xref: /openbmc/qemu/hw/ppc/spapr.c (revision ee3a71e36654317b14ede0290e87628f8b79f850)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  */
26 
27 #include "qemu/osdep.h"
28 #include "qemu-common.h"
29 #include "qapi/error.h"
30 #include "qapi/visitor.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/hostmem.h"
33 #include "sysemu/numa.h"
34 #include "sysemu/qtest.h"
35 #include "sysemu/reset.h"
36 #include "sysemu/runstate.h"
37 #include "qemu/log.h"
38 #include "hw/fw-path-provider.h"
39 #include "elf.h"
40 #include "net/net.h"
41 #include "sysemu/device_tree.h"
42 #include "sysemu/cpus.h"
43 #include "sysemu/hw_accel.h"
44 #include "kvm_ppc.h"
45 #include "migration/misc.h"
46 #include "migration/qemu-file-types.h"
47 #include "migration/global_state.h"
48 #include "migration/register.h"
49 #include "migration/blocker.h"
50 #include "mmu-hash64.h"
51 #include "mmu-book3s-v3.h"
52 #include "cpu-models.h"
53 #include "hw/core/cpu.h"
54 
55 #include "hw/boards.h"
56 #include "hw/ppc/ppc.h"
57 #include "hw/loader.h"
58 
59 #include "hw/ppc/fdt.h"
60 #include "hw/ppc/spapr.h"
61 #include "hw/ppc/spapr_vio.h"
62 #include "hw/qdev-properties.h"
63 #include "hw/pci-host/spapr.h"
64 #include "hw/pci/msi.h"
65 
66 #include "hw/pci/pci.h"
67 #include "hw/scsi/scsi.h"
68 #include "hw/virtio/virtio-scsi.h"
69 #include "hw/virtio/vhost-scsi-common.h"
70 
71 #include "exec/address-spaces.h"
72 #include "exec/ram_addr.h"
73 #include "hw/usb.h"
74 #include "qemu/config-file.h"
75 #include "qemu/error-report.h"
76 #include "trace.h"
77 #include "hw/nmi.h"
78 #include "hw/intc/intc.h"
79 
80 #include "hw/ppc/spapr_cpu_core.h"
81 #include "hw/mem/memory-device.h"
82 #include "hw/ppc/spapr_tpm_proxy.h"
83 #include "hw/ppc/spapr_nvdimm.h"
84 
85 #include "monitor/monitor.h"
86 
87 #include <libfdt.h>
88 
89 /* SLOF memory layout:
90  *
91  * SLOF raw image loaded at 0, copies its romfs right below the flat
92  * device-tree, then position SLOF itself 31M below that
93  *
94  * So we set FW_OVERHEAD to 40MB which should account for all of that
95  * and more
96  *
97  * We load our kernel at 4M, leaving space for SLOF initial image
98  */
99 #define FDT_MAX_SIZE            0x100000
100 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
101 #define FW_MAX_SIZE             0x400000
102 #define FW_FILE_NAME            "slof.bin"
103 #define FW_OVERHEAD             0x2800000
104 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
105 
106 #define MIN_RMA_SLOF            128UL
107 
108 #define PHANDLE_INTC            0x00001111
109 
110 /* These two functions implement the VCPU id numbering: one to compute them
111  * all and one to identify thread 0 of a VCORE. Any change to the first one
112  * is likely to have an impact on the second one, so let's keep them close.
113  */
114 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
115 {
116     MachineState *ms = MACHINE(spapr);
117     unsigned int smp_threads = ms->smp.threads;
118 
119     assert(spapr->vsmt);
120     return
121         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
122 }
123 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
124                                       PowerPCCPU *cpu)
125 {
126     assert(spapr->vsmt);
127     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 int spapr_max_server_number(SpaprMachineState *spapr)
165 {
166     MachineState *ms = MACHINE(spapr);
167 
168     assert(spapr->vsmt);
169     return DIV_ROUND_UP(ms->smp.max_cpus * spapr->vsmt, ms->smp.threads);
170 }
171 
172 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
173                                   int smt_threads)
174 {
175     int i, ret = 0;
176     uint32_t servers_prop[smt_threads];
177     uint32_t gservers_prop[smt_threads * 2];
178     int index = spapr_get_vcpu_id(cpu);
179 
180     if (cpu->compat_pvr) {
181         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
182         if (ret < 0) {
183             return ret;
184         }
185     }
186 
187     /* Build interrupt servers and gservers properties */
188     for (i = 0; i < smt_threads; i++) {
189         servers_prop[i] = cpu_to_be32(index + i);
190         /* Hack, direct the group queues back to cpu 0 */
191         gservers_prop[i*2] = cpu_to_be32(index + i);
192         gservers_prop[i*2 + 1] = 0;
193     }
194     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
195                       servers_prop, sizeof(servers_prop));
196     if (ret < 0) {
197         return ret;
198     }
199     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
200                       gservers_prop, sizeof(gservers_prop));
201 
202     return ret;
203 }
204 
205 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
206 {
207     int index = spapr_get_vcpu_id(cpu);
208     uint32_t associativity[] = {cpu_to_be32(0x5),
209                                 cpu_to_be32(0x0),
210                                 cpu_to_be32(0x0),
211                                 cpu_to_be32(0x0),
212                                 cpu_to_be32(cpu->node_id),
213                                 cpu_to_be32(index)};
214 
215     /* Advertise NUMA via ibm,associativity */
216     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
217                           sizeof(associativity));
218 }
219 
220 /* Populate the "ibm,pa-features" property */
221 static void spapr_populate_pa_features(SpaprMachineState *spapr,
222                                        PowerPCCPU *cpu,
223                                        void *fdt, int offset)
224 {
225     uint8_t pa_features_206[] = { 6, 0,
226         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
227     uint8_t pa_features_207[] = { 24, 0,
228         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
229         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
230         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
231         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
232     uint8_t pa_features_300[] = { 66, 0,
233         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
234         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
235         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
236         /* 6: DS207 */
237         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
238         /* 16: Vector */
239         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
240         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
241         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
242         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
243         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
244         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
245         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
246         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
247         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
248         /* 42: PM, 44: PC RA, 46: SC vec'd */
249         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
250         /* 48: SIMD, 50: QP BFP, 52: String */
251         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
252         /* 54: DecFP, 56: DecI, 58: SHA */
253         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
254         /* 60: NM atomic, 62: RNG */
255         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
256     };
257     uint8_t *pa_features = NULL;
258     size_t pa_size;
259 
260     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
261         pa_features = pa_features_206;
262         pa_size = sizeof(pa_features_206);
263     }
264     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
265         pa_features = pa_features_207;
266         pa_size = sizeof(pa_features_207);
267     }
268     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
269         pa_features = pa_features_300;
270         pa_size = sizeof(pa_features_300);
271     }
272     if (!pa_features) {
273         return;
274     }
275 
276     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
277         /*
278          * Note: we keep CI large pages off by default because a 64K capable
279          * guest provisioned with large pages might otherwise try to map a qemu
280          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
281          * even if that qemu runs on a 4k host.
282          * We dd this bit back here if we are confident this is not an issue
283          */
284         pa_features[3] |= 0x20;
285     }
286     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
287         pa_features[24] |= 0x80;    /* Transactional memory support */
288     }
289     if (spapr->cas_pre_isa3_guest && pa_size > 40) {
290         /* Workaround for broken kernels that attempt (guest) radix
291          * mode when they can't handle it, if they see the radix bit set
292          * in pa-features. So hide it from them. */
293         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
294     }
295 
296     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
297 }
298 
299 static hwaddr spapr_node0_size(MachineState *machine)
300 {
301     if (machine->numa_state->num_nodes) {
302         int i;
303         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
304             if (machine->numa_state->nodes[i].node_mem) {
305                 return MIN(pow2floor(machine->numa_state->nodes[i].node_mem),
306                            machine->ram_size);
307             }
308         }
309     }
310     return machine->ram_size;
311 }
312 
313 static void add_str(GString *s, const gchar *s1)
314 {
315     g_string_append_len(s, s1, strlen(s1) + 1);
316 }
317 
318 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
319                                        hwaddr size)
320 {
321     uint32_t associativity[] = {
322         cpu_to_be32(0x4), /* length */
323         cpu_to_be32(0x0), cpu_to_be32(0x0),
324         cpu_to_be32(0x0), cpu_to_be32(nodeid)
325     };
326     char mem_name[32];
327     uint64_t mem_reg_property[2];
328     int off;
329 
330     mem_reg_property[0] = cpu_to_be64(start);
331     mem_reg_property[1] = cpu_to_be64(size);
332 
333     sprintf(mem_name, "memory@%" HWADDR_PRIx, start);
334     off = fdt_add_subnode(fdt, 0, mem_name);
335     _FDT(off);
336     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
337     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
338                       sizeof(mem_reg_property))));
339     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
340                       sizeof(associativity))));
341     return off;
342 }
343 
344 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
345 {
346     MachineState *machine = MACHINE(spapr);
347     hwaddr mem_start, node_size;
348     int i, nb_nodes = machine->numa_state->num_nodes;
349     NodeInfo *nodes = machine->numa_state->nodes;
350 
351     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
352         if (!nodes[i].node_mem) {
353             continue;
354         }
355         if (mem_start >= machine->ram_size) {
356             node_size = 0;
357         } else {
358             node_size = nodes[i].node_mem;
359             if (node_size > machine->ram_size - mem_start) {
360                 node_size = machine->ram_size - mem_start;
361             }
362         }
363         if (!mem_start) {
364             /* spapr_machine_init() checks for rma_size <= node0_size
365              * already */
366             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
367             mem_start += spapr->rma_size;
368             node_size -= spapr->rma_size;
369         }
370         for ( ; node_size; ) {
371             hwaddr sizetmp = pow2floor(node_size);
372 
373             /* mem_start != 0 here */
374             if (ctzl(mem_start) < ctzl(sizetmp)) {
375                 sizetmp = 1ULL << ctzl(mem_start);
376             }
377 
378             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
379             node_size -= sizetmp;
380             mem_start += sizetmp;
381         }
382     }
383 
384     return 0;
385 }
386 
387 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
388                                   SpaprMachineState *spapr)
389 {
390     MachineState *ms = MACHINE(spapr);
391     PowerPCCPU *cpu = POWERPC_CPU(cs);
392     CPUPPCState *env = &cpu->env;
393     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
394     int index = spapr_get_vcpu_id(cpu);
395     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
396                        0xffffffff, 0xffffffff};
397     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
398         : SPAPR_TIMEBASE_FREQ;
399     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
400     uint32_t page_sizes_prop[64];
401     size_t page_sizes_prop_size;
402     unsigned int smp_threads = ms->smp.threads;
403     uint32_t vcpus_per_socket = smp_threads * ms->smp.cores;
404     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
405     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
406     SpaprDrc *drc;
407     int drc_index;
408     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
409     int i;
410 
411     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
412     if (drc) {
413         drc_index = spapr_drc_index(drc);
414         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
415     }
416 
417     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
418     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
419 
420     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
421     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
422                            env->dcache_line_size)));
423     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
424                            env->dcache_line_size)));
425     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
426                            env->icache_line_size)));
427     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
428                            env->icache_line_size)));
429 
430     if (pcc->l1_dcache_size) {
431         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
432                                pcc->l1_dcache_size)));
433     } else {
434         warn_report("Unknown L1 dcache size for cpu");
435     }
436     if (pcc->l1_icache_size) {
437         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
438                                pcc->l1_icache_size)));
439     } else {
440         warn_report("Unknown L1 icache size for cpu");
441     }
442 
443     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
444     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
445     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
446     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
447     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
448     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
449 
450     if (env->spr_cb[SPR_PURR].oea_read) {
451         _FDT((fdt_setprop_cell(fdt, offset, "ibm,purr", 1)));
452     }
453     if (env->spr_cb[SPR_SPURR].oea_read) {
454         _FDT((fdt_setprop_cell(fdt, offset, "ibm,spurr", 1)));
455     }
456 
457     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
458         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
459                           segs, sizeof(segs))));
460     }
461 
462     /* Advertise VSX (vector extensions) if available
463      *   1               == VMX / Altivec available
464      *   2               == VSX available
465      *
466      * Only CPUs for which we create core types in spapr_cpu_core.c
467      * are possible, and all of those have VMX */
468     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
469         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
470     } else {
471         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
472     }
473 
474     /* Advertise DFP (Decimal Floating Point) if available
475      *   0 / no property == no DFP
476      *   1               == DFP available */
477     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
478         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
479     }
480 
481     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
482                                                       sizeof(page_sizes_prop));
483     if (page_sizes_prop_size) {
484         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
485                           page_sizes_prop, page_sizes_prop_size)));
486     }
487 
488     spapr_populate_pa_features(spapr, cpu, fdt, offset);
489 
490     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
491                            cs->cpu_index / vcpus_per_socket)));
492 
493     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
494                       pft_size_prop, sizeof(pft_size_prop))));
495 
496     if (ms->numa_state->num_nodes > 1) {
497         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
498     }
499 
500     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
501 
502     if (pcc->radix_page_info) {
503         for (i = 0; i < pcc->radix_page_info->count; i++) {
504             radix_AP_encodings[i] =
505                 cpu_to_be32(pcc->radix_page_info->entries[i]);
506         }
507         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
508                           radix_AP_encodings,
509                           pcc->radix_page_info->count *
510                           sizeof(radix_AP_encodings[0]))));
511     }
512 
513     /*
514      * We set this property to let the guest know that it can use the large
515      * decrementer and its width in bits.
516      */
517     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
518         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
519                               pcc->lrg_decr_bits)));
520 }
521 
522 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
523 {
524     CPUState **rev;
525     CPUState *cs;
526     int n_cpus;
527     int cpus_offset;
528     char *nodename;
529     int i;
530 
531     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
532     _FDT(cpus_offset);
533     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
534     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
535 
536     /*
537      * We walk the CPUs in reverse order to ensure that CPU DT nodes
538      * created by fdt_add_subnode() end up in the right order in FDT
539      * for the guest kernel the enumerate the CPUs correctly.
540      *
541      * The CPU list cannot be traversed in reverse order, so we need
542      * to do extra work.
543      */
544     n_cpus = 0;
545     rev = NULL;
546     CPU_FOREACH(cs) {
547         rev = g_renew(CPUState *, rev, n_cpus + 1);
548         rev[n_cpus++] = cs;
549     }
550 
551     for (i = n_cpus - 1; i >= 0; i--) {
552         CPUState *cs = rev[i];
553         PowerPCCPU *cpu = POWERPC_CPU(cs);
554         int index = spapr_get_vcpu_id(cpu);
555         DeviceClass *dc = DEVICE_GET_CLASS(cs);
556         int offset;
557 
558         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
559             continue;
560         }
561 
562         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
563         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
564         g_free(nodename);
565         _FDT(offset);
566         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
567     }
568 
569     g_free(rev);
570 }
571 
572 static int spapr_rng_populate_dt(void *fdt)
573 {
574     int node;
575     int ret;
576 
577     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
578     if (node <= 0) {
579         return -1;
580     }
581     ret = fdt_setprop_string(fdt, node, "device_type",
582                              "ibm,platform-facilities");
583     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
584     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
585 
586     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
587     if (node <= 0) {
588         return -1;
589     }
590     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
591 
592     return ret ? -1 : 0;
593 }
594 
595 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
596 {
597     MemoryDeviceInfoList *info;
598 
599     for (info = list; info; info = info->next) {
600         MemoryDeviceInfo *value = info->value;
601 
602         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
603             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
604 
605             if (addr >= pcdimm_info->addr &&
606                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
607                 return pcdimm_info->node;
608             }
609         }
610     }
611 
612     return -1;
613 }
614 
615 struct sPAPRDrconfCellV2 {
616      uint32_t seq_lmbs;
617      uint64_t base_addr;
618      uint32_t drc_index;
619      uint32_t aa_index;
620      uint32_t flags;
621 } QEMU_PACKED;
622 
623 typedef struct DrconfCellQueue {
624     struct sPAPRDrconfCellV2 cell;
625     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
626 } DrconfCellQueue;
627 
628 static DrconfCellQueue *
629 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
630                       uint32_t drc_index, uint32_t aa_index,
631                       uint32_t flags)
632 {
633     DrconfCellQueue *elem;
634 
635     elem = g_malloc0(sizeof(*elem));
636     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
637     elem->cell.base_addr = cpu_to_be64(base_addr);
638     elem->cell.drc_index = cpu_to_be32(drc_index);
639     elem->cell.aa_index = cpu_to_be32(aa_index);
640     elem->cell.flags = cpu_to_be32(flags);
641 
642     return elem;
643 }
644 
645 /* ibm,dynamic-memory-v2 */
646 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
647                                    int offset, MemoryDeviceInfoList *dimms)
648 {
649     MachineState *machine = MACHINE(spapr);
650     uint8_t *int_buf, *cur_index;
651     int ret;
652     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
653     uint64_t addr, cur_addr, size;
654     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
655     uint64_t mem_end = machine->device_memory->base +
656                        memory_region_size(&machine->device_memory->mr);
657     uint32_t node, buf_len, nr_entries = 0;
658     SpaprDrc *drc;
659     DrconfCellQueue *elem, *next;
660     MemoryDeviceInfoList *info;
661     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
662         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
663 
664     /* Entry to cover RAM and the gap area */
665     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
666                                  SPAPR_LMB_FLAGS_RESERVED |
667                                  SPAPR_LMB_FLAGS_DRC_INVALID);
668     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
669     nr_entries++;
670 
671     cur_addr = machine->device_memory->base;
672     for (info = dimms; info; info = info->next) {
673         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
674 
675         addr = di->addr;
676         size = di->size;
677         node = di->node;
678 
679         /*
680          * The NVDIMM area is hotpluggable after the NVDIMM is unplugged. The
681          * area is marked hotpluggable in the next iteration for the bigger
682          * chunk including the NVDIMM occupied area.
