xref: /openbmc/qemu/hw/ppc/spapr.c (revision fa59483d)
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 "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
46 #include "cpu-models.h"
47 #include "qom/cpu.h"
48 
49 #include "hw/boards.h"
50 #include "hw/ppc/ppc.h"
51 #include "hw/loader.h"
52 
53 #include "hw/ppc/fdt.h"
54 #include "hw/ppc/spapr.h"
55 #include "hw/ppc/spapr_vio.h"
56 #include "hw/pci-host/spapr.h"
57 #include "hw/pci/msi.h"
58 
59 #include "hw/pci/pci.h"
60 #include "hw/scsi/scsi.h"
61 #include "hw/virtio/virtio-scsi.h"
62 #include "hw/virtio/vhost-scsi-common.h"
63 
64 #include "exec/address-spaces.h"
65 #include "exec/ram_addr.h"
66 #include "hw/usb.h"
67 #include "qemu/config-file.h"
68 #include "qemu/error-report.h"
69 #include "trace.h"
70 #include "hw/nmi.h"
71 #include "hw/intc/intc.h"
72 
73 #include "qemu/cutils.h"
74 #include "hw/ppc/spapr_cpu_core.h"
75 #include "hw/mem/memory-device.h"
76 
77 #include <libfdt.h>
78 
79 /* SLOF memory layout:
80  *
81  * SLOF raw image loaded at 0, copies its romfs right below the flat
82  * device-tree, then position SLOF itself 31M below that
83  *
84  * So we set FW_OVERHEAD to 40MB which should account for all of that
85  * and more
86  *
87  * We load our kernel at 4M, leaving space for SLOF initial image
88  */
89 #define FDT_MAX_SIZE            0x100000
90 #define RTAS_MAX_SIZE           0x10000
91 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
92 #define FW_MAX_SIZE             0x400000
93 #define FW_FILE_NAME            "slof.bin"
94 #define FW_OVERHEAD             0x2800000
95 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
96 
97 #define MIN_RMA_SLOF            128UL
98 
99 #define PHANDLE_INTC            0x00001111
100 
101 /* These two functions implement the VCPU id numbering: one to compute them
102  * all and one to identify thread 0 of a VCORE. Any change to the first one
103  * is likely to have an impact on the second one, so let's keep them close.
104  */
105 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
106 {
107     assert(spapr->vsmt);
108     return
109         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
110 }
111 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
112                                       PowerPCCPU *cpu)
113 {
114     assert(spapr->vsmt);
115     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
116 }
117 
118 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
119 {
120     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
121      * and newer QEMUs don't even have them. In both cases, we don't want
122      * to send anything on the wire.
123      */
124     return false;
125 }
126 
127 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
128     .name = "icp/server",
129     .version_id = 1,
130     .minimum_version_id = 1,
131     .needed = pre_2_10_vmstate_dummy_icp_needed,
132     .fields = (VMStateField[]) {
133         VMSTATE_UNUSED(4), /* uint32_t xirr */
134         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
135         VMSTATE_UNUSED(1), /* uint8_t mfrr */
136         VMSTATE_END_OF_LIST()
137     },
138 };
139 
140 static void pre_2_10_vmstate_register_dummy_icp(int i)
141 {
142     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
143                      (void *)(uintptr_t) i);
144 }
145 
146 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
147 {
148     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
149                        (void *)(uintptr_t) i);
150 }
151 
152 int spapr_max_server_number(sPAPRMachineState *spapr)
153 {
154     assert(spapr->vsmt);
155     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
156 }
157 
158 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
159                                   int smt_threads)
160 {
161     int i, ret = 0;
162     uint32_t servers_prop[smt_threads];
163     uint32_t gservers_prop[smt_threads * 2];
164     int index = spapr_get_vcpu_id(cpu);
165 
166     if (cpu->compat_pvr) {
167         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
168         if (ret < 0) {
169             return ret;
170         }
171     }
172 
173     /* Build interrupt servers and gservers properties */
174     for (i = 0; i < smt_threads; i++) {
175         servers_prop[i] = cpu_to_be32(index + i);
176         /* Hack, direct the group queues back to cpu 0 */
177         gservers_prop[i*2] = cpu_to_be32(index + i);
178         gservers_prop[i*2 + 1] = 0;
179     }
180     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
181                       servers_prop, sizeof(servers_prop));
182     if (ret < 0) {
183         return ret;
184     }
185     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
186                       gservers_prop, sizeof(gservers_prop));
187 
188     return ret;
189 }
190 
191 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
192 {
193     int index = spapr_get_vcpu_id(cpu);
194     uint32_t associativity[] = {cpu_to_be32(0x5),
195                                 cpu_to_be32(0x0),
196                                 cpu_to_be32(0x0),
197                                 cpu_to_be32(0x0),
198                                 cpu_to_be32(cpu->node_id),
199                                 cpu_to_be32(index)};
200 
201     /* Advertise NUMA via ibm,associativity */
202     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
203                           sizeof(associativity));
204 }
205 
206 /* Populate the "ibm,pa-features" property */
207 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
208                                        PowerPCCPU *cpu,
209                                        void *fdt, int offset,
210                                        bool legacy_guest)
211 {
212     uint8_t pa_features_206[] = { 6, 0,
213         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214     uint8_t pa_features_207[] = { 24, 0,
215         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219     uint8_t pa_features_300[] = { 66, 0,
220         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
223         /* 6: DS207 */
224         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
225         /* 16: Vector */
226         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235         /* 42: PM, 44: PC RA, 46: SC vec'd */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237         /* 48: SIMD, 50: QP BFP, 52: String */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239         /* 54: DecFP, 56: DecI, 58: SHA */
240         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241         /* 60: NM atomic, 62: RNG */
242         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
243     };
244     uint8_t *pa_features = NULL;
245     size_t pa_size;
246 
247     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
248         pa_features = pa_features_206;
249         pa_size = sizeof(pa_features_206);
250     }
251     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
252         pa_features = pa_features_207;
253         pa_size = sizeof(pa_features_207);
254     }
255     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
256         pa_features = pa_features_300;
257         pa_size = sizeof(pa_features_300);
258     }
259     if (!pa_features) {
260         return;
261     }
262 
263     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
264         /*
265          * Note: we keep CI large pages off by default because a 64K capable
266          * guest provisioned with large pages might otherwise try to map a qemu
267          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268          * even if that qemu runs on a 4k host.
269          * We dd this bit back here if we are confident this is not an issue
270          */
271         pa_features[3] |= 0x20;
272     }
273     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
274         pa_features[24] |= 0x80;    /* Transactional memory support */
275     }
276     if (legacy_guest && pa_size > 40) {
277         /* Workaround for broken kernels that attempt (guest) radix
278          * mode when they can't handle it, if they see the radix bit set
279          * in pa-features. So hide it from them. */
280         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
281     }
282 
283     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
284 }
285 
286 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
287 {
288     int ret = 0, offset, cpus_offset;
289     CPUState *cs;
290     char cpu_model[32];
291     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
292 
293     CPU_FOREACH(cs) {
294         PowerPCCPU *cpu = POWERPC_CPU(cs);
295         DeviceClass *dc = DEVICE_GET_CLASS(cs);
296         int index = spapr_get_vcpu_id(cpu);
297         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
298 
299         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
300             continue;
301         }
302 
303         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
304 
305         cpus_offset = fdt_path_offset(fdt, "/cpus");
306         if (cpus_offset < 0) {
307             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
308             if (cpus_offset < 0) {
309                 return cpus_offset;
310             }
311         }
312         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
313         if (offset < 0) {
314             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
315             if (offset < 0) {
316                 return offset;
317             }
318         }
319 
320         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
321                           pft_size_prop, sizeof(pft_size_prop));
322         if (ret < 0) {
323             return ret;
324         }
325 
326         if (nb_numa_nodes > 1) {
327             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
328             if (ret < 0) {
329                 return ret;
330             }
331         }
332 
333         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
334         if (ret < 0) {
335             return ret;
336         }
337 
338         spapr_populate_pa_features(spapr, cpu, fdt, offset,
339                                    spapr->cas_legacy_guest_workaround);
340     }
341     return ret;
342 }
343 
344 static hwaddr spapr_node0_size(MachineState *machine)
345 {
346     if (nb_numa_nodes) {
347         int i;
348         for (i = 0; i < nb_numa_nodes; ++i) {
349             if (numa_info[i].node_mem) {
350                 return MIN(pow2floor(numa_info[i].node_mem),
351                            machine->ram_size);
352             }
353         }
354     }
355     return machine->ram_size;
356 }
357 
358 static void add_str(GString *s, const gchar *s1)
359 {
360     g_string_append_len(s, s1, strlen(s1) + 1);
361 }
362 
363 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
364                                        hwaddr size)
365 {
366     uint32_t associativity[] = {
367         cpu_to_be32(0x4), /* length */
368         cpu_to_be32(0x0), cpu_to_be32(0x0),
369         cpu_to_be32(0x0), cpu_to_be32(nodeid)
370     };
371     char mem_name[32];
372     uint64_t mem_reg_property[2];
373     int off;
374 
375     mem_reg_property[0] = cpu_to_be64(start);
376     mem_reg_property[1] = cpu_to_be64(size);
377 
378     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
379     off = fdt_add_subnode(fdt, 0, mem_name);
380     _FDT(off);
381     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
382     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
383                       sizeof(mem_reg_property))));
384     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
385                       sizeof(associativity))));
386     return off;
387 }
388 
389 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
390 {
391     MachineState *machine = MACHINE(spapr);
392     hwaddr mem_start, node_size;
393     int i, nb_nodes = nb_numa_nodes;
394     NodeInfo *nodes = numa_info;
395     NodeInfo ramnode;
396 
397     /* No NUMA nodes, assume there is just one node with whole RAM */
398     if (!nb_numa_nodes) {
399         nb_nodes = 1;
400         ramnode.node_mem = machine->ram_size;
401         nodes = &ramnode;
402     }
403 
404     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
405         if (!nodes[i].node_mem) {
406             continue;
407         }
408         if (mem_start >= machine->ram_size) {
409             node_size = 0;
410         } else {
411             node_size = nodes[i].node_mem;
412             if (node_size > machine->ram_size - mem_start) {
413                 node_size = machine->ram_size - mem_start;
414             }
415         }
416         if (!mem_start) {
417             /* spapr_machine_init() checks for rma_size <= node0_size
418              * already */
419             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
420             mem_start += spapr->rma_size;
421             node_size -= spapr->rma_size;
422         }
423         for ( ; node_size; ) {
424             hwaddr sizetmp = pow2floor(node_size);
425 
426             /* mem_start != 0 here */
427             if (ctzl(mem_start) < ctzl(sizetmp)) {
428                 sizetmp = 1ULL << ctzl(mem_start);
429             }
430 
431             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
432             node_size -= sizetmp;
433             mem_start += sizetmp;
434         }
435     }
436 
437     return 0;
438 }
439 
440 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
441                                   sPAPRMachineState *spapr)
442 {
443     PowerPCCPU *cpu = POWERPC_CPU(cs);
444     CPUPPCState *env = &cpu->env;
445     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
446     int index = spapr_get_vcpu_id(cpu);
447     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
448                        0xffffffff, 0xffffffff};
449     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
450         : SPAPR_TIMEBASE_FREQ;
451     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
452     uint32_t page_sizes_prop[64];
453     size_t page_sizes_prop_size;
454     uint32_t vcpus_per_socket = smp_threads * smp_cores;
455     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
456     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
457     sPAPRDRConnector *drc;
458     int drc_index;
459     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
460     int i;
461 
462     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
463     if (drc) {
464         drc_index = spapr_drc_index(drc);
465         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
466     }
467 
468     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
469     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
470 
471     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
472     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
473                            env->dcache_line_size)));
474     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
475                            env->dcache_line_size)));
476     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
477                            env->icache_line_size)));
478     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
479                            env->icache_line_size)));
480 
481     if (pcc->l1_dcache_size) {
482         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
483                                pcc->l1_dcache_size)));
484     } else {
485         warn_report("Unknown L1 dcache size for cpu");
486     }
487     if (pcc->l1_icache_size) {
488         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
489                                pcc->l1_icache_size)));
490     } else {
491         warn_report("Unknown L1 icache size for cpu");
492     }
493 
494     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
495     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
496     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
497     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
498     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
499     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
500 
501     if (env->spr_cb[SPR_PURR].oea_read) {
502         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
503     }
504 
505     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
506         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
507                           segs, sizeof(segs))));
508     }
509 
510     /* Advertise VSX (vector extensions) if available
511      *   1               == VMX / Altivec available
512      *   2               == VSX available
513      *
514      * Only CPUs for which we create core types in spapr_cpu_core.c
515      * are possible, and all of those have VMX */
516     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
517         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
518     } else {
519         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
520     }
521 
522     /* Advertise DFP (Decimal Floating Point) if available
523      *   0 / no property == no DFP
524      *   1               == DFP available */
525     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
526         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
527     }
528 
529     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
530                                                       sizeof(page_sizes_prop));
531     if (page_sizes_prop_size) {
532         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
533                           page_sizes_prop, page_sizes_prop_size)));
534     }
535 
536     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
537 
538     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
539                            cs->cpu_index / vcpus_per_socket)));
540 
541     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
542                       pft_size_prop, sizeof(pft_size_prop))));
543 
544     if (nb_numa_nodes > 1) {
545         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
546     }
547 
548     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
549 
550     if (pcc->radix_page_info) {
551         for (i = 0; i < pcc->radix_page_info->count; i++) {
552             radix_AP_encodings[i] =
553                 cpu_to_be32(pcc->radix_page_info->entries[i]);
554         }
555         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
556                           radix_AP_encodings,
557                           pcc->radix_page_info->count *
558                           sizeof(radix_AP_encodings[0]))));
559     }
560 }
561 
562 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
563 {
564     CPUState **rev;
565     CPUState *cs;
566     int n_cpus;
567     int cpus_offset;
568     char *nodename;
569     int i;
570 
571     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
572     _FDT(cpus_offset);
573     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
574     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
575 
576     /*
577      * We walk the CPUs in reverse order to ensure that CPU DT nodes
578      * created by fdt_add_subnode() end up in the right order in FDT
579      * for the guest kernel the enumerate the CPUs correctly.
580      *
581      * The CPU list cannot be traversed in reverse order, so we need
582      * to do extra work.
