xref: /openbmc/qemu/hw/ppc/spapr.c (revision ad265631c0a0addc06ec3c4f133e746f4dcc872a)
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 #include "hw/virtio/vhost-scsi-common.h"
61 
62 #include "exec/address-spaces.h"
63 #include "hw/usb.h"
64 #include "qemu/config-file.h"
65 #include "qemu/error-report.h"
66 #include "trace.h"
67 #include "hw/nmi.h"
68 #include "hw/intc/intc.h"
69 
70 #include "hw/compat.h"
71 #include "qemu/cutils.h"
72 #include "hw/ppc/spapr_cpu_core.h"
73 #include "qmp-commands.h"
74 
75 #include <libfdt.h>
76 
77 /* SLOF memory layout:
78  *
79  * SLOF raw image loaded at 0, copies its romfs right below the flat
80  * device-tree, then position SLOF itself 31M below that
81  *
82  * So we set FW_OVERHEAD to 40MB which should account for all of that
83  * and more
84  *
85  * We load our kernel at 4M, leaving space for SLOF initial image
86  */
87 #define FDT_MAX_SIZE            0x100000
88 #define RTAS_MAX_SIZE           0x10000
89 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
90 #define FW_MAX_SIZE             0x400000
91 #define FW_FILE_NAME            "slof.bin"
92 #define FW_OVERHEAD             0x2800000
93 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
94 
95 #define MIN_RMA_SLOF            128UL
96 
97 #define PHANDLE_XICP            0x00001111
98 
99 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
100 
101 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
102                                   const char *type_ics,
103                                   int nr_irqs, Error **errp)
104 {
105     Error *local_err = NULL;
106     Object *obj;
107 
108     obj = object_new(type_ics);
109     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
110     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
111                                    &error_abort);
112     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
113     if (local_err) {
114         goto error;
115     }
116     object_property_set_bool(obj, true, "realized", &local_err);
117     if (local_err) {
118         goto error;
119     }
120 
121     return ICS_SIMPLE(obj);
122 
123 error:
124     error_propagate(errp, local_err);
125     return NULL;
126 }
127 
128 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
129 {
130     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
131 
132     if (kvm_enabled()) {
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, errp);
137         }
138         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
139             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
140             return;
141         }
142     }
143 
144     if (!spapr->ics) {
145         xics_spapr_init(spapr);
146         spapr->icp_type = TYPE_ICP;
147         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
148         if (!spapr->ics) {
149             return;
150         }
151     }
152 }
153 
154 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
155                                   int smt_threads)
156 {
157     int i, ret = 0;
158     uint32_t servers_prop[smt_threads];
159     uint32_t gservers_prop[smt_threads * 2];
160     int index = ppc_get_vcpu_dt_id(cpu);
161 
162     if (cpu->compat_pvr) {
163         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
164         if (ret < 0) {
165             return ret;
166         }
167     }
168 
169     /* Build interrupt servers and gservers properties */
170     for (i = 0; i < smt_threads; i++) {
171         servers_prop[i] = cpu_to_be32(index + i);
172         /* Hack, direct the group queues back to cpu 0 */
173         gservers_prop[i*2] = cpu_to_be32(index + i);
174         gservers_prop[i*2 + 1] = 0;
175     }
176     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
177                       servers_prop, sizeof(servers_prop));
178     if (ret < 0) {
179         return ret;
180     }
181     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
182                       gservers_prop, sizeof(gservers_prop));
183 
184     return ret;
185 }
186 
187 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
188 {
189     int index = ppc_get_vcpu_dt_id(cpu);
190     uint32_t associativity[] = {cpu_to_be32(0x5),
191                                 cpu_to_be32(0x0),
192                                 cpu_to_be32(0x0),
193                                 cpu_to_be32(0x0),
194                                 cpu_to_be32(cpu->node_id),
195                                 cpu_to_be32(index)};
196 
197     /* Advertise NUMA via ibm,associativity */
198     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
199                           sizeof(associativity));
200 }
201 
202 /* Populate the "ibm,pa-features" property */
203 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset,
204                                       bool legacy_guest)
205 {
206     uint8_t pa_features_206[] = { 6, 0,
207         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
208     uint8_t pa_features_207[] = { 24, 0,
209         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
210         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
211         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
212         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
213     uint8_t pa_features_300[] = { 66, 0,
214         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
215         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
216         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
217         /* 6: DS207 */
218         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
219         /* 16: Vector */
220         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
221         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
222         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
223         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
224         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
225         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
226         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
227         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
228         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
229         /* 42: PM, 44: PC RA, 46: SC vec'd */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
231         /* 48: SIMD, 50: QP BFP, 52: String */
232         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
233         /* 54: DecFP, 56: DecI, 58: SHA */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
235         /* 60: NM atomic, 62: RNG */
236         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
237     };
238     uint8_t *pa_features;
239     size_t pa_size;
240 
241     switch (POWERPC_MMU_VER(env->mmu_model)) {
242     case POWERPC_MMU_VER_2_06:
243         pa_features = pa_features_206;
244         pa_size = sizeof(pa_features_206);
245         break;
246     case POWERPC_MMU_VER_2_07:
247         pa_features = pa_features_207;
248         pa_size = sizeof(pa_features_207);
249         break;
250     case POWERPC_MMU_VER_3_00:
251         pa_features = pa_features_300;
252         pa_size = sizeof(pa_features_300);
253         break;
254     default:
255         return;
256     }
257 
258     if (env->ci_large_pages) {
259         /*
260          * Note: we keep CI large pages off by default because a 64K capable
261          * guest provisioned with large pages might otherwise try to map a qemu
262          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
263          * even if that qemu runs on a 4k host.
264          * We dd this bit back here if we are confident this is not an issue
265          */
266         pa_features[3] |= 0x20;
267     }
268     if (kvmppc_has_cap_htm() && pa_size > 24) {
269         pa_features[24] |= 0x80;    /* Transactional memory support */
270     }
271     if (legacy_guest && pa_size > 40) {
272         /* Workaround for broken kernels that attempt (guest) radix
273          * mode when they can't handle it, if they see the radix bit set
274          * in pa-features. So hide it from them. */
275         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
276     }
277 
278     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
279 }
280 
281 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
282 {
283     int ret = 0, offset, cpus_offset;
284     CPUState *cs;
285     char cpu_model[32];
286     int smt = kvmppc_smt_threads();
287     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
288 
289     CPU_FOREACH(cs) {
290         PowerPCCPU *cpu = POWERPC_CPU(cs);
291         CPUPPCState *env = &cpu->env;
292         DeviceClass *dc = DEVICE_GET_CLASS(cs);
293         int index = ppc_get_vcpu_dt_id(cpu);
294         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
295 
296         if ((index % smt) != 0) {
297             continue;
298         }
299 
300         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
301 
302         cpus_offset = fdt_path_offset(fdt, "/cpus");
303         if (cpus_offset < 0) {
304             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
305                                           "cpus");
306             if (cpus_offset < 0) {
307                 return cpus_offset;
308             }
309         }
310         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
311         if (offset < 0) {
312             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
313             if (offset < 0) {
314                 return offset;
315             }
316         }
317 
318         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
319                           pft_size_prop, sizeof(pft_size_prop));
320         if (ret < 0) {
321             return ret;
322         }
323 
324         if (nb_numa_nodes > 1) {
325             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
326             if (ret < 0) {
327                 return ret;
328             }
329         }
330 
331         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
332         if (ret < 0) {
333             return ret;
334         }
335 
336         spapr_populate_pa_features(env, fdt, offset,
337                                          spapr->cas_legacy_guest_workaround);
338     }
339     return ret;
340 }
341 
342 static hwaddr spapr_node0_size(void)
343 {
344     MachineState *machine = MACHINE(qdev_get_machine());
345 
346     if (nb_numa_nodes) {
347         int i;
348         for (i = 0; i < nb_numa_nodes; ++i) {
349             if (numa_info[i].node_mem) {
350                 return MIN(pow2floor(numa_info[i].node_mem),
351                            machine->ram_size);
352             }
353         }
354     }
355     return machine->ram_size;
356 }
357 
358 static void add_str(GString *s, const gchar *s1)
359 {
360     g_string_append_len(s, s1, strlen(s1) + 1);
361 }
362 
363 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
364                                        hwaddr size)
365 {
366     uint32_t associativity[] = {
367         cpu_to_be32(0x4), /* length */
368         cpu_to_be32(0x0), cpu_to_be32(0x0),
369         cpu_to_be32(0x0), cpu_to_be32(nodeid)
370     };
371     char mem_name[32];
372     uint64_t mem_reg_property[2];
373     int off;
374 
375     mem_reg_property[0] = cpu_to_be64(start);
376     mem_reg_property[1] = cpu_to_be64(size);
377 
378     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
379     off = fdt_add_subnode(fdt, 0, mem_name);
380     _FDT(off);
381     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
382     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
383                       sizeof(mem_reg_property))));
384     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
385                       sizeof(associativity))));
386     return off;
387 }
388 
389 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
390 {
391     MachineState *machine = MACHINE(spapr);
392     hwaddr mem_start, node_size;
393     int i, nb_nodes = nb_numa_nodes;
394     NodeInfo *nodes = numa_info;
395     NodeInfo ramnode;
396 
397     /* No NUMA nodes, assume there is just one node with whole RAM */
398     if (!nb_numa_nodes) {
399         nb_nodes = 1;
400         ramnode.node_mem = machine->ram_size;
401         nodes = &ramnode;
402     }
403 
404     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
405         if (!nodes[i].node_mem) {
406             continue;
407         }
408         if (mem_start >= machine->ram_size) {
409             node_size = 0;
410         } else {
411             node_size = nodes[i].node_mem;
412             if (node_size > machine->ram_size - mem_start) {
413                 node_size = machine->ram_size - mem_start;
414             }
415         }
416         if (!mem_start) {
417             /* ppc_spapr_init() checks for rma_size <= node0_size already */
418             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
419             mem_start += spapr->rma_size;
420             node_size -= spapr->rma_size;
421         }
422         for ( ; node_size; ) {
423             hwaddr sizetmp = pow2floor(node_size);
424 
425             /* mem_start != 0 here */
426             if (ctzl(mem_start) < ctzl(sizetmp)) {
427                 sizetmp = 1ULL << ctzl(mem_start);
428             }
429 
430             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
431             node_size -= sizetmp;
432             mem_start += sizetmp;
433         }
434     }
435 
436     return 0;
437 }
438 
439 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
440                                   sPAPRMachineState *spapr)
441 {
442     PowerPCCPU *cpu = POWERPC_CPU(cs);
443     CPUPPCState *env = &cpu->env;
444     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
445     int index = ppc_get_vcpu_dt_id(cpu);
446     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
447                        0xffffffff, 0xffffffff};
448     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
449         : SPAPR_TIMEBASE_FREQ;
450     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
451     uint32_t page_sizes_prop[64];
452     size_t page_sizes_prop_size;
453     uint32_t vcpus_per_socket = smp_threads * smp_cores;
454     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
455     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
456     sPAPRDRConnector *drc;
457     int drc_index;
458     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
459     int i;
460 
461     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
462     if (drc) {
463         drc_index = spapr_drc_index(drc);
464         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
465     }
466 
467     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
468     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
469 
470     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
471     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
472                            env->dcache_line_size)));
473     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
474                            env->dcache_line_size)));
475     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
476                            env->icache_line_size)));
477     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
478                            env->icache_line_size)));
479 
480     if (pcc->l1_dcache_size) {
481         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
482                                pcc->l1_dcache_size)));
483     } else {
484         error_report("Warning: Unknown L1 dcache size for cpu");
485     }
486     if (pcc->l1_icache_size) {
487         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
488                                pcc->l1_icache_size)));
489     } else {
490         error_report("Warning: Unknown L1 icache size for cpu");
491     }
492 
493     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
494     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
495     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
496     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
497     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
498     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
499 
500     if (env->spr_cb[SPR_PURR].oea_read) {
501         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
502     }
503 
504     if (env->mmu_model & POWERPC_MMU_1TSEG) {
505         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
506                           segs, sizeof(segs))));
507     }
508 
509     /* Advertise VMX/VSX (vector extensions) if available
510      *   0 / no property == no vector extensions
511      *   1               == VMX / Altivec available
512      *   2               == VSX available */
513     if (env->insns_flags & PPC_ALTIVEC) {
514         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
515 
516         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
517     }
518 
519     /* Advertise DFP (Decimal Floating Point) if available
520      *   0 / no property == no DFP
521      *   1               == DFP available */
522     if (env->insns_flags2 & PPC2_DFP) {
523         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
524     }
525 
526     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
527                                                   sizeof(page_sizes_prop));
528     if (page_sizes_prop_size) {
529         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
530                           page_sizes_prop, page_sizes_prop_size)));
531     }
532 
533     spapr_populate_pa_features(env, fdt, offset, false);
534 
535     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
536                            cs->cpu_index / vcpus_per_socket)));
537 
538     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
539                       pft_size_prop, sizeof(pft_size_prop))));
540 
541     if (nb_numa_nodes > 1) {
542         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
543     }
544 
545     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
546 
547     if (pcc->radix_page_info) {
548         for (i = 0; i < pcc->radix_page_info->count; i++) {
549             radix_AP_encodings[i] =
550                 cpu_to_be32(pcc->radix_page_info->entries[i]);
551         }
552         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
553                           radix_AP_encodings,
554                           pcc->radix_page_info->count *
555                           sizeof(radix_AP_encodings[0]))));
556     }
557 }
558 
559 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
560 {
561     CPUState *cs;
562     int cpus_offset;
563     char *nodename;
564     int smt = kvmppc_smt_threads();
565 
566     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
567     _FDT(cpus_offset);
568     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
569     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
570 
571     /*
572      * We walk the CPUs in reverse order to ensure that CPU DT nodes
573      * created by fdt_add_subnode() end up in the right order in FDT
574      * for the guest kernel the enumerate the CPUs correctly.
