xref: /openbmc/qemu/hw/ppc/spapr.c (revision de86eccc0c836adfa8dbb94848096720177f5ccb)
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 "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
31 #include "hw/hw.h"
32 #include "qemu/log.h"
33 #include "hw/fw-path-provider.h"
34 #include "elf.h"
35 #include "net/net.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/migration.h"
42 #include "mmu-hash64.h"
43 #include "mmu-book3s-v3.h"
44 #include "qom/cpu.h"
45 
46 #include "hw/boards.h"
47 #include "hw/ppc/ppc.h"
48 #include "hw/loader.h"
49 
50 #include "hw/ppc/fdt.h"
51 #include "hw/ppc/spapr.h"
52 #include "hw/ppc/spapr_vio.h"
53 #include "hw/pci-host/spapr.h"
54 #include "hw/ppc/xics.h"
55 #include "hw/pci/msi.h"
56 
57 #include "hw/pci/pci.h"
58 #include "hw/scsi/scsi.h"
59 #include "hw/virtio/virtio-scsi.h"
60 
61 #include "exec/address-spaces.h"
62 #include "hw/usb.h"
63 #include "qemu/config-file.h"
64 #include "qemu/error-report.h"
65 #include "trace.h"
66 #include "hw/nmi.h"
67 #include "hw/intc/intc.h"
68 
69 #include "hw/compat.h"
70 #include "qemu/cutils.h"
71 #include "hw/ppc/spapr_cpu_core.h"
72 #include "qmp-commands.h"
73 
74 #include <libfdt.h>
75 
76 /* SLOF memory layout:
77  *
78  * SLOF raw image loaded at 0, copies its romfs right below the flat
79  * device-tree, then position SLOF itself 31M below that
80  *
81  * So we set FW_OVERHEAD to 40MB which should account for all of that
82  * and more
83  *
84  * We load our kernel at 4M, leaving space for SLOF initial image
85  */
86 #define FDT_MAX_SIZE            0x100000
87 #define RTAS_MAX_SIZE           0x10000
88 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
89 #define FW_MAX_SIZE             0x400000
90 #define FW_FILE_NAME            "slof.bin"
91 #define FW_OVERHEAD             0x2800000
92 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
93 
94 #define MIN_RMA_SLOF            128UL
95 
96 #define PHANDLE_XICP            0x00001111
97 
98 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
99 
100 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
101                                   const char *type_ics,
102                                   int nr_irqs, Error **errp)
103 {
104     Error *local_err = NULL;
105     Object *obj;
106 
107     obj = object_new(type_ics);
108     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
109     object_property_add_const_link(obj, "xics", OBJECT(spapr), &error_abort);
110     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
111     if (local_err) {
112         goto error;
113     }
114     object_property_set_bool(obj, true, "realized", &local_err);
115     if (local_err) {
116         goto error;
117     }
118 
119     return ICS_SIMPLE(obj);
120 
121 error:
122     error_propagate(errp, local_err);
123     return NULL;
124 }
125 
126 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
127 {
128     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
129 
130     if (kvm_enabled()) {
131         Error *err = NULL;
132 
133         if (machine_kernel_irqchip_allowed(machine) &&
134             !xics_kvm_init(spapr, errp)) {
135             spapr->icp_type = TYPE_KVM_ICP;
136             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, &err);
137         }
138         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
139             error_reportf_err(err,
140                               "kernel_irqchip requested but unavailable: ");
141         } else {
142             error_free(err);
143         }
144     }
145 
146     if (!spapr->ics) {
147         xics_spapr_init(spapr);
148         spapr->icp_type = TYPE_ICP;
149         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
150     }
151 }
152 
153 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
154                                   int smt_threads)
155 {
156     int i, ret = 0;
157     uint32_t servers_prop[smt_threads];
158     uint32_t gservers_prop[smt_threads * 2];
159     int index = ppc_get_vcpu_dt_id(cpu);
160 
161     if (cpu->compat_pvr) {
162         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
163         if (ret < 0) {
164             return ret;
165         }
166     }
167 
168     /* Build interrupt servers and gservers properties */
169     for (i = 0; i < smt_threads; i++) {
170         servers_prop[i] = cpu_to_be32(index + i);
171         /* Hack, direct the group queues back to cpu 0 */
172         gservers_prop[i*2] = cpu_to_be32(index + i);
173         gservers_prop[i*2 + 1] = 0;
174     }
175     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
176                       servers_prop, sizeof(servers_prop));
177     if (ret < 0) {
178         return ret;
179     }
180     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
181                       gservers_prop, sizeof(gservers_prop));
182 
183     return ret;
184 }
185 
186 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
187 {
188     int ret = 0;
189     PowerPCCPU *cpu = POWERPC_CPU(cs);
190     int index = ppc_get_vcpu_dt_id(cpu);
191     uint32_t associativity[] = {cpu_to_be32(0x5),
192                                 cpu_to_be32(0x0),
193                                 cpu_to_be32(0x0),
194                                 cpu_to_be32(0x0),
195                                 cpu_to_be32(cs->numa_node),
196                                 cpu_to_be32(index)};
197 
198     /* Advertise NUMA via ibm,associativity */
199     if (nb_numa_nodes > 1) {
200         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
201                           sizeof(associativity));
202     }
203 
204     return ret;
205 }
206 
207 /* Populate the "ibm,pa-features" property */
208 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset,
209                                       bool legacy_guest)
210 {
211     uint8_t pa_features_206[] = { 6, 0,
212         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
213     uint8_t pa_features_207[] = { 24, 0,
214         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
215         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
216         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
217         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
218     uint8_t pa_features_300[] = { 66, 0,
219         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
220         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
221         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
222         /* 6: DS207 */
223         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
224         /* 16: Vector */
225         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
226         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
227         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
228         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
229         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
230         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
231         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
232         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
233         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
234         /* 42: PM, 44: PC RA, 46: SC vec'd */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
236         /* 48: SIMD, 50: QP BFP, 52: String */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
238         /* 54: DecFP, 56: DecI, 58: SHA */
239         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
240         /* 60: NM atomic, 62: RNG */
241         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
242     };
243     uint8_t *pa_features;
244     size_t pa_size;
245 
246     switch (POWERPC_MMU_VER(env->mmu_model)) {
247     case POWERPC_MMU_VER_2_06:
248         pa_features = pa_features_206;
249         pa_size = sizeof(pa_features_206);
250         break;
251     case POWERPC_MMU_VER_2_07:
252         pa_features = pa_features_207;
253         pa_size = sizeof(pa_features_207);
254         break;
255     case POWERPC_MMU_VER_3_00:
256         pa_features = pa_features_300;
257         pa_size = sizeof(pa_features_300);
258         break;
259     default:
260         return;
261     }
262 
263     if (env->ci_large_pages) {
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 (kvmppc_has_cap_htm() && 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     int smt = kvmppc_smt_threads();
292     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
293 
294     CPU_FOREACH(cs) {
295         PowerPCCPU *cpu = POWERPC_CPU(cs);
296         CPUPPCState *env = &cpu->env;
297         DeviceClass *dc = DEVICE_GET_CLASS(cs);
298         int index = ppc_get_vcpu_dt_id(cpu);
299         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
300 
301         if ((index % smt) != 0) {
302             continue;
303         }
304 
305         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
306 
307         cpus_offset = fdt_path_offset(fdt, "/cpus");
308         if (cpus_offset < 0) {
309             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
310                                           "cpus");
311             if (cpus_offset < 0) {
312                 return cpus_offset;
313             }
314         }
315         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
316         if (offset < 0) {
317             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
318             if (offset < 0) {
319                 return offset;
320             }
321         }
322 
323         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
324                           pft_size_prop, sizeof(pft_size_prop));
325         if (ret < 0) {
326             return ret;
327         }
328 
329         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
330         if (ret < 0) {
331             return ret;
332         }
333 
334         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
335         if (ret < 0) {
336             return ret;
337         }
338 
339         spapr_populate_pa_features(env, fdt, offset,
340                                          spapr->cas_legacy_guest_workaround);
341     }
342     return ret;
343 }
344 
345 static hwaddr spapr_node0_size(void)
346 {
347     MachineState *machine = MACHINE(qdev_get_machine());
348 
349     if (nb_numa_nodes) {
350         int i;
351         for (i = 0; i < nb_numa_nodes; ++i) {
352             if (numa_info[i].node_mem) {
353                 return MIN(pow2floor(numa_info[i].node_mem),
354                            machine->ram_size);
355             }
356         }
357     }
358     return machine->ram_size;
359 }
360 
361 static void add_str(GString *s, const gchar *s1)
362 {
363     g_string_append_len(s, s1, strlen(s1) + 1);
364 }
365 
366 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
367                                        hwaddr size)
368 {
369     uint32_t associativity[] = {
370         cpu_to_be32(0x4), /* length */
371         cpu_to_be32(0x0), cpu_to_be32(0x0),
372         cpu_to_be32(0x0), cpu_to_be32(nodeid)
373     };
374     char mem_name[32];
375     uint64_t mem_reg_property[2];
376     int off;
377 
378     mem_reg_property[0] = cpu_to_be64(start);
379     mem_reg_property[1] = cpu_to_be64(size);
380 
381     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
382     off = fdt_add_subnode(fdt, 0, mem_name);
383     _FDT(off);
384     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
385     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
386                       sizeof(mem_reg_property))));
387     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
388                       sizeof(associativity))));
389     return off;
390 }
391 
392 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
393 {
394     MachineState *machine = MACHINE(spapr);
395     hwaddr mem_start, node_size;
396     int i, nb_nodes = nb_numa_nodes;
397     NodeInfo *nodes = numa_info;
398     NodeInfo ramnode;
399 
400     /* No NUMA nodes, assume there is just one node with whole RAM */
401     if (!nb_numa_nodes) {
402         nb_nodes = 1;
403         ramnode.node_mem = machine->ram_size;
404         nodes = &ramnode;
405     }
406 
407     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
408         if (!nodes[i].node_mem) {
409             continue;
410         }
411         if (mem_start >= machine->ram_size) {
412             node_size = 0;
413         } else {
414             node_size = nodes[i].node_mem;
415             if (node_size > machine->ram_size - mem_start) {
416                 node_size = machine->ram_size - mem_start;
417             }
418         }
419         if (!mem_start) {
420             /* ppc_spapr_init() checks for rma_size <= node0_size already */
421             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
422             mem_start += spapr->rma_size;
423             node_size -= spapr->rma_size;
424         }
425         for ( ; node_size; ) {
426             hwaddr sizetmp = pow2floor(node_size);
427 
428             /* mem_start != 0 here */
429             if (ctzl(mem_start) < ctzl(sizetmp)) {
430                 sizetmp = 1ULL << ctzl(mem_start);
431             }
432 
433             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
434             node_size -= sizetmp;
435             mem_start += sizetmp;
436         }
437     }
438 
439     return 0;
440 }
441 
442 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
443                                   sPAPRMachineState *spapr)
444 {
445     PowerPCCPU *cpu = POWERPC_CPU(cs);
446     CPUPPCState *env = &cpu->env;
447     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
448     int index = ppc_get_vcpu_dt_id(cpu);
449     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
450                        0xffffffff, 0xffffffff};
451     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
452         : SPAPR_TIMEBASE_FREQ;
453     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
454     uint32_t page_sizes_prop[64];
455     size_t page_sizes_prop_size;
456     uint32_t vcpus_per_socket = smp_threads * smp_cores;
457     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
458     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
459     sPAPRDRConnector *drc;
460     sPAPRDRConnectorClass *drck;
461     int drc_index;
462     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
463     int i;
464 
465     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
466     if (drc) {
467         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
468         drc_index = drck->get_index(drc);
469         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
470     }
471 
472     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
473     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
474 
475     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
476     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
477                            env->dcache_line_size)));
478     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
479                            env->dcache_line_size)));
480     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
481                            env->icache_line_size)));
482     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
483                            env->icache_line_size)));
484 
485     if (pcc->l1_dcache_size) {
486         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
487                                pcc->l1_dcache_size)));
488     } else {
489         error_report("Warning: Unknown L1 dcache size for cpu");
490     }
491     if (pcc->l1_icache_size) {
492         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
493                                pcc->l1_icache_size)));
494     } else {
495         error_report("Warning: Unknown L1 icache size for cpu");
496     }
497 
498     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
499     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
500     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
501     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
502     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
503     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
504 
505     if (env->spr_cb[SPR_PURR].oea_read) {
506         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
507     }
508 
509     if (env->mmu_model & POWERPC_MMU_1TSEG) {
510         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
511                           segs, sizeof(segs))));
512     }
513 
514     /* Advertise VMX/VSX (vector extensions) if available
515      *   0 / no property == no vector extensions
516      *   1               == VMX / Altivec available
517      *   2               == VSX available */
518     if (env->insns_flags & PPC_ALTIVEC) {
519         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
520 
521         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
522     }
523 
524     /* Advertise DFP (Decimal Floating Point) if available
525      *   0 / no property == no DFP
526      *   1               == DFP available */
527     if (env->insns_flags2 & PPC2_DFP) {
528         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
529     }
530 
531     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
532                                                   sizeof(page_sizes_prop));
533     if (page_sizes_prop_size) {
534         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
535                           page_sizes_prop, page_sizes_prop_size)));
536     }
537 
538     spapr_populate_pa_features(env, fdt, offset, false);
539 
540     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
541                            cs->cpu_index / vcpus_per_socket)));
542 
543     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
544                       pft_size_prop, sizeof(pft_size_prop))));
545 
546     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
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 *cs;
565     int cpus_offset;
566     char *nodename;
567     int smt = kvmppc_smt_threads();
568 
569     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
570     _FDT(cpus_offset);
571     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
572     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
573 
574     /*
575      * We walk the CPUs in reverse order to ensure that CPU DT nodes
576      * created by fdt_add_subnode() end up in the right order in FDT
577      * for the guest kernel the enumerate the CPUs correctly.
