xref: /openbmc/qemu/hw/ppc/spapr.c (revision c95f6161de0b92efbdd64c9ddfdad58843b8760e)
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/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
46 #include "qom/cpu.h"
47 
48 #include "hw/boards.h"
49 #include "hw/ppc/ppc.h"
50 #include "hw/loader.h"
51 
52 #include "hw/ppc/fdt.h"
53 #include "hw/ppc/spapr.h"
54 #include "hw/ppc/spapr_vio.h"
55 #include "hw/pci-host/spapr.h"
56 #include "hw/ppc/xics.h"
57 #include "hw/pci/msi.h"
58 
59 #include "hw/pci/pci.h"
60 #include "hw/scsi/scsi.h"
61 #include "hw/virtio/virtio-scsi.h"
62 #include "hw/virtio/vhost-scsi-common.h"
63 
64 #include "exec/address-spaces.h"
65 #include "hw/usb.h"
66 #include "qemu/config-file.h"
67 #include "qemu/error-report.h"
68 #include "trace.h"
69 #include "hw/nmi.h"
70 #include "hw/intc/intc.h"
71 
72 #include "hw/compat.h"
73 #include "qemu/cutils.h"
74 #include "hw/ppc/spapr_cpu_core.h"
75 #include "qmp-commands.h"
76 
77 #include <libfdt.h>
78 
79 /* SLOF memory layout:
80  *
81  * SLOF raw image loaded at 0, copies its romfs right below the flat
82  * device-tree, then position SLOF itself 31M below that
83  *
84  * So we set FW_OVERHEAD to 40MB which should account for all of that
85  * and more
86  *
87  * We load our kernel at 4M, leaving space for SLOF initial image
88  */
89 #define FDT_MAX_SIZE            0x100000
90 #define RTAS_MAX_SIZE           0x10000
91 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
92 #define FW_MAX_SIZE             0x400000
93 #define FW_FILE_NAME            "slof.bin"
94 #define FW_OVERHEAD             0x2800000
95 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
96 
97 #define MIN_RMA_SLOF            128UL
98 
99 #define PHANDLE_XICP            0x00001111
100 
101 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
102 
103 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
104                                   const char *type_ics,
105                                   int nr_irqs, Error **errp)
106 {
107     Error *local_err = NULL;
108     Object *obj;
109 
110     obj = object_new(type_ics);
111     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
112     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
113                                    &error_abort);
114     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
115     if (local_err) {
116         goto error;
117     }
118     object_property_set_bool(obj, true, "realized", &local_err);
119     if (local_err) {
120         goto error;
121     }
122 
123     return ICS_SIMPLE(obj);
124 
125 error:
126     error_propagate(errp, local_err);
127     return NULL;
128 }
129 
130 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
131 {
132     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
133 
134     if (kvm_enabled()) {
135         if (machine_kernel_irqchip_allowed(machine) &&
136             !xics_kvm_init(spapr, errp)) {
137             spapr->icp_type = TYPE_KVM_ICP;
138             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
139         }
140         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
141             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
142             return;
143         }
144     }
145 
146     if (!spapr->ics) {
147         xics_spapr_init(spapr);
148         spapr->icp_type = TYPE_ICP;
149         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
150         if (!spapr->ics) {
151             return;
152         }
153     }
154 }
155 
156 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
157                                   int smt_threads)
158 {
159     int i, ret = 0;
160     uint32_t servers_prop[smt_threads];
161     uint32_t gservers_prop[smt_threads * 2];
162     int index = ppc_get_vcpu_dt_id(cpu);
163 
164     if (cpu->compat_pvr) {
165         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
166         if (ret < 0) {
167             return ret;
168         }
169     }
170 
171     /* Build interrupt servers and gservers properties */
172     for (i = 0; i < smt_threads; i++) {
173         servers_prop[i] = cpu_to_be32(index + i);
174         /* Hack, direct the group queues back to cpu 0 */
175         gservers_prop[i*2] = cpu_to_be32(index + i);
176         gservers_prop[i*2 + 1] = 0;
177     }
178     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
179                       servers_prop, sizeof(servers_prop));
180     if (ret < 0) {
181         return ret;
182     }
183     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
184                       gservers_prop, sizeof(gservers_prop));
185 
186     return ret;
187 }
188 
189 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
190 {
191     int index = ppc_get_vcpu_dt_id(cpu);
192     uint32_t associativity[] = {cpu_to_be32(0x5),
193                                 cpu_to_be32(0x0),
194                                 cpu_to_be32(0x0),
195                                 cpu_to_be32(0x0),
196                                 cpu_to_be32(cpu->node_id),
197                                 cpu_to_be32(index)};
198 
199     /* Advertise NUMA via ibm,associativity */
200     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
201                           sizeof(associativity));
202 }
203 
204 /* Populate the "ibm,pa-features" property */
205 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset,
206                                       bool legacy_guest)
207 {
208     uint8_t pa_features_206[] = { 6, 0,
209         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
210     uint8_t pa_features_207[] = { 24, 0,
211         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
212         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
213         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
214         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
215     uint8_t pa_features_300[] = { 66, 0,
216         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
217         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
218         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
219         /* 6: DS207 */
220         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
221         /* 16: Vector */
222         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
223         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
224         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
225         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
226         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
227         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
228         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
229         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
231         /* 42: PM, 44: PC RA, 46: SC vec'd */
232         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
233         /* 48: SIMD, 50: QP BFP, 52: String */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
235         /* 54: DecFP, 56: DecI, 58: SHA */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
237         /* 60: NM atomic, 62: RNG */
238         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
239     };
240     uint8_t *pa_features;
241     size_t pa_size;
242 
243     switch (POWERPC_MMU_VER(env->mmu_model)) {
244     case POWERPC_MMU_VER_2_06:
245         pa_features = pa_features_206;
246         pa_size = sizeof(pa_features_206);
247         break;
248     case POWERPC_MMU_VER_2_07:
249         pa_features = pa_features_207;
250         pa_size = sizeof(pa_features_207);
251         break;
252     case POWERPC_MMU_VER_3_00:
253         pa_features = pa_features_300;
254         pa_size = sizeof(pa_features_300);
255         break;
256     default:
257         return;
258     }
259 
260     if (env->ci_large_pages) {
261         /*
262          * Note: we keep CI large pages off by default because a 64K capable
263          * guest provisioned with large pages might otherwise try to map a qemu
264          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
265          * even if that qemu runs on a 4k host.
266          * We dd this bit back here if we are confident this is not an issue
267          */
268         pa_features[3] |= 0x20;
269     }
270     if (kvmppc_has_cap_htm() && pa_size > 24) {
271         pa_features[24] |= 0x80;    /* Transactional memory support */
272     }
273     if (legacy_guest && pa_size > 40) {
274         /* Workaround for broken kernels that attempt (guest) radix
275          * mode when they can't handle it, if they see the radix bit set
276          * in pa-features. So hide it from them. */
277         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
278     }
279 
280     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
281 }
282 
283 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
284 {
285     int ret = 0, offset, cpus_offset;
286     CPUState *cs;
287     char cpu_model[32];
288     int smt = kvmppc_smt_threads();
289     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
290 
291     CPU_FOREACH(cs) {
292         PowerPCCPU *cpu = POWERPC_CPU(cs);
293         CPUPPCState *env = &cpu->env;
294         DeviceClass *dc = DEVICE_GET_CLASS(cs);
295         int index = ppc_get_vcpu_dt_id(cpu);
296         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
297 
298         if ((index % smt) != 0) {
299             continue;
300         }
301 
302         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
303 
304         cpus_offset = fdt_path_offset(fdt, "/cpus");
305         if (cpus_offset < 0) {
306             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
307                                           "cpus");
308             if (cpus_offset < 0) {
309                 return cpus_offset;
310             }
311         }
312         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
313         if (offset < 0) {
314             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
315             if (offset < 0) {
316                 return offset;
317             }
318         }
319 
320         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
321                           pft_size_prop, sizeof(pft_size_prop));
322         if (ret < 0) {
323             return ret;
324         }
325 
326         if (nb_numa_nodes > 1) {
327             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
328             if (ret < 0) {
329                 return ret;
330             }
331         }
332 
333         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
334         if (ret < 0) {
335             return ret;
336         }
337 
338         spapr_populate_pa_features(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     int drc_index;
460     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
461     int i;
462 
463     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
464     if (drc) {
465         drc_index = spapr_drc_index(drc);
466         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
467     }
468 
469     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
470     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
471 
472     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
473     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
474                            env->dcache_line_size)));
475     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
476                            env->dcache_line_size)));
477     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
478                            env->icache_line_size)));
479     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
480                            env->icache_line_size)));
481 
482     if (pcc->l1_dcache_size) {
483         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
484                                pcc->l1_dcache_size)));
485     } else {
486         error_report("Warning: Unknown L1 dcache size for cpu");
487     }
488     if (pcc->l1_icache_size) {
489         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
490                                pcc->l1_icache_size)));
491     } else {
492         error_report("Warning: Unknown L1 icache size for cpu");
493     }
494 
495     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
496     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
497     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
498     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
499     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
500     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
501 
502     if (env->spr_cb[SPR_PURR].oea_read) {
503         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
504     }
505 
506     if (env->mmu_model & POWERPC_MMU_1TSEG) {
507         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
508                           segs, sizeof(segs))));
509     }
510 
511     /* Advertise VMX/VSX (vector extensions) if available
512      *   0 / no property == no vector extensions
513      *   1               == VMX / Altivec available
514      *   2               == VSX available */
515     if (env->insns_flags & PPC_ALTIVEC) {
516         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
517 
518         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
519     }
520 
521     /* Advertise DFP (Decimal Floating Point) if available
522      *   0 / no property == no DFP
523      *   1               == DFP available */
524     if (env->insns_flags2 & PPC2_DFP) {
525         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
526     }
527 
528     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
529                                                   sizeof(page_sizes_prop));
530     if (page_sizes_prop_size) {
531         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
532                           page_sizes_prop, page_sizes_prop_size)));
533     }
534 
535     spapr_populate_pa_features(env, fdt, offset, false);
536 
537     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
538                            cs->cpu_index / vcpus_per_socket)));
539 
540     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
541                       pft_size_prop, sizeof(pft_size_prop))));
542 
543     if (nb_numa_nodes > 1) {
544         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
545     }
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 
658             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
659             g_assert(drc);
660 
661             dynamic_memory[0] = cpu_to_be32(addr >> 32);
662             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
663             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
664             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
665             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
666             if (memory_region_present(get_system_memory(), addr)) {
667                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
668             } else {
669                 dynamic_memory[5] = cpu_to_be32(0);
670             }
671         } else {
672             /*
673              * LMB information for RMA, boot time RAM and gap b/n RAM and
674              * hotplug memory region -- all these are marked as reserved
675              * and as having no valid DRC.
