               Error **errp)
3470 {
3471     Error *local_err = NULL;
3472     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3473     PCDIMMDevice *dimm = PC_DIMM(dev);
3474     uint64_t size, addr, slot;
3475     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3476 
3477     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3478 
3479     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3480     if (local_err) {
3481         goto out;
3482     }
3483 
3484     if (!is_nvdimm) {
3485         addr = object_property_get_uint(OBJECT(dimm),
3486                                         PC_DIMM_ADDR_PROP, &local_err);
3487         if (local_err) {
3488             goto out_unplug;
3489         }
3490         spapr_add_lmbs(dev, addr, size,
3491                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3492                        &local_err);
3493     } else {
3494         slot = object_property_get_uint(OBJECT(dimm),
3495                                         PC_DIMM_SLOT_PROP, &local_err);
3496         if (local_err) {
3497             goto out_unplug;
3498         }
3499         spapr_add_nvdimm(dev, slot, &local_err);
3500     }
3501 
3502     if (local_err) {
3503         goto out_unplug;
3504     }
3505 
3506     return;
3507 
3508 out_unplug:
3509     pc_dimm_unplug(dimm, MACHINE(ms));
3510 out:
3511     error_propagate(errp, local_err);
3512 }
3513 
3514 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3515                                   Error **errp)
3516 {
3517     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3518     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3519     const MachineClass *mc = MACHINE_CLASS(smc);
3520     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3521     PCDIMMDevice *dimm = PC_DIMM(dev);
3522     Error *local_err = NULL;
3523     uint64_t size;
3524     Object *memdev;
3525     hwaddr pagesize;
3526 
3527     if (!smc->dr_lmb_enabled) {
3528         error_setg(errp, "Memory hotplug not supported for this machine");
3529         return;
3530     }
3531 
3532     if (is_nvdimm && !mc->nvdimm_supported) {
3533         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3534         return;
3535     }
3536 
3537     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3538     if (local_err) {
3539         error_propagate(errp, local_err);
3540         return;
3541     }
3542 
3543     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3544         error_setg(errp, "Hotplugged memory size must be a multiple of "
3545                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3546         return;
3547     } else if (is_nvdimm) {
3548         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3549         if (local_err) {
3550             error_propagate(errp, local_err);
3551             return;
3552         }
3553     }
3554 
3555     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3556                                       &error_abort);
3557     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3558     spapr_check_pagesize(spapr, pagesize, &local_err);
3559     if (local_err) {
3560         error_propagate(errp, local_err);
3561         return;
3562     }
3563 
3564     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3565 }
3566 
3567 struct SpaprDimmState {
3568     PCDIMMDevice *dimm;
3569     uint32_t nr_lmbs;
3570     QTAILQ_ENTRY(SpaprDimmState) next;
3571 };
3572 
3573 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3574                                                        PCDIMMDevice *dimm)
3575 {
3576     SpaprDimmState *dimm_state = NULL;
3577 
3578     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3579         if (dimm_state->dimm == dimm) {
3580             break;
3581         }
3582     }
3583     return dimm_state;
3584 }
3585 
3586 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3587                                                       uint32_t nr_lmbs,
3588                                                       PCDIMMDevice *dimm)
3589 {
3590     SpaprDimmState *ds = NULL;
3591 
3592     /*
3593      * If this request is for a DIMM whose removal had failed earlier
3594      * (due to guest's refusal to remove the LMBs), we would have this
3595      * dimm already in the pending_dimm_unplugs list. In that
3596      * case don't add again.
3597      */
3598     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3599     if (!ds) {
3600         ds = g_malloc0(sizeof(SpaprDimmState));
3601         ds->nr_lmbs = nr_lmbs;
3602         ds->dimm = dimm;
3603         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3604     }
3605     return ds;
3606 }
3607 
3608 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3609                                               SpaprDimmState *dimm_state)
3610 {
3611     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3612     g_free(dimm_state);
3613 }
3614 
3615 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3616                                                         PCDIMMDevice *dimm)
3617 {
3618     SpaprDrc *drc;
3619     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3620                                                   &error_abort);
3621     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3622     uint32_t avail_lmbs = 0;
3623     uint64_t addr_start, addr;
3624     int i;
3625 
3626     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3627                                          &error_abort);
3628 
3629     addr = addr_start;
3630     for (i = 0; i < nr_lmbs; i++) {
3631         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3632                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3633         g_assert(drc);
3634         if (drc->dev) {
3635             avail_lmbs++;
3636         }
3637         addr += SPAPR_MEMORY_BLOCK_SIZE;
3638     }
3639 
3640     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3641 }
3642 
3643 /* Callback to be called during DRC release. */
3644 void spapr_lmb_release(DeviceState *dev)
3645 {
3646     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3647     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3648     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3649 
3650     /* This information will get lost if a migration occurs
3651      * during the unplug process. In this case recover it. */
3652     if (ds == NULL) {
3653         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3654         g_assert(ds);
3655         /* The DRC being examined by the caller at least must be counted */
3656         g_assert(ds->nr_lmbs);
3657     }
3658 
3659     if (--ds->nr_lmbs) {
3660         return;
3661     }
3662 
3663     /*
3664      * Now that all the LMBs have been removed by the guest, call the
3665      * unplug handler chain. This can never fail.