578      */
579     CPU_FOREACH_REVERSE(cs) {
580         PowerPCCPU *cpu = POWERPC_CPU(cs);
581         int index = ppc_get_vcpu_dt_id(cpu);
582         DeviceClass *dc = DEVICE_GET_CLASS(cs);
583         int offset;
584 
585         if ((index % smt) != 0) {
586             continue;
587         }
588 
589         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
590         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
591         g_free(nodename);
592         _FDT(offset);
593         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
594     }
595 
596 }
597 
598 /*
599  * Adds ibm,dynamic-reconfiguration-memory node.
600  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
601  * of this device tree node.
602  */
603 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
604 {
605     MachineState *machine = MACHINE(spapr);
606     int ret, i, offset;
607     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
608     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
609     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
610     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
611                        memory_region_size(&spapr->hotplug_memory.mr)) /
612                        lmb_size;
613     uint32_t *int_buf, *cur_index, buf_len;
614     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
615 
616     /*
617      * Don't create the node if there is no hotpluggable memory
618      */
619     if (machine->ram_size == machine->maxram_size) {
620         return 0;
621     }
622 
623     /*
624      * Allocate enough buffer size to fit in ibm,dynamic-memory
625      * or ibm,associativity-lookup-arrays
626      */
627     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
628               * sizeof(uint32_t);
629     cur_index = int_buf = g_malloc0(buf_len);
630 
631     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
632 
633     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
634                     sizeof(prop_lmb_size));
635     if (ret < 0) {
636         goto out;
637     }
638 
639     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
640     if (ret < 0) {
641         goto out;
642     }
643 
644     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
645     if (ret < 0) {
646         goto out;
647     }
648 
649     /* ibm,dynamic-memory */
650     int_buf[0] = cpu_to_be32(nr_lmbs);
651     cur_index++;
652     for (i = 0; i < nr_lmbs; i++) {
653         uint64_t addr = i * lmb_size;
654         uint32_t *dynamic_memory = cur_index;
655 
656         if (i >= hotplug_lmb_start) {
657             sPAPRDRConnector *drc;
658             sPAPRDRConnectorClass *drck;
659 
660             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
661             g_assert(drc);
662             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
663 
664             dynamic_memory[0] = cpu_to_be32(addr >> 32);
665             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
666             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
667             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
668             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
669             if (memory_region_present(get_system_memory(), addr)) {
670                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
671             } else {
672                 dynamic_memory[5] = cpu_to_be32(0);
673             }
674         } else {
675             /*
676              * LMB information for RMA, boot time RAM and gap b/n RAM and
677              * hotplug memory region -- all these are marked as reserved
678              * and as having no valid DRC.
679              */
680             dynamic_memory[0] = cpu_to_be32(addr >> 32);
681             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
682             dynamic_memory[2] = cpu_to_be32(0);
683             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
684             dynamic_memory[4] = cpu_to_be32(-1);
685             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
686                                             SPAPR_LMB_FLAGS_DRC_INVALID);
687         }
688 
689         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
690     }
691     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
692     if (ret < 0) {
693         goto out;
694     }
695 
696     /* ibm,associativity-lookup-arrays */
697     cur_index = int_buf;
698     int_buf[0] = cpu_to_be32(nr_nodes);
699     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
700     cur_index += 2;
701     for (i = 0; i < nr_nodes; i++) {
702         uint32_t associativity[] = {
703             cpu_to_be32(0x0),
704             cpu_to_be32(0x0),
705             cpu_to_be32(0x0),
706             cpu_to_be32(i)
707         };
708         memcpy(cur_index, associativity, sizeof(associativity));
709         cur_index += 4;
710     }
711     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
712             (cur_index - int_buf) * sizeof(uint32_t));
713 out:
714     g_free(int_buf);
715     return ret;
716 }
717 
718 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
719                                 sPAPROptionVector *ov5_updates)
720 {
721     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
722     int ret = 0, offset;
723 
724     /* Generate ibm,dynamic-reconfiguration-memory node if required */
725     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
726         g_assert(smc->dr_lmb_enabled);
727         ret = spapr_populate_drconf_memory(spapr, fdt);
728         if (ret) {
729             goto out;
730         }
731     }
732 
733     offset = fdt_path_offset(fdt, "/chosen");
734     if (offset < 0) {
735         offset = fdt_add_subnode(fdt, 0, "chosen");
736         if (offset < 0) {
737             return offset;
738         }
739     }
740     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
741                                  "ibm,architecture-vec-5");
742 
743 out:
744     return ret;
745 }
746 
747 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
748                                  target_ulong addr, target_ulong size,
749                                  sPAPROptionVector *ov5_updates)
750 {
751     void *fdt, *fdt_skel;
752     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
753 
754     size -= sizeof(hdr);
755 
756     /* Create sceleton */
757     fdt_skel = g_malloc0(size);
758     _FDT((fdt_create(fdt_skel, size)));
759     _FDT((fdt_begin_node(fdt_skel, "")));
760     _FDT((fdt_end_node(fdt_skel)));
761     _FDT((fdt_finish(fdt_skel)));
762     fdt = g_malloc0(size);
763     _FDT((fdt_open_into(fdt_skel, fdt, size)));
764     g_free(fdt_skel);
765 
766     /* Fixup cpu nodes */
767     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
768 
769     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
770         return -1;
771     }
772 
773     /* Pack resulting tree */
774     _FDT((fdt_pack(fdt)));
775 
776     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
777         trace_spapr_cas_failed(size);
778         return -1;
779     }
780 
781     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
782     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
783     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
784     g_free(fdt);
785 
786     return 0;
787 }
788 
789 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
790 {
791     int rtas;
792     GString *hypertas = g_string_sized_new(256);
793     GString *qemu_hypertas = g_string_sized_new(256);
794     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
795     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
796         memory_region_size(&spapr->hotplug_memory.mr);
797     uint32_t lrdr_capacity[] = {
798         cpu_to_be32(max_hotplug_addr >> 32),
799         cpu_to_be32(max_hotplug_addr & 0xffffffff),
800         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
801         cpu_to_be32(max_cpus / smp_threads),
802     };
803 
804     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
805 
806     /* hypertas */
807     add_str(hypertas, "hcall-pft");
808     add_str(hypertas, "hcall-term");
809     add_str(hypertas, "hcall-dabr");
810     add_str(hypertas, "hcall-interrupt");
811     add_str(hypertas, "hcall-tce");
812     add_str(hypertas, "hcall-vio");
813     add_str(hypertas, "hcall-splpar");
814     add_str(hypertas, "hcall-bulk");
815     add_str(hypertas, "hcall-set-mode");
816     add_str(hypertas, "hcall-sprg0");
817     add_str(hypertas, "hcall-copy");
818     add_str(hypertas, "hcall-debug");
819     add_str(qemu_hypertas, "hcall-memop1");
820 
821     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
822         add_str(hypertas, "hcall-multi-tce");
823     }
824     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
825                      hypertas->str, hypertas->len));
826     g_string_free(hypertas, TRUE);
827     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
828                      qemu_hypertas->str, qemu_hypertas->len));
829     g_string_free(qemu_hypertas, TRUE);
830 
831     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
832                      refpoints, sizeof(refpoints)));
833 
834     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
835                           RTAS_ERROR_LOG_MAX));
836     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
837                           RTAS_EVENT_SCAN_RATE));
838 
839     if (msi_nonbroken) {
840         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
841     }
842 
843     /*
844      * According to PAPR, rtas ibm,os-term does not guarantee a return
845      * back to the guest cpu.
846      *
847      * While an additional ibm,extended-os-term property indicates
848      * that rtas call return will always occur. Set this property.
849      */
850     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
851 
852     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
853                      lrdr_capacity, sizeof(lrdr_capacity)));
854 
855     spapr_dt_rtas_tokens(fdt, rtas);
856 }
857 
858 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
859  * that the guest may request and thus the valid values for bytes 24..26 of
860  * option vector 5: */
861 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
862 {
863     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
864 
865     char val[2 * 3] = {
866         24, 0x00, /* Hash/Radix, filled in below. */
867         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
868         26, 0x40, /* Radix options: GTSE == yes. */
869     };
870 
871     if (kvm_enabled()) {
872         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
873             val[1] = 0x80; /* OV5_MMU_BOTH */
874         } else if (kvmppc_has_cap_mmu_radix()) {
875             val[1] = 0x40; /* OV5_MMU_RADIX_300 */
876         } else {
877             val[1] = 0x00; /* Hash */
878         }
879     } else {
880         if (first_ppc_cpu->env.mmu_model & POWERPC_MMU_V3) {
881             /* V3 MMU supports both hash and radix (with dynamic switching) */
882             val[1] = 0xC0;
883         } else {
884             /* Otherwise we can only do hash */
885             val[1] = 0x00;
886         }
887     }
888     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
889                      val, sizeof(val)));
890 }
891 
892 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
893 {
894     MachineState *machine = MACHINE(spapr);
895     int chosen;
896     const char *boot_device = machine->boot_order;
897     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
898     size_t cb = 0;
899     char *bootlist = get_boot_devices_list(&cb, true);
900 
901     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
902 
903     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
904     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
905                           spapr->initrd_base));
906     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
907                           spapr->initrd_base + spapr->initrd_size));
908 
909     if (spapr->kernel_size) {
910         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
911                               cpu_to_be64(spapr->kernel_size) };
912 
913         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
914                          &kprop, sizeof(kprop)));
915         if (spapr->kernel_le) {
916             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
917         }
918     }
919     if (boot_menu) {
920         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
921     }
922     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
923     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
924     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
925 
926     if (cb && bootlist) {
927         int i;
928 
929         for (i = 0; i < cb; i++) {
930             if (bootlist[i] == '\n') {
931                 bootlist[i] = ' ';
932             }
933         }
934         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
935     }
936 
937     if (boot_device && strlen(boot_device)) {
938         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
939     }
940 
941     if (!spapr->has_graphics && stdout_path) {
942         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
943     }
944 
945     spapr_dt_ov5_platform_support(fdt, chosen);
946 
947     g_free(stdout_path);
948     g_free(bootlist);
949 }
950 
951 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
952 {
953     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
954      * KVM to work under pHyp with some guest co-operation */
955     int hypervisor;
956     uint8_t hypercall[16];
957 
958     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
959     /* indicate KVM hypercall interface */
960     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
961     if (kvmppc_has_cap_fixup_hcalls()) {
962         /*
963          * Older KVM versions with older guest kernels were broken
964          * with the magic page, don't allow the guest to map it.
