xref: /qemu/hw/ppc/spapr.c (revision 8897ea5a9fc0aafa5ed7eee1e0c49893b91a2d87)
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(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 (kvm_enabled()) {
1589         hwaddr vrma_limit = kvmppc_vrma_limit(spapr->htab_shift);
1590 
1591         /* Check our RMA fits in the possible VRMA */
1592         if (vrma_limit < spapr->rma_size) {
1593             error_report("Unable to create %" HWADDR_PRIu
1594                          "MiB RMA (VRMA only allows %" HWADDR_PRIu "MiB",
1595                          spapr->rma_size / MiB, vrma_limit / MiB);
1596             exit(EXIT_FAILURE);
1597         }
1598     }
1599 }
1600 
1601 static int spapr_reset_drcs(Object *child, void *opaque)
1602 {
1603     SpaprDrc *drc =
1604         (SpaprDrc *) object_dynamic_cast(child,
1605                                                  TYPE_SPAPR_DR_CONNECTOR);
1606 
1607     if (drc) {
1608         spapr_drc_reset(drc);
1609     }
1610 
1611     return 0;
1612 }
1613 
1614 static void spapr_machine_reset(MachineState *machine)
1615 {
1616     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1617     PowerPCCPU *first_ppc_cpu;
1618     hwaddr fdt_addr;
1619     void *fdt;
1620     int rc;
1621 
1622     kvmppc_svm_off(&error_fatal);
1623     spapr_caps_apply(spapr);
1624 
1625     first_ppc_cpu = POWERPC_CPU(first_cpu);
1626     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1627         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1628                               spapr->max_compat_pvr)) {
1629         /*
1630          * If using KVM with radix mode available, VCPUs can be started
1631          * without a HPT because KVM will start them in radix mode.
1632          * Set the GR bit in PATE so that we know there is no HPT.
1633          */
1634         spapr->patb_entry = PATE1_GR;
1635         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1636     } else {
1637         spapr_setup_hpt(spapr);
1638     }
1639 
1640     qemu_devices_reset();
1641 
1642     /*
1643      * If this reset wasn't generated by CAS, we should reset our
1644      * negotiated options and start from scratch
1645      */
1646     if (!spapr->cas_reboot) {
1647         spapr_ovec_cleanup(spapr->ov5_cas);
1648         spapr->ov5_cas = spapr_ovec_new();
1649 
1650         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1651     }
1652 
1653     /*
1654      * This is fixing some of the default configuration of the XIVE
1655      * devices. To be called after the reset of the machine devices.
1656      */
1657     spapr_irq_reset(spapr, &error_fatal);
1658 
1659     /*
1660      * There is no CAS under qtest. Simulate one to please the code that
1661      * depends on spapr->ov5_cas. This is especially needed to test device
1662      * unplug, so we do that before resetting the DRCs.
1663      */
1664     if (qtest_enabled()) {
1665         spapr_ovec_cleanup(spapr->ov5_cas);
1666         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1667     }
1668 
1669     /* DRC reset may cause a device to be unplugged. This will cause troubles
1670      * if this device is used by another device (eg, a running vhost backend
1671      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1672      * situations, we reset DRCs after all devices have been reset.
1673      */
1674     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1675 
1676     spapr_clear_pending_events(spapr);
1677 
1678     /*
1679      * We place the device tree and RTAS just below either the top of the RMA,
1680      * or just below 2GB, whichever is lower, so that it can be
1681      * processed with 32-bit real mode code if necessary
1682      */
1683     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1684 
1685     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1686 
1687     rc = fdt_pack(fdt);
1688 
1689     /* Should only fail if we've built a corrupted tree */
1690     assert(rc == 0);
1691 
1692     /* Load the fdt */
1693     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1694     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1695     g_free(spapr->fdt_blob);
1696     spapr->fdt_size = fdt_totalsize(fdt);
1697     spapr->fdt_initial_size = spapr->fdt_size;
1698     spapr->fdt_blob = fdt;
1699 
1700     /* Set up the entry state */
1701     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1702     first_ppc_cpu->env.gpr[5] = 0;
1703 
1704     spapr->cas_reboot = false;
1705 
1706     spapr->mc_status = -1;
1707     spapr->guest_machine_check_addr = -1;
1708 
1709     /* Signal all vCPUs waiting on this condition */
1710     qemu_cond_broadcast(&spapr->mc_delivery_cond);
1711 
1712     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1713 }
1714 
1715 static void spapr_create_nvram(SpaprMachineState *spapr)
1716 {
1717     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1718     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1719 
1720     if (dinfo) {
1721         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1722                             &error_fatal);
1723     }
1724 
1725     qdev_init_nofail(dev);
1726 
1727     spapr->nvram = (struct SpaprNvram *)dev;
1728 }
1729 
1730 static void spapr_rtc_create(SpaprMachineState *spapr)
1731 {
1732     object_initialize_child(OBJECT(spapr), "rtc",
1733                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1734                             &error_fatal, NULL);
1735     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1736                               &error_fatal);
1737     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1738                               "date", &error_fatal);
1739 }
1740 
1741 /* Returns whether we want to use VGA or not */
1742 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1743 {
1744     switch (vga_interface_type) {
1745     case VGA_NONE:
1746         return false;
1747     case VGA_DEVICE:
1748         return true;
1749     case VGA_STD:
1750     case VGA_VIRTIO:
1751     case VGA_CIRRUS:
1752         return pci_vga_init(pci_bus) != NULL;
1753     default:
1754         error_setg(errp,
1755                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1756         return false;
1757     }
1758 }
1759 
1760 static int spapr_pre_load(void *opaque)
1761 {
1762     int rc;
1763 
1764     rc = spapr_caps_pre_load(opaque);
1765     if (rc) {
1766         return rc;
1767     }
1768 
1769     return 0;
1770 }
1771 
1772 static int spapr_post_load(void *opaque, int version_id)
1773 {
1774     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1775     int err = 0;
1776 
1777     err = spapr_caps_post_migration(spapr);
1778     if (err) {
1779         return err;
1780     }
1781 
1782     /*
1783      * In earlier versions, there was no separate qdev for the PAPR
1784      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1785      * So when migrating from those versions, poke the incoming offset
1786      * value into the RTC device
1787      */
1788     if (version_id < 3) {
1789         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1790         if (err) {
1791             return err;
1792         }
1793     }
1794 
1795     if (kvm_enabled() && spapr->patb_entry) {
1796         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1797         bool radix = !!(spapr->patb_entry & PATE1_GR);
1798         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1799 
1800         /*
1801          * Update LPCR:HR and UPRT as they may not be set properly in
1802          * the stream
1803          */
1804         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1805                             LPCR_HR | LPCR_UPRT);
1806 
1807         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1808         if (err) {
1809             error_report("Process table config unsupported by the host");
1810             return -EINVAL;
1811         }
1812     }
1813 
1814     err = spapr_irq_post_load(spapr, version_id);
1815     if (err) {
1816         return err;
1817     }
1818 
1819     return err;
1820 }
1821 
1822 static int spapr_pre_save(void *opaque)
1823 {
1824     int rc;
1825 
1826     rc = spapr_caps_pre_save(opaque);
1827     if (rc) {
1828         return rc;
1829     }
1830 
1831     return 0;
1832 }
1833 
1834 static bool version_before_3(void *opaque, int version_id)
1835 {
1836     return version_id < 3;
1837 }
1838 
1839 static bool spapr_pending_events_needed(void *opaque)
1840 {
1841     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1842     return !QTAILQ_EMPTY(&spapr->pending_events);
1843 }
1844 
1845 static const VMStateDescription vmstate_spapr_event_entry = {
1846     .name = "spapr_event_log_entry",
1847     .version_id = 1,
1848     .minimum_version_id = 1,
1849     .fields = (VMStateField[]) {
1850         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1851         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1852         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1853                                      NULL, extended_length),
1854         VMSTATE_END_OF_LIST()
1855     },
1856 };
1857 
1858 static const VMStateDescription vmstate_spapr_pending_events = {
1859     .name = "spapr_pending_events",
1860     .version_id = 1,
1861     .minimum_version_id = 1,
1862     .needed = spapr_pending_events_needed,
1863     .fields = (VMStateField[]) {
1864         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1865                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1866         VMSTATE_END_OF_LIST()
1867     },
1868 };
1869 
1870 static bool spapr_ov5_cas_needed(void *opaque)
1871 {
1872     SpaprMachineState *spapr = opaque;
1873     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1874     bool cas_needed;
1875 
1876     /* Prior to the introduction of SpaprOptionVector, we had two option
1877      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1878      * Both of these options encode machine topology into the device-tree
1879      * in such a way that the now-booted OS should still be able to interact
1880      * appropriately with QEMU regardless of what options were actually
1881      * negotiatied on the source side.
1882      *
1883      * As such, we can avoid migrating the CAS-negotiated options if these
1884      * are the only options available on the current machine/platform.
1885      * Since these are the only options available for pseries-2.7 and
1886      * earlier, this allows us to maintain old->new/new->old migration
1887      * compatibility.
1888      *
1889      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1890      * via default pseries-2.8 machines and explicit command-line parameters.
1891      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1892      * of the actual CAS-negotiated values to continue working properly. For
1893      * example, availability of memory unplug depends on knowing whether
1894      * OV5_HP_EVT was negotiated via CAS.
1895      *
1896      * Thus, for any cases where the set of available CAS-negotiatable
1897      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1898      * include the CAS-negotiated options in the migration stream, unless
1899      * if they affect boot time behaviour only.
1900      */
1901     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1902     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1903     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1904 
1905     /* We need extra information if we have any bits outside the mask
1906      * defined above */
1907     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1908 
1909     spapr_ovec_cleanup(ov5_mask);
1910 
1911     return cas_needed;
1912 }
1913 
1914 static const VMStateDescription vmstate_spapr_ov5_cas = {
1915     .name = "spapr_option_vector_ov5_cas",
1916     .version_id = 1,
1917     .minimum_version_id = 1,
1918     .needed = spapr_ov5_cas_needed,
1919     .fields = (VMStateField[]) {
1920         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1921                                  vmstate_spapr_ovec, SpaprOptionVector),
1922         VMSTATE_END_OF_LIST()
1923     },
1924 };
1925 
1926 static bool spapr_patb_entry_needed(void *opaque)
1927 {
1928     SpaprMachineState *spapr = opaque;
1929 
1930     return !!spapr->patb_entry;
1931 }
1932 
1933 static const VMStateDescription vmstate_spapr_patb_entry = {
1934     .name = "spapr_patb_entry",
1935     .version_id = 1,
1936     .minimum_version_id = 1,
1937     .needed = spapr_patb_entry_needed,
1938     .fields = (VMStateField[]) {
1939         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1940         VMSTATE_END_OF_LIST()
1941     },
1942 };
1943 
1944 static bool spapr_irq_map_needed(void *opaque)
1945 {
1946     SpaprMachineState *spapr = opaque;
1947 
1948     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1949 }
1950 
1951 static const VMStateDescription vmstate_spapr_irq_map = {
1952     .name = "spapr_irq_map",
1953     .version_id = 1,
1954     .minimum_version_id = 1,
1955     .needed = spapr_irq_map_needed,
1956     .fields = (VMStateField[]) {
1957         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1958         VMSTATE_END_OF_LIST()
1959     },
1960 };
1961 
1962 static bool spapr_dtb_needed(void *opaque)
1963 {
1964     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1965 
1966     return smc->update_dt_enabled;
1967 }
1968 
1969 static int spapr_dtb_pre_load(void *opaque)
1970 {
1971     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1972 
1973     g_free(spapr->fdt_blob);
1974     spapr->fdt_blob = NULL;
1975     spapr->fdt_size = 0;
1976 
1977     return 0;
1978 }
1979 
1980 static const VMStateDescription vmstate_spapr_dtb = {
1981     .name = "spapr_dtb",
1982     .version_id = 1,
1983     .minimum_version_id = 1,
1984     .needed = spapr_dtb_needed,
1985     .pre_load = spapr_dtb_pre_load,
1986     .fields = (VMStateField[]) {
1987         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1988         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1989         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1990                                      fdt_size),
1991         VMSTATE_END_OF_LIST()
1992     },
1993 };
1994 
1995 static bool spapr_fwnmi_needed(void *opaque)
1996 {
1997     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1998 
1999     return spapr->guest_machine_check_addr != -1;
2000 }
2001 
2002 static int spapr_fwnmi_pre_save(void *opaque)
2003 {
2004     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2005 
2006     /*
2007      * Check if machine check handling is in progress and print a
2008      * warning message.
