xref: /openbmc/qemu/hw/ppc/spapr.c (revision c7b95171)
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_XICP            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, buf_len;
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, 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(qemu_hypertas, "hcall-memop1");
1052 
1053     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1054         add_str(hypertas, "hcall-multi-tce");
1055     }
1056 
1057     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1058         add_str(hypertas, "hcall-hpt-resize");
1059     }
1060 
1061     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1062                      hypertas->str, hypertas->len));
1063     g_string_free(hypertas, TRUE);
1064     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1065                      qemu_hypertas->str, qemu_hypertas->len));
1066     g_string_free(qemu_hypertas, TRUE);
1067 
1068     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1069                      refpoints, sizeof(refpoints)));
1070 
1071     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1072                      maxdomains, sizeof(maxdomains)));
1073 
1074     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1075                           RTAS_ERROR_LOG_MAX));
1076     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1077                           RTAS_EVENT_SCAN_RATE));
1078 
1079     g_assert(msi_nonbroken);
1080     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1081 
1082     /*
1083      * According to PAPR, rtas ibm,os-term does not guarantee a return
1084      * back to the guest cpu.
1085      *
1086      * While an additional ibm,extended-os-term property indicates
1087      * that rtas call return will always occur. Set this property.
1088      */
1089     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1090 
1091     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1092                      lrdr_capacity, sizeof(lrdr_capacity)));
1093 
1094     spapr_dt_rtas_tokens(fdt, rtas);
1095 }
1096 
1097 /*
1098  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1099  * and the XIVE features that the guest may request and thus the valid
1100  * values for bytes 23..26 of option vector 5:
1101  */
1102 static void spapr_dt_ov5_platform_support(sPAPRMachineState *spapr, void *fdt,
1103                                           int chosen)
1104 {
1105     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1106 
1107     char val[2 * 4] = {
1108         23, spapr->irq->ov5, /* Xive mode. */
1109         24, 0x00, /* Hash/Radix, filled in below. */
1110         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1111         26, 0x40, /* Radix options: GTSE == yes. */
1112     };
1113 
1114     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1115                           first_ppc_cpu->compat_pvr)) {
1116         /*
1117          * If we're in a pre POWER9 compat mode then the guest should
1118          * do hash and use the legacy interrupt mode
1119          */
1120         val[1] = 0x00; /* XICS */
1121         val[3] = 0x00; /* Hash */
1122     } else if (kvm_enabled()) {
1123         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1124             val[3] = 0x80; /* OV5_MMU_BOTH */
1125         } else if (kvmppc_has_cap_mmu_radix()) {
1126             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1127         } else {
1128             val[3] = 0x00; /* Hash */
1129         }
1130     } else {
1131         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1132         val[3] = 0xC0;
1133     }
1134     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1135                      val, sizeof(val)));
1136 }
1137 
1138 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1139 {
1140     MachineState *machine = MACHINE(spapr);
1141     int chosen;
1142     const char *boot_device = machine->boot_order;
1143     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1144     size_t cb = 0;
1145     char *bootlist = get_boot_devices_list(&cb);
1146 
1147     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1148 
1149     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1150     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1151                           spapr->initrd_base));
1152     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1153                           spapr->initrd_base + spapr->initrd_size));
1154 
1155     if (spapr->kernel_size) {
1156         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1157                               cpu_to_be64(spapr->kernel_size) };
1158 
1159         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1160                          &kprop, sizeof(kprop)));
1161         if (spapr->kernel_le) {
1162             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1163         }
1164     }
1165     if (boot_menu) {
1166         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1167     }
1168     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1169     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1170     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1171 
1172     if (cb && bootlist) {
1173         int i;
1174 
1175         for (i = 0; i < cb; i++) {
1176             if (bootlist[i] == '\n') {
1177                 bootlist[i] = ' ';
1178             }
1179         }
1180         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1181     }
1182 
1183     if (boot_device && strlen(boot_device)) {
1184         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1185     }
1186 
1187     if (!spapr->has_graphics && stdout_path) {
1188         /*
1189          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1190          * kernel. New platforms should only use the "stdout-path" property. Set
1191          * the new property and continue using older property to remain
1192          * compatible with the existing firmware.
1193          */
1194         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1195         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1196     }
1197 
1198     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1199 
1200     g_free(stdout_path);
1201     g_free(bootlist);
1202 }
1203 
1204 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1205 {
1206     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1207      * KVM to work under pHyp with some guest co-operation */
1208     int hypervisor;
1209     uint8_t hypercall[16];
1210 
1211     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1212     /* indicate KVM hypercall interface */
1213     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1214     if (kvmppc_has_cap_fixup_hcalls()) {
1215         /*
1216          * Older KVM versions with older guest kernels were broken
1217          * with the magic page, don't allow the guest to map it.
1218          */
1219         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1220                                   sizeof(hypercall))) {
1221             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1222                              hypercall, sizeof(hypercall)));
1223         }
1224     }
1225 }
1226 
1227 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1228                              hwaddr rtas_addr,
1229                              hwaddr rtas_size)
1230 {
1231     MachineState *machine = MACHINE(spapr);
1232     MachineClass *mc = MACHINE_GET_CLASS(machine);
1233     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1234     int ret;
1235     void *fdt;
1236     sPAPRPHBState *phb;
1237     char *buf;
1238 
1239     fdt = g_malloc0(FDT_MAX_SIZE);
1240     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1241 
1242     /* Root node */
1243     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1244     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1245     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1246 
1247     /*
1248      * Add info to guest to indentify which host is it being run on
1249      * and what is the uuid of the guest
1250      */
1251     if (kvmppc_get_host_model(&buf)) {
1252         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1253         g_free(buf);
1254     }
1255     if (kvmppc_get_host_serial(&buf)) {
1256         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1257         g_free(buf);
1258     }
1259 
1260     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1261 
1262     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1263     if (qemu_uuid_set) {
1264         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1265     }
1266     g_free(buf);
1267 
1268     if (qemu_get_vm_name()) {
1269         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1270                                 qemu_get_vm_name()));
1271     }
1272 
1273     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1274     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1275 
1276     /* /interrupt controller */
1277     spapr->irq->dt_populate(spapr, spapr_max_server_number(spapr), fdt,
1278                           PHANDLE_XICP);
1279 
1280     ret = spapr_populate_memory(spapr, fdt);
1281     if (ret < 0) {
1282         error_report("couldn't setup memory nodes in fdt");
1283         exit(1);
1284     }
1285 
1286     /* /vdevice */
1287     spapr_dt_vdevice(spapr->vio_bus, fdt);
1288 
1289     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1290         ret = spapr_rng_populate_dt(fdt);
1291         if (ret < 0) {
1292             error_report("could not set up rng device in the fdt");
1293             exit(1);
1294         }
1295     }
1296 
1297     QLIST_FOREACH(phb, &spapr->phbs, list) {
1298         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt,
1299                                     spapr->irq->nr_msis);
1300         if (ret < 0) {
1301             error_report("couldn't setup PCI devices in fdt");
1302             exit(1);
1303         }
1304     }
1305 
1306     /* cpus */
1307     spapr_populate_cpus_dt_node(fdt, spapr);
1308 
1309     if (smc->dr_lmb_enabled) {
1310         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1311     }
1312 
1313     if (mc->has_hotpluggable_cpus) {
1314         int offset = fdt_path_offset(fdt, "/cpus");
1315         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1316                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1317         if (ret < 0) {
1318             error_report("Couldn't set up CPU DR device tree properties");
1319             exit(1);
1320         }
1321     }
1322 
1323     /* /event-sources */
1324     spapr_dt_events(spapr, fdt);
1325 
1326     /* /rtas */
1327     spapr_dt_rtas(spapr, fdt);
1328 
1329     /* /chosen */
1330     spapr_dt_chosen(spapr, fdt);
1331 
1332     /* /hypervisor */
1333     if (kvm_enabled()) {
1334         spapr_dt_hypervisor(spapr, fdt);
1335     }
1336 
1337     /* Build memory reserve map */
1338     if (spapr->kernel_size) {
1339         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1340     }
1341     if (spapr->initrd_size) {
1342         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1343     }
1344 
1345     /* ibm,client-architecture-support updates */
1346     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1347     if (ret < 0) {
1348         error_report("couldn't setup CAS properties fdt");
1349         exit(1);
1350     }
1351 
1352     return fdt;
1353 }
1354 
1355 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1356 {
1357     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1358 }
1359 
1360 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1361                                     PowerPCCPU *cpu)
1362 {
1363     CPUPPCState *env = &cpu->env;
1364 
1365     /* The TCG path should also be holding the BQL at this point */
1366     g_assert(qemu_mutex_iothread_locked());
1367 
1368     if (msr_pr) {
1369         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1370         env->gpr[3] = H_PRIVILEGE;
1371     } else {
1372         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1373     }
1374 }
1375 
1376 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1377 {
1378     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1379 
1380     return spapr->patb_entry;
1381 }
1382 
1383 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1384 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1385 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1386 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1387 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1388 
1389 /*
1390  * Get the fd to access the kernel htab, re-opening it if necessary
1391  */
1392 static int get_htab_fd(sPAPRMachineState *spapr)
1393 {
1394     Error *local_err = NULL;
1395 
1396     if (spapr->htab_fd >= 0) {
1397         return spapr->htab_fd;
1398     }
1399 
1400     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1401     if (spapr->htab_fd < 0) {
1402         error_report_err(local_err);
1403     }
1404 
1405     return spapr->htab_fd;
1406 }
1407 
1408 void close_htab_fd(sPAPRMachineState *spapr)
1409 {
1410     if (spapr->htab_fd >= 0) {
1411         close(spapr->htab_fd);
1412     }
1413     spapr->htab_fd = -1;
1414 }
1415 
1416 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1417 {
1418     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1419 
1420     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1421 }
1422 
1423 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1424 {
1425     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1426 
1427     assert(kvm_enabled());
1428 
1429     if (!spapr->htab) {
1430         return 0;
1431     }
1432 
1433     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1434 }
1435 
1436 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1437                                                 hwaddr ptex, int n)
1438 {
1439     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1440     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1441 
1442     if (!spapr->htab) {
1443         /*
1444          * HTAB is controlled by KVM. Fetch into temporary buffer
1445          */
1446         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1447         kvmppc_read_hptes(hptes, ptex, n);
1448         return hptes;
1449     }
1450 
1451     /*
1452      * HTAB is controlled by QEMU. Just point to the internally
1453      * accessible PTEG.