965          */
966         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
967                                   sizeof(hypercall))) {
968             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
969                              hypercall, sizeof(hypercall)));
970         }
971     }
972 }
973 
974 static void *spapr_build_fdt(sPAPRMachineState *spapr,
975                              hwaddr rtas_addr,
976                              hwaddr rtas_size)
977 {
978     MachineState *machine = MACHINE(qdev_get_machine());
979     MachineClass *mc = MACHINE_GET_CLASS(machine);
980     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
981     int ret;
982     void *fdt;
983     sPAPRPHBState *phb;
984     char *buf;
985     int smt = kvmppc_smt_threads();
986 
987     fdt = g_malloc0(FDT_MAX_SIZE);
988     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
989 
990     /* Root node */
991     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
992     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
993     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
994 
995     /*
996      * Add info to guest to indentify which host is it being run on
997      * and what is the uuid of the guest
998      */
999     if (kvmppc_get_host_model(&buf)) {
1000         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1001         g_free(buf);
1002     }
1003     if (kvmppc_get_host_serial(&buf)) {
1004         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1005         g_free(buf);
1006     }
1007 
1008     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1009 
1010     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1011     if (qemu_uuid_set) {
1012         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1013     }
1014     g_free(buf);
1015 
1016     if (qemu_get_vm_name()) {
1017         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1018                                 qemu_get_vm_name()));
1019     }
1020 
1021     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1022     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1023 
1024     /* /interrupt controller */
1025     spapr_dt_xics(DIV_ROUND_UP(max_cpus * smt, smp_threads), fdt, PHANDLE_XICP);
1026 
1027     ret = spapr_populate_memory(spapr, fdt);
1028     if (ret < 0) {
1029         error_report("couldn't setup memory nodes in fdt");
1030         exit(1);
1031     }
1032 
1033     /* /vdevice */
1034     spapr_dt_vdevice(spapr->vio_bus, fdt);
1035 
1036     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1037         ret = spapr_rng_populate_dt(fdt);
1038         if (ret < 0) {
1039             error_report("could not set up rng device in the fdt");
1040             exit(1);
1041         }
1042     }
1043 
1044     QLIST_FOREACH(phb, &spapr->phbs, list) {
1045         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1046         if (ret < 0) {
1047             error_report("couldn't setup PCI devices in fdt");
1048             exit(1);
1049         }
1050     }
1051 
1052     /* cpus */
1053     spapr_populate_cpus_dt_node(fdt, spapr);
1054 
1055     if (smc->dr_lmb_enabled) {
1056         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1057     }
1058 
1059     if (mc->has_hotpluggable_cpus) {
1060         int offset = fdt_path_offset(fdt, "/cpus");
1061         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1062                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1063         if (ret < 0) {
1064             error_report("Couldn't set up CPU DR device tree properties");
1065             exit(1);
1066         }
1067     }
1068 
1069     /* /event-sources */
1070     spapr_dt_events(spapr, fdt);
1071 
1072     /* /rtas */
1073     spapr_dt_rtas(spapr, fdt);
1074 
1075     /* /chosen */
1076     spapr_dt_chosen(spapr, fdt);
1077 
1078     /* /hypervisor */
1079     if (kvm_enabled()) {
1080         spapr_dt_hypervisor(spapr, fdt);
1081     }
1082 
1083     /* Build memory reserve map */
1084     if (spapr->kernel_size) {
1085         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1086     }
1087     if (spapr->initrd_size) {
1088         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1089     }
1090 
1091     /* ibm,client-architecture-support updates */
1092     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1093     if (ret < 0) {
1094         error_report("couldn't setup CAS properties fdt");
1095         exit(1);
1096     }
1097 
1098     return fdt;
1099 }
1100 
1101 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1102 {
1103     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1104 }
1105 
1106 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1107                                     PowerPCCPU *cpu)
1108 {
1109     CPUPPCState *env = &cpu->env;
1110 
1111     /* The TCG path should also be holding the BQL at this point */
1112     g_assert(qemu_mutex_iothread_locked());
1113 
1114     if (msr_pr) {
1115         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1116         env->gpr[3] = H_PRIVILEGE;
1117     } else {
1118         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1119     }
1120 }
1121 
1122 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1123 {
1124     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1125 
1126     return spapr->patb_entry;
1127 }
1128 
1129 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1130 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1131 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1132 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1133 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1134 
1135 /*
1136  * Get the fd to access the kernel htab, re-opening it if necessary
1137  */
1138 static int get_htab_fd(sPAPRMachineState *spapr)
1139 {
1140     if (spapr->htab_fd >= 0) {
1141         return spapr->htab_fd;
1142     }
1143 
1144     spapr->htab_fd = kvmppc_get_htab_fd(false);
1145     if (spapr->htab_fd < 0) {
1146         error_report("Unable to open fd for reading hash table from KVM: %s",
1147                      strerror(errno));
1148     }
1149 
1150     return spapr->htab_fd;
1151 }
1152 
1153 void close_htab_fd(sPAPRMachineState *spapr)
1154 {
1155     if (spapr->htab_fd >= 0) {
1156         close(spapr->htab_fd);
1157     }
1158     spapr->htab_fd = -1;
1159 }
1160 
1161 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1162 {
1163     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1164 
1165     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1166 }
1167 
1168 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1169                                                 hwaddr ptex, int n)
1170 {
1171     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1172     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1173 
1174     if (!spapr->htab) {
1175         /*
1176          * HTAB is controlled by KVM. Fetch into temporary buffer
1177          */
1178         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1179         kvmppc_read_hptes(hptes, ptex, n);
1180         return hptes;
1181     }
1182 
1183     /*
1184      * HTAB is controlled by QEMU. Just point to the internally
1185      * accessible PTEG.
1186      */
1187     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1188 }
1189 
1190 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1191                               const ppc_hash_pte64_t *hptes,
1192                               hwaddr ptex, int n)
1193 {
1194     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1195 
1196     if (!spapr->htab) {
1197         g_free((void *)hptes);
1198     }
1199 
1200     /* Nothing to do for qemu managed HPT */
1201 }
1202 
1203 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1204                              uint64_t pte0, uint64_t pte1)
1205 {
1206     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1207     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1208 
1209     if (!spapr->htab) {
1210         kvmppc_write_hpte(ptex, pte0, pte1);
1211     } else {
1212         stq_p(spapr->htab + offset, pte0);
1213         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1214     }
1215 }
1216 
1217 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1218 {
1219     int shift;
1220 
1221     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1222      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1223      * that's much more than is needed for Linux guests */
1224     shift = ctz64(pow2ceil(ramsize)) - 7;
1225     shift = MAX(shift, 18); /* Minimum architected size */
1226     shift = MIN(shift, 46); /* Maximum architected size */
1227     return shift;
1228 }
1229 
1230 void spapr_free_hpt(sPAPRMachineState *spapr)
1231 {
1232     g_free(spapr->htab);
1233     spapr->htab = NULL;
1234     spapr->htab_shift = 0;
1235     close_htab_fd(spapr);
1236 }
1237 
1238 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1239                                  Error **errp)
1240 {
1241     long rc;
1242 
1243     /* Clean up any HPT info from a previous boot */
1244     spapr_free_hpt(spapr);
1245 
1246     rc = kvmppc_reset_htab(shift);
1247     if (rc < 0) {
1248         /* kernel-side HPT needed, but couldn't allocate one */
1249         error_setg_errno(errp, errno,
1250                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1251                          shift);
1252         /* This is almost certainly fatal, but if the caller really
1253          * wants to carry on with shift == 0, it's welcome to try */
1254     } else if (rc > 0) {
1255         /* kernel-side HPT allocated */
1256         if (rc != shift) {
1257             error_setg(errp,
1258                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1259                        shift, rc);
1260         }
1261 
1262         spapr->htab_shift = shift;
1263         spapr->htab = NULL;
1264     } else {
1265         /* kernel-side HPT not needed, allocate in userspace instead */
1266         size_t size = 1ULL << shift;
1267         int i;
1268 
1269         spapr->htab = qemu_memalign(size, size);
1270         if (!spapr->htab) {
1271             error_setg_errno(errp, errno,
1272                              "Could not allocate HPT of order %d", shift);
1273             return;
1274         }
1275 
1276         memset(spapr->htab, 0, size);
1277         spapr->htab_shift = shift;
1278 
1279         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1280             DIRTY_HPTE(HPTE(spapr->htab, i));
1281         }
1282     }
1283 }
1284 
1285 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1286 {
1287     spapr_reallocate_hpt(spapr,
1288                      spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size),
1289                      &error_fatal);
1290     if (spapr->vrma_adjust) {
1291         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1292                                           spapr->htab_shift);
1293     }
1294     /* We're setting up a hash table, so that means we're not radix */
1295     spapr->patb_entry = 0;
1296 }
1297 
1298 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1299 {
1300     bool matched = false;
1301 
1302     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1303         matched = true;
1304     }
1305 
1306     if (!matched) {
1307         error_report("Device %s is not supported by this machine yet.",
1308                      qdev_fw_name(DEVICE(sbdev)));
1309         exit(1);
1310     }
1311 }
1312 
1313 static void ppc_spapr_reset(void)
1314 {
1315     MachineState *machine = MACHINE(qdev_get_machine());
1316     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1317     PowerPCCPU *first_ppc_cpu;
1318     uint32_t rtas_limit;
1319     hwaddr rtas_addr, fdt_addr;
1320     void *fdt;
1321     int rc;
1322 
1323     /* Check for unknown sysbus devices */
1324     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1325 
1326     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1327         /* If using KVM with radix mode available, VCPUs can be started
1328          * without a HPT because KVM will start them in radix mode.
1329          * Set the GR bit in PATB so that we know there is no HPT. */
1330         spapr->patb_entry = PATBE1_GR;
1331     } else {
1332         spapr->patb_entry = 0;
1333         spapr_setup_hpt_and_vrma(spapr);
1334     }
1335 
1336     qemu_devices_reset();
1337 
1338     /*
1339      * We place the device tree and RTAS just below either the top of the RMA,
1340      * or just below 2GB, whichever is lowere, so that it can be
1341      * processed with 32-bit real mode code if necessary
1342      */
1343     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1344     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1345     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1346 
1347     /* if this reset wasn't generated by CAS, we should reset our
1348      * negotiated options and start from scratch */
1349     if (!spapr->cas_reboot) {
1350         spapr_ovec_cleanup(spapr->ov5_cas);
1351         spapr->ov5_cas = spapr_ovec_new();
1352     }
1353 
1354     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1355 
1356     spapr_load_rtas(spapr, fdt, rtas_addr);
1357 
1358     rc = fdt_pack(fdt);
1359 
1360     /* Should only fail if we've built a corrupted tree */
1361     assert(rc == 0);
1362 
1363     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1364         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1365                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1366         exit(1);
1367     }
1368 
1369     /* Load the fdt */
1370     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1371     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1372     g_free(fdt);
1373 
1374     /* Set up the entry state */
1375     first_ppc_cpu = POWERPC_CPU(first_cpu);
1376     first_ppc_cpu->env.gpr[3] = fdt_addr;
1377     first_ppc_cpu->env.gpr[5] = 0;
1378     first_cpu->halted = 0;
1379     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1380 
1381     spapr->cas_reboot = false;
1382 }
1383 
1384 static void spapr_create_nvram(sPAPRMachineState *spapr)
1385 {
1386     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1387     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1388 
1389     if (dinfo) {
1390         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1391                             &error_fatal);
1392     }
1393 
1394     qdev_init_nofail(dev);
1395 
1396     spapr->nvram = (struct sPAPRNVRAM *)dev;
1397 }
1398 
1399 static void spapr_rtc_create(sPAPRMachineState *spapr)
1400 {
1401     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1402     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1403                               &error_fatal);
1404     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1405                               &error_fatal);
1406     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1407                               "date", &error_fatal);
1408 }
1409 
1410 /* Returns whether we want to use VGA or not */
1411 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1412 {
1413     switch (vga_interface_type) {
1414     case VGA_NONE:
1415         return false;
1416     case VGA_DEVICE:
1417         return true;
1418     case VGA_STD:
1419     case VGA_VIRTIO:
1420         return pci_vga_init(pci_bus) != NULL;
1421     default:
1422         error_setg(errp,
1423                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1424         return false;
1425     }
1426 }
1427 
1428 static int spapr_post_load(void *opaque, int version_id)
1429 {
1430     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1431     int err = 0;
1432 
1433     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1434         CPUState *cs;
1435         CPU_FOREACH(cs) {
1436             PowerPCCPU *cpu = POWERPC_CPU(cs);
1437             icp_resend(ICP(cpu->intc));
1438         }
1439     }
1440 
1441     /* In earlier versions, there was no separate qdev for the PAPR
1442      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1443      * So when migrating from those versions, poke the incoming offset
1444      * value into the RTC device */
1445     if (version_id < 3) {
1446         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1447     }
1448 
1449     return err;
1450 }
1451 
1452 static bool version_before_3(void *opaque, int version_id)
1453 {
1454     return version_id < 3;
1455 }
1456 
1457 static bool spapr_ov5_cas_needed(void *opaque)
1458 {
1459     sPAPRMachineState *spapr = opaque;
1460     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1461     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1462     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1463     bool cas_needed;
1464 
1465     /* Prior to the introduction of sPAPROptionVector, we had two option
1466      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1467      * Both of these options encode machine topology into the device-tree
1468      * in such a way that the now-booted OS should still be able to interact
1469      * appropriately with QEMU regardless of what options were actually
1470      * negotiatied on the source side.