575      */
576     CPU_FOREACH_REVERSE(cs) {
577         PowerPCCPU *cpu = POWERPC_CPU(cs);
578         int index = ppc_get_vcpu_dt_id(cpu);
579         DeviceClass *dc = DEVICE_GET_CLASS(cs);
580         int offset;
581 
582         if ((index % smt) != 0) {
583             continue;
584         }
585 
586         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
587         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
588         g_free(nodename);
589         _FDT(offset);
590         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
591     }
592 
593 }
594 
595 /*
596  * Adds ibm,dynamic-reconfiguration-memory node.
597  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
598  * of this device tree node.
599  */
600 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
601 {
602     MachineState *machine = MACHINE(spapr);
603     int ret, i, offset;
604     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
605     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
606     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
607     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
608                        memory_region_size(&spapr->hotplug_memory.mr)) /
609                        lmb_size;
610     uint32_t *int_buf, *cur_index, buf_len;
611     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
612 
613     /*
614      * Don't create the node if there is no hotpluggable memory
615      */
616     if (machine->ram_size == machine->maxram_size) {
617         return 0;
618     }
619 
620     /*
621      * Allocate enough buffer size to fit in ibm,dynamic-memory
622      * or ibm,associativity-lookup-arrays
623      */
624     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
625               * sizeof(uint32_t);
626     cur_index = int_buf = g_malloc0(buf_len);
627 
628     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
629 
630     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
631                     sizeof(prop_lmb_size));
632     if (ret < 0) {
633         goto out;
634     }
635 
636     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
637     if (ret < 0) {
638         goto out;
639     }
640 
641     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
642     if (ret < 0) {
643         goto out;
644     }
645 
646     /* ibm,dynamic-memory */
647     int_buf[0] = cpu_to_be32(nr_lmbs);
648     cur_index++;
649     for (i = 0; i < nr_lmbs; i++) {
650         uint64_t addr = i * lmb_size;
651         uint32_t *dynamic_memory = cur_index;
652 
653         if (i >= hotplug_lmb_start) {
654             sPAPRDRConnector *drc;
655 
656             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
657             g_assert(drc);
658 
659             dynamic_memory[0] = cpu_to_be32(addr >> 32);
660             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
661             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
662             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
663             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
664             if (memory_region_present(get_system_memory(), addr)) {
665                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
666             } else {
667                 dynamic_memory[5] = cpu_to_be32(0);
668             }
669         } else {
670             /*
671              * LMB information for RMA, boot time RAM and gap b/n RAM and
672              * hotplug memory region -- all these are marked as reserved
673              * and as having no valid DRC.
674              */
675             dynamic_memory[0] = cpu_to_be32(addr >> 32);
676             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
677             dynamic_memory[2] = cpu_to_be32(0);
678             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
679             dynamic_memory[4] = cpu_to_be32(-1);
680             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
681                                             SPAPR_LMB_FLAGS_DRC_INVALID);
682         }
683 
684         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
685     }
686     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
687     if (ret < 0) {
688         goto out;
689     }
690 
691     /* ibm,associativity-lookup-arrays */
692     cur_index = int_buf;
693     int_buf[0] = cpu_to_be32(nr_nodes);
694     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
695     cur_index += 2;
696     for (i = 0; i < nr_nodes; i++) {
697         uint32_t associativity[] = {
698             cpu_to_be32(0x0),
699             cpu_to_be32(0x0),
700             cpu_to_be32(0x0),
701             cpu_to_be32(i)
702         };
703         memcpy(cur_index, associativity, sizeof(associativity));
704         cur_index += 4;
705     }
706     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
707             (cur_index - int_buf) * sizeof(uint32_t));
708 out:
709     g_free(int_buf);
710     return ret;
711 }
712 
713 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
714                                 sPAPROptionVector *ov5_updates)
715 {
716     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
717     int ret = 0, offset;
718 
719     /* Generate ibm,dynamic-reconfiguration-memory node if required */
720     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
721         g_assert(smc->dr_lmb_enabled);
722         ret = spapr_populate_drconf_memory(spapr, fdt);
723         if (ret) {
724             goto out;
725         }
726     }
727 
728     offset = fdt_path_offset(fdt, "/chosen");
729     if (offset < 0) {
730         offset = fdt_add_subnode(fdt, 0, "chosen");
731         if (offset < 0) {
732             return offset;
733         }
734     }
735     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
736                                  "ibm,architecture-vec-5");
737 
738 out:
739     return ret;
740 }
741 
742 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
743                                  target_ulong addr, target_ulong size,
744                                  sPAPROptionVector *ov5_updates)
745 {
746     void *fdt, *fdt_skel;
747     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
748 
749     size -= sizeof(hdr);
750 
751     /* Create sceleton */
752     fdt_skel = g_malloc0(size);
753     _FDT((fdt_create(fdt_skel, size)));
754     _FDT((fdt_begin_node(fdt_skel, "")));
755     _FDT((fdt_end_node(fdt_skel)));
756     _FDT((fdt_finish(fdt_skel)));
757     fdt = g_malloc0(size);
758     _FDT((fdt_open_into(fdt_skel, fdt, size)));
759     g_free(fdt_skel);
760 
761     /* Fixup cpu nodes */
762     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
763 
764     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
765         return -1;
766     }
767 
768     /* Pack resulting tree */
769     _FDT((fdt_pack(fdt)));
770 
771     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
772         trace_spapr_cas_failed(size);
773         return -1;
774     }
775 
776     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
777     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
778     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
779     g_free(fdt);
780 
781     return 0;
782 }
783 
784 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
785 {
786     int rtas;
787     GString *hypertas = g_string_sized_new(256);
788     GString *qemu_hypertas = g_string_sized_new(256);
789     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
790     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
791         memory_region_size(&spapr->hotplug_memory.mr);
792     uint32_t lrdr_capacity[] = {
793         cpu_to_be32(max_hotplug_addr >> 32),
794         cpu_to_be32(max_hotplug_addr & 0xffffffff),
795         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
796         cpu_to_be32(max_cpus / smp_threads),
797     };
798 
799     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
800 
801     /* hypertas */
802     add_str(hypertas, "hcall-pft");
803     add_str(hypertas, "hcall-term");
804     add_str(hypertas, "hcall-dabr");
805     add_str(hypertas, "hcall-interrupt");
806     add_str(hypertas, "hcall-tce");
807     add_str(hypertas, "hcall-vio");
808     add_str(hypertas, "hcall-splpar");
809     add_str(hypertas, "hcall-bulk");
810     add_str(hypertas, "hcall-set-mode");
811     add_str(hypertas, "hcall-sprg0");
812     add_str(hypertas, "hcall-copy");
813     add_str(hypertas, "hcall-debug");
814     add_str(qemu_hypertas, "hcall-memop1");
815 
816     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
817         add_str(hypertas, "hcall-multi-tce");
818     }
819     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
820                      hypertas->str, hypertas->len));
821     g_string_free(hypertas, TRUE);
822     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
823                      qemu_hypertas->str, qemu_hypertas->len));
824     g_string_free(qemu_hypertas, TRUE);
825 
826     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
827                      refpoints, sizeof(refpoints)));
828 
829     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
830                           RTAS_ERROR_LOG_MAX));
831     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
832                           RTAS_EVENT_SCAN_RATE));
833 
834     if (msi_nonbroken) {
835         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
836     }
837 
838     /*
839      * According to PAPR, rtas ibm,os-term does not guarantee a return
840      * back to the guest cpu.
841      *
842      * While an additional ibm,extended-os-term property indicates
843      * that rtas call return will always occur. Set this property.
844      */
845     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
846 
847     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
848                      lrdr_capacity, sizeof(lrdr_capacity)));
849 
850     spapr_dt_rtas_tokens(fdt, rtas);
851 }
852 
853 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
854  * that the guest may request and thus the valid values for bytes 24..26 of
855  * option vector 5: */
856 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
857 {
858     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
859 
860     char val[2 * 3] = {
861         24, 0x00, /* Hash/Radix, filled in below. */
862         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
863         26, 0x40, /* Radix options: GTSE == yes. */
864     };
865 
866     if (kvm_enabled()) {
867         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
868             val[1] = 0x80; /* OV5_MMU_BOTH */
869         } else if (kvmppc_has_cap_mmu_radix()) {
870             val[1] = 0x40; /* OV5_MMU_RADIX_300 */
871         } else {
872             val[1] = 0x00; /* Hash */
873         }
874     } else {
875         if (first_ppc_cpu->env.mmu_model & POWERPC_MMU_V3) {
876             /* V3 MMU supports both hash and radix (with dynamic switching) */
877             val[1] = 0xC0;
878         } else {
879             /* Otherwise we can only do hash */
880             val[1] = 0x00;
881         }
882     }
883     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
884                      val, sizeof(val)));
885 }
886 
887 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
888 {
889     MachineState *machine = MACHINE(spapr);
890     int chosen;
891     const char *boot_device = machine->boot_order;
892     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
893     size_t cb = 0;
894     char *bootlist = get_boot_devices_list(&cb, true);
895 
896     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
897 
898     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
899     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
900                           spapr->initrd_base));
901     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
902                           spapr->initrd_base + spapr->initrd_size));
903 
904     if (spapr->kernel_size) {
905         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
906                               cpu_to_be64(spapr->kernel_size) };
907 
908         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
909                          &kprop, sizeof(kprop)));
910         if (spapr->kernel_le) {
911             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
912         }
913     }
914     if (boot_menu) {
915         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
916     }
917     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
918     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
919     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
920 
921     if (cb && bootlist) {
922         int i;
923 
924         for (i = 0; i < cb; i++) {
925             if (bootlist[i] == '\n') {
926                 bootlist[i] = ' ';
927             }
928         }
929         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
930     }
931 
932     if (boot_device && strlen(boot_device)) {
933         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
934     }
935 
936     if (!spapr->has_graphics && stdout_path) {
937         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
938     }
939 
940     spapr_dt_ov5_platform_support(fdt, chosen);
941 
942     g_free(stdout_path);
943     g_free(bootlist);
944 }
945 
946 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
947 {
948     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
949      * KVM to work under pHyp with some guest co-operation */
950     int hypervisor;
951     uint8_t hypercall[16];
952 
953     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
954     /* indicate KVM hypercall interface */
955     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
956     if (kvmppc_has_cap_fixup_hcalls()) {
957         /*
958          * Older KVM versions with older guest kernels were broken
959          * with the magic page, don't allow the guest to map it.
