xref: /openbmc/qemu/hw/ppc/spapr.c (revision 8cf108c5)
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 "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "sysemu/qtest.h"
33 #include "hw/hw.h"
34 #include "qemu/log.h"
35 #include "hw/fw-path-provider.h"
36 #include "elf.h"
37 #include "net/net.h"
38 #include "sysemu/device_tree.h"
39 #include "sysemu/cpus.h"
40 #include "sysemu/hw_accel.h"
41 #include "kvm_ppc.h"
42 #include "migration/misc.h"
43 #include "migration/global_state.h"
44 #include "migration/register.h"
45 #include "mmu-hash64.h"
46 #include "mmu-book3s-v3.h"
47 #include "cpu-models.h"
48 #include "qom/cpu.h"
49 
50 #include "hw/boards.h"
51 #include "hw/ppc/ppc.h"
52 #include "hw/loader.h"
53 
54 #include "hw/ppc/fdt.h"
55 #include "hw/ppc/spapr.h"
56 #include "hw/ppc/spapr_vio.h"
57 #include "hw/pci-host/spapr.h"
58 #include "hw/pci/msi.h"
59 
60 #include "hw/pci/pci.h"
61 #include "hw/scsi/scsi.h"
62 #include "hw/virtio/virtio-scsi.h"
63 #include "hw/virtio/vhost-scsi-common.h"
64 
65 #include "exec/address-spaces.h"
66 #include "exec/ram_addr.h"
67 #include "hw/usb.h"
68 #include "qemu/config-file.h"
69 #include "qemu/error-report.h"
70 #include "trace.h"
71 #include "hw/nmi.h"
72 #include "hw/intc/intc.h"
73 
74 #include "qemu/cutils.h"
75 #include "hw/ppc/spapr_cpu_core.h"
76 #include "hw/mem/memory-device.h"
77 
78 #include <libfdt.h>
79 
80 /* SLOF memory layout:
81  *
82  * SLOF raw image loaded at 0, copies its romfs right below the flat
83  * device-tree, then position SLOF itself 31M below that
84  *
85  * So we set FW_OVERHEAD to 40MB which should account for all of that
86  * and more
87  *
88  * We load our kernel at 4M, leaving space for SLOF initial image
89  */
90 #define FDT_MAX_SIZE            0x100000
91 #define RTAS_MAX_SIZE           0x10000
92 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
93 #define FW_MAX_SIZE             0x400000
94 #define FW_FILE_NAME            "slof.bin"
95 #define FW_OVERHEAD             0x2800000
96 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
97 
98 #define MIN_RMA_SLOF            128UL
99 
100 #define PHANDLE_INTC            0x00001111
101 
102 /* These two functions implement the VCPU id numbering: one to compute them
103  * all and one to identify thread 0 of a VCORE. Any change to the first one
104  * is likely to have an impact on the second one, so let's keep them close.
105  */
106 static int spapr_vcpu_id(SpaprMachineState *spapr, int cpu_index)
107 {
108     assert(spapr->vsmt);
109     return
110         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
111 }
112 static bool spapr_is_thread0_in_vcore(SpaprMachineState *spapr,
113                                       PowerPCCPU *cpu)
114 {
115     assert(spapr->vsmt);
116     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
117 }
118 
119 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
120 {
121     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
122      * and newer QEMUs don't even have them. In both cases, we don't want
123      * to send anything on the wire.
124      */
125     return false;
126 }
127 
128 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
129     .name = "icp/server",
130     .version_id = 1,
131     .minimum_version_id = 1,
132     .needed = pre_2_10_vmstate_dummy_icp_needed,
133     .fields = (VMStateField[]) {
134         VMSTATE_UNUSED(4), /* uint32_t xirr */
135         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
136         VMSTATE_UNUSED(1), /* uint8_t mfrr */
137         VMSTATE_END_OF_LIST()
138     },
139 };
140 
141 static void pre_2_10_vmstate_register_dummy_icp(int i)
142 {
143     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
144                      (void *)(uintptr_t) i);
145 }
146 
147 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
148 {
149     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
150                        (void *)(uintptr_t) i);
151 }
152 
153 int spapr_max_server_number(SpaprMachineState *spapr)
154 {
155     assert(spapr->vsmt);
156     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
157 }
158 
159 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
160                                   int smt_threads)
161 {
162     int i, ret = 0;
163     uint32_t servers_prop[smt_threads];
164     uint32_t gservers_prop[smt_threads * 2];
165     int index = spapr_get_vcpu_id(cpu);
166 
167     if (cpu->compat_pvr) {
168         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
169         if (ret < 0) {
170             return ret;
171         }
172     }
173 
174     /* Build interrupt servers and gservers properties */
175     for (i = 0; i < smt_threads; i++) {
176         servers_prop[i] = cpu_to_be32(index + i);
177         /* Hack, direct the group queues back to cpu 0 */
178         gservers_prop[i*2] = cpu_to_be32(index + i);
179         gservers_prop[i*2 + 1] = 0;
180     }
181     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
182                       servers_prop, sizeof(servers_prop));
183     if (ret < 0) {
184         return ret;
185     }
186     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
187                       gservers_prop, sizeof(gservers_prop));
188 
189     return ret;
190 }
191 
192 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
193 {
194     int index = spapr_get_vcpu_id(cpu);
195     uint32_t associativity[] = {cpu_to_be32(0x5),
196                                 cpu_to_be32(0x0),
197                                 cpu_to_be32(0x0),
198                                 cpu_to_be32(0x0),
199                                 cpu_to_be32(cpu->node_id),
200                                 cpu_to_be32(index)};
201 
202     /* Advertise NUMA via ibm,associativity */
203     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
204                           sizeof(associativity));
205 }
206 
207 /* Populate the "ibm,pa-features" property */
208 static void spapr_populate_pa_features(SpaprMachineState *spapr,
209                                        PowerPCCPU *cpu,
210                                        void *fdt, int offset,
211                                        bool legacy_guest)
212 {
213     uint8_t pa_features_206[] = { 6, 0,
214         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
215     uint8_t pa_features_207[] = { 24, 0,
216         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
217         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
218         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
219         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
220     uint8_t pa_features_300[] = { 66, 0,
221         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
222         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
223         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
224         /* 6: DS207 */
225         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
226         /* 16: Vector */
227         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
228         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
229         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
230         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
231         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
232         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
233         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
234         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
236         /* 42: PM, 44: PC RA, 46: SC vec'd */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
238         /* 48: SIMD, 50: QP BFP, 52: String */
239         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
240         /* 54: DecFP, 56: DecI, 58: SHA */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
242         /* 60: NM atomic, 62: RNG */
243         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
244     };
245     uint8_t *pa_features = NULL;
246     size_t pa_size;
247 
248     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
249         pa_features = pa_features_206;
250         pa_size = sizeof(pa_features_206);
251     }
252     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
253         pa_features = pa_features_207;
254         pa_size = sizeof(pa_features_207);
255     }
256     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
257         pa_features = pa_features_300;
258         pa_size = sizeof(pa_features_300);
259     }
260     if (!pa_features) {
261         return;
262     }
263 
264     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
265         /*
266          * Note: we keep CI large pages off by default because a 64K capable
267          * guest provisioned with large pages might otherwise try to map a qemu
268          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
269          * even if that qemu runs on a 4k host.
270          * We dd this bit back here if we are confident this is not an issue
271          */
272         pa_features[3] |= 0x20;
273     }
274     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
275         pa_features[24] |= 0x80;    /* Transactional memory support */
276     }
277     if (legacy_guest && pa_size > 40) {
278         /* Workaround for broken kernels that attempt (guest) radix
279          * mode when they can't handle it, if they see the radix bit set
280          * in pa-features. So hide it from them. */
281         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
282     }
283 
284     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
285 }
286 
287 static int spapr_fixup_cpu_dt(void *fdt, SpaprMachineState *spapr)
288 {
289     int ret = 0, offset, cpus_offset;
290     CPUState *cs;
291     char cpu_model[32];
292     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
293 
294     CPU_FOREACH(cs) {
295         PowerPCCPU *cpu = POWERPC_CPU(cs);
296         DeviceClass *dc = DEVICE_GET_CLASS(cs);
297         int index = spapr_get_vcpu_id(cpu);
298         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
299 
300         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
301             continue;
302         }
303 
304         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
305 
306         cpus_offset = fdt_path_offset(fdt, "/cpus");
307         if (cpus_offset < 0) {
308             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
309             if (cpus_offset < 0) {
310                 return cpus_offset;
311             }
312         }
313         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
314         if (offset < 0) {
315             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
316             if (offset < 0) {
317                 return offset;
318             }
319         }
320 
321         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
322                           pft_size_prop, sizeof(pft_size_prop));
323         if (ret < 0) {
324             return ret;
325         }
326 
327         if (nb_numa_nodes > 1) {
328             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
329             if (ret < 0) {
330                 return ret;
331             }
332         }
333 
334         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
335         if (ret < 0) {
336             return ret;
337         }
338 
339         spapr_populate_pa_features(spapr, cpu, fdt, offset,
340                                    spapr->cas_legacy_guest_workaround);
341     }
342     return ret;
343 }
344 
345 static hwaddr spapr_node0_size(MachineState *machine)
346 {
347     if (nb_numa_nodes) {
348         int i;
349         for (i = 0; i < nb_numa_nodes; ++i) {
350             if (numa_info[i].node_mem) {
351                 return MIN(pow2floor(numa_info[i].node_mem),
352                            machine->ram_size);
353             }
354         }
355     }
356     return machine->ram_size;
357 }
358 
359 static void add_str(GString *s, const gchar *s1)
360 {
361     g_string_append_len(s, s1, strlen(s1) + 1);
362 }
363 
364 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
365                                        hwaddr size)
366 {
367     uint32_t associativity[] = {
368         cpu_to_be32(0x4), /* length */
369         cpu_to_be32(0x0), cpu_to_be32(0x0),
370         cpu_to_be32(0x0), cpu_to_be32(nodeid)
371     };
372     char mem_name[32];
373     uint64_t mem_reg_property[2];
374     int off;
375 
376     mem_reg_property[0] = cpu_to_be64(start);
377     mem_reg_property[1] = cpu_to_be64(size);
378 
379     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
380     off = fdt_add_subnode(fdt, 0, mem_name);
381     _FDT(off);
382     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
383     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
384                       sizeof(mem_reg_property))));
385     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
386                       sizeof(associativity))));
387     return off;
388 }
389 
390 static int spapr_populate_memory(SpaprMachineState *spapr, void *fdt)
391 {
392     MachineState *machine = MACHINE(spapr);
393     hwaddr mem_start, node_size;
394     int i, nb_nodes = nb_numa_nodes;
395     NodeInfo *nodes = numa_info;
396     NodeInfo ramnode;
397 
398     /* No NUMA nodes, assume there is just one node with whole RAM */
399     if (!nb_numa_nodes) {
400         nb_nodes = 1;
401         ramnode.node_mem = machine->ram_size;
402         nodes = &ramnode;
403     }
404 
405     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
406         if (!nodes[i].node_mem) {
407             continue;
408         }
409         if (mem_start >= machine->ram_size) {
410             node_size = 0;
411         } else {
412             node_size = nodes[i].node_mem;
413             if (node_size > machine->ram_size - mem_start) {
414                 node_size = machine->ram_size - mem_start;
415             }
416         }
417         if (!mem_start) {
418             /* spapr_machine_init() checks for rma_size <= node0_size
419              * already */
420             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
421             mem_start += spapr->rma_size;
422             node_size -= spapr->rma_size;
423         }
424         for ( ; node_size; ) {
425             hwaddr sizetmp = pow2floor(node_size);
426 
427             /* mem_start != 0 here */
428             if (ctzl(mem_start) < ctzl(sizetmp)) {
429                 sizetmp = 1ULL << ctzl(mem_start);
430             }
431 
432             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
433             node_size -= sizetmp;
434             mem_start += sizetmp;
435         }
436     }
437 
438     return 0;
439 }
440 
441 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
442                                   SpaprMachineState *spapr)
443 {
444     PowerPCCPU *cpu = POWERPC_CPU(cs);
445     CPUPPCState *env = &cpu->env;
446     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
447     int index = spapr_get_vcpu_id(cpu);
448     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
449                        0xffffffff, 0xffffffff};
450     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
451         : SPAPR_TIMEBASE_FREQ;
452     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
453     uint32_t page_sizes_prop[64];
454     size_t page_sizes_prop_size;
455     uint32_t vcpus_per_socket = smp_threads * smp_cores;
456     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
457     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
458     SpaprDrc *drc;
459     int drc_index;
460     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
461     int i;
462 
463     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
464     if (drc) {
465         drc_index = spapr_drc_index(drc);
466         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
467     }
468 
469     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
470     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
471 
472     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
473     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
474                            env->dcache_line_size)));
475     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
476                            env->dcache_line_size)));
477     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
478                            env->icache_line_size)));
479     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
480                            env->icache_line_size)));
481 
482     if (pcc->l1_dcache_size) {
483         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
484                                pcc->l1_dcache_size)));
485     } else {
486         warn_report("Unknown L1 dcache size for cpu");
487     }
488     if (pcc->l1_icache_size) {
489         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
490                                pcc->l1_icache_size)));
491     } else {
492         warn_report("Unknown L1 icache size for cpu");
493     }
494 
495     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
496     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
497     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
498     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
499     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
500     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
501 
502     if (env->spr_cb[SPR_PURR].oea_read) {
503         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
504     }
505 
506     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
507         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
508                           segs, sizeof(segs))));
509     }
510 
511     /* Advertise VSX (vector extensions) if available
512      *   1               == VMX / Altivec available
513      *   2               == VSX available
514      *
515      * Only CPUs for which we create core types in spapr_cpu_core.c
516      * are possible, and all of those have VMX */
517     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
518         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
519     } else {
520         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
521     }
522 
523     /* Advertise DFP (Decimal Floating Point) if available
524      *   0 / no property == no DFP
525      *   1               == DFP available */
526     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
527         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
528     }
529 
530     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
531                                                       sizeof(page_sizes_prop));
532     if (page_sizes_prop_size) {
533         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
534                           page_sizes_prop, page_sizes_prop_size)));
535     }
536 
537     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
538 
539     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
540                            cs->cpu_index / vcpus_per_socket)));
541 
542     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
543                       pft_size_prop, sizeof(pft_size_prop))));
544 
545     if (nb_numa_nodes > 1) {
546         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
547     }
548 
549     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
550 
551     if (pcc->radix_page_info) {
552         for (i = 0; i < pcc->radix_page_info->count; i++) {
553             radix_AP_encodings[i] =
554                 cpu_to_be32(pcc->radix_page_info->entries[i]);
555         }
556         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
557                           radix_AP_encodings,
558                           pcc->radix_page_info->count *
559                           sizeof(radix_AP_encodings[0]))));
560     }
561 
562     /*
563      * We set this property to let the guest know that it can use the large
564      * decrementer and its width in bits.
565      */
566     if (spapr_get_cap(spapr, SPAPR_CAP_LARGE_DECREMENTER) != SPAPR_CAP_OFF)
567         _FDT((fdt_setprop_u32(fdt, offset, "ibm,dec-bits",
568                               pcc->lrg_decr_bits)));
569 }
570 
571 static void spapr_populate_cpus_dt_node(void *fdt, SpaprMachineState *spapr)
572 {
573     CPUState **rev;
574     CPUState *cs;
575     int n_cpus;
576     int cpus_offset;
577     char *nodename;
578     int i;
579 
580     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
581     _FDT(cpus_offset);
582     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
583     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
584 
585     /*
586      * We walk the CPUs in reverse order to ensure that CPU DT nodes
587      * created by fdt_add_subnode() end up in the right order in FDT
588      * for the guest kernel the enumerate the CPUs correctly.
589      *
590      * The CPU list cannot be traversed in reverse order, so we need
591      * to do extra work.