583      */
584     n_cpus = 0;
585     rev = NULL;
586     CPU_FOREACH(cs) {
587         rev = g_renew(CPUState *, rev, n_cpus + 1);
588         rev[n_cpus++] = cs;
589     }
590 
591     for (i = n_cpus - 1; i >= 0; i--) {
592         CPUState *cs = rev[i];
593         PowerPCCPU *cpu = POWERPC_CPU(cs);
594         int index = spapr_get_vcpu_id(cpu);
595         DeviceClass *dc = DEVICE_GET_CLASS(cs);
596         int offset;
597 
598         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
599             continue;
600         }
601 
602         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
603         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
604         g_free(nodename);
605         _FDT(offset);
606         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
607     }
608 
609     g_free(rev);
610 }
611 
612 static int spapr_rng_populate_dt(void *fdt)
613 {
614     int node;
615     int ret;
616 
617     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
618     if (node <= 0) {
619         return -1;
620     }
621     ret = fdt_setprop_string(fdt, node, "device_type",
622                              "ibm,platform-facilities");
623     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
624     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
625 
626     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
627     if (node <= 0) {
628         return -1;
629     }
630     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
631 
632     return ret ? -1 : 0;
633 }
634 
635 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
636 {
637     MemoryDeviceInfoList *info;
638 
639     for (info = list; info; info = info->next) {
640         MemoryDeviceInfo *value = info->value;
641 
642         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
643             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
644 
645             if (addr >= pcdimm_info->addr &&
646                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
647                 return pcdimm_info->node;
648             }
649         }
650     }
651 
652     return -1;
653 }
654 
655 struct sPAPRDrconfCellV2 {
656      uint32_t seq_lmbs;
657      uint64_t base_addr;
658      uint32_t drc_index;
659      uint32_t aa_index;
660      uint32_t flags;
661 } QEMU_PACKED;
662 
663 typedef struct DrconfCellQueue {
664     struct sPAPRDrconfCellV2 cell;
665     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
666 } DrconfCellQueue;
667 
668 static DrconfCellQueue *
669 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
670                       uint32_t drc_index, uint32_t aa_index,
671                       uint32_t flags)
672 {
673     DrconfCellQueue *elem;
674 
675     elem = g_malloc0(sizeof(*elem));
676     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
677     elem->cell.base_addr = cpu_to_be64(base_addr);
678     elem->cell.drc_index = cpu_to_be32(drc_index);
679     elem->cell.aa_index = cpu_to_be32(aa_index);
680     elem->cell.flags = cpu_to_be32(flags);
681 
682     return elem;
683 }
684 
685 /* ibm,dynamic-memory-v2 */
686 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
687                                    int offset, MemoryDeviceInfoList *dimms)
688 {
689     MachineState *machine = MACHINE(spapr);
690     uint8_t *int_buf, *cur_index;
691     int ret;
692     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
693     uint64_t addr, cur_addr, size;
694     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
695     uint64_t mem_end = machine->device_memory->base +
696                        memory_region_size(&machine->device_memory->mr);
697     uint32_t node, buf_len, nr_entries = 0;
698     sPAPRDRConnector *drc;
699     DrconfCellQueue *elem, *next;
700     MemoryDeviceInfoList *info;
701     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
702         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
703 
704     /* Entry to cover RAM and the gap area */
705     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
706                                  SPAPR_LMB_FLAGS_RESERVED |
707                                  SPAPR_LMB_FLAGS_DRC_INVALID);
708     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
709     nr_entries++;
710 
711     cur_addr = machine->device_memory->base;
712     for (info = dimms; info; info = info->next) {
713         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
714 
715         addr = di->addr;
716         size = di->size;
717         node = di->node;
718 
719         /* Entry for hot-pluggable area */
720         if (cur_addr < addr) {
721             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
722             g_assert(drc);
723             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
724                                          cur_addr, spapr_drc_index(drc), -1, 0);
725             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
726             nr_entries++;
727         }
728 
729         /* Entry for DIMM */
730         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
731         g_assert(drc);
732         elem = spapr_get_drconf_cell(size / lmb_size, addr,
733                                      spapr_drc_index(drc), node,
734                                      SPAPR_LMB_FLAGS_ASSIGNED);
735         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
736         nr_entries++;
737         cur_addr = addr + size;
738     }
739 
740     /* Entry for remaining hotpluggable area */
741     if (cur_addr < mem_end) {
742         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
743         g_assert(drc);
744         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
745                                      cur_addr, spapr_drc_index(drc), -1, 0);
746         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
747         nr_entries++;
748     }
749 
750     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
751     int_buf = cur_index = g_malloc0(buf_len);
752     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
753     cur_index += sizeof(nr_entries);
754 
755     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
756         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
757         cur_index += sizeof(elem->cell);
758         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
759         g_free(elem);
760     }
761 
762     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
763     g_free(int_buf);
764     if (ret < 0) {
765         return -1;
766     }
767     return 0;
768 }
769 
770 /* ibm,dynamic-memory */
771 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
772                                    int offset, MemoryDeviceInfoList *dimms)
773 {
774     MachineState *machine = MACHINE(spapr);
775     int i, ret;
776     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
777     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
778     uint32_t nr_lmbs = (machine->device_memory->base +
779                        memory_region_size(&machine->device_memory->mr)) /
780                        lmb_size;
781     uint32_t *int_buf, *cur_index, buf_len;
782 
783     /*
784      * Allocate enough buffer size to fit in ibm,dynamic-memory
785      */
786     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
787     cur_index = int_buf = g_malloc0(buf_len);
788     int_buf[0] = cpu_to_be32(nr_lmbs);
789     cur_index++;
790     for (i = 0; i < nr_lmbs; i++) {
791         uint64_t addr = i * lmb_size;
792         uint32_t *dynamic_memory = cur_index;
793 
794         if (i >= device_lmb_start) {
795             sPAPRDRConnector *drc;
796 
797             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
798             g_assert(drc);
799 
800             dynamic_memory[0] = cpu_to_be32(addr >> 32);
801             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
802             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
803             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
804             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
805             if (memory_region_present(get_system_memory(), addr)) {
806                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
807             } else {
808                 dynamic_memory[5] = cpu_to_be32(0);
809             }
810         } else {
811             /*
812              * LMB information for RMA, boot time RAM and gap b/n RAM and
813              * device memory region -- all these are marked as reserved
814              * and as having no valid DRC.
815              */
816             dynamic_memory[0] = cpu_to_be32(addr >> 32);
817             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
818             dynamic_memory[2] = cpu_to_be32(0);
819             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
820             dynamic_memory[4] = cpu_to_be32(-1);
821             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
822                                             SPAPR_LMB_FLAGS_DRC_INVALID);
823         }
824 
825         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
826     }
827     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
828     g_free(int_buf);
829     if (ret < 0) {
830         return -1;
831     }
832     return 0;
833 }
834 
835 /*
836  * Adds ibm,dynamic-reconfiguration-memory node.
837  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
838  * of this device tree node.
839  */
840 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
841 {
842     MachineState *machine = MACHINE(spapr);
843     int ret, i, offset;
844     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
845     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
846     uint32_t *int_buf, *cur_index, buf_len;
847     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
848     MemoryDeviceInfoList *dimms = NULL;
849 
850     /*
851      * Don't create the node if there is no device memory
852      */
853     if (machine->ram_size == machine->maxram_size) {
854         return 0;
855     }
856 
857     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
858 
859     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
860                     sizeof(prop_lmb_size));
861     if (ret < 0) {
862         return ret;
863     }
864 
865     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
866     if (ret < 0) {
867         return ret;
868     }
869 
870     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
871     if (ret < 0) {
872         return ret;
873     }
874 
875     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
876     dimms = qmp_memory_device_list();
877     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
878         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
879     } else {
880         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
881     }
882     qapi_free_MemoryDeviceInfoList(dimms);
883 
884     if (ret < 0) {
885         return ret;
886     }
887 
888     /* ibm,associativity-lookup-arrays */
889     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
890     cur_index = int_buf = g_malloc0(buf_len);
891     int_buf[0] = cpu_to_be32(nr_nodes);
892     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
893     cur_index += 2;
894     for (i = 0; i < nr_nodes; i++) {
895         uint32_t associativity[] = {
896             cpu_to_be32(0x0),
897             cpu_to_be32(0x0),
898             cpu_to_be32(0x0),
899             cpu_to_be32(i)
900         };
901         memcpy(cur_index, associativity, sizeof(associativity));
902         cur_index += 4;
903     }
904     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
905             (cur_index - int_buf) * sizeof(uint32_t));
906     g_free(int_buf);
907 
908     return ret;
909 }
910 
911 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
912                                 sPAPROptionVector *ov5_updates)
913 {
914     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
915     int ret = 0, offset;
916 
917     /* Generate ibm,dynamic-reconfiguration-memory node if required */
918     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
919         g_assert(smc->dr_lmb_enabled);
920         ret = spapr_populate_drconf_memory(spapr, fdt);
921         if (ret) {
922             goto out;
923         }
924     }
925 
926     offset = fdt_path_offset(fdt, "/chosen");
927     if (offset < 0) {
928         offset = fdt_add_subnode(fdt, 0, "chosen");
929         if (offset < 0) {
930             return offset;
931         }
932     }
933     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
934                                  "ibm,architecture-vec-5");
935 
936 out:
937     return ret;
938 }
939 
940 static bool spapr_hotplugged_dev_before_cas(void)
941 {
942     Object *drc_container, *obj;
943     ObjectProperty *prop;
944     ObjectPropertyIterator iter;
945 
946     drc_container = container_get(object_get_root(), "/dr-connector");
947     object_property_iter_init(&iter, drc_container);
948     while ((prop = object_property_iter_next(&iter))) {
949         if (!strstart(prop->type, "link<", NULL)) {
950             continue;
951         }
952         obj = object_property_get_link(drc_container, prop->name, NULL);
953         if (spapr_drc_needed(obj)) {
954             return true;
955         }
956     }
957     return false;
958 }
959 
960 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
961                                  target_ulong addr, target_ulong size,
962                                  sPAPROptionVector *ov5_updates)
963 {
964     void *fdt, *fdt_skel;
965     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
966 
967     if (spapr_hotplugged_dev_before_cas()) {
968         return 1;
969     }
970 
971     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
972         error_report("SLOF provided an unexpected CAS buffer size "
973                      TARGET_FMT_lu " (min: %zu, max: %u)",
974                      size, sizeof(hdr), FW_MAX_SIZE);
975         exit(EXIT_FAILURE);
976     }
977 
978     size -= sizeof(hdr);
979 
980     /* Create skeleton */
981     fdt_skel = g_malloc0(size);
982     _FDT((fdt_create(fdt_skel, size)));
983     _FDT((fdt_finish_reservemap(fdt_skel)));
984     _FDT((fdt_begin_node(fdt_skel, "")));
985     _FDT((fdt_end_node(fdt_skel)));
986     _FDT((fdt_finish(fdt_skel)));
987     fdt = g_malloc0(size);
988     _FDT((fdt_open_into(fdt_skel, fdt, size)));
989     g_free(fdt_skel);
990 
991     /* Fixup cpu nodes */
992     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
993 
994     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
995         return -1;
996     }
997 
998     /* Pack resulting tree */
999     _FDT((fdt_pack(fdt)));
1000 
1001     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1002         trace_spapr_cas_failed(size);
1003         return -1;
1004     }
1005 
1006     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1007     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1008     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1009     g_free(fdt);
1010 
1011     return 0;
1012 }
1013 
1014 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1015 {
1016     int rtas;
1017     GString *hypertas = g_string_sized_new(256);
1018     GString *qemu_hypertas = g_string_sized_new(256);
1019     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1020     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1021         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1022     uint32_t lrdr_capacity[] = {
1023         cpu_to_be32(max_device_addr >> 32),
1024         cpu_to_be32(max_device_addr & 0xffffffff),
1025         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1026         cpu_to_be32(max_cpus / smp_threads),
1027     };
1028     uint32_t maxdomains[] = {
1029         cpu_to_be32(4),
1030         cpu_to_be32(0),
1031         cpu_to_be32(0),
1032         cpu_to_be32(0),
1033         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes : 1),
1034     };
1035 
1036     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1037 
1038     /* hypertas */
1039     add_str(hypertas, "hcall-pft");
1040     add_str(hypertas, "hcall-term");
1041     add_str(hypertas, "hcall-dabr");
1042     add_str(hypertas, "hcall-interrupt");
1043     add_str(hypertas, "hcall-tce");
1044     add_str(hypertas, "hcall-vio");
1045     add_str(hypertas, "hcall-splpar");
1046     add_str(hypertas, "hcall-bulk");
1047     add_str(hypertas, "hcall-set-mode");
1048     add_str(hypertas, "hcall-sprg0");
1049     add_str(hypertas, "hcall-copy");
1050     add_str(hypertas, "hcall-debug");
1051     add_str(hypertas, "hcall-vphn");
1052     add_str(qemu_hypertas, "hcall-memop1");
1053 
1054     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1055         add_str(hypertas, "hcall-multi-tce");
1056     }
1057 
1058     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1059         add_str(hypertas, "hcall-hpt-resize");
1060     }
1061 
1062     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1063                      hypertas->str, hypertas->len));
1064     g_string_free(hypertas, TRUE);
1065     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1066                      qemu_hypertas->str, qemu_hypertas->len));
1067     g_string_free(qemu_hypertas, TRUE);
1068 
1069     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1070                      refpoints, sizeof(refpoints)));
1071 
1072     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1073                      maxdomains, sizeof(maxdomains)));
1074 
1075     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1076                           RTAS_ERROR_LOG_MAX));
1077     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1078                           RTAS_EVENT_SCAN_RATE));
1079 
1080     g_assert(msi_nonbroken);
1081     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1082 
1083     /*
1084      * According to PAPR, rtas ibm,os-term does not guarantee a return
1085      * back to the guest cpu.
1086      *
1087      * While an additional ibm,extended-os-term property indicates
1088      * that rtas call return will always occur. Set this property.
1089      */
1090     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1091 
1092     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1093                      lrdr_capacity, sizeof(lrdr_capacity)));
1094 
1095     spapr_dt_rtas_tokens(fdt, rtas);
1096 }
1097 
1098 /*
1099  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1100  * and the XIVE features that the guest may request and thus the valid
1101  * values for bytes 23..26 of option vector 5:
1102  */
1103 static void spapr_dt_ov5_platform_support(sPAPRMachineState *spapr, void *fdt,
1104                                           int chosen)
1105 {
1106     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1107 
1108     char val[2 * 4] = {
1109         23, spapr->irq->ov5, /* Xive mode. */
1110         24, 0x00, /* Hash/Radix, filled in below. */
1111         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1112         26, 0x40, /* Radix options: GTSE == yes. */
1113     };
1114 
1115     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1116                           first_ppc_cpu->compat_pvr)) {
1117         /*
1118          * If we're in a pre POWER9 compat mode then the guest should
1119          * do hash and use the legacy interrupt mode
1120          */
1121         val[1] = 0x00; /* XICS */
1122         val[3] = 0x00; /* Hash */
1123     } else if (kvm_enabled()) {
1124         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1125             val[3] = 0x80; /* OV5_MMU_BOTH */
1126         } else if (kvmppc_has_cap_mmu_radix()) {
1127             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1128         } else {
1129             val[3] = 0x00; /* Hash */
1130         }
1131     } else {
1132         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1133         val[3] = 0xC0;
1134     }
1135     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1136                      val, sizeof(val)));
1137 }
1138 
1139 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1140 {
1141     MachineState *machine = MACHINE(spapr);
1142     int chosen;
1143     const char *boot_device = machine->boot_order;
1144     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1145     size_t cb = 0;
1146     char *bootlist = get_boot_devices_list(&cb);
1147 
1148     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1149 
1150     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1151     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1152                           spapr->initrd_base));
1153     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1154                           spapr->initrd_base + spapr->initrd_size));
1155 
1156     if (spapr->kernel_size) {
1157         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1158                               cpu_to_be64(spapr->kernel_size) };
1159 
1160         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1161                          &kprop, sizeof(kprop)));
1162         if (spapr->kernel_le) {
1163             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1164         }
1165     }
1166     if (boot_menu) {
1167         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1168     }
1169     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1170     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1171     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1172 
1173     if (cb && bootlist) {
1174         int i;
1175 
1176         for (i = 0; i < cb; i++) {
1177             if (bootlist[i] == '\n') {
1178                 bootlist[i] = ' ';
1179             }
1180         }
1181         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1182     }
1183 
1184     if (boot_device && strlen(boot_device)) {
1185         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1186     }
1187 
1188     if (!spapr->has_graphics && stdout_path) {
1189         /*
1190          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1191          * kernel. New platforms should only use the "stdout-path" property. Set
1192          * the new property and continue using older property to remain
1193          * compatible with the existing firmware.
1194          */
1195         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1196         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1197     }
1198 
1199     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1200 
1201     g_free(stdout_path);
1202     g_free(bootlist);
1203 }
1204 
1205 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1206 {
1207     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1208      * KVM to work under pHyp with some guest co-operation */
1209     int hypervisor;
1210     uint8_t hypercall[16];
1211 
1212     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1213     /* indicate KVM hypercall interface */
1214     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1215     if (kvmppc_has_cap_fixup_hcalls()) {
1216         /*
1217          * Older KVM versions with older guest kernels were broken
1218          * with the magic page, don't allow the guest to map it.