676              */
677             dynamic_memory[0] = cpu_to_be32(addr >> 32);
678             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
679             dynamic_memory[2] = cpu_to_be32(0);
680             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
681             dynamic_memory[4] = cpu_to_be32(-1);
682             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
683                                             SPAPR_LMB_FLAGS_DRC_INVALID);
684         }
685 
686         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
687     }
688     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
689     if (ret < 0) {
690         goto out;
691     }
692 
693     /* ibm,associativity-lookup-arrays */
694     cur_index = int_buf;
695     int_buf[0] = cpu_to_be32(nr_nodes);
696     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
697     cur_index += 2;
698     for (i = 0; i < nr_nodes; i++) {
699         uint32_t associativity[] = {
700             cpu_to_be32(0x0),
701             cpu_to_be32(0x0),
702             cpu_to_be32(0x0),
703             cpu_to_be32(i)
704         };
705         memcpy(cur_index, associativity, sizeof(associativity));
706         cur_index += 4;
707     }
708     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
709             (cur_index - int_buf) * sizeof(uint32_t));
710 out:
711     g_free(int_buf);
712     return ret;
713 }
714 
715 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
716                                 sPAPROptionVector *ov5_updates)
717 {
718     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
719     int ret = 0, offset;
720 
721     /* Generate ibm,dynamic-reconfiguration-memory node if required */
722     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
723         g_assert(smc->dr_lmb_enabled);
724         ret = spapr_populate_drconf_memory(spapr, fdt);
725         if (ret) {
726             goto out;
727         }
728     }
729 
730     offset = fdt_path_offset(fdt, "/chosen");
731     if (offset < 0) {
732         offset = fdt_add_subnode(fdt, 0, "chosen");
733         if (offset < 0) {
734             return offset;
735         }
736     }
737     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
738                                  "ibm,architecture-vec-5");
739 
740 out:
741     return ret;
742 }
743 
744 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
745                                  target_ulong addr, target_ulong size,
746                                  sPAPROptionVector *ov5_updates)
747 {
748     void *fdt, *fdt_skel;
749     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
750 
751     size -= sizeof(hdr);
752 
753     /* Create sceleton */
754     fdt_skel = g_malloc0(size);
755     _FDT((fdt_create(fdt_skel, size)));
756     _FDT((fdt_begin_node(fdt_skel, "")));
757     _FDT((fdt_end_node(fdt_skel)));
758     _FDT((fdt_finish(fdt_skel)));
759     fdt = g_malloc0(size);
760     _FDT((fdt_open_into(fdt_skel, fdt, size)));
761     g_free(fdt_skel);
762 
763     /* Fixup cpu nodes */
764     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
765 
766     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
767         return -1;
768     }
769 
770     /* Pack resulting tree */
771     _FDT((fdt_pack(fdt)));
772 
773     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
774         trace_spapr_cas_failed(size);
775         return -1;
776     }
777 
778     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
779     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
780     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
781     g_free(fdt);
782 
783     return 0;
784 }
785 
786 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
787 {
788     int rtas;
789     GString *hypertas = g_string_sized_new(256);
790     GString *qemu_hypertas = g_string_sized_new(256);
791     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
792     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
793         memory_region_size(&spapr->hotplug_memory.mr);
794     uint32_t lrdr_capacity[] = {
795         cpu_to_be32(max_hotplug_addr >> 32),
796         cpu_to_be32(max_hotplug_addr & 0xffffffff),
797         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
798         cpu_to_be32(max_cpus / smp_threads),
799     };
800 
801     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
802 
803     /* hypertas */
804     add_str(hypertas, "hcall-pft");
805     add_str(hypertas, "hcall-term");
806     add_str(hypertas, "hcall-dabr");
807     add_str(hypertas, "hcall-interrupt");
808     add_str(hypertas, "hcall-tce");
809     add_str(hypertas, "hcall-vio");
810     add_str(hypertas, "hcall-splpar");
811     add_str(hypertas, "hcall-bulk");
812     add_str(hypertas, "hcall-set-mode");
813     add_str(hypertas, "hcall-sprg0");
814     add_str(hypertas, "hcall-copy");
815     add_str(hypertas, "hcall-debug");
816     add_str(qemu_hypertas, "hcall-memop1");
817 
818     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
819         add_str(hypertas, "hcall-multi-tce");
820     }
821     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
822                      hypertas->str, hypertas->len));
823     g_string_free(hypertas, TRUE);
824     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
825                      qemu_hypertas->str, qemu_hypertas->len));
826     g_string_free(qemu_hypertas, TRUE);
827 
828     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
829                      refpoints, sizeof(refpoints)));
830 
831     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
832                           RTAS_ERROR_LOG_MAX));
833     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
834                           RTAS_EVENT_SCAN_RATE));
835 
836     if (msi_nonbroken) {
837         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
838     }
839 
840     /*
841      * According to PAPR, rtas ibm,os-term does not guarantee a return
842      * back to the guest cpu.
843      *
844      * While an additional ibm,extended-os-term property indicates
845      * that rtas call return will always occur. Set this property.
846      */
847     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
848 
849     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
850                      lrdr_capacity, sizeof(lrdr_capacity)));
851 
852     spapr_dt_rtas_tokens(fdt, rtas);
853 }
854 
855 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
856  * that the guest may request and thus the valid values for bytes 24..26 of
857  * option vector 5: */
858 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
859 {
860     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
861 
862     char val[2 * 3] = {
863         24, 0x00, /* Hash/Radix, filled in below. */
864         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
865         26, 0x40, /* Radix options: GTSE == yes. */
866     };
867 
868     if (kvm_enabled()) {
869         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
870             val[1] = 0x80; /* OV5_MMU_BOTH */
871         } else if (kvmppc_has_cap_mmu_radix()) {
872             val[1] = 0x40; /* OV5_MMU_RADIX_300 */
873         } else {
874             val[1] = 0x00; /* Hash */
875         }
876     } else {
877         if (first_ppc_cpu->env.mmu_model & POWERPC_MMU_V3) {
878             /* V3 MMU supports both hash and radix (with dynamic switching) */
879             val[1] = 0xC0;
880         } else {
881             /* Otherwise we can only do hash */
882             val[1] = 0x00;
883         }
884     }
885     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
886                      val, sizeof(val)));
887 }
888 
889 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
890 {
891     MachineState *machine = MACHINE(spapr);
892     int chosen;
893     const char *boot_device = machine->boot_order;
894     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
895     size_t cb = 0;
896     char *bootlist = get_boot_devices_list(&cb, true);
897 
898     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
899 
900     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
901     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
902                           spapr->initrd_base));
903     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
904                           spapr->initrd_base + spapr->initrd_size));
905 
906     if (spapr->kernel_size) {
907         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
908                               cpu_to_be64(spapr->kernel_size) };
909 
910         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
911                          &kprop, sizeof(kprop)));
912         if (spapr->kernel_le) {
913             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
914         }
915     }
916     if (boot_menu) {
917         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
918     }
919     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
920     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
921     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
922 
923     if (cb && bootlist) {
924         int i;
925 
926         for (i = 0; i < cb; i++) {
927             if (bootlist[i] == '\n') {
928                 bootlist[i] = ' ';
929             }
930         }
931         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
932     }
933 
934     if (boot_device && strlen(boot_device)) {
935         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
936     }
937 
938     if (!spapr->has_graphics && stdout_path) {
939         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
940     }
941 
942     spapr_dt_ov5_platform_support(fdt, chosen);
943 
944     g_free(stdout_path);
945     g_free(bootlist);
946 }
947 
948 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
949 {
950     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
951      * KVM to work under pHyp with some guest co-operation */
952     int hypervisor;
953     uint8_t hypercall[16];
954 
955     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
956     /* indicate KVM hypercall interface */
957     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
958     if (kvmppc_has_cap_fixup_hcalls()) {
959         /*
960          * Older KVM versions with older guest kernels were broken
961          * with the magic page, don't allow the guest to map it.