2009      */
2010     if (spapr->mc_status != -1) {
2011         warn_report("A machine check is being handled during migration. The"
2012                 "handler may run and log hardware error on the destination");
2013     }
2014 
2015     return 0;
2016 }
2017 
2018 static const VMStateDescription vmstate_spapr_machine_check = {
2019     .name = "spapr_machine_check",
2020     .version_id = 1,
2021     .minimum_version_id = 1,
2022     .needed = spapr_fwnmi_needed,
2023     .pre_save = spapr_fwnmi_pre_save,
2024     .fields = (VMStateField[]) {
2025         VMSTATE_UINT64(guest_machine_check_addr, SpaprMachineState),
2026         VMSTATE_INT32(mc_status, SpaprMachineState),
2027         VMSTATE_END_OF_LIST()
2028     },
2029 };
2030 
2031 static const VMStateDescription vmstate_spapr = {
2032     .name = "spapr",
2033     .version_id = 3,
2034     .minimum_version_id = 1,
2035     .pre_load = spapr_pre_load,
2036     .post_load = spapr_post_load,
2037     .pre_save = spapr_pre_save,
2038     .fields = (VMStateField[]) {
2039         /* used to be @next_irq */
2040         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2041 
2042         /* RTC offset */
2043         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2044 
2045         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2046         VMSTATE_END_OF_LIST()
2047     },
2048     .subsections = (const VMStateDescription*[]) {
2049         &vmstate_spapr_ov5_cas,
2050         &vmstate_spapr_patb_entry,
2051         &vmstate_spapr_pending_events,
2052         &vmstate_spapr_cap_htm,
2053         &vmstate_spapr_cap_vsx,
2054         &vmstate_spapr_cap_dfp,
2055         &vmstate_spapr_cap_cfpc,
2056         &vmstate_spapr_cap_sbbc,
2057         &vmstate_spapr_cap_ibs,
2058         &vmstate_spapr_cap_hpt_maxpagesize,
2059         &vmstate_spapr_irq_map,
2060         &vmstate_spapr_cap_nested_kvm_hv,
2061         &vmstate_spapr_dtb,
2062         &vmstate_spapr_cap_large_decr,
2063         &vmstate_spapr_cap_ccf_assist,
2064         &vmstate_spapr_cap_fwnmi,
2065         &vmstate_spapr_machine_check,
2066         NULL
2067     }
2068 };
2069 
2070 static int htab_save_setup(QEMUFile *f, void *opaque)
2071 {
2072     SpaprMachineState *spapr = opaque;
2073 
2074     /* "Iteration" header */
2075     if (!spapr->htab_shift) {
2076         qemu_put_be32(f, -1);
2077     } else {
2078         qemu_put_be32(f, spapr->htab_shift);
2079     }
2080 
2081     if (spapr->htab) {
2082         spapr->htab_save_index = 0;
2083         spapr->htab_first_pass = true;
2084     } else {
2085         if (spapr->htab_shift) {
2086             assert(kvm_enabled());
2087         }
2088     }
2089 
2090 
2091     return 0;
2092 }
2093 
2094 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2095                             int chunkstart, int n_valid, int n_invalid)
2096 {
2097     qemu_put_be32(f, chunkstart);
2098     qemu_put_be16(f, n_valid);
2099     qemu_put_be16(f, n_invalid);
2100     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2101                     HASH_PTE_SIZE_64 * n_valid);
2102 }
2103 
2104 static void htab_save_end_marker(QEMUFile *f)
2105 {
2106     qemu_put_be32(f, 0);
2107     qemu_put_be16(f, 0);
2108     qemu_put_be16(f, 0);
2109 }
2110 
2111 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2112                                  int64_t max_ns)
2113 {
2114     bool has_timeout = max_ns != -1;
2115     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2116     int index = spapr->htab_save_index;
2117     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2118 
2119     assert(spapr->htab_first_pass);
2120 
2121     do {
2122         int chunkstart;
2123 
2124         /* Consume invalid HPTEs */
2125         while ((index < htabslots)
2126                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2127             CLEAN_HPTE(HPTE(spapr->htab, index));
2128             index++;
2129         }
2130 
2131         /* Consume valid HPTEs */
2132         chunkstart = index;
2133         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2134                && HPTE_VALID(HPTE(spapr->htab, index))) {
2135             CLEAN_HPTE(HPTE(spapr->htab, index));
2136             index++;
2137         }
2138 
2139         if (index > chunkstart) {
2140             int n_valid = index - chunkstart;
2141 
2142             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2143 
2144             if (has_timeout &&
2145                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2146                 break;
2147             }
2148         }
2149     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2150 
2151     if (index >= htabslots) {
2152         assert(index == htabslots);
2153         index = 0;
2154         spapr->htab_first_pass = false;
2155     }
2156     spapr->htab_save_index = index;
2157 }
2158 
2159 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2160                                 int64_t max_ns)
2161 {
2162     bool final = max_ns < 0;
2163     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2164     int examined = 0, sent = 0;
2165     int index = spapr->htab_save_index;
2166     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2167 
2168     assert(!spapr->htab_first_pass);
2169 
2170     do {
2171         int chunkstart, invalidstart;
2172 
2173         /* Consume non-dirty HPTEs */
2174         while ((index < htabslots)
2175                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2176             index++;
2177             examined++;
2178         }
2179 
2180         chunkstart = index;
2181         /* Consume valid dirty HPTEs */
2182         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2183                && HPTE_DIRTY(HPTE(spapr->htab, index))
2184                && HPTE_VALID(HPTE(spapr->htab, index))) {
2185             CLEAN_HPTE(HPTE(spapr->htab, index));
2186             index++;
2187             examined++;
2188         }
2189 
2190         invalidstart = index;
2191         /* Consume invalid dirty HPTEs */
2192         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2193                && HPTE_DIRTY(HPTE(spapr->htab, index))
2194                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2195             CLEAN_HPTE(HPTE(spapr->htab, index));
2196             index++;
2197             examined++;
2198         }
2199 
2200         if (index > chunkstart) {
2201             int n_valid = invalidstart - chunkstart;
2202             int n_invalid = index - invalidstart;
2203 
2204             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2205             sent += index - chunkstart;
2206 
2207             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2208                 break;
2209             }
2210         }
2211 
2212         if (examined >= htabslots) {
2213             break;
2214         }
2215 
2216         if (index >= htabslots) {
2217             assert(index == htabslots);
2218             index = 0;
2219         }
2220     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2221 
2222     if (index >= htabslots) {
2223         assert(index == htabslots);
2224         index = 0;
2225     }
2226 
2227     spapr->htab_save_index = index;
2228 
2229     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2230 }
2231 
2232 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2233 #define MAX_KVM_BUF_SIZE    2048
2234 
2235 static int htab_save_iterate(QEMUFile *f, void *opaque)
2236 {
2237     SpaprMachineState *spapr = opaque;
2238     int fd;
2239     int rc = 0;
2240 
2241     /* Iteration header */
2242     if (!spapr->htab_shift) {
2243         qemu_put_be32(f, -1);
2244         return 1;
2245     } else {
2246         qemu_put_be32(f, 0);
2247     }
2248 
2249     if (!spapr->htab) {
2250         assert(kvm_enabled());
2251 
2252         fd = get_htab_fd(spapr);
2253         if (fd < 0) {
2254             return fd;
2255         }
2256 
2257         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2258         if (rc < 0) {
2259             return rc;
2260         }
2261     } else  if (spapr->htab_first_pass) {
2262         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2263     } else {
2264         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2265     }
2266 
2267     htab_save_end_marker(f);
2268 
2269     return rc;
2270 }
2271 
2272 static int htab_save_complete(QEMUFile *f, void *opaque)
2273 {
2274     SpaprMachineState *spapr = opaque;
2275     int fd;
2276 
2277     /* Iteration header */
2278     if (!spapr->htab_shift) {
2279         qemu_put_be32(f, -1);
2280         return 0;
2281     } else {
2282         qemu_put_be32(f, 0);
2283     }
2284 
2285     if (!spapr->htab) {
2286         int rc;
2287 
2288         assert(kvm_enabled());
2289 
2290         fd = get_htab_fd(spapr);
2291         if (fd < 0) {
2292             return fd;
2293         }
2294 
2295         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2296         if (rc < 0) {
2297             return rc;
2298         }
2299     } else {
2300         if (spapr->htab_first_pass) {
2301             htab_save_first_pass(f, spapr, -1);
2302         }
2303         htab_save_later_pass(f, spapr, -1);
2304     }
2305 
2306     /* End marker */
2307     htab_save_end_marker(f);
2308 
2309     return 0;
2310 }
2311 
2312 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2313 {
2314     SpaprMachineState *spapr = opaque;
2315     uint32_t section_hdr;
2316     int fd = -1;
2317     Error *local_err = NULL;
2318 
2319     if (version_id < 1 || version_id > 1) {
2320         error_report("htab_load() bad version");
2321         return -EINVAL;
2322     }
2323 
2324     section_hdr = qemu_get_be32(f);
2325 
2326     if (section_hdr == -1) {
2327         spapr_free_hpt(spapr);
2328         return 0;
2329     }
2330 
2331     if (section_hdr) {
2332         /* First section gives the htab size */
2333         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2334         if (local_err) {
2335             error_report_err(local_err);
2336             return -EINVAL;
2337         }
2338         return 0;
2339     }
2340 
2341     if (!spapr->htab) {
2342         assert(kvm_enabled());
2343 
2344         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2345         if (fd < 0) {
2346             error_report_err(local_err);
2347             return fd;
2348         }
2349     }
2350 
2351     while (true) {
2352         uint32_t index;
2353         uint16_t n_valid, n_invalid;
2354 
2355         index = qemu_get_be32(f);
2356         n_valid = qemu_get_be16(f);
2357         n_invalid = qemu_get_be16(f);
2358 
2359         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2360             /* End of Stream */
2361             break;
2362         }
2363 
2364         if ((index + n_valid + n_invalid) >
2365             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2366             /* Bad index in stream */
2367             error_report(
2368                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2369                 index, n_valid, n_invalid, spapr->htab_shift);
2370             return -EINVAL;
2371         }
2372 
2373         if (spapr->htab) {
2374             if (n_valid) {
2375                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2376                                 HASH_PTE_SIZE_64 * n_valid);
2377             }
2378             if (n_invalid) {
2379                 memset(HPTE(spapr->htab, index + n_valid), 0,
2380                        HASH_PTE_SIZE_64 * n_invalid);
2381             }
2382         } else {
2383             int rc;
2384 
2385             assert(fd >= 0);
2386 
2387             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2388             if (rc < 0) {
2389                 return rc;
2390             }
2391         }
2392     }
2393 
2394     if (!spapr->htab) {
2395         assert(fd >= 0);
2396         close(fd);
2397     }
2398 
2399     return 0;
2400 }
2401 
2402 static void htab_save_cleanup(void *opaque)
2403 {
2404     SpaprMachineState *spapr = opaque;
2405 
2406     close_htab_fd(spapr);
2407 }
2408 
2409 static SaveVMHandlers savevm_htab_handlers = {
2410     .save_setup = htab_save_setup,
2411     .save_live_iterate = htab_save_iterate,
2412     .save_live_complete_precopy = htab_save_complete,
2413     .save_cleanup = htab_save_cleanup,
2414     .load_state = htab_load,
2415 };
2416 
2417 static void spapr_boot_set(void *opaque, const char *boot_device,
2418                            Error **errp)
2419 {
2420     MachineState *machine = MACHINE(opaque);
2421     machine->boot_order = g_strdup(boot_device);
2422 }
2423 
2424 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2425 {
2426     MachineState *machine = MACHINE(spapr);
2427     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2428     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2429     int i;
2430 
2431     for (i = 0; i < nr_lmbs; i++) {
2432         uint64_t addr;
2433 
2434         addr = i * lmb_size + machine->device_memory->base;
2435         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2436                                addr / lmb_size);
2437     }
2438 }
2439 
2440 /*
2441  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2442  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2443  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2444  */
2445 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2446 {
2447     int i;
2448 
2449     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2450         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2451                    " is not aligned to %" PRIu64 " MiB",
2452                    machine->ram_size,
2453                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2454         return;
2455     }
2456 
2457     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2458         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2459                    " is not aligned to %" PRIu64 " MiB",
2460                    machine->ram_size,
2461                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2462         return;
2463     }
2464 
2465     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2466         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2467             error_setg(errp,
2468                        "Node %d memory size 0x%" PRIx64
2469                        " is not aligned to %" PRIu64 " MiB",
2470                        i, machine->numa_state->nodes[i].node_mem,
2471                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2472             return;
2473         }
2474     }
2475 }
2476 
2477 /* find cpu slot in machine->possible_cpus by core_id */
2478 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2479 {
2480     int index = id / ms->smp.threads;
2481 
2482     if (index >= ms->possible_cpus->len) {
2483         return NULL;
2484     }
2485     if (idx) {
2486         *idx = index;
2487     }
2488     return &ms->possible_cpus->cpus[index];
2489 }
2490 
2491 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2492 {
2493     MachineState *ms = MACHINE(spapr);
2494     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2495     Error *local_err = NULL;
2496     bool vsmt_user = !!spapr->vsmt;
2497     int kvm_smt = kvmppc_smt_threads();
2498     int ret;
2499     unsigned int smp_threads = ms->smp.threads;
2500 
2501     if (!kvm_enabled() && (smp_threads > 1)) {
2502         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2503                      "on a pseries machine");
2504         goto out;
2505     }
2506     if (!is_power_of_2(smp_threads)) {
2507         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2508                      "machine because it must be a power of 2", smp_threads);
2509         goto out;
2510     }
2511 
2512     /* Detemine the VSMT mode to use: */
2513     if (vsmt_user) {
2514         if (spapr->vsmt < smp_threads) {
2515             error_setg(&local_err, "Cannot support VSMT mode %d"
2516                          " because it must be >= threads/core (%d)",
2517                          spapr->vsmt, smp_threads);
2518             goto out;
2519         }
2520         /* In this case, spapr->vsmt has been set by the command line */
2521     } else if (!smc->smp_threads_vsmt) {
2522         /*
2523          * Default VSMT value is tricky, because we need it to be as
2524          * consistent as possible (for migration), but this requires
2525          * changing it for at least some existing cases.  We pick 8 as
2526          * the value that we'd get with KVM on POWER8, the
2527          * overwhelmingly common case in production systems.
