xref: /openbmc/qemu/hw/ppc/spapr.c (revision f664b88247487c4cb020d016bef0f3b1daf9f4e5)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
31 #include "hw/hw.h"
32 #include "qemu/log.h"
33 #include "hw/fw-path-provider.h"
34 #include "elf.h"
35 #include "net/net.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/migration.h"
42 #include "mmu-hash64.h"
43 #include "mmu-book3s-v3.h"
44 #include "qom/cpu.h"
45 
46 #include "hw/boards.h"
47 #include "hw/ppc/ppc.h"
48 #include "hw/loader.h"
49 
50 #include "hw/ppc/fdt.h"
51 #include "hw/ppc/spapr.h"
52 #include "hw/ppc/spapr_vio.h"
53 #include "hw/pci-host/spapr.h"
54 #include "hw/ppc/xics.h"
55 #include "hw/pci/msi.h"
56 
57 #include "hw/pci/pci.h"
58 #include "hw/scsi/scsi.h"
59 #include "hw/virtio/virtio-scsi.h"
60 
61 #include "exec/address-spaces.h"
62 #include "hw/usb.h"
63 #include "qemu/config-file.h"
64 #include "qemu/error-report.h"
65 #include "trace.h"
66 #include "hw/nmi.h"
67 #include "hw/intc/intc.h"
68 
69 #include "hw/compat.h"
70 #include "qemu/cutils.h"
71 #include "hw/ppc/spapr_cpu_core.h"
72 #include "qmp-commands.h"
73 
74 #include <libfdt.h>
75 
76 /* SLOF memory layout:
77  *
78  * SLOF raw image loaded at 0, copies its romfs right below the flat
79  * device-tree, then position SLOF itself 31M below that
80  *
81  * So we set FW_OVERHEAD to 40MB which should account for all of that
82  * and more
83  *
84  * We load our kernel at 4M, leaving space for SLOF initial image
85  */
86 #define FDT_MAX_SIZE            0x100000
87 #define RTAS_MAX_SIZE           0x10000
88 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
89 #define FW_MAX_SIZE             0x400000
90 #define FW_FILE_NAME            "slof.bin"
91 #define FW_OVERHEAD             0x2800000
92 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
93 
94 #define MIN_RMA_SLOF            128UL
95 
96 #define PHANDLE_XICP            0x00001111
97 
98 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
99 
100 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
101                                   const char *type_ics,
102                                   int nr_irqs, Error **errp)
103 {
104     Error *local_err = NULL;
105     Object *obj;
106 
107     obj = object_new(type_ics);
108     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
109     object_property_add_const_link(obj, "xics", OBJECT(spapr), &error_abort);
110     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
111     if (local_err) {
112         goto error;
113     }
114     object_property_set_bool(obj, true, "realized", &local_err);
115     if (local_err) {
116         goto error;
117     }
118 
119     return ICS_SIMPLE(obj);
120 
121 error:
122     error_propagate(errp, local_err);
123     return NULL;
124 }
125 
126 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
127 {
128     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
129 
130     if (kvm_enabled()) {
131         if (machine_kernel_irqchip_allowed(machine) &&
132             !xics_kvm_init(spapr, errp)) {
133             spapr->icp_type = TYPE_KVM_ICP;
134             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
135         }
136         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
137             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
138             return;
139         }
140     }
141 
142     if (!spapr->ics) {
143         xics_spapr_init(spapr);
144         spapr->icp_type = TYPE_ICP;
145         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
146         if (!spapr->ics) {
147             return;
148         }
149     }
150 }
151 
152 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
153                                   int smt_threads)
154 {
155     int i, ret = 0;
156     uint32_t servers_prop[smt_threads];
157     uint32_t gservers_prop[smt_threads * 2];
158     int index = ppc_get_vcpu_dt_id(cpu);
159 
160     if (cpu->compat_pvr) {
161         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
162         if (ret < 0) {
163             return ret;
164         }
165     }
166 
167     /* Build interrupt servers and gservers properties */
168     for (i = 0; i < smt_threads; i++) {
169         servers_prop[i] = cpu_to_be32(index + i);
170         /* Hack, direct the group queues back to cpu 0 */
171         gservers_prop[i*2] = cpu_to_be32(index + i);
172         gservers_prop[i*2 + 1] = 0;
173     }
174     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
175                       servers_prop, sizeof(servers_prop));
176     if (ret < 0) {
177         return ret;
178     }
179     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
180                       gservers_prop, sizeof(gservers_prop));
181 
182     return ret;
183 }
184 
185 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
186 {
187     int ret = 0;
188     PowerPCCPU *cpu = POWERPC_CPU(cs);
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(cs->numa_node),
195                                 cpu_to_be32(index)};
196 
197     /* Advertise NUMA via ibm,associativity */
198     if (nb_numa_nodes > 1) {
199         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
200                           sizeof(associativity));
201     }
202 
203     return ret;
204 }
205 
206 /* Populate the "ibm,pa-features" property */
207 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset,
208                                       bool legacy_guest)
209 {
210     uint8_t pa_features_206[] = { 6, 0,
211         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
212     uint8_t pa_features_207[] = { 24, 0,
213         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
214         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
215         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
216         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
217     uint8_t pa_features_300[] = { 66, 0,
218         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
219         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
220         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
221         /* 6: DS207 */
222         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
223         /* 16: Vector */
224         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
225         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
226         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
227         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
228         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
229         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
230         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
231         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
232         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
233         /* 42: PM, 44: PC RA, 46: SC vec'd */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
235         /* 48: SIMD, 50: QP BFP, 52: String */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
237         /* 54: DecFP, 56: DecI, 58: SHA */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
239         /* 60: NM atomic, 62: RNG */
240         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
241     };
242     uint8_t *pa_features;
243     size_t pa_size;
244 
245     switch (POWERPC_MMU_VER(env->mmu_model)) {
246     case POWERPC_MMU_VER_2_06:
247         pa_features = pa_features_206;
248         pa_size = sizeof(pa_features_206);
249         break;
250     case POWERPC_MMU_VER_2_07:
251         pa_features = pa_features_207;
252         pa_size = sizeof(pa_features_207);
253         break;
254     case POWERPC_MMU_VER_3_00:
255         pa_features = pa_features_300;
256         pa_size = sizeof(pa_features_300);
257         break;
258     default:
259         return;
260     }
261 
262     if (env->ci_large_pages) {
263         /*
264          * Note: we keep CI large pages off by default because a 64K capable
265          * guest provisioned with large pages might otherwise try to map a qemu
266          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
267          * even if that qemu runs on a 4k host.
268          * We dd this bit back here if we are confident this is not an issue
269          */
270         pa_features[3] |= 0x20;
271     }
272     if (kvmppc_has_cap_htm() && pa_size > 24) {
273         pa_features[24] |= 0x80;    /* Transactional memory support */
274     }
275     if (legacy_guest && pa_size > 40) {
276         /* Workaround for broken kernels that attempt (guest) radix
277          * mode when they can't handle it, if they see the radix bit set
278          * in pa-features. So hide it from them. */
279         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
280     }
281 
282     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
283 }
284 
285 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
286 {
287     int ret = 0, offset, cpus_offset;
288     CPUState *cs;
289     char cpu_model[32];
290     int smt = kvmppc_smt_threads();
291     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
292 
293     CPU_FOREACH(cs) {
294         PowerPCCPU *cpu = POWERPC_CPU(cs);
295         CPUPPCState *env = &cpu->env;
296         DeviceClass *dc = DEVICE_GET_CLASS(cs);
297         int index = ppc_get_vcpu_dt_id(cpu);
298         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
299 
300         if ((index % smt) != 0) {
301             continue;
302         }
303 
304         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
305 
306         cpus_offset = fdt_path_offset(fdt, "/cpus");
307         if (cpus_offset < 0) {
308             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
309                                           "cpus");
310             if (cpus_offset < 0) {
311                 return cpus_offset;
312             }
313         }
314         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
315         if (offset < 0) {
316             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
317             if (offset < 0) {
318                 return offset;
319             }
320         }
321 
322         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
323                           pft_size_prop, sizeof(pft_size_prop));
324         if (ret < 0) {
325             return ret;
326         }
327 
328         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
329         if (ret < 0) {
330             return ret;
331         }
332 
333         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
334         if (ret < 0) {
335             return ret;
336         }
337 
338         spapr_populate_pa_features(env, fdt, offset,
339                                          spapr->cas_legacy_guest_workaround);
340     }
341     return ret;
342 }
343 
344 static hwaddr spapr_node0_size(void)
345 {
346     MachineState *machine = MACHINE(qdev_get_machine());
347 
348     if (nb_numa_nodes) {
349         int i;
350         for (i = 0; i < nb_numa_nodes; ++i) {
351             if (numa_info[i].node_mem) {
352                 return MIN(pow2floor(numa_info[i].node_mem),
353                            machine->ram_size);
354             }
355         }
356     }
357     return machine->ram_size;
358 }
359 
360 static void add_str(GString *s, const gchar *s1)
361 {
362     g_string_append_len(s, s1, strlen(s1) + 1);
363 }
364 
365 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
366                                        hwaddr size)
367 {
368     uint32_t associativity[] = {
369         cpu_to_be32(0x4), /* length */
370         cpu_to_be32(0x0), cpu_to_be32(0x0),
371         cpu_to_be32(0x0), cpu_to_be32(nodeid)
372     };
373     char mem_name[32];
374     uint64_t mem_reg_property[2];
375     int off;
376 
377     mem_reg_property[0] = cpu_to_be64(start);
378     mem_reg_property[1] = cpu_to_be64(size);
379 
380     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
381     off = fdt_add_subnode(fdt, 0, mem_name);
382     _FDT(off);
383     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
384     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
385                       sizeof(mem_reg_property))));
386     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
387                       sizeof(associativity))));
388     return off;
389 }
390 
391 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
392 {
393     MachineState *machine = MACHINE(spapr);
394     hwaddr mem_start, node_size;
395     int i, nb_nodes = nb_numa_nodes;
396     NodeInfo *nodes = numa_info;
397     NodeInfo ramnode;
398 
399     /* No NUMA nodes, assume there is just one node with whole RAM */
400     if (!nb_numa_nodes) {
401         nb_nodes = 1;
402         ramnode.node_mem = machine->ram_size;
403         nodes = &ramnode;
404     }
405 
406     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
407         if (!nodes[i].node_mem) {
408             continue;
409         }
410         if (mem_start >= machine->ram_size) {
411             node_size = 0;
412         } else {
413             node_size = nodes[i].node_mem;
414             if (node_size > machine->ram_size - mem_start) {
415                 node_size = machine->ram_size - mem_start;
416             }
417         }
418         if (!mem_start) {
419             /* ppc_spapr_init() checks for rma_size <= node0_size already */
420             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
421             mem_start += spapr->rma_size;
422             node_size -= spapr->rma_size;
423         }
424         for ( ; node_size; ) {
425             hwaddr sizetmp = pow2floor(node_size);
426 
427             /* mem_start != 0 here */
428             if (ctzl(mem_start) < ctzl(sizetmp)) {
429                 sizetmp = 1ULL << ctzl(mem_start);
430             }
431 
432             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
433             node_size -= sizetmp;
434             mem_start += sizetmp;
435         }
436     }
437 
438     return 0;
439 }
440 
441 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
442                                   sPAPRMachineState *spapr)
443 {
444     PowerPCCPU *cpu = POWERPC_CPU(cs);
445     CPUPPCState *env = &cpu->env;
446     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
447     int index = ppc_get_vcpu_dt_id(cpu);
448     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
449                        0xffffffff, 0xffffffff};
450     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
451         : SPAPR_TIMEBASE_FREQ;
452     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
453     uint32_t page_sizes_prop[64];
454     size_t page_sizes_prop_size;
455     uint32_t vcpus_per_socket = smp_threads * smp_cores;
456     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
457     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
458     sPAPRDRConnector *drc;
459     sPAPRDRConnectorClass *drck;
460     int drc_index;
461     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
462     int i;
463 
464     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
465     if (drc) {
466         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
467         drc_index = drck->get_index(drc);
468         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
469     }
470 
471     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
472     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
473 
474     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
475     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
476                            env->dcache_line_size)));
477     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
478                            env->dcache_line_size)));
479     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
480                            env->icache_line_size)));
481     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
482                            env->icache_line_size)));
483 
484     if (pcc->l1_dcache_size) {
485         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
486                                pcc->l1_dcache_size)));
487     } else {
488         error_report("Warning: Unknown L1 dcache size for cpu");
489     }
490     if (pcc->l1_icache_size) {
491         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
492                                pcc->l1_icache_size)));
493     } else {
494         error_report("Warning: Unknown L1 icache size for cpu");
495     }
496 
497     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
498     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
499     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
500     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
501     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
502     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
503 
504     if (env->spr_cb[SPR_PURR].oea_read) {
505         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
506     }
507 
508     if (env->mmu_model & POWERPC_MMU_1TSEG) {
509         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
510                           segs, sizeof(segs))));
511     }
512 
513     /* Advertise VMX/VSX (vector extensions) if available
514      *   0 / no property == no vector extensions
515      *   1               == VMX / Altivec available
516      *   2               == VSX available */
517     if (env->insns_flags & PPC_ALTIVEC) {
518         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
519 
520         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
521     }
522 
523     /* Advertise DFP (Decimal Floating Point) if available
524      *   0 / no property == no DFP
525      *   1               == DFP available */
526     if (env->insns_flags2 & PPC2_DFP) {
527         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
528     }
529 
530     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
531                                                   sizeof(page_sizes_prop));
532     if (page_sizes_prop_size) {
533         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
534                           page_sizes_prop, page_sizes_prop_size)));
535     }
536 
537     spapr_populate_pa_features(env, fdt, offset, false);
538 
539     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
540                            cs->cpu_index / vcpus_per_socket)));
541 
542     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
543                       pft_size_prop, sizeof(pft_size_prop))));
544 
545     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
546 
547     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
548 
549     if (pcc->radix_page_info) {
550         for (i = 0; i < pcc->radix_page_info->count; i++) {
551             radix_AP_encodings[i] =
552                 cpu_to_be32(pcc->radix_page_info->entries[i]);
553         }
554         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
555                           radix_AP_encodings,
556                           pcc->radix_page_info->count *
557                           sizeof(radix_AP_encodings[0]))));
558     }
559 }
560 
561 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
562 {
563     CPUState *cs;
564     int cpus_offset;
565     char *nodename;
566     int smt = kvmppc_smt_threads();
567 
568     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
569     _FDT(cpus_offset);
570     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
571     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
572 
573     /*
574      * We walk the CPUs in reverse order to ensure that CPU DT nodes
575      * created by fdt_add_subnode() end up in the right order in FDT
576      * for the guest kernel the enumerate the CPUs correctly.