962          */
963         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
964                                   sizeof(hypercall))) {
965             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
966                              hypercall, sizeof(hypercall)));
967         }
968     }
969 }
970 
971 static void *spapr_build_fdt(sPAPRMachineState *spapr,
972                              hwaddr rtas_addr,
973                              hwaddr rtas_size)
974 {
975     MachineState *machine = MACHINE(qdev_get_machine());
976     MachineClass *mc = MACHINE_GET_CLASS(machine);
977     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
978     int ret;
979     void *fdt;
980     sPAPRPHBState *phb;
981     char *buf;
982     int smt = kvmppc_smt_threads();
983 
984     fdt = g_malloc0(FDT_MAX_SIZE);
985     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
986 
987     /* Root node */
988     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
989     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
990     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
991 
992     /*
993      * Add info to guest to indentify which host is it being run on
994      * and what is the uuid of the guest
995      */
996     if (kvmppc_get_host_model(&buf)) {
997         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
998         g_free(buf);
999     }
1000     if (kvmppc_get_host_serial(&buf)) {
1001         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1002         g_free(buf);
1003     }
1004 
1005     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1006 
1007     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1008     if (qemu_uuid_set) {
1009         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1010     }
1011     g_free(buf);
1012 
1013     if (qemu_get_vm_name()) {
1014         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1015                                 qemu_get_vm_name()));
1016     }
1017 
1018     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1019     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1020 
1021     /* /interrupt controller */
1022     spapr_dt_xics(DIV_ROUND_UP(max_cpus * smt, smp_threads), fdt, PHANDLE_XICP);
1023 
1024     ret = spapr_populate_memory(spapr, fdt);
1025     if (ret < 0) {
1026         error_report("couldn't setup memory nodes in fdt");
1027         exit(1);
1028     }
1029 
1030     /* /vdevice */
1031     spapr_dt_vdevice(spapr->vio_bus, fdt);
1032 
1033     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1034         ret = spapr_rng_populate_dt(fdt);
1035         if (ret < 0) {
1036             error_report("could not set up rng device in the fdt");
1037             exit(1);
1038         }
1039     }
1040 
1041     QLIST_FOREACH(phb, &spapr->phbs, list) {
1042         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1043         if (ret < 0) {
1044             error_report("couldn't setup PCI devices in fdt");
1045             exit(1);
1046         }
1047     }
1048 
1049     /* cpus */
1050     spapr_populate_cpus_dt_node(fdt, spapr);
1051 
1052     if (smc->dr_lmb_enabled) {
1053         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1054     }
1055 
1056     if (mc->has_hotpluggable_cpus) {
1057         int offset = fdt_path_offset(fdt, "/cpus");
1058         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1059                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1060         if (ret < 0) {
1061             error_report("Couldn't set up CPU DR device tree properties");
1062             exit(1);
1063         }
1064     }
1065 
1066     /* /event-sources */
1067     spapr_dt_events(spapr, fdt);
1068 
1069     /* /rtas */
1070     spapr_dt_rtas(spapr, fdt);
1071 
1072     /* /chosen */
1073     spapr_dt_chosen(spapr, fdt);
1074 
1075     /* /hypervisor */
1076     if (kvm_enabled()) {
1077         spapr_dt_hypervisor(spapr, fdt);
1078     }
1079 
1080     /* Build memory reserve map */
1081     if (spapr->kernel_size) {
1082         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1083     }
1084     if (spapr->initrd_size) {
1085         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1086     }
1087 
1088     /* ibm,client-architecture-support updates */
1089     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1090     if (ret < 0) {
1091         error_report("couldn't setup CAS properties fdt");
1092         exit(1);
1093     }
1094 
1095     return fdt;
1096 }
1097 
1098 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1099 {
1100     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1101 }
1102 
1103 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1104                                     PowerPCCPU *cpu)
1105 {
1106     CPUPPCState *env = &cpu->env;
1107 
1108     /* The TCG path should also be holding the BQL at this point */
1109     g_assert(qemu_mutex_iothread_locked());
1110 
1111     if (msr_pr) {
1112         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1113         env->gpr[3] = H_PRIVILEGE;
1114     } else {
1115         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1116     }
1117 }
1118 
1119 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1120 {
1121     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1122 
1123     return spapr->patb_entry;
1124 }
1125 
1126 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1127 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1128 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1129 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1130 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1131 
1132 /*
1133  * Get the fd to access the kernel htab, re-opening it if necessary
1134  */
1135 static int get_htab_fd(sPAPRMachineState *spapr)
1136 {
1137     if (spapr->htab_fd >= 0) {
1138         return spapr->htab_fd;
1139     }
1140 
1141     spapr->htab_fd = kvmppc_get_htab_fd(false);
1142     if (spapr->htab_fd < 0) {
1143         error_report("Unable to open fd for reading hash table from KVM: %s",
1144                      strerror(errno));
1145     }
1146 
1147     return spapr->htab_fd;
1148 }
1149 
1150 void close_htab_fd(sPAPRMachineState *spapr)
1151 {
1152     if (spapr->htab_fd >= 0) {
1153         close(spapr->htab_fd);
1154     }
1155     spapr->htab_fd = -1;
1156 }
1157 
1158 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1159 {
1160     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1161 
1162     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1163 }
1164 
1165 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1166                                                 hwaddr ptex, int n)
1167 {
1168     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1169     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1170 
1171     if (!spapr->htab) {
1172         /*
1173          * HTAB is controlled by KVM. Fetch into temporary buffer
1174          */
1175         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1176         kvmppc_read_hptes(hptes, ptex, n);
1177         return hptes;
1178     }
1179 
1180     /*
1181      * HTAB is controlled by QEMU. Just point to the internally
1182      * accessible PTEG.
1183      */
1184     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1185 }
1186 
1187 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1188                               const ppc_hash_pte64_t *hptes,
1189                               hwaddr ptex, int n)
1190 {
1191     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1192 
1193     if (!spapr->htab) {
1194         g_free((void *)hptes);
1195     }
1196 
1197     /* Nothing to do for qemu managed HPT */
1198 }
1199 
1200 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1201                              uint64_t pte0, uint64_t pte1)
1202 {
1203     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1204     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1205 
1206     if (!spapr->htab) {
1207         kvmppc_write_hpte(ptex, pte0, pte1);
1208     } else {
1209         stq_p(spapr->htab + offset, pte0);
1210         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1211     }
1212 }
1213 
1214 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1215 {
1216     int shift;
1217 
1218     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1219      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1220      * that's much more than is needed for Linux guests */
1221     shift = ctz64(pow2ceil(ramsize)) - 7;
1222     shift = MAX(shift, 18); /* Minimum architected size */
1223     shift = MIN(shift, 46); /* Maximum architected size */
1224     return shift;
1225 }
1226 
1227 void spapr_free_hpt(sPAPRMachineState *spapr)
1228 {
1229     g_free(spapr->htab);
1230     spapr->htab = NULL;
1231     spapr->htab_shift = 0;
1232     close_htab_fd(spapr);
1233 }
1234 
1235 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1236                                  Error **errp)
1237 {
1238     long rc;
1239 
1240     /* Clean up any HPT info from a previous boot */
1241     spapr_free_hpt(spapr);
1242 
1243     rc = kvmppc_reset_htab(shift);
1244     if (rc < 0) {
1245         /* kernel-side HPT needed, but couldn't allocate one */
1246         error_setg_errno(errp, errno,
1247                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1248                          shift);
1249         /* This is almost certainly fatal, but if the caller really
1250          * wants to carry on with shift == 0, it's welcome to try */
1251     } else if (rc > 0) {
1252         /* kernel-side HPT allocated */
1253         if (rc != shift) {
1254             error_setg(errp,
1255                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1256                        shift, rc);
1257         }
1258 
1259         spapr->htab_shift = shift;
1260         spapr->htab = NULL;
1261     } else {
1262         /* kernel-side HPT not needed, allocate in userspace instead */
1263         size_t size = 1ULL << shift;
1264         int i;
1265 
1266         spapr->htab = qemu_memalign(size, size);
1267         if (!spapr->htab) {
1268             error_setg_errno(errp, errno,
1269                              "Could not allocate HPT of order %d", shift);
1270             return;
1271         }
1272 
1273         memset(spapr->htab, 0, size);
1274         spapr->htab_shift = shift;
1275 
1276         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1277             DIRTY_HPTE(HPTE(spapr->htab, i));
1278         }
1279     }
1280 }
1281 
1282 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1283 {
1284     spapr_reallocate_hpt(spapr,
1285                      spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size),
1286                      &error_fatal);
1287     if (spapr->vrma_adjust) {
1288         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1289                                           spapr->htab_shift);
1290     }
1291     /* We're setting up a hash table, so that means we're not radix */
1292     spapr->patb_entry = 0;
1293 }
1294 
1295 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1296 {
1297     bool matched = false;
1298 
1299     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1300         matched = true;
1301     }
1302 
1303     if (!matched) {
1304         error_report("Device %s is not supported by this machine yet.",
1305                      qdev_fw_name(DEVICE(sbdev)));
1306         exit(1);
1307     }
1308 }
1309 
1310 static void ppc_spapr_reset(void)
1311 {
1312     MachineState *machine = MACHINE(qdev_get_machine());
1313     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1314     PowerPCCPU *first_ppc_cpu;
1315     uint32_t rtas_limit;
1316     hwaddr rtas_addr, fdt_addr;
1317     void *fdt;
1318     int rc;
1319 
1320     /* Check for unknown sysbus devices */
1321     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1322 
1323     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1324         /* If using KVM with radix mode available, VCPUs can be started
1325          * without a HPT because KVM will start them in radix mode.
1326          * Set the GR bit in PATB so that we know there is no HPT. */
1327         spapr->patb_entry = PATBE1_GR;
1328     } else {
1329         spapr->patb_entry = 0;
1330         spapr_setup_hpt_and_vrma(spapr);
1331     }
1332 
1333     qemu_devices_reset();
1334 
1335     /*
1336      * We place the device tree and RTAS just below either the top of the RMA,
1337      * or just below 2GB, whichever is lowere, so that it can be
1338      * processed with 32-bit real mode code if necessary
1339      */
1340     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1341     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1342     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1343 
1344     /* if this reset wasn't generated by CAS, we should reset our
1345      * negotiated options and start from scratch */
1346     if (!spapr->cas_reboot) {
1347         spapr_ovec_cleanup(spapr->ov5_cas);
1348         spapr->ov5_cas = spapr_ovec_new();
1349 
1350         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
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     if (spapr->patb_entry) {
1449         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1450         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1451         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1452 
1453         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1454         if (err) {
1455             error_report("Process table config unsupported by the host");
1456             return -EINVAL;
1457         }
1458     }
1459 
1460     return err;
1461 }
1462 
1463 static bool version_before_3(void *opaque, int version_id)
1464 {
1465     return version_id < 3;
1466 }
1467 
1468 static bool spapr_ov5_cas_needed(void *opaque)
1469 {
1470     sPAPRMachineState *spapr = opaque;
1471     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1472     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1473     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1474     bool cas_needed;
1475 
1476     /* Prior to the introduction of sPAPROptionVector, we had two option
1477      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1478      * Both of these options encode machine topology into the device-tree
1479      * in such a way that the now-booted OS should still be able to interact
1480      * appropriately with QEMU regardless of what options were actually
1481      * negotiatied on the source side.