592      */
593     n_cpus = 0;
594     rev = NULL;
595     CPU_FOREACH(cs) {
596         rev = g_renew(CPUState *, rev, n_cpus + 1);
597         rev[n_cpus++] = cs;
598     }
599 
600     for (i = n_cpus - 1; i >= 0; i--) {
601         CPUState *cs = rev[i];
602         PowerPCCPU *cpu = POWERPC_CPU(cs);
603         int index = spapr_get_vcpu_id(cpu);
604         DeviceClass *dc = DEVICE_GET_CLASS(cs);
605         int offset;
606 
607         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
608             continue;
609         }
610 
611         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
612         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
613         g_free(nodename);
614         _FDT(offset);
615         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
616     }
617 
618     g_free(rev);
619 }
620 
621 static int spapr_rng_populate_dt(void *fdt)
622 {
623     int node;
624     int ret;
625 
626     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
627     if (node <= 0) {
628         return -1;
629     }
630     ret = fdt_setprop_string(fdt, node, "device_type",
631                              "ibm,platform-facilities");
632     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
633     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
634 
635     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
636     if (node <= 0) {
637         return -1;
638     }
639     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
640 
641     return ret ? -1 : 0;
642 }
643 
644 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
645 {
646     MemoryDeviceInfoList *info;
647 
648     for (info = list; info; info = info->next) {
649         MemoryDeviceInfo *value = info->value;
650 
651         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
652             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
653 
654             if (addr >= pcdimm_info->addr &&
655                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
656                 return pcdimm_info->node;
657             }
658         }
659     }
660 
661     return -1;
662 }
663 
664 struct sPAPRDrconfCellV2 {
665      uint32_t seq_lmbs;
666      uint64_t base_addr;
667      uint32_t drc_index;
668      uint32_t aa_index;
669      uint32_t flags;
670 } QEMU_PACKED;
671 
672 typedef struct DrconfCellQueue {
673     struct sPAPRDrconfCellV2 cell;
674     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
675 } DrconfCellQueue;
676 
677 static DrconfCellQueue *
678 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
679                       uint32_t drc_index, uint32_t aa_index,
680                       uint32_t flags)
681 {
682     DrconfCellQueue *elem;
683 
684     elem = g_malloc0(sizeof(*elem));
685     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
686     elem->cell.base_addr = cpu_to_be64(base_addr);
687     elem->cell.drc_index = cpu_to_be32(drc_index);
688     elem->cell.aa_index = cpu_to_be32(aa_index);
689     elem->cell.flags = cpu_to_be32(flags);
690 
691     return elem;
692 }
693 
694 /* ibm,dynamic-memory-v2 */
695 static int spapr_populate_drmem_v2(SpaprMachineState *spapr, void *fdt,
696                                    int offset, MemoryDeviceInfoList *dimms)
697 {
698     MachineState *machine = MACHINE(spapr);
699     uint8_t *int_buf, *cur_index;
700     int ret;
701     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
702     uint64_t addr, cur_addr, size;
703     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
704     uint64_t mem_end = machine->device_memory->base +
705                        memory_region_size(&machine->device_memory->mr);
706     uint32_t node, buf_len, nr_entries = 0;
707     SpaprDrc *drc;
708     DrconfCellQueue *elem, *next;
709     MemoryDeviceInfoList *info;
710     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
711         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
712 
713     /* Entry to cover RAM and the gap area */
714     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
715                                  SPAPR_LMB_FLAGS_RESERVED |
716                                  SPAPR_LMB_FLAGS_DRC_INVALID);
717     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
718     nr_entries++;
719 
720     cur_addr = machine->device_memory->base;
721     for (info = dimms; info; info = info->next) {
722         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
723 
724         addr = di->addr;
725         size = di->size;
726         node = di->node;
727 
728         /* Entry for hot-pluggable area */
729         if (cur_addr < addr) {
730             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
731             g_assert(drc);
732             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
733                                          cur_addr, spapr_drc_index(drc), -1, 0);
734             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
735             nr_entries++;
736         }
737 
738         /* Entry for DIMM */
739         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
740         g_assert(drc);
741         elem = spapr_get_drconf_cell(size / lmb_size, addr,
742                                      spapr_drc_index(drc), node,
743                                      SPAPR_LMB_FLAGS_ASSIGNED);
744         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
745         nr_entries++;
746         cur_addr = addr + size;
747     }
748 
749     /* Entry for remaining hotpluggable area */
750     if (cur_addr < mem_end) {
751         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
752         g_assert(drc);
753         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
754                                      cur_addr, spapr_drc_index(drc), -1, 0);
755         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
756         nr_entries++;
757     }
758 
759     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
760     int_buf = cur_index = g_malloc0(buf_len);
761     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
762     cur_index += sizeof(nr_entries);
763 
764     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
765         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
766         cur_index += sizeof(elem->cell);
767         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
768         g_free(elem);
769     }
770 
771     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
772     g_free(int_buf);
773     if (ret < 0) {
774         return -1;
775     }
776     return 0;
777 }
778 
779 /* ibm,dynamic-memory */
780 static int spapr_populate_drmem_v1(SpaprMachineState *spapr, void *fdt,
781                                    int offset, MemoryDeviceInfoList *dimms)
782 {
783     MachineState *machine = MACHINE(spapr);
784     int i, ret;
785     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
786     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
787     uint32_t nr_lmbs = (machine->device_memory->base +
788                        memory_region_size(&machine->device_memory->mr)) /
789                        lmb_size;
790     uint32_t *int_buf, *cur_index, buf_len;
791 
792     /*
793      * Allocate enough buffer size to fit in ibm,dynamic-memory
794      */
795     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
796     cur_index = int_buf = g_malloc0(buf_len);
797     int_buf[0] = cpu_to_be32(nr_lmbs);
798     cur_index++;
799     for (i = 0; i < nr_lmbs; i++) {
800         uint64_t addr = i * lmb_size;
801         uint32_t *dynamic_memory = cur_index;
802 
803         if (i >= device_lmb_start) {
804             SpaprDrc *drc;
805 
806             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
807             g_assert(drc);
808 
809             dynamic_memory[0] = cpu_to_be32(addr >> 32);
810             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
811             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
812             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
813             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
814             if (memory_region_present(get_system_memory(), addr)) {
815                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
816             } else {
817                 dynamic_memory[5] = cpu_to_be32(0);
818             }
819         } else {
820             /*
821              * LMB information for RMA, boot time RAM and gap b/n RAM and
822              * device memory region -- all these are marked as reserved
823              * and as having no valid DRC.
824              */
825             dynamic_memory[0] = cpu_to_be32(addr >> 32);
826             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
827             dynamic_memory[2] = cpu_to_be32(0);
828             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
829             dynamic_memory[4] = cpu_to_be32(-1);
830             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
831                                             SPAPR_LMB_FLAGS_DRC_INVALID);
832         }
833 
834         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
835     }
836     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
837     g_free(int_buf);
838     if (ret < 0) {
839         return -1;
840     }
841     return 0;
842 }
843 
844 /*
845  * Adds ibm,dynamic-reconfiguration-memory node.
846  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
847  * of this device tree node.
848  */
849 static int spapr_populate_drconf_memory(SpaprMachineState *spapr, void *fdt)
850 {
851     MachineState *machine = MACHINE(spapr);
852     int ret, i, offset;
853     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
854     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
855     uint32_t *int_buf, *cur_index, buf_len;
856     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
857     MemoryDeviceInfoList *dimms = NULL;
858 
859     /*
860      * Don't create the node if there is no device memory
861      */
862     if (machine->ram_size == machine->maxram_size) {
863         return 0;
864     }
865 
866     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
867 
868     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
869                     sizeof(prop_lmb_size));
870     if (ret < 0) {
871         return ret;
872     }
873 
874     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
875     if (ret < 0) {
876         return ret;
877     }
878 
879     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
880     if (ret < 0) {
881         return ret;
882     }
883 
884     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
885     dimms = qmp_memory_device_list();
886     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
887         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
888     } else {
889         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
890     }
891     qapi_free_MemoryDeviceInfoList(dimms);
892 
893     if (ret < 0) {
894         return ret;
895     }
896 
897     /* ibm,associativity-lookup-arrays */
898     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
899     cur_index = int_buf = g_malloc0(buf_len);
900     int_buf[0] = cpu_to_be32(nr_nodes);
901     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
902     cur_index += 2;
903     for (i = 0; i < nr_nodes; i++) {
904         uint32_t associativity[] = {
905             cpu_to_be32(0x0),
906             cpu_to_be32(0x0),
907             cpu_to_be32(0x0),
908             cpu_to_be32(i)
909         };
910         memcpy(cur_index, associativity, sizeof(associativity));
911         cur_index += 4;
912     }
913     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
914             (cur_index - int_buf) * sizeof(uint32_t));
915     g_free(int_buf);
916 
917     return ret;
918 }
919 
920 static int spapr_dt_cas_updates(SpaprMachineState *spapr, void *fdt,
921                                 SpaprOptionVector *ov5_updates)
922 {
923     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
924     int ret = 0, offset;
925 
926     /* Generate ibm,dynamic-reconfiguration-memory node if required */
927     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
928         g_assert(smc->dr_lmb_enabled);
929         ret = spapr_populate_drconf_memory(spapr, fdt);
930         if (ret) {
931             goto out;
932         }
933     }
934 
935     offset = fdt_path_offset(fdt, "/chosen");
936     if (offset < 0) {
937         offset = fdt_add_subnode(fdt, 0, "chosen");
938         if (offset < 0) {
939             return offset;
940         }
941     }
942     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
943                                  "ibm,architecture-vec-5");
944 
945 out:
946     return ret;
947 }
948 
949 static bool spapr_hotplugged_dev_before_cas(void)
950 {
951     Object *drc_container, *obj;
952     ObjectProperty *prop;
953     ObjectPropertyIterator iter;
954 
955     drc_container = container_get(object_get_root(), "/dr-connector");
956     object_property_iter_init(&iter, drc_container);
957     while ((prop = object_property_iter_next(&iter))) {
958         if (!strstart(prop->type, "link<", NULL)) {
959             continue;
960         }
961         obj = object_property_get_link(drc_container, prop->name, NULL);
962         if (spapr_drc_needed(obj)) {
963             return true;
964         }
965     }
966     return false;
967 }
968 
969 int spapr_h_cas_compose_response(SpaprMachineState *spapr,
970                                  target_ulong addr, target_ulong size,
971                                  SpaprOptionVector *ov5_updates)
972 {
973     void *fdt, *fdt_skel;
974     SpaprDeviceTreeUpdateHeader hdr = { .version_id = 1 };
975 
976     if (spapr_hotplugged_dev_before_cas()) {
977         return 1;
978     }
979 
980     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
981         error_report("SLOF provided an unexpected CAS buffer size "
982                      TARGET_FMT_lu " (min: %zu, max: %u)",
983                      size, sizeof(hdr), FW_MAX_SIZE);
984         exit(EXIT_FAILURE);
985     }
986 
987     size -= sizeof(hdr);
988 
989     /* Create skeleton */
990     fdt_skel = g_malloc0(size);
991     _FDT((fdt_create(fdt_skel, size)));
992     _FDT((fdt_finish_reservemap(fdt_skel)));
993     _FDT((fdt_begin_node(fdt_skel, "")));
994     _FDT((fdt_end_node(fdt_skel)));
995     _FDT((fdt_finish(fdt_skel)));
996     fdt = g_malloc0(size);
997     _FDT((fdt_open_into(fdt_skel, fdt, size)));
998     g_free(fdt_skel);
999 
1000     /* Fixup cpu nodes */
1001     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
1002 
1003     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
1004         return -1;
1005     }
1006 
1007     /* Pack resulting tree */
1008     _FDT((fdt_pack(fdt)));
1009 
1010     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1011         trace_spapr_cas_failed(size);
1012         return -1;
1013     }
1014 
1015     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1016     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1017     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1018     g_free(fdt);
1019 
1020     return 0;
1021 }
1022 
1023 static void spapr_dt_rtas(SpaprMachineState *spapr, void *fdt)
1024 {
1025     int rtas;
1026     GString *hypertas = g_string_sized_new(256);
1027     GString *qemu_hypertas = g_string_sized_new(256);
1028     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1029     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1030         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1031     uint32_t lrdr_capacity[] = {
1032         cpu_to_be32(max_device_addr >> 32),
1033         cpu_to_be32(max_device_addr & 0xffffffff),
1034         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1035         cpu_to_be32(max_cpus / smp_threads),
1036     };
1037     uint32_t maxdomains[] = {
1038         cpu_to_be32(4),
1039         cpu_to_be32(0),
1040         cpu_to_be32(0),
1041         cpu_to_be32(0),
1042         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes : 1),
1043     };
1044 
1045     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1046 
1047     /* hypertas */
1048     add_str(hypertas, "hcall-pft");
1049     add_str(hypertas, "hcall-term");
1050     add_str(hypertas, "hcall-dabr");
1051     add_str(hypertas, "hcall-interrupt");
1052     add_str(hypertas, "hcall-tce");
1053     add_str(hypertas, "hcall-vio");
1054     add_str(hypertas, "hcall-splpar");
1055     add_str(hypertas, "hcall-bulk");
1056     add_str(hypertas, "hcall-set-mode");
1057     add_str(hypertas, "hcall-sprg0");
1058     add_str(hypertas, "hcall-copy");
1059     add_str(hypertas, "hcall-debug");
1060     add_str(hypertas, "hcall-vphn");
1061     add_str(qemu_hypertas, "hcall-memop1");
1062 
1063     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1064         add_str(hypertas, "hcall-multi-tce");
1065     }
1066 
1067     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1068         add_str(hypertas, "hcall-hpt-resize");
1069     }
1070 
1071     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1072                      hypertas->str, hypertas->len));
1073     g_string_free(hypertas, TRUE);
1074     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1075                      qemu_hypertas->str, qemu_hypertas->len));
1076     g_string_free(qemu_hypertas, TRUE);
1077 
1078     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1079                      refpoints, sizeof(refpoints)));
1080 
1081     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1082                      maxdomains, sizeof(maxdomains)));
1083 
1084     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1085                           RTAS_ERROR_LOG_MAX));
1086     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1087                           RTAS_EVENT_SCAN_RATE));
1088 
1089     g_assert(msi_nonbroken);
1090     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1091 
1092     /*
1093      * According to PAPR, rtas ibm,os-term does not guarantee a return
1094      * back to the guest cpu.
1095      *
1096      * While an additional ibm,extended-os-term property indicates
1097      * that rtas call return will always occur. Set this property.
1098      */
1099     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1100 
1101     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1102                      lrdr_capacity, sizeof(lrdr_capacity)));
1103 
1104     spapr_dt_rtas_tokens(fdt, rtas);
1105 }
1106 
1107 /*
1108  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1109  * and the XIVE features that the guest may request and thus the valid
1110  * values for bytes 23..26 of option vector 5:
1111  */
1112 static void spapr_dt_ov5_platform_support(SpaprMachineState *spapr, void *fdt,
1113                                           int chosen)
1114 {
1115     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1116 
1117     char val[2 * 4] = {
1118         23, spapr->irq->ov5, /* Xive mode. */
1119         24, 0x00, /* Hash/Radix, filled in below. */
1120         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1121         26, 0x40, /* Radix options: GTSE == yes. */
1122     };
1123 
1124     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1125                           first_ppc_cpu->compat_pvr)) {
1126         /*
1127          * If we're in a pre POWER9 compat mode then the guest should
1128          * do hash and use the legacy interrupt mode
1129          */
1130         val[1] = 0x00; /* XICS */
1131         val[3] = 0x00; /* Hash */
1132     } else if (kvm_enabled()) {
1133         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1134             val[3] = 0x80; /* OV5_MMU_BOTH */
1135         } else if (kvmppc_has_cap_mmu_radix()) {
1136             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1137         } else {
1138             val[3] = 0x00; /* Hash */
1139         }
1140     } else {
1141         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1142         val[3] = 0xC0;
1143     }
1144     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1145                      val, sizeof(val)));
1146 }
1147 
1148 static void spapr_dt_chosen(SpaprMachineState *spapr, void *fdt)
1149 {
1150     MachineState *machine = MACHINE(spapr);
1151     int chosen;
1152     const char *boot_device = machine->boot_order;
1153     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1154     size_t cb = 0;
1155     char *bootlist = get_boot_devices_list(&cb);
1156 
1157     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1158 
1159     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1160     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1161                           spapr->initrd_base));
1162     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1163                           spapr->initrd_base + spapr->initrd_size));
1164 
1165     if (spapr->kernel_size) {
1166         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1167                               cpu_to_be64(spapr->kernel_size) };
1168 
1169         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1170                          &kprop, sizeof(kprop)));
1171         if (spapr->kernel_le) {
1172             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1173         }
1174     }
1175     if (boot_menu) {
1176         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1177     }
1178     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1179     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1180     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1181 
1182     if (cb && bootlist) {
1183         int i;
1184 
1185         for (i = 0; i < cb; i++) {
1186             if (bootlist[i] == '\n') {
1187                 bootlist[i] = ' ';
1188             }
1189         }
1190         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1191     }
1192 
1193     if (boot_device && strlen(boot_device)) {
1194         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1195     }
1196 
1197     if (!spapr->has_graphics && stdout_path) {
1198         /*
1199          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1200          * kernel. New platforms should only use the "stdout-path" property. Set
1201          * the new property and continue using older property to remain
1202          * compatible with the existing firmware.