683          */
684         if (info->value->type == MEMORY_DEVICE_INFO_KIND_NVDIMM)
685             continue;
686 
687         /* Entry for hot-pluggable area */
688         if (cur_addr < addr) {
689             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
690             g_assert(drc);
691             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
692                                          cur_addr, spapr_drc_index(drc), -1, 0);
693             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
694             nr_entries++;
695         }
696 
697         /* Entry for DIMM */
698         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
699         g_assert(drc);
700         elem = spapr_get_drconf_cell(size / lmb_size, addr,
701                                      spapr_drc_index(drc), node,
702                                      SPAPR_LMB_FLAGS_ASSIGNED);
703         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
704         nr_entries++;
705         cur_addr = addr + size;
706     }
707 
708     /* Entry for remaining hotpluggable area */
709     if (cur_addr < mem_end) {
710         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
711         g_assert(drc);
712         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
713                                      cur_addr, spapr_drc_index(drc), -1, 0);
714         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
715         nr_entries++;
716     }
717 
718     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
719     int_buf = cur_index = g_malloc0(buf_len);
720     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
721     cur_index += sizeof(nr_entries);
722 
723     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
724         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
725         cur_index += sizeof(elem->cell);
726         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
727         g_free(elem);
728     }
729 
730     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
731     g_free(int_buf);
732     if (ret < 0) {
733         return -1;
734     }
735     return 0;
736 }
737 
738 /* ibm,dynamic-memory */
739 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
740                                    int offset, MemoryDeviceInfoList *dimms)
741 {
742     MachineState *machine = MACHINE(spapr);
743     int i, ret;
744     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
745     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
746     uint32_t nr_lmbs = (machine->device_memory->base +
747                        memory_region_size(&machine->device_memory->mr)) /
748                        lmb_size;
749     uint32_t *int_buf, *cur_index, buf_len;
750 
751     /*
752      * Allocate enough buffer size to fit in ibm,dynamic-memory
753      */
754     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
755     cur_index = int_buf = g_malloc0(buf_len);
756     int_buf[0] = cpu_to_be32(nr_lmbs);
757     cur_index++;
758     for (i = 0; i < nr_lmbs; i++) {
759         uint64_t addr = i * lmb_size;
760         uint32_t *dynamic_memory = cur_index;
761 
762         if (i >= device_lmb_start) {
763             SpaprDrc *drc;
764 
765             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
766             g_assert(drc);
767 
768             dynamic_memory[0] = cpu_to_be32(addr >> 32);
769             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
770             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
771             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
772             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
773             if (memory_region_present(get_system_memory(), addr)) {
774                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
775             } else {
776                 dynamic_memory[5] = cpu_to_be32(0);
777             }
778         } else {
779             /*
780              * LMB information for RMA, boot time RAM and gap b/n RAM and
781              * device memory region -- all these are marked as reserved
782              * and as having no valid DRC.
783              */
784             dynamic_memory[0] = cpu_to_be32(addr >> 32);
785             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
786             dynamic_memory[2] = cpu_to_be32(0);
787             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
788             dynamic_memory[4] = cpu_to_be32(-1);
789             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
790                                             SPAPR_LMB_FLAGS_DRC_INVALID);
791         }
792 
793         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
794     }
795     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
796     g_free(int_buf);
797     if (ret < 0) {
798         return -1;
799     }
800     return 0;
801 }
802 
803 /*
804  * Adds ibm,dynamic-reconfiguration-memory node.
805  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
806  * of this device tree node.
807  */
808 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
809 {
810     MachineState *machine = MACHINE(spapr);
811     int nb_numa_nodes = machine->numa_state->num_nodes;
812     int ret, i, offset;
813     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
814     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
815     uint32_t *int_buf, *cur_index, buf_len;
816     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
817     MemoryDeviceInfoList *dimms = NULL;
818 
819     /*
820      * Don't create the node if there is no device memory
821      */
822     if (machine->ram_size == machine->maxram_size) {
823         return 0;
824     }
825 
826     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
827 
828     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
829                     sizeof(prop_lmb_size));
830     if (ret < 0) {
831         return ret;
832     }
833 
834     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
835     if (ret < 0) {
836         return ret;
837     }
838 
839     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
840     if (ret < 0) {
841         return ret;
842     }
843 
844     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
845     dimms = qmp_memory_device_list();
846     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
847         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
848     } else {
849         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
850     }
851     qapi_free_MemoryDeviceInfoList(dimms);
852 
853     if (ret < 0) {
854         return ret;
855     }
856 
857     /* ibm,associativity-lookup-arrays */
858     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
859     cur_index = int_buf = g_malloc0(buf_len);
860     int_buf[0] = cpu_to_be32(nr_nodes);
861     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
862     cur_index += 2;
863     for (i = 0; i < nr_nodes; i++) {
864         uint32_t associativity[] = {
865             cpu_to_be32(0x0),
866             cpu_to_be32(0x0),
867             cpu_to_be32(0x0),
868             cpu_to_be32(i)
869         };
870         memcpy(cur_index, associativity, sizeof(associativity));
871         cur_index += 4;
872     }
873     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
874             (cur_index - int_buf) * sizeof(uint32_t));
875     g_free(int_buf);
876 
877     return ret;
878 }
879 
880 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
881                                 SpaprOptionVector *ov5_updates)
882 {
883     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
884     int ret = 0, offset;
885 
886     /* Generate ibm,dynamic-reconfiguration-memory node if required */
887     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
888         g_assert(smc->dr_lmb_enabled);
889         ret = spapr_populate_drconf_memory(spapr, fdt);
890         if (ret) {
891             return ret;
892         }
893     }
894 
895     offset = fdt_path_offset(fdt, "/chosen");
896     if (offset < 0) {
897         offset = fdt_add_subnode(fdt, 0, "chosen");
898         if (offset < 0) {
899             return offset;
900         }
901     }
902     return spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
903                                   "ibm,architecture-vec-5");
904 }
905 
906 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
907 {
908     MachineState *ms = MACHINE(spapr);
909     int rtas;
910     GString *hypertas = g_string_sized_new(256);
911     GString *qemu_hypertas = g_string_sized_new(256);
912     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
913     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
914         memory_region_size(&MACHINE(spapr)->device_memory->mr);
915     uint32_t lrdr_capacity[] = {
916         cpu_to_be32(max_device_addr >> 32),
917         cpu_to_be32(max_device_addr & 0xffffffff),
918         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
919         cpu_to_be32(ms->smp.max_cpus / ms->smp.threads),
920     };
921     uint32_t maxdomain = cpu_to_be32(spapr->gpu_numa_id > 1 ? 1 : 0);
922     uint32_t maxdomains[] = {
923         cpu_to_be32(4),
924         maxdomain,
925         maxdomain,
926         maxdomain,
927         cpu_to_be32(spapr->gpu_numa_id),
928     };
929 
930     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
931 
932     /* hypertas */
933     add_str(hypertas, "hcall-pft");
934     add_str(hypertas, "hcall-term");
935     add_str(hypertas, "hcall-dabr");
936     add_str(hypertas, "hcall-interrupt");
937     add_str(hypertas, "hcall-tce");
938     add_str(hypertas, "hcall-vio");
939     add_str(hypertas, "hcall-splpar");
940     add_str(hypertas, "hcall-join");
941     add_str(hypertas, "hcall-bulk");
942     add_str(hypertas, "hcall-set-mode");
943     add_str(hypertas, "hcall-sprg0");
944     add_str(hypertas, "hcall-copy");
945     add_str(hypertas, "hcall-debug");
946     add_str(hypertas, "hcall-vphn");
947     add_str(qemu_hypertas, "hcall-memop1");
948 
949     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
950         add_str(hypertas, "hcall-multi-tce");
951     }
952 
953     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
954         add_str(hypertas, "hcall-hpt-resize");
955     }
956 
957     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
958                      hypertas->str, hypertas->len));
959     g_string_free(hypertas, TRUE);
960     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
961                      qemu_hypertas->str, qemu_hypertas->len));
962     g_string_free(qemu_hypertas, TRUE);
963 
964     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
965                      refpoints, sizeof(refpoints)));
966 
967     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
968                      maxdomains, sizeof(maxdomains)));
969 
970     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
971                           RTAS_ERROR_LOG_MAX));
972     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
973                           RTAS_EVENT_SCAN_RATE));
974 
975     g_assert(msi_nonbroken);
976     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
977 
978     /*
979      * According to PAPR, rtas ibm,os-term does not guarantee a return
980      * back to the guest cpu.
981      *
982      * While an additional ibm,extended-os-term property indicates
983      * that rtas call return will always occur. Set this property.
984      */
985     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
986 
987     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
988                      lrdr_capacity, sizeof(lrdr_capacity)));
989 
990     spapr_dt_rtas_tokens(fdt, rtas);
991 }
992 
993 /*
994  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
995  * and the XIVE features that the guest may request and thus the valid
996  * values for bytes 23..26 of option vector 5:
997  */
998 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
999                                           int chosen)
1000 {
1001     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1002 
1003     char val[2 * 4] = {
1004         23, 0x00, /* XICS / XIVE mode */
1005         24, 0x00, /* Hash/Radix, filled in below. */
1006         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1007         26, 0x40, /* Radix options: GTSE == yes. */
1008     };
1009 
1010     if (spapr->irq->xics && spapr->irq->xive) {
1011         val[1] = SPAPR_OV5_XIVE_BOTH;
1012     } else if (spapr->irq->xive) {
1013         val[1] = SPAPR_OV5_XIVE_EXPLOIT;
1014     } else {
1015         assert(spapr->irq->xics);
1016         val[1] = SPAPR_OV5_XIVE_LEGACY;
1017     }
1018 
1019     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1020                           first_ppc_cpu->compat_pvr)) {
1021         /*
1022          * If we're in a pre POWER9 compat mode then the guest should
1023          * do hash and use the legacy interrupt mode
1024          */
1025         val[1] = SPAPR_OV5_XIVE_LEGACY; /* XICS */
1026         val[3] = 0x00; /* Hash */
1027     } else if (kvm_enabled()) {
1028         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1029             val[3] = 0x80; /* OV5_MMU_BOTH */
1030         } else if (kvmppc_has_cap_mmu_radix()) {
1031             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1032         } else {
1033             val[3] = 0x00; /* Hash */
1034         }
1035     } else {
1036         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1037         val[3] = 0xC0;
1038     }
1039     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1040                      val, sizeof(val)));
1041 }
1042 
1043 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1044 {
1045     MachineState *machine = MACHINE(spapr);
1046     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1047     int chosen;
1048     const char *boot_device = machine->boot_order;
1049     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1050     size_t cb = 0;
1051     char *bootlist = get_boot_devices_list(&cb);
1052 
1053     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1054 
1055     if (machine->kernel_cmdline && machine->kernel_cmdline[0]) {
1056         _FDT(fdt_setprop_string(fdt, chosen, "bootargs",
1057                                 machine->kernel_cmdline));
1058     }
1059     if (spapr->initrd_size) {
1060         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1061                               spapr->initrd_base));
1062         _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1063                               spapr->initrd_base + spapr->initrd_size));
1064     }
1065 
1066     if (spapr->kernel_size) {
1067         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_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, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1256         }
1257         if (spapr->initrd_size) {
1258             _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base,
1259                                   spapr->initrd_size)));
1260         }
1261     }
1262 
1263     /* ibm,client-architecture-support updates */
1264     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1265     if (ret < 0) {
1266         error_report("couldn't setup CAS properties fdt");
1267         exit(1);
1268     }
1269 
1270     if (smc->dr_phb_enabled) {
1271         ret = spapr_dt_drc(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1272         if (ret < 0) {
1273             error_report("Couldn't set up PHB DR device tree properties");
1274             exit(1);
1275         }
1276     }
1277 
1278     /* NVDIMM devices */
1279     if (mc->nvdimm_supported) {
1280         spapr_dt_persistent_memory(fdt);
1281     }
1282 
1283     return fdt;
1284 }
1285 
1286 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1287 {
1288     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1289 }
1290 
1291 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1292                                     PowerPCCPU *cpu)
1293 {
1294     CPUPPCState *env = &cpu->env;
1295 
1296     /* The TCG path should also be holding the BQL at this point */
1297     g_assert(qemu_mutex_iothread_locked());
1298 
1299     if (msr_pr) {
1300         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1301         env->gpr[3] = H_PRIVILEGE;
1302     } else {
1303         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1304     }
1305 }
1306 
1307 struct LPCRSyncState {
1308     target_ulong value;
1309     target_ulong mask;
1310 };
1311 
1312 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1313 {
1314     struct LPCRSyncState *s = arg.host_ptr;
1315     PowerPCCPU *cpu = POWERPC_CPU(cs);
1316     CPUPPCState *env = &cpu->env;
1317     target_ulong lpcr;
1318 
1319     cpu_synchronize_state(cs);
1320     lpcr = env->spr[SPR_LPCR];
1321     lpcr &= ~s->mask;
1322     lpcr |= s->value;
1323     ppc_store_lpcr(cpu, lpcr);
1324 }
1325 
1326 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1327 {
1328     CPUState *cs;
1329     struct LPCRSyncState s = {
1330         .value = value,
1331         .mask = mask
1332     };
1333     CPU_FOREACH(cs) {
1334         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1335     }
1336 }
1337 
1338 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1339 {
1340     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1341 
1342     /* Copy PATE1:GR into PATE0:HR */
1343     entry->dw0 = spapr->patb_entry & PATE0_HR;
1344     entry->dw1 = spapr->patb_entry;
1345 }
1346 
1347 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1348 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1349 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1350 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1351 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1352 
1353 /*
1354  * Get the fd to access the kernel htab, re-opening it if necessary
1355  */
1356 static int get_htab_fd(SpaprMachineState *spapr)
1357 {
1358     Error *local_err = NULL;
1359 
1360     if (spapr->htab_fd >= 0) {
1361         return spapr->htab_fd;
1362     }
1363 
1364     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1365     if (spapr->htab_fd < 0) {
1366         error_report_err(local_err);
1367     }
1368 
1369     return spapr->htab_fd;
1370 }
1371 
1372 void close_htab_fd(SpaprMachineState *spapr)
1373 {
1374     if (spapr->htab_fd >= 0) {
1375         close(spapr->htab_fd);
1376     }
1377     spapr->htab_fd = -1;
1378 }
1379 
1380 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1381 {
1382     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1383 
1384     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1385 }
1386 
1387 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1388 {
1389     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1390 
1391     assert(kvm_enabled());
1392 
1393     if (!spapr->htab) {
1394         return 0;
1395     }
1396 
1397     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1398 }
1399 
1400 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1401                                                 hwaddr ptex, int n)
1402 {
1403     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1404     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1405 
1406     if (!spapr->htab) {
1407         /*
1408          * HTAB is controlled by KVM. Fetch into temporary buffer
1409          */
1410         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1411         kvmppc_read_hptes(hptes, ptex, n);
1412         return hptes;
1413     }
1414 
1415     /*
1416      * HTAB is controlled by QEMU. Just point to the internally
1417      * accessible PTEG.