1482      *
1483      * As such, we can avoid migrating the CAS-negotiated options if these
1484      * are the only options available on the current machine/platform.
1485      * Since these are the only options available for pseries-2.7 and
1486      * earlier, this allows us to maintain old->new/new->old migration
1487      * compatibility.
1488      *
1489      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1490      * via default pseries-2.8 machines and explicit command-line parameters.
1491      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1492      * of the actual CAS-negotiated values to continue working properly. For
1493      * example, availability of memory unplug depends on knowing whether
1494      * OV5_HP_EVT was negotiated via CAS.
1495      *
1496      * Thus, for any cases where the set of available CAS-negotiatable
1497      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1498      * include the CAS-negotiated options in the migration stream.
1499      */
1500     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1501     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1502 
1503     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1504      * the mask itself since in the future it's possible "legacy" bits may be
1505      * removed via machine options, which could generate a false positive
1506      * that breaks migration.
1507      */
1508     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1509     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1510 
1511     spapr_ovec_cleanup(ov5_mask);
1512     spapr_ovec_cleanup(ov5_legacy);
1513     spapr_ovec_cleanup(ov5_removed);
1514 
1515     return cas_needed;
1516 }
1517 
1518 static const VMStateDescription vmstate_spapr_ov5_cas = {
1519     .name = "spapr_option_vector_ov5_cas",
1520     .version_id = 1,
1521     .minimum_version_id = 1,
1522     .needed = spapr_ov5_cas_needed,
1523     .fields = (VMStateField[]) {
1524         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1525                                  vmstate_spapr_ovec, sPAPROptionVector),
1526         VMSTATE_END_OF_LIST()
1527     },
1528 };
1529 
1530 static bool spapr_patb_entry_needed(void *opaque)
1531 {
1532     sPAPRMachineState *spapr = opaque;
1533 
1534     return !!spapr->patb_entry;
1535 }
1536 
1537 static const VMStateDescription vmstate_spapr_patb_entry = {
1538     .name = "spapr_patb_entry",
1539     .version_id = 1,
1540     .minimum_version_id = 1,
1541     .needed = spapr_patb_entry_needed,
1542     .fields = (VMStateField[]) {
1543         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1544         VMSTATE_END_OF_LIST()
1545     },
1546 };
1547 
1548 static const VMStateDescription vmstate_spapr = {
1549     .name = "spapr",
1550     .version_id = 3,
1551     .minimum_version_id = 1,
1552     .post_load = spapr_post_load,
1553     .fields = (VMStateField[]) {
1554         /* used to be @next_irq */
1555         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1556 
1557         /* RTC offset */
1558         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1559 
1560         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1561         VMSTATE_END_OF_LIST()
1562     },
1563     .subsections = (const VMStateDescription*[]) {
1564         &vmstate_spapr_ov5_cas,
1565         &vmstate_spapr_patb_entry,
1566         NULL
1567     }
1568 };
1569 
1570 static int htab_save_setup(QEMUFile *f, void *opaque)
1571 {
1572     sPAPRMachineState *spapr = opaque;
1573 
1574     /* "Iteration" header */
1575     if (!spapr->htab_shift) {
1576         qemu_put_be32(f, -1);
1577     } else {
1578         qemu_put_be32(f, spapr->htab_shift);
1579     }
1580 
1581     if (spapr->htab) {
1582         spapr->htab_save_index = 0;
1583         spapr->htab_first_pass = true;
1584     } else {
1585         if (spapr->htab_shift) {
1586             assert(kvm_enabled());
1587         }
1588     }
1589 
1590 
1591     return 0;
1592 }
1593 
1594 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1595                                  int64_t max_ns)
1596 {
1597     bool has_timeout = max_ns != -1;
1598     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1599     int index = spapr->htab_save_index;
1600     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1601 
1602     assert(spapr->htab_first_pass);
1603 
1604     do {
1605         int chunkstart;
1606 
1607         /* Consume invalid HPTEs */
1608         while ((index < htabslots)
1609                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1610             CLEAN_HPTE(HPTE(spapr->htab, index));
1611             index++;
1612         }
1613 
1614         /* Consume valid HPTEs */
1615         chunkstart = index;
1616         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1617                && HPTE_VALID(HPTE(spapr->htab, index))) {
1618             CLEAN_HPTE(HPTE(spapr->htab, index));
1619             index++;
1620         }
1621 
1622         if (index > chunkstart) {
1623             int n_valid = index - chunkstart;
1624 
1625             qemu_put_be32(f, chunkstart);
1626             qemu_put_be16(f, n_valid);
1627             qemu_put_be16(f, 0);
1628             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1629                             HASH_PTE_SIZE_64 * n_valid);
1630 
1631             if (has_timeout &&
1632                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1633                 break;
1634             }
1635         }
1636     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1637 
1638     if (index >= htabslots) {
1639         assert(index == htabslots);
1640         index = 0;
1641         spapr->htab_first_pass = false;
1642     }
1643     spapr->htab_save_index = index;
1644 }
1645 
1646 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1647                                 int64_t max_ns)
1648 {
1649     bool final = max_ns < 0;
1650     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1651     int examined = 0, sent = 0;
1652     int index = spapr->htab_save_index;
1653     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1654 
1655     assert(!spapr->htab_first_pass);
1656 
1657     do {
1658         int chunkstart, invalidstart;
1659 
1660         /* Consume non-dirty HPTEs */
1661         while ((index < htabslots)
1662                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1663             index++;
1664             examined++;
1665         }
1666 
1667         chunkstart = index;
1668         /* Consume valid dirty HPTEs */
1669         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1670                && HPTE_DIRTY(HPTE(spapr->htab, index))
1671                && HPTE_VALID(HPTE(spapr->htab, index))) {
1672             CLEAN_HPTE(HPTE(spapr->htab, index));
1673             index++;
1674             examined++;
1675         }
1676 
1677         invalidstart = index;
1678         /* Consume invalid dirty HPTEs */
1679         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1680                && HPTE_DIRTY(HPTE(spapr->htab, index))
1681                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1682             CLEAN_HPTE(HPTE(spapr->htab, index));
1683             index++;
1684             examined++;
1685         }
1686 
1687         if (index > chunkstart) {
1688             int n_valid = invalidstart - chunkstart;
1689             int n_invalid = index - invalidstart;
1690 
1691             qemu_put_be32(f, chunkstart);
1692             qemu_put_be16(f, n_valid);
1693             qemu_put_be16(f, n_invalid);
1694             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1695                             HASH_PTE_SIZE_64 * n_valid);
1696             sent += index - chunkstart;
1697 
1698             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1699                 break;
1700             }
1701         }
1702 
1703         if (examined >= htabslots) {
1704             break;
1705         }
1706 
1707         if (index >= htabslots) {
1708             assert(index == htabslots);
1709             index = 0;
1710         }
1711     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1712 
1713     if (index >= htabslots) {
1714         assert(index == htabslots);
1715         index = 0;
1716     }
1717 
1718     spapr->htab_save_index = index;
1719 
1720     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1721 }
1722 
1723 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1724 #define MAX_KVM_BUF_SIZE    2048
1725 
1726 static int htab_save_iterate(QEMUFile *f, void *opaque)
1727 {
1728     sPAPRMachineState *spapr = opaque;
1729     int fd;
1730     int rc = 0;
1731 
1732     /* Iteration header */
1733     if (!spapr->htab_shift) {
1734         qemu_put_be32(f, -1);
1735         return 0;
1736     } else {
1737         qemu_put_be32(f, 0);
1738     }
1739 
1740     if (!spapr->htab) {
1741         assert(kvm_enabled());
1742 
1743         fd = get_htab_fd(spapr);
1744         if (fd < 0) {
1745             return fd;
1746         }
1747 
1748         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1749         if (rc < 0) {
1750             return rc;
1751         }
1752     } else  if (spapr->htab_first_pass) {
1753         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1754     } else {
1755         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1756     }
1757 
1758     /* End marker */
1759     qemu_put_be32(f, 0);
1760     qemu_put_be16(f, 0);
1761     qemu_put_be16(f, 0);
1762 
1763     return rc;
1764 }
1765 
1766 static int htab_save_complete(QEMUFile *f, void *opaque)
1767 {
1768     sPAPRMachineState *spapr = opaque;
1769     int fd;
1770 
1771     /* Iteration header */
1772     if (!spapr->htab_shift) {
1773         qemu_put_be32(f, -1);
1774         return 0;
1775     } else {
1776         qemu_put_be32(f, 0);
1777     }
1778 
1779     if (!spapr->htab) {
1780         int rc;
1781 
1782         assert(kvm_enabled());
1783 
1784         fd = get_htab_fd(spapr);
1785         if (fd < 0) {
1786             return fd;
1787         }
1788 
1789         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1790         if (rc < 0) {
1791             return rc;
1792         }
1793     } else {
1794         if (spapr->htab_first_pass) {
1795             htab_save_first_pass(f, spapr, -1);
1796         }
1797         htab_save_later_pass(f, spapr, -1);
1798     }
1799 
1800     /* End marker */
1801     qemu_put_be32(f, 0);
1802     qemu_put_be16(f, 0);
1803     qemu_put_be16(f, 0);
1804 
1805     return 0;