2528          */
2529         spapr->vsmt = MAX(8, smp_threads);
2530     } else {
2531         spapr->vsmt = smp_threads;
2532     }
2533 
2534     /* KVM: If necessary, set the SMT mode: */
2535     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2536         ret = kvmppc_set_smt_threads(spapr->vsmt);
2537         if (ret) {
2538             /* Looks like KVM isn't able to change VSMT mode */
2539             error_setg(&local_err,
2540                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2541                        spapr->vsmt, ret);
2542             /* We can live with that if the default one is big enough
2543              * for the number of threads, and a submultiple of the one
2544              * we want.  In this case we'll waste some vcpu ids, but
2545              * behaviour will be correct */
2546             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2547                 warn_report_err(local_err);
2548                 local_err = NULL;
2549                 goto out;
2550             } else {
2551                 if (!vsmt_user) {
2552                     error_append_hint(&local_err,
2553                                       "On PPC, a VM with %d threads/core"
2554                                       " on a host with %d threads/core"
2555                                       " requires the use of VSMT mode %d.\n",
2556                                       smp_threads, kvm_smt, spapr->vsmt);
2557                 }
2558                 kvmppc_error_append_smt_possible_hint(&local_err);
2559                 goto out;
2560             }
2561         }
2562     }
2563     /* else TCG: nothing to do currently */
2564 out:
2565     error_propagate(errp, local_err);
2566 }
2567 
2568 static void spapr_init_cpus(SpaprMachineState *spapr)
2569 {
2570     MachineState *machine = MACHINE(spapr);
2571     MachineClass *mc = MACHINE_GET_CLASS(machine);
2572     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2573     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2574     const CPUArchIdList *possible_cpus;
2575     unsigned int smp_cpus = machine->smp.cpus;
2576     unsigned int smp_threads = machine->smp.threads;
2577     unsigned int max_cpus = machine->smp.max_cpus;
2578     int boot_cores_nr = smp_cpus / smp_threads;
2579     int i;
2580 
2581     possible_cpus = mc->possible_cpu_arch_ids(machine);
2582     if (mc->has_hotpluggable_cpus) {
2583         if (smp_cpus % smp_threads) {
2584             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2585                          smp_cpus, smp_threads);
2586             exit(1);
2587         }
2588         if (max_cpus % smp_threads) {
2589             error_report("max_cpus (%u) must be multiple of threads (%u)",
2590                          max_cpus, smp_threads);
2591             exit(1);
2592         }
2593     } else {
2594         if (max_cpus != smp_cpus) {
2595             error_report("This machine version does not support CPU hotplug");
2596             exit(1);
2597         }
2598         boot_cores_nr = possible_cpus->len;
2599     }
2600 
2601     if (smc->pre_2_10_has_unused_icps) {
2602         int i;
2603 
2604         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2605             /* Dummy entries get deregistered when real ICPState objects
2606              * are registered during CPU core hotplug.
2607              */
2608             pre_2_10_vmstate_register_dummy_icp(i);
2609         }
2610     }
2611 
2612     for (i = 0; i < possible_cpus->len; i++) {
2613         int core_id = i * smp_threads;
2614 
2615         if (mc->has_hotpluggable_cpus) {
2616             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2617                                    spapr_vcpu_id(spapr, core_id));
2618         }
2619 
2620         if (i < boot_cores_nr) {
2621             Object *core  = object_new(type);
2622             int nr_threads = smp_threads;
2623 
2624             /* Handle the partially filled core for older machine types */
2625             if ((i + 1) * smp_threads >= smp_cpus) {
2626                 nr_threads = smp_cpus - i * smp_threads;
2627             }
2628 
2629             object_property_set_int(core, nr_threads, "nr-threads",
2630                                     &error_fatal);
2631             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2632                                     &error_fatal);
2633             object_property_set_bool(core, true, "realized", &error_fatal);
2634 
2635             object_unref(core);
2636         }
2637     }
2638 }
2639 
2640 static PCIHostState *spapr_create_default_phb(void)
2641 {
2642     DeviceState *dev;
2643 
2644     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2645     qdev_prop_set_uint32(dev, "index", 0);
2646     qdev_init_nofail(dev);
2647 
2648     return PCI_HOST_BRIDGE(dev);
2649 }
2650 
2651 /* pSeries LPAR / sPAPR hardware init */
2652 static void spapr_machine_init(MachineState *machine)
2653 {
2654     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2655     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2656     MachineClass *mc = MACHINE_GET_CLASS(machine);
2657     const char *kernel_filename = machine->kernel_filename;
2658     const char *initrd_filename = machine->initrd_filename;
2659     PCIHostState *phb;
2660     int i;
2661     MemoryRegion *sysmem = get_system_memory();
2662     hwaddr node0_size = spapr_node0_size(machine);
2663     long load_limit, fw_size;
2664     char *filename;
2665     Error *resize_hpt_err = NULL;
2666 
2667     msi_nonbroken = true;
2668 
2669     QLIST_INIT(&spapr->phbs);
2670     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2671 
2672     /* Determine capabilities to run with */
2673     spapr_caps_init(spapr);
2674 
2675     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2676     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2677         /*
2678          * If the user explicitly requested a mode we should either
2679          * supply it, or fail completely (which we do below).  But if
2680          * it's not set explicitly, we reset our mode to something
2681          * that works
2682          */
2683         if (resize_hpt_err) {
2684             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2685             error_free(resize_hpt_err);
2686             resize_hpt_err = NULL;
2687         } else {
2688             spapr->resize_hpt = smc->resize_hpt_default;
2689         }
2690     }
2691 
2692     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2693 
2694     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2695         /*
2696          * User requested HPT resize, but this host can't supply it.  Bail out
2697          */
2698         error_report_err(resize_hpt_err);
2699         exit(1);
2700     }
2701 
2702     spapr->rma_size = node0_size;
2703 
2704     /* Actually we don't support unbounded RMA anymore since we added
2705      * proper emulation of HV mode. The max we can get is 16G which
2706      * also happens to be what we configure for PAPR mode so make sure
2707      * we don't do anything bigger than that
2708      */
2709     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2710 
2711     if (spapr->rma_size > node0_size) {
2712         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2713                      spapr->rma_size);
2714         exit(1);
2715     }
2716 
2717     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2718     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2719 
2720     /*
2721      * VSMT must be set in order to be able to compute VCPU ids, ie to
2722      * call spapr_max_server_number() or spapr_vcpu_id().
2723      */
2724     spapr_set_vsmt_mode(spapr, &error_fatal);
2725 
2726     /* Set up Interrupt Controller before we create the VCPUs */
2727     spapr_irq_init(spapr, &error_fatal);
2728 
2729     /* Set up containers for ibm,client-architecture-support negotiated options
2730      */
2731     spapr->ov5 = spapr_ovec_new();
2732     spapr->ov5_cas = spapr_ovec_new();
2733 
2734     if (smc->dr_lmb_enabled) {
2735         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2736         spapr_validate_node_memory(machine, &error_fatal);
2737     }
2738 
2739     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2740 
2741     /* advertise support for dedicated HP event source to guests */
2742     if (spapr->use_hotplug_event_source) {
2743         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2744     }
2745 
2746     /* advertise support for HPT resizing */
2747     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2748         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2749     }
2750 
2751     /* advertise support for ibm,dyamic-memory-v2 */
2752     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2753 
2754     /* advertise XIVE on POWER9 machines */
2755     if (spapr->irq->xive) {
2756         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2757     }
2758 
2759     /* init CPUs */
2760     spapr_init_cpus(spapr);
2761 
2762     /*
2763      * check we don't have a memory-less/cpu-less NUMA node
2764      * Firmware relies on the existing memory/cpu topology to provide the
2765      * NUMA topology to the kernel.
2766      * And the linux kernel needs to know the NUMA topology at start
2767      * to be able to hotplug CPUs later.
2768      */
2769     if (machine->numa_state->num_nodes) {
2770         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2771             /* check for memory-less node */
2772             if (machine->numa_state->nodes[i].node_mem == 0) {
2773                 CPUState *cs;
2774                 int found = 0;
2775                 /* check for cpu-less node */
2776                 CPU_FOREACH(cs) {
2777                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2778                     if (cpu->node_id == i) {
2779                         found = 1;
2780                         break;
2781                     }
2782                 }
2783                 /* memory-less and cpu-less node */
2784                 if (!found) {
2785                     error_report(
2786                        "Memory-less/cpu-less nodes are not supported (node %d)",
2787                                  i);
2788                     exit(1);
2789                 }
2790             }
2791         }
2792 
2793     }
2794 
2795     /*
2796      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2797      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2798      * called from vPHB reset handler so we initialize the counter here.