1203          */
1204         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1205         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1206     }
1207 
1208     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1209 
1210     g_free(stdout_path);
1211     g_free(bootlist);
1212 }
1213 
1214 static void spapr_dt_hypervisor(SpaprMachineState *spapr, void *fdt)
1215 {
1216     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1217      * KVM to work under pHyp with some guest co-operation */
1218     int hypervisor;
1219     uint8_t hypercall[16];
1220 
1221     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1222     /* indicate KVM hypercall interface */
1223     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1224     if (kvmppc_has_cap_fixup_hcalls()) {
1225         /*
1226          * Older KVM versions with older guest kernels were broken
1227          * with the magic page, don't allow the guest to map it.
1228          */
1229         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1230                                   sizeof(hypercall))) {
1231             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1232                              hypercall, sizeof(hypercall)));
1233         }
1234     }
1235 }
1236 
1237 static void *spapr_build_fdt(SpaprMachineState *spapr)
1238 {
1239     MachineState *machine = MACHINE(spapr);
1240     MachineClass *mc = MACHINE_GET_CLASS(machine);
1241     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1242     int ret;
1243     void *fdt;
1244     SpaprPhbState *phb;
1245     char *buf;
1246 
1247     fdt = g_malloc0(FDT_MAX_SIZE);
1248     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1249 
1250     /* Root node */
1251     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1252     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1253     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1254 
1255     /* Guest UUID & Name*/
1256     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1257     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1258     if (qemu_uuid_set) {
1259         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1260     }
1261     g_free(buf);
1262 
1263     if (qemu_get_vm_name()) {
1264         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1265                                 qemu_get_vm_name()));
1266     }
1267 
1268     /* Host Model & Serial Number */
1269     if (spapr->host_model) {
1270         _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1271     } else if (smc->broken_host_serial_model && kvmppc_get_host_model(&buf)) {
1272         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1273         g_free(buf);
1274     }
1275 
1276     if (spapr->host_serial) {
1277         _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1278     } else if (smc->broken_host_serial_model && kvmppc_get_host_serial(&buf)) {
1279         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1280         g_free(buf);
1281     }
1282 
1283     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1284     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1285 
1286     /* /interrupt controller */
1287     spapr->irq->dt_populate(spapr, spapr_max_server_number(spapr), fdt,
1288                           PHANDLE_INTC);
1289 
1290     ret = spapr_populate_memory(spapr, fdt);
1291     if (ret < 0) {
1292         error_report("couldn't setup memory nodes in fdt");
1293         exit(1);
1294     }
1295 
1296     /* /vdevice */
1297     spapr_dt_vdevice(spapr->vio_bus, fdt);
1298 
1299     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1300         ret = spapr_rng_populate_dt(fdt);
1301         if (ret < 0) {
1302             error_report("could not set up rng device in the fdt");
1303             exit(1);
1304         }
1305     }
1306 
1307     QLIST_FOREACH(phb, &spapr->phbs, list) {
1308         ret = spapr_populate_pci_dt(phb, PHANDLE_INTC, fdt,
1309                                     spapr->irq->nr_msis, NULL);
1310         if (ret < 0) {
1311             error_report("couldn't setup PCI devices in fdt");
1312             exit(1);
1313         }
1314     }
1315 
1316     /* cpus */
1317     spapr_populate_cpus_dt_node(fdt, spapr);
1318 
1319     if (smc->dr_lmb_enabled) {
1320         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1321     }
1322 
1323     if (mc->has_hotpluggable_cpus) {
1324         int offset = fdt_path_offset(fdt, "/cpus");
1325         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1326                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1327         if (ret < 0) {
1328             error_report("Couldn't set up CPU DR device tree properties");
1329             exit(1);
1330         }
1331     }
1332 
1333     /* /event-sources */
1334     spapr_dt_events(spapr, fdt);
1335 
1336     /* /rtas */
1337     spapr_dt_rtas(spapr, fdt);
1338 
1339     /* /chosen */
1340     spapr_dt_chosen(spapr, fdt);
1341 
1342     /* /hypervisor */
1343     if (kvm_enabled()) {
1344         spapr_dt_hypervisor(spapr, fdt);
1345     }
1346 
1347     /* Build memory reserve map */
1348     if (spapr->kernel_size) {
1349         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1350     }
1351     if (spapr->initrd_size) {
1352         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1353     }
1354 
1355     /* ibm,client-architecture-support updates */
1356     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1357     if (ret < 0) {
1358         error_report("couldn't setup CAS properties fdt");
1359         exit(1);
1360     }
1361 
1362     if (smc->dr_phb_enabled) {
1363         ret = spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1364         if (ret < 0) {
1365             error_report("Couldn't set up PHB DR device tree properties");
1366             exit(1);
1367         }
1368     }
1369 
1370     return fdt;
1371 }
1372 
1373 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1374 {
1375     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1376 }
1377 
1378 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1379                                     PowerPCCPU *cpu)
1380 {
1381     CPUPPCState *env = &cpu->env;
1382 
1383     /* The TCG path should also be holding the BQL at this point */
1384     g_assert(qemu_mutex_iothread_locked());
1385 
1386     if (msr_pr) {
1387         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1388         env->gpr[3] = H_PRIVILEGE;
1389     } else {
1390         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1391     }
1392 }
1393 
1394 struct LPCRSyncState {
1395     target_ulong value;
1396     target_ulong mask;
1397 };
1398 
1399 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1400 {
1401     struct LPCRSyncState *s = arg.host_ptr;
1402     PowerPCCPU *cpu = POWERPC_CPU(cs);
1403     CPUPPCState *env = &cpu->env;
1404     target_ulong lpcr;
1405 
1406     cpu_synchronize_state(cs);
1407     lpcr = env->spr[SPR_LPCR];
1408     lpcr &= ~s->mask;
1409     lpcr |= s->value;
1410     ppc_store_lpcr(cpu, lpcr);
1411 }
1412 
1413 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1414 {
1415     CPUState *cs;
1416     struct LPCRSyncState s = {
1417         .value = value,
1418         .mask = mask
1419     };
1420     CPU_FOREACH(cs) {
1421         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1422     }
1423 }
1424 
1425 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1426 {
1427     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1428 
1429     /* Copy PATE1:GR into PATE0:HR */
1430     entry->dw0 = spapr->patb_entry & PATE0_HR;
1431     entry->dw1 = spapr->patb_entry;
1432 }
1433 
1434 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1435 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1436 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1437 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1438 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1439 
1440 /*
1441  * Get the fd to access the kernel htab, re-opening it if necessary
1442  */
1443 static int get_htab_fd(SpaprMachineState *spapr)
1444 {
1445     Error *local_err = NULL;
1446 
1447     if (spapr->htab_fd >= 0) {
1448         return spapr->htab_fd;
1449     }
1450 
1451     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1452     if (spapr->htab_fd < 0) {
1453         error_report_err(local_err);
1454     }
1455 
1456     return spapr->htab_fd;
1457 }
1458 
1459 void close_htab_fd(SpaprMachineState *spapr)
1460 {
1461     if (spapr->htab_fd >= 0) {
1462         close(spapr->htab_fd);
1463     }
1464     spapr->htab_fd = -1;
1465 }
1466 
1467 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1468 {
1469     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1470 
1471     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1472 }
1473 
1474 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1475 {
1476     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1477 
1478     assert(kvm_enabled());
1479 
1480     if (!spapr->htab) {
1481         return 0;
1482     }
1483 
1484     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1485 }
1486 
1487 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1488                                                 hwaddr ptex, int n)
1489 {
1490     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1491     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1492 
1493     if (!spapr->htab) {
1494         /*
1495          * HTAB is controlled by KVM. Fetch into temporary buffer
1496          */
1497         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1498         kvmppc_read_hptes(hptes, ptex, n);
1499         return hptes;
1500     }
1501 
1502     /*
1503      * HTAB is controlled by QEMU. Just point to the internally
1504      * accessible PTEG.
1505      */
1506     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1507 }
1508 
1509 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1510                               const ppc_hash_pte64_t *hptes,
1511                               hwaddr ptex, int n)
1512 {
1513     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1514 
1515     if (!spapr->htab) {
1516         g_free((void *)hptes);
1517     }
1518 
1519     /* Nothing to do for qemu managed HPT */
1520 }
1521 
1522 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1523                              uint64_t pte0, uint64_t pte1)
1524 {
1525     SpaprMachineState *spapr = SPAPR_MACHINE(vhyp);
1526     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1527 
1528     if (!spapr->htab) {
1529         kvmppc_write_hpte(ptex, pte0, pte1);
1530     } else {
1531         if (pte0 & HPTE64_V_VALID) {
1532             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1533             /*
1534              * When setting valid, we write PTE1 first. This ensures
1535              * proper synchronization with the reading code in
1536              * ppc_hash64_pteg_search()
1537              */
1538             smp_wmb();
1539             stq_p(spapr->htab + offset, pte0);
1540         } else {
1541             stq_p(spapr->htab + offset, pte0);
1542             /*
1543              * When clearing it we set PTE0 first. This ensures proper
1544              * synchronization with the reading code in
1545              * ppc_hash64_pteg_search()
1546              */
1547             smp_wmb();
1548             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1549         }
1550     }
1551 }
1552 
1553 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1554 {
1555     int shift;
1556 
1557     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1558      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1559      * that's much more than is needed for Linux guests */
1560     shift = ctz64(pow2ceil(ramsize)) - 7;
1561     shift = MAX(shift, 18); /* Minimum architected size */
1562     shift = MIN(shift, 46); /* Maximum architected size */
1563     return shift;
1564 }
1565 
1566 void spapr_free_hpt(SpaprMachineState *spapr)
1567 {
1568     g_free(spapr->htab);
1569     spapr->htab = NULL;
1570     spapr->htab_shift = 0;
1571     close_htab_fd(spapr);
1572 }
1573 
1574 void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
1575                           Error **errp)
1576 {
1577     long rc;
1578 
1579     /* Clean up any HPT info from a previous boot */
1580     spapr_free_hpt(spapr);
1581 
1582     rc = kvmppc_reset_htab(shift);
1583     if (rc < 0) {
1584         /* kernel-side HPT needed, but couldn't allocate one */
1585         error_setg_errno(errp, errno,
1586                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1587                          shift);
1588         /* This is almost certainly fatal, but if the caller really
1589          * wants to carry on with shift == 0, it's welcome to try */
1590     } else if (rc > 0) {
1591         /* kernel-side HPT allocated */
1592         if (rc != shift) {
1593             error_setg(errp,
1594                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1595                        shift, rc);
1596         }
1597 
1598         spapr->htab_shift = shift;
1599         spapr->htab = NULL;
1600     } else {
1601         /* kernel-side HPT not needed, allocate in userspace instead */
1602         size_t size = 1ULL << shift;
1603         int i;
1604 
1605         spapr->htab = qemu_memalign(size, size);
1606         if (!spapr->htab) {
1607             error_setg_errno(errp, errno,
1608                              "Could not allocate HPT of order %d", shift);
1609             return;
1610         }
1611 
1612         memset(spapr->htab, 0, size);
1613         spapr->htab_shift = shift;
1614 
1615         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1616             DIRTY_HPTE(HPTE(spapr->htab, i));
1617         }
1618     }
1619     /* We're setting up a hash table, so that means we're not radix */
1620     spapr->patb_entry = 0;
1621     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1622 }
1623 
1624 void spapr_setup_hpt_and_vrma(SpaprMachineState *spapr)
1625 {
1626     int hpt_shift;
1627 
1628     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1629         || (spapr->cas_reboot
1630             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1631         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1632     } else {
1633         uint64_t current_ram_size;
1634 
1635         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1636         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1637     }
1638     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1639 
1640     if (spapr->vrma_adjust) {
1641         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1642                                           spapr->htab_shift);
1643     }
1644 }
1645 
1646 static int spapr_reset_drcs(Object *child, void *opaque)
1647 {
1648     SpaprDrc *drc =
1649         (SpaprDrc *) object_dynamic_cast(child,
1650                                                  TYPE_SPAPR_DR_CONNECTOR);
1651 
1652     if (drc) {
1653         spapr_drc_reset(drc);
1654     }
1655 
1656     return 0;
1657 }
1658 
1659 static void spapr_machine_reset(void)
1660 {
1661     MachineState *machine = MACHINE(qdev_get_machine());
1662     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
1663     PowerPCCPU *first_ppc_cpu;
1664     uint32_t rtas_limit;
1665     hwaddr rtas_addr, fdt_addr;
1666     void *fdt;
1667     int rc;
1668 
1669     spapr_caps_apply(spapr);
1670 
1671     first_ppc_cpu = POWERPC_CPU(first_cpu);
1672     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1673         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1674                               spapr->max_compat_pvr)) {
1675         /*
1676          * If using KVM with radix mode available, VCPUs can be started
1677          * without a HPT because KVM will start them in radix mode.
1678          * Set the GR bit in PATE so that we know there is no HPT.
1679          */
1680         spapr->patb_entry = PATE1_GR;
1681         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1682     } else {
1683         spapr_setup_hpt_and_vrma(spapr);
1684     }
1685 
1686     /*
1687      * If this reset wasn't generated by CAS, we should reset our
1688      * negotiated options and start from scratch
1689      */
1690     if (!spapr->cas_reboot) {
1691         spapr_ovec_cleanup(spapr->ov5_cas);
1692         spapr->ov5_cas = spapr_ovec_new();
1693 
1694         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1695     }
1696 
1697     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1698         spapr_irq_msi_reset(spapr);
1699     }
1700 
1701     qemu_devices_reset();
1702 
1703     /*
1704      * This is fixing some of the default configuration of the XIVE
1705      * devices. To be called after the reset of the machine devices.
1706      */
1707     spapr_irq_reset(spapr, &error_fatal);
1708 
1709     /*
1710      * There is no CAS under qtest. Simulate one to please the code that
1711      * depends on spapr->ov5_cas. This is especially needed to test device
1712      * unplug, so we do that before resetting the DRCs.
1713      */
1714     if (qtest_enabled()) {
1715         spapr_ovec_cleanup(spapr->ov5_cas);
1716         spapr->ov5_cas = spapr_ovec_clone(spapr->ov5);
1717     }
1718 
1719     /* DRC reset may cause a device to be unplugged. This will cause troubles
1720      * if this device is used by another device (eg, a running vhost backend
1721      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1722      * situations, we reset DRCs after all devices have been reset.