960          */
961         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
962                                   sizeof(hypercall))) {
963             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
964                              hypercall, sizeof(hypercall)));
965         }
966     }
967 }
968 
969 static void *spapr_build_fdt(sPAPRMachineState *spapr,
970                              hwaddr rtas_addr,
971                              hwaddr rtas_size)
972 {
973     MachineState *machine = MACHINE(qdev_get_machine());
974     MachineClass *mc = MACHINE_GET_CLASS(machine);
975     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
976     int ret;
977     void *fdt;
978     sPAPRPHBState *phb;
979     char *buf;
980     int smt = kvmppc_smt_threads();
981 
982     fdt = g_malloc0(FDT_MAX_SIZE);
983     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
984 
985     /* Root node */
986     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
987     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
988     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
989 
990     /*
991      * Add info to guest to indentify which host is it being run on
992      * and what is the uuid of the guest
993      */
994     if (kvmppc_get_host_model(&buf)) {
995         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
996         g_free(buf);
997     }
998     if (kvmppc_get_host_serial(&buf)) {
999         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1000         g_free(buf);
1001     }
1002 
1003     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1004 
1005     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1006     if (qemu_uuid_set) {
1007         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1008     }
1009     g_free(buf);
1010 
1011     if (qemu_get_vm_name()) {
1012         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1013                                 qemu_get_vm_name()));
1014     }
1015 
1016     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1017     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1018 
1019     /* /interrupt controller */
1020     spapr_dt_xics(DIV_ROUND_UP(max_cpus * smt, smp_threads), fdt, PHANDLE_XICP);
1021 
1022     ret = spapr_populate_memory(spapr, fdt);
1023     if (ret < 0) {
1024         error_report("couldn't setup memory nodes in fdt");
1025         exit(1);
1026     }
1027 
1028     /* /vdevice */
1029     spapr_dt_vdevice(spapr->vio_bus, fdt);
1030 
1031     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1032         ret = spapr_rng_populate_dt(fdt);
1033         if (ret < 0) {
1034             error_report("could not set up rng device in the fdt");
1035             exit(1);
1036         }
1037     }
1038 
1039     QLIST_FOREACH(phb, &spapr->phbs, list) {
1040         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1041         if (ret < 0) {
1042             error_report("couldn't setup PCI devices in fdt");
1043             exit(1);
1044         }
1045     }
1046 
1047     /* cpus */
1048     spapr_populate_cpus_dt_node(fdt, spapr);
1049 
1050     if (smc->dr_lmb_enabled) {
1051         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1052     }
1053 
1054     if (mc->has_hotpluggable_cpus) {
1055         int offset = fdt_path_offset(fdt, "/cpus");
1056         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1057                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1058         if (ret < 0) {
1059             error_report("Couldn't set up CPU DR device tree properties");
1060             exit(1);
1061         }
1062     }
1063 
1064     /* /event-sources */
1065     spapr_dt_events(spapr, fdt);
1066 
1067     /* /rtas */
1068     spapr_dt_rtas(spapr, fdt);
1069 
1070     /* /chosen */
1071     spapr_dt_chosen(spapr, fdt);
1072 
1073     /* /hypervisor */
1074     if (kvm_enabled()) {
1075         spapr_dt_hypervisor(spapr, fdt);
1076     }
1077 
1078     /* Build memory reserve map */
1079     if (spapr->kernel_size) {
1080         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1081     }
1082     if (spapr->initrd_size) {
1083         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1084     }
1085 
1086     /* ibm,client-architecture-support updates */
1087     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1088     if (ret < 0) {
1089         error_report("couldn't setup CAS properties fdt");
1090         exit(1);
1091     }
1092 
1093     return fdt;
1094 }
1095 
1096 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1097 {
1098     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1099 }
1100 
1101 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1102                                     PowerPCCPU *cpu)
1103 {
1104     CPUPPCState *env = &cpu->env;
1105 
1106     /* The TCG path should also be holding the BQL at this point */
1107     g_assert(qemu_mutex_iothread_locked());
1108 
1109     if (msr_pr) {
1110         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1111         env->gpr[3] = H_PRIVILEGE;
1112     } else {
1113         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1114     }
1115 }
1116 
1117 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1118 {
1119     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1120 
1121     return spapr->patb_entry;
1122 }
1123 
1124 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1125 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1126 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1127 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1128 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1129 
1130 /*
1131  * Get the fd to access the kernel htab, re-opening it if necessary
1132  */
1133 static int get_htab_fd(sPAPRMachineState *spapr)
1134 {
1135     if (spapr->htab_fd >= 0) {
1136         return spapr->htab_fd;
1137     }
1138 
1139     spapr->htab_fd = kvmppc_get_htab_fd(false);
1140     if (spapr->htab_fd < 0) {
1141         error_report("Unable to open fd for reading hash table from KVM: %s",
1142                      strerror(errno));
1143     }
1144 
1145     return spapr->htab_fd;
1146 }
1147 
1148 void close_htab_fd(sPAPRMachineState *spapr)
1149 {
1150     if (spapr->htab_fd >= 0) {
1151         close(spapr->htab_fd);
1152     }
1153     spapr->htab_fd = -1;
1154 }
1155 
1156 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1157 {
1158     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1159 
1160     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1161 }
1162 
1163 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1164                                                 hwaddr ptex, int n)
1165 {
1166     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1167     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1168 
1169     if (!spapr->htab) {
1170         /*
1171          * HTAB is controlled by KVM. Fetch into temporary buffer
1172          */
1173         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1174         kvmppc_read_hptes(hptes, ptex, n);
1175         return hptes;
1176     }
1177 
1178     /*
1179      * HTAB is controlled by QEMU. Just point to the internally
1180      * accessible PTEG.
1181      */
1182     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1183 }
1184 
1185 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1186                               const ppc_hash_pte64_t *hptes,
1187                               hwaddr ptex, int n)
1188 {
1189     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1190 
1191     if (!spapr->htab) {
1192         g_free((void *)hptes);
1193     }
1194 
1195     /* Nothing to do for qemu managed HPT */
1196 }
1197 
1198 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1199                              uint64_t pte0, uint64_t pte1)
1200 {
1201     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1202     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1203 
1204     if (!spapr->htab) {
1205         kvmppc_write_hpte(ptex, pte0, pte1);
1206     } else {
1207         stq_p(spapr->htab + offset, pte0);
1208         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1209     }
1210 }
1211 
1212 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1213 {
1214     int shift;
1215 
1216     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1217      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1218      * that's much more than is needed for Linux guests */
1219     shift = ctz64(pow2ceil(ramsize)) - 7;
1220     shift = MAX(shift, 18); /* Minimum architected size */
1221     shift = MIN(shift, 46); /* Maximum architected size */
1222     return shift;
1223 }
1224 
1225 void spapr_free_hpt(sPAPRMachineState *spapr)
1226 {
1227     g_free(spapr->htab);
1228     spapr->htab = NULL;
1229     spapr->htab_shift = 0;
1230     close_htab_fd(spapr);
1231 }
1232 
1233 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1234                                  Error **errp)
1235 {
1236     long rc;
1237 
1238     /* Clean up any HPT info from a previous boot */
1239     spapr_free_hpt(spapr);
1240 
1241     rc = kvmppc_reset_htab(shift);
1242     if (rc < 0) {
1243         /* kernel-side HPT needed, but couldn't allocate one */
1244         error_setg_errno(errp, errno,
1245                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1246                          shift);
1247         /* This is almost certainly fatal, but if the caller really
1248          * wants to carry on with shift == 0, it's welcome to try */
1249     } else if (rc > 0) {
1250         /* kernel-side HPT allocated */
1251         if (rc != shift) {
1252             error_setg(errp,
1253                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1254                        shift, rc);
1255         }
1256 
1257         spapr->htab_shift = shift;
1258         spapr->htab = NULL;
1259     } else {
1260         /* kernel-side HPT not needed, allocate in userspace instead */
1261         size_t size = 1ULL << shift;
1262         int i;
1263 
1264         spapr->htab = qemu_memalign(size, size);
1265         if (!spapr->htab) {
1266             error_setg_errno(errp, errno,
1267                              "Could not allocate HPT of order %d", shift);
1268             return;
1269         }
1270 
1271         memset(spapr->htab, 0, size);
1272         spapr->htab_shift = shift;
1273 
1274         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1275             DIRTY_HPTE(HPTE(spapr->htab, i));
1276         }
1277     }
1278 }
1279 
1280 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1281 {
1282     spapr_reallocate_hpt(spapr,
1283                      spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size),
1284                      &error_fatal);
1285     if (spapr->vrma_adjust) {
1286         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1287                                           spapr->htab_shift);
1288     }
1289     /* We're setting up a hash table, so that means we're not radix */
1290     spapr->patb_entry = 0;
1291 }
1292 
1293 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1294 {
1295     bool matched = false;
1296 
1297     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1298         matched = true;
1299     }
1300 
1301     if (!matched) {
1302         error_report("Device %s is not supported by this machine yet.",
1303                      qdev_fw_name(DEVICE(sbdev)));
1304         exit(1);
1305     }
1306 }
1307 
1308 static void ppc_spapr_reset(void)
1309 {
1310     MachineState *machine = MACHINE(qdev_get_machine());
1311     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1312     PowerPCCPU *first_ppc_cpu;
1313     uint32_t rtas_limit;
1314     hwaddr rtas_addr, fdt_addr;
1315     void *fdt;
1316     int rc;
1317 
1318     /* Check for unknown sysbus devices */
1319     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1320 
1321     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1322         /* If using KVM with radix mode available, VCPUs can be started
1323          * without a HPT because KVM will start them in radix mode.
1324          * Set the GR bit in PATB so that we know there is no HPT. */
1325         spapr->patb_entry = PATBE1_GR;
1326     } else {
1327         spapr->patb_entry = 0;
1328         spapr_setup_hpt_and_vrma(spapr);
1329     }
1330 
1331     qemu_devices_reset();
1332 
1333     /*
1334      * We place the device tree and RTAS just below either the top of the RMA,
1335      * or just below 2GB, whichever is lowere, so that it can be
1336      * processed with 32-bit real mode code if necessary
1337      */
1338     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1339     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1340     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1341 
1342     /* if this reset wasn't generated by CAS, we should reset our
1343      * negotiated options and start from scratch */
1344     if (!spapr->cas_reboot) {
1345         spapr_ovec_cleanup(spapr->ov5_cas);
1346         spapr->ov5_cas = spapr_ovec_new();
1347     }
1348 
1349     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1350 
1351     spapr_load_rtas(spapr, fdt, rtas_addr);
1352 
1353     rc = fdt_pack(fdt);
1354 
1355     /* Should only fail if we've built a corrupted tree */
1356     assert(rc == 0);
1357 
1358     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1359         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1360                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1361         exit(1);
1362     }
1363 
1364     /* Load the fdt */
1365     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1366     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1367     g_free(fdt);
1368 
1369     /* Set up the entry state */
1370     first_ppc_cpu = POWERPC_CPU(first_cpu);
1371     first_ppc_cpu->env.gpr[3] = fdt_addr;
1372     first_ppc_cpu->env.gpr[5] = 0;
1373     first_cpu->halted = 0;
1374     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1375 
1376     spapr->cas_reboot = false;
1377 }
1378 
1379 static void spapr_create_nvram(sPAPRMachineState *spapr)
1380 {
1381     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1382     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1383 
1384     if (dinfo) {
1385         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1386                             &error_fatal);
1387     }
1388 
1389     qdev_init_nofail(dev);
1390 
1391     spapr->nvram = (struct sPAPRNVRAM *)dev;
1392 }
1393 
1394 static void spapr_rtc_create(sPAPRMachineState *spapr)
1395 {
1396     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1397     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1398                               &error_fatal);
1399     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1400                               &error_fatal);
1401     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1402                               "date", &error_fatal);
1403 }
1404 
1405 /* Returns whether we want to use VGA or not */
1406 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1407 {
1408     switch (vga_interface_type) {
1409     case VGA_NONE:
1410         return false;
1411     case VGA_DEVICE:
1412         return true;
1413     case VGA_STD:
1414     case VGA_VIRTIO:
1415         return pci_vga_init(pci_bus) != NULL;
1416     default:
1417         error_setg(errp,
1418                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1419         return false;
1420     }
1421 }
1422 
1423 static int spapr_post_load(void *opaque, int version_id)
1424 {
1425     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1426     int err = 0;
1427 
1428     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1429         CPUState *cs;
1430         CPU_FOREACH(cs) {
1431             PowerPCCPU *cpu = POWERPC_CPU(cs);
1432             icp_resend(ICP(cpu->intc));
1433         }
1434     }
1435 
1436     /* In earlier versions, there was no separate qdev for the PAPR
1437      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1438      * So when migrating from those versions, poke the incoming offset
1439      * value into the RTC device */
1440     if (version_id < 3) {
1441         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1442     }
1443 
1444     return err;
1445 }
1446 
1447 static bool version_before_3(void *opaque, int version_id)
1448 {
1449     return version_id < 3;
1450 }
1451 
1452 static bool spapr_ov5_cas_needed(void *opaque)
1453 {
1454     sPAPRMachineState *spapr = opaque;
1455     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1456     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1457     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1458     bool cas_needed;
1459 
1460     /* Prior to the introduction of sPAPROptionVector, we had two option
1461      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1462      * Both of these options encode machine topology into the device-tree
1463      * in such a way that the now-booted OS should still be able to interact
1464      * appropriately with QEMU regardless of what options were actually
1465      * negotiatied on the source side.