1418      */
1419     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1420 }
1421 
1422 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1423                               const ppc_hash_pte64_t *hptes,
1424                               hwaddr ptex, int n)
1425 {
1426     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1427 
1428     if (!spapr->htab) {
1429         g_free((void *)hptes);
1430     }
1431 
1432     /* Nothing to do for qemu managed HPT */
1433 }
1434 
1435 void spapr_store_hpte(PowerPCCPU *cpu, hwaddr ptex,
1436                       uint64_t pte0, uint64_t pte1)
1437 {
1438     SpaprMachineState *spapr = SPAPR_MACHINE(cpu->vhyp);
1439     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1440 
1441     if (!spapr->htab) {
1442         kvmppc_write_hpte(ptex, pte0, pte1);
1443     } else {
1444         if (pte0 & HPTE64_V_VALID) {
1445             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1446             /*
1447              * When setting valid, we write PTE1 first. This ensures
1448              * proper synchronization with the reading code in
1449              * ppc_hash64_pteg_search()
1450              */
1451             smp_wmb();
1452             stq_p(spapr->htab + offset, pte0);
1453         } else {
1454             stq_p(spapr->htab + offset, pte0);
1455             /*
1456              * When clearing it we set PTE0 first. This ensures proper
1457              * synchronization with the reading code in
1458              * ppc_hash64_pteg_search()
1459              */
1460             smp_wmb();
1461             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1462         }
1463     }
1464 }
1465 
1466 static void spapr_hpte_set_c(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1467                              uint64_t pte1)
1468 {
1469     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 15;
1470     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1471 
1472     if (!spapr->htab) {
1473         /* There should always be a hash table when this is called */
1474         error_report("spapr_hpte_set_c called with no hash table !");
1475         return;
1476     }
1477 
1478     /* The HW performs a non-atomic byte update */
1479     stb_p(spapr->htab + offset, (pte1 & 0xff) | 0x80);
1480 }
1481 
1482 static void spapr_hpte_set_r(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1483                              uint64_t pte1)
1484 {
1485     hwaddr offset = ptex * HASH_PTE_SIZE_64 + 14;
1486     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1487 
1488     if (!spapr->htab) {
1489         /* There should always be a hash table when this is called */
1490         error_report("spapr_hpte_set_r called with no hash table !");
1491         return;
1492     }
1493 
1494     /* The HW performs a non-atomic byte update */
1495     stb_p(spapr->htab + offset, ((pte1 >> 8) & 0xff) | 0x01);
1496 }
1497 
1498 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1499 {
1500     int shift;
1501 
1502     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1503      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1504      * that's much more than is needed for Linux guests */
1505     shift = ctz64(pow2ceil(ramsize)) - 7;
1506     shift = MAX(shift, 18); /* Minimum architected size */
1507     shift = MIN(shift, 46); /* Maximum architected size */
1508     return shift;
1509 }
1510 
1511 void spapr_free_hpt(SpaprMachineState *spapr)
1512 {
1513     g_free(spapr->htab);
1514     spapr->htab = NULL;
1515     spapr->htab_shift = 0;
1516     close_htab_fd(spapr);
1517 }
1518 
1519 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1520                           Error **errp)
1521 {
1522     long rc;
1523 
1524     /* Clean up any HPT info from a previous boot */
1525     spapr_free_hpt(spapr);
1526 
1527     rc = kvmppc_reset_htab(shift);
1528     if (rc < 0) {
1529         /* kernel-side HPT needed, but couldn't allocate one */
1530         error_setg_errno(errp, errno,
1531                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1532                          shift);
1533         /* This is almost certainly fatal, but if the caller really
1534          * wants to carry on with shift == 0, it's welcome to try */
1535     } else if (rc > 0) {
1536         /* kernel-side HPT allocated */
1537         if (rc != shift) {
1538             error_setg(errp,
1539                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1540                        shift, rc);
1541         }
1542 
1543         spapr->htab_shift = shift;
1544         spapr->htab = NULL;
1545     } else {
1546         /* kernel-side HPT not needed, allocate in userspace instead */
1547         size_t size = 1ULL << shift;
1548         int i;
1549 
1550         spapr->htab = qemu_memalign(size, size);
1551         if (!spapr->htab) {
1552             error_setg_errno(errp, errno,
1553                              "Could not allocate HPT of order %d", shift);
1554             return;
1555         }
1556 
1557         memset(spapr->htab, 0, size);
1558         spapr->htab_shift = shift;
1559 
1560         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1561             DIRTY_HPTE(HPTE(spapr->htab, i));
1562         }
1563     }
1564     /* We're setting up a hash table, so that means we're not radix */
1565     spapr->patb_entry = 0;
1566     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1567 }
1568 
1569 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1570 {
1571     int hpt_shift;
1572 
1573     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1574         || (spapr->cas_reboot
1575             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1576         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1577     } else {
1578         uint64_t current_ram_size;
1579 
1580         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1581         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1582     }
1583     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1584 
1585     if (spapr->vrma_adjust) {
1586         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1587                                           spapr->htab_shift);
1588     }
1589 }
1590 
1591 static int spapr_reset_drcs(Object *child, void *opaque)
1592 {
1593     SpaprDrc *drc =
1594         (SpaprDrc *) object_dynamic_cast(child,
1595                                                  TYPE_SPAPR_DR_CONNECTOR);
1596 
1597     if (drc) {
1598         spapr_drc_reset(drc);
1599     }
1600 
1601     return 0;
1602 }
1603 
1604 static void spapr_machine_reset(MachineState *machine)
1605 {
1606     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1607     PowerPCCPU *first_ppc_cpu;
1608     hwaddr fdt_addr;
1609     void *fdt;
1610     int rc;
1611 
1612     kvmppc_svm_off(&error_fatal);
1613     spapr_caps_apply(spapr);
1614 
1615     first_ppc_cpu = POWERPC_CPU(first_cpu);
1616     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1617         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1618                               spapr->max_compat_pvr)) {
1619         /*
1620          * If using KVM with radix mode available, VCPUs can be started
1621          * without a HPT because KVM will start them in radix mode.
1622          * Set the GR bit in PATE so that we know there is no HPT.
1623          */
1624         spapr->patb_entry = PATE1_GR;
1625         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1626     } else {
1627         spapr_setup_hpt_and_vrma(spapr);
1628     }
1629 
1630     qemu_devices_reset();
1631 
1632     /*
1633      * If this reset wasn't generated by CAS, we should reset our
1634      * negotiated options and start from scratch
1635      */
1636     if (!spapr->cas_reboot) {
1637         spapr_ovec_cleanup(spapr->ov5_cas);
1638         spapr->ov5_cas = spapr_ovec_new();
1639 
1640         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1641     }
1642 
1643     /*
1644      * This is fixing some of the default configuration of the XIVE
1645      * devices. To be called after the reset of the machine devices.
1646      */
1647     spapr_irq_reset(spapr, &error_fatal);
1648 
1649     /*
1650      * There is no CAS under qtest. Simulate one to please the code that
1651      * depends on spapr->ov5_cas. This is especially needed to test device
1652      * unplug, so we do that before resetting the DRCs.
1653      */
1654     if (qtest_enabled()) {
1655         spapr_ovec_cleanup(spapr->ov5_cas);
1656         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1657     }
1658 
1659     /* DRC reset may cause a device to be unplugged. This will cause troubles
1660      * if this device is used by another device (eg, a running vhost backend
1661      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1662      * situations, we reset DRCs after all devices have been reset.
1663      */
1664     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1665 
1666     spapr_clear_pending_events(spapr);
1667 
1668     /*
1669      * We place the device tree and RTAS just below either the top of the RMA,
1670      * or just below 2GB, whichever is lower, so that it can be
1671      * processed with 32-bit real mode code if necessary
1672      */
1673     fdt_addr = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FDT_MAX_SIZE;
1674 
1675     fdt = spapr_build_fdt(spapr, true, FDT_MAX_SIZE);
1676 
1677     rc = fdt_pack(fdt);
1678 
1679     /* Should only fail if we've built a corrupted tree */
1680     assert(rc == 0);
1681 
1682     /* Load the fdt */
1683     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1684     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1685     g_free(spapr->fdt_blob);
1686     spapr->fdt_size = fdt_totalsize(fdt);
1687     spapr->fdt_initial_size = spapr->fdt_size;
1688     spapr->fdt_blob = fdt;
1689 
1690     /* Set up the entry state */
1691     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1692     first_ppc_cpu->env.gpr[5] = 0;
1693 
1694     spapr->cas_reboot = false;
1695 
1696     spapr->mc_status = -1;
1697     spapr->guest_machine_check_addr = -1;
1698 
1699     /* Signal all vCPUs waiting on this condition */
1700     qemu_cond_broadcast(&spapr->mc_delivery_cond);
1701 
1702     migrate_del_blocker(spapr->fwnmi_migration_blocker);
1703 }
1704 
1705 static void spapr_create_nvram(SpaprMachineState *spapr)
1706 {
1707     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1708     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1709 
1710     if (dinfo) {
1711         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1712                             &error_fatal);
1713     }
1714 
1715     qdev_init_nofail(dev);
1716 
1717     spapr->nvram = (struct SpaprNvram *)dev;
1718 }
1719 
1720 static void spapr_rtc_create(SpaprMachineState *spapr)
1721 {
1722     object_initialize_child(OBJECT(spapr), "rtc",
1723                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1724                             &error_fatal, NULL);
1725     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1726                               &error_fatal);
1727     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1728                               "date", &error_fatal);
1729 }
1730 
1731 /* Returns whether we want to use VGA or not */
1732 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1733 {
1734     switch (vga_interface_type) {
1735     case VGA_NONE:
1736         return false;
1737     case VGA_DEVICE:
1738         return true;
1739     case VGA_STD:
1740     case VGA_VIRTIO:
1741     case VGA_CIRRUS:
1742         return pci_vga_init(pci_bus) != NULL;
1743     default:
1744         error_setg(errp,
1745                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1746         return false;
1747     }
1748 }
1749 
1750 static int spapr_pre_load(void *opaque)
1751 {
1752     int rc;
1753 
1754     rc = spapr_caps_pre_load(opaque);
1755     if (rc) {
1756         return rc;
1757     }
1758 
1759     return 0;
1760 }
1761 
1762 static int spapr_post_load(void *opaque, int version_id)
1763 {
1764     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1765     int err = 0;
1766 
1767     err = spapr_caps_post_migration(spapr);
1768     if (err) {
1769         return err;
1770     }
1771 
1772     /*
1773      * In earlier versions, there was no separate qdev for the PAPR
1774      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1775      * So when migrating from those versions, poke the incoming offset
1776      * value into the RTC device
1777      */
1778     if (version_id < 3) {
1779         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1780         if (err) {
1781             return err;
1782         }
1783     }
1784 
1785     if (kvm_enabled() && spapr->patb_entry) {
1786         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1787         bool radix = !!(spapr->patb_entry & PATE1_GR);
1788         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1789 
1790         /*
1791          * Update LPCR:HR and UPRT as they may not be set properly in
1792          * the stream
1793          */
1794         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1795                             LPCR_HR | LPCR_UPRT);
1796 
1797         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1798         if (err) {
1799             error_report("Process table config unsupported by the host");
1800             return -EINVAL;
1801         }
1802     }
1803 
1804     err = spapr_irq_post_load(spapr, version_id);
1805     if (err) {
1806         return err;
1807     }
1808 
1809     return err;
1810 }
1811 
1812 static int spapr_pre_save(void *opaque)
1813 {
1814     int rc;
1815 
1816     rc = spapr_caps_pre_save(opaque);
1817     if (rc) {
1818         return rc;
1819     }
1820 
1821     return 0;
1822 }
1823 
1824 static bool version_before_3(void *opaque, int version_id)
1825 {
1826     return version_id < 3;
1827 }
1828 
1829 static bool spapr_pending_events_needed(void *opaque)
1830 {
1831     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1832     return !QTAILQ_EMPTY(&spapr->pending_events);
1833 }
1834 
1835 static const VMStateDescription vmstate_spapr_event_entry = {
1836     .name = "spapr_event_log_entry",
1837     .version_id = 1,
1838     .minimum_version_id = 1,
1839     .fields = (VMStateField[]) {
1840         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1841         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1842         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1843                                      NULL, extended_length),
1844         VMSTATE_END_OF_LIST()
1845     },
1846 };
1847 
1848 static const VMStateDescription vmstate_spapr_pending_events = {
1849     .name = "spapr_pending_events",
1850     .version_id = 1,
1851     .minimum_version_id = 1,
1852     .needed = spapr_pending_events_needed,
1853     .fields = (VMStateField[]) {
1854         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1855                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1856         VMSTATE_END_OF_LIST()
1857     },
1858 };
1859 
1860 static bool spapr_ov5_cas_needed(void *opaque)
1861 {
1862     SpaprMachineState *spapr = opaque;
1863     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1864     bool cas_needed;
1865 
1866     /* Prior to the introduction of SpaprOptionVector, we had two option
1867      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1868      * Both of these options encode machine topology into the device-tree
1869      * in such a way that the now-booted OS should still be able to interact
1870      * appropriately with QEMU regardless of what options were actually
1871      * negotiatied on the source side.
1872      *
1873      * As such, we can avoid migrating the CAS-negotiated options if these
1874      * are the only options available on the current machine/platform.
1875      * Since these are the only options available for pseries-2.7 and
1876      * earlier, this allows us to maintain old->new/new->old migration
1877      * compatibility.
1878      *
1879      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1880      * via default pseries-2.8 machines and explicit command-line parameters.
1881      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1882      * of the actual CAS-negotiated values to continue working properly. For
1883      * example, availability of memory unplug depends on knowing whether
1884      * OV5_HP_EVT was negotiated via CAS.
1885      *
1886      * Thus, for any cases where the set of available CAS-negotiatable
1887      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1888      * include the CAS-negotiated options in the migration stream, unless
1889      * if they affect boot time behaviour only.
1890      */
1891     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1892     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1893     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1894 
1895     /* We need extra information if we have any bits outside the mask
1896      * defined above */
1897     cas_needed = !spapr_ovec_subset(spapr->ov5, ov5_mask);
1898 
1899     spapr_ovec_cleanup(ov5_mask);
1900 
1901     return cas_needed;
1902 }
1903 
1904 static const VMStateDescription vmstate_spapr_ov5_cas = {
1905     .name = "spapr_option_vector_ov5_cas",
1906     .version_id = 1,
1907     .minimum_version_id = 1,
1908     .needed = spapr_ov5_cas_needed,
1909     .fields = (VMStateField[]) {
1910         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1911                                  vmstate_spapr_ovec, SpaprOptionVector),
1912         VMSTATE_END_OF_LIST()
1913     },
1914 };
1915 
1916 static bool spapr_patb_entry_needed(void *opaque)
1917 {
1918     SpaprMachineState *spapr = opaque;
1919 
1920     return !!spapr->patb_entry;
1921 }
1922 
1923 static const VMStateDescription vmstate_spapr_patb_entry = {
1924     .name = "spapr_patb_entry",
1925     .version_id = 1,
1926     .minimum_version_id = 1,
1927     .needed = spapr_patb_entry_needed,
1928     .fields = (VMStateField[]) {
1929         VMSTATE_UINT64(patb_entry, SpaprMachineState),
1930         VMSTATE_END_OF_LIST()
1931     },
1932 };
1933 
1934 static bool spapr_irq_map_needed(void *opaque)
1935 {
1936     SpaprMachineState *spapr = opaque;
1937 
1938     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1939 }
1940 
1941 static const VMStateDescription vmstate_spapr_irq_map = {
1942     .name = "spapr_irq_map",
1943     .version_id = 1,
1944     .minimum_version_id = 1,
1945     .needed = spapr_irq_map_needed,
1946     .fields = (VMStateField[]) {
1947         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
1948         VMSTATE_END_OF_LIST()
1949     },
1950 };
1951 
1952 static bool spapr_dtb_needed(void *opaque)
1953 {
1954     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
1955 
1956     return smc->update_dt_enabled;
1957 }
1958 
1959 static int spapr_dtb_pre_load(void *opaque)
1960 {
1961     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1962 
1963     g_free(spapr->fdt_blob);
1964     spapr->fdt_blob = NULL;
1965     spapr->fdt_size = 0;
1966 
1967     return 0;
1968 }
1969 
1970 static const VMStateDescription vmstate_spapr_dtb = {
1971     .name = "spapr_dtb",
1972     .version_id = 1,
1973     .minimum_version_id = 1,
1974     .needed = spapr_dtb_needed,
1975     .pre_load = spapr_dtb_pre_load,
1976     .fields = (VMStateField[]) {
1977         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
1978         VMSTATE_UINT32(fdt_size, SpaprMachineState),
1979         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
1980                                      fdt_size),
1981         VMSTATE_END_OF_LIST()
1982     },
1983 };
1984 
1985 static bool spapr_fwnmi_needed(void *opaque)
1986 {
1987     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1988 
1989     return spapr->guest_machine_check_addr != -1;
1990 }
1991 
1992 static int spapr_fwnmi_pre_save(void *opaque)
1993 {
1994     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1995 
1996     /*
1997      * Check if machine check handling is in progress and print a
1998      * warning message.