3666      */
3667     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3668     object_unparent(OBJECT(dev));
3669 }
3670 
3671 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3672 {
3673     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3674     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3675 
3676     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3677     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3678     spapr_pending_dimm_unplugs_remove(spapr, ds);
3679 }
3680 
3681 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3682                                         DeviceState *dev, Error **errp)
3683 {
3684     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3685     Error *local_err = NULL;
3686     PCDIMMDevice *dimm = PC_DIMM(dev);
3687     uint32_t nr_lmbs;
3688     uint64_t size, addr_start, addr;
3689     int i;
3690     SpaprDrc *drc;
3691 
3692     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3693         error_setg(&local_err,
3694                    "nvdimm device hot unplug is not supported yet.");
3695         goto out;
3696     }
3697 
3698     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3699     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3700 
3701     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3702                                          &local_err);
3703     if (local_err) {
3704         goto out;
3705     }
3706 
3707     /*
3708      * An existing pending dimm state for this DIMM means that there is an
3709      * unplug operation in progress, waiting for the spapr_lmb_release
3710      * callback to complete the job (BQL can't cover that far). In this case,
3711      * bail out to avoid detaching DRCs that were already released.
3712      */
3713     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3714         error_setg(&local_err,
3715                    "Memory unplug already in progress for device %s",
3716                    dev->id);
3717         goto out;
3718     }
3719 
3720     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3721 
3722     addr = addr_start;
3723     for (i = 0; i < nr_lmbs; i++) {
3724         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3725                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3726         g_assert(drc);
3727 
3728         spapr_drc_detach(drc);
3729         addr += SPAPR_MEMORY_BLOCK_SIZE;
3730     }
3731 
3732     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3733                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3734     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3735                                               nr_lmbs, spapr_drc_index(drc));
3736 out:
3737     error_propagate(errp, local_err);
3738 }
3739 
3740 /* Callback to be called during DRC release. */
3741 void spapr_core_release(DeviceState *dev)
3742 {
3743     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3744 
3745     /* Call the unplug handler chain. This can never fail. */
3746     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3747     object_unparent(OBJECT(dev));
3748 }
3749 
3750 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3751 {
3752     MachineState *ms = MACHINE(hotplug_dev);
3753     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3754     CPUCore *cc = CPU_CORE(dev);
3755     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3756 
3757     if (smc->pre_2_10_has_unused_icps) {
3758         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3759         int i;
3760 
3761         for (i = 0; i < cc->nr_threads; i++) {
3762             CPUState *cs = CPU(sc->threads[i]);
3763 
3764             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3765         }
3766     }
3767 
3768     assert(core_slot);
3769     core_slot->cpu = NULL;
3770     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3771 }
3772 
3773 static
3774 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3775                                Error **errp)
3776 {
3777     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3778     int index;
3779     SpaprDrc *drc;
3780     CPUCore *cc = CPU_CORE(dev);
3781 
3782     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3783         error_setg(errp, "Unable to find CPU core with core-id: %d",
3784                    cc->core_id);
3785         return;
3786     }
3787     if (index == 0) {
3788         error_setg(errp, "Boot CPU core may not be unplugged");
3789         return;
3790     }
3791 
3792     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3793                           spapr_vcpu_id(spapr, cc->core_id));
3794     g_assert(drc);
3795 
3796     if (!spapr_drc_unplug_requested(drc)) {
3797         spapr_drc_detach(drc);
3798         spapr_hotplug_req_remove_by_index(drc);
3799     }
3800 }
3801 
3802 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3803                            void *fdt, int *fdt_start_offset, Error **errp)
3804 {
3805     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3806     CPUState *cs = CPU(core->threads[0]);
3807     PowerPCCPU *cpu = POWERPC_CPU(cs);
3808     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3809     int id = spapr_get_vcpu_id(cpu);
3810     char *nodename;
3811     int offset;
3812 
3813     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3814     offset = fdt_add_subnode(fdt, 0, nodename);
3815     g_free(nodename);
3816 
3817     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3818 
3819     *fdt_start_offset = offset;
3820     return 0;
3821 }
3822 
3823 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3824                             Error **errp)
3825 {
3826     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3827     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3828     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3829     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3830     CPUCore *cc = CPU_CORE(dev);
3831     CPUState *cs;
3832     SpaprDrc *drc;
3833     Error *local_err = NULL;
3834     CPUArchId *core_slot;
3835     int index;
3836     bool hotplugged = spapr_drc_hotplugged(dev);
3837     int i;
3838 
3839     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3840     if (!core_slot) {
3841         error_setg(errp, "Unable to find CPU core with core-id: %d",
3842                    cc->core_id);
3843         return;
3844     }
3845     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3846                           spapr_vcpu_id(spapr, cc->core_id));
3847 
3848     g_assert(drc || !mc->has_hotpluggable_cpus);
3849 
3850     if (drc) {
3851         spapr_drc_attach(drc, dev, &local_err);
3852         if (local_err) {
3853             error_propagate(errp, local_err);
3854             return;
3855         }
3856 
3857         if (hotplugged) {
3858             /*
3859              * Send hotplug notification interrupt to the guest only
3860              * in case of hotplugged CPUs.