2799      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2800      * must be equally distant from any other node.
2801      * The final value of spapr->gpu_numa_id is going to be written to
2802      * max-associativity-domains in spapr_build_fdt().
2803      */
2804     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2805 
2806     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2807         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2808                               spapr->max_compat_pvr)) {
2809         /* KVM and TCG always allow GTSE with radix... */
2810         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2811     }
2812     /* ... but not with hash (currently). */
2813 
2814     if (kvm_enabled()) {
2815         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2816         kvmppc_enable_logical_ci_hcalls();
2817         kvmppc_enable_set_mode_hcall();
2818 
2819         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2820         kvmppc_enable_clear_ref_mod_hcalls();
2821 
2822         /* Enable H_PAGE_INIT */
2823         kvmppc_enable_h_page_init();
2824     }
2825 
2826     /* map RAM */
2827     memory_region_add_subregion(sysmem, 0, machine->ram);
2828 
2829     /* always allocate the device memory information */
2830     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2831 
2832     /* initialize hotplug memory address space */
2833     if (machine->ram_size < machine->maxram_size) {
2834         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2835         /*
2836          * Limit the number of hotpluggable memory slots to half the number
2837          * slots that KVM supports, leaving the other half for PCI and other
2838          * devices. However ensure that number of slots doesn't drop below 32.
2839          */
2840         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2841                            SPAPR_MAX_RAM_SLOTS;
2842 
2843         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2844             max_memslots = SPAPR_MAX_RAM_SLOTS;
2845         }
2846         if (machine->ram_slots > max_memslots) {
2847             error_report("Specified number of memory slots %"
2848                          PRIu64" exceeds max supported %d",
2849                          machine->ram_slots, max_memslots);
2850             exit(1);
2851         }
2852 
2853         machine->device_memory->base = ROUND_UP(machine->ram_size,
2854                                                 SPAPR_DEVICE_MEM_ALIGN);
2855         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2856                            "device-memory", device_mem_size);
2857         memory_region_add_subregion(sysmem, machine->device_memory->base,
2858                                     &machine->device_memory->mr);
2859     }
2860 
2861     if (smc->dr_lmb_enabled) {
2862         spapr_create_lmb_dr_connectors(spapr);
2863     }
2864 
2865     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI_MCE) == SPAPR_CAP_ON) {
2866         /* Create the error string for live migration blocker */
2867         error_setg(&spapr->fwnmi_migration_blocker,
2868             "A machine check is being handled during migration. The handler"
2869             "may run and log hardware error on the destination");
2870     }
2871 
2872     if (mc->nvdimm_supported) {
2873         spapr_create_nvdimm_dr_connectors(spapr);
2874     }
2875 
2876     /* Set up RTAS event infrastructure */
2877     spapr_events_init(spapr);
2878 
2879     /* Set up the RTC RTAS interfaces */
2880     spapr_rtc_create(spapr);
2881 
2882     /* Set up VIO bus */
2883     spapr->vio_bus = spapr_vio_bus_init();
2884 
2885     for (i = 0; i < serial_max_hds(); i++) {
2886         if (serial_hd(i)) {
2887             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2888         }
2889     }
2890 
2891     /* We always have at least the nvram device on VIO */
2892     spapr_create_nvram(spapr);
2893 
2894     /*
2895      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2896      * connectors (described in root DT node's "ibm,drc-types" property)
2897      * are pre-initialized here. additional child connectors (such as
2898      * connectors for a PHBs PCI slots) are added as needed during their
2899      * parent's realization.
2900      */
2901     if (smc->dr_phb_enabled) {
2902         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2903             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2904         }
2905     }
2906 
2907     /* Set up PCI */
2908     spapr_pci_rtas_init();
2909 
2910     phb = spapr_create_default_phb();
2911 
2912     for (i = 0; i < nb_nics; i++) {
2913         NICInfo *nd = &nd_table[i];
2914 
2915         if (!nd->model) {
2916             nd->model = g_strdup("spapr-vlan");
2917         }
2918 
2919         if (g_str_equal(nd->model, "spapr-vlan") ||
2920             g_str_equal(nd->model, "ibmveth")) {
2921             spapr_vlan_create(spapr->vio_bus, nd);
2922         } else {
2923             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2924         }
2925     }
2926 
2927     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2928         spapr_vscsi_create(spapr->vio_bus);
2929     }
2930 
2931     /* Graphics */
2932     if (spapr_vga_init(phb->bus, &error_fatal)) {
2933         spapr->has_graphics = true;
2934         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2935     }
2936 
2937     if (machine->usb) {
2938         if (smc->use_ohci_by_default) {
2939             pci_create_simple(phb->bus, -1, "pci-ohci");
2940         } else {
2941             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2942         }
2943 
2944         if (spapr->has_graphics) {
2945             USBBus *usb_bus = usb_bus_find(-1);
2946 
2947             usb_create_simple(usb_bus, "usb-kbd");
2948             usb_create_simple(usb_bus, "usb-mouse");
2949         }
2950     }
2951 
2952     if (spapr->rma_size < MIN_RMA_SLOF) {
2953         error_report(
2954             "pSeries SLOF firmware requires >= %ldMiB guest RMA (Real Mode Area memory)",
2955             MIN_RMA_SLOF / MiB);
2956         exit(1);
2957     }
2958 
2959     if (kernel_filename) {
2960         uint64_t lowaddr = 0;
2961 
2962         spapr->kernel_size = load_elf(kernel_filename, NULL,
2963                                       translate_kernel_address, spapr,
2964                                       NULL, &lowaddr, NULL, NULL, 1,
2965                                       PPC_ELF_MACHINE, 0, 0);
2966         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2967             spapr->kernel_size = load_elf(kernel_filename, NULL,
2968                                           translate_kernel_address, spapr, NULL,
2969                                           &lowaddr, NULL, NULL, 0,
2970                                           PPC_ELF_MACHINE,
2971                                           0, 0);
2972             spapr->kernel_le = spapr->kernel_size > 0;
2973         }
2974         if (spapr->kernel_size < 0) {
2975             error_report("error loading %s: %s", kernel_filename,
2976                          load_elf_strerror(spapr->kernel_size));
2977             exit(1);
2978         }
2979 
2980         /* load initrd */
2981         if (initrd_filename) {
2982             /* Try to locate the initrd in the gap between the kernel
2983              * and the firmware. Add a bit of space just in case
2984              */
2985             spapr->initrd_base = (spapr->kernel_addr + spapr->kernel_size
2986                                   + 0x1ffff) & ~0xffff;
2987             spapr->initrd_size = load_image_targphys(initrd_filename,
2988                                                      spapr->initrd_base,
2989                                                      load_limit
2990                                                      - spapr->initrd_base);
2991             if (spapr->initrd_size < 0) {
2992                 error_report("could not load initial ram disk '%s'",
2993                              initrd_filename);
2994                 exit(1);
2995             }
2996         }
2997     }
2998 
2999     if (bios_name == NULL) {
3000         bios_name = FW_FILE_NAME;
3001     }
3002     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3003     if (!filename) {
3004         error_report("Could not find LPAR firmware '%s'", bios_name);
3005         exit(1);
3006     }
3007     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3008     if (fw_size <= 0) {
3009         error_report("Could not load LPAR firmware '%s'", filename);
3010         exit(1);
3011     }
3012     g_free(filename);
3013 
3014     /* FIXME: Should register things through the MachineState's qdev
3015      * interface, this is a legacy from the sPAPREnvironment structure
3016      * which predated MachineState but had a similar function */
3017     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3018     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3019                          &savevm_htab_handlers, spapr);
3020 
3021     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3022                              &error_fatal);
3023 
3024     qemu_register_boot_set(spapr_boot_set, spapr);
3025 
3026     /*
3027      * Nothing needs to be done to resume a suspended guest because
3028      * suspending does not change the machine state, so no need for
3029      * a ->wakeup method.
3030      */
3031     qemu_register_wakeup_support();
3032 
3033     if (kvm_enabled()) {
3034         /* to stop and start vmclock */
3035         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3036                                          &spapr->tb);
3037 
3038         kvmppc_spapr_enable_inkernel_multitce();
3039     }
3040 
3041     qemu_cond_init(&spapr->mc_delivery_cond);
3042 }
3043 
3044 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3045 {
3046     if (!vm_type) {
3047         return 0;
3048     }
3049 
3050     if (!strcmp(vm_type, "HV")) {
3051         return 1;
3052     }
3053 
3054     if (!strcmp(vm_type, "PR")) {
3055         return 2;
3056     }
3057 
3058     error_report("Unknown kvm-type specified '%s'", vm_type);
3059     exit(1);
3060 }
3061 
3062 /*
3063  * Implementation of an interface to adjust firmware path
3064  * for the bootindex property handling.