1806 }
1807 
1808 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1809 {
1810     sPAPRMachineState *spapr = opaque;
1811     uint32_t section_hdr;
1812     int fd = -1;
1813 
1814     if (version_id < 1 || version_id > 1) {
1815         error_report("htab_load() bad version");
1816         return -EINVAL;
1817     }
1818 
1819     section_hdr = qemu_get_be32(f);
1820 
1821     if (section_hdr == -1) {
1822         spapr_free_hpt(spapr);
1823         return 0;
1824     }
1825 
1826     if (section_hdr) {
1827         Error *local_err = NULL;
1828 
1829         /* First section gives the htab size */
1830         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1831         if (local_err) {
1832             error_report_err(local_err);
1833             return -EINVAL;
1834         }
1835         return 0;
1836     }
1837 
1838     if (!spapr->htab) {
1839         assert(kvm_enabled());
1840 
1841         fd = kvmppc_get_htab_fd(true);
1842         if (fd < 0) {
1843             error_report("Unable to open fd to restore KVM hash table: %s",
1844                          strerror(errno));
1845         }
1846     }
1847 
1848     while (true) {
1849         uint32_t index;
1850         uint16_t n_valid, n_invalid;
1851 
1852         index = qemu_get_be32(f);
1853         n_valid = qemu_get_be16(f);
1854         n_invalid = qemu_get_be16(f);
1855 
1856         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1857             /* End of Stream */
1858             break;
1859         }
1860 
1861         if ((index + n_valid + n_invalid) >
1862             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1863             /* Bad index in stream */
1864             error_report(
1865                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1866                 index, n_valid, n_invalid, spapr->htab_shift);
1867             return -EINVAL;
1868         }
1869 
1870         if (spapr->htab) {
1871             if (n_valid) {
1872                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1873                                 HASH_PTE_SIZE_64 * n_valid);
1874             }
1875             if (n_invalid) {
1876                 memset(HPTE(spapr->htab, index + n_valid), 0,
1877                        HASH_PTE_SIZE_64 * n_invalid);
1878             }
1879         } else {
1880             int rc;
1881 
1882             assert(fd >= 0);
1883 
1884             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1885             if (rc < 0) {
1886                 return rc;
1887             }
1888         }
1889     }
1890 
1891     if (!spapr->htab) {
1892         assert(fd >= 0);
1893         close(fd);
1894     }
1895 
1896     return 0;
1897 }
1898 
1899 static void htab_cleanup(void *opaque)
1900 {
1901     sPAPRMachineState *spapr = opaque;
1902 
1903     close_htab_fd(spapr);
1904 }
1905 
1906 static SaveVMHandlers savevm_htab_handlers = {
1907     .save_live_setup = htab_save_setup,
1908     .save_live_iterate = htab_save_iterate,
1909     .save_live_complete_precopy = htab_save_complete,
1910     .cleanup = htab_cleanup,
1911     .load_state = htab_load,
1912 };
1913 
1914 static void spapr_boot_set(void *opaque, const char *boot_device,
1915                            Error **errp)
1916 {
1917     MachineState *machine = MACHINE(qdev_get_machine());
1918     machine->boot_order = g_strdup(boot_device);
1919 }
1920 
1921 /*
1922  * Reset routine for LMB DR devices.
1923  *
1924  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1925  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1926  * when it walks all its children devices. LMB devices reset occurs
1927  * as part of spapr_ppc_reset().
1928  */
1929 static void spapr_drc_reset(void *opaque)
1930 {
1931     sPAPRDRConnector *drc = opaque;
1932     DeviceState *d = DEVICE(drc);
1933 
1934     if (d) {
1935         device_reset(d);
1936     }
1937 }
1938 
1939 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1940 {
1941     MachineState *machine = MACHINE(spapr);
1942     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1943     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1944     int i;
1945 
1946     for (i = 0; i < nr_lmbs; i++) {
1947         sPAPRDRConnector *drc;
1948         uint64_t addr;
1949 
1950         addr = i * lmb_size + spapr->hotplug_memory.base;
1951         drc = spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
1952                                      addr/lmb_size);
1953         qemu_register_reset(spapr_drc_reset, drc);
1954     }
1955 }
1956 
1957 /*
1958  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1959  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1960  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1961  */
1962 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1963 {
1964     int i;
1965 
1966     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1967         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1968                    " is not aligned to %llu MiB",
1969                    machine->ram_size,
1970                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1971         return;
1972     }
1973 
1974     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1975         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1976                    " is not aligned to %llu MiB",
1977                    machine->ram_size,
1978                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1979         return;
1980     }
1981 
1982     for (i = 0; i < nb_numa_nodes; i++) {
1983         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1984             error_setg(errp,
1985                        "Node %d memory size 0x%" PRIx64
1986                        " is not aligned to %llu MiB",
1987                        i, numa_info[i].node_mem,
1988                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1989             return;
1990         }
1991     }
1992 }
1993 
1994 /* find cpu slot in machine->possible_cpus by core_id */
1995 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1996 {
1997     int index = id / smp_threads;
1998 
1999     if (index >= ms->possible_cpus->len) {
2000         return NULL;
2001     }
2002     if (idx) {
2003         *idx = index;
2004     }
2005     return &ms->possible_cpus->cpus[index];
2006 }
2007 
2008 static void spapr_init_cpus(sPAPRMachineState *spapr)
2009 {
2010     MachineState *machine = MACHINE(spapr);
2011     MachineClass *mc = MACHINE_GET_CLASS(machine);
2012     char *type = spapr_get_cpu_core_type(machine->cpu_model);
2013     int smt = kvmppc_smt_threads();
2014     const CPUArchIdList *possible_cpus;
2015     int boot_cores_nr = smp_cpus / smp_threads;
2016     int i;
2017 
2018     if (!type) {
2019         error_report("Unable to find sPAPR CPU Core definition");
2020         exit(1);
2021     }
2022 
2023     possible_cpus = mc->possible_cpu_arch_ids(machine);
2024     if (mc->has_hotpluggable_cpus) {
2025         if (smp_cpus % smp_threads) {
2026             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2027                          smp_cpus, smp_threads);
2028             exit(1);
2029         }
2030         if (max_cpus % smp_threads) {
2031             error_report("max_cpus (%u) must be multiple of threads (%u)",
2032                          max_cpus, smp_threads);
2033             exit(1);
2034         }
2035     } else {
2036         if (max_cpus != smp_cpus) {
2037             error_report("This machine version does not support CPU hotplug");
2038             exit(1);
2039         }
2040         boot_cores_nr = possible_cpus->len;
2041     }
2042 
2043     for (i = 0; i < possible_cpus->len; i++) {
2044         int core_id = i * smp_threads;
2045 
2046         if (mc->has_hotpluggable_cpus) {
2047             sPAPRDRConnector *drc =
2048                 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2049                                        (core_id / smp_threads) * smt);
2050 
2051             qemu_register_reset(spapr_drc_reset, drc);
2052         }
2053 
2054         if (i < boot_cores_nr) {
2055             Object *core  = object_new(type);
2056             int nr_threads = smp_threads;
2057 
2058             /* Handle the partially filled core for older machine types */
2059             if ((i + 1) * smp_threads >= smp_cpus) {
2060                 nr_threads = smp_cpus - i * smp_threads;
2061             }
2062 
2063             object_property_set_int(core, nr_threads, "nr-threads",
2064                                     &error_fatal);
2065             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2066                                     &error_fatal);
2067             object_property_set_bool(core, true, "realized", &error_fatal);
2068         }
2069     }
2070     g_free(type);
2071 }
2072 
2073 /* pSeries LPAR / sPAPR hardware init */
2074 static void ppc_spapr_init(MachineState *machine)
2075 {
2076     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2077     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2078     const char *kernel_filename = machine->kernel_filename;
2079     const char *initrd_filename = machine->initrd_filename;
2080     PCIHostState *phb;
2081     int i;
2082     MemoryRegion *sysmem = get_system_memory();
2083     MemoryRegion *ram = g_new(MemoryRegion, 1);
2084     MemoryRegion *rma_region;
2085     void *rma = NULL;
2086     hwaddr rma_alloc_size;
2087     hwaddr node0_size = spapr_node0_size();
2088     long load_limit, fw_size;
2089     char *filename;
2090 
2091     msi_nonbroken = true;
2092 
2093     QLIST_INIT(&spapr->phbs);
2094     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2095 
2096     /* Allocate RMA if necessary */
2097     rma_alloc_size = kvmppc_alloc_rma(&rma);
2098 
2099     if (rma_alloc_size == -1) {
2100         error_report("Unable to create RMA");
2101         exit(1);
2102     }
2103 
2104     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2105         spapr->rma_size = rma_alloc_size;
2106     } else {
2107         spapr->rma_size = node0_size;
2108 
2109         /* With KVM, we don't actually know whether KVM supports an
2110          * unbounded RMA (PR KVM) or is limited by the hash table size
2111          * (HV KVM using VRMA), so we always assume the latter
2112          *
2113          * In that case, we also limit the initial allocations for RTAS
2114          * etc... to 256M since we have no way to know what the VRMA size
2115          * is going to be as it depends on the size of the hash table
2116          * isn't determined yet.