1219          */
1220         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1221                                   sizeof(hypercall))) {
1222             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1223                              hypercall, sizeof(hypercall)));
1224         }
1225     }
1226 }
1227 
1228 static void *spapr_build_fdt(sPAPRMachineState *spapr)
1229 {
1230     MachineState *machine = MACHINE(spapr);
1231     MachineClass *mc = MACHINE_GET_CLASS(machine);
1232     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1233     int ret;
1234     void *fdt;
1235     sPAPRPHBState *phb;
1236     char *buf;
1237 
1238     fdt = g_malloc0(FDT_MAX_SIZE);
1239     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1240 
1241     /* Root node */
1242     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1243     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1244     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1245 
1246     /*
1247      * Add info to guest to indentify which host is it being run on
1248      * and what is the uuid of the guest
1249      */
1250     if (spapr->host_model && !g_str_equal(spapr->host_model, "none")) {
1251         if (g_str_equal(spapr->host_model, "passthrough")) {
1252             /* -M host-model=passthrough */
1253             if (kvmppc_get_host_model(&buf)) {
1254                 _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1255                 g_free(buf);
1256             }
1257         } else {
1258             /* -M host-model=<user-string> */
1259             _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1260         }
1261     }
1262 
1263     if (spapr->host_serial && !g_str_equal(spapr->host_serial, "none")) {
1264         if (g_str_equal(spapr->host_serial, "passthrough")) {
1265             /* -M host-serial=passthrough */
1266             if (kvmppc_get_host_serial(&buf)) {
1267                 _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1268                 g_free(buf);
1269             }
1270         } else {
1271             /* -M host-serial=<user-string> */
1272             _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1273         }
1274     }
1275 
1276     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1277 
1278     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1279     if (qemu_uuid_set) {
1280         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1281     }
1282     g_free(buf);
1283 
1284     if (qemu_get_vm_name()) {
1285         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1286                                 qemu_get_vm_name()));
1287     }
1288 
1289     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1290     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1291 
1292     /* /interrupt controller */
1293     spapr->irq->dt_populate(spapr, spapr_max_server_number(spapr), fdt,
1294                           PHANDLE_INTC);
1295 
1296     ret = spapr_populate_memory(spapr, fdt);
1297     if (ret < 0) {
1298         error_report("couldn't setup memory nodes in fdt");
1299         exit(1);
1300     }
1301 
1302     /* /vdevice */
1303     spapr_dt_vdevice(spapr->vio_bus, fdt);
1304 
1305     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1306         ret = spapr_rng_populate_dt(fdt);
1307         if (ret < 0) {
1308             error_report("could not set up rng device in the fdt");
1309             exit(1);
1310         }
1311     }
1312 
1313     QLIST_FOREACH(phb, &spapr->phbs, list) {
1314         ret = spapr_populate_pci_dt(phb, PHANDLE_INTC, fdt,
1315                                     spapr->irq->nr_msis, NULL);
1316         if (ret < 0) {
1317             error_report("couldn't setup PCI devices in fdt");
1318             exit(1);
1319         }
1320     }
1321 
1322     /* cpus */
1323     spapr_populate_cpus_dt_node(fdt, spapr);
1324 
1325     if (smc->dr_lmb_enabled) {
1326         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1327     }
1328 
1329     if (mc->has_hotpluggable_cpus) {
1330         int offset = fdt_path_offset(fdt, "/cpus");
1331         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1332                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1333         if (ret < 0) {
1334             error_report("Couldn't set up CPU DR device tree properties");
1335             exit(1);
1336         }
1337     }
1338 
1339     /* /event-sources */
1340     spapr_dt_events(spapr, fdt);
1341 
1342     /* /rtas */
1343     spapr_dt_rtas(spapr, fdt);
1344 
1345     /* /chosen */
1346     spapr_dt_chosen(spapr, fdt);
1347 
1348     /* /hypervisor */
1349     if (kvm_enabled()) {
1350         spapr_dt_hypervisor(spapr, fdt);
1351     }
1352 
1353     /* Build memory reserve map */
1354     if (spapr->kernel_size) {
1355         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1356     }
1357     if (spapr->initrd_size) {
1358         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1359     }
1360 
1361     /* ibm,client-architecture-support updates */
1362     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1363     if (ret < 0) {
1364         error_report("couldn't setup CAS properties fdt");
1365         exit(1);
1366     }
1367 
1368     if (smc->dr_phb_enabled) {
1369         ret = spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_PHB);
1370         if (ret < 0) {
1371             error_report("Couldn't set up PHB DR device tree properties");
1372             exit(1);
1373         }
1374     }
1375 
1376     return fdt;
1377 }
1378 
1379 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1380 {
1381     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1382 }
1383 
1384 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1385                                     PowerPCCPU *cpu)
1386 {
1387     CPUPPCState *env = &cpu->env;
1388 
1389     /* The TCG path should also be holding the BQL at this point */
1390     g_assert(qemu_mutex_iothread_locked());
1391 
1392     if (msr_pr) {
1393         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1394         env->gpr[3] = H_PRIVILEGE;
1395     } else {
1396         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1397     }
1398 }
1399 
1400 struct LPCRSyncState {
1401     target_ulong value;
1402     target_ulong mask;
1403 };
1404 
1405 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1406 {
1407     struct LPCRSyncState *s = arg.host_ptr;
1408     PowerPCCPU *cpu = POWERPC_CPU(cs);
1409     CPUPPCState *env = &cpu->env;
1410     target_ulong lpcr;
1411 
1412     cpu_synchronize_state(cs);
1413     lpcr = env->spr[SPR_LPCR];
1414     lpcr &= ~s->mask;
1415     lpcr |= s->value;
1416     ppc_store_lpcr(cpu, lpcr);
1417 }
1418 
1419 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1420 {
1421     CPUState *cs;
1422     struct LPCRSyncState s = {
1423         .value = value,
1424         .mask = mask
1425     };
1426     CPU_FOREACH(cs) {
1427         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1428     }
1429 }
1430 
1431 static void spapr_get_pate(PPCVirtualHypervisor *vhyp, ppc_v3_pate_t *entry)
1432 {
1433     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1434 
1435     /* Copy PATE1:GR into PATE0:HR */
1436     entry->dw0 = spapr->patb_entry & PATE0_HR;
1437     entry->dw1 = spapr->patb_entry;
1438 }
1439 
1440 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1441 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1442 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1443 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1444 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1445 
1446 /*
1447  * Get the fd to access the kernel htab, re-opening it if necessary
1448  */
1449 static int get_htab_fd(sPAPRMachineState *spapr)
1450 {
1451     Error *local_err = NULL;
1452 
1453     if (spapr->htab_fd >= 0) {
1454         return spapr->htab_fd;
1455     }
1456 
1457     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1458     if (spapr->htab_fd < 0) {
1459         error_report_err(local_err);
1460     }
1461 
1462     return spapr->htab_fd;
1463 }
1464 
1465 void close_htab_fd(sPAPRMachineState *spapr)
1466 {
1467     if (spapr->htab_fd >= 0) {
1468         close(spapr->htab_fd);
1469     }
1470     spapr->htab_fd = -1;
1471 }
1472 
1473 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1474 {
1475     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1476 
1477     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1478 }
1479 
1480 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1481 {
1482     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1483 
1484     assert(kvm_enabled());
1485 
1486     if (!spapr->htab) {
1487         return 0;
1488     }
1489 
1490     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1491 }
1492 
1493 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1494                                                 hwaddr ptex, int n)
1495 {
1496     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1497     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1498 
1499     if (!spapr->htab) {
1500         /*
1501          * HTAB is controlled by KVM. Fetch into temporary buffer
1502          */
1503         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1504         kvmppc_read_hptes(hptes, ptex, n);
1505         return hptes;
1506     }
1507 
1508     /*
1509      * HTAB is controlled by QEMU. Just point to the internally
1510      * accessible PTEG.
1511      */
1512     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1513 }
1514 
1515 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1516                               const ppc_hash_pte64_t *hptes,
1517                               hwaddr ptex, int n)
1518 {
1519     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1520 
1521     if (!spapr->htab) {
1522         g_free((void *)hptes);
1523     }
1524 
1525     /* Nothing to do for qemu managed HPT */
1526 }
1527 
1528 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1529                              uint64_t pte0, uint64_t pte1)
1530 {
1531     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1532     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1533 
1534     if (!spapr->htab) {
1535         kvmppc_write_hpte(ptex, pte0, pte1);
1536     } else {
1537         if (pte0 & HPTE64_V_VALID) {
1538             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1539             /*
1540              * When setting valid, we write PTE1 first. This ensures
1541              * proper synchronization with the reading code in
1542              * ppc_hash64_pteg_search()
1543              */
1544             smp_wmb();
1545             stq_p(spapr->htab + offset, pte0);
1546         } else {
1547             stq_p(spapr->htab + offset, pte0);
1548             /*
1549              * When clearing it we set PTE0 first. This ensures proper
1550              * synchronization with the reading code in
1551              * ppc_hash64_pteg_search()
1552              */
1553             smp_wmb();
1554             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1555         }
1556     }
1557 }
1558 
1559 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1560 {
1561     int shift;
1562 
1563     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1564      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1565      * that's much more than is needed for Linux guests */
1566     shift = ctz64(pow2ceil(ramsize)) - 7;
1567     shift = MAX(shift, 18); /* Minimum architected size */
1568     shift = MIN(shift, 46); /* Maximum architected size */
1569     return shift;
1570 }
1571 
1572 void spapr_free_hpt(sPAPRMachineState *spapr)
1573 {
1574     g_free(spapr->htab);
1575     spapr->htab = NULL;
1576     spapr->htab_shift = 0;
1577     close_htab_fd(spapr);
1578 }
1579 
1580 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1581                           Error **errp)
1582 {
1583     long rc;
1584 
1585     /* Clean up any HPT info from a previous boot */
1586     spapr_free_hpt(spapr);
1587 
1588     rc = kvmppc_reset_htab(shift);
1589     if (rc < 0) {
1590         /* kernel-side HPT needed, but couldn't allocate one */
1591         error_setg_errno(errp, errno,
1592                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1593                          shift);
1594         /* This is almost certainly fatal, but if the caller really
1595          * wants to carry on with shift == 0, it's welcome to try */
1596     } else if (rc > 0) {
1597         /* kernel-side HPT allocated */
1598         if (rc != shift) {
1599             error_setg(errp,
1600                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1601                        shift, rc);
1602         }
1603 
1604         spapr->htab_shift = shift;
1605         spapr->htab = NULL;
1606     } else {
1607         /* kernel-side HPT not needed, allocate in userspace instead */
1608         size_t size = 1ULL << shift;
1609         int i;
1610 
1611         spapr->htab = qemu_memalign(size, size);
1612         if (!spapr->htab) {
1613             error_setg_errno(errp, errno,
1614                              "Could not allocate HPT of order %d", shift);
1615             return;
1616         }
1617 
1618         memset(spapr->htab, 0, size);
1619         spapr->htab_shift = shift;
1620 
1621         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1622             DIRTY_HPTE(HPTE(spapr->htab, i));
1623         }
1624     }
1625     /* We're setting up a hash table, so that means we're not radix */
1626     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1627 }
1628 
1629 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1630 {
1631     int hpt_shift;
1632 
1633     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1634         || (spapr->cas_reboot
1635             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1636         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1637     } else {
1638         uint64_t current_ram_size;
1639 
1640         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1641         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1642     }
1643     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1644 
1645     if (spapr->vrma_adjust) {
1646         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1647                                           spapr->htab_shift);
1648     }
1649 }
1650 
1651 static int spapr_reset_drcs(Object *child, void *opaque)
1652 {
1653     sPAPRDRConnector *drc =
1654         (sPAPRDRConnector *) object_dynamic_cast(child,
1655                                                  TYPE_SPAPR_DR_CONNECTOR);
1656 
1657     if (drc) {
1658         spapr_drc_reset(drc);
1659     }
1660 
1661     return 0;
1662 }
1663 
1664 static void spapr_machine_reset(void)
1665 {
1666     MachineState *machine = MACHINE(qdev_get_machine());
1667     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1668     PowerPCCPU *first_ppc_cpu;
1669     uint32_t rtas_limit;
1670     hwaddr rtas_addr, fdt_addr;
1671     void *fdt;
1672     int rc;
1673 
1674     spapr_caps_apply(spapr);
1675 
1676     first_ppc_cpu = POWERPC_CPU(first_cpu);
1677     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1678         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1679                               spapr->max_compat_pvr)) {
1680         /*
1681          * If using KVM with radix mode available, VCPUs can be started
1682          * without a HPT because KVM will start them in radix mode.
1683          * Set the GR bit in PATE so that we know there is no HPT.
1684          */
1685         spapr->patb_entry = PATE1_GR;
1686         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1687     } else {
1688         spapr_setup_hpt_and_vrma(spapr);
1689     }
1690 
1691     /*
1692      * If this reset wasn't generated by CAS, we should reset our
1693      * negotiated options and start from scratch
1694      */
1695     if (!spapr->cas_reboot) {
1696         spapr_ovec_cleanup(spapr->ov5_cas);
1697         spapr->ov5_cas = spapr_ovec_new();
1698 
1699         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1700     }
1701 
1702     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1703         spapr_irq_msi_reset(spapr);
1704     }
1705 
1706     qemu_devices_reset();
1707 
1708     /*
1709      * This is fixing some of the default configuration of the XIVE
1710      * devices. To be called after the reset of the machine devices.
1711      */
1712     spapr_irq_reset(spapr, &error_fatal);
1713 
1714     /* DRC reset may cause a device to be unplugged. This will cause troubles
1715      * if this device is used by another device (eg, a running vhost backend
1716      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1717      * situations, we reset DRCs after all devices have been reset.
1718      */
1719     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1720 
1721     spapr_clear_pending_events(spapr);
1722 
1723     /*
1724      * We place the device tree and RTAS just below either the top of the RMA,
1725      * or just below 2GB, whichever is lower, so that it can be
1726      * processed with 32-bit real mode code if necessary
1727      */
1728     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1729     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1730     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1731 
1732     fdt = spapr_build_fdt(spapr);
1733 
1734     spapr_load_rtas(spapr, fdt, rtas_addr);
1735 
1736     rc = fdt_pack(fdt);
1737 
1738     /* Should only fail if we've built a corrupted tree */
1739     assert(rc == 0);
1740 
1741     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1742         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1743                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1744         exit(1);
1745     }
1746 
1747     /* Load the fdt */
1748     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1749     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1750     g_free(spapr->fdt_blob);
1751     spapr->fdt_size = fdt_totalsize(fdt);
1752     spapr->fdt_initial_size = spapr->fdt_size;
1753     spapr->fdt_blob = fdt;
1754 
1755     /* Set up the entry state */
1756     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1757     first_ppc_cpu->env.gpr[5] = 0;
1758 
1759     spapr->cas_reboot = false;
1760 }
1761 
1762 static void spapr_create_nvram(sPAPRMachineState *spapr)
1763 {
1764     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1765     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1766 
1767     if (dinfo) {
1768         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1769                             &error_fatal);
1770     }
1771 
1772     qdev_init_nofail(dev);
1773 
1774     spapr->nvram = (struct sPAPRNVRAM *)dev;
1775 }
1776 
1777 static void spapr_rtc_create(sPAPRMachineState *spapr)
1778 {
1779     object_initialize_child(OBJECT(spapr), "rtc",
1780                             &spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC,
1781                             &error_fatal, NULL);
1782     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1783                               &error_fatal);
1784     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1785                               "date", &error_fatal);
1786 }
1787 
1788 /* Returns whether we want to use VGA or not */
1789 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1790 {
1791     switch (vga_interface_type) {
1792     case VGA_NONE:
1793         return false;
1794     case VGA_DEVICE:
1795         return true;
1796     case VGA_STD:
1797     case VGA_VIRTIO:
1798     case VGA_CIRRUS:
1799         return pci_vga_init(pci_bus) != NULL;
1800     default:
1801         error_setg(errp,
1802                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1803         return false;
1804     }
1805 }
1806 
1807 static int spapr_pre_load(void *opaque)
1808 {
1809     int rc;
1810 
1811     rc = spapr_caps_pre_load(opaque);
1812     if (rc) {
1813         return rc;
1814     }
1815 
1816     return 0;
1817 }
1818 
1819 static int spapr_post_load(void *opaque, int version_id)
1820 {
1821     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1822     int err = 0;
1823 
1824     err = spapr_caps_post_migration(spapr);
1825     if (err) {
1826         return err;
1827     }
1828 
1829     /*
1830      * In earlier versions, there was no separate qdev for the PAPR
1831      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1832      * So when migrating from those versions, poke the incoming offset
1833      * value into the RTC device
1834      */
1835     if (version_id < 3) {
1836         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1837         if (err) {
1838             return err;
1839         }
1840     }
1841 
1842     if (kvm_enabled() && spapr->patb_entry) {
1843         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1844         bool radix = !!(spapr->patb_entry & PATE1_GR);
1845         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1846 
1847         /*
1848          * Update LPCR:HR and UPRT as they may not be set properly in
1849          * the stream
1850          */
1851         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1852                             LPCR_HR | LPCR_UPRT);
1853 
1854         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1855         if (err) {
1856             error_report("Process table config unsupported by the host");
1857             return -EINVAL;
1858         }
1859     }
1860 
1861     err = spapr_irq_post_load(spapr, version_id);
1862     if (err) {
1863         return err;
1864     }
1865 
1866     return err;
1867 }
1868 
1869 static int spapr_pre_save(void *opaque)
1870 {
1871     int rc;
1872 
1873     rc = spapr_caps_pre_save(opaque);
1874     if (rc) {
1875         return rc;
1876     }
1877 
1878     return 0;
1879 }
1880 
1881 static bool version_before_3(void *opaque, int version_id)
1882 {
1883     return version_id < 3;
1884 }
1885 
1886 static bool spapr_pending_events_needed(void *opaque)
1887 {
1888     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1889     return !QTAILQ_EMPTY(&spapr->pending_events);
1890 }
1891 
1892 static const VMStateDescription vmstate_spapr_event_entry = {
1893     .name = "spapr_event_log_entry",
1894     .version_id = 1,
1895     .minimum_version_id = 1,
1896     .fields = (VMStateField[]) {
1897         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1898         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1899         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1900                                      NULL, extended_length),
1901         VMSTATE_END_OF_LIST()
1902     },
1903 };
1904 
1905 static const VMStateDescription vmstate_spapr_pending_events = {
1906     .name = "spapr_pending_events",
1907     .version_id = 1,
1908     .minimum_version_id = 1,
1909     .needed = spapr_pending_events_needed,
1910     .fields = (VMStateField[]) {
1911         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1912                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1913         VMSTATE_END_OF_LIST()
1914     },
1915 };
1916 
1917 static bool spapr_ov5_cas_needed(void *opaque)
1918 {
1919     sPAPRMachineState *spapr = opaque;
1920     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1921     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1922     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1923     bool cas_needed;
1924 
1925     /* Prior to the introduction of sPAPROptionVector, we had two option
1926      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1927      * Both of these options encode machine topology into the device-tree
1928      * in such a way that the now-booted OS should still be able to interact
1929      * appropriately with QEMU regardless of what options were actually
1930      * negotiatied on the source side.