1723      */
1724     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1725 
1726     spapr_clear_pending_events(spapr);
1727 
1728     /*
1729      * We place the device tree and RTAS just below either the top of the RMA,
1730      * or just below 2GB, whichever is lower, so that it can be
1731      * processed with 32-bit real mode code if necessary
1732      */
1733     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1734     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1735     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1736 
1737     fdt = spapr_build_fdt(spapr);
1738 
1739     spapr_load_rtas(spapr, fdt, rtas_addr);
1740 
1741     rc = fdt_pack(fdt);
1742 
1743     /* Should only fail if we've built a corrupted tree */
1744     assert(rc == 0);
1745 
1746     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1747         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1748                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1749         exit(1);
1750     }
1751 
1752     /* Load the fdt */
1753     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1754     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1755     g_free(spapr->fdt_blob);
1756     spapr->fdt_size = fdt_totalsize(fdt);
1757     spapr->fdt_initial_size = spapr->fdt_size;
1758     spapr->fdt_blob = fdt;
1759 
1760     /* Set up the entry state */
1761     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1762     first_ppc_cpu->env.gpr[5] = 0;
1763 
1764     spapr->cas_reboot = false;
1765 }
1766 
1767 static void spapr_create_nvram(SpaprMachineState *spapr)
1768 {
1769     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1770     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1771 
1772     if (dinfo) {
1773         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1774                             &error_fatal);
1775     }
1776 
1777     qdev_init_nofail(dev);
1778 
1779     spapr->nvram = (struct SpaprNvram *)dev;
1780 }
1781 
1782 static void spapr_rtc_create(SpaprMachineState *spapr)
1783 {
1784     object_initialize_child(OBJECT(spapr), "rtc",
1785                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1786                             &error_fatal, NULL);
1787     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1788                               &error_fatal);
1789     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1790                               "date", &error_fatal);
1791 }
1792 
1793 /* Returns whether we want to use VGA or not */
1794 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1795 {
1796     switch (vga_interface_type) {
1797     case VGA_NONE:
1798         return false;
1799     case VGA_DEVICE:
1800         return true;
1801     case VGA_STD:
1802     case VGA_VIRTIO:
1803     case VGA_CIRRUS:
1804         return pci_vga_init(pci_bus) != NULL;
1805     default:
1806         error_setg(errp,
1807                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1808         return false;
1809     }
1810 }
1811 
1812 static int spapr_pre_load(void *opaque)
1813 {
1814     int rc;
1815 
1816     rc = spapr_caps_pre_load(opaque);
1817     if (rc) {
1818         return rc;
1819     }
1820 
1821     return 0;
1822 }
1823 
1824 static int spapr_post_load(void *opaque, int version_id)
1825 {
1826     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1827     int err = 0;
1828 
1829     err = spapr_caps_post_migration(spapr);
1830     if (err) {
1831         return err;
1832     }
1833 
1834     /*
1835      * In earlier versions, there was no separate qdev for the PAPR
1836      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1837      * So when migrating from those versions, poke the incoming offset
1838      * value into the RTC device
1839      */
1840     if (version_id < 3) {
1841         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1842         if (err) {
1843             return err;
1844         }
1845     }
1846 
1847     if (kvm_enabled() && spapr->patb_entry) {
1848         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1849         bool radix = !!(spapr->patb_entry & PATE1_GR);
1850         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1851 
1852         /*
1853          * Update LPCR:HR and UPRT as they may not be set properly in
1854          * the stream
1855          */
1856         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1857                             LPCR_HR | LPCR_UPRT);
1858 
1859         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1860         if (err) {
1861             error_report("Process table config unsupported by the host");
1862             return -EINVAL;
1863         }
1864     }
1865 
1866     err = spapr_irq_post_load(spapr, version_id);
1867     if (err) {
1868         return err;
1869     }
1870 
1871     return err;
1872 }
1873 
1874 static int spapr_pre_save(void *opaque)
1875 {
1876     int rc;
1877 
1878     rc = spapr_caps_pre_save(opaque);
1879     if (rc) {
1880         return rc;
1881     }
1882 
1883     return 0;
1884 }
1885 
1886 static bool version_before_3(void *opaque, int version_id)
1887 {
1888     return version_id < 3;
1889 }
1890 
1891 static bool spapr_pending_events_needed(void *opaque)
1892 {
1893     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
1894     return !QTAILQ_EMPTY(&spapr->pending_events);
1895 }
1896 
1897 static const VMStateDescription vmstate_spapr_event_entry = {
1898     .name = "spapr_event_log_entry",
1899     .version_id = 1,
1900     .minimum_version_id = 1,
1901     .fields = (VMStateField[]) {
1902         VMSTATE_UINT32(summary, SpaprEventLogEntry),
1903         VMSTATE_UINT32(extended_length, SpaprEventLogEntry),
1904         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, SpaprEventLogEntry, 0,
1905                                      NULL, extended_length),
1906         VMSTATE_END_OF_LIST()
1907     },
1908 };
1909 
1910 static const VMStateDescription vmstate_spapr_pending_events = {
1911     .name = "spapr_pending_events",
1912     .version_id = 1,
1913     .minimum_version_id = 1,
1914     .needed = spapr_pending_events_needed,
1915     .fields = (VMStateField[]) {
1916         VMSTATE_QTAILQ_V(pending_events, SpaprMachineState, 1,
1917                          vmstate_spapr_event_entry, SpaprEventLogEntry, next),
1918         VMSTATE_END_OF_LIST()
1919     },
1920 };
1921 
1922 static bool spapr_ov5_cas_needed(void *opaque)
1923 {
1924     SpaprMachineState *spapr = opaque;
1925     SpaprOptionVector *ov5_mask = spapr_ovec_new();
1926     SpaprOptionVector *ov5_legacy = spapr_ovec_new();
1927     SpaprOptionVector *ov5_removed = spapr_ovec_new();
1928     bool cas_needed;
1929 
1930     /* Prior to the introduction of SpaprOptionVector, we had two option
1931      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1932      * Both of these options encode machine topology into the device-tree
1933      * in such a way that the now-booted OS should still be able to interact
1934      * appropriately with QEMU regardless of what options were actually
1935      * negotiatied on the source side.
1936      *
1937      * As such, we can avoid migrating the CAS-negotiated options if these
1938      * are the only options available on the current machine/platform.
1939      * Since these are the only options available for pseries-2.7 and
1940      * earlier, this allows us to maintain old->new/new->old migration
1941      * compatibility.
1942      *
1943      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1944      * via default pseries-2.8 machines and explicit command-line parameters.
1945      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1946      * of the actual CAS-negotiated values to continue working properly. For
1947      * example, availability of memory unplug depends on knowing whether
1948      * OV5_HP_EVT was negotiated via CAS.
1949      *
1950      * Thus, for any cases where the set of available CAS-negotiatable
1951      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1952      * include the CAS-negotiated options in the migration stream, unless
1953      * if they affect boot time behaviour only.
1954      */
1955     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1956     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1957     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1958 
1959     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1960      * the mask itself since in the future it's possible "legacy" bits may be
1961      * removed via machine options, which could generate a false positive
1962      * that breaks migration.
1963      */
1964     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1965     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1966 
1967     spapr_ovec_cleanup(ov5_mask);
1968     spapr_ovec_cleanup(ov5_legacy);
1969     spapr_ovec_cleanup(ov5_removed);
1970 
1971     return cas_needed;
1972 }
1973 
1974 static const VMStateDescription vmstate_spapr_ov5_cas = {
1975     .name = "spapr_option_vector_ov5_cas",
1976     .version_id = 1,
1977     .minimum_version_id = 1,
1978     .needed = spapr_ov5_cas_needed,
1979     .fields = (VMStateField[]) {
1980         VMSTATE_STRUCT_POINTER_V(ov5_cas, SpaprMachineState, 1,
1981                                  vmstate_spapr_ovec, SpaprOptionVector),
1982         VMSTATE_END_OF_LIST()
1983     },
1984 };
1985 
1986 static bool spapr_patb_entry_needed(void *opaque)
1987 {
1988     SpaprMachineState *spapr = opaque;
1989 
1990     return !!spapr->patb_entry;
1991 }
1992 
1993 static const VMStateDescription vmstate_spapr_patb_entry = {
1994     .name = "spapr_patb_entry",
1995     .version_id = 1,
1996     .minimum_version_id = 1,
1997     .needed = spapr_patb_entry_needed,
1998     .fields = (VMStateField[]) {
1999         VMSTATE_UINT64(patb_entry, SpaprMachineState),
2000         VMSTATE_END_OF_LIST()
2001     },
2002 };
2003 
2004 static bool spapr_irq_map_needed(void *opaque)
2005 {
2006     SpaprMachineState *spapr = opaque;
2007 
2008     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2009 }
2010 
2011 static const VMStateDescription vmstate_spapr_irq_map = {
2012     .name = "spapr_irq_map",
2013     .version_id = 1,
2014     .minimum_version_id = 1,
2015     .needed = spapr_irq_map_needed,
2016     .fields = (VMStateField[]) {
2017         VMSTATE_BITMAP(irq_map, SpaprMachineState, 0, irq_map_nr),
2018         VMSTATE_END_OF_LIST()
2019     },
2020 };
2021 
2022 static bool spapr_dtb_needed(void *opaque)
2023 {
2024     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2025 
2026     return smc->update_dt_enabled;
2027 }
2028 
2029 static int spapr_dtb_pre_load(void *opaque)
2030 {
2031     SpaprMachineState *spapr = (SpaprMachineState *)opaque;
2032 
2033     g_free(spapr->fdt_blob);
2034     spapr->fdt_blob = NULL;
2035     spapr->fdt_size = 0;
2036 
2037     return 0;
2038 }
2039 
2040 static const VMStateDescription vmstate_spapr_dtb = {
2041     .name = "spapr_dtb",
2042     .version_id = 1,
2043     .minimum_version_id = 1,
2044     .needed = spapr_dtb_needed,
2045     .pre_load = spapr_dtb_pre_load,
2046     .fields = (VMStateField[]) {
2047         VMSTATE_UINT32(fdt_initial_size, SpaprMachineState),
2048         VMSTATE_UINT32(fdt_size, SpaprMachineState),
2049         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, SpaprMachineState, 0, NULL,
2050                                      fdt_size),
2051         VMSTATE_END_OF_LIST()
2052     },
2053 };
2054 
2055 static const VMStateDescription vmstate_spapr = {
2056     .name = "spapr",
2057     .version_id = 3,
2058     .minimum_version_id = 1,
2059     .pre_load = spapr_pre_load,
2060     .post_load = spapr_post_load,
2061     .pre_save = spapr_pre_save,
2062     .fields = (VMStateField[]) {
2063         /* used to be @next_irq */
2064         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2065 
2066         /* RTC offset */
2067         VMSTATE_UINT64_TEST(rtc_offset, SpaprMachineState, version_before_3),
2068 
2069         VMSTATE_PPC_TIMEBASE_V(tb, SpaprMachineState, 2),
2070         VMSTATE_END_OF_LIST()
2071     },
2072     .subsections = (const VMStateDescription*[]) {
2073         &vmstate_spapr_ov5_cas,
2074         &vmstate_spapr_patb_entry,
2075         &vmstate_spapr_pending_events,
2076         &vmstate_spapr_cap_htm,
2077         &vmstate_spapr_cap_vsx,
2078         &vmstate_spapr_cap_dfp,
2079         &vmstate_spapr_cap_cfpc,
2080         &vmstate_spapr_cap_sbbc,
2081         &vmstate_spapr_cap_ibs,
2082         &vmstate_spapr_irq_map,
2083         &vmstate_spapr_cap_nested_kvm_hv,
2084         &vmstate_spapr_dtb,
2085         &vmstate_spapr_cap_large_decr,
2086         &vmstate_spapr_cap_ccf_assist,
2087         NULL
2088     }
2089 };
2090 
2091 static int htab_save_setup(QEMUFile *f, void *opaque)
2092 {
2093     SpaprMachineState *spapr = opaque;
2094 
2095     /* "Iteration" header */
2096     if (!spapr->htab_shift) {
2097         qemu_put_be32(f, -1);
2098     } else {
2099         qemu_put_be32(f, spapr->htab_shift);
2100     }
2101 
2102     if (spapr->htab) {
2103         spapr->htab_save_index = 0;
2104         spapr->htab_first_pass = true;
2105     } else {
2106         if (spapr->htab_shift) {
2107             assert(kvm_enabled());
2108         }
2109     }
2110 
2111 
2112     return 0;
2113 }
2114 
2115 static void htab_save_chunk(QEMUFile *f, SpaprMachineState *spapr,
2116                             int chunkstart, int n_valid, int n_invalid)
2117 {
2118     qemu_put_be32(f, chunkstart);
2119     qemu_put_be16(f, n_valid);
2120     qemu_put_be16(f, n_invalid);
2121     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2122                     HASH_PTE_SIZE_64 * n_valid);
2123 }
2124 
2125 static void htab_save_end_marker(QEMUFile *f)
2126 {
2127     qemu_put_be32(f, 0);
2128     qemu_put_be16(f, 0);
2129     qemu_put_be16(f, 0);
2130 }
2131 
2132 static void htab_save_first_pass(QEMUFile *f, SpaprMachineState *spapr,
2133                                  int64_t max_ns)
2134 {
2135     bool has_timeout = max_ns != -1;
2136     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2137     int index = spapr->htab_save_index;
2138     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2139 
2140     assert(spapr->htab_first_pass);
2141 
2142     do {
2143         int chunkstart;
2144 
2145         /* Consume invalid HPTEs */
2146         while ((index < htabslots)
2147                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2148             CLEAN_HPTE(HPTE(spapr->htab, index));
2149             index++;
2150         }
2151 
2152         /* Consume valid HPTEs */
2153         chunkstart = index;
2154         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2155                && HPTE_VALID(HPTE(spapr->htab, index))) {
2156             CLEAN_HPTE(HPTE(spapr->htab, index));
2157             index++;
2158         }
2159 
2160         if (index > chunkstart) {
2161             int n_valid = index - chunkstart;
2162 
2163             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2164 
2165             if (has_timeout &&
2166                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2167                 break;
2168             }
2169         }
2170     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2171 
2172     if (index >= htabslots) {
2173         assert(index == htabslots);
2174         index = 0;
2175         spapr->htab_first_pass = false;
2176     }
2177     spapr->htab_save_index = index;
2178 }
2179 
2180 static int htab_save_later_pass(QEMUFile *f, SpaprMachineState *spapr,
2181                                 int64_t max_ns)
2182 {
2183     bool final = max_ns < 0;
2184     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2185     int examined = 0, sent = 0;
2186     int index = spapr->htab_save_index;
2187     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2188 
2189     assert(!spapr->htab_first_pass);
2190 
2191     do {
2192         int chunkstart, invalidstart;
2193 
2194         /* Consume non-dirty HPTEs */
2195         while ((index < htabslots)
2196                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2197             index++;
2198             examined++;
2199         }
2200 
2201         chunkstart = index;
2202         /* Consume valid dirty HPTEs */
2203         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2204                && HPTE_DIRTY(HPTE(spapr->htab, index))
2205                && HPTE_VALID(HPTE(spapr->htab, index))) {
2206             CLEAN_HPTE(HPTE(spapr->htab, index));
2207             index++;
2208             examined++;
2209         }
2210 
2211         invalidstart = index;
2212         /* Consume invalid dirty HPTEs */
2213         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2214                && HPTE_DIRTY(HPTE(spapr->htab, index))
2215                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2216             CLEAN_HPTE(HPTE(spapr->htab, index));
2217             index++;
2218             examined++;
2219         }
2220 
2221         if (index > chunkstart) {
2222             int n_valid = invalidstart - chunkstart;
2223             int n_invalid = index - invalidstart;
2224 
2225             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2226             sent += index - chunkstart;
2227 
2228             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2229                 break;
2230             }
2231         }
2232 
2233         if (examined >= htabslots) {
2234             break;
2235         }
2236 
2237         if (index >= htabslots) {
2238             assert(index == htabslots);
2239             index = 0;
2240         }
2241     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2242 
2243     if (index >= htabslots) {
2244         assert(index == htabslots);
2245         index = 0;
2246     }
2247 
2248     spapr->htab_save_index = index;
2249 
2250     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2251 }
2252 
2253 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2254 #define MAX_KVM_BUF_SIZE    2048
2255 
2256 static int htab_save_iterate(QEMUFile *f, void *opaque)
2257 {
2258     SpaprMachineState *spapr = opaque;
2259     int fd;
2260     int rc = 0;
2261 
2262     /* Iteration header */
2263     if (!spapr->htab_shift) {
2264         qemu_put_be32(f, -1);
2265         return 1;
2266     } else {
2267         qemu_put_be32(f, 0);
2268     }
2269 
2270     if (!spapr->htab) {
2271         assert(kvm_enabled());
2272 
2273         fd = get_htab_fd(spapr);
2274         if (fd < 0) {
2275             return fd;
2276         }
2277 
2278         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2279         if (rc < 0) {
2280             return rc;
2281         }
2282     } else  if (spapr->htab_first_pass) {
2283         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2284     } else {
2285         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2286     }
2287 
2288     htab_save_end_marker(f);
2289 
2290     return rc;
2291 }
2292 
2293 static int htab_save_complete(QEMUFile *f, void *opaque)
2294 {
2295     SpaprMachineState *spapr = opaque;
2296     int fd;
2297 
2298     /* Iteration header */
2299     if (!spapr->htab_shift) {
2300         qemu_put_be32(f, -1);
2301         return 0;
2302     } else {
2303         qemu_put_be32(f, 0);
2304     }
2305 
2306     if (!spapr->htab) {
2307         int rc;
2308 
2309         assert(kvm_enabled());
2310 
2311         fd = get_htab_fd(spapr);
2312         if (fd < 0) {
2313             return fd;
2314         }
2315 
2316         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2317         if (rc < 0) {
2318             return rc;
2319         }
2320     } else {
2321         if (spapr->htab_first_pass) {
2322             htab_save_first_pass(f, spapr, -1);
2323         }
2324         htab_save_later_pass(f, spapr, -1);
2325     }
2326 
2327     /* End marker */
2328     htab_save_end_marker(f);
2329 
2330     return 0;
2331 }
2332 
2333 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2334 {
2335     SpaprMachineState *spapr = opaque;
2336     uint32_t section_hdr;
2337     int fd = -1;
2338     Error *local_err = NULL;
2339 
2340     if (version_id < 1 || version_id > 1) {
2341         error_report("htab_load() bad version");
2342         return -EINVAL;
2343     }
2344 
2345     section_hdr = qemu_get_be32(f);
2346 
2347     if (section_hdr == -1) {
2348         spapr_free_hpt(spapr);
2349         return 0;
2350     }
2351 
2352     if (section_hdr) {