1454      */
1455     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1456 }
1457 
1458 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1459                               const ppc_hash_pte64_t *hptes,
1460                               hwaddr ptex, int n)
1461 {
1462     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1463 
1464     if (!spapr->htab) {
1465         g_free((void *)hptes);
1466     }
1467 
1468     /* Nothing to do for qemu managed HPT */
1469 }
1470 
1471 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1472                              uint64_t pte0, uint64_t pte1)
1473 {
1474     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1475     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1476 
1477     if (!spapr->htab) {
1478         kvmppc_write_hpte(ptex, pte0, pte1);
1479     } else {
1480         stq_p(spapr->htab + offset, pte0);
1481         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1482     }
1483 }
1484 
1485 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1486 {
1487     int shift;
1488 
1489     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1490      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1491      * that's much more than is needed for Linux guests */
1492     shift = ctz64(pow2ceil(ramsize)) - 7;
1493     shift = MAX(shift, 18); /* Minimum architected size */
1494     shift = MIN(shift, 46); /* Maximum architected size */
1495     return shift;
1496 }
1497 
1498 void spapr_free_hpt(sPAPRMachineState *spapr)
1499 {
1500     g_free(spapr->htab);
1501     spapr->htab = NULL;
1502     spapr->htab_shift = 0;
1503     close_htab_fd(spapr);
1504 }
1505 
1506 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1507                           Error **errp)
1508 {
1509     long rc;
1510 
1511     /* Clean up any HPT info from a previous boot */
1512     spapr_free_hpt(spapr);
1513 
1514     rc = kvmppc_reset_htab(shift);
1515     if (rc < 0) {
1516         /* kernel-side HPT needed, but couldn't allocate one */
1517         error_setg_errno(errp, errno,
1518                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1519                          shift);
1520         /* This is almost certainly fatal, but if the caller really
1521          * wants to carry on with shift == 0, it's welcome to try */
1522     } else if (rc > 0) {
1523         /* kernel-side HPT allocated */
1524         if (rc != shift) {
1525             error_setg(errp,
1526                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1527                        shift, rc);
1528         }
1529 
1530         spapr->htab_shift = shift;
1531         spapr->htab = NULL;
1532     } else {
1533         /* kernel-side HPT not needed, allocate in userspace instead */
1534         size_t size = 1ULL << shift;
1535         int i;
1536 
1537         spapr->htab = qemu_memalign(size, size);
1538         if (!spapr->htab) {
1539             error_setg_errno(errp, errno,
1540                              "Could not allocate HPT of order %d", shift);
1541             return;
1542         }
1543 
1544         memset(spapr->htab, 0, size);
1545         spapr->htab_shift = shift;
1546 
1547         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1548             DIRTY_HPTE(HPTE(spapr->htab, i));
1549         }
1550     }
1551     /* We're setting up a hash table, so that means we're not radix */
1552     spapr->patb_entry = 0;
1553 }
1554 
1555 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1556 {
1557     int hpt_shift;
1558 
1559     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1560         || (spapr->cas_reboot
1561             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1562         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1563     } else {
1564         uint64_t current_ram_size;
1565 
1566         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1567         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1568     }
1569     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1570 
1571     if (spapr->vrma_adjust) {
1572         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1573                                           spapr->htab_shift);
1574     }
1575 }
1576 
1577 static int spapr_reset_drcs(Object *child, void *opaque)
1578 {
1579     sPAPRDRConnector *drc =
1580         (sPAPRDRConnector *) object_dynamic_cast(child,
1581                                                  TYPE_SPAPR_DR_CONNECTOR);
1582 
1583     if (drc) {
1584         spapr_drc_reset(drc);
1585     }
1586 
1587     return 0;
1588 }
1589 
1590 static void spapr_machine_reset(void)
1591 {
1592     MachineState *machine = MACHINE(qdev_get_machine());
1593     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1594     PowerPCCPU *first_ppc_cpu;
1595     uint32_t rtas_limit;
1596     hwaddr rtas_addr, fdt_addr;
1597     void *fdt;
1598     int rc;
1599 
1600     spapr_caps_apply(spapr);
1601 
1602     first_ppc_cpu = POWERPC_CPU(first_cpu);
1603     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1604         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1605                               spapr->max_compat_pvr)) {
1606         /* If using KVM with radix mode available, VCPUs can be started
1607          * without a HPT because KVM will start them in radix mode.
1608          * Set the GR bit in PATB so that we know there is no HPT. */
1609         spapr->patb_entry = PATBE1_GR;
1610     } else {
1611         spapr_setup_hpt_and_vrma(spapr);
1612     }
1613 
1614     /* if this reset wasn't generated by CAS, we should reset our
1615      * negotiated options and start from scratch */
1616     if (!spapr->cas_reboot) {
1617         spapr_ovec_cleanup(spapr->ov5_cas);
1618         spapr->ov5_cas = spapr_ovec_new();
1619 
1620         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1621     }
1622 
1623     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1624         spapr_irq_msi_reset(spapr);
1625     }
1626 
1627     qemu_devices_reset();
1628 
1629     /*
1630      * This is fixing some of the default configuration of the XIVE
1631      * devices. To be called after the reset of the machine devices.
1632      */
1633     spapr_irq_reset(spapr, &error_fatal);
1634 
1635     /* DRC reset may cause a device to be unplugged. This will cause troubles
1636      * if this device is used by another device (eg, a running vhost backend
1637      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1638      * situations, we reset DRCs after all devices have been reset.
1639      */
1640     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1641 
1642     spapr_clear_pending_events(spapr);
1643 
1644     /*
1645      * We place the device tree and RTAS just below either the top of the RMA,
1646      * or just below 2GB, whichever is lowere, so that it can be
1647      * processed with 32-bit real mode code if necessary
1648      */
1649     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1650     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1651     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1652 
1653     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1654 
1655     spapr_load_rtas(spapr, fdt, rtas_addr);
1656 
1657     rc = fdt_pack(fdt);
1658 
1659     /* Should only fail if we've built a corrupted tree */
1660     assert(rc == 0);
1661 
1662     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1663         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1664                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1665         exit(1);
1666     }
1667 
1668     /* Load the fdt */
1669     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1670     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1671     g_free(fdt);
1672 
1673     /* Set up the entry state */
1674     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1675     first_ppc_cpu->env.gpr[5] = 0;
1676 
1677     spapr->cas_reboot = false;
1678 }
1679 
1680 static void spapr_create_nvram(sPAPRMachineState *spapr)
1681 {
1682     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1683     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1684 
1685     if (dinfo) {
1686         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1687                             &error_fatal);
1688     }
1689 
1690     qdev_init_nofail(dev);
1691 
1692     spapr->nvram = (struct sPAPRNVRAM *)dev;
1693 }
1694 
1695 static void spapr_rtc_create(sPAPRMachineState *spapr)
1696 {
1697     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1698     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1699                               &error_fatal);
1700     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1701                               &error_fatal);
1702     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1703                               "date", &error_fatal);
1704 }
1705 
1706 /* Returns whether we want to use VGA or not */
1707 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1708 {
1709     switch (vga_interface_type) {
1710     case VGA_NONE:
1711         return false;
1712     case VGA_DEVICE:
1713         return true;
1714     case VGA_STD:
1715     case VGA_VIRTIO:
1716         return pci_vga_init(pci_bus) != NULL;
1717     default:
1718         error_setg(errp,
1719                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1720         return false;
1721     }
1722 }
1723 
1724 static int spapr_pre_load(void *opaque)
1725 {
1726     int rc;
1727 
1728     rc = spapr_caps_pre_load(opaque);
1729     if (rc) {
1730         return rc;
1731     }
1732 
1733     return 0;
1734 }
1735 
1736 static int spapr_post_load(void *opaque, int version_id)
1737 {
1738     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1739     int err = 0;
1740 
1741     err = spapr_caps_post_migration(spapr);
1742     if (err) {
1743         return err;
1744     }
1745 
1746     /* In earlier versions, there was no separate qdev for the PAPR
1747      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1748      * So when migrating from those versions, poke the incoming offset
1749      * value into the RTC device */
1750     if (version_id < 3) {
1751         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1752     }
1753 
1754     if (kvm_enabled() && spapr->patb_entry) {
1755         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1756         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1757         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1758 
1759         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1760         if (err) {
1761             error_report("Process table config unsupported by the host");
1762             return -EINVAL;
1763         }
1764     }
1765 
1766     err = spapr_irq_post_load(spapr, version_id);
1767     if (err) {
1768         return err;
1769     }
1770 
1771     return err;
1772 }
1773 
1774 static int spapr_pre_save(void *opaque)
1775 {
1776     int rc;
1777 
1778     rc = spapr_caps_pre_save(opaque);
1779     if (rc) {
1780         return rc;
1781     }
1782 
1783     return 0;
1784 }
1785 
1786 static bool version_before_3(void *opaque, int version_id)
1787 {
1788     return version_id < 3;
1789 }
1790 
1791 static bool spapr_pending_events_needed(void *opaque)
1792 {
1793     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1794     return !QTAILQ_EMPTY(&spapr->pending_events);
1795 }
1796 
1797 static const VMStateDescription vmstate_spapr_event_entry = {
1798     .name = "spapr_event_log_entry",
1799     .version_id = 1,
1800     .minimum_version_id = 1,
1801     .fields = (VMStateField[]) {
1802         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1803         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1804         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1805                                      NULL, extended_length),
1806         VMSTATE_END_OF_LIST()
1807     },
1808 };
1809 
1810 static const VMStateDescription vmstate_spapr_pending_events = {
1811     .name = "spapr_pending_events",
1812     .version_id = 1,
1813     .minimum_version_id = 1,
1814     .needed = spapr_pending_events_needed,
1815     .fields = (VMStateField[]) {
1816         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1817                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1818         VMSTATE_END_OF_LIST()
1819     },
1820 };
1821 
1822 static bool spapr_ov5_cas_needed(void *opaque)
1823 {
1824     sPAPRMachineState *spapr = opaque;
1825     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1826     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1827     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1828     bool cas_needed;
1829 
1830     /* Prior to the introduction of sPAPROptionVector, we had two option
1831      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1832      * Both of these options encode machine topology into the device-tree
1833      * in such a way that the now-booted OS should still be able to interact
1834      * appropriately with QEMU regardless of what options were actually
1835      * negotiatied on the source side.
1836      *
1837      * As such, we can avoid migrating the CAS-negotiated options if these
1838      * are the only options available on the current machine/platform.
1839      * Since these are the only options available for pseries-2.7 and
1840      * earlier, this allows us to maintain old->new/new->old migration
1841      * compatibility.
1842      *
1843      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1844      * via default pseries-2.8 machines and explicit command-line parameters.
1845      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1846      * of the actual CAS-negotiated values to continue working properly. For
1847      * example, availability of memory unplug depends on knowing whether
1848      * OV5_HP_EVT was negotiated via CAS.
1849      *
1850      * Thus, for any cases where the set of available CAS-negotiatable
1851      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1852      * include the CAS-negotiated options in the migration stream, unless
1853      * if they affect boot time behaviour only.
1854      */
1855     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1856     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1857     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1858 
1859     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1860      * the mask itself since in the future it's possible "legacy" bits may be
1861      * removed via machine options, which could generate a false positive
1862      * that breaks migration.