2117          */
2118         if (kvm_enabled()) {
2119             spapr->vrma_adjust = 1;
2120             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2121         }
2122 
2123         /* Actually we don't support unbounded RMA anymore since we
2124          * added proper emulation of HV mode. The max we can get is
2125          * 16G which also happens to be what we configure for PAPR
2126          * mode so make sure we don't do anything bigger than that
2127          */
2128         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2129     }
2130 
2131     if (spapr->rma_size > node0_size) {
2132         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2133                      spapr->rma_size);
2134         exit(1);
2135     }
2136 
2137     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2138     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2139 
2140     /* Set up Interrupt Controller before we create the VCPUs */
2141     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2142 
2143     /* Set up containers for ibm,client-set-architecture negotiated options */
2144     spapr->ov5 = spapr_ovec_new();
2145     spapr->ov5_cas = spapr_ovec_new();
2146 
2147     if (smc->dr_lmb_enabled) {
2148         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2149         spapr_validate_node_memory(machine, &error_fatal);
2150     }
2151 
2152     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2153     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2154         /* KVM and TCG always allow GTSE with radix... */
2155         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2156     }
2157     /* ... but not with hash (currently). */
2158 
2159     /* advertise support for dedicated HP event source to guests */
2160     if (spapr->use_hotplug_event_source) {
2161         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2162     }
2163 
2164     /* init CPUs */
2165     if (machine->cpu_model == NULL) {
2166         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2167     }
2168 
2169     spapr_cpu_parse_features(spapr);
2170 
2171     spapr_init_cpus(spapr);
2172 
2173     if (kvm_enabled()) {
2174         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2175         kvmppc_enable_logical_ci_hcalls();
2176         kvmppc_enable_set_mode_hcall();
2177 
2178         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2179         kvmppc_enable_clear_ref_mod_hcalls();
2180     }
2181 
2182     /* allocate RAM */
2183     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2184                                          machine->ram_size);
2185     memory_region_add_subregion(sysmem, 0, ram);
2186 
2187     if (rma_alloc_size && rma) {
2188         rma_region = g_new(MemoryRegion, 1);
2189         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2190                                    rma_alloc_size, rma);
2191         vmstate_register_ram_global(rma_region);
2192         memory_region_add_subregion(sysmem, 0, rma_region);
2193     }
2194 
2195     /* initialize hotplug memory address space */
2196     if (machine->ram_size < machine->maxram_size) {
2197         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2198         /*
2199          * Limit the number of hotpluggable memory slots to half the number
2200          * slots that KVM supports, leaving the other half for PCI and other
2201          * devices. However ensure that number of slots doesn't drop below 32.
2202          */
2203         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2204                            SPAPR_MAX_RAM_SLOTS;
2205 
2206         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2207             max_memslots = SPAPR_MAX_RAM_SLOTS;
2208         }
2209         if (machine->ram_slots > max_memslots) {
2210             error_report("Specified number of memory slots %"
2211                          PRIu64" exceeds max supported %d",
2212                          machine->ram_slots, max_memslots);
2213             exit(1);
2214         }
2215 
2216         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2217                                               SPAPR_HOTPLUG_MEM_ALIGN);
2218         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2219                            "hotplug-memory", hotplug_mem_size);
2220         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2221                                     &spapr->hotplug_memory.mr);
2222     }
2223 
2224     if (smc->dr_lmb_enabled) {
2225         spapr_create_lmb_dr_connectors(spapr);
2226     }
2227 
2228     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2229     if (!filename) {
2230         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2231         exit(1);
2232     }
2233     spapr->rtas_size = get_image_size(filename);
2234     if (spapr->rtas_size < 0) {
2235         error_report("Could not get size of LPAR rtas '%s'", filename);
2236         exit(1);
2237     }
2238     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2239     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2240         error_report("Could not load LPAR rtas '%s'", filename);
2241         exit(1);
2242     }
2243     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2244         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2245                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2246         exit(1);
2247     }
2248     g_free(filename);
2249 
2250     /* Set up RTAS event infrastructure */
2251     spapr_events_init(spapr);
2252 
2253     /* Set up the RTC RTAS interfaces */
2254     spapr_rtc_create(spapr);
2255 
2256     /* Set up VIO bus */
2257     spapr->vio_bus = spapr_vio_bus_init();
2258 
2259     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2260         if (serial_hds[i]) {
2261             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2262         }
2263     }
2264 
2265     /* We always have at least the nvram device on VIO */
2266     spapr_create_nvram(spapr);
2267 
2268     /* Set up PCI */
2269     spapr_pci_rtas_init();
2270 
2271     phb = spapr_create_phb(spapr, 0);
2272 
2273     for (i = 0; i < nb_nics; i++) {
2274         NICInfo *nd = &nd_table[i];
2275 
2276         if (!nd->model) {
2277             nd->model = g_strdup("ibmveth");
2278         }
2279 
2280         if (strcmp(nd->model, "ibmveth") == 0) {
2281             spapr_vlan_create(spapr->vio_bus, nd);
2282         } else {
2283             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2284         }
2285     }
2286 
2287     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2288         spapr_vscsi_create(spapr->vio_bus);
2289     }
2290 
2291     /* Graphics */
2292     if (spapr_vga_init(phb->bus, &error_fatal)) {
2293         spapr->has_graphics = true;
2294         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2295     }
2296 
2297     if (machine->usb) {
2298         if (smc->use_ohci_by_default) {
2299             pci_create_simple(phb->bus, -1, "pci-ohci");
2300         } else {
2301             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2302         }
2303 
2304         if (spapr->has_graphics) {
2305             USBBus *usb_bus = usb_bus_find(-1);
2306 
2307             usb_create_simple(usb_bus, "usb-kbd");
2308             usb_create_simple(usb_bus, "usb-mouse");
2309         }
2310     }
2311 
2312     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2313         error_report(
2314             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2315             MIN_RMA_SLOF);
2316         exit(1);
2317     }
2318 
2319     if (kernel_filename) {
2320         uint64_t lowaddr = 0;
2321 
2322         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2323                                       NULL, NULL, &lowaddr, NULL, 1,
2324                                       PPC_ELF_MACHINE, 0, 0);
2325         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2326             spapr->kernel_size = load_elf(kernel_filename,
2327                                           translate_kernel_address, NULL, NULL,
2328                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2329                                           0, 0);
2330             spapr->kernel_le = spapr->kernel_size > 0;
2331         }
2332         if (spapr->kernel_size < 0) {
2333             error_report("error loading %s: %s", kernel_filename,
2334                          load_elf_strerror(spapr->kernel_size));
2335             exit(1);
2336         }
2337 
2338         /* load initrd */
2339         if (initrd_filename) {
2340             /* Try to locate the initrd in the gap between the kernel
2341              * and the firmware. Add a bit of space just in case
2342              */
2343             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2344                                   + 0x1ffff) & ~0xffff;
2345             spapr->initrd_size = load_image_targphys(initrd_filename,
2346                                                      spapr->initrd_base,
2347                                                      load_limit
2348                                                      - spapr->initrd_base);
2349             if (spapr->initrd_size < 0) {
2350                 error_report("could not load initial ram disk '%s'",
2351                              initrd_filename);
2352                 exit(1);
2353             }
2354         }
2355     }
2356 
2357     if (bios_name == NULL) {
2358         bios_name = FW_FILE_NAME;
2359     }
2360     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2361     if (!filename) {
2362         error_report("Could not find LPAR firmware '%s'", bios_name);
2363         exit(1);
2364     }
2365     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2366     if (fw_size <= 0) {
2367         error_report("Could not load LPAR firmware '%s'", filename);
2368         exit(1);
2369     }
2370     g_free(filename);
2371 
2372     /* FIXME: Should register things through the MachineState's qdev
2373      * interface, this is a legacy from the sPAPREnvironment structure
2374      * which predated MachineState but had a similar function */
2375     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2376     register_savevm_live(NULL, "spapr/htab", -1, 1,
2377                          &savevm_htab_handlers, spapr);
2378 
2379     qemu_register_boot_set(spapr_boot_set, spapr);
2380 
2381     if (kvm_enabled()) {
2382         /* to stop and start vmclock */
2383         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2384                                          &spapr->tb);
2385 
2386         kvmppc_spapr_enable_inkernel_multitce();
2387     }
2388 }
2389 
2390 static int spapr_kvm_type(const char *vm_type)
2391 {
2392     if (!vm_type) {
2393         return 0;
2394     }
2395 
2396     if (!strcmp(vm_type, "HV")) {
2397         return 1;
2398     }
2399 
2400     if (!strcmp(vm_type, "PR")) {
2401         return 2;
2402     }
2403 
2404     error_report("Unknown kvm-type specified '%s'", vm_type);
2405     exit(1);
2406 }
2407 
2408 /*
2409  * Implementation of an interface to adjust firmware path
2410  * for the bootindex property handling.