2353         /* First section gives the htab size */
2354         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2355         if (local_err) {
2356             error_report_err(local_err);
2357             return -EINVAL;
2358         }
2359         return 0;
2360     }
2361 
2362     if (!spapr->htab) {
2363         assert(kvm_enabled());
2364 
2365         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2366         if (fd < 0) {
2367             error_report_err(local_err);
2368             return fd;
2369         }
2370     }
2371 
2372     while (true) {
2373         uint32_t index;
2374         uint16_t n_valid, n_invalid;
2375 
2376         index = qemu_get_be32(f);
2377         n_valid = qemu_get_be16(f);
2378         n_invalid = qemu_get_be16(f);
2379 
2380         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2381             /* End of Stream */
2382             break;
2383         }
2384 
2385         if ((index + n_valid + n_invalid) >
2386             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2387             /* Bad index in stream */
2388             error_report(
2389                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2390                 index, n_valid, n_invalid, spapr->htab_shift);
2391             return -EINVAL;
2392         }
2393 
2394         if (spapr->htab) {
2395             if (n_valid) {
2396                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2397                                 HASH_PTE_SIZE_64 * n_valid);
2398             }
2399             if (n_invalid) {
2400                 memset(HPTE(spapr->htab, index + n_valid), 0,
2401                        HASH_PTE_SIZE_64 * n_invalid);
2402             }
2403         } else {
2404             int rc;
2405 
2406             assert(fd >= 0);
2407 
2408             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2409             if (rc < 0) {
2410                 return rc;
2411             }
2412         }
2413     }
2414 
2415     if (!spapr->htab) {
2416         assert(fd >= 0);
2417         close(fd);
2418     }
2419 
2420     return 0;
2421 }
2422 
2423 static void htab_save_cleanup(void *opaque)
2424 {
2425     SpaprMachineState *spapr = opaque;
2426 
2427     close_htab_fd(spapr);
2428 }
2429 
2430 static SaveVMHandlers savevm_htab_handlers = {
2431     .save_setup = htab_save_setup,
2432     .save_live_iterate = htab_save_iterate,
2433     .save_live_complete_precopy = htab_save_complete,
2434     .save_cleanup = htab_save_cleanup,
2435     .load_state = htab_load,
2436 };
2437 
2438 static void spapr_boot_set(void *opaque, const char *boot_device,
2439                            Error **errp)
2440 {
2441     MachineState *machine = MACHINE(opaque);
2442     machine->boot_order = g_strdup(boot_device);
2443 }
2444 
2445 static void spapr_create_lmb_dr_connectors(SpaprMachineState *spapr)
2446 {
2447     MachineState *machine = MACHINE(spapr);
2448     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2449     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2450     int i;
2451 
2452     for (i = 0; i < nr_lmbs; i++) {
2453         uint64_t addr;
2454 
2455         addr = i * lmb_size + machine->device_memory->base;
2456         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2457                                addr / lmb_size);
2458     }
2459 }
2460 
2461 /*
2462  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2463  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2464  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2465  */
2466 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2467 {
2468     int i;
2469 
2470     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2471         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2472                    " is not aligned to %" PRIu64 " MiB",
2473                    machine->ram_size,
2474                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2475         return;
2476     }
2477 
2478     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2479         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2480                    " is not aligned to %" PRIu64 " MiB",
2481                    machine->ram_size,
2482                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2483         return;
2484     }
2485 
2486     for (i = 0; i < nb_numa_nodes; i++) {
2487         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2488             error_setg(errp,
2489                        "Node %d memory size 0x%" PRIx64
2490                        " is not aligned to %" PRIu64 " MiB",
2491                        i, numa_info[i].node_mem,
2492                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2493             return;
2494         }
2495     }
2496 }
2497 
2498 /* find cpu slot in machine->possible_cpus by core_id */
2499 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2500 {
2501     int index = id / smp_threads;
2502 
2503     if (index >= ms->possible_cpus->len) {
2504         return NULL;
2505     }
2506     if (idx) {
2507         *idx = index;
2508     }
2509     return &ms->possible_cpus->cpus[index];
2510 }
2511 
2512 static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
2513 {
2514     Error *local_err = NULL;
2515     bool vsmt_user = !!spapr->vsmt;
2516     int kvm_smt = kvmppc_smt_threads();
2517     int ret;
2518 
2519     if (!kvm_enabled() && (smp_threads > 1)) {
2520         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2521                      "on a pseries machine");
2522         goto out;
2523     }
2524     if (!is_power_of_2(smp_threads)) {
2525         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2526                      "machine because it must be a power of 2", smp_threads);
2527         goto out;
2528     }
2529 
2530     /* Detemine the VSMT mode to use: */
2531     if (vsmt_user) {
2532         if (spapr->vsmt < smp_threads) {
2533             error_setg(&local_err, "Cannot support VSMT mode %d"
2534                          " because it must be >= threads/core (%d)",
2535                          spapr->vsmt, smp_threads);
2536             goto out;
2537         }
2538         /* In this case, spapr->vsmt has been set by the command line */
2539     } else {
2540         /*
2541          * Default VSMT value is tricky, because we need it to be as
2542          * consistent as possible (for migration), but this requires
2543          * changing it for at least some existing cases.  We pick 8 as
2544          * the value that we'd get with KVM on POWER8, the
2545          * overwhelmingly common case in production systems.
2546          */
2547         spapr->vsmt = MAX(8, smp_threads);
2548     }
2549 
2550     /* KVM: If necessary, set the SMT mode: */
2551     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2552         ret = kvmppc_set_smt_threads(spapr->vsmt);
2553         if (ret) {
2554             /* Looks like KVM isn't able to change VSMT mode */
2555             error_setg(&local_err,
2556                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2557                        spapr->vsmt, ret);
2558             /* We can live with that if the default one is big enough
2559              * for the number of threads, and a submultiple of the one
2560              * we want.  In this case we'll waste some vcpu ids, but
2561              * behaviour will be correct */
2562             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2563                 warn_report_err(local_err);
2564                 local_err = NULL;
2565                 goto out;
2566             } else {
2567                 if (!vsmt_user) {
2568                     error_append_hint(&local_err,
2569                                       "On PPC, a VM with %d threads/core"
2570                                       " on a host with %d threads/core"
2571                                       " requires the use of VSMT mode %d.\n",
2572                                       smp_threads, kvm_smt, spapr->vsmt);
2573                 }
2574                 kvmppc_hint_smt_possible(&local_err);
2575                 goto out;
2576             }
2577         }
2578     }
2579     /* else TCG: nothing to do currently */
2580 out:
2581     error_propagate(errp, local_err);
2582 }
2583 
2584 static void spapr_init_cpus(SpaprMachineState *spapr)
2585 {
2586     MachineState *machine = MACHINE(spapr);
2587     MachineClass *mc = MACHINE_GET_CLASS(machine);
2588     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2589     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2590     const CPUArchIdList *possible_cpus;
2591     int boot_cores_nr = smp_cpus / smp_threads;
2592     int i;
2593 
2594     possible_cpus = mc->possible_cpu_arch_ids(machine);
2595     if (mc->has_hotpluggable_cpus) {
2596         if (smp_cpus % smp_threads) {
2597             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2598                          smp_cpus, smp_threads);
2599             exit(1);
2600         }
2601         if (max_cpus % smp_threads) {
2602             error_report("max_cpus (%u) must be multiple of threads (%u)",
2603                          max_cpus, smp_threads);
2604             exit(1);
2605         }
2606     } else {
2607         if (max_cpus != smp_cpus) {
2608             error_report("This machine version does not support CPU hotplug");
2609             exit(1);
2610         }
2611         boot_cores_nr = possible_cpus->len;
2612     }
2613 
2614     if (smc->pre_2_10_has_unused_icps) {
2615         int i;
2616 
2617         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2618             /* Dummy entries get deregistered when real ICPState objects
2619              * are registered during CPU core hotplug.
2620              */
2621             pre_2_10_vmstate_register_dummy_icp(i);
2622         }
2623     }
2624 
2625     for (i = 0; i < possible_cpus->len; i++) {
2626         int core_id = i * smp_threads;
2627 
2628         if (mc->has_hotpluggable_cpus) {
2629             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2630                                    spapr_vcpu_id(spapr, core_id));
2631         }
2632 
2633         if (i < boot_cores_nr) {
2634             Object *core  = object_new(type);
2635             int nr_threads = smp_threads;
2636 
2637             /* Handle the partially filled core for older machine types */
2638             if ((i + 1) * smp_threads >= smp_cpus) {
2639                 nr_threads = smp_cpus - i * smp_threads;
2640             }
2641 
2642             object_property_set_int(core, nr_threads, "nr-threads",
2643                                     &error_fatal);
2644             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2645                                     &error_fatal);
2646             object_property_set_bool(core, true, "realized", &error_fatal);
2647 
2648             object_unref(core);
2649         }
2650     }
2651 }
2652 
2653 static PCIHostState *spapr_create_default_phb(void)
2654 {
2655     DeviceState *dev;
2656 
2657     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2658     qdev_prop_set_uint32(dev, "index", 0);
2659     qdev_init_nofail(dev);
2660 
2661     return PCI_HOST_BRIDGE(dev);
2662 }
2663 
2664 /* pSeries LPAR / sPAPR hardware init */
2665 static void spapr_machine_init(MachineState *machine)
2666 {
2667     SpaprMachineState *spapr = SPAPR_MACHINE(machine);
2668     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2669     const char *kernel_filename = machine->kernel_filename;
2670     const char *initrd_filename = machine->initrd_filename;
2671     PCIHostState *phb;
2672     int i;
2673     MemoryRegion *sysmem = get_system_memory();
2674     MemoryRegion *ram = g_new(MemoryRegion, 1);
2675     hwaddr node0_size = spapr_node0_size(machine);
2676     long load_limit, fw_size;
2677     char *filename;
2678     Error *resize_hpt_err = NULL;
2679 
2680     msi_nonbroken = true;
2681 
2682     QLIST_INIT(&spapr->phbs);
2683     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2684 
2685     /* Determine capabilities to run with */
2686     spapr_caps_init(spapr);
2687 
2688     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2689     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2690         /*
2691          * If the user explicitly requested a mode we should either
2692          * supply it, or fail completely (which we do below).  But if
2693          * it's not set explicitly, we reset our mode to something
2694          * that works
2695          */
2696         if (resize_hpt_err) {
2697             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2698             error_free(resize_hpt_err);
2699             resize_hpt_err = NULL;
2700         } else {
2701             spapr->resize_hpt = smc->resize_hpt_default;
2702         }
2703     }
2704 
2705     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2706 
2707     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2708         /*
2709          * User requested HPT resize, but this host can't supply it.  Bail out
2710          */
2711         error_report_err(resize_hpt_err);
2712         exit(1);
2713     }
2714 
2715     spapr->rma_size = node0_size;
2716 
2717     /* With KVM, we don't actually know whether KVM supports an
2718      * unbounded RMA (PR KVM) or is limited by the hash table size
2719      * (HV KVM using VRMA), so we always assume the latter
2720      *
2721      * In that case, we also limit the initial allocations for RTAS
2722      * etc... to 256M since we have no way to know what the VRMA size
2723      * is going to be as it depends on the size of the hash table
2724      * which isn't determined yet.
2725      */
2726     if (kvm_enabled()) {
2727         spapr->vrma_adjust = 1;
2728         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2729     }
2730 
2731     /* Actually we don't support unbounded RMA anymore since we added
2732      * proper emulation of HV mode. The max we can get is 16G which
2733      * also happens to be what we configure for PAPR mode so make sure
2734      * we don't do anything bigger than that
2735      */
2736     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2737 
2738     if (spapr->rma_size > node0_size) {
2739         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2740                      spapr->rma_size);
2741         exit(1);
2742     }
2743 
2744     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2745     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2746 
2747     /*
2748      * VSMT must be set in order to be able to compute VCPU ids, ie to
2749      * call spapr_max_server_number() or spapr_vcpu_id().
2750      */
2751     spapr_set_vsmt_mode(spapr, &error_fatal);
2752 
2753     /* Set up Interrupt Controller before we create the VCPUs */
2754     spapr_irq_init(spapr, &error_fatal);
2755 
2756     /* Set up containers for ibm,client-architecture-support negotiated options
2757      */
2758     spapr->ov5 = spapr_ovec_new();
2759     spapr->ov5_cas = spapr_ovec_new();
2760 
2761     if (smc->dr_lmb_enabled) {
2762         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2763         spapr_validate_node_memory(machine, &error_fatal);
2764     }
2765 
2766     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2767 
2768     /* advertise support for dedicated HP event source to guests */
2769     if (spapr->use_hotplug_event_source) {
2770         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2771     }
2772 
2773     /* advertise support for HPT resizing */
2774     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2775         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2776     }
2777 
2778     /* advertise support for ibm,dyamic-memory-v2 */
2779     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2780 
2781     /* advertise XIVE on POWER9 machines */
2782     if (spapr->irq->ov5 & (SPAPR_OV5_XIVE_EXPLOIT | SPAPR_OV5_XIVE_BOTH)) {
2783         spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2784     }
2785 
2786     /* init CPUs */
2787     spapr_init_cpus(spapr);
2788 
2789     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2790         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2791                               spapr->max_compat_pvr)) {
2792         /* KVM and TCG always allow GTSE with radix... */
2793         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2794     }
2795     /* ... but not with hash (currently). */
2796 
2797     if (kvm_enabled()) {
2798         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2799         kvmppc_enable_logical_ci_hcalls();
2800         kvmppc_enable_set_mode_hcall();
2801 
2802         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2803         kvmppc_enable_clear_ref_mod_hcalls();
2804 
2805         /* Enable H_PAGE_INIT */
2806         kvmppc_enable_h_page_init();
2807     }
2808 
2809     /* allocate RAM */
2810     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2811                                          machine->ram_size);
2812     memory_region_add_subregion(sysmem, 0, ram);
2813 
2814     /* always allocate the device memory information */
2815     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2816 
2817     /* initialize hotplug memory address space */
2818     if (machine->ram_size < machine->maxram_size) {
2819         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2820         /*
2821          * Limit the number of hotpluggable memory slots to half the number
2822          * slots that KVM supports, leaving the other half for PCI and other
2823          * devices. However ensure that number of slots doesn't drop below 32.
2824          */
2825         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2826                            SPAPR_MAX_RAM_SLOTS;
2827 
2828         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2829             max_memslots = SPAPR_MAX_RAM_SLOTS;
2830         }
2831         if (machine->ram_slots > max_memslots) {
2832             error_report("Specified number of memory slots %"
2833                          PRIu64" exceeds max supported %d",
2834                          machine->ram_slots, max_memslots);
2835             exit(1);
2836         }
2837 
2838         machine->device_memory->base = ROUND_UP(machine->ram_size,
2839                                                 SPAPR_DEVICE_MEM_ALIGN);
2840         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2841                            "device-memory", device_mem_size);
2842         memory_region_add_subregion(sysmem, machine->device_memory->base,
2843                                     &machine->device_memory->mr);
2844     }
2845 
2846     if (smc->dr_lmb_enabled) {
2847         spapr_create_lmb_dr_connectors(spapr);
2848     }
2849 
2850     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2851     if (!filename) {
2852         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2853         exit(1);
2854     }
2855     spapr->rtas_size = get_image_size(filename);
2856     if (spapr->rtas_size < 0) {
2857         error_report("Could not get size of LPAR rtas '%s'", filename);
2858         exit(1);
2859     }
2860     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2861     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2862         error_report("Could not load LPAR rtas '%s'", filename);
2863         exit(1);
2864     }
2865     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2866         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2867                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2868         exit(1);
2869     }
2870     g_free(filename);
2871 
2872     /* Set up RTAS event infrastructure */
2873     spapr_events_init(spapr);
2874 
2875     /* Set up the RTC RTAS interfaces */
2876     spapr_rtc_create(spapr);
2877 
2878     /* Set up VIO bus */
2879     spapr->vio_bus = spapr_vio_bus_init();
2880 
2881     for (i = 0; i < serial_max_hds(); i++) {
2882         if (serial_hd(i)) {
2883             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2884         }
2885     }
2886 
2887     /* We always have at least the nvram device on VIO */
2888     spapr_create_nvram(spapr);
2889 
2890     /*
2891      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2892      * connectors (described in root DT node's "ibm,drc-types" property)
2893      * are pre-initialized here. additional child connectors (such as
2894      * connectors for a PHBs PCI slots) are added as needed during their
2895      * parent's realization.