577      */
578     CPU_FOREACH_REVERSE(cs) {
579         PowerPCCPU *cpu = POWERPC_CPU(cs);
580         int index = ppc_get_vcpu_dt_id(cpu);
581         DeviceClass *dc = DEVICE_GET_CLASS(cs);
582         int offset;
583 
584         if ((index % smt) != 0) {
585             continue;
586         }
587 
588         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
589         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
590         g_free(nodename);
591         _FDT(offset);
592         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
593     }
594 
595 }
596 
597 /*
598  * Adds ibm,dynamic-reconfiguration-memory node.
599  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
600  * of this device tree node.
601  */
602 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
603 {
604     MachineState *machine = MACHINE(spapr);
605     int ret, i, offset;
606     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
607     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
608     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
609     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
610                        memory_region_size(&spapr->hotplug_memory.mr)) /
611                        lmb_size;
612     uint32_t *int_buf, *cur_index, buf_len;
613     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
614 
615     /*
616      * Don't create the node if there is no hotpluggable memory
617      */
618     if (machine->ram_size == machine->maxram_size) {
619         return 0;
620     }
621 
622     /*
623      * Allocate enough buffer size to fit in ibm,dynamic-memory
624      * or ibm,associativity-lookup-arrays
625      */
626     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
627               * sizeof(uint32_t);
628     cur_index = int_buf = g_malloc0(buf_len);
629 
630     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
631 
632     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
633                     sizeof(prop_lmb_size));
634     if (ret < 0) {
635         goto out;
636     }
637 
638     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
639     if (ret < 0) {
640         goto out;
641     }
642 
643     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
644     if (ret < 0) {
645         goto out;
646     }
647 
648     /* ibm,dynamic-memory */
649     int_buf[0] = cpu_to_be32(nr_lmbs);
650     cur_index++;
651     for (i = 0; i < nr_lmbs; i++) {
652         uint64_t addr = i * lmb_size;
653         uint32_t *dynamic_memory = cur_index;
654 
655         if (i >= hotplug_lmb_start) {
656             sPAPRDRConnector *drc;
657             sPAPRDRConnectorClass *drck;
658 
659             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
660             g_assert(drc);
661             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
662 
663             dynamic_memory[0] = cpu_to_be32(addr >> 32);
664             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
665             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
666             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
667             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
668             if (memory_region_present(get_system_memory(), addr)) {
669                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
670             } else {
671                 dynamic_memory[5] = cpu_to_be32(0);
672             }
673         } else {
674             /*
675              * LMB information for RMA, boot time RAM and gap b/n RAM and
676              * hotplug memory region -- all these are marked as reserved
677              * and as having no valid DRC.
678              */
679             dynamic_memory[0] = cpu_to_be32(addr >> 32);
680             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
681             dynamic_memory[2] = cpu_to_be32(0);
682             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
683             dynamic_memory[4] = cpu_to_be32(-1);
684             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
685                                             SPAPR_LMB_FLAGS_DRC_INVALID);
686         }
687 
688         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
689     }
690     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
691     if (ret < 0) {
692         goto out;
693     }
694 
695     /* ibm,associativity-lookup-arrays */
696     cur_index = int_buf;
697     int_buf[0] = cpu_to_be32(nr_nodes);
698     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
699     cur_index += 2;
700     for (i = 0; i < nr_nodes; i++) {
701         uint32_t associativity[] = {
702             cpu_to_be32(0x0),
703             cpu_to_be32(0x0),
704             cpu_to_be32(0x0),
705             cpu_to_be32(i)
706         };
707         memcpy(cur_index, associativity, sizeof(associativity));
708         cur_index += 4;
709     }
710     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
711             (cur_index - int_buf) * sizeof(uint32_t));
712 out:
713     g_free(int_buf);
714     return ret;
715 }
716 
717 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
718                                 sPAPROptionVector *ov5_updates)
719 {
720     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
721     int ret = 0, offset;
722 
723     /* Generate ibm,dynamic-reconfiguration-memory node if required */
724     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
725         g_assert(smc->dr_lmb_enabled);
726         ret = spapr_populate_drconf_memory(spapr, fdt);
727         if (ret) {
728             goto out;
729         }
730     }
731 
732     offset = fdt_path_offset(fdt, "/chosen");
733     if (offset < 0) {
734         offset = fdt_add_subnode(fdt, 0, "chosen");
735         if (offset < 0) {
736             return offset;
737         }
738     }
739     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
740                                  "ibm,architecture-vec-5");
741 
742 out:
743     return ret;
744 }
745 
746 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
747                                  target_ulong addr, target_ulong size,
748                                  sPAPROptionVector *ov5_updates)
749 {
750     void *fdt, *fdt_skel;
751     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
752 
753     size -= sizeof(hdr);
754 
755     /* Create sceleton */
756     fdt_skel = g_malloc0(size);
757     _FDT((fdt_create(fdt_skel, size)));
758     _FDT((fdt_begin_node(fdt_skel, "")));
759     _FDT((fdt_end_node(fdt_skel)));
760     _FDT((fdt_finish(fdt_skel)));
761     fdt = g_malloc0(size);
762     _FDT((fdt_open_into(fdt_skel, fdt, size)));
763     g_free(fdt_skel);
764 
765     /* Fixup cpu nodes */
766     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
767 
768     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
769         return -1;
770     }
771 
772     /* Pack resulting tree */
773     _FDT((fdt_pack(fdt)));
774 
775     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
776         trace_spapr_cas_failed(size);
777         return -1;
778     }
779 
780     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
781     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
782     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
783     g_free(fdt);
784 
785     return 0;
786 }
787 
788 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
789 {
790     int rtas;
791     GString *hypertas = g_string_sized_new(256);
792     GString *qemu_hypertas = g_string_sized_new(256);
793     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
794     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
795         memory_region_size(&spapr->hotplug_memory.mr);
796     uint32_t lrdr_capacity[] = {
797         cpu_to_be32(max_hotplug_addr >> 32),
798         cpu_to_be32(max_hotplug_addr & 0xffffffff),
799         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
800         cpu_to_be32(max_cpus / smp_threads),
801     };
802 
803     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
804 
805     /* hypertas */
806     add_str(hypertas, "hcall-pft");
807     add_str(hypertas, "hcall-term");
808     add_str(hypertas, "hcall-dabr");
809     add_str(hypertas, "hcall-interrupt");
810     add_str(hypertas, "hcall-tce");
811     add_str(hypertas, "hcall-vio");
812     add_str(hypertas, "hcall-splpar");
813     add_str(hypertas, "hcall-bulk");
814     add_str(hypertas, "hcall-set-mode");
815     add_str(hypertas, "hcall-sprg0");
816     add_str(hypertas, "hcall-copy");
817     add_str(hypertas, "hcall-debug");
818     add_str(qemu_hypertas, "hcall-memop1");
819 
820     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
821         add_str(hypertas, "hcall-multi-tce");
822     }
823     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
824                      hypertas->str, hypertas->len));
825     g_string_free(hypertas, TRUE);
826     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
827                      qemu_hypertas->str, qemu_hypertas->len));
828     g_string_free(qemu_hypertas, TRUE);
829 
830     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
831                      refpoints, sizeof(refpoints)));
832 
833     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
834                           RTAS_ERROR_LOG_MAX));
835     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
836                           RTAS_EVENT_SCAN_RATE));
837 
838     if (msi_nonbroken) {
839         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
840     }
841 
842     /*
843      * According to PAPR, rtas ibm,os-term does not guarantee a return
844      * back to the guest cpu.
845      *
846      * While an additional ibm,extended-os-term property indicates
847      * that rtas call return will always occur. Set this property.
848      */
849     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
850 
851     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
852                      lrdr_capacity, sizeof(lrdr_capacity)));
853 
854     spapr_dt_rtas_tokens(fdt, rtas);
855 }
856 
857 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
858  * that the guest may request and thus the valid values for bytes 24..26 of
859  * option vector 5: */
860 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
861 {
862     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
863 
864     char val[2 * 3] = {
865         24, 0x00, /* Hash/Radix, filled in below. */
866         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
867         26, 0x40, /* Radix options: GTSE == yes. */
868     };
869 
870     if (kvm_enabled()) {
871         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
872             val[1] = 0x80; /* OV5_MMU_BOTH */
873         } else if (kvmppc_has_cap_mmu_radix()) {
874             val[1] = 0x40; /* OV5_MMU_RADIX_300 */
875         } else {
876             val[1] = 0x00; /* Hash */
877         }
878     } else {
879         if (first_ppc_cpu->env.mmu_model & POWERPC_MMU_V3) {
880             /* V3 MMU supports both hash and radix (with dynamic switching) */
881             val[1] = 0xC0;
882         } else {
883             /* Otherwise we can only do hash */
884             val[1] = 0x00;
885         }
886     }
887     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
888                      val, sizeof(val)));
889 }
890 
891 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
892 {
893     MachineState *machine = MACHINE(spapr);
894     int chosen;
895     const char *boot_device = machine->boot_order;
896     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
897     size_t cb = 0;
898     char *bootlist = get_boot_devices_list(&cb, true);
899 
900     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
901 
902     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
903     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
904                           spapr->initrd_base));
905     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
906                           spapr->initrd_base + spapr->initrd_size));
907 
908     if (spapr->kernel_size) {
909         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
910                               cpu_to_be64(spapr->kernel_size) };
911 
912         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
913                          &kprop, sizeof(kprop)));
914         if (spapr->kernel_le) {
915             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
916         }
917     }
918     if (boot_menu) {
919         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
920     }
921     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
922     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
923     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
924 
925     if (cb && bootlist) {
926         int i;
927 
928         for (i = 0; i < cb; i++) {
929             if (bootlist[i] == '\n') {
930                 bootlist[i] = ' ';
931             }
932         }
933         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
934     }
935 
936     if (boot_device && strlen(boot_device)) {
937         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
938     }
939 
940     if (!spapr->has_graphics && stdout_path) {
941         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
942     }
943 
944     spapr_dt_ov5_platform_support(fdt, chosen);
945 
946     g_free(stdout_path);
947     g_free(bootlist);
948 }
949 
950 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
951 {
952     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
953      * KVM to work under pHyp with some guest co-operation */
954     int hypervisor;
955     uint8_t hypercall[16];
956 
957     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
958     /* indicate KVM hypercall interface */
959     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
960     if (kvmppc_has_cap_fixup_hcalls()) {
961         /*
962          * Older KVM versions with older guest kernels were broken
963          * with the magic page, don't allow the guest to map it.