3861              */
3862             spapr_hotplug_req_add_by_index(drc);
3863         } else {
3864             spapr_drc_reset(drc);
3865         }
3866     }
3867 
3868     core_slot->cpu = OBJECT(dev);
3869 
3870     if (smc->pre_2_10_has_unused_icps) {
3871         for (i = 0; i < cc->nr_threads; i++) {
3872             cs = CPU(core->threads[i]);
3873             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3874         }
3875     }
3876 
3877     /*
3878      * Set compatibility mode to match the boot CPU, which was either set
3879      * by the machine reset code or by CAS.
3880      */
3881     if (hotplugged) {
3882         for (i = 0; i < cc->nr_threads; i++) {
3883             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3884                            &local_err);
3885             if (local_err) {
3886                 error_propagate(errp, local_err);
3887                 return;
3888             }
3889         }
3890     }
3891 }
3892 
3893 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3894                                 Error **errp)
3895 {
3896     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3897     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3898     Error *local_err = NULL;
3899     CPUCore *cc = CPU_CORE(dev);
3900     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3901     const char *type = object_get_typename(OBJECT(dev));
3902     CPUArchId *core_slot;
3903     int index;
3904     unsigned int smp_threads = machine->smp.threads;
3905 
3906     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3907         error_setg(&local_err, "CPU hotplug not supported for this machine");
3908         goto out;
3909     }
3910 
3911     if (strcmp(base_core_type, type)) {
3912         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3913         goto out;
3914     }
3915 
3916     if (cc->core_id % smp_threads) {
3917         error_setg(&local_err, "invalid core id %d", cc->core_id);
3918         goto out;
3919     }
3920 
3921     /*
3922      * In general we should have homogeneous threads-per-core, but old
3923      * (pre hotplug support) machine types allow the last core to have
3924      * reduced threads as a compatibility hack for when we allowed
3925      * total vcpus not a multiple of threads-per-core.
3926      */
3927     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3928         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3929                    cc->nr_threads, smp_threads);
3930         goto out;
3931     }
3932 
3933     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3934     if (!core_slot) {
3935         error_setg(&local_err, "core id %d out of range", cc->core_id);
3936         goto out;
3937     }
3938 
3939     if (core_slot->cpu) {
3940         error_setg(&local_err, "core %d already populated", cc->core_id);
3941         goto out;
3942     }
3943 
3944     numa_cpu_pre_plug(core_slot, dev, &local_err);
3945 
3946 out:
3947     error_propagate(errp, local_err);
3948 }
3949 
3950 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3951                           void *fdt, int *fdt_start_offset, Error **errp)
3952 {
3953     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3954     int intc_phandle;
3955 
3956     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3957     if (intc_phandle <= 0) {
3958         return -1;
3959     }
3960 
3961     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3962         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3963         return -1;
3964     }
3965 
3966     /* generally SLOF creates these, for hotplug it's up to QEMU */
3967     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3968 
3969     return 0;
3970 }
3971 
3972 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3973                                Error **errp)
3974 {
3975     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3976     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3977     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3978     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3979 
3980     if (dev->hotplugged && !smc->dr_phb_enabled) {
3981         error_setg(errp, "PHB hotplug not supported for this machine");
3982         return;
3983     }
3984 
3985     if (sphb->index == (uint32_t)-1) {
3986         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3987         return;
3988     }
3989 
3990     /*
3991      * This will check that sphb->index doesn't exceed the maximum number of
3992      * PHBs for the current machine type.