1863      */
1864     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1865     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1866 
1867     spapr_ovec_cleanup(ov5_mask);
1868     spapr_ovec_cleanup(ov5_legacy);
1869     spapr_ovec_cleanup(ov5_removed);
1870 
1871     return cas_needed;
1872 }
1873 
1874 static const VMStateDescription vmstate_spapr_ov5_cas = {
1875     .name = "spapr_option_vector_ov5_cas",
1876     .version_id = 1,
1877     .minimum_version_id = 1,
1878     .needed = spapr_ov5_cas_needed,
1879     .fields = (VMStateField[]) {
1880         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1881                                  vmstate_spapr_ovec, sPAPROptionVector),
1882         VMSTATE_END_OF_LIST()
1883     },
1884 };
1885 
1886 static bool spapr_patb_entry_needed(void *opaque)
1887 {
1888     sPAPRMachineState *spapr = opaque;
1889 
1890     return !!spapr->patb_entry;
1891 }
1892 
1893 static const VMStateDescription vmstate_spapr_patb_entry = {
1894     .name = "spapr_patb_entry",
1895     .version_id = 1,
1896     .minimum_version_id = 1,
1897     .needed = spapr_patb_entry_needed,
1898     .fields = (VMStateField[]) {
1899         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1900         VMSTATE_END_OF_LIST()
1901     },
1902 };
1903 
1904 static bool spapr_irq_map_needed(void *opaque)
1905 {
1906     sPAPRMachineState *spapr = opaque;
1907 
1908     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1909 }
1910 
1911 static const VMStateDescription vmstate_spapr_irq_map = {
1912     .name = "spapr_irq_map",
1913     .version_id = 1,
1914     .minimum_version_id = 1,
1915     .needed = spapr_irq_map_needed,
1916     .fields = (VMStateField[]) {
1917         VMSTATE_BITMAP(irq_map, sPAPRMachineState, 0, irq_map_nr),
1918         VMSTATE_END_OF_LIST()
1919     },
1920 };
1921 
1922 static const VMStateDescription vmstate_spapr = {
1923     .name = "spapr",
1924     .version_id = 3,
1925     .minimum_version_id = 1,
1926     .pre_load = spapr_pre_load,
1927     .post_load = spapr_post_load,
1928     .pre_save = spapr_pre_save,
1929     .fields = (VMStateField[]) {
1930         /* used to be @next_irq */
1931         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1932 
1933         /* RTC offset */
1934         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1935 
1936         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1937         VMSTATE_END_OF_LIST()
1938     },
1939     .subsections = (const VMStateDescription*[]) {
1940         &vmstate_spapr_ov5_cas,
1941         &vmstate_spapr_patb_entry,
1942         &vmstate_spapr_pending_events,
1943         &vmstate_spapr_cap_htm,
1944         &vmstate_spapr_cap_vsx,
1945         &vmstate_spapr_cap_dfp,
1946         &vmstate_spapr_cap_cfpc,
1947         &vmstate_spapr_cap_sbbc,
1948         &vmstate_spapr_cap_ibs,
1949         &vmstate_spapr_irq_map,
1950         &vmstate_spapr_cap_nested_kvm_hv,
1951         NULL
1952     }
1953 };
1954 
1955 static int htab_save_setup(QEMUFile *f, void *opaque)
1956 {
1957     sPAPRMachineState *spapr = opaque;
1958 
1959     /* "Iteration" header */
1960     if (!spapr->htab_shift) {
1961         qemu_put_be32(f, -1);
1962     } else {
1963         qemu_put_be32(f, spapr->htab_shift);
1964     }
1965 
1966     if (spapr->htab) {
1967         spapr->htab_save_index = 0;
1968         spapr->htab_first_pass = true;
1969     } else {
1970         if (spapr->htab_shift) {
1971             assert(kvm_enabled());
1972         }
1973     }
1974 
1975 
1976     return 0;
1977 }
1978 
1979 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1980                             int chunkstart, int n_valid, int n_invalid)
1981 {
1982     qemu_put_be32(f, chunkstart);
1983     qemu_put_be16(f, n_valid);
1984     qemu_put_be16(f, n_invalid);
1985     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1986                     HASH_PTE_SIZE_64 * n_valid);
1987 }
1988 
1989 static void htab_save_end_marker(QEMUFile *f)
1990 {
1991     qemu_put_be32(f, 0);
1992     qemu_put_be16(f, 0);
1993     qemu_put_be16(f, 0);
1994 }
1995 
1996 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1997                                  int64_t max_ns)
1998 {
1999     bool has_timeout = max_ns != -1;
2000     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2001     int index = spapr->htab_save_index;
2002     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2003 
2004     assert(spapr->htab_first_pass);
2005 
2006     do {
2007         int chunkstart;
2008 
2009         /* Consume invalid HPTEs */
2010         while ((index < htabslots)
2011                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2012             CLEAN_HPTE(HPTE(spapr->htab, index));
2013             index++;
2014         }
2015 
2016         /* Consume valid HPTEs */
2017         chunkstart = index;
2018         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2019                && HPTE_VALID(HPTE(spapr->htab, index))) {
2020             CLEAN_HPTE(HPTE(spapr->htab, index));
2021             index++;
2022         }
2023 
2024         if (index > chunkstart) {
2025             int n_valid = index - chunkstart;
2026 
2027             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2028 
2029             if (has_timeout &&
2030                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2031                 break;
2032             }
2033         }
2034     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2035 
2036     if (index >= htabslots) {
2037         assert(index == htabslots);
2038         index = 0;
2039         spapr->htab_first_pass = false;
2040     }
2041     spapr->htab_save_index = index;
2042 }
2043 
2044 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2045                                 int64_t max_ns)
2046 {
2047     bool final = max_ns < 0;
2048     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2049     int examined = 0, sent = 0;
2050     int index = spapr->htab_save_index;
2051     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2052 
2053     assert(!spapr->htab_first_pass);
2054 
2055     do {
2056         int chunkstart, invalidstart;
2057 
2058         /* Consume non-dirty HPTEs */
2059         while ((index < htabslots)
2060                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2061             index++;
2062             examined++;
2063         }
2064 
2065         chunkstart = index;
2066         /* Consume valid dirty HPTEs */
2067         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2068                && HPTE_DIRTY(HPTE(spapr->htab, index))
2069                && HPTE_VALID(HPTE(spapr->htab, index))) {
2070             CLEAN_HPTE(HPTE(spapr->htab, index));
2071             index++;
2072             examined++;
2073         }
2074 
2075         invalidstart = index;
2076         /* Consume invalid dirty HPTEs */
2077         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2078                && HPTE_DIRTY(HPTE(spapr->htab, index))
2079                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2080             CLEAN_HPTE(HPTE(spapr->htab, index));
2081             index++;
2082             examined++;
2083         }
2084 
2085         if (index > chunkstart) {
2086             int n_valid = invalidstart - chunkstart;
2087             int n_invalid = index - invalidstart;
2088 
2089             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2090             sent += index - chunkstart;
2091 
2092             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2093                 break;
2094             }
2095         }
2096 
2097         if (examined >= htabslots) {
2098             break;
2099         }
2100 
2101         if (index >= htabslots) {
2102             assert(index == htabslots);
2103             index = 0;
2104         }
2105     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2106 
2107     if (index >= htabslots) {
2108         assert(index == htabslots);
2109         index = 0;
2110     }
2111 
2112     spapr->htab_save_index = index;
2113 
2114     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2115 }
2116 
2117 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2118 #define MAX_KVM_BUF_SIZE    2048
2119 
2120 static int htab_save_iterate(QEMUFile *f, void *opaque)
2121 {
2122     sPAPRMachineState *spapr = opaque;
2123     int fd;
2124     int rc = 0;
2125 
2126     /* Iteration header */
2127     if (!spapr->htab_shift) {
2128         qemu_put_be32(f, -1);
2129         return 1;
2130     } else {
2131         qemu_put_be32(f, 0);
2132     }
2133 
2134     if (!spapr->htab) {
2135         assert(kvm_enabled());
2136 
2137         fd = get_htab_fd(spapr);
2138         if (fd < 0) {
2139             return fd;
2140         }
2141 
2142         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2143         if (rc < 0) {
2144             return rc;
2145         }
2146     } else  if (spapr->htab_first_pass) {
2147         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2148     } else {
2149         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2150     }
2151 
2152     htab_save_end_marker(f);
2153 
2154     return rc;
2155 }
2156 
2157 static int htab_save_complete(QEMUFile *f, void *opaque)
2158 {
2159     sPAPRMachineState *spapr = opaque;
2160     int fd;
2161 
2162     /* Iteration header */
2163     if (!spapr->htab_shift) {
2164         qemu_put_be32(f, -1);
2165         return 0;
2166     } else {
2167         qemu_put_be32(f, 0);
2168     }
2169 
2170     if (!spapr->htab) {
2171         int rc;
2172 
2173         assert(kvm_enabled());
2174 
2175         fd = get_htab_fd(spapr);
2176         if (fd < 0) {
2177             return fd;
2178         }
2179 
2180         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2181         if (rc < 0) {
2182             return rc;
2183         }
2184     } else {
2185         if (spapr->htab_first_pass) {
2186             htab_save_first_pass(f, spapr, -1);
2187         }
2188         htab_save_later_pass(f, spapr, -1);
2189     }
2190 
2191     /* End marker */
2192     htab_save_end_marker(f);
2193 
2194     return 0;
2195 }
2196 
2197 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2198 {
2199     sPAPRMachineState *spapr = opaque;
2200     uint32_t section_hdr;
2201     int fd = -1;
2202     Error *local_err = NULL;
2203 
2204     if (version_id < 1 || version_id > 1) {
2205         error_report("htab_load() bad version");
2206         return -EINVAL;
2207     }
2208 
2209     section_hdr = qemu_get_be32(f);
2210 
2211     if (section_hdr == -1) {
2212         spapr_free_hpt(spapr);
2213         return 0;
2214     }
2215 
2216     if (section_hdr) {
2217         /* First section gives the htab size */
2218         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2219         if (local_err) {
2220             error_report_err(local_err);
2221             return -EINVAL;
2222         }
2223         return 0;
2224     }
2225 
2226     if (!spapr->htab) {
2227         assert(kvm_enabled());
2228 
2229         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2230         if (fd < 0) {
2231             error_report_err(local_err);
2232             return fd;
2233         }
2234     }
2235 
2236     while (true) {
2237         uint32_t index;
2238         uint16_t n_valid, n_invalid;
2239 
2240         index = qemu_get_be32(f);
2241         n_valid = qemu_get_be16(f);
2242         n_invalid = qemu_get_be16(f);
2243 
2244         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2245             /* End of Stream */
2246             break;
2247         }
2248 
2249         if ((index + n_valid + n_invalid) >
2250             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2251             /* Bad index in stream */
2252             error_report(
2253                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2254                 index, n_valid, n_invalid, spapr->htab_shift);
2255             return -EINVAL;
2256         }
2257 
2258         if (spapr->htab) {
2259             if (n_valid) {
2260                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2261                                 HASH_PTE_SIZE_64 * n_valid);
2262             }
2263             if (n_invalid) {
2264                 memset(HPTE(spapr->htab, index + n_valid), 0,
2265                        HASH_PTE_SIZE_64 * n_invalid);
2266             }
2267         } else {
2268             int rc;
2269 
2270             assert(fd >= 0);
2271 
2272             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2273             if (rc < 0) {
2274                 return rc;
2275             }
2276         }
2277     }
2278 
2279     if (!spapr->htab) {
2280         assert(fd >= 0);
2281         close(fd);
2282     }
2283 
2284     return 0;
2285 }
2286 
2287 static void htab_save_cleanup(void *opaque)
2288 {
2289     sPAPRMachineState *spapr = opaque;
2290 
2291     close_htab_fd(spapr);
2292 }
2293 
2294 static SaveVMHandlers savevm_htab_handlers = {
2295     .save_setup = htab_save_setup,
2296     .save_live_iterate = htab_save_iterate,
2297     .save_live_complete_precopy = htab_save_complete,
2298     .save_cleanup = htab_save_cleanup,
2299     .load_state = htab_load,
2300 };
2301 
2302 static void spapr_boot_set(void *opaque, const char *boot_device,
2303                            Error **errp)
2304 {
2305     MachineState *machine = MACHINE(opaque);
2306     machine->boot_order = g_strdup(boot_device);
2307 }
2308 
2309 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2310 {
2311     MachineState *machine = MACHINE(spapr);
2312     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2313     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2314     int i;
2315 
2316     for (i = 0; i < nr_lmbs; i++) {
2317         uint64_t addr;
2318 
2319         addr = i * lmb_size + machine->device_memory->base;
2320         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2321                                addr / lmb_size);
2322     }
2323 }
2324 
2325 /*
2326  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2327  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2328  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2329  */
2330 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2331 {
2332     int i;
2333 
2334     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2335         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2336                    " is not aligned to %" PRIu64 " MiB",
2337                    machine->ram_size,
2338                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2339         return;
2340     }
2341 
2342     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2343         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2344                    " is not aligned to %" PRIu64 " MiB",
2345                    machine->ram_size,
2346                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2347         return;
2348     }
2349 
2350     for (i = 0; i < nb_numa_nodes; i++) {
2351         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2352             error_setg(errp,
2353                        "Node %d memory size 0x%" PRIx64
2354                        " is not aligned to %" PRIu64 " MiB",
2355                        i, numa_info[i].node_mem,
2356                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2357             return;
2358         }
2359     }
2360 }
2361 
2362 /* find cpu slot in machine->possible_cpus by core_id */
2363 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2364 {
2365     int index = id / smp_threads;
2366 
2367     if (index >= ms->possible_cpus->len) {
2368         return NULL;
2369     }
2370     if (idx) {
2371         *idx = index;
2372     }
2373     return &ms->possible_cpus->cpus[index];
2374 }
2375 
2376 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2377 {
2378     Error *local_err = NULL;
2379     bool vsmt_user = !!spapr->vsmt;
2380     int kvm_smt = kvmppc_smt_threads();
2381     int ret;
2382 
2383     if (!kvm_enabled() && (smp_threads > 1)) {
2384         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2385                      "on a pseries machine");
2386         goto out;
2387     }
2388     if (!is_power_of_2(smp_threads)) {
2389         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2390                      "machine because it must be a power of 2", smp_threads);
2391         goto out;
2392     }
2393 
2394     /* Detemine the VSMT mode to use: */
2395     if (vsmt_user) {
2396         if (spapr->vsmt < smp_threads) {
2397             error_setg(&local_err, "Cannot support VSMT mode %d"
2398                          " because it must be >= threads/core (%d)",
2399                          spapr->vsmt, smp_threads);
2400             goto out;
2401         }
2402         /* In this case, spapr->vsmt has been set by the command line */
2403     } else {
2404         /*
2405          * Default VSMT value is tricky, because we need it to be as
2406          * consistent as possible (for migration), but this requires
2407          * changing it for at least some existing cases.  We pick 8 as
2408          * the value that we'd get with KVM on POWER8, the
2409          * overwhelmingly common case in production systems.