2896      */
2897     if (smc->dr_phb_enabled) {
2898         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2899             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2900         }
2901     }
2902 
2903     /* Set up PCI */
2904     spapr_pci_rtas_init();
2905 
2906     phb = spapr_create_default_phb();
2907 
2908     for (i = 0; i < nb_nics; i++) {
2909         NICInfo *nd = &nd_table[i];
2910 
2911         if (!nd->model) {
2912             nd->model = g_strdup("spapr-vlan");
2913         }
2914 
2915         if (g_str_equal(nd->model, "spapr-vlan") ||
2916             g_str_equal(nd->model, "ibmveth")) {
2917             spapr_vlan_create(spapr->vio_bus, nd);
2918         } else {
2919             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2920         }
2921     }
2922 
2923     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2924         spapr_vscsi_create(spapr->vio_bus);
2925     }
2926 
2927     /* Graphics */
2928     if (spapr_vga_init(phb->bus, &error_fatal)) {
2929         spapr->has_graphics = true;
2930         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2931     }
2932 
2933     if (machine->usb) {
2934         if (smc->use_ohci_by_default) {
2935             pci_create_simple(phb->bus, -1, "pci-ohci");
2936         } else {
2937             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2938         }
2939 
2940         if (spapr->has_graphics) {
2941             USBBus *usb_bus = usb_bus_find(-1);
2942 
2943             usb_create_simple(usb_bus, "usb-kbd");
2944             usb_create_simple(usb_bus, "usb-mouse");
2945         }
2946     }
2947 
2948     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2949         error_report(
2950             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2951             MIN_RMA_SLOF);
2952         exit(1);
2953     }
2954 
2955     if (kernel_filename) {
2956         uint64_t lowaddr = 0;
2957 
2958         spapr->kernel_size = load_elf(kernel_filename, NULL,
2959                                       translate_kernel_address, NULL,
2960                                       NULL, &lowaddr, NULL, 1,
2961                                       PPC_ELF_MACHINE, 0, 0);
2962         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2963             spapr->kernel_size = load_elf(kernel_filename, NULL,
2964                                           translate_kernel_address, NULL, NULL,
2965                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2966                                           0, 0);
2967             spapr->kernel_le = spapr->kernel_size > 0;
2968         }
2969         if (spapr->kernel_size < 0) {
2970             error_report("error loading %s: %s", kernel_filename,
2971                          load_elf_strerror(spapr->kernel_size));
2972             exit(1);
2973         }
2974 
2975         /* load initrd */
2976         if (initrd_filename) {
2977             /* Try to locate the initrd in the gap between the kernel
2978              * and the firmware. Add a bit of space just in case
2979              */
2980             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2981                                   + 0x1ffff) & ~0xffff;
2982             spapr->initrd_size = load_image_targphys(initrd_filename,
2983                                                      spapr->initrd_base,
2984                                                      load_limit
2985                                                      - spapr->initrd_base);
2986             if (spapr->initrd_size < 0) {
2987                 error_report("could not load initial ram disk '%s'",
2988                              initrd_filename);
2989                 exit(1);
2990             }
2991         }
2992     }
2993 
2994     if (bios_name == NULL) {
2995         bios_name = FW_FILE_NAME;
2996     }
2997     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2998     if (!filename) {
2999         error_report("Could not find LPAR firmware '%s'", bios_name);
3000         exit(1);
3001     }
3002     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
3003     if (fw_size <= 0) {
3004         error_report("Could not load LPAR firmware '%s'", filename);
3005         exit(1);
3006     }
3007     g_free(filename);
3008 
3009     /* FIXME: Should register things through the MachineState's qdev
3010      * interface, this is a legacy from the sPAPREnvironment structure
3011      * which predated MachineState but had a similar function */
3012     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3013     register_savevm_live(NULL, "spapr/htab", -1, 1,
3014                          &savevm_htab_handlers, spapr);
3015 
3016     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3017                              &error_fatal);
3018 
3019     qemu_register_boot_set(spapr_boot_set, spapr);
3020 
3021     if (kvm_enabled()) {
3022         /* to stop and start vmclock */
3023         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3024                                          &spapr->tb);
3025 
3026         kvmppc_spapr_enable_inkernel_multitce();
3027     }
3028 }
3029 
3030 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3031 {
3032     if (!vm_type) {
3033         return 0;
3034     }
3035 
3036     if (!strcmp(vm_type, "HV")) {
3037         return 1;
3038     }
3039 
3040     if (!strcmp(vm_type, "PR")) {
3041         return 2;
3042     }
3043 
3044     error_report("Unknown kvm-type specified '%s'", vm_type);
3045     exit(1);
3046 }
3047 
3048 /*
3049  * Implementation of an interface to adjust firmware path
3050  * for the bootindex property handling.
3051  */
3052 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3053                                    DeviceState *dev)
3054 {
3055 #define CAST(type, obj, name) \
3056     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3057     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3058     SpaprPhbState *phb = CAST(SpaprPhbState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3059     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3060 
3061     if (d) {
3062         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3063         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3064         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3065 
3066         if (spapr) {
3067             /*
3068              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3069              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3070              * 0x8000 | (target << 8) | (bus << 5) | lun
3071              * (see the "Logical unit addressing format" table in SAM5)
3072              */
3073             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3074             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3075                                    (uint64_t)id << 48);
3076         } else if (virtio) {
3077             /*
3078              * We use SRP luns of the form 01000000 | (target << 8) | lun
3079              * in the top 32 bits of the 64-bit LUN
3080              * Note: the quote above is from SLOF and it is wrong,
3081              * the actual binding is:
3082              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3083              */
3084             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3085             if (d->lun >= 256) {
3086                 /* Use the LUN "flat space addressing method" */
3087                 id |= 0x4000;
3088             }
3089             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3090                                    (uint64_t)id << 32);
3091         } else if (usb) {
3092             /*
3093              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3094              * in the top 32 bits of the 64-bit LUN
3095              */
3096             unsigned usb_port = atoi(usb->port->path);
3097             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3098             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3099                                    (uint64_t)id << 32);
3100         }
3101     }
3102 
3103     /*
3104      * SLOF probes the USB devices, and if it recognizes that the device is a
3105      * storage device, it changes its name to "storage" instead of "usb-host",
3106      * and additionally adds a child node for the SCSI LUN, so the correct
3107      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3108      */
3109     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3110         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3111         if (usb_host_dev_is_scsi_storage(usbdev)) {
3112             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3113         }
3114     }
3115 
3116     if (phb) {
3117         /* Replace "pci" with "pci@800000020000000" */
3118         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3119     }
3120 
3121     if (vsc) {
3122         /* Same logic as virtio above */
3123         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3124         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3125     }
3126 
3127     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3128         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3129         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3130         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3131     }
3132 
3133     return NULL;
3134 }
3135 
3136 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3137 {
3138     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3139 
3140     return g_strdup(spapr->kvm_type);
3141 }
3142 
3143 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3144 {
3145     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3146 
3147     g_free(spapr->kvm_type);
3148     spapr->kvm_type = g_strdup(value);
3149 }
3150 
3151 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3152 {
3153     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3154 
3155     return spapr->use_hotplug_event_source;
3156 }
3157 
3158 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3159                                             Error **errp)
3160 {
3161     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3162 
3163     spapr->use_hotplug_event_source = value;
3164 }
3165 
3166 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3167 {
3168     return true;
3169 }
3170 
3171 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3172 {
3173     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3174 
3175     switch (spapr->resize_hpt) {
3176     case SPAPR_RESIZE_HPT_DEFAULT:
3177         return g_strdup("default");
3178     case SPAPR_RESIZE_HPT_DISABLED:
3179         return g_strdup("disabled");
3180     case SPAPR_RESIZE_HPT_ENABLED:
3181         return g_strdup("enabled");
3182     case SPAPR_RESIZE_HPT_REQUIRED:
3183         return g_strdup("required");
3184     }
3185     g_assert_not_reached();
3186 }
3187 
3188 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3189 {
3190     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3191 
3192     if (strcmp(value, "default") == 0) {
3193         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3194     } else if (strcmp(value, "disabled") == 0) {
3195         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3196     } else if (strcmp(value, "enabled") == 0) {
3197         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3198     } else if (strcmp(value, "required") == 0) {
3199         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3200     } else {
3201         error_setg(errp, "Bad value for \"resize-hpt\" property");
3202     }
3203 }
3204 
3205 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3206                                    void *opaque, Error **errp)
3207 {
3208     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3209 }
3210 
3211 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3212                                    void *opaque, Error **errp)
3213 {
3214     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3215 }
3216 
3217 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3218 {
3219     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3220 
3221     if (spapr->irq == &spapr_irq_xics_legacy) {
3222         return g_strdup("legacy");
3223     } else if (spapr->irq == &spapr_irq_xics) {
3224         return g_strdup("xics");
3225     } else if (spapr->irq == &spapr_irq_xive) {
3226         return g_strdup("xive");
3227     } else if (spapr->irq == &spapr_irq_dual) {
3228         return g_strdup("dual");
3229     }
3230     g_assert_not_reached();
3231 }
3232 
3233 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3234 {
3235     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3236 
3237     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3238         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3239         return;
3240     }
3241 
3242     /* The legacy IRQ backend can not be set */
3243     if (strcmp(value, "xics") == 0) {
3244         spapr->irq = &spapr_irq_xics;
3245     } else if (strcmp(value, "xive") == 0) {
3246         spapr->irq = &spapr_irq_xive;
3247     } else if (strcmp(value, "dual") == 0) {
3248         spapr->irq = &spapr_irq_dual;
3249     } else {
3250         error_setg(errp, "Bad value for \"ic-mode\" property");
3251     }
3252 }
3253 
3254 static char *spapr_get_host_model(Object *obj, Error **errp)
3255 {
3256     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3257 
3258     return g_strdup(spapr->host_model);
3259 }
3260 
3261 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3262 {
3263     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3264 
3265     g_free(spapr->host_model);
3266     spapr->host_model = g_strdup(value);
3267 }
3268 
3269 static char *spapr_get_host_serial(Object *obj, Error **errp)
3270 {
3271     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3272 
3273     return g_strdup(spapr->host_serial);
3274 }
3275 
3276 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3277 {
3278     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3279 
3280     g_free(spapr->host_serial);
3281     spapr->host_serial = g_strdup(value);
3282 }
3283 
3284 static void spapr_instance_init(Object *obj)
3285 {
3286     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3287     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3288 
3289     spapr->htab_fd = -1;
3290     spapr->use_hotplug_event_source = true;
3291     object_property_add_str(obj, "kvm-type",
3292                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3293     object_property_set_description(obj, "kvm-type",
3294                                     "Specifies the KVM virtualization mode (HV, PR)",
3295                                     NULL);
3296     object_property_add_bool(obj, "modern-hotplug-events",
3297                             spapr_get_modern_hotplug_events,
3298                             spapr_set_modern_hotplug_events,
3299                             NULL);
3300     object_property_set_description(obj, "modern-hotplug-events",
3301                                     "Use dedicated hotplug event mechanism in"
3302                                     " place of standard EPOW events when possible"
3303                                     " (required for memory hot-unplug support)",
3304                                     NULL);
3305     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3306                             "Maximum permitted CPU compatibility mode",
3307                             &error_fatal);
3308 
3309     object_property_add_str(obj, "resize-hpt",
3310                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3311     object_property_set_description(obj, "resize-hpt",
3312                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3313                                     NULL);
3314     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3315                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3316     object_property_set_description(obj, "vsmt",
3317                                     "Virtual SMT: KVM behaves as if this were"
3318                                     " the host's SMT mode", &error_abort);
3319     object_property_add_bool(obj, "vfio-no-msix-emulation",
3320                              spapr_get_msix_emulation, NULL, NULL);
3321 
3322     /* The machine class defines the default interrupt controller mode */
3323     spapr->irq = smc->irq;
3324     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3325                             spapr_set_ic_mode, NULL);
3326     object_property_set_description(obj, "ic-mode",
3327                  "Specifies the interrupt controller mode (xics, xive, dual)",
3328                  NULL);
3329 
3330     object_property_add_str(obj, "host-model",
3331         spapr_get_host_model, spapr_set_host_model,
3332         &error_abort);
3333     object_property_set_description(obj, "host-model",
3334         "Host model to advertise in guest device tree", &error_abort);
3335     object_property_add_str(obj, "host-serial",
3336         spapr_get_host_serial, spapr_set_host_serial,
3337         &error_abort);
3338     object_property_set_description(obj, "host-serial",
3339         "Host serial number to advertise in guest device tree", &error_abort);
3340 }
3341 
3342 static void spapr_machine_finalizefn(Object *obj)
3343 {
3344     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
3345 
3346     g_free(spapr->kvm_type);
3347 }
3348 
3349 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3350 {
3351     cpu_synchronize_state(cs);
3352     ppc_cpu_do_system_reset(cs);
3353 }
3354 
3355 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3356 {
3357     CPUState *cs;
3358 
3359     CPU_FOREACH(cs) {
3360         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3361     }
3362 }
3363 
3364 int spapr_lmb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3365                           void *fdt, int *fdt_start_offset, Error **errp)
3366 {
3367     uint64_t addr;
3368     uint32_t node;
3369 
3370     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3371     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3372                                     &error_abort);
3373     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3374                                                    SPAPR_MEMORY_BLOCK_SIZE);
3375     return 0;
3376 }
3377 
3378 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3379                            bool dedicated_hp_event_source, Error **errp)
3380 {
3381     SpaprDrc *drc;
3382     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3383     int i;
3384     uint64_t addr = addr_start;
3385     bool hotplugged = spapr_drc_hotplugged(dev);
3386     Error *local_err = NULL;
3387 
3388     for (i = 0; i < nr_lmbs; i++) {
3389         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3390                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3391         g_assert(drc);
3392 
3393         spapr_drc_attach(drc, dev, &local_err);
3394         if (local_err) {
3395             while (addr > addr_start) {
3396                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3397                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3398                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3399                 spapr_drc_detach(drc);
3400             }
3401             error_propagate(errp, local_err);
3402             return;
3403         }
3404         if (!hotplugged) {
3405             spapr_drc_reset(drc);
3406         }
3407         addr += SPAPR_MEMORY_BLOCK_SIZE;
3408     }
3409     /* send hotplug notification to the
3410      * guest only in case of hotplugged memory
3411      */
3412     if (hotplugged) {
3413         if (dedicated_hp_event_source) {
3414             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3415                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3416             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3417                                                    nr_lmbs,
3418                                                    spapr_drc_index(drc));
3419         } else {
3420             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3421                                            nr_lmbs);
3422         }
3423     }
3424 }
3425 
3426 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3427                               Error **errp)
3428 {
3429     Error *local_err = NULL;
3430     SpaprMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3431     PCDIMMDevice *dimm = PC_DIMM(dev);
3432     uint64_t size, addr;
3433 
3434     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3435 
3436     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3437     if (local_err) {
3438         goto out;
3439     }
3440 
3441     addr = object_property_get_uint(OBJECT(dimm),
3442                                     PC_DIMM_ADDR_PROP, &local_err);
3443     if (local_err) {
3444         goto out_unplug;
3445     }
3446 
3447     spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3448                    &local_err);
3449     if (local_err) {
3450         goto out_unplug;
3451     }
3452 
3453     return;
3454 
3455 out_unplug:
3456     pc_dimm_unplug(dimm, MACHINE(ms));
3457 out:
3458     error_propagate(errp, local_err);
3459 }
3460 
3461 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3462                                   Error **errp)
3463 {
3464     const SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3465     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3466     PCDIMMDevice *dimm = PC_DIMM(dev);
3467     Error *local_err = NULL;
3468     uint64_t size;
3469     Object *memdev;
3470     hwaddr pagesize;
3471 
3472     if (!smc->dr_lmb_enabled) {
3473         error_setg(errp, "Memory hotplug not supported for this machine");
3474         return;
3475     }
3476 
3477     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3478     if (local_err) {
3479         error_propagate(errp, local_err);
3480         return;
3481     }
3482 
3483     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3484         error_setg(errp, "Hotplugged memory size must be a multiple of "
3485                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3486         return;
3487     }
3488 
3489     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3490                                       &error_abort);
3491     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3492     spapr_check_pagesize(spapr, pagesize, &local_err);
3493     if (local_err) {
3494         error_propagate(errp, local_err);
3495         return;
3496     }
3497 
3498     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3499 }
3500 
3501 struct SpaprDimmState {
3502     PCDIMMDevice *dimm;
3503     uint32_t nr_lmbs;
3504     QTAILQ_ENTRY(SpaprDimmState) next;
3505 };
3506 
3507 static SpaprDimmState *spapr_pending_dimm_unplugs_find(SpaprMachineState *s,
3508                                                        PCDIMMDevice *dimm)
3509 {
3510     SpaprDimmState *dimm_state = NULL;
3511 
3512     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3513         if (dimm_state->dimm == dimm) {
3514             break;
3515         }
3516     }
3517     return dimm_state;
3518 }
3519 
3520 static SpaprDimmState *spapr_pending_dimm_unplugs_add(SpaprMachineState *spapr,
3521                                                       uint32_t nr_lmbs,
3522                                                       PCDIMMDevice *dimm)
3523 {
3524     SpaprDimmState *ds = NULL;
3525 
3526     /*
3527      * If this request is for a DIMM whose removal had failed earlier
3528      * (due to guest's refusal to remove the LMBs), we would have this
3529      * dimm already in the pending_dimm_unplugs list. In that
3530      * case don't add again.