1466      *
1467      * As such, we can avoid migrating the CAS-negotiated options if these
1468      * are the only options available on the current machine/platform.
1469      * Since these are the only options available for pseries-2.7 and
1470      * earlier, this allows us to maintain old->new/new->old migration
1471      * compatibility.
1472      *
1473      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1474      * via default pseries-2.8 machines and explicit command-line parameters.
1475      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1476      * of the actual CAS-negotiated values to continue working properly. For
1477      * example, availability of memory unplug depends on knowing whether
1478      * OV5_HP_EVT was negotiated via CAS.
1479      *
1480      * Thus, for any cases where the set of available CAS-negotiatable
1481      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1482      * include the CAS-negotiated options in the migration stream.
1483      */
1484     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1485     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1486 
1487     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1488      * the mask itself since in the future it's possible "legacy" bits may be
1489      * removed via machine options, which could generate a false positive
1490      * that breaks migration.
1491      */
1492     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1493     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1494 
1495     spapr_ovec_cleanup(ov5_mask);
1496     spapr_ovec_cleanup(ov5_legacy);
1497     spapr_ovec_cleanup(ov5_removed);
1498 
1499     return cas_needed;
1500 }
1501 
1502 static const VMStateDescription vmstate_spapr_ov5_cas = {
1503     .name = "spapr_option_vector_ov5_cas",
1504     .version_id = 1,
1505     .minimum_version_id = 1,
1506     .needed = spapr_ov5_cas_needed,
1507     .fields = (VMStateField[]) {
1508         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1509                                  vmstate_spapr_ovec, sPAPROptionVector),
1510         VMSTATE_END_OF_LIST()
1511     },
1512 };
1513 
1514 static bool spapr_patb_entry_needed(void *opaque)
1515 {
1516     sPAPRMachineState *spapr = opaque;
1517 
1518     return !!spapr->patb_entry;
1519 }
1520 
1521 static const VMStateDescription vmstate_spapr_patb_entry = {
1522     .name = "spapr_patb_entry",
1523     .version_id = 1,
1524     .minimum_version_id = 1,
1525     .needed = spapr_patb_entry_needed,
1526     .fields = (VMStateField[]) {
1527         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1528         VMSTATE_END_OF_LIST()
1529     },
1530 };
1531 
1532 static const VMStateDescription vmstate_spapr = {
1533     .name = "spapr",
1534     .version_id = 3,
1535     .minimum_version_id = 1,
1536     .post_load = spapr_post_load,
1537     .fields = (VMStateField[]) {
1538         /* used to be @next_irq */
1539         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1540 
1541         /* RTC offset */
1542         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1543 
1544         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1545         VMSTATE_END_OF_LIST()
1546     },
1547     .subsections = (const VMStateDescription*[]) {
1548         &vmstate_spapr_ov5_cas,
1549         &vmstate_spapr_patb_entry,
1550         NULL
1551     }
1552 };
1553 
1554 static int htab_save_setup(QEMUFile *f, void *opaque)
1555 {
1556     sPAPRMachineState *spapr = opaque;
1557 
1558     /* "Iteration" header */
1559     qemu_put_be32(f, spapr->htab_shift);
1560 
1561     if (spapr->htab) {
1562         spapr->htab_save_index = 0;
1563         spapr->htab_first_pass = true;
1564     } else {
1565         assert(kvm_enabled());
1566     }
1567 
1568 
1569     return 0;
1570 }
1571 
1572 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1573                                  int64_t max_ns)
1574 {
1575     bool has_timeout = max_ns != -1;
1576     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1577     int index = spapr->htab_save_index;
1578     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1579 
1580     assert(spapr->htab_first_pass);
1581 
1582     do {
1583         int chunkstart;
1584 
1585         /* Consume invalid HPTEs */
1586         while ((index < htabslots)
1587                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1588             CLEAN_HPTE(HPTE(spapr->htab, index));
1589             index++;
1590         }
1591 
1592         /* Consume valid HPTEs */
1593         chunkstart = index;
1594         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1595                && HPTE_VALID(HPTE(spapr->htab, index))) {
1596             CLEAN_HPTE(HPTE(spapr->htab, index));
1597             index++;
1598         }
1599 
1600         if (index > chunkstart) {
1601             int n_valid = index - chunkstart;
1602 
1603             qemu_put_be32(f, chunkstart);
1604             qemu_put_be16(f, n_valid);
1605             qemu_put_be16(f, 0);
1606             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1607                             HASH_PTE_SIZE_64 * n_valid);
1608 
1609             if (has_timeout &&
1610                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1611                 break;
1612             }
1613         }
1614     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1615 
1616     if (index >= htabslots) {
1617         assert(index == htabslots);
1618         index = 0;
1619         spapr->htab_first_pass = false;
1620     }
1621     spapr->htab_save_index = index;
1622 }
1623 
1624 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1625                                 int64_t max_ns)
1626 {
1627     bool final = max_ns < 0;
1628     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1629     int examined = 0, sent = 0;
1630     int index = spapr->htab_save_index;
1631     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1632 
1633     assert(!spapr->htab_first_pass);
1634 
1635     do {
1636         int chunkstart, invalidstart;
1637 
1638         /* Consume non-dirty HPTEs */
1639         while ((index < htabslots)
1640                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1641             index++;
1642             examined++;
1643         }
1644 
1645         chunkstart = index;
1646         /* Consume valid dirty HPTEs */
1647         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1648                && HPTE_DIRTY(HPTE(spapr->htab, index))
1649                && HPTE_VALID(HPTE(spapr->htab, index))) {
1650             CLEAN_HPTE(HPTE(spapr->htab, index));
1651             index++;
1652             examined++;
1653         }
1654 
1655         invalidstart = index;
1656         /* Consume invalid dirty HPTEs */
1657         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1658                && HPTE_DIRTY(HPTE(spapr->htab, index))
1659                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1660             CLEAN_HPTE(HPTE(spapr->htab, index));
1661             index++;
1662             examined++;
1663         }
1664 
1665         if (index > chunkstart) {
1666             int n_valid = invalidstart - chunkstart;
1667             int n_invalid = index - invalidstart;
1668 
1669             qemu_put_be32(f, chunkstart);
1670             qemu_put_be16(f, n_valid);
1671             qemu_put_be16(f, n_invalid);
1672             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1673                             HASH_PTE_SIZE_64 * n_valid);
1674             sent += index - chunkstart;
1675 
1676             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1677                 break;
1678             }
1679         }
1680 
1681         if (examined >= htabslots) {
1682             break;
1683         }
1684 
1685         if (index >= htabslots) {
1686             assert(index == htabslots);
1687             index = 0;
1688         }
1689     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1690 
1691     if (index >= htabslots) {
1692         assert(index == htabslots);
1693         index = 0;
1694     }
1695 
1696     spapr->htab_save_index = index;
1697 
1698     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1699 }
1700 
1701 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1702 #define MAX_KVM_BUF_SIZE    2048
1703 
1704 static int htab_save_iterate(QEMUFile *f, void *opaque)
1705 {
1706     sPAPRMachineState *spapr = opaque;
1707     int fd;
1708     int rc = 0;
1709 
1710     /* Iteration header */
1711     qemu_put_be32(f, 0);
1712 
1713     if (!spapr->htab) {
1714         assert(kvm_enabled());
1715 
1716         fd = get_htab_fd(spapr);
1717         if (fd < 0) {
1718             return fd;
1719         }
1720 
1721         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1722         if (rc < 0) {
1723             return rc;
1724         }
1725     } else  if (spapr->htab_first_pass) {
1726         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1727     } else {
1728         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1729     }
1730 
1731     /* End marker */
1732     qemu_put_be32(f, 0);
1733     qemu_put_be16(f, 0);
1734     qemu_put_be16(f, 0);
1735 
1736     return rc;
1737 }
1738 
1739 static int htab_save_complete(QEMUFile *f, void *opaque)
1740 {
1741     sPAPRMachineState *spapr = opaque;
1742     int fd;
1743 
1744     /* Iteration header */
1745     qemu_put_be32(f, 0);
1746 
1747     if (!spapr->htab) {
1748         int rc;
1749 
1750         assert(kvm_enabled());
1751 
1752         fd = get_htab_fd(spapr);
1753         if (fd < 0) {
1754             return fd;
1755         }
1756 
1757         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1758         if (rc < 0) {
1759             return rc;
1760         }
1761     } else {
1762         if (spapr->htab_first_pass) {
1763             htab_save_first_pass(f, spapr, -1);
1764         }
1765         htab_save_later_pass(f, spapr, -1);
1766     }
1767 
1768     /* End marker */
1769     qemu_put_be32(f, 0);
1770     qemu_put_be16(f, 0);
1771     qemu_put_be16(f, 0);
1772 
1773     return 0;
1774 }
1775 
1776 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1777 {
1778     sPAPRMachineState *spapr = opaque;
1779     uint32_t section_hdr;
1780     int fd = -1;
1781 
1782     if (version_id < 1 || version_id > 1) {
1783         error_report("htab_load() bad version");
1784         return -EINVAL;
1785     }
1786 
1787     section_hdr = qemu_get_be32(f);
1788 
1789     if (section_hdr) {
1790         Error *local_err = NULL;
1791 
1792         /* First section gives the htab size */
1793         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1794         if (local_err) {
1795             error_report_err(local_err);
1796             return -EINVAL;
1797         }
1798         return 0;
1799     }
1800 
1801     if (!spapr->htab) {
1802         assert(kvm_enabled());
1803 
1804         fd = kvmppc_get_htab_fd(true);
1805         if (fd < 0) {
1806             error_report("Unable to open fd to restore KVM hash table: %s",
1807                          strerror(errno));
1808         }
1809     }
1810 
1811     while (true) {
1812         uint32_t index;
1813         uint16_t n_valid, n_invalid;
1814 
1815         index = qemu_get_be32(f);
1816         n_valid = qemu_get_be16(f);
1817         n_invalid = qemu_get_be16(f);
1818 
1819         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1820             /* End of Stream */
1821             break;
1822         }
1823 
1824         if ((index + n_valid + n_invalid) >
1825             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1826             /* Bad index in stream */
1827             error_report(
1828                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1829                 index, n_valid, n_invalid, spapr->htab_shift);
1830             return -EINVAL;
1831         }
1832 
1833         if (spapr->htab) {
1834             if (n_valid) {
1835                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1836                                 HASH_PTE_SIZE_64 * n_valid);
1837             }
1838             if (n_invalid) {
1839                 memset(HPTE(spapr->htab, index + n_valid), 0,
1840                        HASH_PTE_SIZE_64 * n_invalid);
1841             }
1842         } else {
1843             int rc;
1844 
1845             assert(fd >= 0);
1846 
1847             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1848             if (rc < 0) {
1849                 return rc;
1850             }
1851         }
1852     }
1853 
1854     if (!spapr->htab) {
1855         assert(fd >= 0);
1856         close(fd);
1857     }
1858 
1859     return 0;
1860 }
1861 
1862 static void htab_cleanup(void *opaque)
1863 {
1864     sPAPRMachineState *spapr = opaque;
1865 
1866     close_htab_fd(spapr);
1867 }
1868 
1869 static SaveVMHandlers savevm_htab_handlers = {
1870     .save_live_setup = htab_save_setup,
1871     .save_live_iterate = htab_save_iterate,
1872     .save_live_complete_precopy = htab_save_complete,
1873     .cleanup = htab_cleanup,
1874     .load_state = htab_load,
1875 };
1876 
1877 static void spapr_boot_set(void *opaque, const char *boot_device,
1878                            Error **errp)
1879 {
1880     MachineState *machine = MACHINE(qdev_get_machine());
1881     machine->boot_order = g_strdup(boot_device);
1882 }
1883 
1884 /*
1885  * Reset routine for LMB DR devices.
1886  *
1887  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1888  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1889  * when it walks all its children devices. LMB devices reset occurs
1890  * as part of spapr_ppc_reset().