964          */
965         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
966                                   sizeof(hypercall))) {
967             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
968                              hypercall, sizeof(hypercall)));
969         }
970     }
971 }
972 
973 static void *spapr_build_fdt(sPAPRMachineState *spapr,
974                              hwaddr rtas_addr,
975                              hwaddr rtas_size)
976 {
977     MachineState *machine = MACHINE(qdev_get_machine());
978     MachineClass *mc = MACHINE_GET_CLASS(machine);
979     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
980     int ret;
981     void *fdt;
982     sPAPRPHBState *phb;
983     char *buf;
984     int smt = kvmppc_smt_threads();
985 
986     fdt = g_malloc0(FDT_MAX_SIZE);
987     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
988 
989     /* Root node */
990     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
991     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
992     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
993 
994     /*
995      * Add info to guest to indentify which host is it being run on
996      * and what is the uuid of the guest
997      */
998     if (kvmppc_get_host_model(&buf)) {
999         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1000         g_free(buf);
1001     }
1002     if (kvmppc_get_host_serial(&buf)) {
1003         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1004         g_free(buf);
1005     }
1006 
1007     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1008 
1009     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1010     if (qemu_uuid_set) {
1011         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1012     }
1013     g_free(buf);
1014 
1015     if (qemu_get_vm_name()) {
1016         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1017                                 qemu_get_vm_name()));
1018     }
1019 
1020     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1021     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1022 
1023     /* /interrupt controller */
1024     spapr_dt_xics(DIV_ROUND_UP(max_cpus * smt, smp_threads), fdt, PHANDLE_XICP);
1025 
1026     ret = spapr_populate_memory(spapr, fdt);
1027     if (ret < 0) {
1028         error_report("couldn't setup memory nodes in fdt");
1029         exit(1);
1030     }
1031 
1032     /* /vdevice */
1033     spapr_dt_vdevice(spapr->vio_bus, fdt);
1034 
1035     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1036         ret = spapr_rng_populate_dt(fdt);
1037         if (ret < 0) {
1038             error_report("could not set up rng device in the fdt");
1039             exit(1);
1040         }
1041     }
1042 
1043     QLIST_FOREACH(phb, &spapr->phbs, list) {
1044         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1045         if (ret < 0) {
1046             error_report("couldn't setup PCI devices in fdt");
1047             exit(1);
1048         }
1049     }
1050 
1051     /* cpus */
1052     spapr_populate_cpus_dt_node(fdt, spapr);
1053 
1054     if (smc->dr_lmb_enabled) {
1055         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1056     }
1057 
1058     if (mc->has_hotpluggable_cpus) {
1059         int offset = fdt_path_offset(fdt, "/cpus");
1060         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1061                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1062         if (ret < 0) {
1063             error_report("Couldn't set up CPU DR device tree properties");
1064             exit(1);
1065         }
1066     }
1067 
1068     /* /event-sources */
1069     spapr_dt_events(spapr, fdt);
1070 
1071     /* /rtas */
1072     spapr_dt_rtas(spapr, fdt);
1073 
1074     /* /chosen */
1075     spapr_dt_chosen(spapr, fdt);
1076 
1077     /* /hypervisor */
1078     if (kvm_enabled()) {
1079         spapr_dt_hypervisor(spapr, fdt);
1080     }
1081 
1082     /* Build memory reserve map */
1083     if (spapr->kernel_size) {
1084         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1085     }
1086     if (spapr->initrd_size) {
1087         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1088     }
1089 
1090     /* ibm,client-architecture-support updates */
1091     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1092     if (ret < 0) {
1093         error_report("couldn't setup CAS properties fdt");
1094         exit(1);
1095     }
1096 
1097     return fdt;
1098 }
1099 
1100 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1101 {
1102     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1103 }
1104 
1105 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1106                                     PowerPCCPU *cpu)
1107 {
1108     CPUPPCState *env = &cpu->env;
1109 
1110     /* The TCG path should also be holding the BQL at this point */
1111     g_assert(qemu_mutex_iothread_locked());
1112 
1113     if (msr_pr) {
1114         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1115         env->gpr[3] = H_PRIVILEGE;
1116     } else {
1117         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1118     }
1119 }
1120 
1121 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1122 {
1123     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1124 
1125     return spapr->patb_entry;
1126 }
1127 
1128 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1129 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1130 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1131 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1132 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1133 
1134 /*
1135  * Get the fd to access the kernel htab, re-opening it if necessary
1136  */
1137 static int get_htab_fd(sPAPRMachineState *spapr)
1138 {
1139     if (spapr->htab_fd >= 0) {
1140         return spapr->htab_fd;
1141     }
1142 
1143     spapr->htab_fd = kvmppc_get_htab_fd(false);
1144     if (spapr->htab_fd < 0) {
1145         error_report("Unable to open fd for reading hash table from KVM: %s",
1146                      strerror(errno));
1147     }
1148 
1149     return spapr->htab_fd;
1150 }
1151 
1152 void close_htab_fd(sPAPRMachineState *spapr)
1153 {
1154     if (spapr->htab_fd >= 0) {
1155         close(spapr->htab_fd);
1156     }
1157     spapr->htab_fd = -1;
1158 }
1159 
1160 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1161 {
1162     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1163 
1164     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1165 }
1166 
1167 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1168                                                 hwaddr ptex, int n)
1169 {
1170     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1171     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1172 
1173     if (!spapr->htab) {
1174         /*
1175          * HTAB is controlled by KVM. Fetch into temporary buffer
1176          */
1177         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1178         kvmppc_read_hptes(hptes, ptex, n);
1179         return hptes;
1180     }
1181 
1182     /*
1183      * HTAB is controlled by QEMU. Just point to the internally
1184      * accessible PTEG.
1185      */
1186     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1187 }
1188 
1189 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1190                               const ppc_hash_pte64_t *hptes,
1191                               hwaddr ptex, int n)
1192 {
1193     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1194 
1195     if (!spapr->htab) {
1196         g_free((void *)hptes);
1197     }
1198 
1199     /* Nothing to do for qemu managed HPT */
1200 }
1201 
1202 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1203                              uint64_t pte0, uint64_t pte1)
1204 {
1205     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1206     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1207 
1208     if (!spapr->htab) {
1209         kvmppc_write_hpte(ptex, pte0, pte1);
1210     } else {
1211         stq_p(spapr->htab + offset, pte0);
1212         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1213     }
1214 }
1215 
1216 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1217 {
1218     int shift;
1219 
1220     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1221      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1222      * that's much more than is needed for Linux guests */
1223     shift = ctz64(pow2ceil(ramsize)) - 7;
1224     shift = MAX(shift, 18); /* Minimum architected size */
1225     shift = MIN(shift, 46); /* Maximum architected size */
1226     return shift;
1227 }
1228 
1229 void spapr_free_hpt(sPAPRMachineState *spapr)
1230 {
1231     g_free(spapr->htab);
1232     spapr->htab = NULL;
1233     spapr->htab_shift = 0;
1234     close_htab_fd(spapr);
1235 }
1236 
1237 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1238                                  Error **errp)
1239 {
1240     long rc;
1241 
1242     /* Clean up any HPT info from a previous boot */
1243     spapr_free_hpt(spapr);
1244 
1245     rc = kvmppc_reset_htab(shift);
1246     if (rc < 0) {
1247         /* kernel-side HPT needed, but couldn't allocate one */
1248         error_setg_errno(errp, errno,
1249                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1250                          shift);
1251         /* This is almost certainly fatal, but if the caller really
1252          * wants to carry on with shift == 0, it's welcome to try */
1253     } else if (rc > 0) {
1254         /* kernel-side HPT allocated */
1255         if (rc != shift) {
1256             error_setg(errp,
1257                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1258                        shift, rc);
1259         }
1260 
1261         spapr->htab_shift = shift;
1262         spapr->htab = NULL;
1263     } else {
1264         /* kernel-side HPT not needed, allocate in userspace instead */
1265         size_t size = 1ULL << shift;
1266         int i;
1267 
1268         spapr->htab = qemu_memalign(size, size);
1269         if (!spapr->htab) {
1270             error_setg_errno(errp, errno,
1271                              "Could not allocate HPT of order %d", shift);
1272             return;
1273         }
1274 
1275         memset(spapr->htab, 0, size);
1276         spapr->htab_shift = shift;
1277 
1278         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1279             DIRTY_HPTE(HPTE(spapr->htab, i));
1280         }
1281     }
1282 }
1283 
1284 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1285 {
1286     spapr_reallocate_hpt(spapr,
1287                      spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size),
1288                      &error_fatal);
1289     if (spapr->vrma_adjust) {
1290         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1291                                           spapr->htab_shift);
1292     }
1293     /* We're setting up a hash table, so that means we're not radix */
1294     spapr->patb_entry = 0;
1295 }
1296 
1297 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1298 {
1299     bool matched = false;
1300 
1301     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1302         matched = true;
1303     }
1304 
1305     if (!matched) {
1306         error_report("Device %s is not supported by this machine yet.",
1307                      qdev_fw_name(DEVICE(sbdev)));
1308         exit(1);
1309     }
1310 }
1311 
1312 static void ppc_spapr_reset(void)
1313 {
1314     MachineState *machine = MACHINE(qdev_get_machine());
1315     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1316     PowerPCCPU *first_ppc_cpu;
1317     uint32_t rtas_limit;
1318     hwaddr rtas_addr, fdt_addr;
1319     void *fdt;
1320     int rc;
1321 
1322     /* Check for unknown sysbus devices */
1323     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1324 
1325     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1326         /* If using KVM with radix mode available, VCPUs can be started
1327          * without a HPT because KVM will start them in radix mode.
1328          * Set the GR bit in PATB so that we know there is no HPT. */
1329         spapr->patb_entry = PATBE1_GR;
1330     } else {
1331         spapr->patb_entry = 0;
1332         spapr_setup_hpt_and_vrma(spapr);
1333     }
1334 
1335     qemu_devices_reset();
1336 
1337     /*
1338      * We place the device tree and RTAS just below either the top of the RMA,
1339      * or just below 2GB, whichever is lowere, so that it can be
1340      * processed with 32-bit real mode code if necessary
1341      */
1342     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1343     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1344     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1345 
1346     /* if this reset wasn't generated by CAS, we should reset our
1347      * negotiated options and start from scratch */
1348     if (!spapr->cas_reboot) {
1349         spapr_ovec_cleanup(spapr->ov5_cas);
1350         spapr->ov5_cas = spapr_ovec_new();
1351     }
1352 
1353     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1354 
1355     spapr_load_rtas(spapr, fdt, rtas_addr);
1356 
1357     rc = fdt_pack(fdt);
1358 
1359     /* Should only fail if we've built a corrupted tree */
1360     assert(rc == 0);
1361 
1362     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1363         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1364                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1365         exit(1);
1366     }
1367 
1368     /* Load the fdt */
1369     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1370     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1371     g_free(fdt);
1372 
1373     /* Set up the entry state */
1374     first_ppc_cpu = POWERPC_CPU(first_cpu);
1375     first_ppc_cpu->env.gpr[3] = fdt_addr;
1376     first_ppc_cpu->env.gpr[5] = 0;
1377     first_cpu->halted = 0;
1378     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1379 
1380     spapr->cas_reboot = false;
1381 }
1382 
1383 static void spapr_create_nvram(sPAPRMachineState *spapr)
1384 {
1385     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1386     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1387 
1388     if (dinfo) {
1389         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1390                             &error_fatal);
1391     }
1392 
1393     qdev_init_nofail(dev);
1394 
1395     spapr->nvram = (struct sPAPRNVRAM *)dev;
1396 }
1397 
1398 static void spapr_rtc_create(sPAPRMachineState *spapr)
1399 {
1400     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1401     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1402                               &error_fatal);
1403     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1404                               &error_fatal);
1405     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1406                               "date", &error_fatal);
1407 }
1408 
1409 /* Returns whether we want to use VGA or not */
1410 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1411 {
1412     switch (vga_interface_type) {
1413     case VGA_NONE:
1414         return false;
1415     case VGA_DEVICE:
1416         return true;
1417     case VGA_STD:
1418     case VGA_VIRTIO:
1419         return pci_vga_init(pci_bus) != NULL;
1420     default:
1421         error_setg(errp,
1422                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1423         return false;
1424     }
1425 }
1426 
1427 static int spapr_post_load(void *opaque, int version_id)
1428 {
1429     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1430     int err = 0;
1431 
1432     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1433         CPUState *cs;
1434         CPU_FOREACH(cs) {
1435             PowerPCCPU *cpu = POWERPC_CPU(cs);
1436             icp_resend(ICP(cpu->intc));
1437         }
1438     }
1439 
1440     /* In earlier versions, there was no separate qdev for the PAPR
1441      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1442      * So when migrating from those versions, poke the incoming offset
1443      * value into the RTC device */
1444     if (version_id < 3) {
1445         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1446     }
1447 
1448     return err;
1449 }
1450 
1451 static bool version_before_3(void *opaque, int version_id)
1452 {
1453     return version_id < 3;
1454 }
1455 
1456 static bool spapr_ov5_cas_needed(void *opaque)
1457 {
1458     sPAPRMachineState *spapr = opaque;
1459     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1460     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1461     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1462     bool cas_needed;
1463 
1464     /* Prior to the introduction of sPAPROptionVector, we had two option
1465      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1466      * Both of these options encode machine topology into the device-tree
1467      * in such a way that the now-booted OS should still be able to interact
1468      * appropriately with QEMU regardless of what options were actually
1469      * negotiatied on the source side.