1931      *
1932      * As such, we can avoid migrating the CAS-negotiated options if these
1933      * are the only options available on the current machine/platform.
1934      * Since these are the only options available for pseries-2.7 and
1935      * earlier, this allows us to maintain old->new/new->old migration
1936      * compatibility.
1937      *
1938      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1939      * via default pseries-2.8 machines and explicit command-line parameters.
1940      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1941      * of the actual CAS-negotiated values to continue working properly. For
1942      * example, availability of memory unplug depends on knowing whether
1943      * OV5_HP_EVT was negotiated via CAS.
1944      *
1945      * Thus, for any cases where the set of available CAS-negotiatable
1946      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1947      * include the CAS-negotiated options in the migration stream, unless
1948      * if they affect boot time behaviour only.
1949      */
1950     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1951     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1952     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1953 
1954     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1955      * the mask itself since in the future it's possible "legacy" bits may be
1956      * removed via machine options, which could generate a false positive
1957      * that breaks migration.
1958      */
1959     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1960     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1961 
1962     spapr_ovec_cleanup(ov5_mask);
1963     spapr_ovec_cleanup(ov5_legacy);
1964     spapr_ovec_cleanup(ov5_removed);
1965 
1966     return cas_needed;
1967 }
1968 
1969 static const VMStateDescription vmstate_spapr_ov5_cas = {
1970     .name = "spapr_option_vector_ov5_cas",
1971     .version_id = 1,
1972     .minimum_version_id = 1,
1973     .needed = spapr_ov5_cas_needed,
1974     .fields = (VMStateField[]) {
1975         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1976                                  vmstate_spapr_ovec, sPAPROptionVector),
1977         VMSTATE_END_OF_LIST()
1978     },
1979 };
1980 
1981 static bool spapr_patb_entry_needed(void *opaque)
1982 {
1983     sPAPRMachineState *spapr = opaque;
1984 
1985     return !!spapr->patb_entry;
1986 }
1987 
1988 static const VMStateDescription vmstate_spapr_patb_entry = {
1989     .name = "spapr_patb_entry",
1990     .version_id = 1,
1991     .minimum_version_id = 1,
1992     .needed = spapr_patb_entry_needed,
1993     .fields = (VMStateField[]) {
1994         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1995         VMSTATE_END_OF_LIST()
1996     },
1997 };
1998 
1999 static bool spapr_irq_map_needed(void *opaque)
2000 {
2001     sPAPRMachineState *spapr = opaque;
2002 
2003     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
2004 }
2005 
2006 static const VMStateDescription vmstate_spapr_irq_map = {
2007     .name = "spapr_irq_map",
2008     .version_id = 1,
2009     .minimum_version_id = 1,
2010     .needed = spapr_irq_map_needed,
2011     .fields = (VMStateField[]) {
2012         VMSTATE_BITMAP(irq_map, sPAPRMachineState, 0, irq_map_nr),
2013         VMSTATE_END_OF_LIST()
2014     },
2015 };
2016 
2017 static bool spapr_dtb_needed(void *opaque)
2018 {
2019     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2020 
2021     return smc->update_dt_enabled;
2022 }
2023 
2024 static int spapr_dtb_pre_load(void *opaque)
2025 {
2026     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
2027 
2028     g_free(spapr->fdt_blob);
2029     spapr->fdt_blob = NULL;
2030     spapr->fdt_size = 0;
2031 
2032     return 0;
2033 }
2034 
2035 static const VMStateDescription vmstate_spapr_dtb = {
2036     .name = "spapr_dtb",
2037     .version_id = 1,
2038     .minimum_version_id = 1,
2039     .needed = spapr_dtb_needed,
2040     .pre_load = spapr_dtb_pre_load,
2041     .fields = (VMStateField[]) {
2042         VMSTATE_UINT32(fdt_initial_size, sPAPRMachineState),
2043         VMSTATE_UINT32(fdt_size, sPAPRMachineState),
2044         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, sPAPRMachineState, 0, NULL,
2045                                      fdt_size),
2046         VMSTATE_END_OF_LIST()
2047     },
2048 };
2049 
2050 static const VMStateDescription vmstate_spapr = {
2051     .name = "spapr",
2052     .version_id = 3,
2053     .minimum_version_id = 1,
2054     .pre_load = spapr_pre_load,
2055     .post_load = spapr_post_load,
2056     .pre_save = spapr_pre_save,
2057     .fields = (VMStateField[]) {
2058         /* used to be @next_irq */
2059         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2060 
2061         /* RTC offset */
2062         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
2063 
2064         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
2065         VMSTATE_END_OF_LIST()
2066     },
2067     .subsections = (const VMStateDescription*[]) {
2068         &vmstate_spapr_ov5_cas,
2069         &vmstate_spapr_patb_entry,
2070         &vmstate_spapr_pending_events,
2071         &vmstate_spapr_cap_htm,
2072         &vmstate_spapr_cap_vsx,
2073         &vmstate_spapr_cap_dfp,
2074         &vmstate_spapr_cap_cfpc,
2075         &vmstate_spapr_cap_sbbc,
2076         &vmstate_spapr_cap_ibs,
2077         &vmstate_spapr_irq_map,
2078         &vmstate_spapr_cap_nested_kvm_hv,
2079         &vmstate_spapr_dtb,
2080         NULL
2081     }
2082 };
2083 
2084 static int htab_save_setup(QEMUFile *f, void *opaque)
2085 {
2086     sPAPRMachineState *spapr = opaque;
2087 
2088     /* "Iteration" header */
2089     if (!spapr->htab_shift) {
2090         qemu_put_be32(f, -1);
2091     } else {
2092         qemu_put_be32(f, spapr->htab_shift);
2093     }
2094 
2095     if (spapr->htab) {
2096         spapr->htab_save_index = 0;
2097         spapr->htab_first_pass = true;
2098     } else {
2099         if (spapr->htab_shift) {
2100             assert(kvm_enabled());
2101         }
2102     }
2103 
2104 
2105     return 0;
2106 }
2107 
2108 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
2109                             int chunkstart, int n_valid, int n_invalid)
2110 {
2111     qemu_put_be32(f, chunkstart);
2112     qemu_put_be16(f, n_valid);
2113     qemu_put_be16(f, n_invalid);
2114     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2115                     HASH_PTE_SIZE_64 * n_valid);
2116 }
2117 
2118 static void htab_save_end_marker(QEMUFile *f)
2119 {
2120     qemu_put_be32(f, 0);
2121     qemu_put_be16(f, 0);
2122     qemu_put_be16(f, 0);
2123 }
2124 
2125 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
2126                                  int64_t max_ns)
2127 {
2128     bool has_timeout = max_ns != -1;
2129     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2130     int index = spapr->htab_save_index;
2131     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2132 
2133     assert(spapr->htab_first_pass);
2134 
2135     do {
2136         int chunkstart;
2137 
2138         /* Consume invalid HPTEs */
2139         while ((index < htabslots)
2140                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2141             CLEAN_HPTE(HPTE(spapr->htab, index));
2142             index++;
2143         }
2144 
2145         /* Consume valid HPTEs */
2146         chunkstart = index;
2147         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2148                && HPTE_VALID(HPTE(spapr->htab, index))) {
2149             CLEAN_HPTE(HPTE(spapr->htab, index));
2150             index++;
2151         }
2152 
2153         if (index > chunkstart) {
2154             int n_valid = index - chunkstart;
2155 
2156             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2157 
2158             if (has_timeout &&
2159                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2160                 break;
2161             }
2162         }
2163     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2164 
2165     if (index >= htabslots) {
2166         assert(index == htabslots);
2167         index = 0;
2168         spapr->htab_first_pass = false;
2169     }
2170     spapr->htab_save_index = index;
2171 }
2172 
2173 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2174                                 int64_t max_ns)
2175 {
2176     bool final = max_ns < 0;
2177     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2178     int examined = 0, sent = 0;
2179     int index = spapr->htab_save_index;
2180     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2181 
2182     assert(!spapr->htab_first_pass);
2183 
2184     do {
2185         int chunkstart, invalidstart;
2186 
2187         /* Consume non-dirty HPTEs */
2188         while ((index < htabslots)
2189                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2190             index++;
2191             examined++;
2192         }
2193 
2194         chunkstart = index;
2195         /* Consume valid dirty HPTEs */
2196         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2197                && HPTE_DIRTY(HPTE(spapr->htab, index))
2198                && HPTE_VALID(HPTE(spapr->htab, index))) {
2199             CLEAN_HPTE(HPTE(spapr->htab, index));
2200             index++;
2201             examined++;
2202         }
2203 
2204         invalidstart = index;
2205         /* Consume invalid dirty HPTEs */
2206         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2207                && HPTE_DIRTY(HPTE(spapr->htab, index))
2208                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2209             CLEAN_HPTE(HPTE(spapr->htab, index));
2210             index++;
2211             examined++;
2212         }
2213 
2214         if (index > chunkstart) {
2215             int n_valid = invalidstart - chunkstart;
2216             int n_invalid = index - invalidstart;
2217 
2218             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2219             sent += index - chunkstart;
2220 
2221             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2222                 break;
2223             }
2224         }
2225 
2226         if (examined >= htabslots) {
2227             break;
2228         }
2229 
2230         if (index >= htabslots) {
2231             assert(index == htabslots);
2232             index = 0;
2233         }
2234     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2235 
2236     if (index >= htabslots) {
2237         assert(index == htabslots);
2238         index = 0;
2239     }
2240 
2241     spapr->htab_save_index = index;
2242 
2243     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2244 }
2245 
2246 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2247 #define MAX_KVM_BUF_SIZE    2048
2248 
2249 static int htab_save_iterate(QEMUFile *f, void *opaque)
2250 {
2251     sPAPRMachineState *spapr = opaque;
2252     int fd;
2253     int rc = 0;
2254 
2255     /* Iteration header */
2256     if (!spapr->htab_shift) {
2257         qemu_put_be32(f, -1);
2258         return 1;
2259     } else {
2260         qemu_put_be32(f, 0);
2261     }
2262 
2263     if (!spapr->htab) {
2264         assert(kvm_enabled());
2265 
2266         fd = get_htab_fd(spapr);
2267         if (fd < 0) {
2268             return fd;
2269         }
2270 
2271         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2272         if (rc < 0) {
2273             return rc;
2274         }
2275     } else  if (spapr->htab_first_pass) {
2276         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2277     } else {
2278         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2279     }
2280 
2281     htab_save_end_marker(f);
2282 
2283     return rc;
2284 }
2285 
2286 static int htab_save_complete(QEMUFile *f, void *opaque)
2287 {
2288     sPAPRMachineState *spapr = opaque;
2289     int fd;
2290 
2291     /* Iteration header */
2292     if (!spapr->htab_shift) {
2293         qemu_put_be32(f, -1);
2294         return 0;
2295     } else {
2296         qemu_put_be32(f, 0);
2297     }
2298 
2299     if (!spapr->htab) {
2300         int rc;
2301 
2302         assert(kvm_enabled());
2303 
2304         fd = get_htab_fd(spapr);
2305         if (fd < 0) {
2306             return fd;
2307         }
2308 
2309         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2310         if (rc < 0) {
2311             return rc;
2312         }
2313     } else {
2314         if (spapr->htab_first_pass) {
2315             htab_save_first_pass(f, spapr, -1);
2316         }
2317         htab_save_later_pass(f, spapr, -1);
2318     }
2319 
2320     /* End marker */
2321     htab_save_end_marker(f);
2322 
2323     return 0;
2324 }
2325 
2326 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2327 {
2328     sPAPRMachineState *spapr = opaque;
2329     uint32_t section_hdr;
2330     int fd = -1;
2331     Error *local_err = NULL;
2332 
2333     if (version_id < 1 || version_id > 1) {
2334         error_report("htab_load() bad version");
2335         return -EINVAL;
2336     }
2337 
2338     section_hdr = qemu_get_be32(f);
2339 
2340     if (section_hdr == -1) {
2341         spapr_free_hpt(spapr);
2342         return 0;
2343     }
2344 
2345     if (section_hdr) {
2346         /* First section gives the htab size */
2347         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2348         if (local_err) {
2349             error_report_err(local_err);
2350             return -EINVAL;
2351         }
2352         return 0;
2353     }
2354 
2355     if (!spapr->htab) {
2356         assert(kvm_enabled());
2357 
2358         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2359         if (fd < 0) {
2360             error_report_err(local_err);
2361             return fd;
2362         }
2363     }
2364 
2365     while (true) {
2366         uint32_t index;
2367         uint16_t n_valid, n_invalid;
2368 
2369         index = qemu_get_be32(f);
2370         n_valid = qemu_get_be16(f);
2371         n_invalid = qemu_get_be16(f);
2372 
2373         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2374             /* End of Stream */
2375             break;
2376         }
2377 
2378         if ((index + n_valid + n_invalid) >
2379             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2380             /* Bad index in stream */
2381             error_report(
2382                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2383                 index, n_valid, n_invalid, spapr->htab_shift);
2384             return -EINVAL;
2385         }
2386 
2387         if (spapr->htab) {
2388             if (n_valid) {
2389                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2390                                 HASH_PTE_SIZE_64 * n_valid);
2391             }
2392             if (n_invalid) {
2393                 memset(HPTE(spapr->htab, index + n_valid), 0,
2394                        HASH_PTE_SIZE_64 * n_invalid);
2395             }
2396         } else {
2397             int rc;
2398 
2399             assert(fd >= 0);
2400 
2401             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2402             if (rc < 0) {
2403                 return rc;
2404             }
2405         }
2406     }
2407 
2408     if (!spapr->htab) {
2409         assert(fd >= 0);
2410         close(fd);
2411     }
2412 
2413     return 0;
2414 }
2415 
2416 static void htab_save_cleanup(void *opaque)
2417 {
2418     sPAPRMachineState *spapr = opaque;
2419 
2420     close_htab_fd(spapr);
2421 }
2422 
2423 static SaveVMHandlers savevm_htab_handlers = {
2424     .save_setup = htab_save_setup,
2425     .save_live_iterate = htab_save_iterate,
2426     .save_live_complete_precopy = htab_save_complete,
2427     .save_cleanup = htab_save_cleanup,
2428     .load_state = htab_load,
2429 };
2430 
2431 static void spapr_boot_set(void *opaque, const char *boot_device,
2432                            Error **errp)
2433 {
2434     MachineState *machine = MACHINE(opaque);
2435     machine->boot_order = g_strdup(boot_device);
2436 }
2437 
2438 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2439 {
2440     MachineState *machine = MACHINE(spapr);
2441     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2442     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2443     int i;
2444 
2445     for (i = 0; i < nr_lmbs; i++) {
2446         uint64_t addr;
2447 
2448         addr = i * lmb_size + machine->device_memory->base;
2449         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2450                                addr / lmb_size);
2451     }
2452 }
2453 
2454 /*
2455  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2456  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2457  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2458  */
2459 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2460 {
2461     int i;
2462 
2463     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2464         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2465                    " is not aligned to %" PRIu64 " MiB",
2466                    machine->ram_size,
2467                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2468         return;
2469     }
2470 
2471     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2472         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2473                    " is not aligned to %" PRIu64 " MiB",
2474                    machine->ram_size,
2475                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2476         return;
2477     }
2478 
2479     for (i = 0; i < nb_numa_nodes; i++) {
2480         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2481             error_setg(errp,
2482                        "Node %d memory size 0x%" PRIx64
2483                        " is not aligned to %" PRIu64 " MiB",
2484                        i, numa_info[i].node_mem,
2485                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2486             return;
2487         }
2488     }
2489 }
2490 
2491 /* find cpu slot in machine->possible_cpus by core_id */
2492 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2493 {
2494     int index = id / smp_threads;
2495 
2496     if (index >= ms->possible_cpus->len) {
2497         return NULL;
2498     }
2499     if (idx) {
2500         *idx = index;
2501     }
2502     return &ms->possible_cpus->cpus[index];
2503 }
2504 
2505 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2506 {
2507     Error *local_err = NULL;
2508     bool vsmt_user = !!spapr->vsmt;
2509     int kvm_smt = kvmppc_smt_threads();
2510     int ret;
2511 
2512     if (!kvm_enabled() && (smp_threads > 1)) {
2513         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2514                      "on a pseries machine");
2515         goto out;
2516     }
2517     if (!is_power_of_2(smp_threads)) {
2518         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2519                      "machine because it must be a power of 2", smp_threads);
2520         goto out;
2521     }
2522 
2523     /* Detemine the VSMT mode to use: */
2524     if (vsmt_user) {
2525         if (spapr->vsmt < smp_threads) {
2526             error_setg(&local_err, "Cannot support VSMT mode %d"
2527                          " because it must be >= threads/core (%d)",
2528                          spapr->vsmt, smp_threads);
2529             goto out;
2530         }
2531         /* In this case, spapr->vsmt has been set by the command line */
2532     } else {
2533         /*
2534          * Default VSMT value is tricky, because we need it to be as
2535          * consistent as possible (for migration), but this requires
2536          * changing it for at least some existing cases.  We pick 8 as
2537          * the value that we'd get with KVM on POWER8, the
2538          * overwhelmingly common case in production systems.