1471      *
1472      * As such, we can avoid migrating the CAS-negotiated options if these
1473      * are the only options available on the current machine/platform.
1474      * Since these are the only options available for pseries-2.7 and
1475      * earlier, this allows us to maintain old->new/new->old migration
1476      * compatibility.
1477      *
1478      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1479      * via default pseries-2.8 machines and explicit command-line parameters.
1480      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1481      * of the actual CAS-negotiated values to continue working properly. For
1482      * example, availability of memory unplug depends on knowing whether
1483      * OV5_HP_EVT was negotiated via CAS.
1484      *
1485      * Thus, for any cases where the set of available CAS-negotiatable
1486      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1487      * include the CAS-negotiated options in the migration stream.
1488      */
1489     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1490     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1491 
1492     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1493      * the mask itself since in the future it's possible "legacy" bits may be
1494      * removed via machine options, which could generate a false positive
1495      * that breaks migration.
1496      */
1497     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1498     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1499 
1500     spapr_ovec_cleanup(ov5_mask);
1501     spapr_ovec_cleanup(ov5_legacy);
1502     spapr_ovec_cleanup(ov5_removed);
1503 
1504     return cas_needed;
1505 }
1506 
1507 static const VMStateDescription vmstate_spapr_ov5_cas = {
1508     .name = "spapr_option_vector_ov5_cas",
1509     .version_id = 1,
1510     .minimum_version_id = 1,
1511     .needed = spapr_ov5_cas_needed,
1512     .fields = (VMStateField[]) {
1513         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1514                                  vmstate_spapr_ovec, sPAPROptionVector),
1515         VMSTATE_END_OF_LIST()
1516     },
1517 };
1518 
1519 static bool spapr_patb_entry_needed(void *opaque)
1520 {
1521     sPAPRMachineState *spapr = opaque;
1522 
1523     return !!spapr->patb_entry;
1524 }
1525 
1526 static const VMStateDescription vmstate_spapr_patb_entry = {
1527     .name = "spapr_patb_entry",
1528     .version_id = 1,
1529     .minimum_version_id = 1,
1530     .needed = spapr_patb_entry_needed,
1531     .fields = (VMStateField[]) {
1532         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1533         VMSTATE_END_OF_LIST()
1534     },
1535 };
1536 
1537 static const VMStateDescription vmstate_spapr = {
1538     .name = "spapr",
1539     .version_id = 3,
1540     .minimum_version_id = 1,
1541     .post_load = spapr_post_load,
1542     .fields = (VMStateField[]) {
1543         /* used to be @next_irq */
1544         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1545 
1546         /* RTC offset */
1547         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1548 
1549         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1550         VMSTATE_END_OF_LIST()
1551     },
1552     .subsections = (const VMStateDescription*[]) {
1553         &vmstate_spapr_ov5_cas,
1554         &vmstate_spapr_patb_entry,
1555         NULL
1556     }
1557 };
1558 
1559 static int htab_save_setup(QEMUFile *f, void *opaque)
1560 {
1561     sPAPRMachineState *spapr = opaque;
1562 
1563     /* "Iteration" header */
1564     qemu_put_be32(f, spapr->htab_shift);
1565 
1566     if (spapr->htab) {
1567         spapr->htab_save_index = 0;
1568         spapr->htab_first_pass = true;
1569     } else {
1570         assert(kvm_enabled());
1571     }
1572 
1573 
1574     return 0;
1575 }
1576 
1577 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1578                                  int64_t max_ns)
1579 {
1580     bool has_timeout = max_ns != -1;
1581     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1582     int index = spapr->htab_save_index;
1583     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1584 
1585     assert(spapr->htab_first_pass);
1586 
1587     do {
1588         int chunkstart;
1589 
1590         /* Consume invalid HPTEs */
1591         while ((index < htabslots)
1592                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1593             CLEAN_HPTE(HPTE(spapr->htab, index));
1594             index++;
1595         }
1596 
1597         /* Consume valid HPTEs */
1598         chunkstart = index;
1599         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1600                && HPTE_VALID(HPTE(spapr->htab, index))) {
1601             CLEAN_HPTE(HPTE(spapr->htab, index));
1602             index++;
1603         }
1604 
1605         if (index > chunkstart) {
1606             int n_valid = index - chunkstart;
1607 
1608             qemu_put_be32(f, chunkstart);
1609             qemu_put_be16(f, n_valid);
1610             qemu_put_be16(f, 0);
1611             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1612                             HASH_PTE_SIZE_64 * n_valid);
1613 
1614             if (has_timeout &&
1615                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1616                 break;
1617             }
1618         }
1619     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1620 
1621     if (index >= htabslots) {
1622         assert(index == htabslots);
1623         index = 0;
1624         spapr->htab_first_pass = false;
1625     }
1626     spapr->htab_save_index = index;
1627 }
1628 
1629 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1630                                 int64_t max_ns)
1631 {
1632     bool final = max_ns < 0;
1633     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1634     int examined = 0, sent = 0;
1635     int index = spapr->htab_save_index;
1636     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1637 
1638     assert(!spapr->htab_first_pass);
1639 
1640     do {
1641         int chunkstart, invalidstart;
1642 
1643         /* Consume non-dirty HPTEs */
1644         while ((index < htabslots)
1645                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1646             index++;
1647             examined++;
1648         }
1649 
1650         chunkstart = index;
1651         /* Consume valid dirty HPTEs */
1652         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1653                && HPTE_DIRTY(HPTE(spapr->htab, index))
1654                && HPTE_VALID(HPTE(spapr->htab, index))) {
1655             CLEAN_HPTE(HPTE(spapr->htab, index));
1656             index++;
1657             examined++;
1658         }
1659 
1660         invalidstart = index;
1661         /* Consume invalid dirty HPTEs */
1662         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1663                && HPTE_DIRTY(HPTE(spapr->htab, index))
1664                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1665             CLEAN_HPTE(HPTE(spapr->htab, index));
1666             index++;
1667             examined++;
1668         }
1669 
1670         if (index > chunkstart) {
1671             int n_valid = invalidstart - chunkstart;
1672             int n_invalid = index - invalidstart;
1673 
1674             qemu_put_be32(f, chunkstart);
1675             qemu_put_be16(f, n_valid);
1676             qemu_put_be16(f, n_invalid);
1677             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1678                             HASH_PTE_SIZE_64 * n_valid);
1679             sent += index - chunkstart;
1680 
1681             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1682                 break;
1683             }
1684         }
1685 
1686         if (examined >= htabslots) {
1687             break;
1688         }
1689 
1690         if (index >= htabslots) {
1691             assert(index == htabslots);
1692             index = 0;
1693         }
1694     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1695 
1696     if (index >= htabslots) {
1697         assert(index == htabslots);
1698         index = 0;
1699     }
1700 
1701     spapr->htab_save_index = index;
1702 
1703     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1704 }
1705 
1706 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1707 #define MAX_KVM_BUF_SIZE    2048
1708 
1709 static int htab_save_iterate(QEMUFile *f, void *opaque)
1710 {
1711     sPAPRMachineState *spapr = opaque;
1712     int fd;
1713     int rc = 0;
1714 
1715     /* Iteration header */
1716     qemu_put_be32(f, 0);
1717 
1718     if (!spapr->htab) {
1719         assert(kvm_enabled());
1720 
1721         fd = get_htab_fd(spapr);
1722         if (fd < 0) {
1723             return fd;
1724         }
1725 
1726         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1727         if (rc < 0) {
1728             return rc;
1729         }
1730     } else  if (spapr->htab_first_pass) {
1731         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1732     } else {
1733         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1734     }
1735 
1736     /* End marker */
1737     qemu_put_be32(f, 0);
1738     qemu_put_be16(f, 0);
1739     qemu_put_be16(f, 0);
1740 
1741     return rc;
1742 }
1743 
1744 static int htab_save_complete(QEMUFile *f, void *opaque)
1745 {
1746     sPAPRMachineState *spapr = opaque;
1747     int fd;
1748 
1749     /* Iteration header */
1750     qemu_put_be32(f, 0);
1751 
1752     if (!spapr->htab) {
1753         int rc;
1754 
1755         assert(kvm_enabled());
1756 
1757         fd = get_htab_fd(spapr);
1758         if (fd < 0) {
1759             return fd;
1760         }
1761 
1762         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1763         if (rc < 0) {
1764             return rc;
1765         }
1766     } else {
1767         if (spapr->htab_first_pass) {
1768             htab_save_first_pass(f, spapr, -1);
1769         }
1770         htab_save_later_pass(f, spapr, -1);
1771     }
1772 
1773     /* End marker */
1774     qemu_put_be32(f, 0);
1775     qemu_put_be16(f, 0);
1776     qemu_put_be16(f, 0);
1777 
1778     return 0;
1779 }
1780 
1781 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1782 {
1783     sPAPRMachineState *spapr = opaque;
1784     uint32_t section_hdr;
1785     int fd = -1;
1786 
1787     if (version_id < 1 || version_id > 1) {
1788         error_report("htab_load() bad version");
1789         return -EINVAL;
1790     }
1791 
1792     section_hdr = qemu_get_be32(f);
1793 
1794     if (section_hdr) {
1795         Error *local_err = NULL;
1796 
1797         /* First section gives the htab size */
1798         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1799         if (local_err) {
1800             error_report_err(local_err);
1801             return -EINVAL;
1802         }
1803         return 0;
1804     }
1805 
1806     if (!spapr->htab) {
1807         assert(kvm_enabled());
1808 
1809         fd = kvmppc_get_htab_fd(true);
1810         if (fd < 0) {
1811             error_report("Unable to open fd to restore KVM hash table: %s",
1812                          strerror(errno));
1813         }
1814     }
1815 
1816     while (true) {
1817         uint32_t index;
1818         uint16_t n_valid, n_invalid;
1819 
1820         index = qemu_get_be32(f);
1821         n_valid = qemu_get_be16(f);
1822         n_invalid = qemu_get_be16(f);
1823 
1824         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1825             /* End of Stream */
1826             break;
1827         }
1828 
1829         if ((index + n_valid + n_invalid) >
1830             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1831             /* Bad index in stream */
1832             error_report(
1833                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1834                 index, n_valid, n_invalid, spapr->htab_shift);
1835             return -EINVAL;
1836         }
1837 
1838         if (spapr->htab) {
1839             if (n_valid) {
1840                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1841                                 HASH_PTE_SIZE_64 * n_valid);
1842             }
1843             if (n_invalid) {
1844                 memset(HPTE(spapr->htab, index + n_valid), 0,
1845                        HASH_PTE_SIZE_64 * n_invalid);
1846             }
1847         } else {
1848             int rc;
1849 
1850             assert(fd >= 0);
1851 
1852             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1853             if (rc < 0) {
1854                 return rc;
1855             }
1856         }
1857     }
1858 
1859     if (!spapr->htab) {
1860         assert(fd >= 0);
1861         close(fd);
1862     }
1863 
1864     return 0;
1865 }
1866 
1867 static void htab_cleanup(void *opaque)
1868 {
1869     sPAPRMachineState *spapr = opaque;
1870 
1871     close_htab_fd(spapr);
1872 }
1873 
1874 static SaveVMHandlers savevm_htab_handlers = {
1875     .save_live_setup = htab_save_setup,
1876     .save_live_iterate = htab_save_iterate,
1877     .save_live_complete_precopy = htab_save_complete,
1878     .cleanup = htab_cleanup,
1879     .load_state = htab_load,
1880 };
1881 
1882 static void spapr_boot_set(void *opaque, const char *boot_device,
1883                            Error **errp)
1884 {
1885     MachineState *machine = MACHINE(qdev_get_machine());
1886     machine->boot_order = g_strdup(boot_device);
1887 }
1888 
1889 /*
1890  * Reset routine for LMB DR devices.