3993      */
3994     smc->phb_placement(spapr, sphb->index,
3995                        &sphb->buid, &sphb->io_win_addr,
3996                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3997                        windows_supported, sphb->dma_liobn,
3998                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3999                        errp);
4000 }
4001 
4002 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4003                            Error **errp)
4004 {
4005     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4006     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
4007     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4008     SpaprDrc *drc;
4009     bool hotplugged = spapr_drc_hotplugged(dev);
4010     Error *local_err = NULL;
4011 
4012     if (!smc->dr_phb_enabled) {
4013         return;
4014     }
4015 
4016     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4017     /* hotplug hooks should check it's enabled before getting this far */
4018     assert(drc);
4019 
4020     spapr_drc_attach(drc, DEVICE(dev), &local_err);
4021     if (local_err) {
4022         error_propagate(errp, local_err);
4023         return;
4024     }
4025 
4026     if (hotplugged) {
4027         spapr_hotplug_req_add_by_index(drc);
4028     } else {
4029         spapr_drc_reset(drc);
4030     }
4031 }
4032 
4033 void spapr_phb_release(DeviceState *dev)
4034 {
4035     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4036 
4037     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4038     object_unparent(OBJECT(dev));
4039 }
4040 
4041 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4042 {
4043     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4044 }
4045 
4046 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4047                                      DeviceState *dev, Error **errp)
4048 {
4049     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4050     SpaprDrc *drc;
4051 
4052     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4053     assert(drc);
4054 
4055     if (!spapr_drc_unplug_requested(drc)) {
4056         spapr_drc_detach(drc);
4057         spapr_hotplug_req_remove_by_index(drc);
4058     }
4059 }
4060 
4061 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4062                                  Error **errp)
4063 {
4064     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4065     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4066 
4067     if (spapr->tpm_proxy != NULL) {
4068         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4069         return;
4070     }
4071 
4072     spapr->tpm_proxy = tpm_proxy;
4073 }
4074 
4075 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4076 {
4077     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4078 
4079     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4080     object_unparent(OBJECT(dev));
4081     spapr->tpm_proxy = NULL;
4082 }
4083 
4084 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4085                                       DeviceState *dev, Error **errp)
4086 {
4087     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4088         spapr_memory_plug(hotplug_dev, dev, errp);
4089     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4090         spapr_core_plug(hotplug_dev, dev, errp);
4091     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4092         spapr_phb_plug(hotplug_dev, dev, errp);
4093     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4094         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4095     }
4096 }
4097 
4098 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4099                                         DeviceState *dev, Error **errp)
4100 {
4101     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4102         spapr_memory_unplug(hotplug_dev, dev);
4103     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4104         spapr_core_unplug(hotplug_dev, dev);
4105     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4106         spapr_phb_unplug(hotplug_dev, dev);
4107     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4108         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4109     }
4110 }
4111 
4112 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4113                                                 DeviceState *dev, Error **errp)
4114 {
4115     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4116     MachineClass *mc = MACHINE_GET_CLASS(sms);
4117     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4118 
4119     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4120         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4121             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4122         } else {
4123             /* NOTE: this means there is a window after guest reset, prior to
4124              * CAS negotiation, where unplug requests will fail due to the
4125              * capability not being detected yet. This is a bit different than
4126              * the case with PCI unplug, where the events will be queued and
4127              * eventually handled by the guest after boot
4128              */
4129             error_setg(errp, "Memory hot unplug not supported for this guest");
4130         }
4131     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4132         if (!mc->has_hotpluggable_cpus) {
4133             error_setg(errp, "CPU hot unplug not supported on this machine");
4134             return;
4135         }
4136         spapr_core_unplug_request(hotplug_dev, dev, errp);
4137     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4138         if (!smc->dr_phb_enabled) {
4139             error_setg(errp, "PHB hot unplug not supported on this machine");
4140             return;
4141         }
4142         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4143     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4144         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4145     }
4146 }
4147 
4148 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4149                                           DeviceState *dev, Error **errp)
4150 {
4151     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4152         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4153     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4154         spapr_core_pre_plug(hotplug_dev, dev, errp);
4155     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4156         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4157     }
4158 }
4159 
4160 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4161                                                  DeviceState *dev)
4162 {
4163     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4164         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4165         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4166         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4167         return HOTPLUG_HANDLER(machine);
4168     }
4169     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4170         PCIDevice *pcidev = PCI_DEVICE(dev);
4171         PCIBus *root = pci_device_root_bus(pcidev);
4172         SpaprPhbState *phb =
4173             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4174                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4175 
4176         if (phb) {
4177             return HOTPLUG_HANDLER(phb);
4178         }
4179     }
4180     return NULL;
4181 }
4182 
4183 static CpuInstanceProperties
4184 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4185 {
4186     CPUArchId *core_slot;
4187     MachineClass *mc = MACHINE_GET_CLASS(machine);
4188 
4189     /* make sure possible_cpu are intialized */
4190     mc->possible_cpu_arch_ids(machine);
4191     /* get CPU core slot containing thread that matches cpu_index */
4192     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4193     assert(core_slot);
4194     return core_slot->props;
4195 }
4196 
4197 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4198 {
4199     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4200 }
4201 
4202 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4203 {
4204     int i;
4205     unsigned int smp_threads = machine->smp.threads;
4206     unsigned int smp_cpus = machine->smp.cpus;
4207     const char *core_type;
4208     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4209     MachineClass *mc = MACHINE_GET_CLASS(machine);
4210 
4211     if (!mc->has_hotpluggable_cpus) {
4212         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4213     }
4214     if (machine->possible_cpus) {
4215         assert(machine->possible_cpus->len == spapr_max_cores);
4216         return machine->possible_cpus;
4217     }
4218 
4219     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4220     if (!core_type) {
4221         error_report("Unable to find sPAPR CPU Core definition");
4222         exit(1);
4223     }
4224 
4225     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4226                              sizeof(CPUArchId) * spapr_max_cores);
4227     machine->possible_cpus->len = spapr_max_cores;
4228     for (i = 0; i < machine->possible_cpus->len; i++) {
4229         int core_id = i * smp_threads;
4230 
4231         machine->possible_cpus->cpus[i].type = core_type;
4232         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4233         machine->possible_cpus->cpus[i].arch_id = core_id;
4234         machine->possible_cpus->cpus[i].props.has_core_id = true;
4235         machine->possible_cpus->cpus[i].props.core_id = core_id;
4236     }
4237     return machine->possible_cpus;
4238 }
4239 
4240 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4241                                 uint64_t *buid, hwaddr *pio,
4242                                 hwaddr *mmio32, hwaddr *mmio64,
4243                                 unsigned n_dma, uint32_t *liobns,
4244                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4245 {
4246     /*
4247      * New-style PHB window placement.