2539          */
2540         spapr->vsmt = MAX(8, smp_threads);
2541     }
2542 
2543     /* KVM: If necessary, set the SMT mode: */
2544     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2545         ret = kvmppc_set_smt_threads(spapr->vsmt);
2546         if (ret) {
2547             /* Looks like KVM isn't able to change VSMT mode */
2548             error_setg(&local_err,
2549                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2550                        spapr->vsmt, ret);
2551             /* We can live with that if the default one is big enough
2552              * for the number of threads, and a submultiple of the one
2553              * we want.  In this case we'll waste some vcpu ids, but
2554              * behaviour will be correct */
2555             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2556                 warn_report_err(local_err);
2557                 local_err = NULL;
2558                 goto out;
2559             } else {
2560                 if (!vsmt_user) {
2561                     error_append_hint(&local_err,
2562                                       "On PPC, a VM with %d threads/core"
2563                                       " on a host with %d threads/core"
2564                                       " requires the use of VSMT mode %d.\n",
2565                                       smp_threads, kvm_smt, spapr->vsmt);
2566                 }
2567                 kvmppc_hint_smt_possible(&local_err);
2568                 goto out;
2569             }
2570         }
2571     }
2572     /* else TCG: nothing to do currently */
2573 out:
2574     error_propagate(errp, local_err);
2575 }
2576 
2577 static void spapr_init_cpus(sPAPRMachineState *spapr)
2578 {
2579     MachineState *machine = MACHINE(spapr);
2580     MachineClass *mc = MACHINE_GET_CLASS(machine);
2581     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2582     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2583     const CPUArchIdList *possible_cpus;
2584     int boot_cores_nr = smp_cpus / smp_threads;
2585     int i;
2586 
2587     possible_cpus = mc->possible_cpu_arch_ids(machine);
2588     if (mc->has_hotpluggable_cpus) {
2589         if (smp_cpus % smp_threads) {
2590             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2591                          smp_cpus, smp_threads);
2592             exit(1);
2593         }
2594         if (max_cpus % smp_threads) {
2595             error_report("max_cpus (%u) must be multiple of threads (%u)",
2596                          max_cpus, smp_threads);
2597             exit(1);
2598         }
2599     } else {
2600         if (max_cpus != smp_cpus) {
2601             error_report("This machine version does not support CPU hotplug");
2602             exit(1);
2603         }
2604         boot_cores_nr = possible_cpus->len;
2605     }
2606 
2607     if (smc->pre_2_10_has_unused_icps) {
2608         int i;
2609 
2610         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2611             /* Dummy entries get deregistered when real ICPState objects
2612              * are registered during CPU core hotplug.
2613              */
2614             pre_2_10_vmstate_register_dummy_icp(i);
2615         }
2616     }
2617 
2618     for (i = 0; i < possible_cpus->len; i++) {
2619         int core_id = i * smp_threads;
2620 
2621         if (mc->has_hotpluggable_cpus) {
2622             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2623                                    spapr_vcpu_id(spapr, core_id));
2624         }
2625 
2626         if (i < boot_cores_nr) {
2627             Object *core  = object_new(type);
2628             int nr_threads = smp_threads;
2629 
2630             /* Handle the partially filled core for older machine types */
2631             if ((i + 1) * smp_threads >= smp_cpus) {
2632                 nr_threads = smp_cpus - i * smp_threads;
2633             }
2634 
2635             object_property_set_int(core, nr_threads, "nr-threads",
2636                                     &error_fatal);
2637             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2638                                     &error_fatal);
2639             object_property_set_bool(core, true, "realized", &error_fatal);
2640 
2641             object_unref(core);
2642         }
2643     }
2644 }
2645 
2646 static PCIHostState *spapr_create_default_phb(void)
2647 {
2648     DeviceState *dev;
2649 
2650     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2651     qdev_prop_set_uint32(dev, "index", 0);
2652     qdev_init_nofail(dev);
2653 
2654     return PCI_HOST_BRIDGE(dev);
2655 }
2656 
2657 /* pSeries LPAR / sPAPR hardware init */
2658 static void spapr_machine_init(MachineState *machine)
2659 {
2660     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2661     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2662     const char *kernel_filename = machine->kernel_filename;
2663     const char *initrd_filename = machine->initrd_filename;
2664     PCIHostState *phb;
2665     int i;
2666     MemoryRegion *sysmem = get_system_memory();
2667     MemoryRegion *ram = g_new(MemoryRegion, 1);
2668     hwaddr node0_size = spapr_node0_size(machine);
2669     long load_limit, fw_size;
2670     char *filename;
2671     Error *resize_hpt_err = NULL;
2672 
2673     msi_nonbroken = true;
2674 
2675     QLIST_INIT(&spapr->phbs);
2676     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2677 
2678     /* Determine capabilities to run with */
2679     spapr_caps_init(spapr);
2680 
2681     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2682     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2683         /*
2684          * If the user explicitly requested a mode we should either
2685          * supply it, or fail completely (which we do below).  But if
2686          * it's not set explicitly, we reset our mode to something
2687          * that works
2688          */
2689         if (resize_hpt_err) {
2690             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2691             error_free(resize_hpt_err);
2692             resize_hpt_err = NULL;
2693         } else {
2694             spapr->resize_hpt = smc->resize_hpt_default;
2695         }
2696     }
2697 
2698     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2699 
2700     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2701         /*
2702          * User requested HPT resize, but this host can't supply it.  Bail out
2703          */
2704         error_report_err(resize_hpt_err);
2705         exit(1);
2706     }
2707 
2708     spapr->rma_size = node0_size;
2709 
2710     /* With KVM, we don't actually know whether KVM supports an
2711      * unbounded RMA (PR KVM) or is limited by the hash table size
2712      * (HV KVM using VRMA), so we always assume the latter
2713      *
2714      * In that case, we also limit the initial allocations for RTAS
2715      * etc... to 256M since we have no way to know what the VRMA size
2716      * is going to be as it depends on the size of the hash table
2717      * which isn't determined yet.
2718      */
2719     if (kvm_enabled()) {
2720         spapr->vrma_adjust = 1;
2721         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2722     }
2723 
2724     /* Actually we don't support unbounded RMA anymore since we added
2725      * proper emulation of HV mode. The max we can get is 16G which
2726      * also happens to be what we configure for PAPR mode so make sure
2727      * we don't do anything bigger than that
2728      */
2729     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2730 
2731     if (spapr->rma_size > node0_size) {
2732         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2733                      spapr->rma_size);
2734         exit(1);
2735     }
2736 
2737     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2738     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2739 
2740     /*
2741      * VSMT must be set in order to be able to compute VCPU ids, ie to
2742      * call spapr_max_server_number() or spapr_vcpu_id().
2743      */
2744     spapr_set_vsmt_mode(spapr, &error_fatal);
2745 
2746     /* Set up Interrupt Controller before we create the VCPUs */
2747     spapr_irq_init(spapr, &error_fatal);
2748 
2749     /* Set up containers for ibm,client-architecture-support negotiated options
2750      */
2751     spapr->ov5 = spapr_ovec_new();
2752     spapr->ov5_cas = spapr_ovec_new();
2753 
2754     if (smc->dr_lmb_enabled) {
2755         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2756         spapr_validate_node_memory(machine, &error_fatal);
2757     }
2758 
2759     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2760 
2761     /* advertise support for dedicated HP event source to guests */
2762     if (spapr->use_hotplug_event_source) {
2763         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2764     }
2765 
2766     /* advertise support for HPT resizing */
2767     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2768         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2769     }
2770 
2771     /* advertise support for ibm,dyamic-memory-v2 */
2772     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2773 
2774     /* advertise XIVE on POWER9 machines */
2775     if (spapr->irq->ov5 & (SPAPR_OV5_XIVE_EXPLOIT | SPAPR_OV5_XIVE_BOTH)) {
2776         if (ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00,
2777                                   0, spapr->max_compat_pvr)) {
2778             spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2779         } else if (spapr->irq->ov5 & SPAPR_OV5_XIVE_EXPLOIT) {
2780             error_report("XIVE-only machines require a POWER9 CPU");
2781             exit(1);
2782         }
2783     }
2784 
2785     /* init CPUs */
2786     spapr_init_cpus(spapr);
2787 
2788     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2789         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2790                               spapr->max_compat_pvr)) {
2791         /* KVM and TCG always allow GTSE with radix... */
2792         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2793     }
2794     /* ... but not with hash (currently). */
2795 
2796     if (kvm_enabled()) {
2797         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2798         kvmppc_enable_logical_ci_hcalls();
2799         kvmppc_enable_set_mode_hcall();
2800 
2801         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2802         kvmppc_enable_clear_ref_mod_hcalls();
2803     }
2804 
2805     /* allocate RAM */
2806     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2807                                          machine->ram_size);
2808     memory_region_add_subregion(sysmem, 0, ram);
2809 
2810     /* always allocate the device memory information */
2811     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2812 
2813     /* initialize hotplug memory address space */
2814     if (machine->ram_size < machine->maxram_size) {
2815         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2816         /*
2817          * Limit the number of hotpluggable memory slots to half the number
2818          * slots that KVM supports, leaving the other half for PCI and other
2819          * devices. However ensure that number of slots doesn't drop below 32.
2820          */
2821         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2822                            SPAPR_MAX_RAM_SLOTS;
2823 
2824         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2825             max_memslots = SPAPR_MAX_RAM_SLOTS;
2826         }
2827         if (machine->ram_slots > max_memslots) {
2828             error_report("Specified number of memory slots %"
2829                          PRIu64" exceeds max supported %d",
2830                          machine->ram_slots, max_memslots);
2831             exit(1);
2832         }
2833 
2834         machine->device_memory->base = ROUND_UP(machine->ram_size,
2835                                                 SPAPR_DEVICE_MEM_ALIGN);
2836         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2837                            "device-memory", device_mem_size);
2838         memory_region_add_subregion(sysmem, machine->device_memory->base,
2839                                     &machine->device_memory->mr);
2840     }
2841 
2842     if (smc->dr_lmb_enabled) {
2843         spapr_create_lmb_dr_connectors(spapr);
2844     }
2845 
2846     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2847     if (!filename) {
2848         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2849         exit(1);
2850     }
2851     spapr->rtas_size = get_image_size(filename);
2852     if (spapr->rtas_size < 0) {
2853         error_report("Could not get size of LPAR rtas '%s'", filename);
2854         exit(1);
2855     }
2856     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2857     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2858         error_report("Could not load LPAR rtas '%s'", filename);
2859         exit(1);
2860     }
2861     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2862         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2863                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2864         exit(1);
2865     }
2866     g_free(filename);
2867 
2868     /* Set up RTAS event infrastructure */
2869     spapr_events_init(spapr);
2870 
2871     /* Set up the RTC RTAS interfaces */
2872     spapr_rtc_create(spapr);
2873 
2874     /* Set up VIO bus */
2875     spapr->vio_bus = spapr_vio_bus_init();
2876 
2877     for (i = 0; i < serial_max_hds(); i++) {
2878         if (serial_hd(i)) {
2879             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2880         }
2881     }
2882 
2883     /* We always have at least the nvram device on VIO */
2884     spapr_create_nvram(spapr);
2885 
2886     /*
2887      * Setup hotplug / dynamic-reconfiguration connectors. top-level
2888      * connectors (described in root DT node's "ibm,drc-types" property)
2889      * are pre-initialized here. additional child connectors (such as
2890      * connectors for a PHBs PCI slots) are added as needed during their
2891      * parent's realization.