1470      *
1471      * As such, we can avoid migrating the CAS-negotiated options if these
1472      * are the only options available on the current machine/platform.
1473      * Since these are the only options available for pseries-2.7 and
1474      * earlier, this allows us to maintain old->new/new->old migration
1475      * compatibility.
1476      *
1477      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1478      * via default pseries-2.8 machines and explicit command-line parameters.
1479      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1480      * of the actual CAS-negotiated values to continue working properly. For
1481      * example, availability of memory unplug depends on knowing whether
1482      * OV5_HP_EVT was negotiated via CAS.
1483      *
1484      * Thus, for any cases where the set of available CAS-negotiatable
1485      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1486      * include the CAS-negotiated options in the migration stream.
1487      */
1488     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1489     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1490 
1491     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1492      * the mask itself since in the future it's possible "legacy" bits may be
1493      * removed via machine options, which could generate a false positive
1494      * that breaks migration.
1495      */
1496     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1497     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1498 
1499     spapr_ovec_cleanup(ov5_mask);
1500     spapr_ovec_cleanup(ov5_legacy);
1501     spapr_ovec_cleanup(ov5_removed);
1502 
1503     return cas_needed;
1504 }
1505 
1506 static const VMStateDescription vmstate_spapr_ov5_cas = {
1507     .name = "spapr_option_vector_ov5_cas",
1508     .version_id = 1,
1509     .minimum_version_id = 1,
1510     .needed = spapr_ov5_cas_needed,
1511     .fields = (VMStateField[]) {
1512         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1513                                  vmstate_spapr_ovec, sPAPROptionVector),
1514         VMSTATE_END_OF_LIST()
1515     },
1516 };
1517 
1518 static bool spapr_patb_entry_needed(void *opaque)
1519 {
1520     sPAPRMachineState *spapr = opaque;
1521 
1522     return !!spapr->patb_entry;
1523 }
1524 
1525 static const VMStateDescription vmstate_spapr_patb_entry = {
1526     .name = "spapr_patb_entry",
1527     .version_id = 1,
1528     .minimum_version_id = 1,
1529     .needed = spapr_patb_entry_needed,
1530     .fields = (VMStateField[]) {
1531         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1532         VMSTATE_END_OF_LIST()
1533     },
1534 };
1535 
1536 static const VMStateDescription vmstate_spapr = {
1537     .name = "spapr",
1538     .version_id = 3,
1539     .minimum_version_id = 1,
1540     .post_load = spapr_post_load,
1541     .fields = (VMStateField[]) {
1542         /* used to be @next_irq */
1543         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1544 
1545         /* RTC offset */
1546         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1547 
1548         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1549         VMSTATE_END_OF_LIST()
1550     },
1551     .subsections = (const VMStateDescription*[]) {
1552         &vmstate_spapr_ov5_cas,
1553         &vmstate_spapr_patb_entry,
1554         NULL
1555     }
1556 };
1557 
1558 static int htab_save_setup(QEMUFile *f, void *opaque)
1559 {
1560     sPAPRMachineState *spapr = opaque;
1561 
1562     /* "Iteration" header */
1563     qemu_put_be32(f, spapr->htab_shift);
1564 
1565     if (spapr->htab) {
1566         spapr->htab_save_index = 0;
1567         spapr->htab_first_pass = true;
1568     } else {
1569         assert(kvm_enabled());
1570     }
1571 
1572 
1573     return 0;
1574 }
1575 
1576 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1577                                  int64_t max_ns)
1578 {
1579     bool has_timeout = max_ns != -1;
1580     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1581     int index = spapr->htab_save_index;
1582     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1583 
1584     assert(spapr->htab_first_pass);
1585 
1586     do {
1587         int chunkstart;
1588 
1589         /* Consume invalid HPTEs */
1590         while ((index < htabslots)
1591                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1592             CLEAN_HPTE(HPTE(spapr->htab, index));
1593             index++;
1594         }
1595 
1596         /* Consume valid HPTEs */
1597         chunkstart = index;
1598         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1599                && HPTE_VALID(HPTE(spapr->htab, index))) {
1600             CLEAN_HPTE(HPTE(spapr->htab, index));
1601             index++;
1602         }
1603 
1604         if (index > chunkstart) {
1605             int n_valid = index - chunkstart;
1606 
1607             qemu_put_be32(f, chunkstart);
1608             qemu_put_be16(f, n_valid);
1609             qemu_put_be16(f, 0);
1610             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1611                             HASH_PTE_SIZE_64 * n_valid);
1612 
1613             if (has_timeout &&
1614                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1615                 break;
1616             }
1617         }
1618     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1619 
1620     if (index >= htabslots) {
1621         assert(index == htabslots);
1622         index = 0;
1623         spapr->htab_first_pass = false;
1624     }
1625     spapr->htab_save_index = index;
1626 }
1627 
1628 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1629                                 int64_t max_ns)
1630 {
1631     bool final = max_ns < 0;
1632     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1633     int examined = 0, sent = 0;
1634     int index = spapr->htab_save_index;
1635     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1636 
1637     assert(!spapr->htab_first_pass);
1638 
1639     do {
1640         int chunkstart, invalidstart;
1641 
1642         /* Consume non-dirty HPTEs */
1643         while ((index < htabslots)
1644                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1645             index++;
1646             examined++;
1647         }
1648 
1649         chunkstart = index;
1650         /* Consume valid dirty HPTEs */
1651         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1652                && HPTE_DIRTY(HPTE(spapr->htab, index))
1653                && HPTE_VALID(HPTE(spapr->htab, index))) {
1654             CLEAN_HPTE(HPTE(spapr->htab, index));
1655             index++;
1656             examined++;
1657         }
1658 
1659         invalidstart = index;
1660         /* Consume invalid dirty HPTEs */
1661         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1662                && HPTE_DIRTY(HPTE(spapr->htab, index))
1663                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1664             CLEAN_HPTE(HPTE(spapr->htab, index));
1665             index++;
1666             examined++;
1667         }
1668 
1669         if (index > chunkstart) {
1670             int n_valid = invalidstart - chunkstart;
1671             int n_invalid = index - invalidstart;
1672 
1673             qemu_put_be32(f, chunkstart);
1674             qemu_put_be16(f, n_valid);
1675             qemu_put_be16(f, n_invalid);
1676             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1677                             HASH_PTE_SIZE_64 * n_valid);
1678             sent += index - chunkstart;
1679 
1680             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1681                 break;
1682             }
1683         }
1684 
1685         if (examined >= htabslots) {
1686             break;
1687         }
1688 
1689         if (index >= htabslots) {
1690             assert(index == htabslots);
1691             index = 0;
1692         }
1693     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1694 
1695     if (index >= htabslots) {
1696         assert(index == htabslots);
1697         index = 0;
1698     }
1699 
1700     spapr->htab_save_index = index;
1701 
1702     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1703 }
1704 
1705 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1706 #define MAX_KVM_BUF_SIZE    2048
1707 
1708 static int htab_save_iterate(QEMUFile *f, void *opaque)
1709 {
1710     sPAPRMachineState *spapr = opaque;
1711     int fd;
1712     int rc = 0;
1713 
1714     /* Iteration header */
1715     qemu_put_be32(f, 0);
1716 
1717     if (!spapr->htab) {
1718         assert(kvm_enabled());
1719 
1720         fd = get_htab_fd(spapr);
1721         if (fd < 0) {
1722             return fd;
1723         }
1724 
1725         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1726         if (rc < 0) {
1727             return rc;
1728         }
1729     } else  if (spapr->htab_first_pass) {
1730         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1731     } else {
1732         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1733     }
1734 
1735     /* End marker */
1736     qemu_put_be32(f, 0);
1737     qemu_put_be16(f, 0);
1738     qemu_put_be16(f, 0);
1739 
1740     return rc;
1741 }
1742 
1743 static int htab_save_complete(QEMUFile *f, void *opaque)
1744 {
1745     sPAPRMachineState *spapr = opaque;
1746     int fd;
1747 
1748     /* Iteration header */
1749     qemu_put_be32(f, 0);
1750 
1751     if (!spapr->htab) {
1752         int rc;
1753 
1754         assert(kvm_enabled());
1755 
1756         fd = get_htab_fd(spapr);
1757         if (fd < 0) {
1758             return fd;
1759         }
1760 
1761         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1762         if (rc < 0) {
1763             return rc;
1764         }
1765     } else {
1766         if (spapr->htab_first_pass) {
1767             htab_save_first_pass(f, spapr, -1);
1768         }
1769         htab_save_later_pass(f, spapr, -1);
1770     }
1771 
1772     /* End marker */
1773     qemu_put_be32(f, 0);
1774     qemu_put_be16(f, 0);
1775     qemu_put_be16(f, 0);
1776 
1777     return 0;
1778 }
1779 
1780 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1781 {
1782     sPAPRMachineState *spapr = opaque;
1783     uint32_t section_hdr;
1784     int fd = -1;
1785 
1786     if (version_id < 1 || version_id > 1) {
1787         error_report("htab_load() bad version");
1788         return -EINVAL;
1789     }
1790 
1791     section_hdr = qemu_get_be32(f);
1792 
1793     if (section_hdr) {
1794         Error *local_err = NULL;
1795 
1796         /* First section gives the htab size */
1797         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1798         if (local_err) {
1799             error_report_err(local_err);
1800             return -EINVAL;
1801         }
1802         return 0;
1803     }
1804 
1805     if (!spapr->htab) {
1806         assert(kvm_enabled());
1807 
1808         fd = kvmppc_get_htab_fd(true);
1809         if (fd < 0) {
1810             error_report("Unable to open fd to restore KVM hash table: %s",
1811                          strerror(errno));
1812         }
1813     }
1814 
1815     while (true) {
1816         uint32_t index;
1817         uint16_t n_valid, n_invalid;
1818 
1819         index = qemu_get_be32(f);
1820         n_valid = qemu_get_be16(f);
1821         n_invalid = qemu_get_be16(f);
1822 
1823         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1824             /* End of Stream */
1825             break;
1826         }
1827 
1828         if ((index + n_valid + n_invalid) >
1829             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1830             /* Bad index in stream */
1831             error_report(
1832                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1833                 index, n_valid, n_invalid, spapr->htab_shift);
1834             return -EINVAL;
1835         }
1836 
1837         if (spapr->htab) {
1838             if (n_valid) {
1839                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1840                                 HASH_PTE_SIZE_64 * n_valid);
1841             }
1842             if (n_invalid) {
1843                 memset(HPTE(spapr->htab, index + n_valid), 0,
1844                        HASH_PTE_SIZE_64 * n_invalid);
1845             }
1846         } else {
1847             int rc;
1848 
1849             assert(fd >= 0);
1850 
1851             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1852             if (rc < 0) {
1853                 return rc;
1854             }
1855         }
1856     }
1857 
1858     if (!spapr->htab) {
1859         assert(fd >= 0);
1860         close(fd);
1861     }
1862 
1863     return 0;
1864 }
1865 
1866 static void htab_cleanup(void *opaque)
1867 {
1868     sPAPRMachineState *spapr = opaque;
1869 
1870     close_htab_fd(spapr);
1871 }
1872 
1873 static SaveVMHandlers savevm_htab_handlers = {
1874     .save_live_setup = htab_save_setup,
1875     .save_live_iterate = htab_save_iterate,
1876     .save_live_complete_precopy = htab_save_complete,
1877     .cleanup = htab_cleanup,
1878     .load_state = htab_load,
1879 };
1880 
1881 static void spapr_boot_set(void *opaque, const char *boot_device,
1882                            Error **errp)
1883 {
1884     MachineState *machine = MACHINE(qdev_get_machine());
1885     machine->boot_order = g_strdup(boot_device);
1886 }
1887 
1888 /*
1889  * Reset routine for LMB DR devices.