4248      *
4249      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4250      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4251      * windows.
4252      *
4253      * Some guest kernels can't work with MMIO windows above 1<<46
4254      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4255      *
4256      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4257      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4258      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4259      * 1TiB 64-bit MMIO windows for each PHB.
4260      */
4261     const uint64_t base_buid = 0x800000020000000ULL;
4262     int i;
4263 
4264     /* Sanity check natural alignments */
4265     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4266     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4267     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4268     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4269     /* Sanity check bounds */
4270     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4271                       SPAPR_PCI_MEM32_WIN_SIZE);
4272     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4273                       SPAPR_PCI_MEM64_WIN_SIZE);
4274 
4275     if (index >= SPAPR_MAX_PHBS) {
4276         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4277                    SPAPR_MAX_PHBS - 1);
4278         return;
4279     }
4280 
4281     *buid = base_buid + index;
4282     for (i = 0; i < n_dma; ++i) {
4283         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4284     }
4285 
4286     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4287     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4288     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4289 
4290     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4291     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4292 }
4293 
4294 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4295 {
4296     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4297 
4298     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4299 }
4300 
4301 static void spapr_ics_resend(XICSFabric *dev)
4302 {
4303     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4304 
4305     ics_resend(spapr->ics);
4306 }
4307 
4308 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4309 {
4310     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4311 
4312     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4313 }
4314 
4315 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4316                                  Monitor *mon)
4317 {
4318     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4319 
4320     spapr_irq_print_info(spapr, mon);
4321     monitor_printf(mon, "irqchip: %s\n",
4322                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4323 }
4324 
4325 /*
4326  * This is a XIVE only operation
4327  */
4328 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4329                            uint8_t nvt_blk, uint32_t nvt_idx,
4330                            bool cam_ignore, uint8_t priority,
4331                            uint32_t logic_serv, XiveTCTXMatch *match)
4332 {
4333     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4334     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4335     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4336     int count;
4337 
4338     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4339                            priority, logic_serv, match);
4340     if (count < 0) {
4341         return count;
4342     }
4343 
4344     /*
4345      * When we implement the save and restore of the thread interrupt
4346      * contexts in the enter/exit CPU handlers of the machine and the
4347      * escalations in QEMU, we should be able to handle non dispatched
4348      * vCPUs.
4349      *
4350      * Until this is done, the sPAPR machine should find at least one
4351      * matching context always.
4352      */
4353     if (count == 0) {
4354         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4355                       nvt_blk, nvt_idx);
4356     }
4357 
4358     return count;
4359 }
4360 
4361 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4362 {
4363     return cpu->vcpu_id;
4364 }
4365 
4366 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4367 {
4368     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4369     MachineState *ms = MACHINE(spapr);
4370     int vcpu_id;
4371 
4372     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4373 
4374     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4375         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4376         error_append_hint(errp, "Adjust the number of cpus to %d "
4377                           "or try to raise the number of threads per core\n",
4378                           vcpu_id * ms->smp.threads / spapr->vsmt);
4379         return;
4380     }
4381 
4382     cpu->vcpu_id = vcpu_id;
4383 }
4384 
4385 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4386 {
4387     CPUState *cs;
4388 
4389     CPU_FOREACH(cs) {
4390         PowerPCCPU *cpu = POWERPC_CPU(cs);
4391 
4392         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4393             return cpu;
4394         }
4395     }
4396 
4397     return NULL;
4398 }
4399 
4400 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4401 {
4402     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4403 
4404     /* These are only called by TCG, KVM maintains dispatch state */
4405 
4406     spapr_cpu->prod = false;
4407     if (spapr_cpu->vpa_addr) {
4408         CPUState *cs = CPU(cpu);
4409         uint32_t dispatch;
4410 
4411         dispatch = ldl_be_phys(cs->as,
4412                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4413         dispatch++;
4414         if ((dispatch & 1) != 0) {
4415             qemu_log_mask(LOG_GUEST_ERROR,
4416                           "VPA: incorrect dispatch counter value for "
4417                           "dispatched partition %u, correcting.\n", dispatch);
4418             dispatch++;
4419         }
4420         stl_be_phys(cs->as,
4421                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4422     }
4423 }
4424 
4425 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4426 {
4427     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4428 
4429     if (spapr_cpu->vpa_addr) {
4430         CPUState *cs = CPU(cpu);
4431         uint32_t dispatch;
4432 
4433         dispatch = ldl_be_phys(cs->as,
4434                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4435         dispatch++;
4436         if ((dispatch & 1) != 1) {
4437             qemu_log_mask(LOG_GUEST_ERROR,
4438                           "VPA: incorrect dispatch counter value for "