1891  *
1892  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1893  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1894  * when it walks all its children devices. LMB devices reset occurs
1895  * as part of spapr_ppc_reset().
1896  */
1897 static void spapr_drc_reset(void *opaque)
1898 {
1899     sPAPRDRConnector *drc = opaque;
1900     DeviceState *d = DEVICE(drc);
1901 
1902     if (d) {
1903         device_reset(d);
1904     }
1905 }
1906 
1907 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1908 {
1909     MachineState *machine = MACHINE(spapr);
1910     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1911     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1912     int i;
1913 
1914     for (i = 0; i < nr_lmbs; i++) {
1915         sPAPRDRConnector *drc;
1916         uint64_t addr;
1917 
1918         addr = i * lmb_size + spapr->hotplug_memory.base;
1919         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1920                                      addr/lmb_size);
1921         qemu_register_reset(spapr_drc_reset, drc);
1922     }
1923 }
1924 
1925 /*
1926  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1927  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1928  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1929  */
1930 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1931 {
1932     int i;
1933 
1934     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1935         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1936                    " is not aligned to %llu MiB",
1937                    machine->ram_size,
1938                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1939         return;
1940     }
1941 
1942     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1943         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1944                    " is not aligned to %llu MiB",
1945                    machine->ram_size,
1946                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1947         return;
1948     }
1949 
1950     for (i = 0; i < nb_numa_nodes; i++) {
1951         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1952             error_setg(errp,
1953                        "Node %d memory size 0x%" PRIx64
1954                        " is not aligned to %llu MiB",
1955                        i, numa_info[i].node_mem,
1956                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1957             return;
1958         }
1959     }
1960 }
1961 
1962 /* find cpu slot in machine->possible_cpus by core_id */
1963 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1964 {
1965     int index = id / smp_threads;
1966 
1967     if (index >= ms->possible_cpus->len) {
1968         return NULL;
1969     }
1970     if (idx) {
1971         *idx = index;
1972     }
1973     return &ms->possible_cpus->cpus[index];
1974 }
1975 
1976 static void spapr_init_cpus(sPAPRMachineState *spapr)
1977 {
1978     MachineState *machine = MACHINE(spapr);
1979     MachineClass *mc = MACHINE_GET_CLASS(machine);
1980     char *type = spapr_get_cpu_core_type(machine->cpu_model);
1981     int smt = kvmppc_smt_threads();
1982     const CPUArchIdList *possible_cpus;
1983     int boot_cores_nr = smp_cpus / smp_threads;
1984     int i;
1985 
1986     if (!type) {
1987         error_report("Unable to find sPAPR CPU Core definition");
1988         exit(1);
1989     }
1990 
1991     possible_cpus = mc->possible_cpu_arch_ids(machine);
1992     if (mc->has_hotpluggable_cpus) {
1993         if (smp_cpus % smp_threads) {
1994             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1995                          smp_cpus, smp_threads);
1996             exit(1);
1997         }
1998         if (max_cpus % smp_threads) {
1999             error_report("max_cpus (%u) must be multiple of threads (%u)",
2000                          max_cpus, smp_threads);
2001             exit(1);
2002         }
2003     } else {
2004         if (max_cpus != smp_cpus) {
2005             error_report("This machine version does not support CPU hotplug");
2006             exit(1);
2007         }
2008         boot_cores_nr = possible_cpus->len;
2009     }
2010 
2011     for (i = 0; i < possible_cpus->len; i++) {
2012         int core_id = i * smp_threads;
2013 
2014         if (mc->has_hotpluggable_cpus) {
2015             sPAPRDRConnector *drc =
2016                 spapr_dr_connector_new(OBJECT(spapr),
2017                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
2018                                        (core_id / smp_threads) * smt);
2019 
2020             qemu_register_reset(spapr_drc_reset, drc);
2021         }
2022 
2023         if (i < boot_cores_nr) {
2024             Object *core  = object_new(type);
2025             int nr_threads = smp_threads;
2026 
2027             /* Handle the partially filled core for older machine types */
2028             if ((i + 1) * smp_threads >= smp_cpus) {
2029                 nr_threads = smp_cpus - i * smp_threads;
2030             }
2031 
2032             object_property_set_int(core, nr_threads, "nr-threads",
2033                                     &error_fatal);
2034             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2035                                     &error_fatal);
2036             object_property_set_bool(core, true, "realized", &error_fatal);
2037         }
2038     }
2039     g_free(type);
2040 }
2041 
2042 /* pSeries LPAR / sPAPR hardware init */
2043 static void ppc_spapr_init(MachineState *machine)
2044 {
2045     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2046     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2047     const char *kernel_filename = machine->kernel_filename;
2048     const char *initrd_filename = machine->initrd_filename;
2049     PCIHostState *phb;
2050     int i;
2051     MemoryRegion *sysmem = get_system_memory();
2052     MemoryRegion *ram = g_new(MemoryRegion, 1);
2053     MemoryRegion *rma_region;
2054     void *rma = NULL;
2055     hwaddr rma_alloc_size;
2056     hwaddr node0_size = spapr_node0_size();
2057     long load_limit, fw_size;
2058     char *filename;
2059 
2060     msi_nonbroken = true;
2061 
2062     QLIST_INIT(&spapr->phbs);
2063 
2064     /* Allocate RMA if necessary */
2065     rma_alloc_size = kvmppc_alloc_rma(&rma);
2066 
2067     if (rma_alloc_size == -1) {
2068         error_report("Unable to create RMA");
2069         exit(1);
2070     }
2071 
2072     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2073         spapr->rma_size = rma_alloc_size;
2074     } else {
2075         spapr->rma_size = node0_size;
2076 
2077         /* With KVM, we don't actually know whether KVM supports an
2078          * unbounded RMA (PR KVM) or is limited by the hash table size
2079          * (HV KVM using VRMA), so we always assume the latter
2080          *
2081          * In that case, we also limit the initial allocations for RTAS
2082          * etc... to 256M since we have no way to know what the VRMA size
2083          * is going to be as it depends on the size of the hash table
2084          * isn't determined yet.
2085          */
2086         if (kvm_enabled()) {
2087             spapr->vrma_adjust = 1;
2088             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2089         }
2090 
2091         /* Actually we don't support unbounded RMA anymore since we
2092          * added proper emulation of HV mode. The max we can get is
2093          * 16G which also happens to be what we configure for PAPR
2094          * mode so make sure we don't do anything bigger than that
2095          */
2096         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2097     }
2098 
2099     if (spapr->rma_size > node0_size) {
2100         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2101                      spapr->rma_size);
2102         exit(1);
2103     }
2104 
2105     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2106     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2107 
2108     /* Set up Interrupt Controller before we create the VCPUs */
2109     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2110 
2111     /* Set up containers for ibm,client-set-architecture negotiated options */
2112     spapr->ov5 = spapr_ovec_new();
2113     spapr->ov5_cas = spapr_ovec_new();
2114 
2115     if (smc->dr_lmb_enabled) {
2116         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2117         spapr_validate_node_memory(machine, &error_fatal);
2118     }
2119 
2120     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2121     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2122         /* KVM and TCG always allow GTSE with radix... */
2123         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2124     }
2125     /* ... but not with hash (currently). */
2126 
2127     /* advertise support for dedicated HP event source to guests */
2128     if (spapr->use_hotplug_event_source) {
2129         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2130     }
2131 
2132     /* init CPUs */
2133     if (machine->cpu_model == NULL) {
2134         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2135     }
2136 
2137     ppc_cpu_parse_features(machine->cpu_model);
2138 
2139     spapr_init_cpus(spapr);
2140 
2141     if (kvm_enabled()) {
2142         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2143         kvmppc_enable_logical_ci_hcalls();
2144         kvmppc_enable_set_mode_hcall();
2145 
2146         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2147         kvmppc_enable_clear_ref_mod_hcalls();
2148     }
2149 
2150     /* allocate RAM */
2151     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2152                                          machine->ram_size);
2153     memory_region_add_subregion(sysmem, 0, ram);
2154 
2155     if (rma_alloc_size && rma) {
2156         rma_region = g_new(MemoryRegion, 1);
2157         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2158                                    rma_alloc_size, rma);
2159         vmstate_register_ram_global(rma_region);
2160         memory_region_add_subregion(sysmem, 0, rma_region);
2161     }
2162 
2163     /* initialize hotplug memory address space */
2164     if (machine->ram_size < machine->maxram_size) {
2165         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2166         /*
2167          * Limit the number of hotpluggable memory slots to half the number
2168          * slots that KVM supports, leaving the other half for PCI and other
2169          * devices. However ensure that number of slots doesn't drop below 32.
2170          */
2171         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2172                            SPAPR_MAX_RAM_SLOTS;
2173 
2174         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2175             max_memslots = SPAPR_MAX_RAM_SLOTS;
2176         }
2177         if (machine->ram_slots > max_memslots) {
2178             error_report("Specified number of memory slots %"
2179                          PRIu64" exceeds max supported %d",
2180                          machine->ram_slots, max_memslots);
2181             exit(1);
2182         }
2183 
2184         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2185                                               SPAPR_HOTPLUG_MEM_ALIGN);
2186         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2187                            "hotplug-memory", hotplug_mem_size);
2188         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2189                                     &spapr->hotplug_memory.mr);
2190     }
2191 
2192     if (smc->dr_lmb_enabled) {
2193         spapr_create_lmb_dr_connectors(spapr);
2194     }
2195 
2196     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2197     if (!filename) {
2198         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2199         exit(1);
2200     }
2201     spapr->rtas_size = get_image_size(filename);
2202     if (spapr->rtas_size < 0) {
2203         error_report("Could not get size of LPAR rtas '%s'", filename);
2204         exit(1);
2205     }
2206     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2207     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2208         error_report("Could not load LPAR rtas '%s'", filename);
2209         exit(1);
2210     }
2211     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2212         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2213                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2214         exit(1);
2215     }
2216     g_free(filename);
2217 
2218     /* Set up RTAS event infrastructure */
2219     spapr_events_init(spapr);
2220 
2221     /* Set up the RTC RTAS interfaces */
2222     spapr_rtc_create(spapr);
2223 
2224     /* Set up VIO bus */
2225     spapr->vio_bus = spapr_vio_bus_init();
2226 
2227     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2228         if (serial_hds[i]) {
2229             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2230         }
2231     }
2232 
2233     /* We always have at least the nvram device on VIO */
2234     spapr_create_nvram(spapr);
2235 
2236     /* Set up PCI */
2237     spapr_pci_rtas_init();
2238 
2239     phb = spapr_create_phb(spapr, 0);
2240 
2241     for (i = 0; i < nb_nics; i++) {
2242         NICInfo *nd = &nd_table[i];
2243 
2244         if (!nd->model) {
2245             nd->model = g_strdup("ibmveth");
2246         }
2247 
2248         if (strcmp(nd->model, "ibmveth") == 0) {
2249             spapr_vlan_create(spapr->vio_bus, nd);
2250         } else {
2251             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2252         }
2253     }
2254 
2255     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2256         spapr_vscsi_create(spapr->vio_bus);
2257     }
2258 
2259     /* Graphics */
2260     if (spapr_vga_init(phb->bus, &error_fatal)) {
2261         spapr->has_graphics = true;
2262         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2263     }
2264 
2265     if (machine->usb) {
2266         if (smc->use_ohci_by_default) {
2267             pci_create_simple(phb->bus, -1, "pci-ohci");
2268         } else {
2269             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2270         }
2271 
2272         if (spapr->has_graphics) {
2273             USBBus *usb_bus = usb_bus_find(-1);
2274 
2275             usb_create_simple(usb_bus, "usb-kbd");
2276             usb_create_simple(usb_bus, "usb-mouse");
2277         }
2278     }
2279 
2280     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2281         error_report(
2282             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2283             MIN_RMA_SLOF);
2284         exit(1);
2285     }
2286 
2287     if (kernel_filename) {
2288         uint64_t lowaddr = 0;
2289 
2290         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2291                                       NULL, NULL, &lowaddr, NULL, 1,
2292                                       PPC_ELF_MACHINE, 0, 0);
2293         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2294             spapr->kernel_size = load_elf(kernel_filename,
2295                                           translate_kernel_address, NULL, NULL,
2296                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2297                                           0, 0);
2298             spapr->kernel_le = spapr->kernel_size > 0;
2299         }
2300         if (spapr->kernel_size < 0) {
2301             error_report("error loading %s: %s", kernel_filename,
2302                          load_elf_strerror(spapr->kernel_size));
2303             exit(1);
2304         }
2305 
2306         /* load initrd */
2307         if (initrd_filename) {
2308             /* Try to locate the initrd in the gap between the kernel
2309              * and the firmware. Add a bit of space just in case
2310              */
2311             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2312                                   + 0x1ffff) & ~0xffff;
2313             spapr->initrd_size = load_image_targphys(initrd_filename,
2314                                                      spapr->initrd_base,
2315                                                      load_limit
2316                                                      - spapr->initrd_base);
2317             if (spapr->initrd_size < 0) {
2318                 error_report("could not load initial ram disk '%s'",
2319                              initrd_filename);
2320                 exit(1);
2321             }
2322         }
2323     }
2324 
2325     if (bios_name == NULL) {
2326         bios_name = FW_FILE_NAME;
2327     }
2328     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2329     if (!filename) {
2330         error_report("Could not find LPAR firmware '%s'", bios_name);
2331         exit(1);
2332     }
2333     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2334     if (fw_size <= 0) {
2335         error_report("Could not load LPAR firmware '%s'", filename);
2336         exit(1);
2337     }
2338     g_free(filename);
2339 
2340     /* FIXME: Should register things through the MachineState's qdev
2341      * interface, this is a legacy from the sPAPREnvironment structure
2342      * which predated MachineState but had a similar function */
2343     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2344     register_savevm_live(NULL, "spapr/htab", -1, 1,
2345                          &savevm_htab_handlers, spapr);
2346 
2347     /* used by RTAS */
2348     QTAILQ_INIT(&spapr->ccs_list);
2349     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2350 
2351     qemu_register_boot_set(spapr_boot_set, spapr);
2352 
2353     if (kvm_enabled()) {
2354         /* to stop and start vmclock */
2355         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2356                                          &spapr->tb);
2357 
2358         kvmppc_spapr_enable_inkernel_multitce();
2359     }
2360 }
2361 
2362 static int spapr_kvm_type(const char *vm_type)
2363 {
2364     if (!vm_type) {
2365         return 0;
2366     }
2367 
2368     if (!strcmp(vm_type, "HV")) {
2369         return 1;
2370     }
2371 
2372     if (!strcmp(vm_type, "PR")) {
2373         return 2;
2374     }
2375 
2376     error_report("Unknown kvm-type specified '%s'", vm_type);
2377     exit(1);
2378 }
2379 
2380 /*
2381  * Implementation of an interface to adjust firmware path
2382  * for the bootindex property handling.