2411  */
2412 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2413                                    DeviceState *dev)
2414 {
2415 #define CAST(type, obj, name) \
2416     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2417     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2418     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2419     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2420 
2421     if (d) {
2422         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2423         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2424         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2425 
2426         if (spapr) {
2427             /*
2428              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2429              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2430              * in the top 16 bits of the 64-bit LUN
2431              */
2432             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2433             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2434                                    (uint64_t)id << 48);
2435         } else if (virtio) {
2436             /*
2437              * We use SRP luns of the form 01000000 | (target << 8) | lun
2438              * in the top 32 bits of the 64-bit LUN
2439              * Note: the quote above is from SLOF and it is wrong,
2440              * the actual binding is:
2441              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2442              */
2443             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2444             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2445                                    (uint64_t)id << 32);
2446         } else if (usb) {
2447             /*
2448              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2449              * in the top 32 bits of the 64-bit LUN
2450              */
2451             unsigned usb_port = atoi(usb->port->path);
2452             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2453             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2454                                    (uint64_t)id << 32);
2455         }
2456     }
2457 
2458     /*
2459      * SLOF probes the USB devices, and if it recognizes that the device is a
2460      * storage device, it changes its name to "storage" instead of "usb-host",
2461      * and additionally adds a child node for the SCSI LUN, so the correct
2462      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2463      */
2464     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2465         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2466         if (usb_host_dev_is_scsi_storage(usbdev)) {
2467             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2468         }
2469     }
2470 
2471     if (phb) {
2472         /* Replace "pci" with "pci@800000020000000" */
2473         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2474     }
2475 
2476     if (vsc) {
2477         /* Same logic as virtio above */
2478         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2479         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2480     }
2481 
2482     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2483         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2484         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2485         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2486     }
2487 
2488     return NULL;
2489 }
2490 
2491 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2492 {
2493     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2494 
2495     return g_strdup(spapr->kvm_type);
2496 }
2497 
2498 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2499 {
2500     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2501 
2502     g_free(spapr->kvm_type);
2503     spapr->kvm_type = g_strdup(value);
2504 }
2505 
2506 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2507 {
2508     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2509 
2510     return spapr->use_hotplug_event_source;
2511 }
2512 
2513 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2514                                             Error **errp)
2515 {
2516     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2517 
2518     spapr->use_hotplug_event_source = value;
2519 }
2520 
2521 static void spapr_machine_initfn(Object *obj)
2522 {
2523     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2524 
2525     spapr->htab_fd = -1;
2526     spapr->use_hotplug_event_source = true;
2527     object_property_add_str(obj, "kvm-type",
2528                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2529     object_property_set_description(obj, "kvm-type",
2530                                     "Specifies the KVM virtualization mode (HV, PR)",
2531                                     NULL);
2532     object_property_add_bool(obj, "modern-hotplug-events",
2533                             spapr_get_modern_hotplug_events,
2534                             spapr_set_modern_hotplug_events,
2535                             NULL);
2536     object_property_set_description(obj, "modern-hotplug-events",
2537                                     "Use dedicated hotplug event mechanism in"
2538                                     " place of standard EPOW events when possible"
2539                                     " (required for memory hot-unplug support)",
2540                                     NULL);
2541 
2542     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
2543                             "Maximum permitted CPU compatibility mode",
2544                             &error_fatal);
2545 }
2546 
2547 static void spapr_machine_finalizefn(Object *obj)
2548 {
2549     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2550 
2551     g_free(spapr->kvm_type);
2552 }
2553 
2554 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2555 {
2556     cpu_synchronize_state(cs);
2557     ppc_cpu_do_system_reset(cs);
2558 }
2559 
2560 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2561 {
2562     CPUState *cs;
2563 
2564     CPU_FOREACH(cs) {
2565         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2566     }
2567 }
2568 
2569 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2570                            uint32_t node, bool dedicated_hp_event_source,
2571                            Error **errp)
2572 {
2573     sPAPRDRConnector *drc;
2574     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2575     int i, fdt_offset, fdt_size;
2576     void *fdt;
2577     uint64_t addr = addr_start;
2578 
2579     for (i = 0; i < nr_lmbs; i++) {
2580         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2581                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2582         g_assert(drc);
2583 
2584         fdt = create_device_tree(&fdt_size);
2585         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2586                                                 SPAPR_MEMORY_BLOCK_SIZE);
2587 
2588         spapr_drc_attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2589         addr += SPAPR_MEMORY_BLOCK_SIZE;
2590         if (!dev->hotplugged) {
2591             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2592             /* guests expect coldplugged LMBs to be pre-allocated */
2593             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2594             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2595         }
2596     }
2597     /* send hotplug notification to the
2598      * guest only in case of hotplugged memory
2599      */
2600     if (dev->hotplugged) {
2601         if (dedicated_hp_event_source) {
2602             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2603                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2604             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2605                                                    nr_lmbs,
2606                                                    spapr_drc_index(drc));
2607         } else {
2608             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2609                                            nr_lmbs);
2610         }
2611     }
2612 }
2613 
2614 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2615                               uint32_t node, Error **errp)
2616 {
2617     Error *local_err = NULL;
2618     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2619     PCDIMMDevice *dimm = PC_DIMM(dev);
2620     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2621     MemoryRegion *mr = ddc->get_memory_region(dimm);
2622     uint64_t align = memory_region_get_alignment(mr);
2623     uint64_t size = memory_region_size(mr);
2624     uint64_t addr;
2625 
2626     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2627     if (local_err) {
2628         goto out;
2629     }
2630 
2631     addr = object_property_get_uint(OBJECT(dimm),
2632                                     PC_DIMM_ADDR_PROP, &local_err);
2633     if (local_err) {
2634         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2635         goto out;
2636     }
2637 
2638     spapr_add_lmbs(dev, addr, size, node,
2639                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2640                    &error_abort);
2641 
2642 out:
2643     error_propagate(errp, local_err);
2644 }
2645 
2646 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2647                                   Error **errp)
2648 {
2649     PCDIMMDevice *dimm = PC_DIMM(dev);
2650     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2651     MemoryRegion *mr = ddc->get_memory_region(dimm);
2652     uint64_t size = memory_region_size(mr);
2653     char *mem_dev;
2654 
2655     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2656         error_setg(errp, "Hotplugged memory size must be a multiple of "
2657                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2658         return;
2659     }
2660 
2661     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2662     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2663         error_setg(errp, "Memory backend has bad page size. "
2664                    "Use 'memory-backend-file' with correct mem-path.");
2665         goto out;
2666     }
2667 
2668 out:
2669     g_free(mem_dev);
2670 }
2671 
2672 struct sPAPRDIMMState {
2673     PCDIMMDevice *dimm;
2674     uint32_t nr_lmbs;
2675     QTAILQ_ENTRY(sPAPRDIMMState) next;
2676 };
2677 
2678 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
2679                                                        PCDIMMDevice *dimm)
2680 {
2681     sPAPRDIMMState *dimm_state = NULL;
2682 
2683     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
2684         if (dimm_state->dimm == dimm) {
2685             break;
2686         }
2687     }
2688     return dimm_state;
2689 }
2690 
2691 static void spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
2692                                            sPAPRDIMMState *dimm_state)
2693 {
2694     g_assert(!spapr_pending_dimm_unplugs_find(spapr, dimm_state->dimm));
2695     QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, dimm_state, next);
2696 }
2697 
2698 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
2699                                               sPAPRDIMMState *dimm_state)
2700 {
2701     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
2702     g_free(dimm_state);
2703 }
2704 
2705 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
2706                                                         PCDIMMDevice *dimm)
2707 {
2708     sPAPRDRConnector *drc;
2709     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2710     MemoryRegion *mr = ddc->get_memory_region(dimm);
2711     uint64_t size = memory_region_size(mr);
2712     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2713     uint32_t avail_lmbs = 0;
2714     uint64_t addr_start, addr;
2715     int i;
2716     sPAPRDIMMState *ds;
2717 
2718     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2719                                          &error_abort);
2720 
2721     addr = addr_start;
2722     for (i = 0; i < nr_lmbs; i++) {
2723         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2724                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2725         g_assert(drc);
2726         if (drc->dev) {
2727             avail_lmbs++;
2728         }
2729         addr += SPAPR_MEMORY_BLOCK_SIZE;
2730     }
2731 
2732     ds = g_malloc0(sizeof(sPAPRDIMMState));
2733     ds->nr_lmbs = avail_lmbs;
2734     ds->dimm = dimm;
2735     spapr_pending_dimm_unplugs_add(ms, ds);
2736     return ds;
2737 }
2738 
2739 /* Callback to be called during DRC release. */
2740 void spapr_lmb_release(DeviceState *dev)
2741 {
2742     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
2743     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
2744     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
2745 
2746     /* This information will get lost if a migration occurs
2747      * during the unplug process. In this case recover it. */
2748     if (ds == NULL) {
2749         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
2750         /* The DRC being examined by the caller at least must be counted */
2751         g_assert(ds->nr_lmbs);
2752     }
2753 
2754     if (--ds->nr_lmbs) {
2755         return;
2756     }
2757 
2758     spapr_pending_dimm_unplugs_remove(spapr, ds);
2759 
2760     /*
2761      * Now that all the LMBs have been removed by the guest, call the
2762      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2763      */
2764     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2765 }
2766 
2767 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2768                                 Error **errp)
2769 {
2770     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2771     PCDIMMDevice *dimm = PC_DIMM(dev);
2772     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2773     MemoryRegion *mr = ddc->get_memory_region(dimm);
2774 
2775     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2776     object_unparent(OBJECT(dev));
2777 }
2778 
2779 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2780                                         DeviceState *dev, Error **errp)
2781 {
2782     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
2783     Error *local_err = NULL;
2784     PCDIMMDevice *dimm = PC_DIMM(dev);
2785     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2786     MemoryRegion *mr = ddc->get_memory_region(dimm);
2787     uint64_t size = memory_region_size(mr);
2788     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2789     uint64_t addr_start, addr;
2790     int i;
2791     sPAPRDRConnector *drc;
2792     sPAPRDIMMState *ds;
2793 
2794     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2795                                          &local_err);
2796     if (local_err) {
2797         goto out;
2798     }
2799 
2800     ds = g_malloc0(sizeof(sPAPRDIMMState));
2801     ds->nr_lmbs = nr_lmbs;
2802     ds->dimm = dimm;
2803     spapr_pending_dimm_unplugs_add(spapr, ds);
2804 
2805     addr = addr_start;
2806     for (i = 0; i < nr_lmbs; i++) {
2807         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2808                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2809         g_assert(drc);
2810 
2811         spapr_drc_detach(drc, dev, errp);
2812         addr += SPAPR_MEMORY_BLOCK_SIZE;
2813     }
2814 
2815     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2816                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2817     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2818                                               nr_lmbs, spapr_drc_index(drc));
2819 out:
2820     error_propagate(errp, local_err);
2821 }
2822 
2823 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2824                                     sPAPRMachineState *spapr)
2825 {
2826     PowerPCCPU *cpu = POWERPC_CPU(cs);
2827     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2828     int id = ppc_get_vcpu_dt_id(cpu);
2829     void *fdt;
2830     int offset, fdt_size;
2831     char *nodename;
2832 
2833     fdt = create_device_tree(&fdt_size);
2834     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2835     offset = fdt_add_subnode(fdt, 0, nodename);
2836 
2837     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2838     g_free(nodename);
2839 
2840     *fdt_offset = offset;
2841     return fdt;
2842 }
2843 
2844 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2845                               Error **errp)
2846 {
2847     MachineState *ms = MACHINE(qdev_get_machine());
2848     CPUCore *cc = CPU_CORE(dev);
2849     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2850 
2851     assert(core_slot);
2852     core_slot->cpu = NULL;
2853     object_unparent(OBJECT(dev));
2854 }
2855 
2856 /* Callback to be called during DRC release. */
2857 void spapr_core_release(DeviceState *dev)
2858 {
2859     HotplugHandler *hotplug_ctrl;
2860 
2861     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2862     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2863 }
2864 
2865 static
2866 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2867                                Error **errp)
2868 {
2869     int index;
2870     sPAPRDRConnector *drc;
2871     Error *local_err = NULL;
2872     CPUCore *cc = CPU_CORE(dev);
2873     int smt = kvmppc_smt_threads();
2874 
2875     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2876         error_setg(errp, "Unable to find CPU core with core-id: %d",
2877                    cc->core_id);
2878         return;
2879     }
2880     if (index == 0) {
2881         error_setg(errp, "Boot CPU core may not be unplugged");
2882         return;
2883     }
2884 
2885     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
2886     g_assert(drc);
2887 
2888     spapr_drc_detach(drc, dev, &local_err);
2889     if (local_err) {
2890         error_propagate(errp, local_err);
2891         return;
2892     }
2893 
2894     spapr_hotplug_req_remove_by_index(drc);
2895 }
2896 
2897 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2898                             Error **errp)
2899 {
2900     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2901     MachineClass *mc = MACHINE_GET_CLASS(spapr);
2902     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2903     CPUCore *cc = CPU_CORE(dev);
2904     CPUState *cs = CPU(core->threads);
2905     sPAPRDRConnector *drc;
2906     Error *local_err = NULL;
2907     void *fdt = NULL;
2908     int fdt_offset = 0;
2909     int smt = kvmppc_smt_threads();
2910     CPUArchId *core_slot;
2911     int index;
2912 
2913     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2914     if (!core_slot) {
2915         error_setg(errp, "Unable to find CPU core with core-id: %d",
2916                    cc->core_id);
2917         return;
2918     }
2919     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
2920 
2921     g_assert(drc || !mc->has_hotpluggable_cpus);
2922 
2923     /*
2924      * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2925      * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2926      */
2927     if (dev->hotplugged) {
2928         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2929     }
2930 
2931     if (drc) {
2932         spapr_drc_attach(drc, dev, fdt, fdt_offset, !dev->hotplugged,
2933                          &local_err);
2934         if (local_err) {
2935             g_free(fdt);
2936             error_propagate(errp, local_err);
2937             return;
2938         }
2939     }
2940 
2941     if (dev->hotplugged) {
2942         /*
2943          * Send hotplug notification interrupt to the guest only in case
2944          * of hotplugged CPUs.