4439                           "preempted partition %u, correcting.\n", dispatch);
4440             dispatch++;
4441         }
4442         stl_be_phys(cs->as,
4443                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4444     }
4445 }
4446 
4447 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4448 {
4449     MachineClass *mc = MACHINE_CLASS(oc);
4450     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4451     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4452     NMIClass *nc = NMI_CLASS(oc);
4453     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4454     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4455     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4456     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4457     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4458 
4459     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4460     mc->ignore_boot_device_suffixes = true;
4461 
4462     /*
4463      * We set up the default / latest behaviour here.  The class_init
4464      * functions for the specific versioned machine types can override
4465      * these details for backwards compatibility
4466      */
4467     mc->init = spapr_machine_init;
4468     mc->reset = spapr_machine_reset;
4469     mc->block_default_type = IF_SCSI;
4470     mc->max_cpus = 1024;
4471     mc->no_parallel = 1;
4472     mc->default_boot_order = "";
4473     mc->default_ram_size = 512 * MiB;
4474     mc->default_ram_id = "ppc_spapr.ram";
4475     mc->default_display = "std";
4476     mc->kvm_type = spapr_kvm_type;
4477     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4478     mc->pci_allow_0_address = true;
4479     assert(!mc->get_hotplug_handler);
4480     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4481     hc->pre_plug = spapr_machine_device_pre_plug;
4482     hc->plug = spapr_machine_device_plug;
4483     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4484     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4485     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4486     hc->unplug_request = spapr_machine_device_unplug_request;
4487     hc->unplug = spapr_machine_device_unplug;
4488 
4489     smc->dr_lmb_enabled = true;
4490     smc->update_dt_enabled = true;
4491     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4492     mc->has_hotpluggable_cpus = true;
4493     mc->nvdimm_supported = true;
4494     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4495     fwc->get_dev_path = spapr_get_fw_dev_path;
4496     nc->nmi_monitor_handler = spapr_nmi;
4497     smc->phb_placement = spapr_phb_placement;
4498     vhc->hypercall = emulate_spapr_hypercall;
4499     vhc->hpt_mask = spapr_hpt_mask;
4500     vhc->map_hptes = spapr_map_hptes;
4501     vhc->unmap_hptes = spapr_unmap_hptes;
4502     vhc->hpte_set_c = spapr_hpte_set_c;
4503     vhc->hpte_set_r = spapr_hpte_set_r;
4504     vhc->get_pate = spapr_get_pate;
4505     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4506     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4507     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4508     xic->ics_get = spapr_ics_get;
4509     xic->ics_resend = spapr_ics_resend;
4510     xic->icp_get = spapr_icp_get;
4511     ispc->print_info = spapr_pic_print_info;
4512     /* Force NUMA node memory size to be a multiple of
4513      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4514      * in which LMBs are represented and hot-added
4515      */
4516     mc->numa_mem_align_shift = 28;
4517     mc->numa_mem_supported = true;
4518     mc->auto_enable_numa = true;
4519 
4520     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4521     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4522     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4523     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4524     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4525     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4526     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4527     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4528     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4529     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4530     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_ON;
4531     spapr_caps_add_properties(smc, &error_abort);
4532     smc->irq = &spapr_irq_dual;
4533     smc->dr_phb_enabled = true;
4534     smc->linux_pci_probe = true;
4535     smc->smp_threads_vsmt = true;
4536     smc->nr_xirqs = SPAPR_NR_XIRQS;
4537     xfc->match_nvt = spapr_match_nvt;
4538 }
4539 
4540 static const TypeInfo spapr_machine_info = {
4541     .name          = TYPE_SPAPR_MACHINE,
4542     .parent        = TYPE_MACHINE,
4543     .abstract      = true,
4544     .instance_size = sizeof(SpaprMachineState),
4545     .instance_init = spapr_instance_init,
4546     .instance_finalize = spapr_machine_finalizefn,
4547     .class_size    = sizeof(SpaprMachineClass),
4548     .class_init    = spapr_machine_class_init,
4549     .interfaces = (InterfaceInfo[]) {
4550         { TYPE_FW_PATH_PROVIDER },
4551         { TYPE_NMI },
4552         { TYPE_HOTPLUG_HANDLER },
4553         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4554         { TYPE_XICS_FABRIC },
4555         { TYPE_INTERRUPT_STATS_PROVIDER },
4556         { TYPE_XIVE_FABRIC },
4557         { }
4558     },
4559 };
4560 
4561 static void spapr_machine_latest_class_options(MachineClass *mc)
4562 {
4563     mc->alias = "pseries";
4564     mc->is_default = true;
4565 }
4566 
4567 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4568     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4569                                                     void *data)      \
4570     {                                                                \
4571         MachineClass *mc = MACHINE_CLASS(oc);                        \
4572         spapr_machine_##suffix##_class_options(mc);                  \
4573         if (latest) {                                                \
4574             spapr_machine_latest_class_options(mc);                  \
4575         }                                                            \
4576     }                                                                \
4577     static const TypeInfo spapr_machine_##suffix##_info = {          \
4578         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4579         .parent = TYPE_SPAPR_MACHINE,                                \
4580         .class_init = spapr_machine_##suffix##_class_init,           \
4581     };                                                               \
4582     static void spapr_machine_register_##suffix(void)                \