2892      */
2893     if (smc->dr_phb_enabled) {
2894         for (i = 0; i < SPAPR_MAX_PHBS; i++) {
2895             spapr_dr_connector_new(OBJECT(machine), TYPE_SPAPR_DRC_PHB, i);
2896         }
2897     }
2898 
2899     /* Set up PCI */
2900     spapr_pci_rtas_init();
2901 
2902     phb = spapr_create_default_phb();
2903 
2904     for (i = 0; i < nb_nics; i++) {
2905         NICInfo *nd = &nd_table[i];
2906 
2907         if (!nd->model) {
2908             nd->model = g_strdup("spapr-vlan");
2909         }
2910 
2911         if (g_str_equal(nd->model, "spapr-vlan") ||
2912             g_str_equal(nd->model, "ibmveth")) {
2913             spapr_vlan_create(spapr->vio_bus, nd);
2914         } else {
2915             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2916         }
2917     }
2918 
2919     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2920         spapr_vscsi_create(spapr->vio_bus);
2921     }
2922 
2923     /* Graphics */
2924     if (spapr_vga_init(phb->bus, &error_fatal)) {
2925         spapr->has_graphics = true;
2926         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2927     }
2928 
2929     if (machine->usb) {
2930         if (smc->use_ohci_by_default) {
2931             pci_create_simple(phb->bus, -1, "pci-ohci");
2932         } else {
2933             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2934         }
2935 
2936         if (spapr->has_graphics) {
2937             USBBus *usb_bus = usb_bus_find(-1);
2938 
2939             usb_create_simple(usb_bus, "usb-kbd");
2940             usb_create_simple(usb_bus, "usb-mouse");
2941         }
2942     }
2943 
2944     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2945         error_report(
2946             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2947             MIN_RMA_SLOF);
2948         exit(1);
2949     }
2950 
2951     if (kernel_filename) {
2952         uint64_t lowaddr = 0;
2953 
2954         spapr->kernel_size = load_elf(kernel_filename, NULL,
2955                                       translate_kernel_address, NULL,
2956                                       NULL, &lowaddr, NULL, 1,
2957                                       PPC_ELF_MACHINE, 0, 0);
2958         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2959             spapr->kernel_size = load_elf(kernel_filename, NULL,
2960                                           translate_kernel_address, NULL, NULL,
2961                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2962                                           0, 0);
2963             spapr->kernel_le = spapr->kernel_size > 0;
2964         }
2965         if (spapr->kernel_size < 0) {
2966             error_report("error loading %s: %s", kernel_filename,
2967                          load_elf_strerror(spapr->kernel_size));
2968             exit(1);
2969         }
2970 
2971         /* load initrd */
2972         if (initrd_filename) {
2973             /* Try to locate the initrd in the gap between the kernel
2974              * and the firmware. Add a bit of space just in case
2975              */
2976             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2977                                   + 0x1ffff) & ~0xffff;
2978             spapr->initrd_size = load_image_targphys(initrd_filename,
2979                                                      spapr->initrd_base,
2980                                                      load_limit
2981                                                      - spapr->initrd_base);
2982             if (spapr->initrd_size < 0) {
2983                 error_report("could not load initial ram disk '%s'",
2984                              initrd_filename);
2985                 exit(1);
2986             }
2987         }
2988     }
2989 
2990     if (bios_name == NULL) {
2991         bios_name = FW_FILE_NAME;
2992     }
2993     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2994     if (!filename) {
2995         error_report("Could not find LPAR firmware '%s'", bios_name);
2996         exit(1);
2997     }
2998     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2999     if (fw_size <= 0) {
3000         error_report("Could not load LPAR firmware '%s'", filename);
3001         exit(1);
3002     }
3003     g_free(filename);
3004 
3005     /* FIXME: Should register things through the MachineState's qdev
3006      * interface, this is a legacy from the sPAPREnvironment structure
3007      * which predated MachineState but had a similar function */
3008     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
3009     register_savevm_live(NULL, "spapr/htab", -1, 1,
3010                          &savevm_htab_handlers, spapr);
3011 
3012     qbus_set_hotplug_handler(sysbus_get_default(), OBJECT(machine),
3013                              &error_fatal);
3014 
3015     qemu_register_boot_set(spapr_boot_set, spapr);
3016 
3017     if (kvm_enabled()) {
3018         /* to stop and start vmclock */
3019         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
3020                                          &spapr->tb);
3021 
3022         kvmppc_spapr_enable_inkernel_multitce();
3023     }
3024 }
3025 
3026 static int spapr_kvm_type(MachineState *machine, const char *vm_type)
3027 {
3028     if (!vm_type) {
3029         return 0;
3030     }
3031 
3032     if (!strcmp(vm_type, "HV")) {
3033         return 1;
3034     }
3035 
3036     if (!strcmp(vm_type, "PR")) {
3037         return 2;
3038     }
3039 
3040     error_report("Unknown kvm-type specified '%s'", vm_type);
3041     exit(1);
3042 }
3043 
3044 /*
3045  * Implementation of an interface to adjust firmware path
3046  * for the bootindex property handling.
3047  */
3048 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3049                                    DeviceState *dev)
3050 {
3051 #define CAST(type, obj, name) \
3052     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3053     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3054     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3055     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3056 
3057     if (d) {
3058         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3059         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3060         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3061 
3062         if (spapr) {
3063             /*
3064              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3065              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3066              * 0x8000 | (target << 8) | (bus << 5) | lun
3067              * (see the "Logical unit addressing format" table in SAM5)
3068              */
3069             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3070             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3071                                    (uint64_t)id << 48);
3072         } else if (virtio) {
3073             /*
3074              * We use SRP luns of the form 01000000 | (target << 8) | lun
3075              * in the top 32 bits of the 64-bit LUN
3076              * Note: the quote above is from SLOF and it is wrong,
3077              * the actual binding is:
3078              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3079              */
3080             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3081             if (d->lun >= 256) {
3082                 /* Use the LUN "flat space addressing method" */
3083                 id |= 0x4000;
3084             }
3085             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3086                                    (uint64_t)id << 32);
3087         } else if (usb) {
3088             /*
3089              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3090              * in the top 32 bits of the 64-bit LUN
3091              */
3092             unsigned usb_port = atoi(usb->port->path);
3093             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3094             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3095                                    (uint64_t)id << 32);
3096         }
3097     }
3098 
3099     /*
3100      * SLOF probes the USB devices, and if it recognizes that the device is a
3101      * storage device, it changes its name to "storage" instead of "usb-host",
3102      * and additionally adds a child node for the SCSI LUN, so the correct
3103      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3104      */
3105     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3106         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3107         if (usb_host_dev_is_scsi_storage(usbdev)) {
3108             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3109         }
3110     }
3111 
3112     if (phb) {
3113         /* Replace "pci" with "pci@800000020000000" */
3114         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3115     }
3116 
3117     if (vsc) {
3118         /* Same logic as virtio above */
3119         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3120         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3121     }
3122 
3123     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3124         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3125         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3126         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3127     }
3128 
3129     return NULL;
3130 }
3131 
3132 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3133 {
3134     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3135 
3136     return g_strdup(spapr->kvm_type);
3137 }
3138 
3139 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3140 {
3141     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3142 
3143     g_free(spapr->kvm_type);
3144     spapr->kvm_type = g_strdup(value);
3145 }
3146 
3147 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3148 {
3149     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3150 
3151     return spapr->use_hotplug_event_source;
3152 }
3153 
3154 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3155                                             Error **errp)
3156 {
3157     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3158 
3159     spapr->use_hotplug_event_source = value;
3160 }
3161 
3162 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3163 {
3164     return true;
3165 }
3166 
3167 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3168 {
3169     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3170 
3171     switch (spapr->resize_hpt) {
3172     case SPAPR_RESIZE_HPT_DEFAULT:
3173         return g_strdup("default");
3174     case SPAPR_RESIZE_HPT_DISABLED:
3175         return g_strdup("disabled");
3176     case SPAPR_RESIZE_HPT_ENABLED:
3177         return g_strdup("enabled");
3178     case SPAPR_RESIZE_HPT_REQUIRED:
3179         return g_strdup("required");
3180     }
3181     g_assert_not_reached();
3182 }
3183 
3184 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3185 {
3186     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3187 
3188     if (strcmp(value, "default") == 0) {
3189         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3190     } else if (strcmp(value, "disabled") == 0) {
3191         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3192     } else if (strcmp(value, "enabled") == 0) {
3193         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3194     } else if (strcmp(value, "required") == 0) {
3195         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3196     } else {
3197         error_setg(errp, "Bad value for \"resize-hpt\" property");
3198     }
3199 }
3200 
3201 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3202                                    void *opaque, Error **errp)
3203 {
3204     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3205 }
3206 
3207 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3208                                    void *opaque, Error **errp)
3209 {
3210     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3211 }
3212 
3213 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3214 {
3215     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3216 
3217     if (spapr->irq == &spapr_irq_xics_legacy) {
3218         return g_strdup("legacy");
3219     } else if (spapr->irq == &spapr_irq_xics) {
3220         return g_strdup("xics");
3221     } else if (spapr->irq == &spapr_irq_xive) {
3222         return g_strdup("xive");
3223     } else if (spapr->irq == &spapr_irq_dual) {
3224         return g_strdup("dual");
3225     }
3226     g_assert_not_reached();
3227 }
3228 
3229 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3230 {
3231     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3232 
3233     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3234         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3235         return;
3236     }
3237 
3238     /* The legacy IRQ backend can not be set */
3239     if (strcmp(value, "xics") == 0) {
3240         spapr->irq = &spapr_irq_xics;
3241     } else if (strcmp(value, "xive") == 0) {
3242         spapr->irq = &spapr_irq_xive;
3243     } else if (strcmp(value, "dual") == 0) {
3244         spapr->irq = &spapr_irq_dual;
3245     } else {
3246         error_setg(errp, "Bad value for \"ic-mode\" property");
3247     }
3248 }
3249 
3250 static char *spapr_get_host_model(Object *obj, Error **errp)
3251 {
3252     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3253 
3254     return g_strdup(spapr->host_model);
3255 }
3256 
3257 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3258 {
3259     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3260 
3261     g_free(spapr->host_model);
3262     spapr->host_model = g_strdup(value);
3263 }
3264 
3265 static char *spapr_get_host_serial(Object *obj, Error **errp)
3266 {
3267     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3268 
3269     return g_strdup(spapr->host_serial);
3270 }
3271 
3272 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3273 {
3274     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3275 
3276     g_free(spapr->host_serial);
3277     spapr->host_serial = g_strdup(value);
3278 }
3279 
3280 static void spapr_instance_init(Object *obj)
3281 {
3282     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3283     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3284 
3285     spapr->htab_fd = -1;
3286     spapr->use_hotplug_event_source = true;
3287     object_property_add_str(obj, "kvm-type",
3288                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3289     object_property_set_description(obj, "kvm-type",
3290                                     "Specifies the KVM virtualization mode (HV, PR)",
3291                                     NULL);
3292     object_property_add_bool(obj, "modern-hotplug-events",
3293                             spapr_get_modern_hotplug_events,
3294                             spapr_set_modern_hotplug_events,
3295                             NULL);
3296     object_property_set_description(obj, "modern-hotplug-events",
3297                                     "Use dedicated hotplug event mechanism in"
3298                                     " place of standard EPOW events when possible"
3299                                     " (required for memory hot-unplug support)",
3300                                     NULL);
3301     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3302                             "Maximum permitted CPU compatibility mode",
3303                             &error_fatal);
3304 
3305     object_property_add_str(obj, "resize-hpt",
3306                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3307     object_property_set_description(obj, "resize-hpt",
3308                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3309                                     NULL);
3310     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3311                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3312     object_property_set_description(obj, "vsmt",
3313                                     "Virtual SMT: KVM behaves as if this were"
3314                                     " the host's SMT mode", &error_abort);
3315     object_property_add_bool(obj, "vfio-no-msix-emulation",
3316                              spapr_get_msix_emulation, NULL, NULL);
3317 
3318     /* The machine class defines the default interrupt controller mode */
3319     spapr->irq = smc->irq;
3320     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3321                             spapr_set_ic_mode, NULL);
3322     object_property_set_description(obj, "ic-mode",
3323                  "Specifies the interrupt controller mode (xics, xive, dual)",
3324                  NULL);
3325 
3326     object_property_add_str(obj, "host-model",
3327         spapr_get_host_model, spapr_set_host_model,
3328         &error_abort);
3329     object_property_set_description(obj, "host-model",
3330         "Set host's model-id to use - none|passthrough|string", &error_abort);
3331     object_property_add_str(obj, "host-serial",
3332         spapr_get_host_serial, spapr_set_host_serial,
3333         &error_abort);
3334     object_property_set_description(obj, "host-serial",
3335         "Set host's system-id to use - none|passthrough|string", &error_abort);
3336 }
3337 
3338 static void spapr_machine_finalizefn(Object *obj)
3339 {
3340     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3341 
3342     g_free(spapr->kvm_type);
3343 }
3344 
3345 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3346 {
3347     cpu_synchronize_state(cs);
3348     ppc_cpu_do_system_reset(cs);
3349 }
3350 
3351 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3352 {
3353     CPUState *cs;
3354 
3355     CPU_FOREACH(cs) {
3356         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3357     }
3358 }
3359 
3360 int spapr_lmb_dt_populate(sPAPRDRConnector *drc, sPAPRMachineState *spapr,
3361                           void *fdt, int *fdt_start_offset, Error **errp)
3362 {
3363     uint64_t addr;
3364     uint32_t node;
3365 
3366     addr = spapr_drc_index(drc) * SPAPR_MEMORY_BLOCK_SIZE;
3367     node = object_property_get_uint(OBJECT(drc->dev), PC_DIMM_NODE_PROP,
3368                                     &error_abort);
3369     *fdt_start_offset = spapr_populate_memory_node(fdt, node, addr,
3370                                                    SPAPR_MEMORY_BLOCK_SIZE);
3371     return 0;
3372 }
3373 
3374 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3375                            bool dedicated_hp_event_source, Error **errp)
3376 {
3377     sPAPRDRConnector *drc;
3378     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3379     int i;
3380     uint64_t addr = addr_start;
3381     bool hotplugged = spapr_drc_hotplugged(dev);
3382     Error *local_err = NULL;
3383 
3384     for (i = 0; i < nr_lmbs; i++) {
3385         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3386                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3387         g_assert(drc);
3388 
3389         spapr_drc_attach(drc, dev, &local_err);
3390         if (local_err) {
3391             while (addr > addr_start) {
3392                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3393                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3394                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3395                 spapr_drc_detach(drc);
3396             }
3397             error_propagate(errp, local_err);
3398             return;
3399         }
3400         if (!hotplugged) {
3401             spapr_drc_reset(drc);
3402         }
3403         addr += SPAPR_MEMORY_BLOCK_SIZE;
3404     }
3405     /* send hotplug notification to the
3406      * guest only in case of hotplugged memory
3407      */
3408     if (hotplugged) {
3409         if (dedicated_hp_event_source) {
3410             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3411                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3412             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3413                                                    nr_lmbs,
3414                                                    spapr_drc_index(drc));
3415         } else {
3416             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3417                                            nr_lmbs);
3418         }
3419     }
3420 }
3421 
3422 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3423                               Error **errp)
3424 {
3425     Error *local_err = NULL;
3426     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3427     PCDIMMDevice *dimm = PC_DIMM(dev);
3428     uint64_t size, addr;
3429 
3430     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3431 
3432     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3433     if (local_err) {
3434         goto out;
3435     }
3436 
3437     addr = object_property_get_uint(OBJECT(dimm),
3438                                     PC_DIMM_ADDR_PROP, &local_err);
3439     if (local_err) {
3440         goto out_unplug;
3441     }
3442 
3443     spapr_add_lmbs(dev, addr, size, spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3444                    &local_err);
3445     if (local_err) {
3446         goto out_unplug;
3447     }
3448 
3449     return;
3450 
3451 out_unplug:
3452     pc_dimm_unplug(dimm, MACHINE(ms));
3453 out:
3454     error_propagate(errp, local_err);
3455 }
3456 
3457 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3458                                   Error **errp)
3459 {
3460     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3461     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3462     PCDIMMDevice *dimm = PC_DIMM(dev);
3463     Error *local_err = NULL;
3464     uint64_t size;
3465     Object *memdev;
3466     hwaddr pagesize;
3467 
3468     if (!smc->dr_lmb_enabled) {
3469         error_setg(errp, "Memory hotplug not supported for this machine");
3470         return;
3471     }
3472 
3473     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3474     if (local_err) {
3475         error_propagate(errp, local_err);
3476         return;
3477     }
3478 
3479     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3480         error_setg(errp, "Hotplugged memory size must be a multiple of "
3481                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3482         return;
3483     }
3484 
3485     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3486                                       &error_abort);
3487     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3488     spapr_check_pagesize(spapr, pagesize, &local_err);
3489     if (local_err) {
3490         error_propagate(errp, local_err);
3491         return;
3492     }
3493 
3494     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3495 }
3496 
3497 struct sPAPRDIMMState {
3498     PCDIMMDevice *dimm;
3499     uint32_t nr_lmbs;
3500     QTAILQ_ENTRY(sPAPRDIMMState) next;
3501 };
3502 
3503 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3504                                                        PCDIMMDevice *dimm)
3505 {
3506     sPAPRDIMMState *dimm_state = NULL;
3507 
3508     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3509         if (dimm_state->dimm == dimm) {
3510             break;
3511         }
3512     }
3513     return dimm_state;
3514 }
3515 
3516 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3517                                                       uint32_t nr_lmbs,
3518                                                       PCDIMMDevice *dimm)
3519 {
3520     sPAPRDIMMState *ds = NULL;
3521 
3522     /*
3523      * If this request is for a DIMM whose removal had failed earlier
3524      * (due to guest's refusal to remove the LMBs), we would have this
3525      * dimm already in the pending_dimm_unplugs list. In that
3526      * case don't add again.