1999      */
2000     if (spapr->mc_status != -1) {
2001         warn_report("A machine check is being handled during migration. The"
2002                 "handler may run and log hardware error on the destination");
2003     }
2004 
2005     return 0;
2006 }
2007 
2008 static const VMStateDescription vmstate_spapr_machine_check = {
2009     .name = "spapr_machine_check",
2010     .version_id = 1,
2011     .minimum_version_id = 1,
2012     .needed = spapr_fwnmi_needed,
2013     .pre_save = spapr_fwnmi_pre_save,
2014     .fields = (VMStateField[]) {
2015         VMSTATE_UINT64(guest_machine_check_addr, SpaprMachineState),
2016         VMSTATE_INT32(mc_status, SpaprMachineState),
2017         VMSTATE_END_OF_LIST()
2018     },
2019 };
2020 
2021 static const VMStateDescription vmstate_spapr = {
2022     .name = "spapr",
2023     .version_id = 3,
2024     .minimum_version_id = 1,
2025     .pre_load = spapr_pre_load,
2026     .post_load = spapr_post_load,
2027     .pre_save = spapr_pre_save,
2028     .fields = (VMStateField[]) {
2029         /* used to be @next_irq */
2030         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2031 
2032         /* RTC offset */
2033         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2034 
2035         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2036         VMSTATE_END_OF_LIST()
2037     },
2038     .subsections = (const VMStateDescription*[]) {
2039         &vmstate_spapr_ov5_cas,
2040         &vmstate_spapr_patb_entry,
2041         &vmstate_spapr_pending_events,
2042         &vmstate_spapr_cap_htm,
2043         &vmstate_spapr_cap_vsx,
2044         &vmstate_spapr_cap_dfp,
2045         &vmstate_spapr_cap_cfpc,
2046         &vmstate_spapr_cap_sbbc,
2047         &vmstate_spapr_cap_ibs,
2048         &vmstate_spapr_cap_hpt_maxpagesize,
2049         &vmstate_spapr_irq_map,
2050         &vmstate_spapr_cap_nested_kvm_hv,
2051         &vmstate_spapr_dtb,
2052         &vmstate_spapr_cap_large_decr,
2053         &vmstate_spapr_cap_ccf_assist,
2054         &vmstate_spapr_cap_fwnmi,
2055         &vmstate_spapr_machine_check,
2056         NULL
2057     }
2058 };
2059 
2060 static int htab_save_setup(QEMUFile *f, void *opaque)
2061 {
2062     SpaprMachineState *spapr = opaque;
2063 
2064     /* "Iteration" header */
2065     if (!spapr->htab_shift) {
2066         qemu_put_be32(f, -1);
2067     } else {
2068         qemu_put_be32(f, spapr->htab_shift);
2069     }
2070 
2071     if (spapr->htab) {
2072         spapr->htab_save_index = 0;
2073         spapr->htab_first_pass = true;
2074     } else {
2075         if (spapr->htab_shift) {
2076             assert(kvm_enabled());
2077         }
2078     }
2079 
2080 
2081     return 0;
2082 }
2083 
2084 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2085                             int chunkstart, int n_valid, int n_invalid)
2086 {
2087     qemu_put_be32(f, chunkstart);
2088     qemu_put_be16(f, n_valid);
2089     qemu_put_be16(f, n_invalid);
2090     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2091                     HASH_PTE_SIZE_64 * n_valid);
2092 }
2093 
2094 static void htab_save_end_marker(QEMUFile *f)
2095 {
2096     qemu_put_be32(f, 0);
2097     qemu_put_be16(f, 0);
2098     qemu_put_be16(f, 0);
2099 }
2100 
2101 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2102                                  int64_t max_ns)
2103 {
2104     bool has_timeout = max_ns != -1;
2105     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2106     int index = spapr->htab_save_index;
2107     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2108 
2109     assert(spapr->htab_first_pass);
2110 
2111     do {
2112         int chunkstart;
2113 
2114         /* Consume invalid HPTEs */
2115         while ((index < htabslots)
2116                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2117             CLEAN_HPTE(HPTE(spapr->htab, index));
2118             index++;
2119         }
2120 
2121         /* Consume valid HPTEs */
2122         chunkstart = index;
2123         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2124                && HPTE_VALID(HPTE(spapr->htab, index))) {
2125             CLEAN_HPTE(HPTE(spapr->htab, index));
2126             index++;
2127         }
2128 
2129         if (index > chunkstart) {
2130             int n_valid = index - chunkstart;
2131 
2132             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2133 
2134             if (has_timeout &&
2135                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2136                 break;
2137             }
2138         }
2139     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2140 
2141     if (index >= htabslots) {
2142         assert(index == htabslots);
2143         index = 0;
2144         spapr->htab_first_pass = false;
2145     }
2146     spapr->htab_save_index = index;
2147 }
2148 
2149 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2150                                 int64_t max_ns)
2151 {
2152     bool final = max_ns < 0;
2153     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2154     int examined = 0, sent = 0;
2155     int index = spapr->htab_save_index;
2156     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2157 
2158     assert(!spapr->htab_first_pass);
2159 
2160     do {
2161         int chunkstart, invalidstart;
2162 
2163         /* Consume non-dirty HPTEs */
2164         while ((index < htabslots)
2165                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2166             index++;
2167             examined++;
2168         }
2169 
2170         chunkstart = index;
2171         /* Consume valid dirty HPTEs */
2172         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2173                && HPTE_DIRTY(HPTE(spapr->htab, index))
2174                && HPTE_VALID(HPTE(spapr->htab, index))) {
2175             CLEAN_HPTE(HPTE(spapr->htab, index));
2176             index++;
2177             examined++;
2178         }
2179 
2180         invalidstart = index;
2181         /* Consume invalid dirty HPTEs */
2182         while ((index < htabslots) && (index - invalidstart < 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         if (index > chunkstart) {
2191             int n_valid = invalidstart - chunkstart;
2192             int n_invalid = index - invalidstart;
2193 
2194             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2195             sent += index - chunkstart;
2196 
2197             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2198                 break;
2199             }
2200         }
2201 
2202         if (examined >= htabslots) {
2203             break;
2204         }
2205 
2206         if (index >= htabslots) {
2207             assert(index == htabslots);
2208             index = 0;
2209         }
2210     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2211 
2212     if (index >= htabslots) {
2213         assert(index == htabslots);
2214         index = 0;
2215     }
2216 
2217     spapr->htab_save_index = index;
2218 
2219     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2220 }
2221 
2222 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2223 #define MAX_KVM_BUF_SIZE    2048
2224 
2225 static int htab_save_iterate(QEMUFile *f, void *opaque)
2226 {
2227     SpaprMachineState *spapr = opaque;
2228     int fd;
2229     int rc = 0;
2230 
2231     /* Iteration header */
2232     if (!spapr->htab_shift) {
2233         qemu_put_be32(f, -1);
2234         return 1;
2235     } else {
2236         qemu_put_be32(f, 0);
2237     }
2238 
2239     if (!spapr->htab) {
2240         assert(kvm_enabled());
2241 
2242         fd = get_htab_fd(spapr);
2243         if (fd < 0) {
2244             return fd;
2245         }
2246 
2247         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2248         if (rc < 0) {
2249             return rc;
2250         }
2251     } else  if (spapr->htab_first_pass) {
2252         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2253     } else {
2254         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2255     }
2256 
2257     htab_save_end_marker(f);
2258 
2259     return rc;
2260 }
2261 
2262 static int htab_save_complete(QEMUFile *f, void *opaque)
2263 {
2264     SpaprMachineState *spapr = opaque;
2265     int fd;
2266 
2267     /* Iteration header */
2268     if (!spapr->htab_shift) {
2269         qemu_put_be32(f, -1);
2270         return 0;
2271     } else {
2272         qemu_put_be32(f, 0);
2273     }
2274 
2275     if (!spapr->htab) {
2276         int rc;
2277 
2278         assert(kvm_enabled());
2279 
2280         fd = get_htab_fd(spapr);
2281         if (fd < 0) {
2282             return fd;
2283         }
2284 
2285         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2286         if (rc < 0) {
2287             return rc;
2288         }
2289     } else {
2290         if (spapr->htab_first_pass) {
2291             htab_save_first_pass(f, spapr, -1);
2292         }
2293         htab_save_later_pass(f, spapr, -1);
2294     }
2295 
2296     /* End marker */
2297     htab_save_end_marker(f);
2298 
2299     return 0;
2300 }
2301 
2302 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2303 {
2304     SpaprMachineState *spapr = opaque;
2305     uint32_t section_hdr;
2306     int fd = -1;
2307     Error *local_err = NULL;
2308 
2309     if (version_id < 1 || version_id > 1) {
2310         error_report("htab_load() bad version");
2311         return -EINVAL;
2312     }
2313 
2314     section_hdr = qemu_get_be32(f);
2315 
2316     if (section_hdr == -1) {
2317         spapr_free_hpt(spapr);
2318         return 0;
2319     }
2320 
2321     if (section_hdr) {
2322         /* First section gives the htab size */
2323         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2324         if (local_err) {
2325             error_report_err(local_err);
2326             return -EINVAL;
2327         }
2328         return 0;
2329     }
2330 
2331     if (!spapr->htab) {
2332         assert(kvm_enabled());
2333 
2334         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2335         if (fd < 0) {
2336             error_report_err(local_err);
2337             return fd;
2338         }
2339     }
2340 
2341     while (true) {
2342         uint32_t index;
2343         uint16_t n_valid, n_invalid;
2344 
2345         index = qemu_get_be32(f);
2346         n_valid = qemu_get_be16(f);
2347         n_invalid = qemu_get_be16(f);
2348 
2349         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2350             /* End of Stream */
2351             break;
2352         }
2353 
2354         if ((index + n_valid + n_invalid) >
2355             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2356             /* Bad index in stream */
2357             error_report(
2358                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2359                 index, n_valid, n_invalid, spapr->htab_shift);
2360             return -EINVAL;
2361         }
2362 
2363         if (spapr->htab) {
2364             if (n_valid) {
2365                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2366                                 HASH_PTE_SIZE_64 * n_valid);
2367             }
2368             if (n_invalid) {
2369                 memset(HPTE(spapr->htab, index + n_valid), 0,
2370                        HASH_PTE_SIZE_64 * n_invalid);
2371             }
2372         } else {
2373             int rc;
2374 
2375             assert(fd >= 0);
2376 
2377             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2378             if (rc < 0) {
2379                 return rc;
2380             }
2381         }
2382     }
2383 
2384     if (!spapr->htab) {
2385         assert(fd >= 0);
2386         close(fd);
2387     }
2388 
2389     return 0;
2390 }
2391 
2392 static void htab_save_cleanup(void *opaque)
2393 {
2394     SpaprMachineState *spapr = opaque;
2395 
2396     close_htab_fd(spapr);
2397 }
2398 
2399 static SaveVMHandlers savevm_htab_handlers = {
2400     .save_setup = htab_save_setup,
2401     .save_live_iterate = htab_save_iterate,
2402     .save_live_complete_precopy = htab_save_complete,
2403     .save_cleanup = htab_save_cleanup,
2404     .load_state = htab_load,
2405 };
2406 
2407 static void spapr_boot_set(void *opaque, const char *boot_device,
2408                            Error **errp)
2409 {
2410     MachineState *machine = MACHINE(opaque);
2411     machine->boot_order = g_strdup(boot_device);
2412 }
2413 
2414 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2415 {
2416     MachineState *machine = MACHINE(spapr);
2417     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2418     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2419     int i;
2420 
2421     for (i = 0; i < nr_lmbs; i++) {
2422         uint64_t addr;
2423 
2424         addr = i * lmb_size + machine->device_memory->base;
2425         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2426                                addr / lmb_size);
2427     }
2428 }
2429 
2430 /*
2431  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2432  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2433  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2434  */
2435 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2436 {
2437     int i;
2438 
2439     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2440         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2441                    " is not aligned to %" PRIu64 " MiB",
2442                    machine->ram_size,
2443                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2444         return;
2445     }
2446 
2447     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2448         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2449                    " is not aligned to %" PRIu64 " MiB",
2450                    machine->ram_size,
2451                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2452         return;
2453     }
2454 
2455     for (i = 0; i < machine->numa_state->num_nodes; i++) {
2456         if (machine->numa_state->nodes[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2457             error_setg(errp,
2458                        "Node %d memory size 0x%" PRIx64
2459                        " is not aligned to %" PRIu64 " MiB",
2460                        i, machine->numa_state->nodes[i].node_mem,
2461                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2462             return;
2463         }
2464     }
2465 }
2466 
2467 /* find cpu slot in machine->possible_cpus by core_id */
2468 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2469 {
2470     int index = id / ms->smp.threads;
2471 
2472     if (index >= ms->possible_cpus->len) {
2473         return NULL;
2474     }
2475     if (idx) {
2476         *idx = index;
2477     }
2478     return &ms->possible_cpus->cpus[index];
2479 }
2480 
2481 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2482 {
2483     MachineState *ms = MACHINE(spapr);
2484     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
2485     Error *local_err = NULL;
2486     bool vsmt_user = !!spapr->vsmt;
2487     int kvm_smt = kvmppc_smt_threads();
2488     int ret;
2489     unsigned int smp_threads = ms->smp.threads;
2490 
2491     if (!kvm_enabled() && (smp_threads > 1)) {
2492         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2493                      "on a pseries machine");
2494         goto out;
2495     }
2496     if (!is_power_of_2(smp_threads)) {
2497         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2498                      "machine because it must be a power of 2", smp_threads);
2499         goto out;
2500     }
2501 
2502     /* Detemine the VSMT mode to use: */
2503     if (vsmt_user) {
2504         if (spapr->vsmt < smp_threads) {
2505             error_setg(&local_err, "Cannot support VSMT mode %d"
2506                          " because it must be >= threads/core (%d)",
2507                          spapr->vsmt, smp_threads);
2508             goto out;
2509         }
2510         /* In this case, spapr->vsmt has been set by the command line */
2511     } else if (!smc->smp_threads_vsmt) {
2512         /*
2513          * Default VSMT value is tricky, because we need it to be as
2514          * consistent as possible (for migration), but this requires
2515          * changing it for at least some existing cases.  We pick 8 as
2516          * the value that we'd get with KVM on POWER8, the
2517          * overwhelmingly common case in production systems.
2518          */
2519         spapr->vsmt = MAX(8, smp_threads);
2520     } else {
2521         spapr->vsmt = smp_threads;
2522     }
2523 
2524     /* KVM: If necessary, set the SMT mode: */
2525     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2526         ret = kvmppc_set_smt_threads(spapr->vsmt);
2527         if (ret) {
2528             /* Looks like KVM isn't able to change VSMT mode */
2529             error_setg(&local_err,
2530                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2531                        spapr->vsmt, ret);
2532             /* We can live with that if the default one is big enough
2533              * for the number of threads, and a submultiple of the one
2534              * we want.  In this case we'll waste some vcpu ids, but
2535              * behaviour will be correct */
2536             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2537                 warn_report_err(local_err);
2538                 local_err = NULL;
2539                 goto out;
2540             } else {
2541                 if (!vsmt_user) {
2542                     error_append_hint(&local_err,
2543                                       "On PPC, a VM with %d threads/core"
2544                                       " on a host with %d threads/core"
2545                                       " requires the use of VSMT mode %d.\n",
2546                                       smp_threads, kvm_smt, spapr->vsmt);
2547                 }
2548                 kvmppc_error_append_smt_possible_hint(&local_err);
2549                 goto out;
2550             }
2551         }
2552     }
2553     /* else TCG: nothing to do currently */
2554 out:
2555     error_propagate(errp, local_err);
2556 }
2557 
2558 static void spapr_init_cpus(SpaprMachineState *spapr)
2559 {
2560     MachineState *machine = MACHINE(spapr);
2561     MachineClass *mc = MACHINE_GET_CLASS(machine);
2562     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2563     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2564     const CPUArchIdList *possible_cpus;
2565     unsigned int smp_cpus = machine->smp.cpus;
2566     unsigned int smp_threads = machine->smp.threads;
2567     unsigned int max_cpus = machine->smp.max_cpus;
2568     int boot_cores_nr = smp_cpus / smp_threads;
2569     int i;
2570 
2571     possible_cpus = mc->possible_cpu_arch_ids(machine);
2572     if (mc->has_hotpluggable_cpus) {
2573         if (smp_cpus % smp_threads) {
2574             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2575                          smp_cpus, smp_threads);
2576             exit(1);
2577         }
2578         if (max_cpus % smp_threads) {
2579             error_report("max_cpus (%u) must be multiple of threads (%u)",
2580                          max_cpus, smp_threads);
2581             exit(1);
2582         }
2583     } else {
2584         if (max_cpus != smp_cpus) {
2585             error_report("This machine version does not support CPU hotplug");
2586             exit(1);
2587         }
2588         boot_cores_nr = possible_cpus->len;
2589     }
2590 
2591     if (smc->pre_2_10_has_unused_icps) {
2592         int i;
2593 
2594         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2595             /* Dummy entries get deregistered when real ICPState objects
2596              * are registered during CPU core hotplug.