1890  *
1891  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1892  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1893  * when it walks all its children devices. LMB devices reset occurs
1894  * as part of spapr_ppc_reset().
1895  */
1896 static void spapr_drc_reset(void *opaque)
1897 {
1898     sPAPRDRConnector *drc = opaque;
1899     DeviceState *d = DEVICE(drc);
1900 
1901     if (d) {
1902         device_reset(d);
1903     }
1904 }
1905 
1906 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1907 {
1908     MachineState *machine = MACHINE(spapr);
1909     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1910     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1911     int i;
1912 
1913     for (i = 0; i < nr_lmbs; i++) {
1914         sPAPRDRConnector *drc;
1915         uint64_t addr;
1916 
1917         addr = i * lmb_size + spapr->hotplug_memory.base;
1918         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1919                                      addr/lmb_size);
1920         qemu_register_reset(spapr_drc_reset, drc);
1921     }
1922 }
1923 
1924 /*
1925  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1926  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1927  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1928  */
1929 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1930 {
1931     int i;
1932 
1933     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1934         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1935                    " is not aligned to %llu MiB",
1936                    machine->ram_size,
1937                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1938         return;
1939     }
1940 
1941     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1942         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1943                    " is not aligned to %llu MiB",
1944                    machine->ram_size,
1945                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1946         return;
1947     }
1948 
1949     for (i = 0; i < nb_numa_nodes; i++) {
1950         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1951             error_setg(errp,
1952                        "Node %d memory size 0x%" PRIx64
1953                        " is not aligned to %llu MiB",
1954                        i, numa_info[i].node_mem,
1955                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1956             return;
1957         }
1958     }
1959 }
1960 
1961 /* find cpu slot in machine->possible_cpus by core_id */
1962 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1963 {
1964     int index = id / smp_threads;
1965 
1966     if (index >= ms->possible_cpus->len) {
1967         return NULL;
1968     }
1969     if (idx) {
1970         *idx = index;
1971     }
1972     return &ms->possible_cpus->cpus[index];
1973 }
1974 
1975 static void spapr_init_cpus(sPAPRMachineState *spapr)
1976 {
1977     MachineState *machine = MACHINE(spapr);
1978     MachineClass *mc = MACHINE_GET_CLASS(machine);
1979     char *type = spapr_get_cpu_core_type(machine->cpu_model);
1980     int smt = kvmppc_smt_threads();
1981     const CPUArchIdList *possible_cpus;
1982     int boot_cores_nr = smp_cpus / smp_threads;
1983     int i;
1984 
1985     if (!type) {
1986         error_report("Unable to find sPAPR CPU Core definition");
1987         exit(1);
1988     }
1989 
1990     possible_cpus = mc->possible_cpu_arch_ids(machine);
1991     if (mc->has_hotpluggable_cpus) {
1992         if (smp_cpus % smp_threads) {
1993             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1994                          smp_cpus, smp_threads);
1995             exit(1);
1996         }
1997         if (max_cpus % smp_threads) {
1998             error_report("max_cpus (%u) must be multiple of threads (%u)",
1999                          max_cpus, smp_threads);
2000             exit(1);
2001         }
2002     } else {
2003         if (max_cpus != smp_cpus) {
2004             error_report("This machine version does not support CPU hotplug");
2005             exit(1);
2006         }
2007         boot_cores_nr = possible_cpus->len;
2008     }
2009 
2010     for (i = 0; i < possible_cpus->len; i++) {
2011         int core_id = i * smp_threads;
2012 
2013         if (mc->has_hotpluggable_cpus) {
2014             sPAPRDRConnector *drc =
2015                 spapr_dr_connector_new(OBJECT(spapr),
2016                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
2017                                        (core_id / smp_threads) * smt);
2018 
2019             qemu_register_reset(spapr_drc_reset, drc);
2020         }
2021 
2022         if (i < boot_cores_nr) {
2023             Object *core  = object_new(type);
2024             int nr_threads = smp_threads;
2025 
2026             /* Handle the partially filled core for older machine types */
2027             if ((i + 1) * smp_threads >= smp_cpus) {
2028                 nr_threads = smp_cpus - i * smp_threads;
2029             }
2030 
2031             object_property_set_int(core, nr_threads, "nr-threads",
2032                                     &error_fatal);
2033             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2034                                     &error_fatal);
2035             object_property_set_bool(core, true, "realized", &error_fatal);
2036         }
2037     }
2038     g_free(type);
2039 }
2040 
2041 /* pSeries LPAR / sPAPR hardware init */
2042 static void ppc_spapr_init(MachineState *machine)
2043 {
2044     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2045     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2046     const char *kernel_filename = machine->kernel_filename;
2047     const char *initrd_filename = machine->initrd_filename;
2048     PCIHostState *phb;
2049     int i;
2050     MemoryRegion *sysmem = get_system_memory();
2051     MemoryRegion *ram = g_new(MemoryRegion, 1);
2052     MemoryRegion *rma_region;
2053     void *rma = NULL;
2054     hwaddr rma_alloc_size;
2055     hwaddr node0_size = spapr_node0_size();
2056     long load_limit, fw_size;
2057     char *filename;
2058 
2059     msi_nonbroken = true;
2060 
2061     QLIST_INIT(&spapr->phbs);
2062     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2063 
2064     /* Allocate RMA if necessary */
2065     rma_alloc_size = kvmppc_alloc_rma(&rma);
2066 
2067     if (rma_alloc_size == -1) {
2068         error_report("Unable to create RMA");
2069         exit(1);
2070     }
2071 
2072     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2073         spapr->rma_size = rma_alloc_size;
2074     } else {
2075         spapr->rma_size = node0_size;
2076 
2077         /* With KVM, we don't actually know whether KVM supports an
2078          * unbounded RMA (PR KVM) or is limited by the hash table size
2079          * (HV KVM using VRMA), so we always assume the latter
2080          *
2081          * In that case, we also limit the initial allocations for RTAS
2082          * etc... to 256M since we have no way to know what the VRMA size
2083          * is going to be as it depends on the size of the hash table
2084          * isn't determined yet.
2085          */
2086         if (kvm_enabled()) {
2087             spapr->vrma_adjust = 1;
2088             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2089         }
2090 
2091         /* Actually we don't support unbounded RMA anymore since we
2092          * added proper emulation of HV mode. The max we can get is
2093          * 16G which also happens to be what we configure for PAPR
2094          * mode so make sure we don't do anything bigger than that
2095          */
2096         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2097     }
2098 
2099     if (spapr->rma_size > node0_size) {
2100         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2101                      spapr->rma_size);
2102         exit(1);
2103     }
2104 
2105     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2106     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2107 
2108     /* Set up Interrupt Controller before we create the VCPUs */
2109     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2110 
2111     /* Set up containers for ibm,client-set-architecture negotiated options */
2112     spapr->ov5 = spapr_ovec_new();
2113     spapr->ov5_cas = spapr_ovec_new();
2114 
2115     if (smc->dr_lmb_enabled) {
2116         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2117         spapr_validate_node_memory(machine, &error_fatal);
2118     }
2119 
2120     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2121     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2122         /* KVM and TCG always allow GTSE with radix... */
2123         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2124     }
2125     /* ... but not with hash (currently). */
2126 
2127     /* advertise support for dedicated HP event source to guests */
2128     if (spapr->use_hotplug_event_source) {
2129         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2130     }
2131 
2132     /* init CPUs */
2133     if (machine->cpu_model == NULL) {
2134         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2135     }
2136 
2137     ppc_cpu_parse_features(machine->cpu_model);
2138 
2139     spapr_init_cpus(spapr);
2140 
2141     if (kvm_enabled()) {
2142         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2143         kvmppc_enable_logical_ci_hcalls();
2144         kvmppc_enable_set_mode_hcall();
2145 
2146         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2147         kvmppc_enable_clear_ref_mod_hcalls();
2148     }
2149 
2150     /* allocate RAM */
2151     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2152                                          machine->ram_size);
2153     memory_region_add_subregion(sysmem, 0, ram);
2154 
2155     if (rma_alloc_size && rma) {
2156         rma_region = g_new(MemoryRegion, 1);
2157         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2158                                    rma_alloc_size, rma);
2159         vmstate_register_ram_global(rma_region);
2160         memory_region_add_subregion(sysmem, 0, rma_region);
2161     }
2162 
2163     /* initialize hotplug memory address space */
2164     if (machine->ram_size < machine->maxram_size) {
2165         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2166         /*
2167          * Limit the number of hotpluggable memory slots to half the number
2168          * slots that KVM supports, leaving the other half for PCI and other
2169          * devices. However ensure that number of slots doesn't drop below 32.
2170          */
2171         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2172                            SPAPR_MAX_RAM_SLOTS;
2173 
2174         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2175             max_memslots = SPAPR_MAX_RAM_SLOTS;
2176         }
2177         if (machine->ram_slots > max_memslots) {
2178             error_report("Specified number of memory slots %"
2179                          PRIu64" exceeds max supported %d",
2180                          machine->ram_slots, max_memslots);
2181             exit(1);
2182         }
2183 
2184         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2185                                               SPAPR_HOTPLUG_MEM_ALIGN);
2186         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2187                            "hotplug-memory", hotplug_mem_size);
2188         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2189                                     &spapr->hotplug_memory.mr);
2190     }
2191 
2192     if (smc->dr_lmb_enabled) {
2193         spapr_create_lmb_dr_connectors(spapr);
2194     }
2195 
2196     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2197     if (!filename) {
2198         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2199         exit(1);
2200     }
2201     spapr->rtas_size = get_image_size(filename);
2202     if (spapr->rtas_size < 0) {
2203         error_report("Could not get size of LPAR rtas '%s'", filename);
2204         exit(1);
2205     }
2206     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2207     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2208         error_report("Could not load LPAR rtas '%s'", filename);
2209         exit(1);
2210     }
2211     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2212         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2213                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2214         exit(1);
2215     }
2216     g_free(filename);
2217 
2218     /* Set up RTAS event infrastructure */
2219     spapr_events_init(spapr);
2220 
2221     /* Set up the RTC RTAS interfaces */
2222     spapr_rtc_create(spapr);
2223 
2224     /* Set up VIO bus */
2225     spapr->vio_bus = spapr_vio_bus_init();
2226 
2227     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2228         if (serial_hds[i]) {
2229             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2230         }
2231     }
2232 
2233     /* We always have at least the nvram device on VIO */
2234     spapr_create_nvram(spapr);
2235 
2236     /* Set up PCI */
2237     spapr_pci_rtas_init();
2238 
2239     phb = spapr_create_phb(spapr, 0);
2240 
2241     for (i = 0; i < nb_nics; i++) {
2242         NICInfo *nd = &nd_table[i];
2243 
2244         if (!nd->model) {
2245             nd->model = g_strdup("ibmveth");
2246         }
2247 
2248         if (strcmp(nd->model, "ibmveth") == 0) {
2249             spapr_vlan_create(spapr->vio_bus, nd);
2250         } else {
2251             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2252         }
2253     }
2254 
2255     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2256         spapr_vscsi_create(spapr->vio_bus);
2257     }
2258 
2259     /* Graphics */
2260     if (spapr_vga_init(phb->bus, &error_fatal)) {
2261         spapr->has_graphics = true;
2262         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2263     }
2264 
2265     if (machine->usb) {
2266         if (smc->use_ohci_by_default) {
2267             pci_create_simple(phb->bus, -1, "pci-ohci");
2268         } else {
2269             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2270         }
2271 
2272         if (spapr->has_graphics) {
2273             USBBus *usb_bus = usb_bus_find(-1);
2274 
2275             usb_create_simple(usb_bus, "usb-kbd");
2276             usb_create_simple(usb_bus, "usb-mouse");
2277         }
2278     }
2279 
2280     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2281         error_report(
2282             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2283             MIN_RMA_SLOF);
2284         exit(1);
2285     }
2286 
2287     if (kernel_filename) {
2288         uint64_t lowaddr = 0;
2289 
2290         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2291                                       NULL, NULL, &lowaddr, NULL, 1,
2292                                       PPC_ELF_MACHINE, 0, 0);
2293         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2294             spapr->kernel_size = load_elf(kernel_filename,
2295                                           translate_kernel_address, NULL, NULL,
2296                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2297                                           0, 0);
2298             spapr->kernel_le = spapr->kernel_size > 0;
2299         }
2300         if (spapr->kernel_size < 0) {
2301             error_report("error loading %s: %s", kernel_filename,
2302                          load_elf_strerror(spapr->kernel_size));
2303             exit(1);
2304         }
2305 
2306         /* load initrd */
2307         if (initrd_filename) {
2308             /* Try to locate the initrd in the gap between the kernel
2309              * and the firmware. Add a bit of space just in case
2310              */
2311             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2312                                   + 0x1ffff) & ~0xffff;
2313             spapr->initrd_size = load_image_targphys(initrd_filename,
2314                                                      spapr->initrd_base,
2315                                                      load_limit
2316                                                      - spapr->initrd_base);
2317             if (spapr->initrd_size < 0) {
2318                 error_report("could not load initial ram disk '%s'",
2319                              initrd_filename);
2320                 exit(1);
2321             }
2322         }
2323     }
2324 
2325     if (bios_name == NULL) {
2326         bios_name = FW_FILE_NAME;
2327     }
2328     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2329     if (!filename) {
2330         error_report("Could not find LPAR firmware '%s'", bios_name);
2331         exit(1);
2332     }
2333     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2334     if (fw_size <= 0) {
2335         error_report("Could not load LPAR firmware '%s'", filename);
2336         exit(1);
2337     }
2338     g_free(filename);
2339 
2340     /* FIXME: Should register things through the MachineState's qdev
2341      * interface, this is a legacy from the sPAPREnvironment structure
2342      * which predated MachineState but had a similar function */
2343     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2344     register_savevm_live(NULL, "spapr/htab", -1, 1,
2345                          &savevm_htab_handlers, spapr);
2346 
2347     /* used by RTAS */
2348     QTAILQ_INIT(&spapr->ccs_list);
2349     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2350 
2351     qemu_register_boot_set(spapr_boot_set, spapr);
2352 
2353     if (kvm_enabled()) {
2354         /* to stop and start vmclock */
2355         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2356                                          &spapr->tb);
2357 
2358         kvmppc_spapr_enable_inkernel_multitce();
2359     }
2360 }
2361 
2362 static int spapr_kvm_type(const char *vm_type)
2363 {
2364     if (!vm_type) {
2365         return 0;
2366     }
2367 
2368     if (!strcmp(vm_type, "HV")) {
2369         return 1;
2370     }
2371 
2372     if (!strcmp(vm_type, "PR")) {
2373         return 2;
2374     }
2375 
2376     error_report("Unknown kvm-type specified '%s'", vm_type);
2377     exit(1);
2378 }
2379 
2380 /*
2381  * Implementation of an interface to adjust firmware path
2382  * for the bootindex property handling.