3531      */
3532     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3533     if (!ds) {
3534         ds = g_malloc0(sizeof(SpaprDimmState));
3535         ds->nr_lmbs = nr_lmbs;
3536         ds->dimm = dimm;
3537         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3538     }
3539     return ds;
3540 }
3541 
3542 static void spapr_pending_dimm_unplugs_remove(SpaprMachineState *spapr,
3543                                               SpaprDimmState *dimm_state)
3544 {
3545     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3546     g_free(dimm_state);
3547 }
3548 
3549 static SpaprDimmState *spapr_recover_pending_dimm_state(SpaprMachineState *ms,
3550                                                         PCDIMMDevice *dimm)
3551 {
3552     SpaprDrc *drc;
3553     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3554                                                   &error_abort);
3555     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3556     uint32_t avail_lmbs = 0;
3557     uint64_t addr_start, addr;
3558     int i;
3559 
3560     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3561                                          &error_abort);
3562 
3563     addr = addr_start;
3564     for (i = 0; i < nr_lmbs; i++) {
3565         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3566                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3567         g_assert(drc);
3568         if (drc->dev) {
3569             avail_lmbs++;
3570         }
3571         addr += SPAPR_MEMORY_BLOCK_SIZE;
3572     }
3573 
3574     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3575 }
3576 
3577 /* Callback to be called during DRC release. */
3578 void spapr_lmb_release(DeviceState *dev)
3579 {
3580     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3581     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3582     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3583 
3584     /* This information will get lost if a migration occurs
3585      * during the unplug process. In this case recover it. */
3586     if (ds == NULL) {
3587         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3588         g_assert(ds);
3589         /* The DRC being examined by the caller at least must be counted */
3590         g_assert(ds->nr_lmbs);
3591     }
3592 
3593     if (--ds->nr_lmbs) {
3594         return;
3595     }
3596 
3597     /*
3598      * Now that all the LMBs have been removed by the guest, call the
3599      * unplug handler chain. This can never fail.
3600      */
3601     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3602     object_unparent(OBJECT(dev));
3603 }
3604 
3605 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3606 {
3607     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3608     SpaprDimmState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3609 
3610     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3611     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3612     spapr_pending_dimm_unplugs_remove(spapr, ds);
3613 }
3614 
3615 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3616                                         DeviceState *dev, Error **errp)
3617 {
3618     SpaprMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3619     Error *local_err = NULL;
3620     PCDIMMDevice *dimm = PC_DIMM(dev);
3621     uint32_t nr_lmbs;
3622     uint64_t size, addr_start, addr;
3623     int i;
3624     SpaprDrc *drc;
3625 
3626     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3627     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3628 
3629     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3630                                          &local_err);
3631     if (local_err) {
3632         goto out;
3633     }
3634 
3635     /*
3636      * An existing pending dimm state for this DIMM means that there is an
3637      * unplug operation in progress, waiting for the spapr_lmb_release
3638      * callback to complete the job (BQL can't cover that far). In this case,
3639      * bail out to avoid detaching DRCs that were already released.
3640      */
3641     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3642         error_setg(&local_err,
3643                    "Memory unplug already in progress for device %s",
3644                    dev->id);
3645         goto out;
3646     }
3647 
3648     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3649 
3650     addr = addr_start;
3651     for (i = 0; i < nr_lmbs; i++) {
3652         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3653                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3654         g_assert(drc);
3655 
3656         spapr_drc_detach(drc);
3657         addr += SPAPR_MEMORY_BLOCK_SIZE;
3658     }
3659 
3660     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3661                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3662     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3663                                               nr_lmbs, spapr_drc_index(drc));
3664 out:
3665     error_propagate(errp, local_err);
3666 }
3667 
3668 /* Callback to be called during DRC release. */
3669 void spapr_core_release(DeviceState *dev)
3670 {
3671     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3672 
3673     /* Call the unplug handler chain. This can never fail. */
3674     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3675     object_unparent(OBJECT(dev));
3676 }
3677 
3678 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3679 {
3680     MachineState *ms = MACHINE(hotplug_dev);
3681     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3682     CPUCore *cc = CPU_CORE(dev);
3683     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3684 
3685     if (smc->pre_2_10_has_unused_icps) {
3686         SpaprCpuCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3687         int i;
3688 
3689         for (i = 0; i < cc->nr_threads; i++) {
3690             CPUState *cs = CPU(sc->threads[i]);
3691 
3692             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3693         }
3694     }
3695 
3696     assert(core_slot);
3697     core_slot->cpu = NULL;
3698     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3699 }
3700 
3701 static
3702 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3703                                Error **errp)
3704 {
3705     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3706     int index;
3707     SpaprDrc *drc;
3708     CPUCore *cc = CPU_CORE(dev);
3709 
3710     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3711         error_setg(errp, "Unable to find CPU core with core-id: %d",
3712                    cc->core_id);
3713         return;
3714     }
3715     if (index == 0) {
3716         error_setg(errp, "Boot CPU core may not be unplugged");
3717         return;
3718     }
3719 
3720     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3721                           spapr_vcpu_id(spapr, cc->core_id));
3722     g_assert(drc);
3723 
3724     spapr_drc_detach(drc);
3725 
3726     spapr_hotplug_req_remove_by_index(drc);
3727 }
3728 
3729 int spapr_core_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3730                            void *fdt, int *fdt_start_offset, Error **errp)
3731 {
3732     SpaprCpuCore *core = SPAPR_CPU_CORE(drc->dev);
3733     CPUState *cs = CPU(core->threads[0]);
3734     PowerPCCPU *cpu = POWERPC_CPU(cs);
3735     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3736     int id = spapr_get_vcpu_id(cpu);
3737     char *nodename;
3738     int offset;
3739 
3740     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3741     offset = fdt_add_subnode(fdt, 0, nodename);
3742     g_free(nodename);
3743 
3744     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3745 
3746     *fdt_start_offset = offset;
3747     return 0;
3748 }
3749 
3750 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3751                             Error **errp)
3752 {
3753     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3754     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3755     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3756     SpaprCpuCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3757     CPUCore *cc = CPU_CORE(dev);
3758     CPUState *cs;
3759     SpaprDrc *drc;
3760     Error *local_err = NULL;
3761     CPUArchId *core_slot;
3762     int index;
3763     bool hotplugged = spapr_drc_hotplugged(dev);
3764 
3765     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3766     if (!core_slot) {
3767         error_setg(errp, "Unable to find CPU core with core-id: %d",
3768                    cc->core_id);
3769         return;
3770     }
3771     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3772                           spapr_vcpu_id(spapr, cc->core_id));
3773 
3774     g_assert(drc || !mc->has_hotpluggable_cpus);
3775 
3776     if (drc) {
3777         spapr_drc_attach(drc, dev, &local_err);
3778         if (local_err) {
3779             error_propagate(errp, local_err);
3780             return;
3781         }
3782 
3783         if (hotplugged) {
3784             /*
3785              * Send hotplug notification interrupt to the guest only
3786              * in case of hotplugged CPUs.
3787              */
3788             spapr_hotplug_req_add_by_index(drc);
3789         } else {
3790             spapr_drc_reset(drc);
3791         }
3792     }
3793 
3794     core_slot->cpu = OBJECT(dev);
3795 
3796     if (smc->pre_2_10_has_unused_icps) {
3797         int i;
3798 
3799         for (i = 0; i < cc->nr_threads; i++) {
3800             cs = CPU(core->threads[i]);
3801             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3802         }
3803     }
3804 }
3805 
3806 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3807                                 Error **errp)
3808 {
3809     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3810     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3811     Error *local_err = NULL;
3812     CPUCore *cc = CPU_CORE(dev);
3813     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3814     const char *type = object_get_typename(OBJECT(dev));
3815     CPUArchId *core_slot;
3816     int index;
3817 
3818     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3819         error_setg(&local_err, "CPU hotplug not supported for this machine");
3820         goto out;
3821     }
3822 
3823     if (strcmp(base_core_type, type)) {
3824         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3825         goto out;
3826     }
3827 
3828     if (cc->core_id % smp_threads) {
3829         error_setg(&local_err, "invalid core id %d", cc->core_id);
3830         goto out;
3831     }
3832 
3833     /*
3834      * In general we should have homogeneous threads-per-core, but old
3835      * (pre hotplug support) machine types allow the last core to have
3836      * reduced threads as a compatibility hack for when we allowed
3837      * total vcpus not a multiple of threads-per-core.
3838      */
3839     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3840         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3841                    cc->nr_threads, smp_threads);
3842         goto out;
3843     }
3844 
3845     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3846     if (!core_slot) {
3847         error_setg(&local_err, "core id %d out of range", cc->core_id);
3848         goto out;
3849     }
3850 
3851     if (core_slot->cpu) {
3852         error_setg(&local_err, "core %d already populated", cc->core_id);
3853         goto out;
3854     }
3855 
3856     numa_cpu_pre_plug(core_slot, dev, &local_err);
3857 
3858 out:
3859     error_propagate(errp, local_err);
3860 }
3861 
3862 int spapr_phb_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr,
3863                           void *fdt, int *fdt_start_offset, Error **errp)
3864 {
3865     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3866     int intc_phandle;
3867 
3868     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3869     if (intc_phandle <= 0) {
3870         return -1;
3871     }
3872 
3873     if (spapr_populate_pci_dt(sphb, intc_phandle, fdt, spapr->irq->nr_msis,
3874                               fdt_start_offset)) {
3875         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3876         return -1;
3877     }
3878 
3879     /* generally SLOF creates these, for hotplug it's up to QEMU */
3880     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3881 
3882     return 0;
3883 }
3884 
3885 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3886                                Error **errp)
3887 {
3888     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3889     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3890     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3891     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3892 
3893     if (dev->hotplugged && !smc->dr_phb_enabled) {
3894         error_setg(errp, "PHB hotplug not supported for this machine");
3895         return;
3896     }
3897 
3898     if (sphb->index == (uint32_t)-1) {
3899         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3900         return;
3901     }
3902 
3903     /*
3904      * This will check that sphb->index doesn't exceed the maximum number of
3905      * PHBs for the current machine type.
3906      */
3907     smc->phb_placement(spapr, sphb->index,
3908                        &sphb->buid, &sphb->io_win_addr,
3909                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3910                        windows_supported, sphb->dma_liobn, errp);
3911 }
3912 
3913 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3914                            Error **errp)
3915 {
3916     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3917     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3918     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3919     SpaprDrc *drc;
3920     bool hotplugged = spapr_drc_hotplugged(dev);
3921     Error *local_err = NULL;
3922 
3923     if (!smc->dr_phb_enabled) {
3924         return;
3925     }
3926 
3927     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3928     /* hotplug hooks should check it's enabled before getting this far */
3929     assert(drc);
3930 
3931     spapr_drc_attach(drc, DEVICE(dev), &local_err);
3932     if (local_err) {
3933         error_propagate(errp, local_err);
3934         return;
3935     }
3936 
3937     if (hotplugged) {
3938         spapr_hotplug_req_add_by_index(drc);
3939     } else {
3940         spapr_drc_reset(drc);
3941     }
3942 }
3943 
3944 void spapr_phb_release(DeviceState *dev)
3945 {
3946     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3947 
3948     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3949     object_unparent(OBJECT(dev));
3950 }
3951 
3952 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3953 {
3954     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3955 }
3956 
3957 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3958                                      DeviceState *dev, Error **errp)
3959 {
3960     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3961     SpaprDrc *drc;
3962 
3963     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3964     assert(drc);
3965 
3966     if (!spapr_drc_unplug_requested(drc)) {
3967         spapr_drc_detach(drc);
3968         spapr_hotplug_req_remove_by_index(drc);
3969     }
3970 }
3971 
3972 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3973                                       DeviceState *dev, Error **errp)
3974 {
3975     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3976         spapr_memory_plug(hotplug_dev, dev, errp);
3977     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3978         spapr_core_plug(hotplug_dev, dev, errp);
3979     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
3980         spapr_phb_plug(hotplug_dev, dev, errp);
3981     }
3982 }
3983 
3984 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3985                                         DeviceState *dev, Error **errp)
3986 {
3987     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3988         spapr_memory_unplug(hotplug_dev, dev);
3989     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3990         spapr_core_unplug(hotplug_dev, dev);
3991     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
3992         spapr_phb_unplug(hotplug_dev, dev);
3993     }
3994 }
3995 
3996 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3997                                                 DeviceState *dev, Error **errp)
3998 {
3999     SpaprMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
4000     MachineClass *mc = MACHINE_GET_CLASS(sms);
4001     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4002 
4003     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4004         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4005             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4006         } else {
4007             /* NOTE: this means there is a window after guest reset, prior to
4008              * CAS negotiation, where unplug requests will fail due to the
4009              * capability not being detected yet. This is a bit different than
4010              * the case with PCI unplug, where the events will be queued and
4011              * eventually handled by the guest after boot
4012              */
4013             error_setg(errp, "Memory hot unplug not supported for this guest");
4014         }
4015     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4016         if (!mc->has_hotpluggable_cpus) {
4017             error_setg(errp, "CPU hot unplug not supported on this machine");
4018             return;
4019         }
4020         spapr_core_unplug_request(hotplug_dev, dev, errp);
4021     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4022         if (!smc->dr_phb_enabled) {
4023             error_setg(errp, "PHB hot unplug not supported on this machine");
4024             return;
4025         }
4026         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4027     }
4028 }
4029 
4030 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4031                                           DeviceState *dev, Error **errp)
4032 {
4033     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4034         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4035     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4036         spapr_core_pre_plug(hotplug_dev, dev, errp);
4037     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4038         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4039     }
4040 }
4041 
4042 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4043                                                  DeviceState *dev)
4044 {
4045     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4046         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4047         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4048         return HOTPLUG_HANDLER(machine);
4049     }
4050     return NULL;
4051 }
4052 
4053 static CpuInstanceProperties
4054 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4055 {
4056     CPUArchId *core_slot;
4057     MachineClass *mc = MACHINE_GET_CLASS(machine);
4058 
4059     /* make sure possible_cpu are intialized */
4060     mc->possible_cpu_arch_ids(machine);
4061     /* get CPU core slot containing thread that matches cpu_index */
4062     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4063     assert(core_slot);
4064     return core_slot->props;
4065 }
4066 
4067 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4068 {
4069     return idx / smp_cores % nb_numa_nodes;
4070 }
4071 
4072 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4073 {
4074     int i;
4075     const char *core_type;
4076     int spapr_max_cores = max_cpus / smp_threads;
4077     MachineClass *mc = MACHINE_GET_CLASS(machine);
4078 
4079     if (!mc->has_hotpluggable_cpus) {
4080         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4081     }
4082     if (machine->possible_cpus) {
4083         assert(machine->possible_cpus->len == spapr_max_cores);
4084         return machine->possible_cpus;
4085     }
4086 
4087     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4088     if (!core_type) {
4089         error_report("Unable to find sPAPR CPU Core definition");
4090         exit(1);
4091     }
4092 
4093     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4094                              sizeof(CPUArchId) * spapr_max_cores);
4095     machine->possible_cpus->len = spapr_max_cores;
4096     for (i = 0; i < machine->possible_cpus->len; i++) {
4097         int core_id = i * smp_threads;
4098 
4099         machine->possible_cpus->cpus[i].type = core_type;
4100         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4101         machine->possible_cpus->cpus[i].arch_id = core_id;
4102         machine->possible_cpus->cpus[i].props.has_core_id = true;
4103         machine->possible_cpus->cpus[i].props.core_id = core_id;
4104     }
4105     return machine->possible_cpus;
4106 }
4107 
4108 static void spapr_phb_placement(SpaprMachineState *spapr, uint32_t index,
4109                                 uint64_t *buid, hwaddr *pio,
4110                                 hwaddr *mmio32, hwaddr *mmio64,
4111                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4112 {
4113     /*
4114      * New-style PHB window placement.