3527      */
3528     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3529     if (!ds) {
3530         ds = g_malloc0(sizeof(sPAPRDIMMState));
3531         ds->nr_lmbs = nr_lmbs;
3532         ds->dimm = dimm;
3533         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3534     }
3535     return ds;
3536 }
3537 
3538 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3539                                               sPAPRDIMMState *dimm_state)
3540 {
3541     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3542     g_free(dimm_state);
3543 }
3544 
3545 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3546                                                         PCDIMMDevice *dimm)
3547 {
3548     sPAPRDRConnector *drc;
3549     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3550                                                   &error_abort);
3551     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3552     uint32_t avail_lmbs = 0;
3553     uint64_t addr_start, addr;
3554     int i;
3555 
3556     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3557                                          &error_abort);
3558 
3559     addr = addr_start;
3560     for (i = 0; i < nr_lmbs; i++) {
3561         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3562                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3563         g_assert(drc);
3564         if (drc->dev) {
3565             avail_lmbs++;
3566         }
3567         addr += SPAPR_MEMORY_BLOCK_SIZE;
3568     }
3569 
3570     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3571 }
3572 
3573 /* Callback to be called during DRC release. */
3574 void spapr_lmb_release(DeviceState *dev)
3575 {
3576     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3577     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3578     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3579 
3580     /* This information will get lost if a migration occurs
3581      * during the unplug process. In this case recover it. */
3582     if (ds == NULL) {
3583         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3584         g_assert(ds);
3585         /* The DRC being examined by the caller at least must be counted */
3586         g_assert(ds->nr_lmbs);
3587     }
3588 
3589     if (--ds->nr_lmbs) {
3590         return;
3591     }
3592 
3593     /*
3594      * Now that all the LMBs have been removed by the guest, call the
3595      * unplug handler chain. This can never fail.
3596      */
3597     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3598     object_unparent(OBJECT(dev));
3599 }
3600 
3601 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3602 {
3603     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3604     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3605 
3606     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3607     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3608     spapr_pending_dimm_unplugs_remove(spapr, ds);
3609 }
3610 
3611 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3612                                         DeviceState *dev, Error **errp)
3613 {
3614     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3615     Error *local_err = NULL;
3616     PCDIMMDevice *dimm = PC_DIMM(dev);
3617     uint32_t nr_lmbs;
3618     uint64_t size, addr_start, addr;
3619     int i;
3620     sPAPRDRConnector *drc;
3621 
3622     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3623     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3624 
3625     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3626                                          &local_err);
3627     if (local_err) {
3628         goto out;
3629     }
3630 
3631     /*
3632      * An existing pending dimm state for this DIMM means that there is an
3633      * unplug operation in progress, waiting for the spapr_lmb_release
3634      * callback to complete the job (BQL can't cover that far). In this case,
3635      * bail out to avoid detaching DRCs that were already released.
3636      */
3637     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3638         error_setg(&local_err,
3639                    "Memory unplug already in progress for device %s",
3640                    dev->id);
3641         goto out;
3642     }
3643 
3644     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3645 
3646     addr = addr_start;
3647     for (i = 0; i < nr_lmbs; i++) {
3648         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3649                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3650         g_assert(drc);
3651 
3652         spapr_drc_detach(drc);
3653         addr += SPAPR_MEMORY_BLOCK_SIZE;
3654     }
3655 
3656     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3657                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3658     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3659                                               nr_lmbs, spapr_drc_index(drc));
3660 out:
3661     error_propagate(errp, local_err);
3662 }
3663 
3664 /* Callback to be called during DRC release. */
3665 void spapr_core_release(DeviceState *dev)
3666 {
3667     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3668 
3669     /* Call the unplug handler chain. This can never fail. */
3670     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3671     object_unparent(OBJECT(dev));
3672 }
3673 
3674 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3675 {
3676     MachineState *ms = MACHINE(hotplug_dev);
3677     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3678     CPUCore *cc = CPU_CORE(dev);
3679     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3680 
3681     if (smc->pre_2_10_has_unused_icps) {
3682         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3683         int i;
3684 
3685         for (i = 0; i < cc->nr_threads; i++) {
3686             CPUState *cs = CPU(sc->threads[i]);
3687 
3688             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3689         }
3690     }
3691 
3692     assert(core_slot);
3693     core_slot->cpu = NULL;
3694     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3695 }
3696 
3697 static
3698 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3699                                Error **errp)
3700 {
3701     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3702     int index;
3703     sPAPRDRConnector *drc;
3704     CPUCore *cc = CPU_CORE(dev);
3705 
3706     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3707         error_setg(errp, "Unable to find CPU core with core-id: %d",
3708                    cc->core_id);
3709         return;
3710     }
3711     if (index == 0) {
3712         error_setg(errp, "Boot CPU core may not be unplugged");
3713         return;
3714     }
3715 
3716     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3717                           spapr_vcpu_id(spapr, cc->core_id));
3718     g_assert(drc);
3719 
3720     spapr_drc_detach(drc);
3721 
3722     spapr_hotplug_req_remove_by_index(drc);
3723 }
3724 
3725 int spapr_core_dt_populate(sPAPRDRConnector *drc, sPAPRMachineState *spapr,
3726                            void *fdt, int *fdt_start_offset, Error **errp)
3727 {
3728     sPAPRCPUCore *core = SPAPR_CPU_CORE(drc->dev);
3729     CPUState *cs = CPU(core->threads[0]);
3730     PowerPCCPU *cpu = POWERPC_CPU(cs);
3731     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3732     int id = spapr_get_vcpu_id(cpu);
3733     char *nodename;
3734     int offset;
3735 
3736     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3737     offset = fdt_add_subnode(fdt, 0, nodename);
3738     g_free(nodename);
3739 
3740     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3741 
3742     *fdt_start_offset = offset;
3743     return 0;
3744 }
3745 
3746 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3747                             Error **errp)
3748 {
3749     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3750     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3751     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3752     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3753     CPUCore *cc = CPU_CORE(dev);
3754     CPUState *cs;
3755     sPAPRDRConnector *drc;
3756     Error *local_err = NULL;
3757     CPUArchId *core_slot;
3758     int index;
3759     bool hotplugged = spapr_drc_hotplugged(dev);
3760 
3761     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3762     if (!core_slot) {
3763         error_setg(errp, "Unable to find CPU core with core-id: %d",
3764                    cc->core_id);
3765         return;
3766     }
3767     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3768                           spapr_vcpu_id(spapr, cc->core_id));
3769 
3770     g_assert(drc || !mc->has_hotpluggable_cpus);
3771 
3772     if (drc) {
3773         spapr_drc_attach(drc, dev, &local_err);
3774         if (local_err) {
3775             error_propagate(errp, local_err);
3776             return;
3777         }
3778 
3779         if (hotplugged) {
3780             /*
3781              * Send hotplug notification interrupt to the guest only
3782              * in case of hotplugged CPUs.
3783              */
3784             spapr_hotplug_req_add_by_index(drc);
3785         } else {
3786             spapr_drc_reset(drc);
3787         }
3788     }
3789 
3790     core_slot->cpu = OBJECT(dev);
3791 
3792     if (smc->pre_2_10_has_unused_icps) {
3793         int i;
3794 
3795         for (i = 0; i < cc->nr_threads; i++) {
3796             cs = CPU(core->threads[i]);
3797             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3798         }
3799     }
3800 }
3801 
3802 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3803                                 Error **errp)
3804 {
3805     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3806     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3807     Error *local_err = NULL;
3808     CPUCore *cc = CPU_CORE(dev);
3809     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3810     const char *type = object_get_typename(OBJECT(dev));
3811     CPUArchId *core_slot;
3812     int index;
3813 
3814     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3815         error_setg(&local_err, "CPU hotplug not supported for this machine");
3816         goto out;
3817     }
3818 
3819     if (strcmp(base_core_type, type)) {
3820         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3821         goto out;
3822     }
3823 
3824     if (cc->core_id % smp_threads) {
3825         error_setg(&local_err, "invalid core id %d", cc->core_id);
3826         goto out;
3827     }
3828 
3829     /*
3830      * In general we should have homogeneous threads-per-core, but old
3831      * (pre hotplug support) machine types allow the last core to have
3832      * reduced threads as a compatibility hack for when we allowed
3833      * total vcpus not a multiple of threads-per-core.
3834      */
3835     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3836         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3837                    cc->nr_threads, smp_threads);
3838         goto out;
3839     }
3840 
3841     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3842     if (!core_slot) {
3843         error_setg(&local_err, "core id %d out of range", cc->core_id);
3844         goto out;
3845     }
3846 
3847     if (core_slot->cpu) {
3848         error_setg(&local_err, "core %d already populated", cc->core_id);
3849         goto out;
3850     }
3851 
3852     numa_cpu_pre_plug(core_slot, dev, &local_err);
3853 
3854 out:
3855     error_propagate(errp, local_err);
3856 }
3857 
3858 int spapr_phb_dt_populate(sPAPRDRConnector *drc, sPAPRMachineState *spapr,
3859                           void *fdt, int *fdt_start_offset, Error **errp)
3860 {
3861     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(drc->dev);
3862     int intc_phandle;
3863 
3864     intc_phandle = spapr_irq_get_phandle(spapr, spapr->fdt_blob, errp);
3865     if (intc_phandle <= 0) {
3866         return -1;
3867     }
3868 
3869     if (spapr_populate_pci_dt(sphb, intc_phandle, fdt, spapr->irq->nr_msis,
3870                               fdt_start_offset)) {
3871         error_setg(errp, "unable to create FDT node for PHB %d", sphb->index);
3872         return -1;
3873     }
3874 
3875     /* generally SLOF creates these, for hotplug it's up to QEMU */
3876     _FDT(fdt_setprop_string(fdt, *fdt_start_offset, "name", "pci"));
3877 
3878     return 0;
3879 }
3880 
3881 static void spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3882                                Error **errp)
3883 {
3884     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3885     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3886     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3887     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
3888 
3889     if (dev->hotplugged && !smc->dr_phb_enabled) {
3890         error_setg(errp, "PHB hotplug not supported for this machine");
3891         return;
3892     }
3893 
3894     if (sphb->index == (uint32_t)-1) {
3895         error_setg(errp, "\"index\" for PAPR PHB is mandatory");
3896         return;
3897     }
3898 
3899     /*
3900      * This will check that sphb->index doesn't exceed the maximum number of
3901      * PHBs for the current machine type.
3902      */
3903     smc->phb_placement(spapr, sphb->index,
3904                        &sphb->buid, &sphb->io_win_addr,
3905                        &sphb->mem_win_addr, &sphb->mem64_win_addr,
3906                        windows_supported, sphb->dma_liobn, errp);
3907 }
3908 
3909 static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3910                            Error **errp)
3911 {
3912     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3913     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3914     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3915     sPAPRDRConnector *drc;
3916     bool hotplugged = spapr_drc_hotplugged(dev);
3917     Error *local_err = NULL;
3918 
3919     if (!smc->dr_phb_enabled) {
3920         return;
3921     }
3922 
3923     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3924     /* hotplug hooks should check it's enabled before getting this far */
3925     assert(drc);
3926 
3927     spapr_drc_attach(drc, DEVICE(dev), &local_err);
3928     if (local_err) {
3929         error_propagate(errp, local_err);
3930         return;
3931     }
3932 
3933     if (hotplugged) {
3934         spapr_hotplug_req_add_by_index(drc);
3935     } else {
3936         spapr_drc_reset(drc);
3937     }
3938 }
3939 
3940 void spapr_phb_release(DeviceState *dev)
3941 {
3942     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3943 
3944     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3945     object_unparent(OBJECT(dev));
3946 }
3947 
3948 static void spapr_phb_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3949 {
3950     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
3951 }
3952 
3953 static void spapr_phb_unplug_request(HotplugHandler *hotplug_dev,
3954                                      DeviceState *dev, Error **errp)
3955 {
3956     sPAPRPHBState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
3957     sPAPRDRConnector *drc;
3958 
3959     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
3960     assert(drc);
3961 
3962     if (!spapr_drc_unplug_requested(drc)) {
3963         spapr_drc_detach(drc);
3964         spapr_hotplug_req_remove_by_index(drc);
3965     }
3966 }
3967 
3968 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3969                                       DeviceState *dev, Error **errp)
3970 {
3971     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3972         spapr_memory_plug(hotplug_dev, dev, errp);
3973     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3974         spapr_core_plug(hotplug_dev, dev, errp);
3975     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
3976         spapr_phb_plug(hotplug_dev, dev, errp);
3977     }
3978 }
3979 
3980 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3981                                         DeviceState *dev, Error **errp)
3982 {
3983     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3984         spapr_memory_unplug(hotplug_dev, dev);
3985     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3986         spapr_core_unplug(hotplug_dev, dev);
3987     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
3988         spapr_phb_unplug(hotplug_dev, dev);
3989     }
3990 }
3991 
3992 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3993                                                 DeviceState *dev, Error **errp)
3994 {
3995     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3996     MachineClass *mc = MACHINE_GET_CLASS(sms);
3997     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3998 
3999     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4000         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
4001             spapr_memory_unplug_request(hotplug_dev, dev, errp);
4002         } else {
4003             /* NOTE: this means there is a window after guest reset, prior to
4004              * CAS negotiation, where unplug requests will fail due to the
4005              * capability not being detected yet. This is a bit different than
4006              * the case with PCI unplug, where the events will be queued and
4007              * eventually handled by the guest after boot
4008              */
4009             error_setg(errp, "Memory hot unplug not supported for this guest");
4010         }
4011     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4012         if (!mc->has_hotpluggable_cpus) {
4013             error_setg(errp, "CPU hot unplug not supported on this machine");
4014             return;
4015         }
4016         spapr_core_unplug_request(hotplug_dev, dev, errp);
4017     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4018         if (!smc->dr_phb_enabled) {
4019             error_setg(errp, "PHB hot unplug not supported on this machine");
4020             return;
4021         }
4022         spapr_phb_unplug_request(hotplug_dev, dev, errp);
4023     }
4024 }
4025 
4026 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
4027                                           DeviceState *dev, Error **errp)
4028 {
4029     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
4030         spapr_memory_pre_plug(hotplug_dev, dev, errp);
4031     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
4032         spapr_core_pre_plug(hotplug_dev, dev, errp);
4033     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4034         spapr_phb_pre_plug(hotplug_dev, dev, errp);
4035     }
4036 }
4037 
4038 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
4039                                                  DeviceState *dev)
4040 {
4041     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
4042         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE) ||
4043         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
4044         return HOTPLUG_HANDLER(machine);
4045     }
4046     return NULL;
4047 }
4048 
4049 static CpuInstanceProperties
4050 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
4051 {
4052     CPUArchId *core_slot;
4053     MachineClass *mc = MACHINE_GET_CLASS(machine);
4054 
4055     /* make sure possible_cpu are intialized */
4056     mc->possible_cpu_arch_ids(machine);
4057     /* get CPU core slot containing thread that matches cpu_index */
4058     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
4059     assert(core_slot);
4060     return core_slot->props;
4061 }
4062 
4063 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
4064 {
4065     return idx / smp_cores % nb_numa_nodes;
4066 }
4067 
4068 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
4069 {
4070     int i;
4071     const char *core_type;
4072     int spapr_max_cores = max_cpus / smp_threads;
4073     MachineClass *mc = MACHINE_GET_CLASS(machine);
4074 
4075     if (!mc->has_hotpluggable_cpus) {
4076         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
4077     }
4078     if (machine->possible_cpus) {
4079         assert(machine->possible_cpus->len == spapr_max_cores);
4080         return machine->possible_cpus;
4081     }
4082 
4083     core_type = spapr_get_cpu_core_type(machine->cpu_type);
4084     if (!core_type) {
4085         error_report("Unable to find sPAPR CPU Core definition");
4086         exit(1);
4087     }
4088 
4089     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
4090                              sizeof(CPUArchId) * spapr_max_cores);
4091     machine->possible_cpus->len = spapr_max_cores;
4092     for (i = 0; i < machine->possible_cpus->len; i++) {
4093         int core_id = i * smp_threads;
4094 
4095         machine->possible_cpus->cpus[i].type = core_type;
4096         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
4097         machine->possible_cpus->cpus[i].arch_id = core_id;
4098         machine->possible_cpus->cpus[i].props.has_core_id = true;
4099         machine->possible_cpus->cpus[i].props.core_id = core_id;
4100     }
4101     return machine->possible_cpus;
4102 }
4103 
4104 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
4105                                 uint64_t *buid, hwaddr *pio,
4106                                 hwaddr *mmio32, hwaddr *mmio64,
4107                                 unsigned n_dma, uint32_t *liobns, Error **errp)
4108 {
4109     /*
4110      * New-style PHB window placement.