3065  */
3066 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3067                                    DeviceState *dev)
3068 {
3069 #define CAST(type, obj, name) \
3070     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3071     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3072     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3073     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3074 
3075     if (d) {
3076         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3077         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3078         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3079 
3080         if (spapr) {
3081             /*
3082              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3083              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3084              * 0x8000 | (target << 8) | (bus << 5) | lun
3085              * (see the "Logical unit addressing format" table in SAM5)
3086              */
3087             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3088             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3089                                    (uint64_t)id << 48);
3090         } else if (virtio) {
3091             /*
3092              * We use SRP luns of the form 01000000 | (target << 8) | lun
3093              * in the top 32 bits of the 64-bit LUN
3094              * Note: the quote above is from SLOF and it is wrong,
3095              * the actual binding is:
3096              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3097              */
3098             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3099             if (d->lun >= 256) {
3100                 /* Use the LUN "flat space addressing method" */
3101                 id |= 0x4000;
3102             }
3103             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3104                                    (uint64_t)id << 32);
3105         } else if (usb) {
3106             /*
3107              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3108              * in the top 32 bits of the 64-bit LUN
3109              */
3110             unsigned usb_port = atoi(usb->port->path);
3111             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3112             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3113                                    (uint64_t)id << 32);
3114         }
3115     }
3116 
3117     /*
3118      * SLOF probes the USB devices, and if it recognizes that the device is a
3119      * storage device, it changes its name to "storage" instead of "usb-host",
3120      * and additionally adds a child node for the SCSI LUN, so the correct
3121      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3122      */
3123     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3124         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3125         if (usb_host_dev_is_scsi_storage(usbdev)) {
3126             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3127         }
3128     }
3129 
3130     if (phb) {
3131         /* Replace "pci" with "pci@800000020000000" */
3132         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3133     }
3134 
3135     if (vsc) {
3136         /* Same logic as virtio above */
3137         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3138         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3139     }
3140 
3141     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3142         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3143         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3144         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3145     }
3146 
3147     return NULL;
3148 }
3149 
3150 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3151 {
3152     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3153 
3154     return g_strdup(spapr->kvm_type);
3155 }
3156 
3157 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3158 {
3159     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3160 
3161     g_free(spapr->kvm_type);
3162     spapr->kvm_type = g_strdup(value);
3163 }
3164 
3165 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3166 {
3167     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3168 
3169     return spapr->use_hotplug_event_source;
3170 }
3171 
3172 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3173                                             Error **errp)
3174 {
3175     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3176 
3177     spapr->use_hotplug_event_source = value;
3178 }
3179 
3180 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3181 {
3182     return true;
3183 }
3184 
3185 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3186 {
3187     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3188 
3189     switch (spapr->resize_hpt) {
3190     case SPAPR_RESIZE_HPT_DEFAULT:
3191         return g_strdup("default");
3192     case SPAPR_RESIZE_HPT_DISABLED:
3193         return g_strdup("disabled");
3194     case SPAPR_RESIZE_HPT_ENABLED:
3195         return g_strdup("enabled");
3196     case SPAPR_RESIZE_HPT_REQUIRED:
3197         return g_strdup("required");
3198     }
3199     g_assert_not_reached();
3200 }
3201 
3202 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3203 {
3204     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3205 
3206     if (strcmp(value, "default") == 0) {
3207         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3208     } else if (strcmp(value, "disabled") == 0) {
3209         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3210     } else if (strcmp(value, "enabled") == 0) {
3211         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3212     } else if (strcmp(value, "required") == 0) {
3213         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3214     } else {
3215         error_setg(errp, "Bad value for \"resize-hpt\" property");
3216     }
3217 }
3218 
3219 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3220                                    void *opaque, Error **errp)
3221 {
3222     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3223 }
3224 
3225 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3226                                    void *opaque, Error **errp)
3227 {
3228     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3229 }
3230 
3231 static void spapr_get_kernel_addr(Object *obj, Visitor *v, const char *name,
3232                                   void *opaque, Error **errp)
3233 {
3234     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3235 }
3236 
3237 static void spapr_set_kernel_addr(Object *obj, Visitor *v, const char *name,
3238                                   void *opaque, Error **errp)
3239 {
3240     visit_type_uint64(v, name, (uint64_t *)opaque, errp);
3241 }
3242 
3243 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3244 {
3245     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3246 
3247     if (spapr->irq == &spapr_irq_xics_legacy) {
3248         return g_strdup("legacy");
3249     } else if (spapr->irq == &spapr_irq_xics) {
3250         return g_strdup("xics");
3251     } else if (spapr->irq == &spapr_irq_xive) {
3252         return g_strdup("xive");
3253     } else if (spapr->irq == &spapr_irq_dual) {
3254         return g_strdup("dual");
3255     }
3256     g_assert_not_reached();
3257 }
3258 
3259 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3260 {
3261     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3262 
3263     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3264         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3265         return;
3266     }
3267 
3268     /* The legacy IRQ backend can not be set */
3269     if (strcmp(value, "xics") == 0) {
3270         spapr->irq = &spapr_irq_xics;
3271     } else if (strcmp(value, "xive") == 0) {
3272         spapr->irq = &spapr_irq_xive;
3273     } else if (strcmp(value, "dual") == 0) {
3274         spapr->irq = &spapr_irq_dual;
3275     } else {
3276         error_setg(errp, "Bad value for \"ic-mode\" property");
3277     }
3278 }
3279 
3280 static char *spapr_get_host_model(Object *obj, Error **errp)
3281 {
3282     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3283 
3284     return g_strdup(spapr->host_model);
3285 }
3286 
3287 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3288 {
3289     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3290 
3291     g_free(spapr->host_model);
3292     spapr->host_model = g_strdup(value);
3293 }
3294 
3295 static char *spapr_get_host_serial(Object *obj, Error **errp)
3296 {
3297     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3298 
3299     return g_strdup(spapr->host_serial);
3300 }
3301 
3302 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3303 {
3304     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3305 
3306     g_free(spapr->host_serial);
3307     spapr->host_serial = g_strdup(value);
3308 }
3309 
3310 static void spapr_instance_init(Object *obj)
3311 {
3312     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3313     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3314 
3315     spapr->htab_fd = -1;
3316     spapr->use_hotplug_event_source = true;
3317     object_property_add_str(obj, "kvm-type",
3318                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3319     object_property_set_description(obj, "kvm-type",
3320                                     "Specifies the KVM virtualization mode (HV, PR)",
3321                                     NULL);
3322     object_property_add_bool(obj, "modern-hotplug-events",
3323                             spapr_get_modern_hotplug_events,
3324                             spapr_set_modern_hotplug_events,
3325                             NULL);
3326     object_property_set_description(obj, "modern-hotplug-events",
3327                                     "Use dedicated hotplug event mechanism in"
3328                                     " place of standard EPOW events when possible"
3329                                     " (required for memory hot-unplug support)",
3330                                     NULL);
3331     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3332                             "Maximum permitted CPU compatibility mode",
3333                             &error_fatal);
3334 
3335     object_property_add_str(obj, "resize-hpt",
3336                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3337     object_property_set_description(obj, "resize-hpt",
3338                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3339                                     NULL);
3340     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3341                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3342     object_property_set_description(obj, "vsmt",
3343                                     "Virtual SMT: KVM behaves as if this were"
3344                                     " the host's SMT mode", &error_abort);
3345     object_property_add_bool(obj, "vfio-no-msix-emulation",
3346                              spapr_get_msix_emulation, NULL, NULL);
3347 
3348     object_property_add(obj, "kernel-addr", "uint64", spapr_get_kernel_addr,
3349                         spapr_set_kernel_addr, NULL, &spapr->kernel_addr,
3350                         &error_abort);
3351     object_property_set_description(obj, "kernel-addr",
3352                                     stringify(KERNEL_LOAD_ADDR)
3353                                     " for -kernel is the default",
3354                                     NULL);
3355     spapr->kernel_addr = KERNEL_LOAD_ADDR;
3356     /* The machine class defines the default interrupt controller mode */
3357     spapr->irq = smc->irq;
3358     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3359                             spapr_set_ic_mode, NULL);
3360     object_property_set_description(obj, "ic-mode",
3361                  "Specifies the interrupt controller mode (xics, xive, dual)",
3362                  NULL);
3363 
3364     object_property_add_str(obj, "host-model",
3365         spapr_get_host_model, spapr_set_host_model,
3366         &error_abort);
3367     object_property_set_description(obj, "host-model",
3368         "Host model to advertise in guest device tree", &error_abort);
3369     object_property_add_str(obj, "host-serial",
3370         spapr_get_host_serial, spapr_set_host_serial,
3371         &error_abort);
3372     object_property_set_description(obj, "host-serial",
3373         "Host serial number to advertise in guest device tree", &error_abort);
3374 }
3375 
3376 static void spapr_machine_finalizefn(Object *obj)
3377 {
3378     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3379 
3380     g_free(spapr->kvm_type);
3381 }
3382 
3383 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3384 {
3385     cpu_synchronize_state(cs);
3386     ppc_cpu_do_system_reset(cs);
3387 }
3388 
3389 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3390 {
3391     CPUState *cs;
3392 
3393     CPU_FOREACH(cs) {
3394         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3395     }
3396 }
3397 
3398 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3399                           void *fdt, int *fdt_start_offset, Error **errp)
3400 {
3401     uint64_t addr;
3402     uint32_t node;
3403 
3404     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3405     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3406                                     &error_abort);
3407     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3408                                                    SPAPR_MEMORY_BLOCK_SIZE);
3409     return 0;
3410 }
3411 
3412 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3413                            bool dedicated_hp_event_source, Error **errp)
3414 {
3415     SpaprDrc *drc;
3416     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3417     int i;
3418     uint64_t addr = addr_start;
3419     bool hotplugged = spapr_drc_hotplugged(dev);
3420     Error *local_err = NULL;
3421 
3422     for (i = 0; i < nr_lmbs; i++) {
3423         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3424                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3425         g_assert(drc);
3426 
3427         spapr_drc_attach(drc, dev, &local_err);
3428         if (local_err) {
3429             while (addr > addr_start) {
3430                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3431                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3432                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3433                 spapr_drc_detach(drc);
3434             }
3435             error_propagate(errp, local_err);
3436             return;
3437         }
3438         if (!hotplugged) {
3439             spapr_drc_reset(drc);
3440         }
3441         addr += SPAPR_MEMORY_BLOCK_SIZE;
3442     }
3443     /* send hotplug notification to the
3444      * guest only in case of hotplugged memory
3445      */
3446     if (hotplugged) {
3447         if (dedicated_hp_event_source) {
3448             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3449                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3450             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3451                                                    nr_lmbs,
3452                                                    spapr_drc_index(drc));
3453         } else {
3454             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3455                                            nr_lmbs);
3456         }
3457     }
3458 }
3459 
3460 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3461                               Error **errp)
3462 {
3463     Error *local_err = NULL;
3464     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3465     PCDIMMDevice *dimm = PC_DIMM(dev);
3466     uint64_t size, addr, slot;
3467     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3468 
3469     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3470 
3471     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3472     if (local_err) {
3473         goto out;
3474     }
3475 
3476     if (!is_nvdimm) {
3477         addr = object_property_get_uint(OBJECT(dimm),
3478                                         PC_DIMM_ADDR_PROP, &local_err);
3479         if (local_err) {
3480             goto out_unplug;
3481         }
3482         spapr_add_lmbs(dev, addr, size,
3483                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3484                        &local_err);
3485     } else {
3486         slot = object_property_get_uint(OBJECT(dimm),
3487                                         PC_DIMM_SLOT_PROP, &local_err);
3488         if (local_err) {
3489             goto out_unplug;
3490         }
3491         spapr_add_nvdimm(dev, slot, &local_err);
3492     }
3493 
3494     if (local_err) {
3495         goto out_unplug;
3496     }
3497 
3498     return;
3499 
3500 out_unplug:
3501     pc_dimm_unplug(dimm, MACHINE(ms));
3502 out:
3503     error_propagate(errp, local_err);
3504 }
3505 
3506 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3507                                   Error **errp)
3508 {
3509     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3510     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3511     const MachineClass *mc = MACHINE_CLASS(smc);
3512     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3513     PCDIMMDevice *dimm = PC_DIMM(dev);
3514     Error *local_err = NULL;
3515     uint64_t size;
3516     Object *memdev;
3517     hwaddr pagesize;
3518 
3519     if (!smc->dr_lmb_enabled) {
3520         error_setg(errp, "Memory hotplug not supported for this machine");
3521         return;
3522     }
3523 
3524     if (is_nvdimm && !mc->nvdimm_supported) {
3525         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3526         return;
3527     }
3528 
3529     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3530     if (local_err) {
3531         error_propagate(errp, local_err);
3532         return;
3533     }
3534 
3535     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3536         error_setg(errp, "Hotplugged memory size must be a multiple of "
3537                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3538         return;
3539     } else if (is_nvdimm) {
3540         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3541         if (local_err) {
3542             error_propagate(errp, local_err);
3543             return;
3544         }
3545     }
3546 
3547     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3548                                       &error_abort);
3549     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3550     spapr_check_pagesize(spapr, pagesize, &local_err);
3551     if (local_err) {
3552         error_propagate(errp, local_err);
3553         return;
3554     }
3555 
3556     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3557 }
3558 
3559 struct SpaprDimmState {
3560     PCDIMMDevice *dimm;
3561     uint32_t nr_lmbs;
3562     QTAILQ_ENTRY(SpaprDimmState) next;
3563 };
3564 
3565 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3566                                                        PCDIMMDevice *dimm)
3567 {
3568     SpaprDimmState *dimm_state = NULL;
3569 
3570     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3571         if (dimm_state->dimm == dimm) {
3572             break;
3573         }
3574     }
3575     return dimm_state;
3576 }
3577 
3578 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3579                                                       uint32_t nr_lmbs,
3580                                                       PCDIMMDevice *dimm)
3581 {
3582     SpaprDimmState *ds = NULL;
3583 
3584     /*
3585      * If this request is for a DIMM whose removal had failed earlier
3586      * (due to guest's refusal to remove the LMBs), we would have this
3587      * dimm already in the pending_dimm_unplugs list. In that
3588      * case don't add again.