2597              */
2598             pre_2_10_vmstate_register_dummy_icp(i);
2599         }
2600     }
2601 
2602     for (i = 0; i < possible_cpus->len; i++) {
2603         int core_id = i * smp_threads;
2604 
2605         if (mc->has_hotpluggable_cpus) {
2606             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2607                                    spapr_vcpu_id(spapr, core_id));
2608         }
2609 
2610         if (i < boot_cores_nr) {
2611             Object *core  = object_new(type);
2612             int nr_threads = smp_threads;
2613 
2614             /* Handle the partially filled core for older machine types */
2615             if ((i + 1) * smp_threads >= smp_cpus) {
2616                 nr_threads = smp_cpus - i * smp_threads;
2617             }
2618 
2619             object_property_set_int(core, nr_threads, "nr-threads",
2620                                     &error_fatal);
2621             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2622                                     &error_fatal);
2623             object_property_set_bool(core, true, "realized", &error_fatal);
2624 
2625             object_unref(core);
2626         }
2627     }
2628 }
2629 
2630 static PCIHostState *spapr_create_default_phb(void)
2631 {
2632     DeviceState *dev;
2633 
2634     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2635     qdev_prop_set_uint32(dev, "index", 0);
2636     qdev_init_nofail(dev);
2637 
2638     return PCI_HOST_BRIDGE(dev);
2639 }
2640 
2641 /* pSeries LPAR / sPAPR hardware init */
2642 static void spapr_machine_init(MachineState *machine)
2643 {
2644     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2645     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2646     MachineClass *mc = MACHINE_GET_CLASS(machine);
2647     const char *kernel_filename = machine->kernel_filename;
2648     const char *initrd_filename = machine->initrd_filename;
2649     PCIHostState *phb;
2650     int i;
2651     MemoryRegion *sysmem = get_system_memory();
2652     MemoryRegion *ram = g_new(MemoryRegion, 1);
2653     hwaddr node0_size = spapr_node0_size(machine);
2654     long load_limit, fw_size;
2655     char *filename;
2656     Error *resize_hpt_err = NULL;
2657 
2658     msi_nonbroken = true;
2659 
2660     QLIST_INIT(&spapr->phbs);
2661     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2662 
2663     /* Determine capabilities to run with */
2664     spapr_caps_init(spapr);
2665 
2666     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2667     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2668         /*
2669          * If the user explicitly requested a mode we should either
2670          * supply it, or fail completely (which we do below).  But if
2671          * it's not set explicitly, we reset our mode to something
2672          * that works
2673          */
2674         if (resize_hpt_err) {
2675             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2676             error_free(resize_hpt_err);
2677             resize_hpt_err = NULL;
2678         } else {
2679             spapr->resize_hpt = smc->resize_hpt_default;
2680         }
2681     }
2682 
2683     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2684 
2685     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2686         /*
2687          * User requested HPT resize, but this host can't supply it.  Bail out
2688          */
2689         error_report_err(resize_hpt_err);
2690         exit(1);
2691     }
2692 
2693     spapr->rma_size = node0_size;
2694 
2695     /* With KVM, we don't actually know whether KVM supports an
2696      * unbounded RMA (PR KVM) or is limited by the hash table size
2697      * (HV KVM using VRMA), so we always assume the latter
2698      *
2699      * In that case, we also limit the initial allocations for RTAS
2700      * etc... to 256M since we have no way to know what the VRMA size
2701      * is going to be as it depends on the size of the hash table
2702      * which isn't determined yet.
2703      */
2704     if (kvm_enabled()) {
2705         spapr->vrma_adjust = 1;
2706         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2707     }
2708 
2709     /* Actually we don't support unbounded RMA anymore since we added
2710      * proper emulation of HV mode. The max we can get is 16G which
2711      * also happens to be what we configure for PAPR mode so make sure
2712      * we don't do anything bigger than that
2713      */
2714     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2715 
2716     if (spapr->rma_size > node0_size) {
2717         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2718                      spapr->rma_size);
2719         exit(1);
2720     }
2721 
2722     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2723     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2724 
2725     /*
2726      * VSMT must be set in order to be able to compute VCPU ids, ie to
2727      * call spapr_max_server_number() or spapr_vcpu_id().
2728      */
2729     spapr_set_vsmt_mode(spapr, &error_fatal);
2730 
2731     /* Set up Interrupt Controller before we create the VCPUs */
2732     spapr_irq_init(spapr, &error_fatal);
2733 
2734     /* Set up containers for ibm,client-architecture-support negotiated options
2735      */
2736     spapr->ov5 = spapr_ovec_new();
2737     spapr->ov5_cas = spapr_ovec_new();
2738 
2739     if (smc->dr_lmb_enabled) {
2740         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2741         spapr_validate_node_memory(machine, &error_fatal);
2742     }
2743 
2744     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2745 
2746     /* advertise support for dedicated HP event source to guests */
2747     if (spapr->use_hotplug_event_source) {
2748         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2749     }
2750 
2751     /* advertise support for HPT resizing */
2752     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2753         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2754     }
2755 
2756     /* advertise support for ibm,dyamic-memory-v2 */
2757     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2758 
2759     /* advertise XIVE on POWER9 machines */
2760     if (spapr->irq->xive) {
2761         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2762     }
2763 
2764     /* init CPUs */
2765     spapr_init_cpus(spapr);
2766 
2767     /*
2768      * check we don't have a memory-less/cpu-less NUMA node
2769      * Firmware relies on the existing memory/cpu topology to provide the
2770      * NUMA topology to the kernel.
2771      * And the linux kernel needs to know the NUMA topology at start
2772      * to be able to hotplug CPUs later.
2773      */
2774     if (machine->numa_state->num_nodes) {
2775         for (i = 0; i < machine->numa_state->num_nodes; ++i) {
2776             /* check for memory-less node */
2777             if (machine->numa_state->nodes[i].node_mem == 0) {
2778                 CPUState *cs;
2779                 int found = 0;
2780                 /* check for cpu-less node */
2781                 CPU_FOREACH(cs) {
2782                     PowerPCCPU *cpu = POWERPC_CPU(cs);
2783                     if (cpu->node_id == i) {
2784                         found = 1;
2785                         break;
2786                     }
2787                 }
2788                 /* memory-less and cpu-less node */
2789                 if (!found) {
2790                     error_report(
2791                        "Memory-less/cpu-less nodes are not supported (node %d)",
2792                                  i);
2793                     exit(1);
2794                 }
2795             }
2796         }
2797 
2798     }
2799 
2800     /*
2801      * NVLink2-connected GPU RAM needs to be placed on a separate NUMA node.
2802      * We assign a new numa ID per GPU in spapr_pci_collect_nvgpu() which is
2803      * called from vPHB reset handler so we initialize the counter here.
2804      * If no NUMA is configured from the QEMU side, we start from 1 as GPU RAM
2805      * must be equally distant from any other node.
2806      * The final value of spapr->gpu_numa_id is going to be written to
2807      * max-associativity-domains in spapr_build_fdt().
2808      */
2809     spapr->gpu_numa_id = MAX(1, machine->numa_state->num_nodes);
2810 
2811     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2812         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2813                               spapr->max_compat_pvr)) {
2814         /* KVM and TCG always allow GTSE with radix... */
2815         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2816     }
2817     /* ... but not with hash (currently). */
2818 
2819     if (kvm_enabled()) {
2820         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2821         kvmppc_enable_logical_ci_hcalls();
2822         kvmppc_enable_set_mode_hcall();
2823 
2824         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2825         kvmppc_enable_clear_ref_mod_hcalls();
2826 
2827         /* Enable H_PAGE_INIT */
2828         kvmppc_enable_h_page_init();
2829     }
2830 
2831     /* allocate RAM */
2832     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2833                                          machine->ram_size);
2834     memory_region_add_subregion(sysmem, 0, ram);
2835 
2836     /* always allocate the device memory information */
2837     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2838 
2839     /* initialize hotplug memory address space */
2840     if (machine->ram_size < machine->maxram_size) {
2841         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2842         /*
2843          * Limit the number of hotpluggable memory slots to half the number
2844          * slots that KVM supports, leaving the other half for PCI and other
2845          * devices. However ensure that number of slots doesn't drop below 32.
2846          */
2847         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2848                            SPAPR_MAX_RAM_SLOTS;
2849 
2850         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2851             max_memslots = SPAPR_MAX_RAM_SLOTS;
2852         }
2853         if (machine->ram_slots > max_memslots) {
2854             error_report("Specified number of memory slots %"
2855                          PRIu64" exceeds max supported %d",
2856                          machine->ram_slots, max_memslots);
2857             exit(1);
2858         }
2859 
2860         machine->device_memory->base = ROUND_UP(machine->ram_size,
2861                                                 SPAPR_DEVICE_MEM_ALIGN);
2862         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2863                            "device-memory", device_mem_size);
2864         memory_region_add_subregion(sysmem, machine->device_memory->base,
2865                                     &machine->device_memory->mr);
2866     }
2867 
2868     if (smc->dr_lmb_enabled) {
2869         spapr_create_lmb_dr_connectors(spapr);
2870     }
2871 
2872     if (spapr_get_cap(spapr, SPAPR_CAP_FWNMI_MCE) == SPAPR_CAP_ON) {
2873         /* Create the error string for live migration blocker */
2874         error_setg(&spapr->fwnmi_migration_blocker,
2875             "A machine check is being handled during migration. The handler"
2876             "may run and log hardware error on the destination");
2877     }
2878 
2879     if (mc->nvdimm_supported) {
2880         spapr_create_nvdimm_dr_connectors(spapr);
2881     }
2882 
2883     /* Set up RTAS event infrastructure */
2884     spapr_events_init(spapr);
2885 
2886     /* Set up the RTC RTAS interfaces */
2887     spapr_rtc_create(spapr);
2888 
2889     /* Set up VIO bus */
2890     spapr->vio_bus = spapr_vio_bus_init();
2891 
2892     for (i = 0; i < serial_max_hds(); i++) {
2893         if (serial_hd(i)) {
2894             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2895         }
2896     }
2897 
2898     /* We always have at least the nvram device on VIO */
2899     spapr_create_nvram(spapr);
2900 
2901     /*
2902      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2903      * connectors (described in root DT node's "ibm,drc-types" property)
2904      * are pre-initialized here. additional child connectors (such as
2905      * connectors for a PHBs PCI slots) are added as needed during their
2906      * parent's realization.
2907      */
2908     if (smc->dr_phb_enabled) {
2909         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2910             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2911         }
2912     }
2913 
2914     /* Set up PCI */
2915     spapr_pci_rtas_init();
2916 
2917     phb = spapr_create_default_phb();
2918 
2919     for (i = 0; i < nb_nics; i++) {
2920         NICInfo *nd = &nd_table[i];
2921 
2922         if (!nd->model) {
2923             nd->model = g_strdup("spapr-vlan");
2924         }
2925 
2926         if (g_str_equal(nd->model, "spapr-vlan") ||
2927             g_str_equal(nd->model, "ibmveth")) {
2928             spapr_vlan_create(spapr->vio_bus, nd);
2929         } else {
2930             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2931         }
2932     }
2933 
2934     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2935         spapr_vscsi_create(spapr->vio_bus);
2936     }
2937 
2938     /* Graphics */
2939     if (spapr_vga_init(phb->bus, &error_fatal)) {
2940         spapr->has_graphics = true;
2941         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2942     }
2943 
2944     if (machine->usb) {
2945         if (smc->use_ohci_by_default) {
2946             pci_create_simple(phb->bus, -1, "pci-ohci");
2947         } else {
2948             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2949         }
2950 
2951         if (spapr->has_graphics) {
2952             USBBus *usb_bus = usb_bus_find(-1);
2953 
2954             usb_create_simple(usb_bus, "usb-kbd");
2955             usb_create_simple(usb_bus, "usb-mouse");
2956         }
2957     }
2958 
2959     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2960         error_report(
2961             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2962             MIN_RMA_SLOF);
2963         exit(1);
2964     }
2965 
2966     if (kernel_filename) {
2967         uint64_t lowaddr = 0;
2968 
2969         spapr->kernel_size = load_elf(kernel_filename, NULL,
2970                                       translate_kernel_address, NULL,
2971                                       NULL, &lowaddr, NULL, NULL, 1,
2972                                       PPC_ELF_MACHINE, 0, 0);
2973         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2974             spapr->kernel_size = load_elf(kernel_filename, NULL,
2975                                           translate_kernel_address, NULL, NULL,
2976                                           &lowaddr, NULL, NULL, 0,
2977                                           PPC_ELF_MACHINE, 0, 0);
2978             spapr->kernel_le = spapr->kernel_size > 0;
2979         }
2980         if (spapr->kernel_size < 0) {
2981             error_report("error loading %s: %s", kernel_filename,
2982                          load_elf_strerror(spapr->kernel_size));
2983             exit(1);
2984         }
2985 
2986         /* load initrd */
2987         if (initrd_filename) {
2988             /* Try to locate the initrd in the gap between the kernel
2989              * and the firmware. Add a bit of space just in case
2990              */
2991             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2992                                   + 0x1ffff) & ~0xffff;
2993             spapr->initrd_size = load_image_targphys(initrd_filename,
2994                                                      spapr->initrd_base,
2995                                                      load_limit
2996                                                      - spapr->initrd_base);
2997             if (spapr->initrd_size < 0) {
2998                 error_report("could not load initial ram disk '%s'",
2999                              initrd_filename);
3000                 exit(1);
3001             }
3002         }
3003     }
3004 
3005     if (bios_name == NULL) {
3006         bios_name = FW_FILE_NAME;
3007     }
3008     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
3009     if (!filename) {
3010         error_report("Could not find LPAR firmware '%s'", bios_name);
3011         exit(1);
3012     }
3013     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3014     if (fw_size <= 0) {
3015         error_report("Could not load LPAR firmware '%s'", filename);
3016         exit(1);
3017     }
3018     g_free(filename);
3019 
3020     /* FIXME: Should register things through the MachineState's qdev
3021      * interface, this is a legacy from the sPAPREnvironment structure
3022      * which predated MachineState but had a similar function */
3023     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3024     register_savevm_live("spapr/htab", VMSTATE_INSTANCE_ID_ANY, 1,
3025                          &savevm_htab_handlers, spapr);
3026 
3027     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3028                              &error_fatal);
3029 
3030     qemu_register_boot_set(spapr_boot_set, spapr);
3031 
3032     /*
3033      * Nothing needs to be done to resume a suspended guest because
3034      * suspending does not change the machine state, so no need for
3035      * a ->wakeup method.
3036      */
3037     qemu_register_wakeup_support();
3038 
3039     if (kvm_enabled()) {
3040         /* to stop and start vmclock */
3041         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3042                                          &spapr->tb);
3043 
3044         kvmppc_spapr_enable_inkernel_multitce();
3045     }
3046 
3047     qemu_cond_init(&spapr->mc_delivery_cond);
3048 }
3049 
3050 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3051 {
3052     if (!vm_type) {
3053         return 0;
3054     }
3055 
3056     if (!strcmp(vm_type, "HV")) {
3057         return 1;
3058     }
3059 
3060     if (!strcmp(vm_type, "PR")) {
3061         return 2;
3062     }
3063 
3064     error_report("Unknown kvm-type specified '%s'", vm_type);
3065     exit(1);
3066 }
3067 
3068 /*
3069  * Implementation of an interface to adjust firmware path
3070  * for the bootindex property handling.