1891  */
1892 static void spapr_drc_reset(void *opaque)
1893 {
1894     sPAPRDRConnector *drc = opaque;
1895     DeviceState *d = DEVICE(drc);
1896 
1897     if (d) {
1898         device_reset(d);
1899     }
1900 }
1901 
1902 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1903 {
1904     MachineState *machine = MACHINE(spapr);
1905     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1906     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1907     int i;
1908 
1909     for (i = 0; i < nr_lmbs; i++) {
1910         sPAPRDRConnector *drc;
1911         uint64_t addr;
1912 
1913         addr = i * lmb_size + spapr->hotplug_memory.base;
1914         drc = spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
1915                                      addr/lmb_size);
1916         qemu_register_reset(spapr_drc_reset, drc);
1917     }
1918 }
1919 
1920 /*
1921  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1922  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1923  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1924  */
1925 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1926 {
1927     int i;
1928 
1929     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1930         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1931                    " is not aligned to %llu MiB",
1932                    machine->ram_size,
1933                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1934         return;
1935     }
1936 
1937     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1938         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1939                    " is not aligned to %llu MiB",
1940                    machine->ram_size,
1941                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1942         return;
1943     }
1944 
1945     for (i = 0; i < nb_numa_nodes; i++) {
1946         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1947             error_setg(errp,
1948                        "Node %d memory size 0x%" PRIx64
1949                        " is not aligned to %llu MiB",
1950                        i, numa_info[i].node_mem,
1951                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1952             return;
1953         }
1954     }
1955 }
1956 
1957 /* find cpu slot in machine->possible_cpus by core_id */
1958 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1959 {
1960     int index = id / smp_threads;
1961 
1962     if (index >= ms->possible_cpus->len) {
1963         return NULL;
1964     }
1965     if (idx) {
1966         *idx = index;
1967     }
1968     return &ms->possible_cpus->cpus[index];
1969 }
1970 
1971 static void spapr_init_cpus(sPAPRMachineState *spapr)
1972 {
1973     MachineState *machine = MACHINE(spapr);
1974     MachineClass *mc = MACHINE_GET_CLASS(machine);
1975     char *type = spapr_get_cpu_core_type(machine->cpu_model);
1976     int smt = kvmppc_smt_threads();
1977     const CPUArchIdList *possible_cpus;
1978     int boot_cores_nr = smp_cpus / smp_threads;
1979     int i;
1980 
1981     if (!type) {
1982         error_report("Unable to find sPAPR CPU Core definition");
1983         exit(1);
1984     }
1985 
1986     possible_cpus = mc->possible_cpu_arch_ids(machine);
1987     if (mc->has_hotpluggable_cpus) {
1988         if (smp_cpus % smp_threads) {
1989             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1990                          smp_cpus, smp_threads);
1991             exit(1);
1992         }
1993         if (max_cpus % smp_threads) {
1994             error_report("max_cpus (%u) must be multiple of threads (%u)",
1995                          max_cpus, smp_threads);
1996             exit(1);
1997         }
1998     } else {
1999         if (max_cpus != smp_cpus) {
2000             error_report("This machine version does not support CPU hotplug");
2001             exit(1);
2002         }
2003         boot_cores_nr = possible_cpus->len;
2004     }
2005 
2006     for (i = 0; i < possible_cpus->len; i++) {
2007         int core_id = i * smp_threads;
2008 
2009         if (mc->has_hotpluggable_cpus) {
2010             sPAPRDRConnector *drc =
2011                 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2012                                        (core_id / smp_threads) * smt);
2013 
2014             qemu_register_reset(spapr_drc_reset, drc);
2015         }
2016 
2017         if (i < boot_cores_nr) {
2018             Object *core  = object_new(type);
2019             int nr_threads = smp_threads;
2020 
2021             /* Handle the partially filled core for older machine types */
2022             if ((i + 1) * smp_threads >= smp_cpus) {
2023                 nr_threads = smp_cpus - i * smp_threads;
2024             }
2025 
2026             object_property_set_int(core, nr_threads, "nr-threads",
2027                                     &error_fatal);
2028             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2029                                     &error_fatal);
2030             object_property_set_bool(core, true, "realized", &error_fatal);
2031         }
2032     }
2033     g_free(type);
2034 }
2035 
2036 /* pSeries LPAR / sPAPR hardware init */
2037 static void ppc_spapr_init(MachineState *machine)
2038 {
2039     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2040     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2041     const char *kernel_filename = machine->kernel_filename;
2042     const char *initrd_filename = machine->initrd_filename;
2043     PCIHostState *phb;
2044     int i;
2045     MemoryRegion *sysmem = get_system_memory();
2046     MemoryRegion *ram = g_new(MemoryRegion, 1);
2047     MemoryRegion *rma_region;
2048     void *rma = NULL;
2049     hwaddr rma_alloc_size;
2050     hwaddr node0_size = spapr_node0_size();
2051     long load_limit, fw_size;
2052     char *filename;
2053 
2054     msi_nonbroken = true;
2055 
2056     QLIST_INIT(&spapr->phbs);
2057     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2058 
2059     /* Allocate RMA if necessary */
2060     rma_alloc_size = kvmppc_alloc_rma(&rma);
2061 
2062     if (rma_alloc_size == -1) {
2063         error_report("Unable to create RMA");
2064         exit(1);
2065     }
2066 
2067     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2068         spapr->rma_size = rma_alloc_size;
2069     } else {
2070         spapr->rma_size = node0_size;
2071 
2072         /* With KVM, we don't actually know whether KVM supports an
2073          * unbounded RMA (PR KVM) or is limited by the hash table size
2074          * (HV KVM using VRMA), so we always assume the latter
2075          *
2076          * In that case, we also limit the initial allocations for RTAS
2077          * etc... to 256M since we have no way to know what the VRMA size
2078          * is going to be as it depends on the size of the hash table
2079          * isn't determined yet.
2080          */
2081         if (kvm_enabled()) {
2082             spapr->vrma_adjust = 1;
2083             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2084         }
2085 
2086         /* Actually we don't support unbounded RMA anymore since we
2087          * added proper emulation of HV mode. The max we can get is
2088          * 16G which also happens to be what we configure for PAPR
2089          * mode so make sure we don't do anything bigger than that
2090          */
2091         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2092     }
2093 
2094     if (spapr->rma_size > node0_size) {
2095         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2096                      spapr->rma_size);
2097         exit(1);
2098     }
2099 
2100     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2101     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2102 
2103     /* Set up Interrupt Controller before we create the VCPUs */
2104     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2105 
2106     /* Set up containers for ibm,client-set-architecture negotiated options */
2107     spapr->ov5 = spapr_ovec_new();
2108     spapr->ov5_cas = spapr_ovec_new();
2109 
2110     if (smc->dr_lmb_enabled) {
2111         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2112         spapr_validate_node_memory(machine, &error_fatal);
2113     }
2114 
2115     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2116     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2117         /* KVM and TCG always allow GTSE with radix... */
2118         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2119     }
2120     /* ... but not with hash (currently). */
2121 
2122     /* advertise support for dedicated HP event source to guests */
2123     if (spapr->use_hotplug_event_source) {
2124         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2125     }
2126 
2127     /* init CPUs */
2128     if (machine->cpu_model == NULL) {
2129         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2130     }
2131 
2132     ppc_cpu_parse_features(machine->cpu_model);
2133 
2134     spapr_init_cpus(spapr);
2135 
2136     if (kvm_enabled()) {
2137         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2138         kvmppc_enable_logical_ci_hcalls();
2139         kvmppc_enable_set_mode_hcall();
2140 
2141         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2142         kvmppc_enable_clear_ref_mod_hcalls();
2143     }
2144 
2145     /* allocate RAM */
2146     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2147                                          machine->ram_size);
2148     memory_region_add_subregion(sysmem, 0, ram);
2149 
2150     if (rma_alloc_size && rma) {
2151         rma_region = g_new(MemoryRegion, 1);
2152         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2153                                    rma_alloc_size, rma);
2154         vmstate_register_ram_global(rma_region);
2155         memory_region_add_subregion(sysmem, 0, rma_region);
2156     }
2157 
2158     /* initialize hotplug memory address space */
2159     if (machine->ram_size < machine->maxram_size) {
2160         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2161         /*
2162          * Limit the number of hotpluggable memory slots to half the number
2163          * slots that KVM supports, leaving the other half for PCI and other
2164          * devices. However ensure that number of slots doesn't drop below 32.
2165          */
2166         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2167                            SPAPR_MAX_RAM_SLOTS;
2168 
2169         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2170             max_memslots = SPAPR_MAX_RAM_SLOTS;
2171         }
2172         if (machine->ram_slots > max_memslots) {
2173             error_report("Specified number of memory slots %"
2174                          PRIu64" exceeds max supported %d",
2175                          machine->ram_slots, max_memslots);
2176             exit(1);
2177         }
2178 
2179         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2180                                               SPAPR_HOTPLUG_MEM_ALIGN);
2181         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2182                            "hotplug-memory", hotplug_mem_size);
2183         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2184                                     &spapr->hotplug_memory.mr);
2185     }
2186 
2187     if (smc->dr_lmb_enabled) {
2188         spapr_create_lmb_dr_connectors(spapr);
2189     }
2190 
2191     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2192     if (!filename) {
2193         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2194         exit(1);
2195     }
2196     spapr->rtas_size = get_image_size(filename);
2197     if (spapr->rtas_size < 0) {
2198         error_report("Could not get size of LPAR rtas '%s'", filename);
2199         exit(1);
2200     }
2201     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2202     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2203         error_report("Could not load LPAR rtas '%s'", filename);
2204         exit(1);
2205     }
2206     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2207         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2208                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2209         exit(1);
2210     }
2211     g_free(filename);
2212 
2213     /* Set up RTAS event infrastructure */
2214     spapr_events_init(spapr);
2215 
2216     /* Set up the RTC RTAS interfaces */
2217     spapr_rtc_create(spapr);
2218 
2219     /* Set up VIO bus */
2220     spapr->vio_bus = spapr_vio_bus_init();
2221 
2222     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2223         if (serial_hds[i]) {
2224             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2225         }
2226     }
2227 
2228     /* We always have at least the nvram device on VIO */
2229     spapr_create_nvram(spapr);
2230 
2231     /* Set up PCI */
2232     spapr_pci_rtas_init();
2233 
2234     phb = spapr_create_phb(spapr, 0);
2235 
2236     for (i = 0; i < nb_nics; i++) {
2237         NICInfo *nd = &nd_table[i];
2238 
2239         if (!nd->model) {
2240             nd->model = g_strdup("ibmveth");
2241         }
2242 
2243         if (strcmp(nd->model, "ibmveth") == 0) {
2244             spapr_vlan_create(spapr->vio_bus, nd);
2245         } else {
2246             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2247         }
2248     }
2249 
2250     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2251         spapr_vscsi_create(spapr->vio_bus);
2252     }
2253 
2254     /* Graphics */
2255     if (spapr_vga_init(phb->bus, &error_fatal)) {
2256         spapr->has_graphics = true;
2257         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2258     }
2259 
2260     if (machine->usb) {
2261         if (smc->use_ohci_by_default) {
2262             pci_create_simple(phb->bus, -1, "pci-ohci");
2263         } else {
2264             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2265         }
2266 
2267         if (spapr->has_graphics) {
2268             USBBus *usb_bus = usb_bus_find(-1);
2269 
2270             usb_create_simple(usb_bus, "usb-kbd");
2271             usb_create_simple(usb_bus, "usb-mouse");
2272         }
2273     }
2274 
2275     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2276         error_report(
2277             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2278             MIN_RMA_SLOF);
2279         exit(1);
2280     }
2281 
2282     if (kernel_filename) {
2283         uint64_t lowaddr = 0;
2284 
2285         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2286                                       NULL, NULL, &lowaddr, NULL, 1,
2287                                       PPC_ELF_MACHINE, 0, 0);
2288         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2289             spapr->kernel_size = load_elf(kernel_filename,
2290                                           translate_kernel_address, NULL, NULL,
2291                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2292                                           0, 0);
2293             spapr->kernel_le = spapr->kernel_size > 0;
2294         }
2295         if (spapr->kernel_size < 0) {
2296             error_report("error loading %s: %s", kernel_filename,
2297                          load_elf_strerror(spapr->kernel_size));
2298             exit(1);
2299         }
2300 
2301         /* load initrd */
2302         if (initrd_filename) {
2303             /* Try to locate the initrd in the gap between the kernel
2304              * and the firmware. Add a bit of space just in case
2305              */
2306             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2307                                   + 0x1ffff) & ~0xffff;
2308             spapr->initrd_size = load_image_targphys(initrd_filename,
2309                                                      spapr->initrd_base,
2310                                                      load_limit
2311                                                      - spapr->initrd_base);
2312             if (spapr->initrd_size < 0) {
2313                 error_report("could not load initial ram disk '%s'",
2314                              initrd_filename);
2315                 exit(1);
2316             }
2317         }
2318     }
2319 
2320     if (bios_name == NULL) {
2321         bios_name = FW_FILE_NAME;
2322     }
2323     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2324     if (!filename) {
2325         error_report("Could not find LPAR firmware '%s'", bios_name);
2326         exit(1);
2327     }
2328     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2329     if (fw_size <= 0) {
2330         error_report("Could not load LPAR firmware '%s'", filename);
2331         exit(1);
2332     }
2333     g_free(filename);
2334 
2335     /* FIXME: Should register things through the MachineState's qdev
2336      * interface, this is a legacy from the sPAPREnvironment structure
2337      * which predated MachineState but had a similar function */
2338     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2339     register_savevm_live(NULL, "spapr/htab", -1, 1,
2340                          &savevm_htab_handlers, spapr);
2341 
2342     qemu_register_boot_set(spapr_boot_set, spapr);
2343 
2344     if (kvm_enabled()) {
2345         /* to stop and start vmclock */
2346         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2347                                          &spapr->tb);
2348 
2349         kvmppc_spapr_enable_inkernel_multitce();
2350     }
2351 }
2352 
2353 static int spapr_kvm_type(const char *vm_type)
2354 {
2355     if (!vm_type) {
2356         return 0;
2357     }
2358 
2359     if (!strcmp(vm_type, "HV")) {
2360         return 1;
2361     }
2362 
2363     if (!strcmp(vm_type, "PR")) {
2364         return 2;
2365     }
2366 
2367     error_report("Unknown kvm-type specified '%s'", vm_type);
2368     exit(1);
2369 }
2370 
2371 /*
2372  * Implementation of an interface to adjust firmware path
2373  * for the bootindex property handling.