3589      */
3590     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3591     if (!ds) {
3592         ds = g_malloc0(sizeof(SpaprDimmState));
3593         ds->nr_lmbs = nr_lmbs;
3594         ds->dimm = dimm;
3595         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3596     }
3597     return ds;
3598 }
3599 
3600 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3601                                               SpaprDimmState *dimm_state)
3602 {
3603     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3604     g_free(dimm_state);
3605 }
3606 
3607 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3608                                                         PCDIMMDevice *dimm)
3609 {
3610     SpaprDrc *drc;
3611     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3612                                                   &error_abort);
3613     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3614     uint32_t avail_lmbs = 0;
3615     uint64_t addr_start, addr;
3616     int i;
3617 
3618     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3619                                          &error_abort);
3620 
3621     addr = addr_start;
3622     for (i = 0; i < nr_lmbs; i++) {
3623         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3624                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3625         g_assert(drc);
3626         if (drc->dev) {
3627             avail_lmbs++;
3628         }
3629         addr += SPAPR_MEMORY_BLOCK_SIZE;
3630     }
3631 
3632     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3633 }
3634 
3635 /* Callback to be called during DRC release. */
3636 void spapr_lmb_release(DeviceState *dev)
3637 {
3638     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3639     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3640     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3641 
3642     /* This information will get lost if a migration occurs
3643      * during the unplug process. In this case recover it. */
3644     if (ds == NULL) {
3645         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3646         g_assert(ds);
3647         /* The DRC being examined by the caller at least must be counted */
3648         g_assert(ds->nr_lmbs);
3649     }
3650 
3651     if (--ds->nr_lmbs) {
3652         return;
3653     }
3654 
3655     /*
3656      * Now that all the LMBs have been removed by the guest, call the
3657      * unplug handler chain. This can never fail.
3658      */
3659     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3660     object_unparent(OBJECT(dev));
3661 }
3662 
3663 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3664 {
3665     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3666     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3667 
3668     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3669     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3670     spapr_pending_dimm_unplugs_remove(spapr, ds);
3671 }
3672 
3673 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3674                                         DeviceState *dev, Error **errp)
3675 {
3676     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3677     Error *local_err = NULL;
3678     PCDIMMDevice *dimm = PC_DIMM(dev);
3679     uint32_t nr_lmbs;
3680     uint64_t size, addr_start, addr;
3681     int i;
3682     SpaprDrc *drc;
3683 
3684     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3685         error_setg(&local_err,
3686                    "nvdimm device hot unplug is not supported yet.");
3687         goto out;
3688     }
3689 
3690     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3691     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3692 
3693     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3694                                          &local_err);
3695     if (local_err) {
3696         goto out;
3697     }
3698 
3699     /*
3700      * An existing pending dimm state for this DIMM means that there is an
3701      * unplug operation in progress, waiting for the spapr_lmb_release
3702      * callback to complete the job (BQL can't cover that far). In this case,
3703      * bail out to avoid detaching DRCs that were already released.
3704      */
3705     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3706         error_setg(&local_err,
3707                    "Memory unplug already in progress for device %s",
3708                    dev->id);
3709         goto out;
3710     }
3711 
3712     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3713 
3714     addr = addr_start;
3715     for (i = 0; i < nr_lmbs; i++) {
3716         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3717                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3718         g_assert(drc);
3719 
3720         spapr_drc_detach(drc);
3721         addr += SPAPR_MEMORY_BLOCK_SIZE;
3722     }
3723 
3724     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3725                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3726     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3727                                               nr_lmbs, spapr_drc_index(drc));
3728 out:
3729     error_propagate(errp, local_err);
3730 }
3731 
3732 /* Callback to be called during DRC release. */
3733 void spapr_core_release(DeviceState *dev)
3734 {
3735     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3736 
3737     /* Call the unplug handler chain. This can never fail. */
3738     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3739     object_unparent(OBJECT(dev));
3740 }
3741 
3742 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3743 {
3744     MachineState *ms = MACHINE(hotplug_dev);
3745     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3746     CPUCore *cc = CPU_CORE(dev);
3747     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3748 
3749     if (smc->pre_2_10_has_unused_icps) {
3750         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3751         int i;
3752 
3753         for (i = 0; i < cc->nr_threads; i++) {
3754             CPUState *cs = CPU(sc->threads[i]);
3755 
3756             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3757         }
3758     }
3759 
3760     assert(core_slot);
3761     core_slot->cpu = NULL;
3762     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3763 }
3764 
3765 static
3766 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3767                                Error **errp)
3768 {
3769     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3770     int index;
3771     SpaprDrc *drc;
3772     CPUCore *cc = CPU_CORE(dev);
3773 
3774     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3775         error_setg(errp, "Unable to find CPU core with core-id: %d",
3776                    cc->core_id);
3777         return;
3778     }
3779     if (index == 0) {
3780         error_setg(errp, "Boot CPU core may not be unplugged");
3781         return;
3782     }
3783 
3784     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3785                           spapr_vcpu_id(spapr, cc->core_id));
3786     g_assert(drc);
3787 
3788     if (!spapr_drc_unplug_requested(drc)) {
3789         spapr_drc_detach(drc);
3790         spapr_hotplug_req_remove_by_index(drc);
3791     }
3792 }
3793 
3794 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3795                            void *fdt, int *fdt_start_offset, Error **errp)
3796 {
3797     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3798     CPUState *cs = CPU(core->threads[0]);
3799     PowerPCCPU *cpu = POWERPC_CPU(cs);
3800     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3801     int id = spapr_get_vcpu_id(cpu);
3802     char *nodename;
3803     int offset;
3804 
3805     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3806     offset = fdt_add_subnode(fdt, 0, nodename);
3807     g_free(nodename);
3808 
3809     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3810 
3811     *fdt_start_offset = offset;
3812     return 0;
3813 }
3814 
3815 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3816                             Error **errp)
3817 {
3818     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3819     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3820     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3821     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3822     CPUCore *cc = CPU_CORE(dev);
3823     CPUState *cs;
3824     SpaprDrc *drc;
3825     Error *local_err = NULL;
3826     CPUArchId *core_slot;
3827     int index;
3828     bool hotplugged = spapr_drc_hotplugged(dev);
3829     int i;
3830 
3831     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3832     if (!core_slot) {
3833         error_setg(errp, "Unable to find CPU core with core-id: %d",
3834                    cc->core_id);
3835         return;
3836     }
3837     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3838                           spapr_vcpu_id(spapr, cc->core_id));
3839 
3840     g_assert(drc || !mc->has_hotpluggable_cpus);
3841 
3842     if (drc) {
3843         spapr_drc_attach(drc, dev, &local_err);
3844         if (local_err) {
3845             error_propagate(errp, local_err);
3846             return;
3847         }
3848 
3849         if (hotplugged) {
3850             /*
3851              * Send hotplug notification interrupt to the guest only
3852              * in case of hotplugged CPUs.
3853              */
3854             spapr_hotplug_req_add_by_index(drc);
3855         } else {
3856             spapr_drc_reset(drc);
3857         }
3858     }
3859 
3860     core_slot->cpu = OBJECT(dev);
3861 
3862     if (smc->pre_2_10_has_unused_icps) {
3863         for (i = 0; i < cc->nr_threads; i++) {
3864             cs = CPU(core->threads[i]);
3865             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3866         }
3867     }
3868 
3869     /*
3870      * Set compatibility mode to match the boot CPU, which was either set
3871      * by the machine reset code or by CAS.
3872      */
3873     if (hotplugged) {
3874         for (i = 0; i < cc->nr_threads; i++) {
3875             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3876                            &local_err);
3877             if (local_err) {
3878                 error_propagate(errp, local_err);
3879                 return;
3880             }
3881         }
3882     }
3883 }
3884 
3885 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3886                                 Error **errp)
3887 {
3888     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3889     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3890     Error *local_err = NULL;
3891     CPUCore *cc = CPU_CORE(dev);
3892     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3893     const char *type = object_get_typename(OBJECT(dev));
3894     CPUArchId *core_slot;
3895     int index;
3896     unsigned int smp_threads = machine->smp.threads;
3897 
3898     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3899         error_setg(&local_err, "CPU hotplug not supported for this machine");
3900         goto out;
3901     }
3902 
3903     if (strcmp(base_core_type, type)) {
3904         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3905         goto out;
3906     }
3907 
3908     if (cc->core_id % smp_threads) {
3909         error_setg(&local_err, "invalid core id %d", cc->core_id);
3910         goto out;
3911     }
3912 
3913     /*
3914      * In general we should have homogeneous threads-per-core, but old
3915      * (pre hotplug support) machine types allow the last core to have
3916      * reduced threads as a compatibility hack for when we allowed
3917      * total vcpus not a multiple of threads-per-core.
3918      */
3919     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3920         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3921                    cc->nr_threads, smp_threads);
3922         goto out;
3923     }
3924 
3925     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3926     if (!core_slot) {
3927         error_setg(&local_err, "core id %d out of range", cc->core_id);
3928         goto out;
3929     }
3930 
3931     if (core_slot->cpu) {
3932         error_setg(&local_err, "core %d already populated", cc->core_id);
3933         goto out;
3934     }
3935 
3936     numa_cpu_pre_plug(core_slot, dev, &local_err);
3937 
3938 out:
3939     error_propagate(errp, local_err);
3940 }
3941 
3942 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3943                           void *fdt, int *fdt_start_offset, Error **errp)
3944 {
3945     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3946     int intc_phandle;
3947 
3948     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3949     if (intc_phandle <= 0) {
3950         return -1;
3951     }
3952 
3953     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3954         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3955         return -1;
3956     }
3957 
3958     /* generally SLOF creates these, for hotplug it's up to QEMU */
3959     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3960 
3961     return 0;
3962 }
3963 
3964 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3965                                Error **errp)
3966 {
3967     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3968     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3969     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3970     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3971 
3972     if (dev->hotplugged && !smc->dr_phb_enabled) {
3973         error_setg(errp, "PHB hotplug not supported for this machine");
3974         return;
3975     }
3976 
3977     if (sphb->index == (uint32_t)-1) {
3978         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3979         return;
3980     }
3981 
3982     /*
3983      * This will check that sphb->index doesn't exceed the maximum number of
3984      * PHBs for the current machine type.