2410          */
2411         spapr->vsmt = MAX(8, smp_threads);
2412     }
2413 
2414     /* KVM: If necessary, set the SMT mode: */
2415     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2416         ret = kvmppc_set_smt_threads(spapr->vsmt);
2417         if (ret) {
2418             /* Looks like KVM isn't able to change VSMT mode */
2419             error_setg(&local_err,
2420                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2421                        spapr->vsmt, ret);
2422             /* We can live with that if the default one is big enough
2423              * for the number of threads, and a submultiple of the one
2424              * we want.  In this case we'll waste some vcpu ids, but
2425              * behaviour will be correct */
2426             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2427                 warn_report_err(local_err);
2428                 local_err = NULL;
2429                 goto out;
2430             } else {
2431                 if (!vsmt_user) {
2432                     error_append_hint(&local_err,
2433                                       "On PPC, a VM with %d threads/core"
2434                                       " on a host with %d threads/core"
2435                                       " requires the use of VSMT mode %d.\n",
2436                                       smp_threads, kvm_smt, spapr->vsmt);
2437                 }
2438                 kvmppc_hint_smt_possible(&local_err);
2439                 goto out;
2440             }
2441         }
2442     }
2443     /* else TCG: nothing to do currently */
2444 out:
2445     error_propagate(errp, local_err);
2446 }
2447 
2448 static void spapr_init_cpus(sPAPRMachineState *spapr)
2449 {
2450     MachineState *machine = MACHINE(spapr);
2451     MachineClass *mc = MACHINE_GET_CLASS(machine);
2452     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2453     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2454     const CPUArchIdList *possible_cpus;
2455     int boot_cores_nr = smp_cpus / smp_threads;
2456     int i;
2457 
2458     possible_cpus = mc->possible_cpu_arch_ids(machine);
2459     if (mc->has_hotpluggable_cpus) {
2460         if (smp_cpus % smp_threads) {
2461             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2462                          smp_cpus, smp_threads);
2463             exit(1);
2464         }
2465         if (max_cpus % smp_threads) {
2466             error_report("max_cpus (%u) must be multiple of threads (%u)",
2467                          max_cpus, smp_threads);
2468             exit(1);
2469         }
2470     } else {
2471         if (max_cpus != smp_cpus) {
2472             error_report("This machine version does not support CPU hotplug");
2473             exit(1);
2474         }
2475         boot_cores_nr = possible_cpus->len;
2476     }
2477 
2478     if (smc->pre_2_10_has_unused_icps) {
2479         int i;
2480 
2481         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2482             /* Dummy entries get deregistered when real ICPState objects
2483              * are registered during CPU core hotplug.
2484              */
2485             pre_2_10_vmstate_register_dummy_icp(i);
2486         }
2487     }
2488 
2489     for (i = 0; i < possible_cpus->len; i++) {
2490         int core_id = i * smp_threads;
2491 
2492         if (mc->has_hotpluggable_cpus) {
2493             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2494                                    spapr_vcpu_id(spapr, core_id));
2495         }
2496 
2497         if (i < boot_cores_nr) {
2498             Object *core  = object_new(type);
2499             int nr_threads = smp_threads;
2500 
2501             /* Handle the partially filled core for older machine types */
2502             if ((i + 1) * smp_threads >= smp_cpus) {
2503                 nr_threads = smp_cpus - i * smp_threads;
2504             }
2505 
2506             object_property_set_int(core, nr_threads, "nr-threads",
2507                                     &error_fatal);
2508             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2509                                     &error_fatal);
2510             object_property_set_bool(core, true, "realized", &error_fatal);
2511 
2512             object_unref(core);
2513         }
2514     }
2515 }
2516 
2517 /* pSeries LPAR / sPAPR hardware init */
2518 static void spapr_machine_init(MachineState *machine)
2519 {
2520     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2521     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2522     const char *kernel_filename = machine->kernel_filename;
2523     const char *initrd_filename = machine->initrd_filename;
2524     PCIHostState *phb;
2525     int i;
2526     MemoryRegion *sysmem = get_system_memory();
2527     MemoryRegion *ram = g_new(MemoryRegion, 1);
2528     hwaddr node0_size = spapr_node0_size(machine);
2529     long load_limit, fw_size;
2530     char *filename;
2531     Error *resize_hpt_err = NULL;
2532 
2533     msi_nonbroken = true;
2534 
2535     QLIST_INIT(&spapr->phbs);
2536     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2537 
2538     /* Determine capabilities to run with */
2539     spapr_caps_init(spapr);
2540 
2541     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2542     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2543         /*
2544          * If the user explicitly requested a mode we should either
2545          * supply it, or fail completely (which we do below).  But if
2546          * it's not set explicitly, we reset our mode to something
2547          * that works
2548          */
2549         if (resize_hpt_err) {
2550             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2551             error_free(resize_hpt_err);
2552             resize_hpt_err = NULL;
2553         } else {
2554             spapr->resize_hpt = smc->resize_hpt_default;
2555         }
2556     }
2557 
2558     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2559 
2560     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2561         /*
2562          * User requested HPT resize, but this host can't supply it.  Bail out
2563          */
2564         error_report_err(resize_hpt_err);
2565         exit(1);
2566     }
2567 
2568     spapr->rma_size = node0_size;
2569 
2570     /* With KVM, we don't actually know whether KVM supports an
2571      * unbounded RMA (PR KVM) or is limited by the hash table size
2572      * (HV KVM using VRMA), so we always assume the latter
2573      *
2574      * In that case, we also limit the initial allocations for RTAS
2575      * etc... to 256M since we have no way to know what the VRMA size
2576      * is going to be as it depends on the size of the hash table
2577      * which isn't determined yet.
2578      */
2579     if (kvm_enabled()) {
2580         spapr->vrma_adjust = 1;
2581         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2582     }
2583 
2584     /* Actually we don't support unbounded RMA anymore since we added
2585      * proper emulation of HV mode. The max we can get is 16G which
2586      * also happens to be what we configure for PAPR mode so make sure
2587      * we don't do anything bigger than that
2588      */
2589     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2590 
2591     if (spapr->rma_size > node0_size) {
2592         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2593                      spapr->rma_size);
2594         exit(1);
2595     }
2596 
2597     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2598     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2599 
2600     /*
2601      * VSMT must be set in order to be able to compute VCPU ids, ie to
2602      * call spapr_max_server_number() or spapr_vcpu_id().
2603      */
2604     spapr_set_vsmt_mode(spapr, &error_fatal);
2605 
2606     /* Set up Interrupt Controller before we create the VCPUs */
2607     spapr_irq_init(spapr, &error_fatal);
2608 
2609     /* Set up containers for ibm,client-architecture-support negotiated options
2610      */
2611     spapr->ov5 = spapr_ovec_new();
2612     spapr->ov5_cas = spapr_ovec_new();
2613 
2614     if (smc->dr_lmb_enabled) {
2615         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2616         spapr_validate_node_memory(machine, &error_fatal);
2617     }
2618 
2619     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2620 
2621     /* advertise support for dedicated HP event source to guests */
2622     if (spapr->use_hotplug_event_source) {
2623         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2624     }
2625 
2626     /* advertise support for HPT resizing */
2627     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2628         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2629     }
2630 
2631     /* advertise support for ibm,dyamic-memory-v2 */
2632     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2633 
2634     /* advertise XIVE on POWER9 machines */
2635     if (spapr->irq->ov5 & SPAPR_OV5_XIVE_EXPLOIT) {
2636         if (ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00,
2637                                   0, spapr->max_compat_pvr)) {
2638             spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2639         } else {
2640             error_report("XIVE-only machines require a POWER9 CPU");
2641             exit(1);
2642         }
2643     }
2644 
2645     /* init CPUs */
2646     spapr_init_cpus(spapr);
2647 
2648     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2649         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2650                               spapr->max_compat_pvr)) {
2651         /* KVM and TCG always allow GTSE with radix... */
2652         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2653     }
2654     /* ... but not with hash (currently). */
2655 
2656     if (kvm_enabled()) {
2657         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2658         kvmppc_enable_logical_ci_hcalls();
2659         kvmppc_enable_set_mode_hcall();
2660 
2661         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2662         kvmppc_enable_clear_ref_mod_hcalls();
2663     }
2664 
2665     /* allocate RAM */
2666     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2667                                          machine->ram_size);
2668     memory_region_add_subregion(sysmem, 0, ram);
2669 
2670     /* always allocate the device memory information */
2671     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2672 
2673     /* initialize hotplug memory address space */
2674     if (machine->ram_size < machine->maxram_size) {
2675         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2676         /*
2677          * Limit the number of hotpluggable memory slots to half the number
2678          * slots that KVM supports, leaving the other half for PCI and other
2679          * devices. However ensure that number of slots doesn't drop below 32.