3071  */
3072 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3073                                    DeviceState *dev)
3074 {
3075 #define CAST(type, obj, name) \
3076     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3077     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3078     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3079     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3080 
3081     if (d) {
3082         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3083         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3084         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3085 
3086         if (spapr) {
3087             /*
3088              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3089              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3090              * 0x8000 | (target << 8) | (bus << 5) | lun
3091              * (see the "Logical unit addressing format" table in SAM5)
3092              */
3093             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3094             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3095                                    (uint64_t)id << 48);
3096         } else if (virtio) {
3097             /*
3098              * We use SRP luns of the form 01000000 | (target << 8) | lun
3099              * in the top 32 bits of the 64-bit LUN
3100              * Note: the quote above is from SLOF and it is wrong,
3101              * the actual binding is:
3102              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3103              */
3104             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3105             if (d->lun >= 256) {
3106                 /* Use the LUN "flat space addressing method" */
3107                 id |= 0x4000;
3108             }
3109             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3110                                    (uint64_t)id << 32);
3111         } else if (usb) {
3112             /*
3113              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3114              * in the top 32 bits of the 64-bit LUN
3115              */
3116             unsigned usb_port = atoi(usb->port->path);
3117             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3118             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3119                                    (uint64_t)id << 32);
3120         }
3121     }
3122 
3123     /*
3124      * SLOF probes the USB devices, and if it recognizes that the device is a
3125      * storage device, it changes its name to "storage" instead of "usb-host",
3126      * and additionally adds a child node for the SCSI LUN, so the correct
3127      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3128      */
3129     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3130         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3131         if (usb_host_dev_is_scsi_storage(usbdev)) {
3132             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3133         }
3134     }
3135 
3136     if (phb) {
3137         /* Replace "pci" with "pci@800000020000000" */
3138         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3139     }
3140 
3141     if (vsc) {
3142         /* Same logic as virtio above */
3143         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3144         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3145     }
3146 
3147     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3148         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3149         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3150         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3151     }
3152 
3153     return NULL;
3154 }
3155 
3156 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3157 {
3158     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3159 
3160     return g_strdup(spapr->kvm_type);
3161 }
3162 
3163 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3164 {
3165     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3166 
3167     g_free(spapr->kvm_type);
3168     spapr->kvm_type = g_strdup(value);
3169 }
3170 
3171 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3172 {
3173     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3174 
3175     return spapr->use_hotplug_event_source;
3176 }
3177 
3178 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3179                                             Error **errp)
3180 {
3181     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3182 
3183     spapr->use_hotplug_event_source = value;
3184 }
3185 
3186 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3187 {
3188     return true;
3189 }
3190 
3191 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3192 {
3193     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3194 
3195     switch (spapr->resize_hpt) {
3196     case SPAPR_RESIZE_HPT_DEFAULT:
3197         return g_strdup("default");
3198     case SPAPR_RESIZE_HPT_DISABLED:
3199         return g_strdup("disabled");
3200     case SPAPR_RESIZE_HPT_ENABLED:
3201         return g_strdup("enabled");
3202     case SPAPR_RESIZE_HPT_REQUIRED:
3203         return g_strdup("required");
3204     }
3205     g_assert_not_reached();
3206 }
3207 
3208 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3209 {
3210     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3211 
3212     if (strcmp(value, "default") == 0) {
3213         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3214     } else if (strcmp(value, "disabled") == 0) {
3215         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3216     } else if (strcmp(value, "enabled") == 0) {
3217         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3218     } else if (strcmp(value, "required") == 0) {
3219         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3220     } else {
3221         error_setg(errp, "Bad value for \"resize-hpt\" property");
3222     }
3223 }
3224 
3225 static void spapr_get_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_set_vsmt(Object *obj, Visitor *v, const char *name,
3232                                    void *opaque, Error **errp)
3233 {
3234     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3235 }
3236 
3237 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3238 {
3239     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3240 
3241     if (spapr->irq == &spapr_irq_xics_legacy) {
3242         return g_strdup("legacy");
3243     } else if (spapr->irq == &spapr_irq_xics) {
3244         return g_strdup("xics");
3245     } else if (spapr->irq == &spapr_irq_xive) {
3246         return g_strdup("xive");
3247     } else if (spapr->irq == &spapr_irq_dual) {
3248         return g_strdup("dual");
3249     }
3250     g_assert_not_reached();
3251 }
3252 
3253 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3254 {
3255     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3256 
3257     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3258         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3259         return;
3260     }
3261 
3262     /* The legacy IRQ backend can not be set */
3263     if (strcmp(value, "xics") == 0) {
3264         spapr->irq = &spapr_irq_xics;
3265     } else if (strcmp(value, "xive") == 0) {
3266         spapr->irq = &spapr_irq_xive;
3267     } else if (strcmp(value, "dual") == 0) {
3268         spapr->irq = &spapr_irq_dual;
3269     } else {
3270         error_setg(errp, "Bad value for \"ic-mode\" property");
3271     }
3272 }
3273 
3274 static char *spapr_get_host_model(Object *obj, Error **errp)
3275 {
3276     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3277 
3278     return g_strdup(spapr->host_model);
3279 }
3280 
3281 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3282 {
3283     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3284 
3285     g_free(spapr->host_model);
3286     spapr->host_model = g_strdup(value);
3287 }
3288 
3289 static char *spapr_get_host_serial(Object *obj, Error **errp)
3290 {
3291     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3292 
3293     return g_strdup(spapr->host_serial);
3294 }
3295 
3296 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3297 {
3298     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3299 
3300     g_free(spapr->host_serial);
3301     spapr->host_serial = g_strdup(value);
3302 }
3303 
3304 static void spapr_instance_init(Object *obj)
3305 {
3306     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3307     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3308 
3309     spapr->htab_fd = -1;
3310     spapr->use_hotplug_event_source = true;
3311     object_property_add_str(obj, "kvm-type",
3312                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3313     object_property_set_description(obj, "kvm-type",
3314                                     "Specifies the KVM virtualization mode (HV, PR)",
3315                                     NULL);
3316     object_property_add_bool(obj, "modern-hotplug-events",
3317                             spapr_get_modern_hotplug_events,
3318                             spapr_set_modern_hotplug_events,
3319                             NULL);
3320     object_property_set_description(obj, "modern-hotplug-events",
3321                                     "Use dedicated hotplug event mechanism in"
3322                                     " place of standard EPOW events when possible"
3323                                     " (required for memory hot-unplug support)",
3324                                     NULL);
3325     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3326                             "Maximum permitted CPU compatibility mode",
3327                             &error_fatal);
3328 
3329     object_property_add_str(obj, "resize-hpt",
3330                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3331     object_property_set_description(obj, "resize-hpt",
3332                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3333                                     NULL);
3334     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3335                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3336     object_property_set_description(obj, "vsmt",
3337                                     "Virtual SMT: KVM behaves as if this were"
3338                                     " the host's SMT mode", &error_abort);
3339     object_property_add_bool(obj, "vfio-no-msix-emulation",
3340                              spapr_get_msix_emulation, NULL, NULL);
3341 
3342     /* The machine class defines the default interrupt controller mode */
3343     spapr->irq = smc->irq;
3344     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3345                             spapr_set_ic_mode, NULL);
3346     object_property_set_description(obj, "ic-mode",
3347                  "Specifies the interrupt controller mode (xics, xive, dual)",
3348                  NULL);
3349 
3350     object_property_add_str(obj, "host-model",
3351         spapr_get_host_model, spapr_set_host_model,
3352         &error_abort);
3353     object_property_set_description(obj, "host-model",
3354         "Host model to advertise in guest device tree", &error_abort);
3355     object_property_add_str(obj, "host-serial",
3356         spapr_get_host_serial, spapr_set_host_serial,
3357         &error_abort);
3358     object_property_set_description(obj, "host-serial",
3359         "Host serial number to advertise in guest device tree", &error_abort);
3360 }
3361 
3362 static void spapr_machine_finalizefn(Object *obj)
3363 {
3364     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3365 
3366     g_free(spapr->kvm_type);
3367 }
3368 
3369 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3370 {
3371     cpu_synchronize_state(cs);
3372     ppc_cpu_do_system_reset(cs);
3373 }
3374 
3375 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3376 {
3377     CPUState *cs;
3378 
3379     CPU_FOREACH(cs) {
3380         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3381     }
3382 }
3383 
3384 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3385                           void *fdt, int *fdt_start_offset, Error **errp)
3386 {
3387     uint64_t addr;
3388     uint32_t node;
3389 
3390     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3391     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3392                                     &error_abort);
3393     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3394                                                    SPAPR_MEMORY_BLOCK_SIZE);
3395     return 0;
3396 }
3397 
3398 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3399                            bool dedicated_hp_event_source, Error **errp)
3400 {
3401     SpaprDrc *drc;
3402     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3403     int i;
3404     uint64_t addr = addr_start;
3405     bool hotplugged = spapr_drc_hotplugged(dev);
3406     Error *local_err = NULL;
3407 
3408     for (i = 0; i < nr_lmbs; i++) {
3409         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3410                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3411         g_assert(drc);
3412 
3413         spapr_drc_attach(drc, dev, &local_err);
3414         if (local_err) {
3415             while (addr > addr_start) {
3416                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3417                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3418                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3419                 spapr_drc_detach(drc);
3420             }
3421             error_propagate(errp, local_err);
3422             return;
3423         }
3424         if (!hotplugged) {
3425             spapr_drc_reset(drc);
3426         }
3427         addr += SPAPR_MEMORY_BLOCK_SIZE;
3428     }
3429     /* send hotplug notification to the
3430      * guest only in case of hotplugged memory
3431      */
3432     if (hotplugged) {
3433         if (dedicated_hp_event_source) {
3434             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3435                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3436             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3437                                                    nr_lmbs,
3438                                                    spapr_drc_index(drc));
3439         } else {
3440             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3441                                            nr_lmbs);
3442         }
3443     }
3444 }
3445 
3446 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3447                               Error **errp)
3448 {
3449     Error *local_err = NULL;
3450     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3451     PCDIMMDevice *dimm = PC_DIMM(dev);
3452     uint64_t size, addr, slot;
3453     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3454 
3455     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3456 
3457     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3458     if (local_err) {
3459         goto out;
3460     }
3461 
3462     if (!is_nvdimm) {
3463         addr = object_property_get_uint(OBJECT(dimm),
3464                                         PC_DIMM_ADDR_PROP, &local_err);
3465         if (local_err) {
3466             goto out_unplug;
3467         }
3468         spapr_add_lmbs(dev, addr, size,
3469                        spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3470                        &local_err);
3471     } else {
3472         slot = object_property_get_uint(OBJECT(dimm),
3473                                         PC_DIMM_SLOT_PROP, &local_err);
3474         if (local_err) {
3475             goto out_unplug;
3476         }
3477         spapr_add_nvdimm(dev, slot, &local_err);
3478     }
3479 
3480     if (local_err) {
3481         goto out_unplug;
3482     }
3483 
3484     return;
3485 
3486 out_unplug:
3487     pc_dimm_unplug(dimm, MACHINE(ms));
3488 out:
3489     error_propagate(errp, local_err);
3490 }
3491 
3492 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3493                                   Error **errp)
3494 {
3495     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3496     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3497     const MachineClass *mc = MACHINE_CLASS(smc);
3498     bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
3499     PCDIMMDevice *dimm = PC_DIMM(dev);
3500     Error *local_err = NULL;
3501     uint64_t size;
3502     Object *memdev;
3503     hwaddr pagesize;
3504 
3505     if (!smc->dr_lmb_enabled) {
3506         error_setg(errp, "Memory hotplug not supported for this machine");
3507         return;
3508     }
3509 
3510     if (is_nvdimm && !mc->nvdimm_supported) {
3511         error_setg(errp, "NVDIMM hotplug not supported for this machine");
3512         return;
3513     }
3514 
3515     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3516     if (local_err) {
3517         error_propagate(errp, local_err);
3518         return;
3519     }
3520 
3521     if (!is_nvdimm && size % SPAPR_MEMORY_BLOCK_SIZE) {
3522         error_setg(errp, "Hotplugged memory size must be a multiple of "
3523                    "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3524         return;
3525     } else if (is_nvdimm) {
3526         spapr_nvdimm_validate_opts(NVDIMM(dev), size, &local_err);
3527         if (local_err) {
3528             error_propagate(errp, local_err);
3529             return;
3530         }
3531     }
3532 
3533     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3534                                       &error_abort);
3535     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3536     spapr_check_pagesize(spapr, pagesize, &local_err);
3537     if (local_err) {
3538         error_propagate(errp, local_err);
3539         return;
3540     }
3541 
3542     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3543 }
3544 
3545 struct SpaprDimmState {
3546     PCDIMMDevice *dimm;
3547     uint32_t nr_lmbs;
3548     QTAILQ_ENTRY(SpaprDimmState) next;
3549 };
3550 
3551 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3552                                                        PCDIMMDevice *dimm)
3553 {
3554     SpaprDimmState *dimm_state = NULL;
3555 
3556     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3557         if (dimm_state->dimm == dimm) {
3558             break;
3559         }
3560     }
3561     return dimm_state;
3562 }
3563 
3564 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3565                                                       uint32_t nr_lmbs,
3566                                                       PCDIMMDevice *dimm)
3567 {
3568     SpaprDimmState *ds = NULL;
3569 
3570     /*
3571      * If this request is for a DIMM whose removal had failed earlier
3572      * (due to guest's refusal to remove the LMBs), we would have this
3573      * dimm already in the pending_dimm_unplugs list. In that
3574      * case don't add again.
3575      */
3576     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3577     if (!ds) {
3578         ds = g_malloc0(sizeof(SpaprDimmState));
3579         ds->nr_lmbs = nr_lmbs;
3580         ds->dimm = dimm;
3581         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3582     }
3583     return ds;
3584 }
3585 
3586 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3587                                               SpaprDimmState *dimm_state)
3588 {
3589     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3590     g_free(dimm_state);
3591 }
3592 
3593 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3594                                                         PCDIMMDevice *dimm)
3595 {
3596     SpaprDrc *drc;
3597     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3598                                                   &error_abort);
3599     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3600     uint32_t avail_lmbs = 0;
3601     uint64_t addr_start, addr;
3602     int i;
3603 
3604     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3605                                          &error_abort);
3606 
3607     addr = addr_start;
3608     for (i = 0; i < nr_lmbs; i++) {
3609         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3610                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3611         g_assert(drc);
3612         if (drc->dev) {
3613             avail_lmbs++;
3614         }
3615         addr += SPAPR_MEMORY_BLOCK_SIZE;
3616     }
3617 
3618     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3619 }
3620 
3621 /* Callback to be called during DRC release. */
3622 void spapr_lmb_release(DeviceState *dev)
3623 {
3624     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3625     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3626     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3627 
3628     /* This information will get lost if a migration occurs
3629      * during the unplug process. In this case recover it. */
3630     if (ds == NULL) {
3631         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3632         g_assert(ds);
3633         /* The DRC being examined by the caller at least must be counted */
3634         g_assert(ds->nr_lmbs);
3635     }
3636 
3637     if (--ds->nr_lmbs) {
3638         return;
3639     }
3640 
3641     /*
3642      * Now that all the LMBs have been removed by the guest, call the
3643      * unplug handler chain. This can never fail.
3644      */
3645     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3646     object_unparent(OBJECT(dev));
3647 }
3648 
3649 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3650 {
3651     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3652     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3653 
3654     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3655     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3656     spapr_pending_dimm_unplugs_remove(spapr, ds);
3657 }
3658 
3659 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3660                                         DeviceState *dev, Error **errp)
3661 {
3662     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3663     Error *local_err = NULL;
3664     PCDIMMDevice *dimm = PC_DIMM(dev);
3665     uint32_t nr_lmbs;
3666     uint64_t size, addr_start, addr;
3667     int i;
3668     SpaprDrc *drc;
3669 
3670     if (object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM)) {
3671         error_setg(&local_err,
3672                    "nvdimm device hot unplug is not supported yet.");
3673         goto out;
3674     }
3675 
3676     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3677     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3678 
3679     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3680                                          &local_err);
3681     if (local_err) {
3682         goto out;
3683     }
3684 
3685     /*
3686      * An existing pending dimm state for this DIMM means that there is an
3687      * unplug operation in progress, waiting for the spapr_lmb_release
3688      * callback to complete the job (BQL can't cover that far). In this case,
3689      * bail out to avoid detaching DRCs that were already released.
3690      */
3691     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3692         error_setg(&local_err,
3693                    "Memory unplug already in progress for device %s",
3694                    dev->id);
3695         goto out;
3696     }
3697 
3698     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3699 
3700     addr = addr_start;
3701     for (i = 0; i < nr_lmbs; i++) {
3702         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3703                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3704         g_assert(drc);
3705 
3706         spapr_drc_detach(drc);
3707         addr += SPAPR_MEMORY_BLOCK_SIZE;
3708     }
3709 
3710     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3711                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3712     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3713                                               nr_lmbs, spapr_drc_index(drc));
3714 out:
3715     error_propagate(errp, local_err);
3716 }
3717 
3718 /* Callback to be called during DRC release. */
3719 void spapr_core_release(DeviceState *dev)
3720 {
3721     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3722 
3723     /* Call the unplug handler chain. This can never fail. */
3724     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3725     object_unparent(OBJECT(dev));
3726 }
3727 
3728 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3729 {
3730     MachineState *ms = MACHINE(hotplug_dev);
3731     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3732     CPUCore *cc = CPU_CORE(dev);
3733     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3734 
3735     if (smc->pre_2_10_has_unused_icps) {
3736         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3737         int i;
3738 
3739         for (i = 0; i < cc->nr_threads; i++) {
3740             CPUState *cs = CPU(sc->threads[i]);
3741 
3742             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3743         }
3744     }
3745 
3746     assert(core_slot);
3747     core_slot->cpu = NULL;
3748     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3749 }
3750 
3751 static
3752 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3753                                Error **errp)
3754 {
3755     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3756     int index;
3757     SpaprDrc *drc;
3758     CPUCore *cc = CPU_CORE(dev);
3759 
3760     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3761         error_setg(errp, "Unable to find CPU core with core-id: %d",
3762                    cc->core_id);
3763         return;
3764     }
3765     if (index == 0) {
3766         error_setg(errp, "Boot CPU core may not be unplugged");
3767         return;
3768     }
3769 
3770     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3771                           spapr_vcpu_id(spapr, cc->core_id));
3772     g_assert(drc);
3773 
3774     if (!spapr_drc_unplug_requested(drc)) {
3775         spapr_drc_detach(drc);
3776         spapr_hotplug_req_remove_by_index(drc);
3777     }
3778 }
3779 
3780 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3781                            void *fdt, int *fdt_start_offset, Error **errp)
3782 {
3783     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3784     CPUState *cs = CPU(core->threads[0]);
3785     PowerPCCPU *cpu = POWERPC_CPU(cs);
3786     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3787     int id = spapr_get_vcpu_id(cpu);
3788     char *nodename;
3789     int offset;
3790 
3791     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3792     offset = fdt_add_subnode(fdt, 0, nodename);
3793     g_free(nodename);
3794 
3795     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3796 
3797     *fdt_start_offset = offset;
3798     return 0;
3799 }
3800 
3801 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3802                             Error **errp)
3803 {
3804     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3805     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3806     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3807     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3808     CPUCore *cc = CPU_CORE(dev);
3809     CPUState *cs;
3810     SpaprDrc *drc;
3811     Error *local_err = NULL;
3812     CPUArchId *core_slot;
3813     int index;
3814     bool hotplugged = spapr_drc_hotplugged(dev);
3815     int i;
3816 
3817     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3818     if (!core_slot) {
3819         error_setg(errp, "Unable to find CPU core with core-id: %d",
3820                    cc->core_id);
3821         return;
3822     }
3823     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3824                           spapr_vcpu_id(spapr, cc->core_id));
3825 
3826     g_assert(drc || !mc->has_hotpluggable_cpus);
3827 
3828     if (drc) {
3829         spapr_drc_attach(drc, dev, &local_err);
3830         if (local_err) {
3831             error_propagate(errp, local_err);
3832             return;
3833         }
3834 
3835         if (hotplugged) {
3836             /*
3837              * Send hotplug notification interrupt to the guest only
3838              * in case of hotplugged CPUs.