4583     {                                                                \
4584         type_register(&spapr_machine_##suffix##_info);               \
4585     }                                                                \
4586     type_init(spapr_machine_register_##suffix)
4587 
4588 /*
4589  * pseries-5.0
4590  */
4591 static void spapr_machine_5_0_class_options(MachineClass *mc)
4592 {
4593     /* Defaults for the latest behaviour inherited from the base class */
4594 }
4595 
4596 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4597 
4598 /*
4599  * pseries-4.2
4600  */
4601 static void spapr_machine_4_2_class_options(MachineClass *mc)
4602 {
4603     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4604 
4605     spapr_machine_5_0_class_options(mc);
4606     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4607     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4608     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_OFF;
4609     mc->nvdimm_supported = false;
4610 }
4611 
4612 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4613 
4614 /*
4615  * pseries-4.1
4616  */
4617 static void spapr_machine_4_1_class_options(MachineClass *mc)
4618 {
4619     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4620     static GlobalProperty compat[] = {
4621         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4622         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4623     };
4624 
4625     spapr_machine_4_2_class_options(mc);
4626     smc->linux_pci_probe = false;
4627     smc->smp_threads_vsmt = false;
4628     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4629     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4630 }
4631 
4632 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4633 
4634 /*
4635  * pseries-4.0
4636  */
4637 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4638                               uint64_t *buid, hwaddr *pio,
4639                               hwaddr *mmio32, hwaddr *mmio64,
4640                               unsigned n_dma, uint32_t *liobns,
4641                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4642 {
4643     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4644                         nv2gpa, nv2atsd, errp);
4645     *nv2gpa = 0;
4646     *nv2atsd = 0;
4647 }
4648 
4649 static void spapr_machine_4_0_class_options(MachineClass *mc)
4650 {
4651     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4652 
4653     spapr_machine_4_1_class_options(mc);
4654     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4655     smc->phb_placement = phb_placement_4_0;
4656     smc->irq = &spapr_irq_xics;
4657     smc->pre_4_1_migration = true;
4658 }
4659 
4660 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4661 
4662 /*
4663  * pseries-3.1
4664  */
4665 static void spapr_machine_3_1_class_options(MachineClass *mc)
4666 {
4667     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4668 
4669     spapr_machine_4_0_class_options(mc);
4670     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4671 
4672     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4673     smc->update_dt_enabled = false;
4674     smc->dr_phb_enabled = false;
4675     smc->broken_host_serial_model = true;
4676     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4677     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4678     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4679     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4680 }
4681 
4682 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4683 
4684 /*
4685  * pseries-3.0
4686  */
4687 
4688 static void spapr_machine_3_0_class_options(MachineClass *mc)
4689 {
4690     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4691 
4692     spapr_machine_3_1_class_options(mc);
4693     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4694 
4695     smc->legacy_irq_allocation = true;
4696     smc->nr_xirqs = 0x400;
4697     smc->irq = &spapr_irq_xics_legacy;
4698 }
4699 
4700 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4701 
4702 /*
4703  * pseries-2.12
4704  */
4705 static void spapr_machine_2_12_class_options(MachineClass *mc)
4706 {
4707     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4708     static GlobalProperty compat[] = {
4709         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4710         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4711     };
4712 
4713     spapr_machine_3_0_class_options(mc);
4714     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4715     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4716 
4717     /* We depend on kvm_enabled() to choose a default value for the
4718      * hpt-max-page-size capability. Of course we can't do it here
4719      * because this is too early and the HW accelerator isn't initialzed
4720      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4721      */
4722     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4723 }
4724 
4725 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4726 
4727 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4728 {
4729     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4730 
4731     spapr_machine_2_12_class_options(mc);
4732     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4733     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4734     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4735 }
4736 
4737 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4738 
4739 /*
4740  * pseries-2.11
4741  */
4742 
4743 static void spapr_machine_2_11_class_options(MachineClass *mc)
4744 {
4745     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4746 
4747     spapr_machine_2_12_class_options(mc);
4748     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4749     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4750 }
4751 
4752 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4753 
4754 /*
4755  * pseries-2.10
4756  */
4757 
4758 static void spapr_machine_2_10_class_options(MachineClass *mc)
4759 {
4760     spapr_machine_2_11_class_options(mc);
4761     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4762 }
4763 
4764 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4765 
4766 /*
4767  * pseries-2.9
4768  */
4769 
4770 static void spapr_machine_2_9_class_options(MachineClass *mc)
4771 {
4772     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4773     static GlobalProperty compat[] = {
4774         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4775     };
4776 
4777     spapr_machine_2_10_class_options(mc);
4778     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4779     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4780     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4781     smc->pre_2_10_has_unused_icps = true;