4111      *
4112      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
4113      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
4114      * windows.
4115      *
4116      * Some guest kernels can't work with MMIO windows above 1<<46
4117      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
4118      *
4119      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
4120      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
4121      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
4122      * 1TiB 64-bit MMIO windows for each PHB.
4123      */
4124     const uint64_t base_buid = 0x800000020000000ULL;
4125     int i;
4126 
4127     /* Sanity check natural alignments */
4128     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4129     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
4130     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
4131     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
4132     /* Sanity check bounds */
4133     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
4134                       SPAPR_PCI_MEM32_WIN_SIZE);
4135     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
4136                       SPAPR_PCI_MEM64_WIN_SIZE);
4137 
4138     if (index >= SPAPR_MAX_PHBS) {
4139         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
4140                    SPAPR_MAX_PHBS - 1);
4141         return;
4142     }
4143 
4144     *buid = base_buid + index;
4145     for (i = 0; i < n_dma; ++i) {
4146         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4147     }
4148 
4149     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
4150     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
4151     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
4152 }
4153 
4154 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4155 {
4156     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
4157 
4158     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4159 }
4160 
4161 static void spapr_ics_resend(XICSFabric *dev)
4162 {
4163     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
4164 
4165     ics_resend(spapr->ics);
4166 }
4167 
4168 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4169 {
4170     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4171 
4172     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4173 }
4174 
4175 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4176                                  Monitor *mon)
4177 {
4178     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
4179 
4180     spapr->irq->print_info(spapr, mon);
4181 }
4182 
4183 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4184 {
4185     return cpu->vcpu_id;
4186 }
4187 
4188 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4189 {
4190     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4191     int vcpu_id;
4192 
4193     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4194 
4195     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4196         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4197         error_append_hint(errp, "Adjust the number of cpus to %d "
4198                           "or try to raise the number of threads per core\n",
4199                           vcpu_id * smp_threads / spapr->vsmt);
4200         return;
4201     }
4202 
4203     cpu->vcpu_id = vcpu_id;
4204 }
4205 
4206 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4207 {
4208     CPUState *cs;
4209 
4210     CPU_FOREACH(cs) {
4211         PowerPCCPU *cpu = POWERPC_CPU(cs);
4212 
4213         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4214             return cpu;
4215         }
4216     }
4217 
4218     return NULL;
4219 }
4220 
4221 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4222 {
4223     MachineClass *mc = MACHINE_CLASS(oc);
4224     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4225     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4226     NMIClass *nc = NMI_CLASS(oc);
4227     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4228     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4229     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4230     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4231 
4232     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4233     mc->ignore_boot_device_suffixes = true;
4234 
4235     /*
4236      * We set up the default / latest behaviour here.  The class_init
4237      * functions for the specific versioned machine types can override
4238      * these details for backwards compatibility
4239      */
4240     mc->init = spapr_machine_init;
4241     mc->reset = spapr_machine_reset;
4242     mc->block_default_type = IF_SCSI;
4243     mc->max_cpus = 1024;
4244     mc->no_parallel = 1;
4245     mc->default_boot_order = "";
4246     mc->default_ram_size = 512 * MiB;
4247     mc->default_display = "std";
4248     mc->kvm_type = spapr_kvm_type;
4249     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4250     mc->pci_allow_0_address = true;
4251     assert(!mc->get_hotplug_handler);
4252     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4253     hc->pre_plug = spapr_machine_device_pre_plug;
4254     hc->plug = spapr_machine_device_plug;
4255     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4256     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4257     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4258     hc->unplug_request = spapr_machine_device_unplug_request;
4259     hc->unplug = spapr_machine_device_unplug;
4260 
4261     smc->dr_lmb_enabled = true;
4262     smc->update_dt_enabled = true;
4263     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4264     mc->has_hotpluggable_cpus = true;
4265     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4266     fwc->get_dev_path = spapr_get_fw_dev_path;
4267     nc->nmi_monitor_handler = spapr_nmi;
4268     smc->phb_placement = spapr_phb_placement;
4269     vhc->hypercall = emulate_spapr_hypercall;
4270     vhc->hpt_mask = spapr_hpt_mask;
4271     vhc->map_hptes = spapr_map_hptes;
4272     vhc->unmap_hptes = spapr_unmap_hptes;
4273     vhc->store_hpte = spapr_store_hpte;
4274     vhc->get_pate = spapr_get_pate;
4275     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4276     xic->ics_get = spapr_ics_get;
4277     xic->ics_resend = spapr_ics_resend;
4278     xic->icp_get = spapr_icp_get;
4279     ispc->print_info = spapr_pic_print_info;
4280     /* Force NUMA node memory size to be a multiple of
4281      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4282      * in which LMBs are represented and hot-added
4283      */
4284     mc->numa_mem_align_shift = 28;
4285 
4286     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4287     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4288     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4289     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4290     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4291     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4292     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4293     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4294     spapr_caps_add_properties(smc, &error_abort);
4295     smc->irq = &spapr_irq_xics;
4296     smc->dr_phb_enabled = true;
4297 }
4298 
4299 static const TypeInfo spapr_machine_info = {
4300     .name          = TYPE_SPAPR_MACHINE,
4301     .parent        = TYPE_MACHINE,
4302     .abstract      = true,
4303     .instance_size = sizeof(sPAPRMachineState),
4304     .instance_init = spapr_instance_init,
4305     .instance_finalize = spapr_machine_finalizefn,
4306     .class_size    = sizeof(sPAPRMachineClass),
4307     .class_init    = spapr_machine_class_init,
4308     .interfaces = (InterfaceInfo[]) {
4309         { TYPE_FW_PATH_PROVIDER },
4310         { TYPE_NMI },
4311         { TYPE_HOTPLUG_HANDLER },
4312         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4313         { TYPE_XICS_FABRIC },
4314         { TYPE_INTERRUPT_STATS_PROVIDER },
4315         { }
4316     },
4317 };
4318 
4319 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4320     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4321                                                     void *data)      \
4322     {                                                                \
4323         MachineClass *mc = MACHINE_CLASS(oc);                        \
4324         spapr_machine_##suffix##_class_options(mc);                  \
4325         if (latest) {                                                \
4326             mc->alias = "pseries";                                   \
4327             mc->is_default = 1;                                      \
4328         }                                                            \
4329     }                                                                \
4330     static const TypeInfo spapr_machine_##suffix##_info = {          \
4331         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4332         .parent = TYPE_SPAPR_MACHINE,                                \
4333         .class_init = spapr_machine_##suffix##_class_init,           \
4334     };                                                               \
4335     static void spapr_machine_register_##suffix(void)                \
4336     {                                                                \
4337         type_register(&spapr_machine_##suffix##_info);               \
4338     }                                                                \
4339     type_init(spapr_machine_register_##suffix)
4340 
4341 /*
4342  * pseries-4.0
4343  */
4344 static void spapr_machine_4_0_class_options(MachineClass *mc)
4345 {
4346     /* Defaults for the latest behaviour inherited from the base class */
4347 }
4348 
4349 DEFINE_SPAPR_MACHINE(4_0, "4.0", true);
4350 
4351 /*
4352  * pseries-3.1
4353  */
4354 static void spapr_machine_3_1_class_options(MachineClass *mc)
4355 {
4356     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4357     static GlobalProperty compat[] = {
4358         { TYPE_SPAPR_MACHINE, "host-model", "passthrough" },
4359         { TYPE_SPAPR_MACHINE, "host-serial", "passthrough" },
4360     };
4361 
4362     spapr_machine_4_0_class_options(mc);
4363     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4364     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4365 
4366     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4367     smc->update_dt_enabled = false;
4368     smc->dr_phb_enabled = false;
4369 }
4370 
4371 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4372 
4373 /*
4374  * pseries-3.0
4375  */
4376 
4377 static void spapr_machine_3_0_class_options(MachineClass *mc)
4378 {
4379     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4380 
4381     spapr_machine_3_1_class_options(mc);
4382     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4383 
4384     smc->legacy_irq_allocation = true;
4385     smc->irq = &spapr_irq_xics_legacy;
4386 }
4387 
4388 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4389 
4390 /*
4391  * pseries-2.12
4392  */
4393 static void spapr_machine_2_12_class_options(MachineClass *mc)
4394 {
4395     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4396     static GlobalProperty compat[] = {
4397         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4398         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4399     };
4400 
4401     spapr_machine_3_0_class_options(mc);
4402     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4403     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4404 
4405     /* We depend on kvm_enabled() to choose a default value for the
4406      * hpt-max-page-size capability. Of course we can't do it here
4407      * because this is too early and the HW accelerator isn't initialzed
4408      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4409      */
4410     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4411 }
4412 
4413 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4414 
4415 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4416 {
4417     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4418 
4419     spapr_machine_2_12_class_options(mc);
4420     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4421     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4422     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4423 }
4424 
4425 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4426 
4427 /*
4428  * pseries-2.11
4429  */
4430 
4431 static void spapr_machine_2_11_class_options(MachineClass *mc)
4432 {
4433     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4434 
4435     spapr_machine_2_12_class_options(mc);
4436     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4437     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4438 }
4439 
4440 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4441 
4442 /*
4443  * pseries-2.10
4444  */
4445 
4446 static void spapr_machine_2_10_class_options(MachineClass *mc)
4447 {
4448     spapr_machine_2_11_class_options(mc);
4449     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4450 }
4451 
4452 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4453 
4454 /*
4455  * pseries-2.9
4456  */
4457 
4458 static void spapr_machine_2_9_class_options(MachineClass *mc)
4459 {
4460     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4461     static GlobalProperty compat[] = {
4462         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4463     };
4464 
4465     spapr_machine_2_10_class_options(mc);
4466     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4467     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4468     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4469     smc->pre_2_10_has_unused_icps = true;
4470     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4471 }
4472 
4473 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4474 
4475 /*
4476  * pseries-2.8
4477  */
4478 
4479 static void spapr_machine_2_8_class_options(MachineClass *mc)
4480 {
4481     static GlobalProperty compat[] = {
4482         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4483     };
4484 
4485     spapr_machine_2_9_class_options(mc);
4486     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4487     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4488     mc->numa_mem_align_shift = 23;
4489 }
4490 
4491 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4492 
4493 /*
4494  * pseries-2.7
4495  */
4496 
4497 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4498                               uint64_t *buid, hwaddr *pio,
4499                               hwaddr *mmio32, hwaddr *mmio64,
4500                               unsigned n_dma, uint32_t *liobns, Error **errp)
4501 {
4502     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4503     const uint64_t base_buid = 0x800000020000000ULL;
4504     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4505     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4506     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4507     const uint32_t max_index = 255;
4508     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4509 
4510     uint64_t ram_top = MACHINE(spapr)->ram_size;
4511     hwaddr phb0_base, phb_base;
4512     int i;
4513 
4514     /* Do we have device memory? */
4515     if (MACHINE(spapr)->maxram_size > ram_top) {
4516         /* Can't just use maxram_size, because there may be an
4517          * alignment gap between normal and device memory regions
4518          */
4519         ram_top = MACHINE(spapr)->device_memory->base +
4520             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4521     }
4522 
4523     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4524 
4525     if (index > max_index) {
4526         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4527                    max_index);
4528         return;
4529     }
4530 
4531     *buid = base_buid + index;
4532     for (i = 0; i < n_dma; ++i) {
4533         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4534     }
4535 
4536     phb_base = phb0_base + index * phb_spacing;
4537     *pio = phb_base + pio_offset;
4538     *mmio32 = phb_base + mmio_offset;
4539     /*
4540      * We don't set the 64-bit MMIO window, relying on the PHB's
4541      * fallback behaviour of automatically splitting a large "32-bit"
4542      * window into contiguous 32-bit and 64-bit windows
4543      */
4544 }
4545 
4546 static void spapr_machine_2_7_class_options(MachineClass *mc)
4547 {
4548     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4549     static GlobalProperty compat[] = {
4550         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4551         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4552         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4553         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4554     };
4555 
4556     spapr_machine_2_8_class_options(mc);
4557     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4558     mc->default_machine_opts = "modern-hotplug-events=off";
4559     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4560     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4561     smc->phb_placement = phb_placement_2_7;
4562 }
4563 
4564 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4565 
4566 /*
4567  * pseries-2.6
4568  */
4569 
4570 static void spapr_machine_2_6_class_options(MachineClass *mc)
4571 {
4572     static GlobalProperty compat[] = {
4573         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4574     };
4575 
4576     spapr_machine_2_7_class_options(mc);
4577     mc->has_hotpluggable_cpus = false;
4578     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4579     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4580 }
4581 
4582 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4583 
4584 /*
4585  * pseries-2.5
4586  */
4587 
4588 static void spapr_machine_2_5_class_options(MachineClass *mc)
4589 {
4590     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4591     static GlobalProperty compat[] = {
4592         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4593     };
4594 
4595     spapr_machine_2_6_class_options(mc);
4596     smc->use_ohci_by_default = true;
4597     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4598     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4599 }
4600 
4601 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4602 
4603 /*
4604  * pseries-2.4
4605  */
4606 
4607 static void spapr_machine_2_4_class_options(MachineClass *mc)
4608 {
4609     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4610 
4611     spapr_machine_2_5_class_options(mc);
4612     smc->dr_lmb_enabled = false;
4613     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4614 }
4615 
4616 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4617 
4618 /*
4619  * pseries-2.3
4620  */
4621 
4622 static void spapr_machine_2_3_class_options(MachineClass *mc)
4623 {
4624     static GlobalProperty compat[] = {
4625         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4626     };
4627     spapr_machine_2_4_class_options(mc);
4628     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4629     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4630 }
4631 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4632 
4633 /*
4634  * pseries-2.2
4635  */
4636 
4637 static void spapr_machine_2_2_class_options(MachineClass *mc)
4638 {
4639     static GlobalProperty compat[] = {
4640         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4641     };
4642 
4643     spapr_machine_2_3_class_options(mc);
4644     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4645     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4646     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4647 }
4648 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4649 
4650 /*
4651  * pseries-2.1
4652  */
4653 
4654 static void spapr_machine_2_1_class_options(MachineClass *mc)
4655 {
4656     spapr_machine_2_2_class_options(mc);
4657     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4658 }
4659 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4660 
4661 static void spapr_machine_register_types(void)
4662 {
4663     type_register_static(&spapr_machine_info);
4664 }
4665 
4666 type_init(spapr_machine_register_types)
4667