4115      *
4116      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4117      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4118      * windows.
4119      *
4120      * Some guest kernels can't work with MMIO windows above 1<<46
4121      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4122      *
4123      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4124      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4125      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4126      * 1TiB 64-bit MMIO windows for each PHB.
4127      */
4128     const uint64_t base_buid = 0x800000020000000ULL;
4129     int i;
4130 
4131     /* Sanity check natural alignments */
4132     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4133     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4134     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4135     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4136     /* Sanity check bounds */
4137     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4138                       SPAPR_PCI_MEM32_WIN_SIZE);
4139     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4140                       SPAPR_PCI_MEM64_WIN_SIZE);
4141 
4142     if (index >= SPAPR_MAX_PHBS) {
4143         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4144                    SPAPR_MAX_PHBS - 1);
4145         return;
4146     }
4147 
4148     *buid = base_buid + index;
4149     for (i = 0; i < n_dma; ++i) {
4150         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4151     }
4152 
4153     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4154     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4155     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4156 }
4157 
4158 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4159 {
4160     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4161 
4162     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4163 }
4164 
4165 static void spapr_ics_resend(XICSFabric *dev)
4166 {
4167     SpaprMachineState *spapr = SPAPR_MACHINE(dev);
4168 
4169     ics_resend(spapr->ics);
4170 }
4171 
4172 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4173 {
4174     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4175 
4176     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4177 }
4178 
4179 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4180                                  Monitor *mon)
4181 {
4182     SpaprMachineState *spapr = SPAPR_MACHINE(obj);
4183 
4184     spapr->irq->print_info(spapr, mon);
4185 }
4186 
4187 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4188 {
4189     return cpu->vcpu_id;
4190 }
4191 
4192 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4193 {
4194     SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4195     int vcpu_id;
4196 
4197     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4198 
4199     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4200         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4201         error_append_hint(errp, "Adjust the number of cpus to %d "
4202                           "or try to raise the number of threads per core\n",
4203                           vcpu_id * smp_threads / spapr->vsmt);
4204         return;
4205     }
4206 
4207     cpu->vcpu_id = vcpu_id;
4208 }
4209 
4210 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4211 {
4212     CPUState *cs;
4213 
4214     CPU_FOREACH(cs) {
4215         PowerPCCPU *cpu = POWERPC_CPU(cs);
4216 
4217         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4218             return cpu;
4219         }
4220     }
4221 
4222     return NULL;
4223 }
4224 
4225 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4226 {
4227     MachineClass *mc = MACHINE_CLASS(oc);
4228     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4229     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4230     NMIClass *nc = NMI_CLASS(oc);
4231     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4232     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4233     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4234     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4235 
4236     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4237     mc->ignore_boot_device_suffixes = true;
4238 
4239     /*
4240      * We set up the default / latest behaviour here.  The class_init
4241      * functions for the specific versioned machine types can override
4242      * these details for backwards compatibility
4243      */
4244     mc->init = spapr_machine_init;
4245     mc->reset = spapr_machine_reset;
4246     mc->block_default_type = IF_SCSI;
4247     mc->max_cpus = 1024;
4248     mc->no_parallel = 1;
4249     mc->default_boot_order = "";
4250     mc->default_ram_size = 512 * MiB;
4251     mc->default_display = "std";
4252     mc->kvm_type = spapr_kvm_type;
4253     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4254     mc->pci_allow_0_address = true;
4255     assert(!mc->get_hotplug_handler);
4256     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4257     hc->pre_plug = spapr_machine_device_pre_plug;
4258     hc->plug = spapr_machine_device_plug;
4259     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4260     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4261     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4262     hc->unplug_request = spapr_machine_device_unplug_request;
4263     hc->unplug = spapr_machine_device_unplug;
4264 
4265     smc->dr_lmb_enabled = true;
4266     smc->update_dt_enabled = true;
4267     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4268     mc->has_hotpluggable_cpus = true;
4269     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4270     fwc->get_dev_path = spapr_get_fw_dev_path;
4271     nc->nmi_monitor_handler = spapr_nmi;
4272     smc->phb_placement = spapr_phb_placement;
4273     vhc->hypercall = emulate_spapr_hypercall;
4274     vhc->hpt_mask = spapr_hpt_mask;
4275     vhc->map_hptes = spapr_map_hptes;
4276     vhc->unmap_hptes = spapr_unmap_hptes;
4277     vhc->store_hpte = spapr_store_hpte;
4278     vhc->get_pate = spapr_get_pate;
4279     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4280     xic->ics_get = spapr_ics_get;
4281     xic->ics_resend = spapr_ics_resend;
4282     xic->icp_get = spapr_icp_get;
4283     ispc->print_info = spapr_pic_print_info;
4284     /* Force NUMA node memory size to be a multiple of
4285      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4286      * in which LMBs are represented and hot-added
4287      */
4288     mc->numa_mem_align_shift = 28;
4289 
4290     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4291     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4292     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4293     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4294     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4295     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_WORKAROUND;
4296     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4297     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4298     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_ON;
4299     smc->default_caps.caps[SPAPR_CAP_CCF_ASSIST] = SPAPR_CAP_OFF;
4300     spapr_caps_add_properties(smc, &error_abort);
4301     smc->irq = &spapr_irq_xics;
4302     smc->dr_phb_enabled = true;
4303 }
4304 
4305 static const TypeInfo spapr_machine_info = {
4306     .name          = TYPE_SPAPR_MACHINE,
4307     .parent        = TYPE_MACHINE,
4308     .abstract      = true,
4309     .instance_size = sizeof(SpaprMachineState),
4310     .instance_init = spapr_instance_init,
4311     .instance_finalize = spapr_machine_finalizefn,
4312     .class_size    = sizeof(SpaprMachineClass),
4313     .class_init    = spapr_machine_class_init,
4314     .interfaces = (InterfaceInfo[]) {
4315         { TYPE_FW_PATH_PROVIDER },
4316         { TYPE_NMI },
4317         { TYPE_HOTPLUG_HANDLER },
4318         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4319         { TYPE_XICS_FABRIC },
4320         { TYPE_INTERRUPT_STATS_PROVIDER },
4321         { }
4322     },
4323 };
4324 
4325 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4326     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4327                                                     void *data)      \
4328     {                                                                \
4329         MachineClass *mc = MACHINE_CLASS(oc);                        \
4330         spapr_machine_##suffix##_class_options(mc);                  \
4331         if (latest) {                                                \
4332             mc->alias = "pseries";                                   \
4333             mc->is_default = 1;                                      \
4334         }                                                            \
4335     }                                                                \
4336     static const TypeInfo spapr_machine_##suffix##_info = {          \
4337         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4338         .parent = TYPE_SPAPR_MACHINE,                                \
4339         .class_init = spapr_machine_##suffix##_class_init,           \
4340     };                                                               \
4341     static void spapr_machine_register_##suffix(void)                \
4342     {                                                                \
4343         type_register(&spapr_machine_##suffix##_info);               \
4344     }                                                                \
4345     type_init(spapr_machine_register_##suffix)
4346 
4347 /*
4348  * pseries-4.1
4349  */
4350 static void spapr_machine_4_1_class_options(MachineClass *mc)
4351 {
4352     /* Defaults for the latest behaviour inherited from the base class */
4353 }
4354 
4355 DEFINE_SPAPR_MACHINE(4_1, "4.1", true);
4356 
4357 /*
4358  * pseries-4.0
4359  */
4360 static void spapr_machine_4_0_class_options(MachineClass *mc)
4361 {
4362     spapr_machine_4_1_class_options(mc);
4363     compat_props_add(mc->compat_props, hw_compat_4_0, hw_compat_4_0_len);
4364 }
4365 
4366 DEFINE_SPAPR_MACHINE(4_0, "4.0", false);
4367 
4368 /*
4369  * pseries-3.1
4370  */
4371 static void spapr_machine_3_1_class_options(MachineClass *mc)
4372 {
4373     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4374 
4375     spapr_machine_4_0_class_options(mc);
4376     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4377 
4378     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4379     smc->update_dt_enabled = false;
4380     smc->dr_phb_enabled = false;
4381     smc->broken_host_serial_model = true;
4382     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4383     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4384     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4385     smc->default_caps.caps[SPAPR_CAP_LARGE_DECREMENTER] = SPAPR_CAP_OFF;
4386 }
4387 
4388 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4389 
4390 /*
4391  * pseries-3.0
4392  */
4393 
4394 static void spapr_machine_3_0_class_options(MachineClass *mc)
4395 {
4396     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4397 
4398     spapr_machine_3_1_class_options(mc);
4399     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4400 
4401     smc->legacy_irq_allocation = true;
4402     smc->irq = &spapr_irq_xics_legacy;
4403 }
4404 
4405 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4406 
4407 /*
4408  * pseries-2.12
4409  */
4410 static void spapr_machine_2_12_class_options(MachineClass *mc)
4411 {
4412     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4413     static GlobalProperty compat[] = {
4414         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4415         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4416     };
4417 
4418     spapr_machine_3_0_class_options(mc);
4419     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4420     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4421 
4422     /* We depend on kvm_enabled() to choose a default value for the
4423      * hpt-max-page-size capability. Of course we can't do it here
4424      * because this is too early and the HW accelerator isn't initialzed
4425      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4426      */
4427     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4428 }
4429 
4430 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4431 
4432 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4433 {
4434     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4435 
4436     spapr_machine_2_12_class_options(mc);
4437     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4438     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4439     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4440 }
4441 
4442 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4443 
4444 /*
4445  * pseries-2.11
4446  */
4447 
4448 static void spapr_machine_2_11_class_options(MachineClass *mc)
4449 {
4450     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4451 
4452     spapr_machine_2_12_class_options(mc);
4453     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4454     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4455 }
4456 
4457 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4458 
4459 /*
4460  * pseries-2.10
4461  */
4462 
4463 static void spapr_machine_2_10_class_options(MachineClass *mc)
4464 {
4465     spapr_machine_2_11_class_options(mc);
4466     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4467 }
4468 
4469 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4470 
4471 /*
4472  * pseries-2.9
4473  */
4474 
4475 static void spapr_machine_2_9_class_options(MachineClass *mc)
4476 {
4477     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4478     static GlobalProperty compat[] = {
4479         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4480     };
4481 
4482     spapr_machine_2_10_class_options(mc);
4483     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4484     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4485     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4486     smc->pre_2_10_has_unused_icps = true;
4487     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4488 }
4489 
4490 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4491 
4492 /*
4493  * pseries-2.8
4494  */
4495 
4496 static void spapr_machine_2_8_class_options(MachineClass *mc)
4497 {
4498     static GlobalProperty compat[] = {
4499         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4500     };
4501 
4502     spapr_machine_2_9_class_options(mc);
4503     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4504     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4505     mc->numa_mem_align_shift = 23;
4506 }
4507 
4508 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4509 
4510 /*
4511  * pseries-2.7
4512  */
4513 
4514 static void phb_placement_2_7(SpaprMachineState *spapr, uint32_t index,
4515                               uint64_t *buid, hwaddr *pio,
4516                               hwaddr *mmio32, hwaddr *mmio64,
4517                               unsigned n_dma, uint32_t *liobns, Error **errp)
4518 {
4519     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4520     const uint64_t base_buid = 0x800000020000000ULL;
4521     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4522     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4523     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4524     const uint32_t max_index = 255;
4525     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4526 
4527     uint64_t ram_top = MACHINE(spapr)->ram_size;
4528     hwaddr phb0_base, phb_base;
4529     int i;
4530 
4531     /* Do we have device memory? */
4532     if (MACHINE(spapr)->maxram_size > ram_top) {
4533         /* Can't just use maxram_size, because there may be an
4534          * alignment gap between normal and device memory regions
4535          */
4536         ram_top = MACHINE(spapr)->device_memory->base +
4537             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4538     }
4539 
4540     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4541 
4542     if (index > max_index) {
4543         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4544                    max_index);
4545         return;
4546     }
4547 
4548     *buid = base_buid + index;
4549     for (i = 0; i < n_dma; ++i) {
4550         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4551     }
4552 
4553     phb_base = phb0_base + index * phb_spacing;
4554     *pio = phb_base + pio_offset;
4555     *mmio32 = phb_base + mmio_offset;
4556     /*
4557      * We don't set the 64-bit MMIO window, relying on the PHB's
4558      * fallback behaviour of automatically splitting a large "32-bit"
4559      * window into contiguous 32-bit and 64-bit windows
4560      */
4561 }
4562 
4563 static void spapr_machine_2_7_class_options(MachineClass *mc)
4564 {
4565     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4566     static GlobalProperty compat[] = {
4567         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4568         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4569         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4570         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4571     };
4572 
4573     spapr_machine_2_8_class_options(mc);
4574     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4575     mc->default_machine_opts = "modern-hotplug-events=off";
4576     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4577     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4578     smc->phb_placement = phb_placement_2_7;
4579 }
4580 
4581 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4582 
4583 /*
4584  * pseries-2.6
4585  */
4586 
4587 static void spapr_machine_2_6_class_options(MachineClass *mc)
4588 {
4589     static GlobalProperty compat[] = {
4590         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4591     };
4592 
4593     spapr_machine_2_7_class_options(mc);
4594     mc->has_hotpluggable_cpus = false;
4595     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4596     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4597 }
4598 
4599 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4600 
4601 /*
4602  * pseries-2.5
4603  */
4604 
4605 static void spapr_machine_2_5_class_options(MachineClass *mc)
4606 {
4607     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4608     static GlobalProperty compat[] = {
4609         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4610     };
4611 
4612     spapr_machine_2_6_class_options(mc);
4613     smc->use_ohci_by_default = true;
4614     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4615     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4616 }
4617 
4618 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4619 
4620 /*
4621  * pseries-2.4
4622  */
4623 
4624 static void spapr_machine_2_4_class_options(MachineClass *mc)
4625 {
4626     SpaprMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4627 
4628     spapr_machine_2_5_class_options(mc);
4629     smc->dr_lmb_enabled = false;
4630     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4631 }
4632 
4633 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4634 
4635 /*
4636  * pseries-2.3
4637  */
4638 
4639 static void spapr_machine_2_3_class_options(MachineClass *mc)
4640 {
4641     static GlobalProperty compat[] = {
4642         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4643     };
4644     spapr_machine_2_4_class_options(mc);
4645     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4646     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4647 }
4648 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4649 
4650 /*
4651  * pseries-2.2
4652  */
4653 
4654 static void spapr_machine_2_2_class_options(MachineClass *mc)
4655 {
4656     static GlobalProperty compat[] = {
4657         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4658     };
4659 
4660     spapr_machine_2_3_class_options(mc);
4661     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4662     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4663     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4664 }
4665 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4666 
4667 /*
4668  * pseries-2.1
4669  */
4670 
4671 static void spapr_machine_2_1_class_options(MachineClass *mc)
4672 {
4673     spapr_machine_2_2_class_options(mc);
4674     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4675 }
4676 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4677 
4678 static void spapr_machine_register_types(void)
4679 {
4680     type_register_static(&spapr_machine_info);
4681 }
4682 
4683 type_init(spapr_machine_register_types)
4684