3985      */
3986     smc->phb_placement(spapr, sphb->index,
3987                        &sphb->buid, &sphb->io_win_addr,
3988                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3989                        windows_supported, sphb->dma_liobn,
3990                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3991                        errp);
3992 }
3993 
3994 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3995                            Error **errp)
3996 {
3997     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3998     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3999     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4000     SpaprDrc *drc;
4001     bool hotplugged = spapr_drc_hotplugged(dev);
4002     Error *local_err = NULL;
4003 
4004     if (!smc->dr_phb_enabled) {
4005         return;
4006     }
4007 
4008     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4009     /* hotplug hooks should check it's enabled before getting this far */
4010     assert(drc);
4011 
4012     spapr_drc_attach(drc, DEVICE(dev), &local_err);
4013     if (local_err) {
4014         error_propagate(errp, local_err);
4015         return;
4016     }
4017 
4018     if (hotplugged) {
4019         spapr_hotplug_req_add_by_index(drc);
4020     } else {
4021         spapr_drc_reset(drc);
4022     }
4023 }
4024 
4025 void spapr_phb_release(DeviceState *dev)
4026 {
4027     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4028 
4029     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4030     object_unparent(OBJECT(dev));
4031 }
4032 
4033 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4034 {
4035     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4036 }
4037 
4038 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4039                                      DeviceState *dev, Error **errp)
4040 {
4041     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4042     SpaprDrc *drc;
4043 
4044     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4045     assert(drc);
4046 
4047     if (!spapr_drc_unplug_requested(drc)) {
4048         spapr_drc_detach(drc);
4049         spapr_hotplug_req_remove_by_index(drc);
4050     }
4051 }
4052 
4053 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4054                                  Error **errp)
4055 {
4056     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4057     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4058 
4059     if (spapr->tpm_proxy != NULL) {
4060         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4061         return;
4062     }
4063 
4064     spapr->tpm_proxy = tpm_proxy;
4065 }
4066 
4067 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4068 {
4069     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4070 
4071     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4072     object_unparent(OBJECT(dev));
4073     spapr->tpm_proxy = NULL;
4074 }
4075 
4076 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4077                                       DeviceState *dev, Error **errp)
4078 {
4079     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4080         spapr_memory_plug(hotplug_dev, dev, errp);
4081     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4082         spapr_core_plug(hotplug_dev, dev, errp);
4083     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4084         spapr_phb_plug(hotplug_dev, dev, errp);
4085     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4086         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4087     }
4088 }
4089 
4090 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4091                                         DeviceState *dev, Error **errp)
4092 {
4093     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4094         spapr_memory_unplug(hotplug_dev, dev);
4095     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4096         spapr_core_unplug(hotplug_dev, dev);
4097     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4098         spapr_phb_unplug(hotplug_dev, dev);
4099     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4100         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4101     }
4102 }
4103 
4104 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4105                                                 DeviceState *dev, Error **errp)
4106 {
4107     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4108     MachineClass *mc = MACHINE_GET_CLASS(sms);
4109     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4110 
4111     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4112         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4113             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4114         } else {
4115             /* NOTE: this means there is a window after guest reset, prior to
4116              * CAS negotiation, where unplug requests will fail due to the
4117              * capability not being detected yet. This is a bit different than
4118              * the case with PCI unplug, where the events will be queued and
4119              * eventually handled by the guest after boot
4120              */
4121             error_setg(errp, "Memory hot unplug not supported for this guest");
4122         }
4123     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4124         if (!mc->has_hotpluggable_cpus) {
4125             error_setg(errp, "CPU hot unplug not supported on this machine");
4126             return;
4127         }
4128         spapr_core_unplug_request(hotplug_dev, dev, errp);
4129     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4130         if (!smc->dr_phb_enabled) {
4131             error_setg(errp, "PHB hot unplug not supported on this machine");
4132             return;
4133         }
4134         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4135     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4136         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4137     }
4138 }
4139 
4140 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4141                                           DeviceState *dev, Error **errp)
4142 {
4143     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4144         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4145     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4146         spapr_core_pre_plug(hotplug_dev, dev, errp);
4147     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4148         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4149     }
4150 }
4151 
4152 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4153                                                  DeviceState *dev)
4154 {
4155     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4156         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4157         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4158         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4159         return HOTPLUG_HANDLER(machine);
4160     }
4161     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4162         PCIDevice *pcidev = PCI_DEVICE(dev);
4163         PCIBus *root = pci_device_root_bus(pcidev);
4164         SpaprPhbState *phb =
4165             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4166                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4167 
4168         if (phb) {
4169             return HOTPLUG_HANDLER(phb);
4170         }
4171     }
4172     return NULL;
4173 }
4174 
4175 static CpuInstanceProperties
4176 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4177 {
4178     CPUArchId *core_slot;
4179     MachineClass *mc = MACHINE_GET_CLASS(machine);
4180 
4181     /* make sure possible_cpu are intialized */
4182     mc->possible_cpu_arch_ids(machine);
4183     /* get CPU core slot containing thread that matches cpu_index */
4184     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4185     assert(core_slot);
4186     return core_slot->props;
4187 }
4188 
4189 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4190 {
4191     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4192 }
4193 
4194 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4195 {
4196     int i;
4197     unsigned int smp_threads = machine->smp.threads;
4198     unsigned int smp_cpus = machine->smp.cpus;
4199     const char *core_type;
4200     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4201     MachineClass *mc = MACHINE_GET_CLASS(machine);
4202 
4203     if (!mc->has_hotpluggable_cpus) {
4204         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4205     }
4206     if (machine->possible_cpus) {
4207         assert(machine->possible_cpus->len == spapr_max_cores);
4208         return machine->possible_cpus;
4209     }
4210 
4211     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4212     if (!core_type) {
4213         error_report("Unable to find sPAPR CPU Core definition");
4214         exit(1);
4215     }
4216 
4217     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4218                              sizeof(CPUArchId) * spapr_max_cores);
4219     machine->possible_cpus->len = spapr_max_cores;
4220     for (i = 0; i < machine->possible_cpus->len; i++) {
4221         int core_id = i * smp_threads;
4222 
4223         machine->possible_cpus->cpus[i].type = core_type;
4224         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4225         machine->possible_cpus->cpus[i].arch_id = core_id;
4226         machine->possible_cpus->cpus[i].props.has_core_id = true;
4227         machine->possible_cpus->cpus[i].props.core_id = core_id;
4228     }
4229     return machine->possible_cpus;
4230 }
4231 
4232 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4233                                 uint64_t *buid, hwaddr *pio,
4234                                 hwaddr *mmio32, hwaddr *mmio64,
4235                                 unsigned n_dma, uint32_t *liobns,
4236                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4237 {
4238     /*
4239      * New-style PHB window placement.
4240      *
4241      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4242      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4243      * windows.
4244      *
4245      * Some guest kernels can't work with MMIO windows above 1<<46
4246      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4247      *
4248      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4249      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4250      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4251      * 1TiB 64-bit MMIO windows for each PHB.
4252      */
4253     const uint64_t base_buid = 0x800000020000000ULL;
4254     int i;
4255 
4256     /* Sanity check natural alignments */
4257     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4258     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4259     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4260     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4261     /* Sanity check bounds */
4262     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4263                       SPAPR_PCI_MEM32_WIN_SIZE);
4264     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4265                       SPAPR_PCI_MEM64_WIN_SIZE);
4266 
4267     if (index >= SPAPR_MAX_PHBS) {
4268         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4269                    SPAPR_MAX_PHBS - 1);
4270         return;
4271     }
4272 
4273     *buid = base_buid + index;
4274     for (i = 0; i < n_dma; ++i) {
4275         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4276     }
4277 
4278     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4279     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4280     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4281 
4282     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4283     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4284 }
4285 
4286 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4287 {
4288     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4289 
4290     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4291 }
4292 
4293 static void spapr_ics_resend(XICSFabric *dev)
4294 {
4295     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4296 
4297     ics_resend(spapr->ics);
4298 }
4299 
4300 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4301 {
4302     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4303 
4304     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4305 }
4306 
4307 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4308                                  Monitor *mon)
4309 {
4310     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4311 
4312     spapr_irq_print_info(spapr, mon);
4313     monitor_printf(mon, "irqchip: %s\n",
4314                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4315 }
4316 
4317 /*
4318  * This is a XIVE only operation
4319  */
4320 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4321                            uint8_t nvt_blk, uint32_t nvt_idx,
4322                            bool cam_ignore, uint8_t priority,
4323                            uint32_t logic_serv, XiveTCTXMatch *match)
4324 {
4325     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4326     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4327     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4328     int count;
4329 
4330     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4331                            priority, logic_serv, match);
4332     if (count < 0) {
4333         return count;
4334     }
4335 
4336     /*
4337      * When we implement the save and restore of the thread interrupt
4338      * contexts in the enter/exit CPU handlers of the machine and the
4339      * escalations in QEMU, we should be able to handle non dispatched
4340      * vCPUs.
4341      *
4342      * Until this is done, the sPAPR machine should find at least one
4343      * matching context always.
4344      */
4345     if (count == 0) {
4346         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4347                       nvt_blk, nvt_idx);
4348     }
4349 
4350     return count;
4351 }
4352 
4353 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4354 {
4355     return cpu->vcpu_id;
4356 }
4357 
4358 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4359 {
4360     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4361     MachineState *ms = MACHINE(spapr);
4362     int vcpu_id;
4363 
4364     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4365 
4366     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4367         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4368         error_append_hint(errp, "Adjust the number of cpus to %d "
4369                           "or try to raise the number of threads per core\n",
4370                           vcpu_id * ms->smp.threads / spapr->vsmt);
4371         return;
4372     }
4373 
4374     cpu->vcpu_id = vcpu_id;
4375 }
4376 
4377 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4378 {
4379     CPUState *cs;
4380 
4381     CPU_FOREACH(cs) {
4382         PowerPCCPU *cpu = POWERPC_CPU(cs);
4383 
4384         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4385             return cpu;
4386         }
4387     }
4388 
4389     return NULL;
4390 }
4391 
4392 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4393 {
4394     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4395 
4396     /* These are only called by TCG, KVM maintains dispatch state */
4397 
4398     spapr_cpu->prod = false;
4399     if (spapr_cpu->vpa_addr) {
4400         CPUState *cs = CPU(cpu);
4401         uint32_t dispatch;
4402 
4403         dispatch = ldl_be_phys(cs->as,
4404                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4405         dispatch++;
4406         if ((dispatch & 1) != 0) {
4407             qemu_log_mask(LOG_GUEST_ERROR,
4408                           "VPA: incorrect dispatch counter value for "
4409                           "dispatched partition %u, correcting.\n", dispatch);
4410             dispatch++;
4411         }
4412         stl_be_phys(cs->as,
4413                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4414     }
4415 }
4416 
4417 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4418 {
4419     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4420 
4421     if (spapr_cpu->vpa_addr) {
4422         CPUState *cs = CPU(cpu);
4423         uint32_t dispatch;
4424 
4425         dispatch = ldl_be_phys(cs->as,
4426                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4427         dispatch++;
4428         if ((dispatch & 1) != 1) {
4429             qemu_log_mask(LOG_GUEST_ERROR,
4430                           "VPA: incorrect dispatch counter value for "
4431                           "preempted partition %u, correcting.\n", dispatch);
4432             dispatch++;
4433         }
4434         stl_be_phys(cs->as,
4435                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4436     }
4437 }
4438 
4439 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4440 {
4441     MachineClass *mc = MACHINE_CLASS(oc);
4442     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4443     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4444     NMIClass *nc = NMI_CLASS(oc);
4445     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4446     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4447     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4448     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4449     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4450 
4451     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4452     mc->ignore_boot_device_suffixes = true;
4453 
4454     /*
4455      * We set up the default / latest behaviour here.  The class_init
4456      * functions for the specific versioned machine types can override
4457      * these details for backwards compatibility
4458      */
4459     mc->init = spapr_machine_init;
4460     mc->reset = spapr_machine_reset;
4461     mc->block_default_type = IF_SCSI;
4462     mc->max_cpus = 1024;
4463     mc->no_parallel = 1;
4464     mc->default_boot_order = "";
4465     mc->default_ram_size = 512 * MiB;
4466     mc->default_ram_id = "ppc_spapr.ram";
4467     mc->default_display = "std";
4468     mc->kvm_type = spapr_kvm_type;
4469     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4470     mc->pci_allow_0_address = true;
4471     assert(!mc->get_hotplug_handler);