2680          */
2681         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2682                            SPAPR_MAX_RAM_SLOTS;
2683 
2684         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2685             max_memslots = SPAPR_MAX_RAM_SLOTS;
2686         }
2687         if (machine->ram_slots > max_memslots) {
2688             error_report("Specified number of memory slots %"
2689                          PRIu64" exceeds max supported %d",
2690                          machine->ram_slots, max_memslots);
2691             exit(1);
2692         }
2693 
2694         machine->device_memory->base = ROUND_UP(machine->ram_size,
2695                                                 SPAPR_DEVICE_MEM_ALIGN);
2696         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2697                            "device-memory", device_mem_size);
2698         memory_region_add_subregion(sysmem, machine->device_memory->base,
2699                                     &machine->device_memory->mr);
2700     }
2701 
2702     if (smc->dr_lmb_enabled) {
2703         spapr_create_lmb_dr_connectors(spapr);
2704     }
2705 
2706     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2707     if (!filename) {
2708         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2709         exit(1);
2710     }
2711     spapr->rtas_size = get_image_size(filename);
2712     if (spapr->rtas_size < 0) {
2713         error_report("Could not get size of LPAR rtas '%s'", filename);
2714         exit(1);
2715     }
2716     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2717     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2718         error_report("Could not load LPAR rtas '%s'", filename);
2719         exit(1);
2720     }
2721     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2722         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2723                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2724         exit(1);
2725     }
2726     g_free(filename);
2727 
2728     /* Set up RTAS event infrastructure */
2729     spapr_events_init(spapr);
2730 
2731     /* Set up the RTC RTAS interfaces */
2732     spapr_rtc_create(spapr);
2733 
2734     /* Set up VIO bus */
2735     spapr->vio_bus = spapr_vio_bus_init();
2736 
2737     for (i = 0; i < serial_max_hds(); i++) {
2738         if (serial_hd(i)) {
2739             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2740         }
2741     }
2742 
2743     /* We always have at least the nvram device on VIO */
2744     spapr_create_nvram(spapr);
2745 
2746     /* Set up PCI */
2747     spapr_pci_rtas_init();
2748 
2749     phb = spapr_create_phb(spapr, 0);
2750 
2751     for (i = 0; i < nb_nics; i++) {
2752         NICInfo *nd = &nd_table[i];
2753 
2754         if (!nd->model) {
2755             nd->model = g_strdup("spapr-vlan");
2756         }
2757 
2758         if (g_str_equal(nd->model, "spapr-vlan") ||
2759             g_str_equal(nd->model, "ibmveth")) {
2760             spapr_vlan_create(spapr->vio_bus, nd);
2761         } else {
2762             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2763         }
2764     }
2765 
2766     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2767         spapr_vscsi_create(spapr->vio_bus);
2768     }
2769 
2770     /* Graphics */
2771     if (spapr_vga_init(phb->bus, &error_fatal)) {
2772         spapr->has_graphics = true;
2773         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2774     }
2775 
2776     if (machine->usb) {
2777         if (smc->use_ohci_by_default) {
2778             pci_create_simple(phb->bus, -1, "pci-ohci");
2779         } else {
2780             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2781         }
2782 
2783         if (spapr->has_graphics) {
2784             USBBus *usb_bus = usb_bus_find(-1);
2785 
2786             usb_create_simple(usb_bus, "usb-kbd");
2787             usb_create_simple(usb_bus, "usb-mouse");
2788         }
2789     }
2790 
2791     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2792         error_report(
2793             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2794             MIN_RMA_SLOF);
2795         exit(1);
2796     }
2797 
2798     if (kernel_filename) {
2799         uint64_t lowaddr = 0;
2800 
2801         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2802                                       NULL, NULL, &lowaddr, NULL, 1,
2803                                       PPC_ELF_MACHINE, 0, 0);
2804         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2805             spapr->kernel_size = load_elf(kernel_filename,
2806                                           translate_kernel_address, NULL, NULL,
2807                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2808                                           0, 0);
2809             spapr->kernel_le = spapr->kernel_size > 0;
2810         }
2811         if (spapr->kernel_size < 0) {
2812             error_report("error loading %s: %s", kernel_filename,
2813                          load_elf_strerror(spapr->kernel_size));
2814             exit(1);
2815         }
2816 
2817         /* load initrd */
2818         if (initrd_filename) {
2819             /* Try to locate the initrd in the gap between the kernel
2820              * and the firmware. Add a bit of space just in case
2821              */
2822             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2823                                   + 0x1ffff) & ~0xffff;
2824             spapr->initrd_size = load_image_targphys(initrd_filename,
2825                                                      spapr->initrd_base,
2826                                                      load_limit
2827                                                      - spapr->initrd_base);
2828             if (spapr->initrd_size < 0) {
2829                 error_report("could not load initial ram disk '%s'",
2830                              initrd_filename);
2831                 exit(1);
2832             }
2833         }
2834     }
2835 
2836     if (bios_name == NULL) {
2837         bios_name = FW_FILE_NAME;
2838     }
2839     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2840     if (!filename) {
2841         error_report("Could not find LPAR firmware '%s'", bios_name);
2842         exit(1);
2843     }
2844     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2845     if (fw_size <= 0) {
2846         error_report("Could not load LPAR firmware '%s'", filename);
2847         exit(1);
2848     }
2849     g_free(filename);
2850 
2851     /* FIXME: Should register things through the MachineState's qdev
2852      * interface, this is a legacy from the sPAPREnvironment structure
2853      * which predated MachineState but had a similar function */
2854     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2855     register_savevm_live(NULL, "spapr/htab", -1, 1,
2856                          &savevm_htab_handlers, spapr);
2857 
2858     qemu_register_boot_set(spapr_boot_set, spapr);
2859 
2860     if (kvm_enabled()) {
2861         /* to stop and start vmclock */
2862         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2863                                          &spapr->tb);
2864 
2865         kvmppc_spapr_enable_inkernel_multitce();
2866     }
2867 }
2868 
2869 static int spapr_kvm_type(const char *vm_type)
2870 {
2871     if (!vm_type) {
2872         return 0;
2873     }
2874 
2875     if (!strcmp(vm_type, "HV")) {
2876         return 1;
2877     }
2878 
2879     if (!strcmp(vm_type, "PR")) {
2880         return 2;
2881     }
2882 
2883     error_report("Unknown kvm-type specified '%s'", vm_type);
2884     exit(1);
2885 }
2886 
2887 /*
2888  * Implementation of an interface to adjust firmware path
2889  * for the bootindex property handling.
2890  */
2891 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2892                                    DeviceState *dev)
2893 {
2894 #define CAST(type, obj, name) \
2895     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2896     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2897     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2898     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2899 
2900     if (d) {
2901         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2902         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2903         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2904 
2905         if (spapr) {
2906             /*
2907              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2908              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2909              * in the top 16 bits of the 64-bit LUN
2910              */
2911             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2912             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2913                                    (uint64_t)id << 48);
2914         } else if (virtio) {
2915             /*
2916              * We use SRP luns of the form 01000000 | (target << 8) | lun
2917              * in the top 32 bits of the 64-bit LUN
2918              * Note: the quote above is from SLOF and it is wrong,
2919              * the actual binding is:
2920              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2921              */
2922             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2923             if (d->lun >= 256) {
2924                 /* Use the LUN "flat space addressing method" */
2925                 id |= 0x4000;
2926             }
2927             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2928                                    (uint64_t)id << 32);
2929         } else if (usb) {
2930             /*
2931              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2932              * in the top 32 bits of the 64-bit LUN
2933              */
2934             unsigned usb_port = atoi(usb->port->path);
2935             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2936             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2937                                    (uint64_t)id << 32);
2938         }
2939     }
2940 
2941     /*
2942      * SLOF probes the USB devices, and if it recognizes that the device is a
2943      * storage device, it changes its name to "storage" instead of "usb-host",
2944      * and additionally adds a child node for the SCSI LUN, so the correct
2945      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2946      */
2947     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2948         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2949         if (usb_host_dev_is_scsi_storage(usbdev)) {
2950             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2951         }
2952     }
2953 
2954     if (phb) {
2955         /* Replace "pci" with "pci@800000020000000" */
2956         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2957     }
2958 
2959     if (vsc) {
2960         /* Same logic as virtio above */
2961         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2962         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2963     }
2964 
2965     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2966         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2967         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2968         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2969     }
2970 
2971     return NULL;
2972 }
2973 
2974 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2975 {
2976     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2977 
2978     return g_strdup(spapr->kvm_type);
2979 }
2980 
2981 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2982 {
2983     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2984 
2985     g_free(spapr->kvm_type);
2986     spapr->kvm_type = g_strdup(value);
2987 }
2988 
2989 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2990 {
2991     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2992 
2993     return spapr->use_hotplug_event_source;
2994 }
2995 
2996 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2997                                             Error **errp)
2998 {
2999     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3000 
3001     spapr->use_hotplug_event_source = value;
3002 }
3003 
3004 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3005 {
3006     return true;
3007 }
3008 
3009 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3010 {
3011     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3012 
3013     switch (spapr->resize_hpt) {
3014     case SPAPR_RESIZE_HPT_DEFAULT:
3015         return g_strdup("default");
3016     case SPAPR_RESIZE_HPT_DISABLED:
3017         return g_strdup("disabled");
3018     case SPAPR_RESIZE_HPT_ENABLED:
3019         return g_strdup("enabled");
3020     case SPAPR_RESIZE_HPT_REQUIRED:
3021         return g_strdup("required");
3022     }
3023     g_assert_not_reached();
3024 }
3025 
3026 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3027 {
3028     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3029 
3030     if (strcmp(value, "default") == 0) {
3031         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3032     } else if (strcmp(value, "disabled") == 0) {
3033         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3034     } else if (strcmp(value, "enabled") == 0) {
3035         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3036     } else if (strcmp(value, "required") == 0) {
3037         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3038     } else {
3039         error_setg(errp, "Bad value for \"resize-hpt\" property");
3040     }
3041 }
3042 
3043 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3044                                    void *opaque, Error **errp)
3045 {
3046     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3047 }
3048 
3049 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3050                                    void *opaque, Error **errp)
3051 {
3052     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3053 }
3054 
3055 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3056 {
3057     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3058 
3059     if (spapr->irq == &spapr_irq_xics_legacy) {
3060         return g_strdup("legacy");
3061     } else if (spapr->irq == &spapr_irq_xics) {
3062         return g_strdup("xics");
3063     } else if (spapr->irq == &spapr_irq_xive) {
3064         return g_strdup("xive");
3065     }
3066     g_assert_not_reached();
3067 }
3068 
3069 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3070 {
3071     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3072 
3073     /* The legacy IRQ backend can not be set */
3074     if (strcmp(value, "xics") == 0) {
3075         spapr->irq = &spapr_irq_xics;
3076     } else if (strcmp(value, "xive") == 0) {
3077         spapr->irq = &spapr_irq_xive;
3078     } else {
3079         error_setg(errp, "Bad value for \"ic-mode\" property");
3080     }
3081 }
3082 
3083 static void spapr_instance_init(Object *obj)
3084 {
3085     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3086     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3087 
3088     spapr->htab_fd = -1;
3089     spapr->use_hotplug_event_source = true;
3090     object_property_add_str(obj, "kvm-type",
3091                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3092     object_property_set_description(obj, "kvm-type",
3093                                     "Specifies the KVM virtualization mode (HV, PR)",
3094                                     NULL);
3095     object_property_add_bool(obj, "modern-hotplug-events",
3096                             spapr_get_modern_hotplug_events,
3097                             spapr_set_modern_hotplug_events,
3098                             NULL);
3099     object_property_set_description(obj, "modern-hotplug-events",
3100                                     "Use dedicated hotplug event mechanism in"
3101                                     " place of standard EPOW events when possible"
3102                                     " (required for memory hot-unplug support)",
3103                                     NULL);
3104     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3105                             "Maximum permitted CPU compatibility mode",
3106                             &error_fatal);
3107 
3108     object_property_add_str(obj, "resize-hpt",
3109                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3110     object_property_set_description(obj, "resize-hpt",
3111                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3112                                     NULL);