3839              */
3840             spapr_hotplug_req_add_by_index(drc);
3841         } else {
3842             spapr_drc_reset(drc);
3843         }
3844     }
3845 
3846     core_slot->cpu = OBJECT(dev);
3847 
3848     if (smc->pre_2_10_has_unused_icps) {
3849         for (i = 0; i < cc->nr_threads; i++) {
3850             cs = CPU(core->threads[i]);
3851             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3852         }
3853     }
3854 
3855     /*
3856      * Set compatibility mode to match the boot CPU, which was either set
3857      * by the machine reset code or by CAS.
3858      */
3859     if (hotplugged) {
3860         for (i = 0; i < cc->nr_threads; i++) {
3861             ppc_set_compat(core->threads[i], POWERPC_CPU(first_cpu)->compat_pvr,
3862                            &local_err);
3863             if (local_err) {
3864                 error_propagate(errp, local_err);
3865                 return;
3866             }
3867         }
3868     }
3869 }
3870 
3871 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3872                                 Error **errp)
3873 {
3874     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3875     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3876     Error *local_err = NULL;
3877     CPUCore *cc = CPU_CORE(dev);
3878     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3879     const char *type = object_get_typename(OBJECT(dev));
3880     CPUArchId *core_slot;
3881     int index;
3882     unsigned int smp_threads = machine->smp.threads;
3883 
3884     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3885         error_setg(&local_err, "CPU hotplug not supported for this machine");
3886         goto out;
3887     }
3888 
3889     if (strcmp(base_core_type, type)) {
3890         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3891         goto out;
3892     }
3893 
3894     if (cc->core_id % smp_threads) {
3895         error_setg(&local_err, "invalid core id %d", cc->core_id);
3896         goto out;
3897     }
3898 
3899     /*
3900      * In general we should have homogeneous threads-per-core, but old
3901      * (pre hotplug support) machine types allow the last core to have
3902      * reduced threads as a compatibility hack for when we allowed
3903      * total vcpus not a multiple of threads-per-core.
3904      */
3905     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3906         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3907                    cc->nr_threads, smp_threads);
3908         goto out;
3909     }
3910 
3911     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3912     if (!core_slot) {
3913         error_setg(&local_err, "core id %d out of range", cc->core_id);
3914         goto out;
3915     }
3916 
3917     if (core_slot->cpu) {
3918         error_setg(&local_err, "core %d already populated", cc->core_id);
3919         goto out;
3920     }
3921 
3922     numa_cpu_pre_plug(core_slot, dev, &local_err);
3923 
3924 out:
3925     error_propagate(errp, local_err);
3926 }
3927 
3928 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3929                           void *fdt, int *fdt_start_offset, Error **errp)
3930 {
3931     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3932     int intc_phandle;
3933 
3934     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3935     if (intc_phandle <= 0) {
3936         return -1;
3937     }
3938 
3939     if (spapr_dt_phb(spapr, sphb, intc_phandle, fdt, fdt_start_offset)) {
3940         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3941         return -1;
3942     }
3943 
3944     /* generally SLOF creates these, for hotplug it's up to QEMU */
3945     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3946 
3947     return 0;
3948 }
3949 
3950 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3951                                Error **errp)
3952 {
3953     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3954     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3955     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3956     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3957 
3958     if (dev->hotplugged && !smc->dr_phb_enabled) {
3959         error_setg(errp, "PHB hotplug not supported for this machine");
3960         return;
3961     }
3962 
3963     if (sphb->index == (uint32_t)-1) {
3964         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3965         return;
3966     }
3967 
3968     /*
3969      * This will check that sphb->index doesn't exceed the maximum number of
3970      * PHBs for the current machine type.
3971      */
3972     smc->phb_placement(spapr, sphb->index,
3973                        &sphb->buid, &sphb->io_win_addr,
3974                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3975                        windows_supported, sphb->dma_liobn,
3976                        &sphb->nv2_gpa_win_addr, &sphb->nv2_atsd_win_addr,
3977                        errp);
3978 }
3979 
3980 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3981                            Error **errp)
3982 {
3983     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3984     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3985     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3986     SpaprDrc *drc;
3987     bool hotplugged = spapr_drc_hotplugged(dev);
3988     Error *local_err = NULL;
3989 
3990     if (!smc->dr_phb_enabled) {
3991         return;
3992     }
3993 
3994     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3995     /* hotplug hooks should check it's enabled before getting this far */
3996     assert(drc);
3997 
3998     spapr_drc_attach(drc, DEVICE(dev), &local_err);
3999     if (local_err) {
4000         error_propagate(errp, local_err);
4001         return;
4002     }
4003 
4004     if (hotplugged) {
4005         spapr_hotplug_req_add_by_index(drc);
4006     } else {
4007         spapr_drc_reset(drc);
4008     }
4009 }
4010 
4011 void spapr_phb_release(DeviceState *dev)
4012 {
4013     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
4014 
4015     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
4016     object_unparent(OBJECT(dev));
4017 }
4018 
4019 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4020 {
4021     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4022 }
4023 
4024 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
4025                                      DeviceState *dev, Error **errp)
4026 {
4027     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
4028     SpaprDrc *drc;
4029 
4030     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
4031     assert(drc);
4032 
4033     if (!spapr_drc_unplug_requested(drc)) {
4034         spapr_drc_detach(drc);
4035         spapr_hotplug_req_remove_by_index(drc);
4036     }
4037 }
4038 
4039 static void spapr_tpm_proxy_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
4040                                  Error **errp)
4041 {
4042     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4043     SpaprTpmProxy *tpm_proxy = SPAPR_TPM_PROXY(dev);
4044 
4045     if (spapr->tpm_proxy != NULL) {
4046         error_setg(errp, "Only one TPM proxy can be specified for this machine");
4047         return;
4048     }
4049 
4050     spapr->tpm_proxy = tpm_proxy;
4051 }
4052 
4053 static void spapr_tpm_proxy_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
4054 {
4055     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
4056 
4057     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
4058     object_unparent(OBJECT(dev));
4059     spapr->tpm_proxy = NULL;
4060 }
4061 
4062 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
4063                                       DeviceState *dev, Error **errp)
4064 {
4065     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4066         spapr_memory_plug(hotplug_dev, dev, errp);
4067     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4068         spapr_core_plug(hotplug_dev, dev, errp);
4069     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4070         spapr_phb_plug(hotplug_dev, dev, errp);
4071     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4072         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
4073     }
4074 }
4075 
4076 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
4077                                         DeviceState *dev, Error **errp)
4078 {
4079     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4080         spapr_memory_unplug(hotplug_dev, dev);
4081     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4082         spapr_core_unplug(hotplug_dev, dev);
4083     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4084         spapr_phb_unplug(hotplug_dev, dev);
4085     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4086         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4087     }
4088 }
4089 
4090 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
4091                                                 DeviceState *dev, Error **errp)
4092 {
4093     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4094     MachineClass *mc = MACHINE_GET_CLASS(sms);
4095     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4096 
4097     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4098         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4099             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4100         } else {
4101             /* NOTE: this means there is a window after guest reset, prior to
4102              * CAS negotiation, where unplug requests will fail due to the
4103              * capability not being detected yet. This is a bit different than
4104              * the case with PCI unplug, where the events will be queued and
4105              * eventually handled by the guest after boot
4106              */
4107             error_setg(errp, "Memory hot unplug not supported for this guest");
4108         }
4109     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4110         if (!mc->has_hotpluggable_cpus) {
4111             error_setg(errp, "CPU hot unplug not supported on this machine");
4112             return;
4113         }
4114         spapr_core_unplug_request(hotplug_dev, dev, errp);
4115     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4116         if (!smc->dr_phb_enabled) {
4117             error_setg(errp, "PHB hot unplug not supported on this machine");
4118             return;
4119         }
4120         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4121     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4122         spapr_tpm_proxy_unplug(hotplug_dev, dev);
4123     }
4124 }
4125 
4126 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4127                                           DeviceState *dev, Error **errp)
4128 {
4129     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4130         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4131     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4132         spapr_core_pre_plug(hotplug_dev, dev, errp);
4133     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4134         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4135     }
4136 }
4137 
4138 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4139                                                  DeviceState *dev)
4140 {
4141     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4142         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4143         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE) ||
4144         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
4145         return HOTPLUG_HANDLER(machine);
4146     }
4147     if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
4148         PCIDevice *pcidev = PCI_DEVICE(dev);
4149         PCIBus *root = pci_device_root_bus(pcidev);
4150         SpaprPhbState *phb =
4151             (SpaprPhbState *)object_dynamic_cast(OBJECT(BUS(root)->parent),
4152                                                  TYPE_SPAPR_PCI_HOST_BRIDGE);
4153 
4154         if (phb) {
4155             return HOTPLUG_HANDLER(phb);
4156         }
4157     }
4158     return NULL;
4159 }
4160 
4161 static CpuInstanceProperties
4162 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4163 {
4164     CPUArchId *core_slot;
4165     MachineClass *mc = MACHINE_GET_CLASS(machine);
4166 
4167     /* make sure possible_cpu are intialized */
4168     mc->possible_cpu_arch_ids(machine);
4169     /* get CPU core slot containing thread that matches cpu_index */
4170     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4171     assert(core_slot);
4172     return core_slot->props;
4173 }
4174 
4175 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4176 {
4177     return idx / ms->smp.cores % ms->numa_state->num_nodes;
4178 }
4179 
4180 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4181 {
4182     int i;
4183     unsigned int smp_threads = machine->smp.threads;
4184     unsigned int smp_cpus = machine->smp.cpus;
4185     const char *core_type;
4186     int spapr_max_cores = machine->smp.max_cpus / smp_threads;
4187     MachineClass *mc = MACHINE_GET_CLASS(machine);
4188 
4189     if (!mc->has_hotpluggable_cpus) {
4190         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4191     }
4192     if (machine->possible_cpus) {
4193         assert(machine->possible_cpus->len == spapr_max_cores);
4194         return machine->possible_cpus;
4195     }
4196 
4197     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4198     if (!core_type) {
4199         error_report("Unable to find sPAPR CPU Core definition");
4200         exit(1);
4201     }
4202 
4203     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4204                              sizeof(CPUArchId) * spapr_max_cores);
4205     machine->possible_cpus->len = spapr_max_cores;
4206     for (i = 0; i < machine->possible_cpus->len; i++) {
4207         int core_id = i * smp_threads;
4208 
4209         machine->possible_cpus->cpus[i].type = core_type;
4210         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4211         machine->possible_cpus->cpus[i].arch_id = core_id;
4212         machine->possible_cpus->cpus[i].props.has_core_id = true;
4213         machine->possible_cpus->cpus[i].props.core_id = core_id;
4214     }
4215     return machine->possible_cpus;
4216 }
4217 
4218 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4219                                 uint64_t *buid, hwaddr *pio,
4220                                 hwaddr *mmio32, hwaddr *mmio64,
4221                                 unsigned n_dma, uint32_t *liobns,
4222                                 hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4223 {
4224     /*
4225      * New-style PHB window placement.
4226      *
4227      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4228      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4229      * windows.
4230      *
4231      * Some guest kernels can't work with MMIO windows above 1<<46
4232      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4233      *
4234      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4235      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4236      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4237      * 1TiB 64-bit MMIO windows for each PHB.
4238      */
4239     const uint64_t base_buid = 0x800000020000000ULL;
4240     int i;
4241 
4242     /* Sanity check natural alignments */
4243     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4244     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4245     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4246     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4247     /* Sanity check bounds */
4248     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4249                       SPAPR_PCI_MEM32_WIN_SIZE);
4250     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4251                       SPAPR_PCI_MEM64_WIN_SIZE);
4252 
4253     if (index >= SPAPR_MAX_PHBS) {
4254         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4255                    SPAPR_MAX_PHBS - 1);
4256         return;
4257     }
4258 
4259     *buid = base_buid + index;
4260     for (i = 0; i < n_dma; ++i) {
4261         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4262     }
4263 
4264     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4265     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4266     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4267 
4268     *nv2gpa = SPAPR_PCI_NV2RAM64_WIN_BASE + index * SPAPR_PCI_NV2RAM64_WIN_SIZE;
4269     *nv2atsd = SPAPR_PCI_NV2ATSD_WIN_BASE + index * SPAPR_PCI_NV2ATSD_WIN_SIZE;
4270 }
4271 
4272 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4273 {
4274     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4275 
4276     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4277 }
4278 
4279 static void spapr_ics_resend(XICSFabric *dev)
4280 {
4281     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4282 
4283     ics_resend(spapr->ics);
4284 }
4285 
4286 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4287 {
4288     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4289 
4290     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4291 }
4292 
4293 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4294                                  Monitor *mon)
4295 {
4296     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4297 
4298     spapr_irq_print_info(spapr, mon);
4299     monitor_printf(mon, "irqchip: %s\n",
4300                    kvm_irqchip_in_kernel() ? "in-kernel" : "emulated");
4301 }
4302 
4303 /*
4304  * This is a XIVE only operation
4305  */
4306 static int spapr_match_nvt(XiveFabric *xfb, uint8_t format,
4307                            uint8_t nvt_blk, uint32_t nvt_idx,
4308                            bool cam_ignore, uint8_t priority,
4309                            uint32_t logic_serv, XiveTCTXMatch *match)
4310 {
4311     SpaprMachineState *spapr = SPAPR_MACHINE(xfb);
4312     XivePresenter *xptr = XIVE_PRESENTER(spapr->active_intc);
4313     XivePresenterClass *xpc = XIVE_PRESENTER_GET_CLASS(xptr);
4314     int count;
4315 
4316     count = xpc->match_nvt(xptr, format, nvt_blk, nvt_idx, cam_ignore,
4317                            priority, logic_serv, match);
4318     if (count < 0) {
4319         return count;
4320     }
4321 
4322     /*
4323      * When we implement the save and restore of the thread interrupt
4324      * contexts in the enter/exit CPU handlers of the machine and the
4325      * escalations in QEMU, we should be able to handle non dispatched
4326      * vCPUs.
4327      *
4328      * Until this is done, the sPAPR machine should find at least one
4329      * matching context always.