2383  */
2384 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2385                                    DeviceState *dev)
2386 {
2387 #define CAST(type, obj, name) \
2388     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2389     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2390     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2391 
2392     if (d) {
2393         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2394         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2395         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2396 
2397         if (spapr) {
2398             /*
2399              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2400              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2401              * in the top 16 bits of the 64-bit LUN
2402              */
2403             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2404             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2405                                    (uint64_t)id << 48);
2406         } else if (virtio) {
2407             /*
2408              * We use SRP luns of the form 01000000 | (target << 8) | lun
2409              * in the top 32 bits of the 64-bit LUN
2410              * Note: the quote above is from SLOF and it is wrong,
2411              * the actual binding is:
2412              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2413              */
2414             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2415             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2416                                    (uint64_t)id << 32);
2417         } else if (usb) {
2418             /*
2419              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2420              * in the top 32 bits of the 64-bit LUN
2421              */
2422             unsigned usb_port = atoi(usb->port->path);
2423             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2424             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2425                                    (uint64_t)id << 32);
2426         }
2427     }
2428 
2429     /*
2430      * SLOF probes the USB devices, and if it recognizes that the device is a
2431      * storage device, it changes its name to "storage" instead of "usb-host",
2432      * and additionally adds a child node for the SCSI LUN, so the correct
2433      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2434      */
2435     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2436         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2437         if (usb_host_dev_is_scsi_storage(usbdev)) {
2438             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2439         }
2440     }
2441 
2442     if (phb) {
2443         /* Replace "pci" with "pci@800000020000000" */
2444         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2445     }
2446 
2447     return NULL;
2448 }
2449 
2450 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2451 {
2452     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2453 
2454     return g_strdup(spapr->kvm_type);
2455 }
2456 
2457 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2458 {
2459     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2460 
2461     g_free(spapr->kvm_type);
2462     spapr->kvm_type = g_strdup(value);
2463 }
2464 
2465 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2466 {
2467     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2468 
2469     return spapr->use_hotplug_event_source;
2470 }
2471 
2472 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2473                                             Error **errp)
2474 {
2475     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2476 
2477     spapr->use_hotplug_event_source = value;
2478 }
2479 
2480 static void spapr_machine_initfn(Object *obj)
2481 {
2482     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2483 
2484     spapr->htab_fd = -1;
2485     spapr->use_hotplug_event_source = true;
2486     object_property_add_str(obj, "kvm-type",
2487                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2488     object_property_set_description(obj, "kvm-type",
2489                                     "Specifies the KVM virtualization mode (HV, PR)",
2490                                     NULL);
2491     object_property_add_bool(obj, "modern-hotplug-events",
2492                             spapr_get_modern_hotplug_events,
2493                             spapr_set_modern_hotplug_events,
2494                             NULL);
2495     object_property_set_description(obj, "modern-hotplug-events",
2496                                     "Use dedicated hotplug event mechanism in"
2497                                     " place of standard EPOW events when possible"
2498                                     " (required for memory hot-unplug support)",
2499                                     NULL);
2500 }
2501 
2502 static void spapr_machine_finalizefn(Object *obj)
2503 {
2504     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2505 
2506     g_free(spapr->kvm_type);
2507 }
2508 
2509 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2510 {
2511     cpu_synchronize_state(cs);
2512     ppc_cpu_do_system_reset(cs);
2513 }
2514 
2515 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2516 {
2517     CPUState *cs;
2518 
2519     CPU_FOREACH(cs) {
2520         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2521     }
2522 }
2523 
2524 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2525                            uint32_t node, bool dedicated_hp_event_source,
2526                            Error **errp)
2527 {
2528     sPAPRDRConnector *drc;
2529     sPAPRDRConnectorClass *drck;
2530     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2531     int i, fdt_offset, fdt_size;
2532     void *fdt;
2533     uint64_t addr = addr_start;
2534 
2535     for (i = 0; i < nr_lmbs; i++) {
2536         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2537                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2538         g_assert(drc);
2539 
2540         fdt = create_device_tree(&fdt_size);
2541         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2542                                                 SPAPR_MEMORY_BLOCK_SIZE);
2543 
2544         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2545         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2546         addr += SPAPR_MEMORY_BLOCK_SIZE;
2547         if (!dev->hotplugged) {
2548             /* guests expect coldplugged LMBs to be pre-allocated */
2549             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2550             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2551         }
2552     }
2553     /* send hotplug notification to the
2554      * guest only in case of hotplugged memory
2555      */
2556     if (dev->hotplugged) {
2557         if (dedicated_hp_event_source) {
2558             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2559                     addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2560             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2561             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2562                                                    nr_lmbs,
2563                                                    drck->get_index(drc));
2564         } else {
2565             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2566                                            nr_lmbs);
2567         }
2568     }
2569 }
2570 
2571 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2572                               uint32_t node, Error **errp)
2573 {
2574     Error *local_err = NULL;
2575     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2576     PCDIMMDevice *dimm = PC_DIMM(dev);
2577     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2578     MemoryRegion *mr = ddc->get_memory_region(dimm);
2579     uint64_t align = memory_region_get_alignment(mr);
2580     uint64_t size = memory_region_size(mr);
2581     uint64_t addr;
2582 
2583     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2584     if (local_err) {
2585         goto out;
2586     }
2587 
2588     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2589     if (local_err) {
2590         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2591         goto out;
2592     }
2593 
2594     spapr_add_lmbs(dev, addr, size, node,
2595                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2596                    &error_abort);
2597 
2598 out:
2599     error_propagate(errp, local_err);
2600 }
2601 
2602 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2603                                   Error **errp)
2604 {
2605     PCDIMMDevice *dimm = PC_DIMM(dev);
2606     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2607     MemoryRegion *mr = ddc->get_memory_region(dimm);
2608     uint64_t size = memory_region_size(mr);
2609     char *mem_dev;
2610 
2611     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2612         error_setg(errp, "Hotplugged memory size must be a multiple of "
2613                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2614         return;
2615     }
2616 
2617     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2618     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2619         error_setg(errp, "Memory backend has bad page size. "
2620                    "Use 'memory-backend-file' with correct mem-path.");
2621         return;
2622     }
2623 }
2624 
2625 struct sPAPRDIMMState {
2626     PCDIMMDevice *dimm;
2627     uint32_t nr_lmbs;
2628     QTAILQ_ENTRY(sPAPRDIMMState) next;
2629 };
2630 
2631 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
2632                                                        PCDIMMDevice *dimm)
2633 {
2634     sPAPRDIMMState *dimm_state = NULL;
2635 
2636     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
2637         if (dimm_state->dimm == dimm) {
2638             break;
2639         }
2640     }
2641     return dimm_state;
2642 }
2643 
2644 static void spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
2645                                            sPAPRDIMMState *dimm_state)
2646 {
2647     g_assert(!spapr_pending_dimm_unplugs_find(spapr, dimm_state->dimm));
2648     QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, dimm_state, next);
2649 }
2650 
2651 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
2652                                               sPAPRDIMMState *dimm_state)
2653 {
2654     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
2655     g_free(dimm_state);
2656 }
2657 
2658 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
2659                                                         PCDIMMDevice *dimm)
2660 {
2661     sPAPRDRConnector *drc;
2662     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2663     MemoryRegion *mr = ddc->get_memory_region(dimm);
2664     uint64_t size = memory_region_size(mr);
2665     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2666     uint32_t avail_lmbs = 0;
2667     uint64_t addr_start, addr;
2668     int i;
2669     sPAPRDIMMState *ds;
2670 
2671     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2672                                          &error_abort);
2673 
2674     addr = addr_start;
2675     for (i = 0; i < nr_lmbs; i++) {
2676         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2677                                        addr / SPAPR_MEMORY_BLOCK_SIZE);
2678         g_assert(drc);
2679         if (drc->indicator_state != SPAPR_DR_INDICATOR_STATE_INACTIVE) {
2680             avail_lmbs++;
2681         }
2682         addr += SPAPR_MEMORY_BLOCK_SIZE;
2683     }
2684 
2685     ds = g_malloc0(sizeof(sPAPRDIMMState));
2686     ds->nr_lmbs = avail_lmbs;
2687     ds->dimm = dimm;
2688     spapr_pending_dimm_unplugs_add(ms, ds);
2689     return ds;
2690 }
2691 
2692 /* Callback to be called during DRC release. */
2693 void spapr_lmb_release(DeviceState *dev)
2694 {
2695     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
2696     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
2697     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
2698 
2699     /* This information will get lost if a migration occurs
2700      * during the unplug process. In this case recover it. */
2701     if (ds == NULL) {
2702         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
2703         if (ds->nr_lmbs) {
2704             return;
2705         }
2706     } else if (--ds->nr_lmbs) {
2707         return;
2708     }
2709 
2710     spapr_pending_dimm_unplugs_remove(spapr, ds);
2711 
2712     /*
2713      * Now that all the LMBs have been removed by the guest, call the
2714      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2715      */
2716     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2717 }
2718 
2719 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2720                                 Error **errp)
2721 {
2722     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2723     PCDIMMDevice *dimm = PC_DIMM(dev);
2724     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2725     MemoryRegion *mr = ddc->get_memory_region(dimm);
2726 
2727     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2728     object_unparent(OBJECT(dev));
2729 }
2730 
2731 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2732                                         DeviceState *dev, Error **errp)
2733 {
2734     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
2735     Error *local_err = NULL;
2736     PCDIMMDevice *dimm = PC_DIMM(dev);
2737     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2738     MemoryRegion *mr = ddc->get_memory_region(dimm);
2739     uint64_t size = memory_region_size(mr);
2740     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2741     uint64_t addr_start, addr;
2742     int i;
2743     sPAPRDRConnector *drc;
2744     sPAPRDRConnectorClass *drck;
2745     sPAPRDIMMState *ds;
2746 
2747     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2748                                          &local_err);
2749     if (local_err) {
2750         goto out;
2751     }
2752 
2753     ds = g_malloc0(sizeof(sPAPRDIMMState));
2754     ds->nr_lmbs = nr_lmbs;
2755     ds->dimm = dimm;
2756     spapr_pending_dimm_unplugs_add(spapr, ds);
2757 
2758     addr = addr_start;
2759     for (i = 0; i < nr_lmbs; i++) {
2760         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2761                 addr / SPAPR_MEMORY_BLOCK_SIZE);
2762         g_assert(drc);
2763 
2764         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2765         drck->detach(drc, dev, errp);
2766         addr += SPAPR_MEMORY_BLOCK_SIZE;
2767     }
2768 
2769     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2770                                    addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2771     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2772     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2773                                               nr_lmbs,
2774                                               drck->get_index(drc));
2775 out:
2776     error_propagate(errp, local_err);
2777 }
2778 
2779 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2780                                     sPAPRMachineState *spapr)
2781 {
2782     PowerPCCPU *cpu = POWERPC_CPU(cs);
2783     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2784     int id = ppc_get_vcpu_dt_id(cpu);
2785     void *fdt;
2786     int offset, fdt_size;
2787     char *nodename;
2788 
2789     fdt = create_device_tree(&fdt_size);
2790     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2791     offset = fdt_add_subnode(fdt, 0, nodename);
2792 
2793     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2794     g_free(nodename);
2795 
2796     *fdt_offset = offset;
2797     return fdt;
2798 }
2799 
2800 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2801                               Error **errp)
2802 {
2803     MachineState *ms = MACHINE(qdev_get_machine());
2804     CPUCore *cc = CPU_CORE(dev);
2805     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2806 
2807     assert(core_slot);
2808     core_slot->cpu = NULL;
2809     object_unparent(OBJECT(dev));
2810 }
2811 
2812 /* Callback to be called during DRC release. */
2813 void spapr_core_release(DeviceState *dev)
2814 {
2815     HotplugHandler *hotplug_ctrl;
2816 
2817     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2818     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2819 }
2820 
2821 static
2822 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2823                                Error **errp)
2824 {
2825     int index;
2826     sPAPRDRConnector *drc;
2827     sPAPRDRConnectorClass *drck;
2828     Error *local_err = NULL;
2829     CPUCore *cc = CPU_CORE(dev);
2830     int smt = kvmppc_smt_threads();
2831 
2832     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2833         error_setg(errp, "Unable to find CPU core with core-id: %d",
2834                    cc->core_id);
2835         return;
2836     }
2837     if (index == 0) {
2838         error_setg(errp, "Boot CPU core may not be unplugged");
2839         return;
2840     }
2841 
2842     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2843     g_assert(drc);
2844 
2845     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2846     drck->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_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_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         sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2891         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &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     sPAPRCPUCore *sc = SPAPR_CPU_CORE(dev);
2926     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2927     const char *type = object_get_typename(OBJECT(dev));
2928     CPUArchId *core_slot;
2929     int node_id;
2930     int index;
2931 
2932     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2933         error_setg(&local_err, "CPU hotplug not supported for this machine");
2934         goto out;
2935     }
2936 
2937     if (strcmp(base_core_type, type)) {
2938         error_setg(&local_err, "CPU core type should be %s", base_core_type);
2939         goto out;
2940     }
2941 
2942     if (cc->core_id % smp_threads) {
2943         error_setg(&local_err, "invalid core id %d", cc->core_id);
2944         goto out;
2945     }
2946 
2947     /*
2948      * In general we should have homogeneous threads-per-core, but old
2949      * (pre hotplug support) machine types allow the last core to have
2950      * reduced threads as a compatibility hack for when we allowed
2951      * total vcpus not a multiple of threads-per-core.