2374  */
2375 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2376                                    DeviceState *dev)
2377 {
2378 #define CAST(type, obj, name) \
2379     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2380     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2381     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2382     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2383 
2384     if (d) {
2385         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2386         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2387         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2388 
2389         if (spapr) {
2390             /*
2391              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2392              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2393              * in the top 16 bits of the 64-bit LUN
2394              */
2395             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2396             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2397                                    (uint64_t)id << 48);
2398         } else if (virtio) {
2399             /*
2400              * We use SRP luns of the form 01000000 | (target << 8) | lun
2401              * in the top 32 bits of the 64-bit LUN
2402              * Note: the quote above is from SLOF and it is wrong,
2403              * the actual binding is:
2404              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2405              */
2406             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2407             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2408                                    (uint64_t)id << 32);
2409         } else if (usb) {
2410             /*
2411              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2412              * in the top 32 bits of the 64-bit LUN
2413              */
2414             unsigned usb_port = atoi(usb->port->path);
2415             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2416             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2417                                    (uint64_t)id << 32);
2418         }
2419     }
2420 
2421     /*
2422      * SLOF probes the USB devices, and if it recognizes that the device is a
2423      * storage device, it changes its name to "storage" instead of "usb-host",
2424      * and additionally adds a child node for the SCSI LUN, so the correct
2425      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2426      */
2427     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2428         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2429         if (usb_host_dev_is_scsi_storage(usbdev)) {
2430             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2431         }
2432     }
2433 
2434     if (phb) {
2435         /* Replace "pci" with "pci@800000020000000" */
2436         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2437     }
2438 
2439     if (vsc) {
2440         /* Same logic as virtio above */
2441         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2442         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2443     }
2444 
2445     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2446         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2447         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2448         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2449     }
2450 
2451     return NULL;
2452 }
2453 
2454 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2455 {
2456     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2457 
2458     return g_strdup(spapr->kvm_type);
2459 }
2460 
2461 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2462 {
2463     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2464 
2465     g_free(spapr->kvm_type);
2466     spapr->kvm_type = g_strdup(value);
2467 }
2468 
2469 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2470 {
2471     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2472 
2473     return spapr->use_hotplug_event_source;
2474 }
2475 
2476 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2477                                             Error **errp)
2478 {
2479     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2480 
2481     spapr->use_hotplug_event_source = value;
2482 }
2483 
2484 static void spapr_machine_initfn(Object *obj)
2485 {
2486     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2487 
2488     spapr->htab_fd = -1;
2489     spapr->use_hotplug_event_source = true;
2490     object_property_add_str(obj, "kvm-type",
2491                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2492     object_property_set_description(obj, "kvm-type",
2493                                     "Specifies the KVM virtualization mode (HV, PR)",
2494                                     NULL);
2495     object_property_add_bool(obj, "modern-hotplug-events",
2496                             spapr_get_modern_hotplug_events,
2497                             spapr_set_modern_hotplug_events,
2498                             NULL);
2499     object_property_set_description(obj, "modern-hotplug-events",
2500                                     "Use dedicated hotplug event mechanism in"
2501                                     " place of standard EPOW events when possible"
2502                                     " (required for memory hot-unplug support)",
2503                                     NULL);
2504 }
2505 
2506 static void spapr_machine_finalizefn(Object *obj)
2507 {
2508     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2509 
2510     g_free(spapr->kvm_type);
2511 }
2512 
2513 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2514 {
2515     cpu_synchronize_state(cs);
2516     ppc_cpu_do_system_reset(cs);
2517 }
2518 
2519 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2520 {
2521     CPUState *cs;
2522 
2523     CPU_FOREACH(cs) {
2524         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2525     }
2526 }
2527 
2528 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2529                            uint32_t node, bool dedicated_hp_event_source,
2530                            Error **errp)
2531 {
2532     sPAPRDRConnector *drc;
2533     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2534     int i, fdt_offset, fdt_size;
2535     void *fdt;
2536     uint64_t addr = addr_start;
2537 
2538     for (i = 0; i < nr_lmbs; i++) {
2539         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2540                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2541         g_assert(drc);
2542 
2543         fdt = create_device_tree(&fdt_size);
2544         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2545                                                 SPAPR_MEMORY_BLOCK_SIZE);
2546 
2547         spapr_drc_attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2548         addr += SPAPR_MEMORY_BLOCK_SIZE;
2549         if (!dev->hotplugged) {
2550             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2551             /* guests expect coldplugged LMBs to be pre-allocated */
2552             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2553             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2554         }
2555     }
2556     /* send hotplug notification to the
2557      * guest only in case of hotplugged memory
2558      */
2559     if (dev->hotplugged) {
2560         if (dedicated_hp_event_source) {
2561             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2562                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2563             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2564                                                    nr_lmbs,
2565                                                    spapr_drc_index(drc));
2566         } else {
2567             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2568                                            nr_lmbs);
2569         }
2570     }
2571 }
2572 
2573 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2574                               uint32_t node, Error **errp)
2575 {
2576     Error *local_err = NULL;
2577     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2578     PCDIMMDevice *dimm = PC_DIMM(dev);
2579     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2580     MemoryRegion *mr = ddc->get_memory_region(dimm);
2581     uint64_t align = memory_region_get_alignment(mr);
2582     uint64_t size = memory_region_size(mr);
2583     uint64_t addr;
2584 
2585     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2586     if (local_err) {
2587         goto out;
2588     }
2589 
2590     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2591     if (local_err) {
2592         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2593         goto out;
2594     }
2595 
2596     spapr_add_lmbs(dev, addr, size, node,
2597                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2598                    &error_abort);
2599 
2600 out:
2601     error_propagate(errp, local_err);
2602 }
2603 
2604 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2605                                   Error **errp)
2606 {
2607     PCDIMMDevice *dimm = PC_DIMM(dev);
2608     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2609     MemoryRegion *mr = ddc->get_memory_region(dimm);
2610     uint64_t size = memory_region_size(mr);
2611     char *mem_dev;
2612 
2613     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2614         error_setg(errp, "Hotplugged memory size must be a multiple of "
2615                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2616         return;
2617     }
2618 
2619     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2620     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2621         error_setg(errp, "Memory backend has bad page size. "
2622                    "Use 'memory-backend-file' with correct mem-path.");
2623         goto out;
2624     }
2625 
2626 out:
2627     g_free(mem_dev);
2628 }
2629 
2630 struct sPAPRDIMMState {
2631     PCDIMMDevice *dimm;
2632     uint32_t nr_lmbs;
2633     QTAILQ_ENTRY(sPAPRDIMMState) next;
2634 };
2635 
2636 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
2637                                                        PCDIMMDevice *dimm)
2638 {
2639     sPAPRDIMMState *dimm_state = NULL;
2640 
2641     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
2642         if (dimm_state->dimm == dimm) {
2643             break;
2644         }
2645     }
2646     return dimm_state;
2647 }
2648 
2649 static void spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
2650                                            sPAPRDIMMState *dimm_state)
2651 {
2652     g_assert(!spapr_pending_dimm_unplugs_find(spapr, dimm_state->dimm));
2653     QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, dimm_state, next);
2654 }
2655 
2656 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
2657                                               sPAPRDIMMState *dimm_state)
2658 {
2659     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
2660     g_free(dimm_state);
2661 }
2662 
2663 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
2664                                                         PCDIMMDevice *dimm)
2665 {
2666     sPAPRDRConnector *drc;
2667     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2668     MemoryRegion *mr = ddc->get_memory_region(dimm);
2669     uint64_t size = memory_region_size(mr);
2670     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2671     uint32_t avail_lmbs = 0;
2672     uint64_t addr_start, addr;
2673     int i;
2674     sPAPRDIMMState *ds;
2675 
2676     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2677                                          &error_abort);
2678 
2679     addr = addr_start;
2680     for (i = 0; i < nr_lmbs; i++) {
2681         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2682                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2683         g_assert(drc);
2684         if (drc->dev) {
2685             avail_lmbs++;
2686         }
2687         addr += SPAPR_MEMORY_BLOCK_SIZE;
2688     }
2689 
2690     ds = g_malloc0(sizeof(sPAPRDIMMState));
2691     ds->nr_lmbs = avail_lmbs;
2692     ds->dimm = dimm;
2693     spapr_pending_dimm_unplugs_add(ms, ds);
2694     return ds;
2695 }
2696 
2697 /* Callback to be called during DRC release. */
2698 void spapr_lmb_release(DeviceState *dev)
2699 {
2700     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
2701     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
2702     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
2703 
2704     /* This information will get lost if a migration occurs
2705      * during the unplug process. In this case recover it. */
2706     if (ds == NULL) {
2707         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
2708         /* The DRC being examined by the caller at least must be counted */
2709         g_assert(ds->nr_lmbs);
2710     }
2711 
2712     if (--ds->nr_lmbs) {
2713         return;
2714     }
2715 
2716     spapr_pending_dimm_unplugs_remove(spapr, ds);
2717 
2718     /*
2719      * Now that all the LMBs have been removed by the guest, call the
2720      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2721      */
2722     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2723 }
2724 
2725 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2726                                 Error **errp)
2727 {
2728     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2729     PCDIMMDevice *dimm = PC_DIMM(dev);
2730     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2731     MemoryRegion *mr = ddc->get_memory_region(dimm);
2732 
2733     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2734     object_unparent(OBJECT(dev));
2735 }
2736 
2737 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2738                                         DeviceState *dev, Error **errp)
2739 {
2740     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
2741     Error *local_err = NULL;
2742     PCDIMMDevice *dimm = PC_DIMM(dev);
2743     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2744     MemoryRegion *mr = ddc->get_memory_region(dimm);
2745     uint64_t size = memory_region_size(mr);
2746     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2747     uint64_t addr_start, addr;
2748     int i;
2749     sPAPRDRConnector *drc;
2750     sPAPRDIMMState *ds;
2751 
2752     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2753                                          &local_err);
2754     if (local_err) {
2755         goto out;
2756     }
2757 
2758     ds = g_malloc0(sizeof(sPAPRDIMMState));
2759     ds->nr_lmbs = nr_lmbs;
2760     ds->dimm = dimm;
2761     spapr_pending_dimm_unplugs_add(spapr, ds);
2762 
2763     addr = addr_start;
2764     for (i = 0; i < nr_lmbs; i++) {
2765         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2766                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2767         g_assert(drc);
2768 
2769         spapr_drc_detach(drc, dev, errp);
2770         addr += SPAPR_MEMORY_BLOCK_SIZE;
2771     }
2772 
2773     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2774                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2775     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2776                                               nr_lmbs, spapr_drc_index(drc));
2777 out:
2778     error_propagate(errp, local_err);
2779 }
2780 
2781 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2782                                     sPAPRMachineState *spapr)
2783 {
2784     PowerPCCPU *cpu = POWERPC_CPU(cs);
2785     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2786     int id = ppc_get_vcpu_dt_id(cpu);
2787     void *fdt;
2788     int offset, fdt_size;
2789     char *nodename;
2790 
2791     fdt = create_device_tree(&fdt_size);
2792     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2793     offset = fdt_add_subnode(fdt, 0, nodename);
2794 
2795     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2796     g_free(nodename);
2797 
2798     *fdt_offset = offset;
2799     return fdt;
2800 }
2801 
2802 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2803                               Error **errp)
2804 {
2805     MachineState *ms = MACHINE(qdev_get_machine());
2806     CPUCore *cc = CPU_CORE(dev);
2807     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2808 
2809     assert(core_slot);
2810     core_slot->cpu = NULL;
2811     object_unparent(OBJECT(dev));
2812 }
2813 
2814 /* Callback to be called during DRC release. */
2815 void spapr_core_release(DeviceState *dev)
2816 {
2817     HotplugHandler *hotplug_ctrl;
2818 
2819     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2820     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2821 }
2822 
2823 static
2824 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2825                                Error **errp)
2826 {
2827     int index;
2828     sPAPRDRConnector *drc;
2829     Error *local_err = NULL;
2830     CPUCore *cc = CPU_CORE(dev);
2831     int smt = kvmppc_smt_threads();
2832 
2833     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2834         error_setg(errp, "Unable to find CPU core with core-id: %d",
2835                    cc->core_id);
2836         return;
2837     }
2838     if (index == 0) {
2839         error_setg(errp, "Boot CPU core may not be unplugged");
2840         return;
2841     }
2842 
2843     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
2844     g_assert(drc);
2845 
2846     spapr_drc_detach(drc, dev, &local_err);
2847     if (local_err) {
2848         error_propagate(errp, local_err);
2849         return;
2850     }
2851 
2852     spapr_hotplug_req_remove_by_index(drc);
2853 }
2854 
2855 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2856                             Error **errp)
2857 {
2858     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2859     MachineClass *mc = MACHINE_GET_CLASS(spapr);
2860     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2861     CPUCore *cc = CPU_CORE(dev);
2862     CPUState *cs = CPU(core->threads);
2863     sPAPRDRConnector *drc;
2864     Error *local_err = NULL;
2865     void *fdt = NULL;
2866     int fdt_offset = 0;
2867     int smt = kvmppc_smt_threads();
2868     CPUArchId *core_slot;
2869     int index;
2870 
2871     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2872     if (!core_slot) {
2873         error_setg(errp, "Unable to find CPU core with core-id: %d",
2874                    cc->core_id);
2875         return;
2876     }
2877     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
2878 
2879     g_assert(drc || !mc->has_hotpluggable_cpus);
2880 
2881     /*
2882      * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2883      * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2884      */
2885     if (dev->hotplugged) {
2886         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2887     }
2888 
2889     if (drc) {
2890         spapr_drc_attach(drc, dev, fdt, fdt_offset, !dev->hotplugged,
2891                          &local_err);
2892         if (local_err) {
2893             g_free(fdt);
2894             error_propagate(errp, local_err);
2895             return;
2896         }
2897     }
2898 
2899     if (dev->hotplugged) {
2900         /*
2901          * Send hotplug notification interrupt to the guest only in case
2902          * of hotplugged CPUs.