2383  */
2384 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2385                                    DeviceState *dev)
2386 {
2387 #define CAST(type, obj, name) \
2388     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2389     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2390     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2391 
2392     if (d) {
2393         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2394         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2395         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2396 
2397         if (spapr) {
2398             /*
2399              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2400              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2401              * in the top 16 bits of the 64-bit LUN
2402              */
2403             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2404             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2405                                    (uint64_t)id << 48);
2406         } else if (virtio) {
2407             /*
2408              * We use SRP luns of the form 01000000 | (target << 8) | lun
2409              * in the top 32 bits of the 64-bit LUN
2410              * Note: the quote above is from SLOF and it is wrong,
2411              * the actual binding is:
2412              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2413              */
2414             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2415             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2416                                    (uint64_t)id << 32);
2417         } else if (usb) {
2418             /*
2419              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2420              * in the top 32 bits of the 64-bit LUN
2421              */
2422             unsigned usb_port = atoi(usb->port->path);
2423             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2424             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2425                                    (uint64_t)id << 32);
2426         }
2427     }
2428 
2429     /*
2430      * SLOF probes the USB devices, and if it recognizes that the device is a
2431      * storage device, it changes its name to "storage" instead of "usb-host",
2432      * and additionally adds a child node for the SCSI LUN, so the correct
2433      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2434      */
2435     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2436         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2437         if (usb_host_dev_is_scsi_storage(usbdev)) {
2438             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2439         }
2440     }
2441 
2442     if (phb) {
2443         /* Replace "pci" with "pci@800000020000000" */
2444         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2445     }
2446 
2447     return NULL;
2448 }
2449 
2450 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2451 {
2452     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2453 
2454     return g_strdup(spapr->kvm_type);
2455 }
2456 
2457 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2458 {
2459     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2460 
2461     g_free(spapr->kvm_type);
2462     spapr->kvm_type = g_strdup(value);
2463 }
2464 
2465 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2466 {
2467     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2468 
2469     return spapr->use_hotplug_event_source;
2470 }
2471 
2472 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2473                                             Error **errp)
2474 {
2475     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2476 
2477     spapr->use_hotplug_event_source = value;
2478 }
2479 
2480 static void spapr_machine_initfn(Object *obj)
2481 {
2482     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2483 
2484     spapr->htab_fd = -1;
2485     spapr->use_hotplug_event_source = true;
2486     object_property_add_str(obj, "kvm-type",
2487                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2488     object_property_set_description(obj, "kvm-type",
2489                                     "Specifies the KVM virtualization mode (HV, PR)",
2490                                     NULL);
2491     object_property_add_bool(obj, "modern-hotplug-events",
2492                             spapr_get_modern_hotplug_events,
2493                             spapr_set_modern_hotplug_events,
2494                             NULL);
2495     object_property_set_description(obj, "modern-hotplug-events",
2496                                     "Use dedicated hotplug event mechanism in"
2497                                     " place of standard EPOW events when possible"
2498                                     " (required for memory hot-unplug support)",
2499                                     NULL);
2500 }
2501 
2502 static void spapr_machine_finalizefn(Object *obj)
2503 {
2504     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2505 
2506     g_free(spapr->kvm_type);
2507 }
2508 
2509 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2510 {
2511     cpu_synchronize_state(cs);
2512     ppc_cpu_do_system_reset(cs);
2513 }
2514 
2515 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2516 {
2517     CPUState *cs;
2518 
2519     CPU_FOREACH(cs) {
2520         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2521     }
2522 }
2523 
2524 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2525                            uint32_t node, bool dedicated_hp_event_source,
2526                            Error **errp)
2527 {
2528     sPAPRDRConnector *drc;
2529     sPAPRDRConnectorClass *drck;
2530     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2531     int i, fdt_offset, fdt_size;
2532     void *fdt;
2533     uint64_t addr = addr_start;
2534 
2535     for (i = 0; i < nr_lmbs; i++) {
2536         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2537                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2538         g_assert(drc);
2539 
2540         fdt = create_device_tree(&fdt_size);
2541         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2542                                                 SPAPR_MEMORY_BLOCK_SIZE);
2543 
2544         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2545         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2546         addr += SPAPR_MEMORY_BLOCK_SIZE;
2547         if (!dev->hotplugged) {
2548             /* guests expect coldplugged LMBs to be pre-allocated */
2549             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2550             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2551         }
2552     }
2553     /* send hotplug notification to the
2554      * guest only in case of hotplugged memory
2555      */
2556     if (dev->hotplugged) {
2557         if (dedicated_hp_event_source) {
2558             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2559                     addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2560             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2561             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2562                                                    nr_lmbs,
2563                                                    drck->get_index(drc));
2564         } else {
2565             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2566                                            nr_lmbs);
2567         }
2568     }
2569 }
2570 
2571 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2572                               uint32_t node, Error **errp)
2573 {
2574     Error *local_err = NULL;
2575     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2576     PCDIMMDevice *dimm = PC_DIMM(dev);
2577     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2578     MemoryRegion *mr = ddc->get_memory_region(dimm);
2579     uint64_t align = memory_region_get_alignment(mr);
2580     uint64_t size = memory_region_size(mr);
2581     uint64_t addr;
2582     char *mem_dev;
2583 
2584     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2585         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2586                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2587         goto out;
2588     }
2589 
2590     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2591     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2592         error_setg(&local_err, "Memory backend has bad page size. "
2593                    "Use 'memory-backend-file' with correct mem-path.");
2594         goto out;
2595     }
2596 
2597     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2598     if (local_err) {
2599         goto out;
2600     }
2601 
2602     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2603     if (local_err) {
2604         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2605         goto out;
2606     }
2607 
2608     spapr_add_lmbs(dev, addr, size, node,
2609                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2610                    &error_abort);
2611 
2612 out:
2613     error_propagate(errp, local_err);
2614 }
2615 
2616 typedef struct sPAPRDIMMState {
2617     uint32_t nr_lmbs;
2618 } sPAPRDIMMState;
2619 
2620 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2621 {
2622     sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2623     HotplugHandler *hotplug_ctrl;
2624 
2625     if (--ds->nr_lmbs) {
2626         return;
2627     }
2628 
2629     g_free(ds);
2630 
2631     /*
2632      * Now that all the LMBs have been removed by the guest, call the
2633      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2634      */
2635     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2636     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2637 }
2638 
2639 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2640                            Error **errp)
2641 {
2642     sPAPRDRConnector *drc;
2643     sPAPRDRConnectorClass *drck;
2644     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2645     int i;
2646     sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2647     uint64_t addr = addr_start;
2648 
2649     ds->nr_lmbs = nr_lmbs;
2650     for (i = 0; i < nr_lmbs; i++) {
2651         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2652                 addr / SPAPR_MEMORY_BLOCK_SIZE);
2653         g_assert(drc);
2654 
2655         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2656         drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2657         addr += SPAPR_MEMORY_BLOCK_SIZE;
2658     }
2659 
2660     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2661                                    addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2662     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2663     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2664                                               nr_lmbs,
2665                                               drck->get_index(drc));
2666 }
2667 
2668 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2669                                 Error **errp)
2670 {
2671     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2672     PCDIMMDevice *dimm = PC_DIMM(dev);
2673     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2674     MemoryRegion *mr = ddc->get_memory_region(dimm);
2675 
2676     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2677     object_unparent(OBJECT(dev));
2678 }
2679 
2680 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2681                                         DeviceState *dev, Error **errp)
2682 {
2683     Error *local_err = NULL;
2684     PCDIMMDevice *dimm = PC_DIMM(dev);
2685     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2686     MemoryRegion *mr = ddc->get_memory_region(dimm);
2687     uint64_t size = memory_region_size(mr);
2688     uint64_t addr;
2689 
2690     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2691     if (local_err) {
2692         goto out;
2693     }
2694 
2695     spapr_del_lmbs(dev, addr, size, &error_abort);
2696 out:
2697     error_propagate(errp, local_err);
2698 }
2699 
2700 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2701                                     sPAPRMachineState *spapr)
2702 {
2703     PowerPCCPU *cpu = POWERPC_CPU(cs);
2704     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2705     int id = ppc_get_vcpu_dt_id(cpu);
2706     void *fdt;
2707     int offset, fdt_size;
2708     char *nodename;
2709 
2710     fdt = create_device_tree(&fdt_size);
2711     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2712     offset = fdt_add_subnode(fdt, 0, nodename);
2713 
2714     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2715     g_free(nodename);
2716 
2717     *fdt_offset = offset;
2718     return fdt;
2719 }
2720 
2721 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2722                               Error **errp)
2723 {
2724     MachineState *ms = MACHINE(qdev_get_machine());
2725     CPUCore *cc = CPU_CORE(dev);
2726     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2727 
2728     core_slot->cpu = NULL;
2729     object_unparent(OBJECT(dev));
2730 }
2731 
2732 static void spapr_core_release(DeviceState *dev, void *opaque)
2733 {
2734     HotplugHandler *hotplug_ctrl;
2735 
2736     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2737     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2738 }
2739 
2740 static
2741 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2742                                Error **errp)
2743 {
2744     int index;
2745     sPAPRDRConnector *drc;
2746     sPAPRDRConnectorClass *drck;
2747     Error *local_err = NULL;
2748     CPUCore *cc = CPU_CORE(dev);
2749     int smt = kvmppc_smt_threads();
2750 
2751     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2752         error_setg(errp, "Unable to find CPU core with core-id: %d",
2753                    cc->core_id);
2754         return;
2755     }
2756     if (index == 0) {
2757         error_setg(errp, "Boot CPU core may not be unplugged");
2758         return;
2759     }
2760 
2761     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2762     g_assert(drc);
2763 
2764     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2765     drck->detach(drc, dev, spapr_core_release, NULL, &local_err);
2766     if (local_err) {
2767         error_propagate(errp, local_err);
2768         return;
2769     }
2770 
2771     spapr_hotplug_req_remove_by_index(drc);
2772 }
2773 
2774 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2775                             Error **errp)
2776 {
2777     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2778     MachineClass *mc = MACHINE_GET_CLASS(spapr);
2779     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2780     CPUCore *cc = CPU_CORE(dev);
2781     CPUState *cs = CPU(core->threads);
2782     sPAPRDRConnector *drc;
2783     Error *local_err = NULL;
2784     void *fdt = NULL;
2785     int fdt_offset = 0;
2786     int smt = kvmppc_smt_threads();
2787     CPUArchId *core_slot;
2788     int index;
2789 
2790     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2791     if (!core_slot) {
2792         error_setg(errp, "Unable to find CPU core with core-id: %d",
2793                    cc->core_id);
2794         return;
2795     }
2796     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2797 
2798     g_assert(drc || !mc->has_hotpluggable_cpus);
2799 
2800     /*
2801      * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2802      * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2803      */
2804     if (dev->hotplugged) {
2805         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2806     }
2807 
2808     if (drc) {
2809         sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2810         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
2811         if (local_err) {
2812             g_free(fdt);
2813             error_propagate(errp, local_err);
2814             return;
2815         }
2816     }
2817 
2818     if (dev->hotplugged) {
2819         /*
2820          * Send hotplug notification interrupt to the guest only in case
2821          * of hotplugged CPUs.