2952      */
2953     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
2954         error_setg(errp, "invalid nr-threads %d, must be %d",
2955                    cc->nr_threads, smp_threads);
2956         return;
2957     }
2958 
2959     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2960     if (!core_slot) {
2961         error_setg(&local_err, "core id %d out of range", cc->core_id);
2962         goto out;
2963     }
2964 
2965     if (core_slot->cpu) {
2966         error_setg(&local_err, "core %d already populated", cc->core_id);
2967         goto out;
2968     }
2969 
2970     node_id = core_slot->props.node_id;
2971     if (!core_slot->props.has_node_id) {
2972         /* by default CPUState::numa_node was 0 if it's not set via CLI
2973          * keep it this way for now but in future we probably should
2974          * refuse to start up with incomplete numa mapping */
2975         node_id = 0;
2976     }
2977     if (sc->node_id == CPU_UNSET_NUMA_NODE_ID) {
2978         sc->node_id = node_id;
2979     } else if (sc->node_id != node_id) {
2980         error_setg(&local_err, "node-id %d must match numa node specified"
2981             "with -numa option for cpu-index %d", sc->node_id, cc->core_id);
2982         goto out;
2983     }
2984 
2985 out:
2986     g_free(base_core_type);
2987     error_propagate(errp, local_err);
2988 }
2989 
2990 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2991                                       DeviceState *dev, Error **errp)
2992 {
2993     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2994 
2995     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2996         int node;
2997 
2998         if (!smc->dr_lmb_enabled) {
2999             error_setg(errp, "Memory hotplug not supported for this machine");
3000             return;
3001         }
3002         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3003         if (*errp) {
3004             return;
3005         }
3006         if (node < 0 || node >= MAX_NODES) {
3007             error_setg(errp, "Invaild node %d", node);
3008             return;
3009         }
3010 
3011         /*
3012          * Currently PowerPC kernel doesn't allow hot-adding memory to
3013          * memory-less node, but instead will silently add the memory
3014          * to the first node that has some memory. This causes two
3015          * unexpected behaviours for the user.
3016          *
3017          * - Memory gets hotplugged to a different node than what the user
3018          *   specified.
3019          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3020          *   to memory-less node, a reboot will set things accordingly
3021          *   and the previously hotplugged memory now ends in the right node.
3022          *   This appears as if some memory moved from one node to another.
3023          *
3024          * So until kernel starts supporting memory hotplug to memory-less
3025          * nodes, just prevent such attempts upfront in QEMU.
3026          */
3027         if (nb_numa_nodes && !numa_info[node].node_mem) {
3028             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3029                        node);
3030             return;
3031         }
3032 
3033         spapr_memory_plug(hotplug_dev, dev, node, errp);
3034     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3035         spapr_core_plug(hotplug_dev, dev, errp);
3036     }
3037 }
3038 
3039 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3040                                       DeviceState *dev, Error **errp)
3041 {
3042     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
3043     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
3044 
3045     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3046         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3047             spapr_memory_unplug(hotplug_dev, dev, errp);
3048         } else {
3049             error_setg(errp, "Memory hot unplug not supported for this guest");
3050         }
3051     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3052         if (!mc->has_hotpluggable_cpus) {
3053             error_setg(errp, "CPU hot unplug not supported on this machine");
3054             return;
3055         }
3056         spapr_core_unplug(hotplug_dev, dev, errp);
3057     }
3058 }
3059 
3060 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3061                                                 DeviceState *dev, Error **errp)
3062 {
3063     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
3064     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
3065 
3066     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3067         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3068             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3069         } else {
3070             /* NOTE: this means there is a window after guest reset, prior to
3071              * CAS negotiation, where unplug requests will fail due to the
3072              * capability not being detected yet. This is a bit different than
3073              * the case with PCI unplug, where the events will be queued and
3074              * eventually handled by the guest after boot
3075              */
3076             error_setg(errp, "Memory hot unplug not supported for this guest");
3077         }
3078     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3079         if (!mc->has_hotpluggable_cpus) {
3080             error_setg(errp, "CPU hot unplug not supported on this machine");
3081             return;
3082         }
3083         spapr_core_unplug_request(hotplug_dev, dev, errp);
3084     }
3085 }
3086 
3087 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3088                                           DeviceState *dev, Error **errp)
3089 {
3090     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3091         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3092     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3093         spapr_core_pre_plug(hotplug_dev, dev, errp);
3094     }
3095 }
3096 
3097 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3098                                                  DeviceState *dev)
3099 {
3100     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3101         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3102         return HOTPLUG_HANDLER(machine);
3103     }
3104     return NULL;
3105 }
3106 
3107 static CpuInstanceProperties
3108 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3109 {
3110     CPUArchId *core_slot;
3111     MachineClass *mc = MACHINE_GET_CLASS(machine);
3112 
3113     /* make sure possible_cpu are intialized */
3114     mc->possible_cpu_arch_ids(machine);
3115     /* get CPU core slot containing thread that matches cpu_index */
3116     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3117     assert(core_slot);
3118     return core_slot->props;
3119 }
3120 
3121 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3122 {
3123     int i;
3124     int spapr_max_cores = max_cpus / smp_threads;
3125     MachineClass *mc = MACHINE_GET_CLASS(machine);
3126 
3127     if (!mc->has_hotpluggable_cpus) {
3128         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3129     }
3130     if (machine->possible_cpus) {
3131         assert(machine->possible_cpus->len == spapr_max_cores);
3132         return machine->possible_cpus;
3133     }
3134 
3135     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3136                              sizeof(CPUArchId) * spapr_max_cores);
3137     machine->possible_cpus->len = spapr_max_cores;
3138     for (i = 0; i < machine->possible_cpus->len; i++) {
3139         int core_id = i * smp_threads;
3140 
3141         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3142         machine->possible_cpus->cpus[i].arch_id = core_id;
3143         machine->possible_cpus->cpus[i].props.has_core_id = true;
3144         machine->possible_cpus->cpus[i].props.core_id = core_id;
3145 
3146         /* default distribution of CPUs over NUMA nodes */
3147         if (nb_numa_nodes) {
3148             /* preset values but do not enable them i.e. 'has_node_id = false',
3149              * numa init code will enable them later if manual mapping wasn't
3150              * present on CLI */
3151             machine->possible_cpus->cpus[i].props.node_id =
3152                 core_id / smp_threads / smp_cores % nb_numa_nodes;
3153         }
3154     }
3155     return machine->possible_cpus;
3156 }
3157 
3158 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3159                                 uint64_t *buid, hwaddr *pio,
3160                                 hwaddr *mmio32, hwaddr *mmio64,
3161                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3162 {
3163     /*
3164      * New-style PHB window placement.
3165      *
3166      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3167      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3168      * windows.
3169      *
3170      * Some guest kernels can't work with MMIO windows above 1<<46
3171      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3172      *
3173      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3174      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3175      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3176      * 1TiB 64-bit MMIO windows for each PHB.