2903          */
2904         spapr_hotplug_req_add_by_index(drc);
2905     } else {
2906         /*
2907          * Set the right DRC states for cold plugged CPU.
2908          */
2909         if (drc) {
2910             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2911             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2912             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2913         }
2914     }
2915     core_slot->cpu = OBJECT(dev);
2916 }
2917 
2918 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2919                                 Error **errp)
2920 {
2921     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2922     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2923     Error *local_err = NULL;
2924     CPUCore *cc = CPU_CORE(dev);
2925     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2926     const char *type = object_get_typename(OBJECT(dev));
2927     CPUArchId *core_slot;
2928     int index;
2929 
2930     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2931         error_setg(&local_err, "CPU hotplug not supported for this machine");
2932         goto out;
2933     }
2934 
2935     if (strcmp(base_core_type, type)) {
2936         error_setg(&local_err, "CPU core type should be %s", base_core_type);
2937         goto out;
2938     }
2939 
2940     if (cc->core_id % smp_threads) {
2941         error_setg(&local_err, "invalid core id %d", cc->core_id);
2942         goto out;
2943     }
2944 
2945     /*
2946      * In general we should have homogeneous threads-per-core, but old
2947      * (pre hotplug support) machine types allow the last core to have
2948      * reduced threads as a compatibility hack for when we allowed
2949      * total vcpus not a multiple of threads-per-core.
2950      */
2951     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
2952         error_setg(errp, "invalid nr-threads %d, must be %d",
2953                    cc->nr_threads, smp_threads);
2954         return;
2955     }
2956 
2957     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2958     if (!core_slot) {
2959         error_setg(&local_err, "core id %d out of range", cc->core_id);
2960         goto out;
2961     }
2962 
2963     if (core_slot->cpu) {
2964         error_setg(&local_err, "core %d already populated", cc->core_id);
2965         goto out;
2966     }
2967 
2968     numa_cpu_pre_plug(core_slot, dev, &local_err);
2969 
2970 out:
2971     g_free(base_core_type);
2972     error_propagate(errp, local_err);
2973 }
2974 
2975 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2976                                       DeviceState *dev, Error **errp)
2977 {
2978     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2979 
2980     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2981         int node;
2982 
2983         if (!smc->dr_lmb_enabled) {
2984             error_setg(errp, "Memory hotplug not supported for this machine");
2985             return;
2986         }
2987         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2988         if (*errp) {
2989             return;
2990         }
2991         if (node < 0 || node >= MAX_NODES) {
2992             error_setg(errp, "Invaild node %d", node);
2993             return;
2994         }
2995 
2996         /*
2997          * Currently PowerPC kernel doesn't allow hot-adding memory to
2998          * memory-less node, but instead will silently add the memory
2999          * to the first node that has some memory. This causes two
3000          * unexpected behaviours for the user.
3001          *
3002          * - Memory gets hotplugged to a different node than what the user
3003          *   specified.
3004          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3005          *   to memory-less node, a reboot will set things accordingly
3006          *   and the previously hotplugged memory now ends in the right node.
3007          *   This appears as if some memory moved from one node to another.
3008          *
3009          * So until kernel starts supporting memory hotplug to memory-less
3010          * nodes, just prevent such attempts upfront in QEMU.
3011          */
3012         if (nb_numa_nodes && !numa_info[node].node_mem) {
3013             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3014                        node);
3015             return;
3016         }
3017 
3018         spapr_memory_plug(hotplug_dev, dev, node, errp);
3019     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3020         spapr_core_plug(hotplug_dev, dev, errp);
3021     }
3022 }
3023 
3024 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3025                                       DeviceState *dev, Error **errp)
3026 {
3027     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
3028     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
3029 
3030     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3031         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3032             spapr_memory_unplug(hotplug_dev, dev, errp);
3033         } else {
3034             error_setg(errp, "Memory hot unplug not supported for this guest");
3035         }
3036     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3037         if (!mc->has_hotpluggable_cpus) {
3038             error_setg(errp, "CPU hot unplug not supported on this machine");
3039             return;
3040         }
3041         spapr_core_unplug(hotplug_dev, dev, errp);
3042     }
3043 }
3044 
3045 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3046                                                 DeviceState *dev, Error **errp)
3047 {
3048     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
3049     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
3050 
3051     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3052         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3053             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3054         } else {
3055             /* NOTE: this means there is a window after guest reset, prior to
3056              * CAS negotiation, where unplug requests will fail due to the
3057              * capability not being detected yet. This is a bit different than
3058              * the case with PCI unplug, where the events will be queued and
3059              * eventually handled by the guest after boot
3060              */
3061             error_setg(errp, "Memory hot unplug not supported for this guest");
3062         }
3063     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3064         if (!mc->has_hotpluggable_cpus) {
3065             error_setg(errp, "CPU hot unplug not supported on this machine");
3066             return;
3067         }
3068         spapr_core_unplug_request(hotplug_dev, dev, errp);
3069     }
3070 }
3071 
3072 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3073                                           DeviceState *dev, Error **errp)
3074 {
3075     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3076         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3077     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3078         spapr_core_pre_plug(hotplug_dev, dev, errp);
3079     }
3080 }
3081 
3082 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3083                                                  DeviceState *dev)
3084 {
3085     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3086         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3087         return HOTPLUG_HANDLER(machine);
3088     }
3089     return NULL;
3090 }
3091 
3092 static CpuInstanceProperties
3093 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3094 {
3095     CPUArchId *core_slot;
3096     MachineClass *mc = MACHINE_GET_CLASS(machine);
3097 
3098     /* make sure possible_cpu are intialized */
3099     mc->possible_cpu_arch_ids(machine);
3100     /* get CPU core slot containing thread that matches cpu_index */
3101     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3102     assert(core_slot);
3103     return core_slot->props;
3104 }
3105 
3106 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3107 {
3108     int i;
3109     int spapr_max_cores = max_cpus / smp_threads;
3110     MachineClass *mc = MACHINE_GET_CLASS(machine);
3111 
3112     if (!mc->has_hotpluggable_cpus) {
3113         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3114     }
3115     if (machine->possible_cpus) {
3116         assert(machine->possible_cpus->len == spapr_max_cores);
3117         return machine->possible_cpus;
3118     }
3119 
3120     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3121                              sizeof(CPUArchId) * spapr_max_cores);
3122     machine->possible_cpus->len = spapr_max_cores;
3123     for (i = 0; i < machine->possible_cpus->len; i++) {
3124         int core_id = i * smp_threads;
3125 
3126         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3127         machine->possible_cpus->cpus[i].arch_id = core_id;
3128         machine->possible_cpus->cpus[i].props.has_core_id = true;
3129         machine->possible_cpus->cpus[i].props.core_id = core_id;
3130 
3131         /* default distribution of CPUs over NUMA nodes */
3132         if (nb_numa_nodes) {
3133             /* preset values but do not enable them i.e. 'has_node_id = false',
3134              * numa init code will enable them later if manual mapping wasn't
3135              * present on CLI */
3136             machine->possible_cpus->cpus[i].props.node_id =
3137                 core_id / smp_threads / smp_cores % nb_numa_nodes;
3138         }
3139     }
3140     return machine->possible_cpus;
3141 }
3142 
3143 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3144                                 uint64_t *buid, hwaddr *pio,
3145                                 hwaddr *mmio32, hwaddr *mmio64,
3146                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3147 {
3148     /*
3149      * New-style PHB window placement.
3150      *
3151      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3152      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3153      * windows.
3154      *
3155      * Some guest kernels can't work with MMIO windows above 1<<46
3156      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3157      *
3158      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3159      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3160      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3161      * 1TiB 64-bit MMIO windows for each PHB.