3113     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3114                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3115     object_property_set_description(obj, "vsmt",
3116                                     "Virtual SMT: KVM behaves as if this were"
3117                                     " the host's SMT mode", &error_abort);
3118     object_property_add_bool(obj, "vfio-no-msix-emulation",
3119                              spapr_get_msix_emulation, NULL, NULL);
3120 
3121     /* The machine class defines the default interrupt controller mode */
3122     spapr->irq = smc->irq;
3123     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3124                             spapr_set_ic_mode, NULL);
3125     object_property_set_description(obj, "ic-mode",
3126                  "Specifies the interrupt controller mode (xics, xive)",
3127                  NULL);
3128 }
3129 
3130 static void spapr_machine_finalizefn(Object *obj)
3131 {
3132     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3133 
3134     g_free(spapr->kvm_type);
3135 }
3136 
3137 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3138 {
3139     cpu_synchronize_state(cs);
3140     ppc_cpu_do_system_reset(cs);
3141 }
3142 
3143 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3144 {
3145     CPUState *cs;
3146 
3147     CPU_FOREACH(cs) {
3148         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3149     }
3150 }
3151 
3152 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3153                            uint32_t node, bool dedicated_hp_event_source,
3154                            Error **errp)
3155 {
3156     sPAPRDRConnector *drc;
3157     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3158     int i, fdt_offset, fdt_size;
3159     void *fdt;
3160     uint64_t addr = addr_start;
3161     bool hotplugged = spapr_drc_hotplugged(dev);
3162     Error *local_err = NULL;
3163 
3164     for (i = 0; i < nr_lmbs; i++) {
3165         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3166                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3167         g_assert(drc);
3168 
3169         fdt = create_device_tree(&fdt_size);
3170         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3171                                                 SPAPR_MEMORY_BLOCK_SIZE);
3172 
3173         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3174         if (local_err) {
3175             while (addr > addr_start) {
3176                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3177                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3178                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3179                 spapr_drc_detach(drc);
3180             }
3181             g_free(fdt);
3182             error_propagate(errp, local_err);
3183             return;
3184         }
3185         if (!hotplugged) {
3186             spapr_drc_reset(drc);
3187         }
3188         addr += SPAPR_MEMORY_BLOCK_SIZE;
3189     }
3190     /* send hotplug notification to the
3191      * guest only in case of hotplugged memory
3192      */
3193     if (hotplugged) {
3194         if (dedicated_hp_event_source) {
3195             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3196                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3197             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3198                                                    nr_lmbs,
3199                                                    spapr_drc_index(drc));
3200         } else {
3201             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3202                                            nr_lmbs);
3203         }
3204     }
3205 }
3206 
3207 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3208                               Error **errp)
3209 {
3210     Error *local_err = NULL;
3211     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3212     PCDIMMDevice *dimm = PC_DIMM(dev);
3213     uint64_t size, addr;
3214     uint32_t node;
3215 
3216     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3217 
3218     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3219     if (local_err) {
3220         goto out;
3221     }
3222 
3223     addr = object_property_get_uint(OBJECT(dimm),
3224                                     PC_DIMM_ADDR_PROP, &local_err);
3225     if (local_err) {
3226         goto out_unplug;
3227     }
3228 
3229     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3230                                     &error_abort);
3231     spapr_add_lmbs(dev, addr, size, node,
3232                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3233                    &local_err);
3234     if (local_err) {
3235         goto out_unplug;
3236     }
3237 
3238     return;
3239 
3240 out_unplug:
3241     pc_dimm_unplug(dimm, MACHINE(ms));
3242 out:
3243     error_propagate(errp, local_err);
3244 }
3245 
3246 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3247                                   Error **errp)
3248 {
3249     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3250     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3251     PCDIMMDevice *dimm = PC_DIMM(dev);
3252     Error *local_err = NULL;
3253     uint64_t size;
3254     Object *memdev;
3255     hwaddr pagesize;
3256 
3257     if (!smc->dr_lmb_enabled) {
3258         error_setg(errp, "Memory hotplug not supported for this machine");
3259         return;
3260     }
3261 
3262     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3263     if (local_err) {
3264         error_propagate(errp, local_err);
3265         return;
3266     }
3267 
3268     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3269         error_setg(errp, "Hotplugged memory size must be a multiple of "
3270                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3271         return;
3272     }
3273 
3274     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3275                                       &error_abort);
3276     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3277     spapr_check_pagesize(spapr, pagesize, &local_err);
3278     if (local_err) {
3279         error_propagate(errp, local_err);
3280         return;
3281     }
3282 
3283     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3284 }
3285 
3286 struct sPAPRDIMMState {
3287     PCDIMMDevice *dimm;
3288     uint32_t nr_lmbs;
3289     QTAILQ_ENTRY(sPAPRDIMMState) next;
3290 };
3291 
3292 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3293                                                        PCDIMMDevice *dimm)
3294 {
3295     sPAPRDIMMState *dimm_state = NULL;
3296 
3297     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3298         if (dimm_state->dimm == dimm) {
3299             break;
3300         }
3301     }
3302     return dimm_state;
3303 }
3304 
3305 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3306                                                       uint32_t nr_lmbs,
3307                                                       PCDIMMDevice *dimm)
3308 {
3309     sPAPRDIMMState *ds = NULL;
3310 
3311     /*
3312      * If this request is for a DIMM whose removal had failed earlier
3313      * (due to guest's refusal to remove the LMBs), we would have this
3314      * dimm already in the pending_dimm_unplugs list. In that
3315      * case don't add again.
3316      */
3317     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3318     if (!ds) {
3319         ds = g_malloc0(sizeof(sPAPRDIMMState));
3320         ds->nr_lmbs = nr_lmbs;
3321         ds->dimm = dimm;
3322         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3323     }
3324     return ds;
3325 }
3326 
3327 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3328                                               sPAPRDIMMState *dimm_state)
3329 {
3330     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3331     g_free(dimm_state);
3332 }
3333 
3334 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3335                                                         PCDIMMDevice *dimm)
3336 {
3337     sPAPRDRConnector *drc;
3338     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3339                                                   &error_abort);
3340     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3341     uint32_t avail_lmbs = 0;
3342     uint64_t addr_start, addr;
3343     int i;
3344 
3345     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3346                                          &error_abort);
3347 
3348     addr = addr_start;
3349     for (i = 0; i < nr_lmbs; i++) {
3350         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3351                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3352         g_assert(drc);
3353         if (drc->dev) {
3354             avail_lmbs++;
3355         }
3356         addr += SPAPR_MEMORY_BLOCK_SIZE;
3357     }
3358 
3359     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3360 }
3361 
3362 /* Callback to be called during DRC release. */
3363 void spapr_lmb_release(DeviceState *dev)
3364 {
3365     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3366     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3367     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3368 
3369     /* This information will get lost if a migration occurs
3370      * during the unplug process. In this case recover it. */
3371     if (ds == NULL) {
3372         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3373         g_assert(ds);
3374         /* The DRC being examined by the caller at least must be counted */
3375         g_assert(ds->nr_lmbs);
3376     }
3377 
3378     if (--ds->nr_lmbs) {
3379         return;
3380     }
3381 
3382     /*
3383      * Now that all the LMBs have been removed by the guest, call the
3384      * unplug handler chain. This can never fail.
3385      */
3386     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3387 }
3388 
3389 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3390 {
3391     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3392     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3393 
3394     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3395     object_unparent(OBJECT(dev));
3396     spapr_pending_dimm_unplugs_remove(spapr, ds);
3397 }
3398 
3399 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3400                                         DeviceState *dev, Error **errp)
3401 {
3402     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3403     Error *local_err = NULL;
3404     PCDIMMDevice *dimm = PC_DIMM(dev);
3405     uint32_t nr_lmbs;
3406     uint64_t size, addr_start, addr;
3407     int i;
3408     sPAPRDRConnector *drc;
3409 
3410     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3411     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3412 
3413     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3414                                          &local_err);
3415     if (local_err) {
3416         goto out;
3417     }
3418 
3419     /*
3420      * An existing pending dimm state for this DIMM means that there is an
3421      * unplug operation in progress, waiting for the spapr_lmb_release
3422      * callback to complete the job (BQL can't cover that far). In this case,
3423      * bail out to avoid detaching DRCs that were already released.
3424      */
3425     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3426         error_setg(&local_err,
3427                    "Memory unplug already in progress for device %s",
3428                    dev->id);
3429         goto out;
3430     }
3431 
3432     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3433 
3434     addr = addr_start;
3435     for (i = 0; i < nr_lmbs; i++) {
3436         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3437                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3438         g_assert(drc);
3439 
3440         spapr_drc_detach(drc);
3441         addr += SPAPR_MEMORY_BLOCK_SIZE;
3442     }
3443 
3444     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3445                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3446     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3447                                               nr_lmbs, spapr_drc_index(drc));
3448 out:
3449     error_propagate(errp, local_err);
3450 }
3451 
3452 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3453                                            sPAPRMachineState *spapr)
3454 {
3455     PowerPCCPU *cpu = POWERPC_CPU(cs);
3456     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3457     int id = spapr_get_vcpu_id(cpu);
3458     void *fdt;
3459     int offset, fdt_size;
3460     char *nodename;
3461 
3462     fdt = create_device_tree(&fdt_size);
3463     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3464     offset = fdt_add_subnode(fdt, 0, nodename);
3465 
3466     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3467     g_free(nodename);
3468 
3469     *fdt_offset = offset;
3470     return fdt;
3471 }
3472 
3473 /* Callback to be called during DRC release. */
3474 void spapr_core_release(DeviceState *dev)
3475 {
3476     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3477 
3478     /* Call the unplug handler chain. This can never fail. */
3479     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3480 }
3481 
3482 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3483 {
3484     MachineState *ms = MACHINE(hotplug_dev);
3485     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3486     CPUCore *cc = CPU_CORE(dev);
3487     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3488 
3489     if (smc->pre_2_10_has_unused_icps) {
3490         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3491         int i;
3492 
3493         for (i = 0; i < cc->nr_threads; i++) {
3494             CPUState *cs = CPU(sc->threads[i]);
3495 
3496             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3497         }
3498     }
3499 
3500     assert(core_slot);
3501     core_slot->cpu = NULL;
3502     object_unparent(OBJECT(dev));
3503 }
3504 
3505 static
3506 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3507                                Error **errp)
3508 {
3509     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3510     int index;
3511     sPAPRDRConnector *drc;
3512     CPUCore *cc = CPU_CORE(dev);
3513 
3514     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3515         error_setg(errp, "Unable to find CPU core with core-id: %d",
3516                    cc->core_id);
3517         return;
3518     }
3519     if (index == 0) {
3520         error_setg(errp, "Boot CPU core may not be unplugged");
3521         return;
3522     }
3523 
3524     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3525                           spapr_vcpu_id(spapr, cc->core_id));
3526     g_assert(drc);
3527 
3528     spapr_drc_detach(drc);
3529 
3530     spapr_hotplug_req_remove_by_index(drc);
3531 }
3532 
3533 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3534                             Error **errp)
3535 {
3536     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3537     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3538     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3539     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3540     CPUCore *cc = CPU_CORE(dev);
3541     CPUState *cs = CPU(core->threads[0]);
3542     sPAPRDRConnector *drc;
3543     Error *local_err = NULL;
3544     CPUArchId *core_slot;
3545     int index;
3546     bool hotplugged = spapr_drc_hotplugged(dev);
3547 
3548     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3549     if (!core_slot) {
3550         error_setg(errp, "Unable to find CPU core with core-id: %d",
3551                    cc->core_id);
3552         return;
3553     }
3554     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3555                           spapr_vcpu_id(spapr, cc->core_id));
3556 
3557     g_assert(drc || !mc->has_hotpluggable_cpus);
3558 
3559     if (drc) {
3560         void *fdt;
3561         int fdt_offset;
3562 
3563         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3564 
3565         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3566         if (local_err) {
3567             g_free(fdt);
3568             error_propagate(errp, local_err);
3569             return;
3570         }
3571 
3572         if (hotplugged) {
3573             /*
3574              * Send hotplug notification interrupt to the guest only
3575              * in case of hotplugged CPUs.