2822          */
2823         spapr_hotplug_req_add_by_index(drc);
2824     } else {
2825         /*
2826          * Set the right DRC states for cold plugged CPU.
2827          */
2828         if (drc) {
2829             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2830             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2831             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2832         }
2833     }
2834     core_slot->cpu = OBJECT(dev);
2835 }
2836 
2837 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2838                                 Error **errp)
2839 {
2840     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2841     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2842     Error *local_err = NULL;
2843     CPUCore *cc = CPU_CORE(dev);
2844     sPAPRCPUCore *sc = SPAPR_CPU_CORE(dev);
2845     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2846     const char *type = object_get_typename(OBJECT(dev));
2847     CPUArchId *core_slot;
2848     int node_id;
2849     int index;
2850 
2851     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2852         error_setg(&local_err, "CPU hotplug not supported for this machine");
2853         goto out;
2854     }
2855 
2856     if (strcmp(base_core_type, type)) {
2857         error_setg(&local_err, "CPU core type should be %s", base_core_type);
2858         goto out;
2859     }
2860 
2861     if (cc->core_id % smp_threads) {
2862         error_setg(&local_err, "invalid core id %d", cc->core_id);
2863         goto out;
2864     }
2865 
2866     if (cc->nr_threads != smp_threads) {
2867         error_setg(errp, "invalid nr-threads %d, must be %d",
2868                    cc->nr_threads, smp_threads);
2869         return;
2870     }
2871 
2872     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2873     if (!core_slot) {
2874         error_setg(&local_err, "core id %d out of range", cc->core_id);
2875         goto out;
2876     }
2877 
2878     if (core_slot->cpu) {
2879         error_setg(&local_err, "core %d already populated", cc->core_id);
2880         goto out;
2881     }
2882 
2883     node_id = core_slot->props.node_id;
2884     if (!core_slot->props.has_node_id) {
2885         /* by default CPUState::numa_node was 0 if it's not set via CLI
2886          * keep it this way for now but in future we probably should
2887          * refuse to start up with incomplete numa mapping */
2888         node_id = 0;
2889     }
2890     if (sc->node_id == CPU_UNSET_NUMA_NODE_ID) {
2891         sc->node_id = node_id;
2892     } else if (sc->node_id != node_id) {
2893         error_setg(&local_err, "node-id %d must match numa node specified"
2894             "with -numa option for cpu-index %d", sc->node_id, cc->core_id);
2895         goto out;
2896     }
2897 
2898 out:
2899     g_free(base_core_type);
2900     error_propagate(errp, local_err);
2901 }
2902 
2903 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2904                                       DeviceState *dev, Error **errp)
2905 {
2906     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2907 
2908     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2909         int node;
2910 
2911         if (!smc->dr_lmb_enabled) {
2912             error_setg(errp, "Memory hotplug not supported for this machine");
2913             return;
2914         }
2915         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2916         if (*errp) {
2917             return;
2918         }
2919         if (node < 0 || node >= MAX_NODES) {
2920             error_setg(errp, "Invaild node %d", node);
2921             return;
2922         }
2923 
2924         /*
2925          * Currently PowerPC kernel doesn't allow hot-adding memory to
2926          * memory-less node, but instead will silently add the memory
2927          * to the first node that has some memory. This causes two
2928          * unexpected behaviours for the user.
2929          *
2930          * - Memory gets hotplugged to a different node than what the user
2931          *   specified.
2932          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2933          *   to memory-less node, a reboot will set things accordingly
2934          *   and the previously hotplugged memory now ends in the right node.
2935          *   This appears as if some memory moved from one node to another.
2936          *
2937          * So until kernel starts supporting memory hotplug to memory-less
2938          * nodes, just prevent such attempts upfront in QEMU.
2939          */
2940         if (nb_numa_nodes && !numa_info[node].node_mem) {
2941             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2942                        node);
2943             return;
2944         }
2945 
2946         spapr_memory_plug(hotplug_dev, dev, node, errp);
2947     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2948         spapr_core_plug(hotplug_dev, dev, errp);
2949     }
2950 }
2951 
2952 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2953                                       DeviceState *dev, Error **errp)
2954 {
2955     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2956     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2957 
2958     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2959         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2960             spapr_memory_unplug(hotplug_dev, dev, errp);
2961         } else {
2962             error_setg(errp, "Memory hot unplug not supported for this guest");
2963         }
2964     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2965         if (!mc->has_hotpluggable_cpus) {
2966             error_setg(errp, "CPU hot unplug not supported on this machine");
2967             return;
2968         }
2969         spapr_core_unplug(hotplug_dev, dev, errp);
2970     }
2971 }
2972 
2973 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2974                                                 DeviceState *dev, Error **errp)
2975 {
2976     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2977     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2978 
2979     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2980         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2981             spapr_memory_unplug_request(hotplug_dev, dev, errp);
2982         } else {
2983             /* NOTE: this means there is a window after guest reset, prior to
2984              * CAS negotiation, where unplug requests will fail due to the
2985              * capability not being detected yet. This is a bit different than
2986              * the case with PCI unplug, where the events will be queued and
2987              * eventually handled by the guest after boot
2988              */
2989             error_setg(errp, "Memory hot unplug not supported for this guest");
2990         }
2991     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2992         if (!mc->has_hotpluggable_cpus) {
2993             error_setg(errp, "CPU hot unplug not supported on this machine");
2994             return;
2995         }
2996         spapr_core_unplug_request(hotplug_dev, dev, errp);
2997     }
2998 }
2999 
3000 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3001                                           DeviceState *dev, Error **errp)
3002 {
3003     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3004         spapr_core_pre_plug(hotplug_dev, dev, errp);
3005     }
3006 }
3007 
3008 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3009                                                  DeviceState *dev)
3010 {
3011     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3012         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3013         return HOTPLUG_HANDLER(machine);
3014     }
3015     return NULL;
3016 }
3017 
3018 static CpuInstanceProperties
3019 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3020 {
3021     CPUArchId *core_slot;
3022     MachineClass *mc = MACHINE_GET_CLASS(machine);
3023 
3024     /* make sure possible_cpu are intialized */
3025     mc->possible_cpu_arch_ids(machine);
3026     /* get CPU core slot containing thread that matches cpu_index */
3027     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3028     assert(core_slot);
3029     return core_slot->props;
3030 }
3031 
3032 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3033 {
3034     int i;
3035     int spapr_max_cores = max_cpus / smp_threads;
3036     MachineClass *mc = MACHINE_GET_CLASS(machine);
3037 
3038     if (!mc->has_hotpluggable_cpus) {
3039         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3040     }
3041     if (machine->possible_cpus) {
3042         assert(machine->possible_cpus->len == spapr_max_cores);
3043         return machine->possible_cpus;
3044     }
3045 
3046     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3047                              sizeof(CPUArchId) * spapr_max_cores);
3048     machine->possible_cpus->len = spapr_max_cores;
3049     for (i = 0; i < machine->possible_cpus->len; i++) {
3050         int core_id = i * smp_threads;
3051 
3052         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3053         machine->possible_cpus->cpus[i].arch_id = core_id;
3054         machine->possible_cpus->cpus[i].props.has_core_id = true;
3055         machine->possible_cpus->cpus[i].props.core_id = core_id;
3056 
3057         /* default distribution of CPUs over NUMA nodes */
3058         if (nb_numa_nodes) {
3059             /* preset values but do not enable them i.e. 'has_node_id = false',
3060              * numa init code will enable them later if manual mapping wasn't
3061              * present on CLI */
3062             machine->possible_cpus->cpus[i].props.node_id =
3063                 core_id / smp_threads / smp_cores % nb_numa_nodes;
3064         }
3065     }
3066     return machine->possible_cpus;
3067 }
3068 
3069 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3070                                 uint64_t *buid, hwaddr *pio,
3071                                 hwaddr *mmio32, hwaddr *mmio64,
3072                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3073 {
3074     /*
3075      * New-style PHB window placement.
3076      *
3077      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3078      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3079      * windows.
3080      *
3081      * Some guest kernels can't work with MMIO windows above 1<<46
3082      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3083      *
3084      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3085      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3086      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3087      * 1TiB 64-bit MMIO windows for each PHB.