3162      */
3163     const uint64_t base_buid = 0x800000020000000ULL;
3164 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3165                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3166     int i;
3167 
3168     /* Sanity check natural alignments */
3169     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3170     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3171     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3172     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3173     /* Sanity check bounds */
3174     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3175                       SPAPR_PCI_MEM32_WIN_SIZE);
3176     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3177                       SPAPR_PCI_MEM64_WIN_SIZE);
3178 
3179     if (index >= SPAPR_MAX_PHBS) {
3180         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3181                    SPAPR_MAX_PHBS - 1);
3182         return;
3183     }
3184 
3185     *buid = base_buid + index;
3186     for (i = 0; i < n_dma; ++i) {
3187         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3188     }
3189 
3190     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3191     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3192     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3193 }
3194 
3195 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3196 {
3197     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3198 
3199     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3200 }
3201 
3202 static void spapr_ics_resend(XICSFabric *dev)
3203 {
3204     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3205 
3206     ics_resend(spapr->ics);
3207 }
3208 
3209 static ICPState *spapr_icp_get(XICSFabric *xi, int cpu_dt_id)
3210 {
3211     PowerPCCPU *cpu = ppc_get_vcpu_by_dt_id(cpu_dt_id);
3212 
3213     return cpu ? ICP(cpu->intc) : NULL;
3214 }
3215 
3216 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3217                                  Monitor *mon)
3218 {
3219     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3220     CPUState *cs;
3221 
3222     CPU_FOREACH(cs) {
3223         PowerPCCPU *cpu = POWERPC_CPU(cs);
3224 
3225         icp_pic_print_info(ICP(cpu->intc), mon);
3226     }
3227 
3228     ics_pic_print_info(spapr->ics, mon);
3229 }
3230 
3231 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3232 {
3233     MachineClass *mc = MACHINE_CLASS(oc);
3234     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3235     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3236     NMIClass *nc = NMI_CLASS(oc);
3237     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3238     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3239     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3240     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3241 
3242     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3243 
3244     /*
3245      * We set up the default / latest behaviour here.  The class_init
3246      * functions for the specific versioned machine types can override
3247      * these details for backwards compatibility
3248      */
3249     mc->init = ppc_spapr_init;
3250     mc->reset = ppc_spapr_reset;
3251     mc->block_default_type = IF_SCSI;
3252     mc->max_cpus = 1024;
3253     mc->no_parallel = 1;
3254     mc->default_boot_order = "";
3255     mc->default_ram_size = 512 * M_BYTE;
3256     mc->kvm_type = spapr_kvm_type;
3257     mc->has_dynamic_sysbus = true;
3258     mc->pci_allow_0_address = true;
3259     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3260     hc->pre_plug = spapr_machine_device_pre_plug;
3261     hc->plug = spapr_machine_device_plug;
3262     hc->unplug = spapr_machine_device_unplug;
3263     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3264     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3265     hc->unplug_request = spapr_machine_device_unplug_request;
3266 
3267     smc->dr_lmb_enabled = true;
3268     smc->tcg_default_cpu = "POWER8";
3269     mc->has_hotpluggable_cpus = true;
3270     fwc->get_dev_path = spapr_get_fw_dev_path;
3271     nc->nmi_monitor_handler = spapr_nmi;
3272     smc->phb_placement = spapr_phb_placement;
3273     vhc->hypercall = emulate_spapr_hypercall;
3274     vhc->hpt_mask = spapr_hpt_mask;
3275     vhc->map_hptes = spapr_map_hptes;
3276     vhc->unmap_hptes = spapr_unmap_hptes;
3277     vhc->store_hpte = spapr_store_hpte;
3278     vhc->get_patbe = spapr_get_patbe;
3279     xic->ics_get = spapr_ics_get;
3280     xic->ics_resend = spapr_ics_resend;
3281     xic->icp_get = spapr_icp_get;
3282     ispc->print_info = spapr_pic_print_info;
3283     /* Force NUMA node memory size to be a multiple of
3284      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3285      * in which LMBs are represented and hot-added
3286      */
3287     mc->numa_mem_align_shift = 28;
3288 }
3289 
3290 static const TypeInfo spapr_machine_info = {
3291     .name          = TYPE_SPAPR_MACHINE,
3292     .parent        = TYPE_MACHINE,
3293     .abstract      = true,
3294     .instance_size = sizeof(sPAPRMachineState),
3295     .instance_init = spapr_machine_initfn,
3296     .instance_finalize = spapr_machine_finalizefn,
3297     .class_size    = sizeof(sPAPRMachineClass),
3298     .class_init    = spapr_machine_class_init,
3299     .interfaces = (InterfaceInfo[]) {
3300         { TYPE_FW_PATH_PROVIDER },
3301         { TYPE_NMI },
3302         { TYPE_HOTPLUG_HANDLER },
3303         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3304         { TYPE_XICS_FABRIC },
3305         { TYPE_INTERRUPT_STATS_PROVIDER },
3306         { }
3307     },
3308 };
3309 
3310 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3311     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3312                                                     void *data)      \
3313     {                                                                \
3314         MachineClass *mc = MACHINE_CLASS(oc);                        \
3315         spapr_machine_##suffix##_class_options(mc);                  \
3316         if (latest) {                                                \
3317             mc->alias = "pseries";                                   \
3318             mc->is_default = 1;                                      \
3319         }                                                            \
3320     }                                                                \
3321     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3322     {                                                                \
3323         MachineState *machine = MACHINE(obj);                        \
3324         spapr_machine_##suffix##_instance_options(machine);          \
3325     }                                                                \
3326     static const TypeInfo spapr_machine_##suffix##_info = {          \
3327         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3328         .parent = TYPE_SPAPR_MACHINE,                                \
3329         .class_init = spapr_machine_##suffix##_class_init,           \
3330         .instance_init = spapr_machine_##suffix##_instance_init,     \
3331     };                                                               \
3332     static void spapr_machine_register_##suffix(void)                \
3333     {                                                                \
3334         type_register(&spapr_machine_##suffix##_info);               \
3335     }                                                                \
3336     type_init(spapr_machine_register_##suffix)
3337 
3338 /*
3339  * pseries-2.10
3340  */
3341 static void spapr_machine_2_10_instance_options(MachineState *machine)
3342 {
3343 }
3344 
3345 static void spapr_machine_2_10_class_options(MachineClass *mc)
3346 {
3347     /* Defaults for the latest behaviour inherited from the base class */
3348 }
3349 
3350 DEFINE_SPAPR_MACHINE(2_10, "2.10", true);
3351 
3352 /*
3353  * pseries-2.9
3354  */
3355 #define SPAPR_COMPAT_2_9                                               \
3356     HW_COMPAT_2_9
3357 
3358 static void spapr_machine_2_9_instance_options(MachineState *machine)
3359 {
3360     spapr_machine_2_10_instance_options(machine);
3361 }
3362 
3363 static void spapr_machine_2_9_class_options(MachineClass *mc)
3364 {
3365     spapr_machine_2_10_class_options(mc);
3366     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3367     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
3368 }
3369 
3370 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3371 
3372 /*
3373  * pseries-2.8
3374  */
3375 #define SPAPR_COMPAT_2_8                                        \
3376     HW_COMPAT_2_8                                               \
3377     {                                                           \
3378         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3379         .property = "pcie-extended-configuration-space",        \
3380         .value    = "off",                                      \
3381     },
3382 
3383 static void spapr_machine_2_8_instance_options(MachineState *machine)
3384 {
3385     spapr_machine_2_9_instance_options(machine);
3386 }
3387 
3388 static void spapr_machine_2_8_class_options(MachineClass *mc)
3389 {
3390     spapr_machine_2_9_class_options(mc);
3391     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3392     mc->numa_mem_align_shift = 23;
3393 }
3394 
3395 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3396 
3397 /*
3398  * pseries-2.7
3399  */
3400 #define SPAPR_COMPAT_2_7                            \
3401     HW_COMPAT_2_7                                   \
3402     {                                               \
3403         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3404         .property = "mem_win_size",                 \
3405         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3406     },                                              \
3407     {                                               \
3408         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3409         .property = "mem64_win_size",               \
3410         .value    = "0",                            \
3411     },                                              \
3412     {                                               \
3413         .driver = TYPE_POWERPC_CPU,                 \
3414         .property = "pre-2.8-migration",            \
3415         .value    = "on",                           \
3416     },                                              \
3417     {                                               \
3418         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3419         .property = "pre-2.8-migration",            \
3420         .value    = "on",                           \
3421     },
3422 
3423 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3424                               uint64_t *buid, hwaddr *pio,
3425                               hwaddr *mmio32, hwaddr *mmio64,
3426                               unsigned n_dma, uint32_t *liobns, Error **errp)
3427 {
3428     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3429     const uint64_t base_buid = 0x800000020000000ULL;
3430     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3431     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3432     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3433     const uint32_t max_index = 255;
3434     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3435 
3436     uint64_t ram_top = MACHINE(spapr)->ram_size;
3437     hwaddr phb0_base, phb_base;
3438     int i;
3439 
3440     /* Do we have hotpluggable memory? */
3441     if (MACHINE(spapr)->maxram_size > ram_top) {
3442         /* Can't just use maxram_size, because there may be an
3443          * alignment gap between normal and hotpluggable memory
3444          * regions */
3445         ram_top = spapr->hotplug_memory.base +
3446             memory_region_size(&spapr->hotplug_memory.mr);
3447     }
3448 
3449     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3450 
3451     if (index > max_index) {
3452         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3453                    max_index);
3454         return;
3455     }
3456 
3457     *buid = base_buid + index;
3458     for (i = 0; i < n_dma; ++i) {
3459         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3460     }
3461 
3462     phb_base = phb0_base + index * phb_spacing;
3463     *pio = phb_base + pio_offset;
3464     *mmio32 = phb_base + mmio_offset;
3465     /*
3466      * We don't set the 64-bit MMIO window, relying on the PHB's
3467      * fallback behaviour of automatically splitting a large "32-bit"
3468      * window into contiguous 32-bit and 64-bit windows
3469      */
3470 }
3471 
3472 static void spapr_machine_2_7_instance_options(MachineState *machine)
3473 {
3474     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3475 
3476     spapr_machine_2_8_instance_options(machine);
3477     spapr->use_hotplug_event_source = false;
3478 }
3479 
3480 static void spapr_machine_2_7_class_options(MachineClass *mc)
3481 {
3482     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3483 
3484     spapr_machine_2_8_class_options(mc);
3485     smc->tcg_default_cpu = "POWER7";
3486     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3487     smc->phb_placement = phb_placement_2_7;
3488 }
3489 
3490 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3491 
3492 /*
3493  * pseries-2.6
3494  */
3495 #define SPAPR_COMPAT_2_6 \
3496     HW_COMPAT_2_6 \
3497     { \
3498         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3499         .property = "ddw",\
3500         .value    = stringify(off),\
3501     },
3502 
3503 static void spapr_machine_2_6_instance_options(MachineState *machine)
3504 {
3505     spapr_machine_2_7_instance_options(machine);
3506 }
3507 
3508 static void spapr_machine_2_6_class_options(MachineClass *mc)
3509 {
3510     spapr_machine_2_7_class_options(mc);
3511     mc->has_hotpluggable_cpus = false;
3512     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3513 }
3514 
3515 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3516 
3517 /*
3518  * pseries-2.5
3519  */
3520 #define SPAPR_COMPAT_2_5 \
3521     HW_COMPAT_2_5 \
3522     { \
3523         .driver   = "spapr-vlan", \
3524         .property = "use-rx-buffer-pools", \
3525         .value    = "off", \
3526     },
3527 
3528 static void spapr_machine_2_5_instance_options(MachineState *machine)
3529 {
3530     spapr_machine_2_6_instance_options(machine);
3531 }
3532 
3533 static void spapr_machine_2_5_class_options(MachineClass *mc)
3534 {
3535     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3536 
3537     spapr_machine_2_6_class_options(mc);
3538     smc->use_ohci_by_default = true;
3539     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3540 }
3541 
3542 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3543 
3544 /*
3545  * pseries-2.4
3546  */
3547 #define SPAPR_COMPAT_2_4 \
3548         HW_COMPAT_2_4
3549 
3550 static void spapr_machine_2_4_instance_options(MachineState *machine)
3551 {
3552     spapr_machine_2_5_instance_options(machine);
3553 }
3554 
3555 static void spapr_machine_2_4_class_options(MachineClass *mc)
3556 {
3557     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3558 
3559     spapr_machine_2_5_class_options(mc);
3560     smc->dr_lmb_enabled = false;
3561     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3562 }
3563 
3564 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3565 
3566 /*
3567  * pseries-2.3
3568  */
3569 #define SPAPR_COMPAT_2_3 \
3570         HW_COMPAT_2_3 \
3571         {\
3572             .driver   = "spapr-pci-host-bridge",\
3573             .property = "dynamic-reconfiguration",\
3574             .value    = "off",\
3575         },
3576 
3577 static void spapr_machine_2_3_instance_options(MachineState *machine)
3578 {
3579     spapr_machine_2_4_instance_options(machine);
3580     savevm_skip_section_footers();
3581     global_state_set_optional();
3582     savevm_skip_configuration();
3583 }
3584 
3585 static void spapr_machine_2_3_class_options(MachineClass *mc)
3586 {
3587     spapr_machine_2_4_class_options(mc);
3588     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3589 }
3590 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3591 
3592 /*
3593  * pseries-2.2
3594  */
3595 
3596 #define SPAPR_COMPAT_2_2 \
3597         HW_COMPAT_2_2 \
3598         {\
3599             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3600             .property = "mem_win_size",\
3601             .value    = "0x20000000",\
3602         },
3603 
3604 static void spapr_machine_2_2_instance_options(MachineState *machine)
3605 {
3606     spapr_machine_2_3_instance_options(machine);
3607     machine->suppress_vmdesc = true;
3608 }
3609 
3610 static void spapr_machine_2_2_class_options(MachineClass *mc)
3611 {
3612     spapr_machine_2_3_class_options(mc);
3613     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3614 }
3615 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3616 
3617 /*
3618  * pseries-2.1
3619  */
3620 #define SPAPR_COMPAT_2_1 \
3621         HW_COMPAT_2_1
3622 
3623 static void spapr_machine_2_1_instance_options(MachineState *machine)
3624 {
3625     spapr_machine_2_2_instance_options(machine);
3626 }
3627 
3628 static void spapr_machine_2_1_class_options(MachineClass *mc)
3629 {
3630     spapr_machine_2_2_class_options(mc);
3631     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3632 }
3633 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3634 
3635 static void spapr_machine_register_types(void)
3636 {
3637     type_register_static(&spapr_machine_info);
3638 }
3639 
3640 type_init(spapr_machine_register_types)
3641