3576              */
3577             spapr_hotplug_req_add_by_index(drc);
3578         } else {
3579             spapr_drc_reset(drc);
3580         }
3581     }
3582 
3583     core_slot->cpu = OBJECT(dev);
3584 
3585     if (smc->pre_2_10_has_unused_icps) {
3586         int i;
3587 
3588         for (i = 0; i < cc->nr_threads; i++) {
3589             cs = CPU(core->threads[i]);
3590             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3591         }
3592     }
3593 }
3594 
3595 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3596                                 Error **errp)
3597 {
3598     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3599     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3600     Error *local_err = NULL;
3601     CPUCore *cc = CPU_CORE(dev);
3602     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3603     const char *type = object_get_typename(OBJECT(dev));
3604     CPUArchId *core_slot;
3605     int index;
3606 
3607     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3608         error_setg(&local_err, "CPU hotplug not supported for this machine");
3609         goto out;
3610     }
3611 
3612     if (strcmp(base_core_type, type)) {
3613         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3614         goto out;
3615     }
3616 
3617     if (cc->core_id % smp_threads) {
3618         error_setg(&local_err, "invalid core id %d", cc->core_id);
3619         goto out;
3620     }
3621 
3622     /*
3623      * In general we should have homogeneous threads-per-core, but old
3624      * (pre hotplug support) machine types allow the last core to have
3625      * reduced threads as a compatibility hack for when we allowed
3626      * total vcpus not a multiple of threads-per-core.
3627      */
3628     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3629         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3630                    cc->nr_threads, smp_threads);
3631         goto out;
3632     }
3633 
3634     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3635     if (!core_slot) {
3636         error_setg(&local_err, "core id %d out of range", cc->core_id);
3637         goto out;
3638     }
3639 
3640     if (core_slot->cpu) {
3641         error_setg(&local_err, "core %d already populated", cc->core_id);
3642         goto out;
3643     }
3644 
3645     numa_cpu_pre_plug(core_slot, dev, &local_err);
3646 
3647 out:
3648     error_propagate(errp, local_err);
3649 }
3650 
3651 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3652                                       DeviceState *dev, Error **errp)
3653 {
3654     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3655         spapr_memory_plug(hotplug_dev, dev, errp);
3656     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3657         spapr_core_plug(hotplug_dev, dev, errp);
3658     }
3659 }
3660 
3661 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3662                                         DeviceState *dev, Error **errp)
3663 {
3664     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3665         spapr_memory_unplug(hotplug_dev, dev);
3666     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3667         spapr_core_unplug(hotplug_dev, dev);
3668     }
3669 }
3670 
3671 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3672                                                 DeviceState *dev, Error **errp)
3673 {
3674     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3675     MachineClass *mc = MACHINE_GET_CLASS(sms);
3676 
3677     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3678         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3679             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3680         } else {
3681             /* NOTE: this means there is a window after guest reset, prior to
3682              * CAS negotiation, where unplug requests will fail due to the
3683              * capability not being detected yet. This is a bit different than
3684              * the case with PCI unplug, where the events will be queued and
3685              * eventually handled by the guest after boot
3686              */
3687             error_setg(errp, "Memory hot unplug not supported for this guest");
3688         }
3689     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3690         if (!mc->has_hotpluggable_cpus) {
3691             error_setg(errp, "CPU hot unplug not supported on this machine");
3692             return;
3693         }
3694         spapr_core_unplug_request(hotplug_dev, dev, errp);
3695     }
3696 }
3697 
3698 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3699                                           DeviceState *dev, Error **errp)
3700 {
3701     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3702         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3703     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3704         spapr_core_pre_plug(hotplug_dev, dev, errp);
3705     }
3706 }
3707 
3708 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3709                                                  DeviceState *dev)
3710 {
3711     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3712         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3713         return HOTPLUG_HANDLER(machine);
3714     }
3715     return NULL;
3716 }
3717 
3718 static CpuInstanceProperties
3719 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3720 {
3721     CPUArchId *core_slot;
3722     MachineClass *mc = MACHINE_GET_CLASS(machine);
3723 
3724     /* make sure possible_cpu are intialized */
3725     mc->possible_cpu_arch_ids(machine);
3726     /* get CPU core slot containing thread that matches cpu_index */
3727     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3728     assert(core_slot);
3729     return core_slot->props;
3730 }
3731 
3732 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3733 {
3734     return idx / smp_cores % nb_numa_nodes;
3735 }
3736 
3737 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3738 {
3739     int i;
3740     const char *core_type;
3741     int spapr_max_cores = max_cpus / smp_threads;
3742     MachineClass *mc = MACHINE_GET_CLASS(machine);
3743 
3744     if (!mc->has_hotpluggable_cpus) {
3745         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3746     }
3747     if (machine->possible_cpus) {
3748         assert(machine->possible_cpus->len == spapr_max_cores);
3749         return machine->possible_cpus;
3750     }
3751 
3752     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3753     if (!core_type) {
3754         error_report("Unable to find sPAPR CPU Core definition");
3755         exit(1);
3756     }
3757 
3758     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3759                              sizeof(CPUArchId) * spapr_max_cores);
3760     machine->possible_cpus->len = spapr_max_cores;
3761     for (i = 0; i < machine->possible_cpus->len; i++) {
3762         int core_id = i * smp_threads;
3763 
3764         machine->possible_cpus->cpus[i].type = core_type;
3765         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3766         machine->possible_cpus->cpus[i].arch_id = core_id;
3767         machine->possible_cpus->cpus[i].props.has_core_id = true;
3768         machine->possible_cpus->cpus[i].props.core_id = core_id;
3769     }
3770     return machine->possible_cpus;
3771 }
3772 
3773 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3774                                 uint64_t *buid, hwaddr *pio,
3775                                 hwaddr *mmio32, hwaddr *mmio64,
3776                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3777 {
3778     /*
3779      * New-style PHB window placement.
3780      *
3781      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3782      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3783      * windows.
3784      *
3785      * Some guest kernels can't work with MMIO windows above 1<<46
3786      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3787      *
3788      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3789      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3790      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3791      * 1TiB 64-bit MMIO windows for each PHB.
3792      */
3793     const uint64_t base_buid = 0x800000020000000ULL;
3794 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3795                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3796     int i;
3797 
3798     /* Sanity check natural alignments */
3799     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3800     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3801     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3802     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3803     /* Sanity check bounds */
3804     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3805                       SPAPR_PCI_MEM32_WIN_SIZE);
3806     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3807                       SPAPR_PCI_MEM64_WIN_SIZE);
3808 
3809     if (index >= SPAPR_MAX_PHBS) {
3810         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3811                    SPAPR_MAX_PHBS - 1);
3812         return;
3813     }
3814 
3815     *buid = base_buid + index;
3816     for (i = 0; i < n_dma; ++i) {
3817         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3818     }
3819 
3820     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3821     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3822     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3823 }
3824 
3825 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3826 {
3827     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3828 
3829     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3830 }
3831 
3832 static void spapr_ics_resend(XICSFabric *dev)
3833 {
3834     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3835 
3836     ics_resend(spapr->ics);
3837 }
3838 
3839 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3840 {
3841     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3842 
3843     return cpu ? ICP(cpu->intc) : NULL;
3844 }
3845 
3846 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3847                                  Monitor *mon)
3848 {
3849     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3850 
3851     spapr->irq->print_info(spapr, mon);
3852 }
3853 
3854 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3855 {
3856     return cpu->vcpu_id;
3857 }
3858 
3859 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3860 {
3861     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3862     int vcpu_id;
3863 
3864     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3865 
3866     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3867         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3868         error_append_hint(errp, "Adjust the number of cpus to %d "
3869                           "or try to raise the number of threads per core\n",
3870                           vcpu_id * smp_threads / spapr->vsmt);
3871         return;
3872     }
3873 
3874     cpu->vcpu_id = vcpu_id;
3875 }
3876 
3877 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3878 {
3879     CPUState *cs;
3880 
3881     CPU_FOREACH(cs) {
3882         PowerPCCPU *cpu = POWERPC_CPU(cs);
3883 
3884         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3885             return cpu;
3886         }
3887     }
3888 
3889     return NULL;
3890 }
3891 
3892 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3893 {
3894     MachineClass *mc = MACHINE_CLASS(oc);
3895     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3896     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3897     NMIClass *nc = NMI_CLASS(oc);
3898     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3899     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3900     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3901     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3902 
3903     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3904     mc->ignore_boot_device_suffixes = true;
3905 
3906     /*
3907      * We set up the default / latest behaviour here.  The class_init
3908      * functions for the specific versioned machine types can override
3909      * these details for backwards compatibility
3910      */
3911     mc->init = spapr_machine_init;
3912     mc->reset = spapr_machine_reset;
3913     mc->block_default_type = IF_SCSI;
3914     mc->max_cpus = 1024;
3915     mc->no_parallel = 1;
3916     mc->default_boot_order = "";
3917     mc->default_ram_size = 512 * MiB;
3918     mc->default_display = "std";
3919     mc->kvm_type = spapr_kvm_type;
3920     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3921     mc->pci_allow_0_address = true;
3922     assert(!mc->get_hotplug_handler);
3923     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3924     hc->pre_plug = spapr_machine_device_pre_plug;
3925     hc->plug = spapr_machine_device_plug;
3926     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3927     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3928     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3929     hc->unplug_request = spapr_machine_device_unplug_request;
3930     hc->unplug = spapr_machine_device_unplug;
3931 
3932     smc->dr_lmb_enabled = true;
3933     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
3934     mc->has_hotpluggable_cpus = true;
3935     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3936     fwc->get_dev_path = spapr_get_fw_dev_path;
3937     nc->nmi_monitor_handler = spapr_nmi;
3938     smc->phb_placement = spapr_phb_placement;
3939     vhc->hypercall = emulate_spapr_hypercall;
3940     vhc->hpt_mask = spapr_hpt_mask;
3941     vhc->map_hptes = spapr_map_hptes;
3942     vhc->unmap_hptes = spapr_unmap_hptes;
3943     vhc->store_hpte = spapr_store_hpte;
3944     vhc->get_patbe = spapr_get_patbe;
3945     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3946     xic->ics_get = spapr_ics_get;
3947     xic->ics_resend = spapr_ics_resend;
3948     xic->icp_get = spapr_icp_get;
3949     ispc->print_info = spapr_pic_print_info;
3950     /* Force NUMA node memory size to be a multiple of
3951      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3952      * in which LMBs are represented and hot-added
3953      */
3954     mc->numa_mem_align_shift = 28;
3955 
3956     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
3957     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