4782     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4783 }
4784 
4785 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4786 
4787 /*
4788  * pseries-2.8
4789  */
4790 
4791 static void spapr_machine_2_8_class_options(MachineClass *mc)
4792 {
4793     static GlobalProperty compat[] = {
4794         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4795     };
4796 
4797     spapr_machine_2_9_class_options(mc);
4798     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4799     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4800     mc->numa_mem_align_shift = 23;
4801 }
4802 
4803 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4804 
4805 /*
4806  * pseries-2.7
4807  */
4808 
4809 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4810                               uint64_t *buid, hwaddr *pio,
4811                               hwaddr *mmio32, hwaddr *mmio64,
4812                               unsigned n_dma, uint32_t *liobns,
4813                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4814 {
4815     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4816     const uint64_t base_buid = 0x800000020000000ULL;
4817     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4818     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4819     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4820     const uint32_t max_index = 255;
4821     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4822 
4823     uint64_t ram_top = MACHINE(spapr)->ram_size;
4824     hwaddr phb0_base, phb_base;
4825     int i;
4826 
4827     /* Do we have device memory? */
4828     if (MACHINE(spapr)->maxram_size > ram_top) {
4829         /* Can't just use maxram_size, because there may be an
4830          * alignment gap between normal and device memory regions
4831          */
4832         ram_top = MACHINE(spapr)->device_memory->base +
4833             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4834     }
4835 
4836     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4837 
4838     if (index > max_index) {
4839         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4840                    max_index);
4841         return;
4842     }
4843 
4844     *buid = base_buid + index;
4845     for (i = 0; i < n_dma; ++i) {
4846         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4847     }
4848 
4849     phb_base = phb0_base + index * phb_spacing;
4850     *pio = phb_base + pio_offset;
4851     *mmio32 = phb_base + mmio_offset;
4852     /*
4853      * We don't set the 64-bit MMIO window, relying on the PHB's
4854      * fallback behaviour of automatically splitting a large "32-bit"
4855      * window into contiguous 32-bit and 64-bit windows
4856      */
4857 
4858     *nv2gpa = 0;
4859     *nv2atsd = 0;
4860 }
4861 
4862 static void spapr_machine_2_7_class_options(MachineClass *mc)
4863 {
4864     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4865     static GlobalProperty compat[] = {
4866         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4867         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4868         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4869         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4870     };
4871 
4872     spapr_machine_2_8_class_options(mc);
4873     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4874     mc->default_machine_opts = "modern-hotplug-events=off";
4875     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4876     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4877     smc->phb_placement = phb_placement_2_7;
4878 }
4879 
4880 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4881 
4882 /*
4883  * pseries-2.6
4884  */
4885 
4886 static void spapr_machine_2_6_class_options(MachineClass *mc)
4887 {
4888     static GlobalProperty compat[] = {
4889         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4890     };
4891 
4892     spapr_machine_2_7_class_options(mc);
4893     mc->has_hotpluggable_cpus = false;
4894     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4895     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4896 }
4897 
4898 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4899 
4900 /*
4901  * pseries-2.5
4902  */
4903 
4904 static void spapr_machine_2_5_class_options(MachineClass *mc)
4905 {
4906     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4907     static GlobalProperty compat[] = {
4908         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4909     };
4910 
4911     spapr_machine_2_6_class_options(mc);
4912     smc->use_ohci_by_default = true;
4913     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4914     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4915 }
4916 
4917 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4918 
4919 /*
4920  * pseries-2.4
4921  */
4922 
4923 static void spapr_machine_2_4_class_options(MachineClass *mc)
4924 {
4925     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4926 
4927     spapr_machine_2_5_class_options(mc);
4928     smc->dr_lmb_enabled = false;
4929     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4930 }
4931 
4932 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4933 
4934 /*
4935  * pseries-2.3
4936  */
4937 
4938 static void spapr_machine_2_3_class_options(MachineClass *mc)
4939 {
4940     static GlobalProperty compat[] = {
4941         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4942     };
4943     spapr_machine_2_4_class_options(mc);
4944     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4945     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4946 }
4947 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4948 
4949 /*
4950  * pseries-2.2
4951  */
4952 
4953 static void spapr_machine_2_2_class_options(MachineClass *mc)
4954 {
4955     static GlobalProperty compat[] = {
4956         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4957     };
4958 
4959     spapr_machine_2_3_class_options(mc);
4960     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4961     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4962     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4963 }
4964 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4965 
4966 /*
4967  * pseries-2.1
4968  */
4969 
4970 static void spapr_machine_2_1_class_options(MachineClass *mc)
4971 {
4972     spapr_machine_2_2_class_options(mc);
4973     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4974 }
4975 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4976 
4977 static void spapr_machine_register_types(void)
4978 {
4979     type_register_static(&spapr_machine_info);
4980 }
4981 
4982 type_init(spapr_machine_register_types)
4983