4330      */
4331     if (count == 0) {
4332         qemu_log_mask(LOG_GUEST_ERROR, "XIVE: NVT %x/%x is not dispatched\n",
4333                       nvt_blk, nvt_idx);
4334     }
4335 
4336     return count;
4337 }
4338 
4339 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4340 {
4341     return cpu->vcpu_id;
4342 }
4343 
4344 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4345 {
4346     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4347     MachineState *ms = MACHINE(spapr);
4348     int vcpu_id;
4349 
4350     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4351 
4352     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4353         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4354         error_append_hint(errp, "Adjust the number of cpus to %d "
4355                           "or try to raise the number of threads per core\n",
4356                           vcpu_id * ms->smp.threads / spapr->vsmt);
4357         return;
4358     }
4359 
4360     cpu->vcpu_id = vcpu_id;
4361 }
4362 
4363 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4364 {
4365     CPUState *cs;
4366 
4367     CPU_FOREACH(cs) {
4368         PowerPCCPU *cpu = POWERPC_CPU(cs);
4369 
4370         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4371             return cpu;
4372         }
4373     }
4374 
4375     return NULL;
4376 }
4377 
4378 static void spapr_cpu_exec_enter(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4379 {
4380     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4381 
4382     /* These are only called by TCG, KVM maintains dispatch state */
4383 
4384     spapr_cpu->prod = false;
4385     if (spapr_cpu->vpa_addr) {
4386         CPUState *cs = CPU(cpu);
4387         uint32_t dispatch;
4388 
4389         dispatch = ldl_be_phys(cs->as,
4390                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4391         dispatch++;
4392         if ((dispatch & 1) != 0) {
4393             qemu_log_mask(LOG_GUEST_ERROR,
4394                           "VPA: incorrect dispatch counter value for "
4395                           "dispatched partition %u, correcting.\n", dispatch);
4396             dispatch++;
4397         }
4398         stl_be_phys(cs->as,
4399                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4400     }
4401 }
4402 
4403 static void spapr_cpu_exec_exit(PPCVirtualHypervisor *vhyp, PowerPCCPU *cpu)
4404 {
4405     SpaprCpuState *spapr_cpu = spapr_cpu_state(cpu);
4406 
4407     if (spapr_cpu->vpa_addr) {
4408         CPUState *cs = CPU(cpu);
4409         uint32_t dispatch;
4410 
4411         dispatch = ldl_be_phys(cs->as,
4412                                spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER);
4413         dispatch++;
4414         if ((dispatch & 1) != 1) {
4415             qemu_log_mask(LOG_GUEST_ERROR,
4416                           "VPA: incorrect dispatch counter value for "
4417                           "preempted partition %u, correcting.\n", dispatch);
4418             dispatch++;
4419         }
4420         stl_be_phys(cs->as,
4421                     spapr_cpu->vpa_addr + VPA_DISPATCH_COUNTER, dispatch);
4422     }
4423 }
4424 
4425 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4426 {
4427     MachineClass *mc = MACHINE_CLASS(oc);
4428     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4429     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4430     NMIClass *nc = NMI_CLASS(oc);
4431     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4432     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4433     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4434     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4435     XiveFabricClass *xfc = XIVE_FABRIC_CLASS(oc);
4436 
4437     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4438     mc->ignore_boot_device_suffixes = true;
4439 
4440     /*
4441      * We set up the default / latest behaviour here.  The class_init
4442      * functions for the specific versioned machine types can override
4443      * these details for backwards compatibility
4444      */
4445     mc->init = spapr_machine_init;
4446     mc->reset = spapr_machine_reset;
4447     mc->block_default_type = IF_SCSI;
4448     mc->max_cpus = 1024;
4449     mc->no_parallel = 1;
4450     mc->default_boot_order = "";
4451     mc->default_ram_size = 512 * MiB;
4452     mc->default_display = "std";
4453     mc->kvm_type = spapr_kvm_type;
4454     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4455     mc->pci_allow_0_address = true;
4456     assert(!mc->get_hotplug_handler);
4457     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4458     hc->pre_plug = spapr_machine_device_pre_plug;
4459     hc->plug = spapr_machine_device_plug;
4460     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4461     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4462     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4463     hc->unplug_request = spapr_machine_device_unplug_request;
4464     hc->unplug = spapr_machine_device_unplug;
4465 
4466     smc->dr_lmb_enabled = true;
4467     smc->update_dt_enabled = true;
4468     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4469     mc->has_hotpluggable_cpus = true;
4470     mc->nvdimm_supported = true;
4471     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4472     fwc->get_dev_path = spapr_get_fw_dev_path;
4473     nc->nmi_monitor_handler = spapr_nmi;
4474     smc->phb_placement = spapr_phb_placement;
4475     vhc->hypercall = emulate_spapr_hypercall;
4476     vhc->hpt_mask = spapr_hpt_mask;
4477     vhc->map_hptes = spapr_map_hptes;
4478     vhc->unmap_hptes = spapr_unmap_hptes;
4479     vhc->hpte_set_c = spapr_hpte_set_c;
4480     vhc->hpte_set_r = spapr_hpte_set_r;
4481     vhc->get_pate = spapr_get_pate;
4482     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4483     vhc->cpu_exec_enter = spapr_cpu_exec_enter;
4484     vhc->cpu_exec_exit = spapr_cpu_exec_exit;
4485     xic->ics_get = spapr_ics_get;
4486     xic->ics_resend = spapr_ics_resend;
4487     xic->icp_get = spapr_icp_get;
4488     ispc->print_info = spapr_pic_print_info;
4489     /* Force NUMA node memory size to be a multiple of
4490      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4491      * in which LMBs are represented and hot-added
4492      */
4493     mc->numa_mem_align_shift = 28;
4494     mc->numa_mem_supported = true;
4495     mc->auto_enable_numa = true;
4496 
4497     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4498     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4499     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4500     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4501     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4502     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4503     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4504     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4505     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4506     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_ON;
4507     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_ON;
4508     spapr_caps_add_properties(smc, &error_abort);
4509     smc->irq = &spapr_irq_dual;
4510     smc->dr_phb_enabled = true;
4511     smc->linux_pci_probe = true;
4512     smc->smp_threads_vsmt = true;
4513     smc->nr_xirqs = SPAPR_NR_XIRQS;
4514     xfc->match_nvt = spapr_match_nvt;
4515 }
4516 
4517 static const TypeInfo spapr_machine_info = {
4518     .name          = TYPE_SPAPR_MACHINE,
4519     .parent        = TYPE_MACHINE,
4520     .abstract      = true,
4521     .instance_size = sizeof(SpaprMachineState),
4522     .instance_init = spapr_instance_init,
4523     .instance_finalize = spapr_machine_finalizefn,
4524     .class_size    = sizeof(SpaprMachineClass),
4525     .class_init    = spapr_machine_class_init,
4526     .interfaces = (InterfaceInfo[]) {
4527         { TYPE_FW_PATH_PROVIDER },
4528         { TYPE_NMI },
4529         { TYPE_HOTPLUG_HANDLER },
4530         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4531         { TYPE_XICS_FABRIC },
4532         { TYPE_INTERRUPT_STATS_PROVIDER },
4533         { TYPE_XIVE_FABRIC },
4534         { }
4535     },
4536 };
4537 
4538 static void spapr_machine_latest_class_options(MachineClass *mc)
4539 {
4540     mc->alias = "pseries";
4541     mc->is_default = 1;
4542 }
4543 
4544 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4545     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4546                                                     void *data)      \
4547     {                                                                \
4548         MachineClass *mc = MACHINE_CLASS(oc);                        \
4549         spapr_machine_##suffix##_class_options(mc);                  \
4550         if (latest) {                                                \
4551             spapr_machine_latest_class_options(mc);                  \
4552         }                                                            \
4553     }                                                                \
4554     static const TypeInfo spapr_machine_##suffix##_info = {          \
4555         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4556         .parent = TYPE_SPAPR_MACHINE,                                \
4557         .class_init = spapr_machine_##suffix##_class_init,           \
4558     };                                                               \
4559     static void spapr_machine_register_##suffix(void)                \
4560     {                                                                \
4561         type_register(&spapr_machine_##suffix##_info);               \
4562     }                                                                \
4563     type_init(spapr_machine_register_##suffix)
4564 
4565 /*
4566  * pseries-5.0
4567  */
4568 static void spapr_machine_5_0_class_options(MachineClass *mc)
4569 {
4570     /* Defaults for the latest behaviour inherited from the base class */
4571 }
4572 
4573 DEFINE_SPAPR_MACHINE(5_0, "5.0", true);
4574 
4575 /*
4576  * pseries-4.2
4577  */
4578 static void spapr_machine_4_2_class_options(MachineClass *mc)
4579 {
4580     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4581 
4582     spapr_machine_5_0_class_options(mc);
4583     compat_props_add(mc->compat_props, hw_compat_4_2, hw_compat_4_2_len);
4584     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4585     smc->default_caps.caps[SPAPR_CAP_FWNMI_MCE] = SPAPR_CAP_OFF;
4586     mc->nvdimm_supported = false;
4587 }
4588 
4589 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
4590 
4591 /*
4592  * pseries-4.1
4593  */
4594 static void spapr_machine_4_1_class_options(MachineClass *mc)
4595 {
4596     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4597     static GlobalProperty compat[] = {
4598         /* Only allow 4kiB and 64kiB IOMMU pagesizes */
4599         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pgsz", "0x11000" },
4600     };
4601 
4602     spapr_machine_4_2_class_options(mc);
4603     smc->linux_pci_probe = false;
4604     smc->smp_threads_vsmt = false;
4605     compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
4606     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4607 }
4608 
4609 DEFINE_SPAPR_MACHINE(4_1, "4.1", false);
4610 
4611 /*
4612  * pseries-4.0
4613  */
4614 static void phb_placement_4_0(SpaprMachineState *spapr, uint32_t index,
4615                               uint64_t *buid, hwaddr *pio,
4616                               hwaddr *mmio32, hwaddr *mmio64,
4617                               unsigned n_dma, uint32_t *liobns,
4618                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4619 {
4620     spapr_phb_placement(spapr, index, buid, pio, mmio32, mmio64, n_dma, liobns,
4621                         nv2gpa, nv2atsd, errp);
4622     *nv2gpa = 0;
4623     *nv2atsd = 0;
4624 }
4625 
4626 static void spapr_machine_4_0_class_options(MachineClass *mc)
4627 {
4628     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4629 
4630     spapr_machine_4_1_class_options(mc);
4631     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4632     smc->phb_placement = phb_placement_4_0;
4633     smc->irq = &spapr_irq_xics;
4634     smc->pre_4_1_migration = true;
4635 }
4636 
4637 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4638 
4639 /*
4640  * pseries-3.1
4641  */
4642 static void spapr_machine_3_1_class_options(MachineClass *mc)
4643 {
4644     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4645 
4646     spapr_machine_4_0_class_options(mc);
4647     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4648 
4649     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4650     smc->update_dt_enabled = false;
4651     smc->dr_phb_enabled = false;
4652     smc->broken_host_serial_model = true;
4653     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4654     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4655     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4656     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4657 }
4658 
4659 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4660 
4661 /*
4662  * pseries-3.0
4663  */
4664 
4665 static void spapr_machine_3_0_class_options(MachineClass *mc)
4666 {
4667     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4668 
4669     spapr_machine_3_1_class_options(mc);
4670     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4671 
4672     smc->legacy_irq_allocation = true;
4673     smc->nr_xirqs = 0x400;
4674     smc->irq = &spapr_irq_xics_legacy;
4675 }
4676 
4677 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4678 
4679 /*
4680  * pseries-2.12
4681  */
4682 static void spapr_machine_2_12_class_options(MachineClass *mc)
4683 {
4684     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4685     static GlobalProperty compat[] = {
4686         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4687         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4688     };
4689 
4690     spapr_machine_3_0_class_options(mc);
4691     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4692     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4693 
4694     /* We depend on kvm_enabled() to choose a default value for the
4695      * hpt-max-page-size capability. Of course we can't do it here
4696      * because this is too early and the HW accelerator isn't initialzed
4697      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4698      */
4699     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4700 }
4701 
4702 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4703 
4704 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4705 {
4706     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4707 
4708     spapr_machine_2_12_class_options(mc);
4709     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4710     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4711     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4712 }
4713 
4714 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4715 
4716 /*
4717  * pseries-2.11
4718  */
4719 
4720 static void spapr_machine_2_11_class_options(MachineClass *mc)
4721 {
4722     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4723 
4724     spapr_machine_2_12_class_options(mc);
4725     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4726     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4727 }
4728 
4729 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4730 
4731 /*
4732  * pseries-2.10
4733  */
4734 
4735 static void spapr_machine_2_10_class_options(MachineClass *mc)
4736 {
4737     spapr_machine_2_11_class_options(mc);
4738     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4739 }
4740 
4741 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4742 
4743 /*
4744  * pseries-2.9
4745  */
4746 
4747 static void spapr_machine_2_9_class_options(MachineClass *mc)
4748 {
4749     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4750     static GlobalProperty compat[] = {
4751         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4752     };
4753 
4754     spapr_machine_2_10_class_options(mc);
4755     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4756     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4757     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4758     smc->pre_2_10_has_unused_icps = true;
4759     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4760 }
4761 
4762 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4763 
4764 /*
4765  * pseries-2.8
4766  */
4767 
4768 static void spapr_machine_2_8_class_options(MachineClass *mc)
4769 {
4770     static GlobalProperty compat[] = {
4771         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4772     };
4773 
4774     spapr_machine_2_9_class_options(mc);
4775     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4776     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4777     mc->numa_mem_align_shift = 23;
4778 }
4779 
4780 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4781 
4782 /*
4783  * pseries-2.7
4784  */
4785 
4786 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4787                               uint64_t *buid, hwaddr *pio,
4788                               hwaddr *mmio32, hwaddr *mmio64,
4789                               unsigned n_dma, uint32_t *liobns,
4790                               hwaddr *nv2gpa, hwaddr *nv2atsd, Error **errp)
4791 {
4792     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4793     const uint64_t base_buid = 0x800000020000000ULL;
4794     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4795     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4796     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4797     const uint32_t max_index = 255;
4798     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4799 
4800     uint64_t ram_top = MACHINE(spapr)->ram_size;
4801     hwaddr phb0_base, phb_base;
4802     int i;
4803 
4804     /* Do we have device memory? */
4805     if (MACHINE(spapr)->maxram_size > ram_top) {
4806         /* Can't just use maxram_size, because there may be an
4807          * alignment gap between normal and device memory regions
4808          */
4809         ram_top = MACHINE(spapr)->device_memory->base +
4810             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4811     }
4812 
4813     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4814 
4815     if (index > max_index) {
4816         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4817                    max_index);
4818         return;
4819     }
4820 
4821     *buid = base_buid + index;
4822     for (i = 0; i < n_dma; ++i) {
4823         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4824     }
4825 
4826     phb_base = phb0_base + index * phb_spacing;
4827     *pio = phb_base + pio_offset;
4828     *mmio32 = phb_base + mmio_offset;
4829     /*
4830      * We don't set the 64-bit MMIO window, relying on the PHB's
4831      * fallback behaviour of automatically splitting a large "32-bit"
4832      * window into contiguous 32-bit and 64-bit windows
4833      */
4834 
4835     *nv2gpa = 0;
4836     *nv2atsd = 0;
4837 }
4838 
4839 static void spapr_machine_2_7_class_options(MachineClass *mc)
4840 {
4841     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4842     static GlobalProperty compat[] = {
4843         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4844         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4845         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4846         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4847     };
4848 
4849     spapr_machine_2_8_class_options(mc);
4850     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4851     mc->default_machine_opts = "modern-hotplug-events=off";
4852     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4853     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4854     smc->phb_placement = phb_placement_2_7;
4855 }
4856 
4857 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4858 
4859 /*
4860  * pseries-2.6
4861  */
4862 
4863 static void spapr_machine_2_6_class_options(MachineClass *mc)
4864 {
4865     static GlobalProperty compat[] = {
4866         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4867     };
4868 
4869     spapr_machine_2_7_class_options(mc);
4870     mc->has_hotpluggable_cpus = false;
4871     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4872     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4873 }
4874 
4875 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4876 
4877 /*
4878  * pseries-2.5
4879  */
4880 
4881 static void spapr_machine_2_5_class_options(MachineClass *mc)
4882 {
4883     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4884     static GlobalProperty compat[] = {
4885         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4886     };
4887 
4888     spapr_machine_2_6_class_options(mc);
4889     smc->use_ohci_by_default = true;
4890     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4891     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4892 }
4893 
4894 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4895 
4896 /*
4897  * pseries-2.4
4898  */
4899 
4900 static void spapr_machine_2_4_class_options(MachineClass *mc)
4901 {
4902     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4903 
4904     spapr_machine_2_5_class_options(mc);
4905     smc->dr_lmb_enabled = false;
4906     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4907 }
4908 
4909 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4910 
4911 /*
4912  * pseries-2.3
4913  */
4914 
4915 static void spapr_machine_2_3_class_options(MachineClass *mc)
4916 {
4917     static GlobalProperty compat[] = {
4918         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4919     };
4920     spapr_machine_2_4_class_options(mc);
4921     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4922     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4923 }
4924 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4925 
4926 /*
4927  * pseries-2.2
4928  */
4929 
4930 static void spapr_machine_2_2_class_options(MachineClass *mc)
4931 {
4932     static GlobalProperty compat[] = {
4933         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4934     };
4935 
4936     spapr_machine_2_3_class_options(mc);
4937     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4938     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4939     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4940 }
4941 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4942 
4943 /*
4944  * pseries-2.1
4945  */
4946 
4947 static void spapr_machine_2_1_class_options(MachineClass *mc)
4948 {
4949     spapr_machine_2_2_class_options(mc);
4950     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4951 }
4952 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4953 
4954 static void spapr_machine_register_types(void)
4955 {
4956     type_register_static(&spapr_machine_info);
4957 }
4958 
4959 type_init(spapr_machine_register_types)
4960