3177      */
3178     const uint64_t base_buid = 0x800000020000000ULL;
3179 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3180                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3181     int i;
3182 
3183     /* Sanity check natural alignments */
3184     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3185     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3186     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3187     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3188     /* Sanity check bounds */
3189     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3190                       SPAPR_PCI_MEM32_WIN_SIZE);
3191     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3192                       SPAPR_PCI_MEM64_WIN_SIZE);
3193 
3194     if (index >= SPAPR_MAX_PHBS) {
3195         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3196                    SPAPR_MAX_PHBS - 1);
3197         return;
3198     }
3199 
3200     *buid = base_buid + index;
3201     for (i = 0; i < n_dma; ++i) {
3202         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3203     }
3204 
3205     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3206     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3207     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3208 }
3209 
3210 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3211 {
3212     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3213 
3214     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3215 }
3216 
3217 static void spapr_ics_resend(XICSFabric *dev)
3218 {
3219     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3220 
3221     ics_resend(spapr->ics);
3222 }
3223 
3224 static ICPState *spapr_icp_get(XICSFabric *xi, int cpu_dt_id)
3225 {
3226     PowerPCCPU *cpu = ppc_get_vcpu_by_dt_id(cpu_dt_id);
3227 
3228     return cpu ? ICP(cpu->intc) : NULL;
3229 }
3230 
3231 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3232                                  Monitor *mon)
3233 {
3234     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3235     CPUState *cs;
3236 
3237     CPU_FOREACH(cs) {
3238         PowerPCCPU *cpu = POWERPC_CPU(cs);
3239 
3240         icp_pic_print_info(ICP(cpu->intc), mon);
3241     }
3242 
3243     ics_pic_print_info(spapr->ics, mon);
3244 }
3245 
3246 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3247 {
3248     MachineClass *mc = MACHINE_CLASS(oc);
3249     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3250     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3251     NMIClass *nc = NMI_CLASS(oc);
3252     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3253     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3254     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3255     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3256 
3257     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3258 
3259     /*
3260      * We set up the default / latest behaviour here.  The class_init
3261      * functions for the specific versioned machine types can override
3262      * these details for backwards compatibility
3263      */
3264     mc->init = ppc_spapr_init;
3265     mc->reset = ppc_spapr_reset;
3266     mc->block_default_type = IF_SCSI;
3267     mc->max_cpus = 1024;
3268     mc->no_parallel = 1;
3269     mc->default_boot_order = "";
3270     mc->default_ram_size = 512 * M_BYTE;
3271     mc->kvm_type = spapr_kvm_type;
3272     mc->has_dynamic_sysbus = true;
3273     mc->pci_allow_0_address = true;
3274     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3275     hc->pre_plug = spapr_machine_device_pre_plug;
3276     hc->plug = spapr_machine_device_plug;
3277     hc->unplug = spapr_machine_device_unplug;
3278     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3279     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3280     hc->unplug_request = spapr_machine_device_unplug_request;
3281 
3282     smc->dr_lmb_enabled = true;
3283     smc->tcg_default_cpu = "POWER8";
3284     mc->has_hotpluggable_cpus = true;
3285     fwc->get_dev_path = spapr_get_fw_dev_path;
3286     nc->nmi_monitor_handler = spapr_nmi;
3287     smc->phb_placement = spapr_phb_placement;
3288     vhc->hypercall = emulate_spapr_hypercall;
3289     vhc->hpt_mask = spapr_hpt_mask;
3290     vhc->map_hptes = spapr_map_hptes;
3291     vhc->unmap_hptes = spapr_unmap_hptes;
3292     vhc->store_hpte = spapr_store_hpte;
3293     vhc->get_patbe = spapr_get_patbe;
3294     xic->ics_get = spapr_ics_get;
3295     xic->ics_resend = spapr_ics_resend;
3296     xic->icp_get = spapr_icp_get;
3297     ispc->print_info = spapr_pic_print_info;
3298     /* Force NUMA node memory size to be a multiple of
3299      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3300      * in which LMBs are represented and hot-added
3301      */
3302     mc->numa_mem_align_shift = 28;
3303 }
3304 
3305 static const TypeInfo spapr_machine_info = {
3306     .name          = TYPE_SPAPR_MACHINE,
3307     .parent        = TYPE_MACHINE,
3308     .abstract      = true,
3309     .instance_size = sizeof(sPAPRMachineState),
3310     .instance_init = spapr_machine_initfn,
3311     .instance_finalize = spapr_machine_finalizefn,
3312     .class_size    = sizeof(sPAPRMachineClass),
3313     .class_init    = spapr_machine_class_init,
3314     .interfaces = (InterfaceInfo[]) {
3315         { TYPE_FW_PATH_PROVIDER },
3316         { TYPE_NMI },
3317         { TYPE_HOTPLUG_HANDLER },
3318         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3319         { TYPE_XICS_FABRIC },
3320         { TYPE_INTERRUPT_STATS_PROVIDER },
3321         { }
3322     },
3323 };
3324 
3325 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3326     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3327                                                     void *data)      \
3328     {                                                                \
3329         MachineClass *mc = MACHINE_CLASS(oc);                        \
3330         spapr_machine_##suffix##_class_options(mc);                  \
3331         if (latest) {                                                \
3332             mc->alias = "pseries";                                   \
3333             mc->is_default = 1;                                      \
3334         }                                                            \
3335     }                                                                \
3336     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3337     {                                                                \
3338         MachineState *machine = MACHINE(obj);                        \
3339         spapr_machine_##suffix##_instance_options(machine);          \
3340     }                                                                \
3341     static const TypeInfo spapr_machine_##suffix##_info = {          \
3342         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3343         .parent = TYPE_SPAPR_MACHINE,                                \
3344         .class_init = spapr_machine_##suffix##_class_init,           \
3345         .instance_init = spapr_machine_##suffix##_instance_init,     \
3346     };                                                               \
3347     static void spapr_machine_register_##suffix(void)                \
3348     {                                                                \
3349         type_register(&spapr_machine_##suffix##_info);               \
3350     }                                                                \
3351     type_init(spapr_machine_register_##suffix)
3352 
3353 /*
3354  * pseries-2.10
3355  */
3356 static void spapr_machine_2_10_instance_options(MachineState *machine)
3357 {
3358 }
3359 
3360 static void spapr_machine_2_10_class_options(MachineClass *mc)
3361 {
3362     /* Defaults for the latest behaviour inherited from the base class */
3363 }
3364 
3365 DEFINE_SPAPR_MACHINE(2_10, "2.10", true);
3366 
3367 /*
3368  * pseries-2.9
3369  */
3370 #define SPAPR_COMPAT_2_9                                               \
3371     HW_COMPAT_2_9
3372 
3373 static void spapr_machine_2_9_instance_options(MachineState *machine)
3374 {
3375     spapr_machine_2_10_instance_options(machine);
3376 }
3377 
3378 static void spapr_machine_2_9_class_options(MachineClass *mc)
3379 {
3380     spapr_machine_2_10_class_options(mc);
3381     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3382     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
3383 }
3384 
3385 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3386 
3387 /*
3388  * pseries-2.8
3389  */
3390 #define SPAPR_COMPAT_2_8                                        \
3391     HW_COMPAT_2_8                                               \
3392     {                                                           \
3393         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3394         .property = "pcie-extended-configuration-space",        \
3395         .value    = "off",                                      \
3396     },
3397 
3398 static void spapr_machine_2_8_instance_options(MachineState *machine)
3399 {
3400     spapr_machine_2_9_instance_options(machine);
3401 }
3402 
3403 static void spapr_machine_2_8_class_options(MachineClass *mc)
3404 {
3405     spapr_machine_2_9_class_options(mc);
3406     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3407     mc->numa_mem_align_shift = 23;
3408 }
3409 
3410 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3411 
3412 /*
3413  * pseries-2.7
3414  */
3415 #define SPAPR_COMPAT_2_7                            \
3416     HW_COMPAT_2_7                                   \
3417     {                                               \
3418         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3419         .property = "mem_win_size",                 \
3420         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3421     },                                              \
3422     {                                               \
3423         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3424         .property = "mem64_win_size",               \
3425         .value    = "0",                            \
3426     },                                              \
3427     {                                               \
3428         .driver = TYPE_POWERPC_CPU,                 \
3429         .property = "pre-2.8-migration",            \
3430         .value    = "on",                           \
3431     },                                              \
3432     {                                               \
3433         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3434         .property = "pre-2.8-migration",            \
3435         .value    = "on",                           \
3436     },
3437 
3438 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3439                               uint64_t *buid, hwaddr *pio,
3440                               hwaddr *mmio32, hwaddr *mmio64,
3441                               unsigned n_dma, uint32_t *liobns, Error **errp)
3442 {
3443     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3444     const uint64_t base_buid = 0x800000020000000ULL;
3445     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3446     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3447     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3448     const uint32_t max_index = 255;
3449     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3450 
3451     uint64_t ram_top = MACHINE(spapr)->ram_size;
3452     hwaddr phb0_base, phb_base;
3453     int i;
3454 
3455     /* Do we have hotpluggable memory? */
3456     if (MACHINE(spapr)->maxram_size > ram_top) {
3457         /* Can't just use maxram_size, because there may be an
3458          * alignment gap between normal and hotpluggable memory
3459          * regions */
3460         ram_top = spapr->hotplug_memory.base +
3461             memory_region_size(&spapr->hotplug_memory.mr);
3462     }
3463 
3464     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3465 
3466     if (index > max_index) {
3467         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3468                    max_index);
3469         return;
3470     }
3471 
3472     *buid = base_buid + index;
3473     for (i = 0; i < n_dma; ++i) {
3474         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3475     }
3476 
3477     phb_base = phb0_base + index * phb_spacing;
3478     *pio = phb_base + pio_offset;
3479     *mmio32 = phb_base + mmio_offset;
3480     /*
3481      * We don't set the 64-bit MMIO window, relying on the PHB's
3482      * fallback behaviour of automatically splitting a large "32-bit"
3483      * window into contiguous 32-bit and 64-bit windows
3484      */
3485 }
3486 
3487 static void spapr_machine_2_7_instance_options(MachineState *machine)
3488 {
3489     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3490 
3491     spapr_machine_2_8_instance_options(machine);
3492     spapr->use_hotplug_event_source = false;
3493 }
3494 
3495 static void spapr_machine_2_7_class_options(MachineClass *mc)
3496 {
3497     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3498 
3499     spapr_machine_2_8_class_options(mc);
3500     smc->tcg_default_cpu = "POWER7";
3501     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3502     smc->phb_placement = phb_placement_2_7;
3503 }
3504 
3505 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3506 
3507 /*
3508  * pseries-2.6
3509  */
3510 #define SPAPR_COMPAT_2_6 \
3511     HW_COMPAT_2_6 \
3512     { \
3513         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3514         .property = "ddw",\
3515         .value    = stringify(off),\
3516     },
3517 
3518 static void spapr_machine_2_6_instance_options(MachineState *machine)
3519 {
3520     spapr_machine_2_7_instance_options(machine);
3521 }
3522 
3523 static void spapr_machine_2_6_class_options(MachineClass *mc)
3524 {
3525     spapr_machine_2_7_class_options(mc);
3526     mc->has_hotpluggable_cpus = false;
3527     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3528 }
3529 
3530 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3531 
3532 /*
3533  * pseries-2.5
3534  */
3535 #define SPAPR_COMPAT_2_5 \
3536     HW_COMPAT_2_5 \
3537     { \
3538         .driver   = "spapr-vlan", \
3539         .property = "use-rx-buffer-pools", \
3540         .value    = "off", \
3541     },
3542 
3543 static void spapr_machine_2_5_instance_options(MachineState *machine)
3544 {
3545     spapr_machine_2_6_instance_options(machine);
3546 }
3547 
3548 static void spapr_machine_2_5_class_options(MachineClass *mc)
3549 {
3550     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3551 
3552     spapr_machine_2_6_class_options(mc);
3553     smc->use_ohci_by_default = true;
3554     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3555 }
3556 
3557 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3558 
3559 /*
3560  * pseries-2.4
3561  */
3562 #define SPAPR_COMPAT_2_4 \
3563         HW_COMPAT_2_4
3564 
3565 static void spapr_machine_2_4_instance_options(MachineState *machine)
3566 {
3567     spapr_machine_2_5_instance_options(machine);
3568 }
3569 
3570 static void spapr_machine_2_4_class_options(MachineClass *mc)
3571 {
3572     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3573 
3574     spapr_machine_2_5_class_options(mc);
3575     smc->dr_lmb_enabled = false;
3576     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3577 }
3578 
3579 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3580 
3581 /*
3582  * pseries-2.3
3583  */
3584 #define SPAPR_COMPAT_2_3 \
3585         HW_COMPAT_2_3 \
3586         {\
3587             .driver   = "spapr-pci-host-bridge",\
3588             .property = "dynamic-reconfiguration",\
3589             .value    = "off",\
3590         },
3591 
3592 static void spapr_machine_2_3_instance_options(MachineState *machine)
3593 {
3594     spapr_machine_2_4_instance_options(machine);
3595     savevm_skip_section_footers();
3596     global_state_set_optional();
3597     savevm_skip_configuration();
3598 }
3599 
3600 static void spapr_machine_2_3_class_options(MachineClass *mc)
3601 {
3602     spapr_machine_2_4_class_options(mc);
3603     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3604 }
3605 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3606 
3607 /*
3608  * pseries-2.2
3609  */
3610 
3611 #define SPAPR_COMPAT_2_2 \
3612         HW_COMPAT_2_2 \
3613         {\
3614             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3615             .property = "mem_win_size",\
3616             .value    = "0x20000000",\
3617         },
3618 
3619 static void spapr_machine_2_2_instance_options(MachineState *machine)
3620 {
3621     spapr_machine_2_3_instance_options(machine);
3622     machine->suppress_vmdesc = true;
3623 }
3624 
3625 static void spapr_machine_2_2_class_options(MachineClass *mc)
3626 {
3627     spapr_machine_2_3_class_options(mc);
3628     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3629 }
3630 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3631 
3632 /*
3633  * pseries-2.1
3634  */
3635 #define SPAPR_COMPAT_2_1 \
3636         HW_COMPAT_2_1
3637 
3638 static void spapr_machine_2_1_instance_options(MachineState *machine)
3639 {
3640     spapr_machine_2_2_instance_options(machine);
3641 }
3642 
3643 static void spapr_machine_2_1_class_options(MachineClass *mc)
3644 {
3645     spapr_machine_2_2_class_options(mc);
3646     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3647 }
3648 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3649 
3650 static void spapr_machine_register_types(void)
3651 {
3652     type_register_static(&spapr_machine_info);
3653 }
3654 
3655 type_init(spapr_machine_register_types)
3656