3088      */
3089     const uint64_t base_buid = 0x800000020000000ULL;
3090 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3091                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3092     int i;
3093 
3094     /* Sanity check natural alignments */
3095     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3096     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3097     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3098     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3099     /* Sanity check bounds */
3100     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3101                       SPAPR_PCI_MEM32_WIN_SIZE);
3102     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3103                       SPAPR_PCI_MEM64_WIN_SIZE);
3104 
3105     if (index >= SPAPR_MAX_PHBS) {
3106         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3107                    SPAPR_MAX_PHBS - 1);
3108         return;
3109     }
3110 
3111     *buid = base_buid + index;
3112     for (i = 0; i < n_dma; ++i) {
3113         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3114     }
3115 
3116     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3117     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3118     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3119 }
3120 
3121 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3122 {
3123     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3124 
3125     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3126 }
3127 
3128 static void spapr_ics_resend(XICSFabric *dev)
3129 {
3130     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3131 
3132     ics_resend(spapr->ics);
3133 }
3134 
3135 static ICPState *spapr_icp_get(XICSFabric *xi, int cpu_dt_id)
3136 {
3137     PowerPCCPU *cpu = ppc_get_vcpu_by_dt_id(cpu_dt_id);
3138 
3139     return cpu ? ICP(cpu->intc) : NULL;
3140 }
3141 
3142 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3143                                  Monitor *mon)
3144 {
3145     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3146     CPUState *cs;
3147 
3148     CPU_FOREACH(cs) {
3149         PowerPCCPU *cpu = POWERPC_CPU(cs);
3150 
3151         icp_pic_print_info(ICP(cpu->intc), mon);
3152     }
3153 
3154     ics_pic_print_info(spapr->ics, mon);
3155 }
3156 
3157 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3158 {
3159     MachineClass *mc = MACHINE_CLASS(oc);
3160     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3161     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3162     NMIClass *nc = NMI_CLASS(oc);
3163     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3164     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3165     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3166     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3167 
3168     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3169 
3170     /*
3171      * We set up the default / latest behaviour here.  The class_init
3172      * functions for the specific versioned machine types can override
3173      * these details for backwards compatibility
3174      */
3175     mc->init = ppc_spapr_init;
3176     mc->reset = ppc_spapr_reset;
3177     mc->block_default_type = IF_SCSI;
3178     mc->max_cpus = 1024;
3179     mc->no_parallel = 1;
3180     mc->default_boot_order = "";
3181     mc->default_ram_size = 512 * M_BYTE;
3182     mc->kvm_type = spapr_kvm_type;
3183     mc->has_dynamic_sysbus = true;
3184     mc->pci_allow_0_address = true;
3185     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3186     hc->pre_plug = spapr_machine_device_pre_plug;
3187     hc->plug = spapr_machine_device_plug;
3188     hc->unplug = spapr_machine_device_unplug;
3189     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3190     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3191     hc->unplug_request = spapr_machine_device_unplug_request;
3192 
3193     smc->dr_lmb_enabled = true;
3194     smc->tcg_default_cpu = "POWER8";
3195     mc->has_hotpluggable_cpus = true;
3196     fwc->get_dev_path = spapr_get_fw_dev_path;
3197     nc->nmi_monitor_handler = spapr_nmi;
3198     smc->phb_placement = spapr_phb_placement;
3199     vhc->hypercall = emulate_spapr_hypercall;
3200     vhc->hpt_mask = spapr_hpt_mask;
3201     vhc->map_hptes = spapr_map_hptes;
3202     vhc->unmap_hptes = spapr_unmap_hptes;
3203     vhc->store_hpte = spapr_store_hpte;
3204     vhc->get_patbe = spapr_get_patbe;
3205     xic->ics_get = spapr_ics_get;
3206     xic->ics_resend = spapr_ics_resend;
3207     xic->icp_get = spapr_icp_get;
3208     ispc->print_info = spapr_pic_print_info;
3209     /* Force NUMA node memory size to be a multiple of
3210      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3211      * in which LMBs are represented and hot-added
3212      */
3213     mc->numa_mem_align_shift = 28;
3214 }
3215 
3216 static const TypeInfo spapr_machine_info = {
3217     .name          = TYPE_SPAPR_MACHINE,
3218     .parent        = TYPE_MACHINE,
3219     .abstract      = true,
3220     .instance_size = sizeof(sPAPRMachineState),
3221     .instance_init = spapr_machine_initfn,
3222     .instance_finalize = spapr_machine_finalizefn,
3223     .class_size    = sizeof(sPAPRMachineClass),
3224     .class_init    = spapr_machine_class_init,
3225     .interfaces = (InterfaceInfo[]) {
3226         { TYPE_FW_PATH_PROVIDER },
3227         { TYPE_NMI },
3228         { TYPE_HOTPLUG_HANDLER },
3229         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3230         { TYPE_XICS_FABRIC },
3231         { TYPE_INTERRUPT_STATS_PROVIDER },
3232         { }
3233     },
3234 };
3235 
3236 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3237     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3238                                                     void *data)      \
3239     {                                                                \
3240         MachineClass *mc = MACHINE_CLASS(oc);                        \
3241         spapr_machine_##suffix##_class_options(mc);                  \
3242         if (latest) {                                                \
3243             mc->alias = "pseries";                                   \
3244             mc->is_default = 1;                                      \
3245         }                                                            \
3246     }                                                                \
3247     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3248     {                                                                \
3249         MachineState *machine = MACHINE(obj);                        \
3250         spapr_machine_##suffix##_instance_options(machine);          \
3251     }                                                                \
3252     static const TypeInfo spapr_machine_##suffix##_info = {          \
3253         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3254         .parent = TYPE_SPAPR_MACHINE,                                \
3255         .class_init = spapr_machine_##suffix##_class_init,           \
3256         .instance_init = spapr_machine_##suffix##_instance_init,     \
3257     };                                                               \
3258     static void spapr_machine_register_##suffix(void)                \
3259     {                                                                \
3260         type_register(&spapr_machine_##suffix##_info);               \
3261     }                                                                \
3262     type_init(spapr_machine_register_##suffix)
3263 
3264 /*
3265  * pseries-2.10
3266  */
3267 static void spapr_machine_2_10_instance_options(MachineState *machine)
3268 {
3269 }
3270 
3271 static void spapr_machine_2_10_class_options(MachineClass *mc)
3272 {
3273     /* Defaults for the latest behaviour inherited from the base class */
3274 }
3275 
3276 DEFINE_SPAPR_MACHINE(2_10, "2.10", true);
3277 
3278 /*
3279  * pseries-2.9
3280  */
3281 #define SPAPR_COMPAT_2_9                                               \
3282     HW_COMPAT_2_9
3283 
3284 static void spapr_machine_2_9_instance_options(MachineState *machine)
3285 {
3286     spapr_machine_2_10_instance_options(machine);
3287 }
3288 
3289 static void spapr_machine_2_9_class_options(MachineClass *mc)
3290 {
3291     spapr_machine_2_10_class_options(mc);
3292     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3293     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
3294 }
3295 
3296 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3297 
3298 /*
3299  * pseries-2.8
3300  */
3301 #define SPAPR_COMPAT_2_8                                        \
3302     HW_COMPAT_2_8                                               \
3303     {                                                           \
3304         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3305         .property = "pcie-extended-configuration-space",        \
3306         .value    = "off",                                      \
3307     },
3308 
3309 static void spapr_machine_2_8_instance_options(MachineState *machine)
3310 {
3311     spapr_machine_2_9_instance_options(machine);
3312 }
3313 
3314 static void spapr_machine_2_8_class_options(MachineClass *mc)
3315 {
3316     spapr_machine_2_9_class_options(mc);
3317     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3318     mc->numa_mem_align_shift = 23;
3319 }
3320 
3321 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3322 
3323 /*
3324  * pseries-2.7
3325  */
3326 #define SPAPR_COMPAT_2_7                            \
3327     HW_COMPAT_2_7                                   \
3328     {                                               \
3329         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3330         .property = "mem_win_size",                 \
3331         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3332     },                                              \
3333     {                                               \
3334         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3335         .property = "mem64_win_size",               \
3336         .value    = "0",                            \
3337     },                                              \
3338     {                                               \
3339         .driver = TYPE_POWERPC_CPU,                 \
3340         .property = "pre-2.8-migration",            \
3341         .value    = "on",                           \
3342     },                                              \
3343     {                                               \
3344         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3345         .property = "pre-2.8-migration",            \
3346         .value    = "on",                           \
3347     },
3348 
3349 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3350                               uint64_t *buid, hwaddr *pio,
3351                               hwaddr *mmio32, hwaddr *mmio64,
3352                               unsigned n_dma, uint32_t *liobns, Error **errp)
3353 {
3354     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3355     const uint64_t base_buid = 0x800000020000000ULL;
3356     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3357     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3358     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3359     const uint32_t max_index = 255;
3360     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3361 
3362     uint64_t ram_top = MACHINE(spapr)->ram_size;
3363     hwaddr phb0_base, phb_base;
3364     int i;
3365 
3366     /* Do we have hotpluggable memory? */
3367     if (MACHINE(spapr)->maxram_size > ram_top) {
3368         /* Can't just use maxram_size, because there may be an
3369          * alignment gap between normal and hotpluggable memory
3370          * regions */
3371         ram_top = spapr->hotplug_memory.base +
3372             memory_region_size(&spapr->hotplug_memory.mr);
3373     }
3374 
3375     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3376 
3377     if (index > max_index) {
3378         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3379                    max_index);
3380         return;
3381     }
3382 
3383     *buid = base_buid + index;
3384     for (i = 0; i < n_dma; ++i) {
3385         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3386     }
3387 
3388     phb_base = phb0_base + index * phb_spacing;
3389     *pio = phb_base + pio_offset;
3390     *mmio32 = phb_base + mmio_offset;
3391     /*
3392      * We don't set the 64-bit MMIO window, relying on the PHB's
3393      * fallback behaviour of automatically splitting a large "32-bit"
3394      * window into contiguous 32-bit and 64-bit windows
3395      */
3396 }
3397 
3398 static void spapr_machine_2_7_instance_options(MachineState *machine)
3399 {
3400     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3401 
3402     spapr_machine_2_8_instance_options(machine);
3403     spapr->use_hotplug_event_source = false;
3404 }
3405 
3406 static void spapr_machine_2_7_class_options(MachineClass *mc)
3407 {
3408     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3409 
3410     spapr_machine_2_8_class_options(mc);
3411     smc->tcg_default_cpu = "POWER7";
3412     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3413     smc->phb_placement = phb_placement_2_7;
3414 }
3415 
3416 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3417 
3418 /*
3419  * pseries-2.6
3420  */
3421 #define SPAPR_COMPAT_2_6 \
3422     HW_COMPAT_2_6 \
3423     { \
3424         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3425         .property = "ddw",\
3426         .value    = stringify(off),\
3427     },
3428 
3429 static void spapr_machine_2_6_instance_options(MachineState *machine)
3430 {
3431     spapr_machine_2_7_instance_options(machine);
3432 }
3433 
3434 static void spapr_machine_2_6_class_options(MachineClass *mc)
3435 {
3436     spapr_machine_2_7_class_options(mc);
3437     mc->has_hotpluggable_cpus = false;
3438     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3439 }
3440 
3441 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3442 
3443 /*
3444  * pseries-2.5
3445  */
3446 #define SPAPR_COMPAT_2_5 \
3447     HW_COMPAT_2_5 \
3448     { \
3449         .driver   = "spapr-vlan", \
3450         .property = "use-rx-buffer-pools", \
3451         .value    = "off", \
3452     },
3453 
3454 static void spapr_machine_2_5_instance_options(MachineState *machine)
3455 {
3456     spapr_machine_2_6_instance_options(machine);
3457 }
3458 
3459 static void spapr_machine_2_5_class_options(MachineClass *mc)
3460 {
3461     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3462 
3463     spapr_machine_2_6_class_options(mc);
3464     smc->use_ohci_by_default = true;
3465     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3466 }
3467 
3468 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3469 
3470 /*
3471  * pseries-2.4
3472  */
3473 #define SPAPR_COMPAT_2_4 \
3474         HW_COMPAT_2_4
3475 
3476 static void spapr_machine_2_4_instance_options(MachineState *machine)
3477 {
3478     spapr_machine_2_5_instance_options(machine);
3479 }
3480 
3481 static void spapr_machine_2_4_class_options(MachineClass *mc)
3482 {
3483     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3484 
3485     spapr_machine_2_5_class_options(mc);
3486     smc->dr_lmb_enabled = false;
3487     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3488 }
3489 
3490 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3491 
3492 /*
3493  * pseries-2.3
3494  */
3495 #define SPAPR_COMPAT_2_3 \
3496         HW_COMPAT_2_3 \
3497         {\
3498             .driver   = "spapr-pci-host-bridge",\
3499             .property = "dynamic-reconfiguration",\
3500             .value    = "off",\
3501         },
3502 
3503 static void spapr_machine_2_3_instance_options(MachineState *machine)
3504 {
3505     spapr_machine_2_4_instance_options(machine);
3506     savevm_skip_section_footers();
3507     global_state_set_optional();
3508     savevm_skip_configuration();
3509 }
3510 
3511 static void spapr_machine_2_3_class_options(MachineClass *mc)
3512 {
3513     spapr_machine_2_4_class_options(mc);
3514     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3515 }
3516 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3517 
3518 /*
3519  * pseries-2.2
3520  */
3521 
3522 #define SPAPR_COMPAT_2_2 \
3523         HW_COMPAT_2_2 \
3524         {\
3525             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3526             .property = "mem_win_size",\
3527             .value    = "0x20000000",\
3528         },
3529 
3530 static void spapr_machine_2_2_instance_options(MachineState *machine)
3531 {
3532     spapr_machine_2_3_instance_options(machine);
3533     machine->suppress_vmdesc = true;
3534 }
3535 
3536 static void spapr_machine_2_2_class_options(MachineClass *mc)
3537 {
3538     spapr_machine_2_3_class_options(mc);
3539     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3540 }
3541 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3542 
3543 /*
3544  * pseries-2.1
3545  */
3546 #define SPAPR_COMPAT_2_1 \
3547         HW_COMPAT_2_1
3548 
3549 static void spapr_machine_2_1_instance_options(MachineState *machine)
3550 {
3551     spapr_machine_2_2_instance_options(machine);
3552 }
3553 
3554 static void spapr_machine_2_1_class_options(MachineClass *mc)
3555 {
3556     spapr_machine_2_2_class_options(mc);
3557     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3558 }
3559 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3560 
3561 static void spapr_machine_register_types(void)
3562 {
3563     type_register_static(&spapr_machine_info);
3564 }
3565 
3566 type_init(spapr_machine_register_types)
3567