3958     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
3959     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
3960     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
3961     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
3962     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
3963     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
3964     spapr_caps_add_properties(smc, &error_abort);
3965     smc->irq = &spapr_irq_xics;
3966 }
3967 
3968 static const TypeInfo spapr_machine_info = {
3969     .name          = TYPE_SPAPR_MACHINE,
3970     .parent        = TYPE_MACHINE,
3971     .abstract      = true,
3972     .instance_size = sizeof(sPAPRMachineState),
3973     .instance_init = spapr_instance_init,
3974     .instance_finalize = spapr_machine_finalizefn,
3975     .class_size    = sizeof(sPAPRMachineClass),
3976     .class_init    = spapr_machine_class_init,
3977     .interfaces = (InterfaceInfo[]) {
3978         { TYPE_FW_PATH_PROVIDER },
3979         { TYPE_NMI },
3980         { TYPE_HOTPLUG_HANDLER },
3981         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3982         { TYPE_XICS_FABRIC },
3983         { TYPE_INTERRUPT_STATS_PROVIDER },
3984         { }
3985     },
3986 };
3987 
3988 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3989     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3990                                                     void *data)      \
3991     {                                                                \
3992         MachineClass *mc = MACHINE_CLASS(oc);                        \
3993         spapr_machine_##suffix##_class_options(mc);                  \
3994         if (latest) {                                                \
3995             mc->alias = "pseries";                                   \
3996             mc->is_default = 1;                                      \
3997         }                                                            \
3998     }                                                                \
3999     static const TypeInfo spapr_machine_##suffix##_info = {          \
4000         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4001         .parent = TYPE_SPAPR_MACHINE,                                \
4002         .class_init = spapr_machine_##suffix##_class_init,           \
4003     };                                                               \
4004     static void spapr_machine_register_##suffix(void)                \
4005     {                                                                \
4006         type_register(&spapr_machine_##suffix##_info);               \
4007     }                                                                \
4008     type_init(spapr_machine_register_##suffix)
4009 
4010 /*
4011  * pseries-4.0
4012  */
4013 static void spapr_machine_4_0_class_options(MachineClass *mc)
4014 {
4015     /* Defaults for the latest behaviour inherited from the base class */
4016 }
4017 
4018 DEFINE_SPAPR_MACHINE(4_0, "4.0", true);
4019 
4020 /*
4021  * pseries-3.1
4022  */
4023 static void spapr_machine_3_1_class_options(MachineClass *mc)
4024 {
4025     spapr_machine_4_0_class_options(mc);
4026     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4027     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4028 }
4029 
4030 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4031 
4032 /*
4033  * pseries-3.0
4034  */
4035 
4036 static void spapr_machine_3_0_class_options(MachineClass *mc)
4037 {
4038     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4039 
4040     spapr_machine_3_1_class_options(mc);
4041     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4042 
4043     smc->legacy_irq_allocation = true;
4044     smc->irq = &spapr_irq_xics_legacy;
4045 }
4046 
4047 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4048 
4049 /*
4050  * pseries-2.12
4051  */
4052 static void spapr_machine_2_12_class_options(MachineClass *mc)
4053 {
4054     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4055     static GlobalProperty compat[] = {
4056         {
4057             .driver = TYPE_POWERPC_CPU,
4058             .property = "pre-3.0-migration",
4059             .value    = "on",
4060         },
4061         {
4062             .driver = TYPE_SPAPR_CPU_CORE,
4063             .property = "pre-3.0-migration",
4064             .value    = "on",
4065         },
4066     };
4067 
4068     spapr_machine_3_0_class_options(mc);
4069     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4070     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4071 
4072     /* We depend on kvm_enabled() to choose a default value for the
4073      * hpt-max-page-size capability. Of course we can't do it here
4074      * because this is too early and the HW accelerator isn't initialzed
4075      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4076      */
4077     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4078 }
4079 
4080 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4081 
4082 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4083 {
4084     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4085 
4086     spapr_machine_2_12_class_options(mc);
4087     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4088     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4089     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4090 }
4091 
4092 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4093 
4094 /*
4095  * pseries-2.11
4096  */
4097 
4098 static void spapr_machine_2_11_class_options(MachineClass *mc)
4099 {
4100     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4101 
4102     spapr_machine_2_12_class_options(mc);
4103     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4104     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4105 }
4106 
4107 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4108 
4109 /*
4110  * pseries-2.10
4111  */
4112 
4113 static void spapr_machine_2_10_class_options(MachineClass *mc)
4114 {
4115     spapr_machine_2_11_class_options(mc);
4116     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4117 }
4118 
4119 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4120 
4121 /*
4122  * pseries-2.9
4123  */
4124 
4125 static void spapr_machine_2_9_class_options(MachineClass *mc)
4126 {
4127     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4128     static GlobalProperty compat[] = {
4129         {
4130             .driver = TYPE_POWERPC_CPU,
4131             .property = "pre-2.10-migration",
4132             .value    = "on",
4133         },
4134     };
4135 
4136     spapr_machine_2_10_class_options(mc);
4137     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4138     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4139     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4140     smc->pre_2_10_has_unused_icps = true;
4141     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4142 }
4143 
4144 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4145 
4146 /*
4147  * pseries-2.8
4148  */
4149 
4150 static void spapr_machine_2_8_class_options(MachineClass *mc)
4151 {
4152     static GlobalProperty compat[] = {
4153         {
4154             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,
4155             .property = "pcie-extended-configuration-space",
4156             .value    = "off",
4157         },
4158     };
4159 
4160     spapr_machine_2_9_class_options(mc);
4161     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4162     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4163     mc->numa_mem_align_shift = 23;
4164 }
4165 
4166 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4167 
4168 /*
4169  * pseries-2.7
4170  */
4171 
4172 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4173                               uint64_t *buid, hwaddr *pio,
4174                               hwaddr *mmio32, hwaddr *mmio64,
4175                               unsigned n_dma, uint32_t *liobns, Error **errp)
4176 {
4177     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4178     const uint64_t base_buid = 0x800000020000000ULL;
4179     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4180     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4181     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4182     const uint32_t max_index = 255;
4183     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4184 
4185     uint64_t ram_top = MACHINE(spapr)->ram_size;
4186     hwaddr phb0_base, phb_base;
4187     int i;
4188 
4189     /* Do we have device memory? */
4190     if (MACHINE(spapr)->maxram_size > ram_top) {
4191         /* Can't just use maxram_size, because there may be an
4192          * alignment gap between normal and device memory regions
4193          */
4194         ram_top = MACHINE(spapr)->device_memory->base +
4195             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4196     }
4197 
4198     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4199 
4200     if (index > max_index) {
4201         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4202                    max_index);
4203         return;
4204     }
4205 
4206     *buid = base_buid + index;
4207     for (i = 0; i < n_dma; ++i) {
4208         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4209     }
4210 
4211     phb_base = phb0_base + index * phb_spacing;
4212     *pio = phb_base + pio_offset;
4213     *mmio32 = phb_base + mmio_offset;
4214     /*
4215      * We don't set the 64-bit MMIO window, relying on the PHB's
4216      * fallback behaviour of automatically splitting a large "32-bit"
4217      * window into contiguous 32-bit and 64-bit windows
4218      */
4219 }
4220 
4221 static void spapr_machine_2_7_class_options(MachineClass *mc)
4222 {
4223     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4224     static GlobalProperty compat[] = {
4225         {
4226             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,
4227             .property = "mem_win_size",
4228             .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),
4229         },
4230         {
4231             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,
4232             .property = "mem64_win_size",
4233             .value    = "0",
4234         },
4235         {
4236             .driver = TYPE_POWERPC_CPU,
4237             .property = "pre-2.8-migration",
4238             .value    = "on",
4239         },
4240         {
4241             .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,
4242             .property = "pre-2.8-migration",
4243             .value    = "on",
4244         },
4245     };
4246 
4247     spapr_machine_2_8_class_options(mc);
4248     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4249     mc->default_machine_opts = "modern-hotplug-events=off";
4250     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4251     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4252     smc->phb_placement = phb_placement_2_7;
4253 }
4254 
4255 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4256 
4257 /*
4258  * pseries-2.6
4259  */
4260 
4261 static void spapr_machine_2_6_class_options(MachineClass *mc)
4262 {
4263     static GlobalProperty compat[] = {
4264         {
4265             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,
4266             .property = "ddw",
4267             .value    = stringify(off),
4268         },
4269     };
4270 
4271     spapr_machine_2_7_class_options(mc);
4272     mc->has_hotpluggable_cpus = false;
4273     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4274     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4275 }
4276 
4277 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4278 
4279 /*
4280  * pseries-2.5
4281  */
4282 
4283 static void spapr_machine_2_5_class_options(MachineClass *mc)
4284 {
4285     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4286     static GlobalProperty compat[] = {
4287         {
4288             .driver   = "spapr-vlan",
4289             .property = "use-rx-buffer-pools",
4290             .value    = "off",
4291         },
4292     };
4293 
4294     spapr_machine_2_6_class_options(mc);
4295     smc->use_ohci_by_default = true;
4296     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4297     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4298 }
4299 
4300 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4301 
4302 /*
4303  * pseries-2.4
4304  */
4305 
4306 static void spapr_machine_2_4_class_options(MachineClass *mc)
4307 {
4308     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4309 
4310     spapr_machine_2_5_class_options(mc);
4311     smc->dr_lmb_enabled = false;
4312     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4313 }
4314 
4315 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4316 
4317 /*
4318  * pseries-2.3
4319  */
4320 
4321 static void spapr_machine_2_3_class_options(MachineClass *mc)
4322 {
4323     static GlobalProperty compat[] = {
4324         {
4325             .driver   = "spapr-pci-host-bridge",
4326             .property = "dynamic-reconfiguration",
4327             .value    = "off",
4328         },
4329     };
4330     spapr_machine_2_4_class_options(mc);
4331     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4332     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4333 }
4334 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4335 
4336 /*
4337  * pseries-2.2
4338  */
4339 
4340 static void spapr_machine_2_2_class_options(MachineClass *mc)
4341 {
4342     static GlobalProperty compat[] = {
4343         {
4344             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,
4345             .property = "mem_win_size",
4346             .value    = "0x20000000",
4347         },
4348     };
4349 
4350     spapr_machine_2_3_class_options(mc);
4351     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4352     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4353     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4354 }
4355 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4356 
4357 /*
4358  * pseries-2.1
4359  */
4360 
4361 static void spapr_machine_2_1_class_options(MachineClass *mc)
4362 {
4363     spapr_machine_2_2_class_options(mc);
4364     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4365 }
4366 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4367 
4368 static void spapr_machine_register_types(void)
4369 {
4370     type_register_static(&spapr_machine_info);
4371 }
4372 
4373 type_init(spapr_machine_register_types)
4374