4472     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4473     hc->pre_plug = spapr_machine_device_pre_plug;
4474     hc->plug = spapr_machine_device_plug;
4475     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4476     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4477     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4478     hc->unplug_request = spapr_machine_device_unplug_request;
4479     hc->unplug = spapr_machine_device_unplug;
4480 
4481     smc->dr_lmb_enabled = true;
4482     smc->update_dt_enabled = true;
4483     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4484     mc->has_hotpluggable_cpus = true;
4485     mc->nvdimm_supported = true;
4486     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4487     fwc->get_dev_path = spapr_get_fw_dev_path;
4488     nc->nmi_monitor_handler = spapr_nmi;
4489     smc->phb_placement = spapr_phb_placement;
4490     vhc->hypercall = emulate_spapr_hypercall;
4491     vhc->hpt_mask = spapr_hpt_mask;
4492     vhc->map_hptes = spapr_map_hptes;
4493     vhc->unmap_hptes = spapr_unmap_hptes;
4494     vhc->hpte_set_c = spapr_hpte_set_c;
4495     vhc->hpte_set_r = spapr_hpte_set_r;
4496     vhc->get_pate = spapr_get_pate;
4497     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4498     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4499     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4500     xic->ics_get = spapr_ics_get;
4501     xic->ics_resend = spapr_ics_resend;
4502     xic->icp_get = spapr_icp_get;
4503     ispc->print_info = spapr_pic_print_info;
4504     /* Force NUMA node memory size to be a multiple of
4505      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4506      * in which LMBs are represented and hot-added
4507      */
4508     mc->numa_mem_align_shift = 28;
4509     mc->numa_mem_supported = true;
4510     mc->auto_enable_numa = true;
4511 
4512     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4513     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4514     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4515     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4516     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4517     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4518     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4519     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4520     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4521     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4522     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_ON;
4523     spapr_caps_add_properties(smc, &error_abort);
4524     smc->irq = &spapr_irq_dual;
4525     smc->dr_phb_enabled = true;
4526     smc->linux_pci_probe = true;
4527     smc->smp_threads_vsmt = true;
4528     smc->nr_xirqs = SPAPR_NR_XIRQS;
4529     xfc->match_nvt = spapr_match_nvt;
4530 }
4531 
4532 static const TypeInfo spapr_machine_info = {
4533     .name          = TYPE_SPAPR_MACHINE,
4534     .parent        = TYPE_MACHINE,
4535     .abstract      = true,
4536     .instance_size = sizeof(SpaprMachineState),
4537     .instance_init = spapr_instance_init,
4538     .instance_finalize = spapr_machine_finalizefn,
4539     .class_size    = sizeof(SpaprMachineClass),
4540     .class_init    = spapr_machine_class_init,
4541     .interfaces = (InterfaceInfo[]) {
4542         { TYPE_FW_PATH_PROVIDER },
4543         { TYPE_NMI },
4544         { TYPE_HOTPLUG_HANDLER },
4545         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4546         { TYPE_XICS_FABRIC },
4547         { TYPE_INTERRUPT_STATS_PROVIDER },
4548         { TYPE_XIVE_FABRIC },
4549         { }
4550     },
4551 };
4552 
4553 static void spapr_machine_latest_class_options(MachineClass *mc)
4554 {
4555     mc->alias = "pseries";
4556     mc->is_default = true;
4557 }
4558 
4559 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4560     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4561                                                     void *data)      \
4562     {                                                                \
4563         MachineClass *mc = MACHINE_CLASS(oc);                        \
4564         spapr_machine_##suffix##_class_options(mc);                  \
4565         if (latest) {                                                \
4566             spapr_machine_latest_class_options(mc);                  \
4567         }                                                            \
4568     }                                                                \
4569     static const TypeInfo spapr_machine_##suffix##_info = {          \
4570         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4571         .parent = TYPE_SPAPR_MACHINE,                                \
4572         .class_init = spapr_machine_##suffix##_class_init,           \
4573     };                                                               \
4574     static void spapr_machine_register_##suffix(void)                \
4575     {                                                                \
4576         type_register(&spapr_machine_##suffix##_info);               \
4577     }                                                                \
4578     type_init(spapr_machine_register_##suffix)
4579 
4580 /*
4581  * pseries-5.0
4582  */
4583 static void spapr_machine_5_0_class_options(MachineClass *mc)
4584 {
4585     /* Defaults for the latest behaviour inherited from the base class */
4586 }
4587 
4588 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4589 
4590 /*
4591  * pseries-4.2
4592  */
4593 static void spapr_machine_4_2_class_options(MachineClass *mc)
4594 {
4595     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4596 
4597     spapr_machine_5_0_class_options(mc);
4598     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4599     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4600     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_OFF;
4601     mc->nvdimm_supported = false;
4602 }
4603 
4604 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4605 
4606 /*
4607  * pseries-4.1
4608  */
4609 static void spapr_machine_4_1_class_options(MachineClass *mc)
4610 {
4611     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4612     static GlobalProperty compat[] = {
4613         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4614         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4615     };
4616 
4617     spapr_machine_4_2_class_options(mc);
4618     smc->linux_pci_probe = false;
4619     smc->smp_threads_vsmt = false;
4620     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4621     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4622 }
4623 
4624 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4625 
4626 /*
4627  * pseries-4.0
4628  */
4629 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4630                               uint64_t *buid, hwaddr *pio,
4631                               hwaddr *mmio32, hwaddr *mmio64,
4632                               unsigned n_dma, uint32_t *liobns,
4633                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4634 {
4635     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4636                         nv2gpa, nv2atsd, errp);
4637     *nv2gpa = 0;
4638     *nv2atsd = 0;
4639 }
4640 
4641 static void spapr_machine_4_0_class_options(MachineClass *mc)
4642 {
4643     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4644 
4645     spapr_machine_4_1_class_options(mc);
4646     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4647     smc->phb_placement = phb_placement_4_0;
4648     smc->irq = &spapr_irq_xics;
4649     smc->pre_4_1_migration = true;
4650 }
4651 
4652 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4653 
4654 /*
4655  * pseries-3.1
4656  */
4657 static void spapr_machine_3_1_class_options(MachineClass *mc)
4658 {
4659     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4660 
4661     spapr_machine_4_0_class_options(mc);
4662     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4663 
4664     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4665     smc->update_dt_enabled = false;
4666     smc->dr_phb_enabled = false;
4667     smc->broken_host_serial_model = true;
4668     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4669     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4670     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4671     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4672 }
4673 
4674 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4675 
4676 /*
4677  * pseries-3.0
4678  */
4679 
4680 static void spapr_machine_3_0_class_options(MachineClass *mc)
4681 {
4682     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4683 
4684     spapr_machine_3_1_class_options(mc);
4685     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4686 
4687     smc->legacy_irq_allocation = true;
4688     smc->nr_xirqs = 0x400;
4689     smc->irq = &spapr_irq_xics_legacy;
4690 }
4691 
4692 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4693 
4694 /*
4695  * pseries-2.12
4696  */
4697 static void spapr_machine_2_12_class_options(MachineClass *mc)
4698 {
4699     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4700     static GlobalProperty compat[] = {
4701         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4702         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4703     };
4704 
4705     spapr_machine_3_0_class_options(mc);
4706     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4707     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4708 
4709     /* We depend on kvm_enabled() to choose a default value for the
4710      * hpt-max-page-size capability. Of course we can't do it here
4711      * because this is too early and the HW accelerator isn't initialzed
4712      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4713      */
4714     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4715 }
4716 
4717 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4718 
4719 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4720 {
4721     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4722 
4723     spapr_machine_2_12_class_options(mc);
4724     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4725     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4726     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4727 }
4728 
4729 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4730 
4731 /*
4732  * pseries-2.11
4733  */
4734 
4735 static void spapr_machine_2_11_class_options(MachineClass *mc)
4736 {
4737     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4738 
4739     spapr_machine_2_12_class_options(mc);
4740     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4741     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4742 }
4743 
4744 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4745 
4746 /*
4747  * pseries-2.10
4748  */
4749 
4750 static void spapr_machine_2_10_class_options(MachineClass *mc)
4751 {
4752     spapr_machine_2_11_class_options(mc);
4753     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4754 }
4755 
4756 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4757 
4758 /*
4759  * pseries-2.9
4760  */
4761 
4762 static void spapr_machine_2_9_class_options(MachineClass *mc)
4763 {
4764     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4765     static GlobalProperty compat[] = {
4766         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4767     };
4768 
4769     spapr_machine_2_10_class_options(mc);
4770     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4771     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4772     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4773     smc->pre_2_10_has_unused_icps = true;
4774     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4775 }
4776 
4777 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4778 
4779 /*
4780  * pseries-2.8
4781  */
4782 
4783 static void spapr_machine_2_8_class_options(MachineClass *mc)
4784 {
4785     static GlobalProperty compat[] = {
4786         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4787     };
4788 
4789     spapr_machine_2_9_class_options(mc);
4790     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4791     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4792     mc->numa_mem_align_shift = 23;
4793 }
4794 
4795 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4796 
4797 /*
4798  * pseries-2.7
4799  */
4800 
4801 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4802                               uint64_t *buid, hwaddr *pio,
4803                               hwaddr *mmio32, hwaddr *mmio64,
4804                               unsigned n_dma, uint32_t *liobns,
4805                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4806 {
4807     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4808     const uint64_t base_buid = 0x800000020000000ULL;
4809     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4810     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4811     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4812     const uint32_t max_index = 255;
4813     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4814 
4815     uint64_t ram_top = MACHINE(spapr)->ram_size;
4816     hwaddr phb0_base, phb_base;
4817     int i;
4818 
4819     /* Do we have device memory? */
4820     if (MACHINE(spapr)->maxram_size > ram_top) {
4821         /* Can't just use maxram_size, because there may be an
4822          * alignment gap between normal and device memory regions
4823          */
4824         ram_top = MACHINE(spapr)->device_memory->base +
4825             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4826     }
4827 
4828     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4829 
4830     if (index > max_index) {
4831         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4832                    max_index);
4833         return;
4834     }
4835 
4836     *buid = base_buid + index;
4837     for (i = 0; i < n_dma; ++i) {
4838         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4839     }
4840 
4841     phb_base = phb0_base + index * phb_spacing;
4842     *pio = phb_base + pio_offset;
4843     *mmio32 = phb_base + mmio_offset;
4844     /*
4845      * We don't set the 64-bit MMIO window, relying on the PHB's
4846      * fallback behaviour of automatically splitting a large "32-bit"
4847      * window into contiguous 32-bit and 64-bit windows
4848      */
4849 
4850     *nv2gpa = 0;
4851     *nv2atsd = 0;
4852 }
4853 
4854 static void spapr_machine_2_7_class_options(MachineClass *mc)
4855 {
4856     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4857     static GlobalProperty compat[] = {
4858         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4859         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4860         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4861         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4862     };
4863 
4864     spapr_machine_2_8_class_options(mc);
4865     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4866     mc->default_machine_opts = "modern-hotplug-events=off";
4867     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4868     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4869     smc->phb_placement = phb_placement_2_7;
4870 }
4871 
4872 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4873 
4874 /*
4875  * pseries-2.6
4876  */
4877 
4878 static void spapr_machine_2_6_class_options(MachineClass *mc)
4879 {
4880     static GlobalProperty compat[] = {
4881         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4882     };
4883 
4884     spapr_machine_2_7_class_options(mc);
4885     mc->has_hotpluggable_cpus = false;
4886     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4887     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4888 }
4889 
4890 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4891 
4892 /*
4893  * pseries-2.5
4894  */
4895 
4896 static void spapr_machine_2_5_class_options(MachineClass *mc)
4897 {
4898     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4899     static GlobalProperty compat[] = {
4900         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4901     };
4902 
4903     spapr_machine_2_6_class_options(mc);
4904     smc->use_ohci_by_default = true;
4905     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4906     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4907 }
4908 
4909 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4910 
4911 /*
4912  * pseries-2.4
4913  */
4914 
4915 static void spapr_machine_2_4_class_options(MachineClass *mc)
4916 {
4917     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4918 
4919     spapr_machine_2_5_class_options(mc);
4920     smc->dr_lmb_enabled = false;
4921     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4922 }
4923 
4924 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4925 
4926 /*
4927  * pseries-2.3
4928  */
4929 
4930 static void spapr_machine_2_3_class_options(MachineClass *mc)
4931 {
4932     static GlobalProperty compat[] = {
4933         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4934     };
4935     spapr_machine_2_4_class_options(mc);
4936     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4937     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4938 }
4939 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4940 
4941 /*
4942  * pseries-2.2
4943  */
4944 
4945 static void spapr_machine_2_2_class_options(MachineClass *mc)
4946 {
4947     static GlobalProperty compat[] = {
4948         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4949     };
4950 
4951     spapr_machine_2_3_class_options(mc);
4952     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4953     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4954     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4955 }
4956 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4957 
4958 /*
4959  * pseries-2.1
4960  */
4961 
4962 static void spapr_machine_2_1_class_options(MachineClass *mc)
4963 {
4964     spapr_machine_2_2_class_options(mc);
4965     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4966 }
4967 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4968 
4969 static void spapr_machine_register_types(void)
4970 {
4971     type_register_static(&spapr_machine_info);
4972 }
4973 
4974 type_init(spapr_machine_register_types)
4975