2945          */
2946         spapr_hotplug_req_add_by_index(drc);
2947     } else {
2948         /*
2949          * Set the right DRC states for cold plugged CPU.
2950          */
2951         if (drc) {
2952             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2953             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2954             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2955         }
2956     }
2957     core_slot->cpu = OBJECT(dev);
2958 }
2959 
2960 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2961                                 Error **errp)
2962 {
2963     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2964     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2965     Error *local_err = NULL;
2966     CPUCore *cc = CPU_CORE(dev);
2967     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2968     const char *type = object_get_typename(OBJECT(dev));
2969     CPUArchId *core_slot;
2970     int index;
2971 
2972     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2973         error_setg(&local_err, "CPU hotplug not supported for this machine");
2974         goto out;
2975     }
2976 
2977     if (strcmp(base_core_type, type)) {
2978         error_setg(&local_err, "CPU core type should be %s", base_core_type);
2979         goto out;
2980     }
2981 
2982     if (cc->core_id % smp_threads) {
2983         error_setg(&local_err, "invalid core id %d", cc->core_id);
2984         goto out;
2985     }
2986 
2987     /*
2988      * In general we should have homogeneous threads-per-core, but old
2989      * (pre hotplug support) machine types allow the last core to have
2990      * reduced threads as a compatibility hack for when we allowed
2991      * total vcpus not a multiple of threads-per-core.
2992      */
2993     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
2994         error_setg(errp, "invalid nr-threads %d, must be %d",
2995                    cc->nr_threads, smp_threads);
2996         return;
2997     }
2998 
2999     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3000     if (!core_slot) {
3001         error_setg(&local_err, "core id %d out of range", cc->core_id);
3002         goto out;
3003     }
3004 
3005     if (core_slot->cpu) {
3006         error_setg(&local_err, "core %d already populated", cc->core_id);
3007         goto out;
3008     }
3009 
3010     numa_cpu_pre_plug(core_slot, dev, &local_err);
3011 
3012 out:
3013     g_free(base_core_type);
3014     error_propagate(errp, local_err);
3015 }
3016 
3017 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3018                                       DeviceState *dev, Error **errp)
3019 {
3020     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
3021 
3022     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3023         int node;
3024 
3025         if (!smc->dr_lmb_enabled) {
3026             error_setg(errp, "Memory hotplug not supported for this machine");
3027             return;
3028         }
3029         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3030         if (*errp) {
3031             return;
3032         }
3033         if (node < 0 || node >= MAX_NODES) {
3034             error_setg(errp, "Invaild node %d", node);
3035             return;
3036         }
3037 
3038         /*
3039          * Currently PowerPC kernel doesn't allow hot-adding memory to
3040          * memory-less node, but instead will silently add the memory
3041          * to the first node that has some memory. This causes two
3042          * unexpected behaviours for the user.
3043          *
3044          * - Memory gets hotplugged to a different node than what the user
3045          *   specified.
3046          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3047          *   to memory-less node, a reboot will set things accordingly
3048          *   and the previously hotplugged memory now ends in the right node.
3049          *   This appears as if some memory moved from one node to another.
3050          *
3051          * So until kernel starts supporting memory hotplug to memory-less
3052          * nodes, just prevent such attempts upfront in QEMU.
3053          */
3054         if (nb_numa_nodes && !numa_info[node].node_mem) {
3055             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3056                        node);
3057             return;
3058         }
3059 
3060         spapr_memory_plug(hotplug_dev, dev, node, errp);
3061     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3062         spapr_core_plug(hotplug_dev, dev, errp);
3063     }
3064 }
3065 
3066 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3067                                       DeviceState *dev, Error **errp)
3068 {
3069     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
3070     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
3071 
3072     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3073         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3074             spapr_memory_unplug(hotplug_dev, dev, errp);
3075         } else {
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(hotplug_dev, dev, errp);
3084     }
3085 }
3086 
3087 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3088                                                 DeviceState *dev, Error **errp)
3089 {
3090     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
3091     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
3092 
3093     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3094         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3095             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3096         } else {
3097             /* NOTE: this means there is a window after guest reset, prior to
3098              * CAS negotiation, where unplug requests will fail due to the
3099              * capability not being detected yet. This is a bit different than
3100              * the case with PCI unplug, where the events will be queued and
3101              * eventually handled by the guest after boot
3102              */
3103             error_setg(errp, "Memory hot unplug not supported for this guest");
3104         }
3105     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3106         if (!mc->has_hotpluggable_cpus) {
3107             error_setg(errp, "CPU hot unplug not supported on this machine");
3108             return;
3109         }
3110         spapr_core_unplug_request(hotplug_dev, dev, errp);
3111     }
3112 }
3113 
3114 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3115                                           DeviceState *dev, Error **errp)
3116 {
3117     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3118         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3119     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3120         spapr_core_pre_plug(hotplug_dev, dev, errp);
3121     }
3122 }
3123 
3124 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3125                                                  DeviceState *dev)
3126 {
3127     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3128         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3129         return HOTPLUG_HANDLER(machine);
3130     }
3131     return NULL;
3132 }
3133 
3134 static CpuInstanceProperties
3135 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3136 {
3137     CPUArchId *core_slot;
3138     MachineClass *mc = MACHINE_GET_CLASS(machine);
3139 
3140     /* make sure possible_cpu are intialized */
3141     mc->possible_cpu_arch_ids(machine);
3142     /* get CPU core slot containing thread that matches cpu_index */
3143     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3144     assert(core_slot);
3145     return core_slot->props;
3146 }
3147 
3148 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3149 {
3150     int i;
3151     int spapr_max_cores = max_cpus / smp_threads;
3152     MachineClass *mc = MACHINE_GET_CLASS(machine);
3153 
3154     if (!mc->has_hotpluggable_cpus) {
3155         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3156     }
3157     if (machine->possible_cpus) {
3158         assert(machine->possible_cpus->len == spapr_max_cores);
3159         return machine->possible_cpus;
3160     }
3161 
3162     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3163                              sizeof(CPUArchId) * spapr_max_cores);
3164     machine->possible_cpus->len = spapr_max_cores;
3165     for (i = 0; i < machine->possible_cpus->len; i++) {
3166         int core_id = i * smp_threads;
3167 
3168         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3169         machine->possible_cpus->cpus[i].arch_id = core_id;
3170         machine->possible_cpus->cpus[i].props.has_core_id = true;
3171         machine->possible_cpus->cpus[i].props.core_id = core_id;
3172 
3173         /* default distribution of CPUs over NUMA nodes */
3174         if (nb_numa_nodes) {
3175             /* preset values but do not enable them i.e. 'has_node_id = false',
3176              * numa init code will enable them later if manual mapping wasn't
3177              * present on CLI */
3178             machine->possible_cpus->cpus[i].props.node_id =
3179                 core_id / smp_threads / smp_cores % nb_numa_nodes;
3180         }
3181     }
3182     return machine->possible_cpus;
3183 }
3184 
3185 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3186                                 uint64_t *buid, hwaddr *pio,
3187                                 hwaddr *mmio32, hwaddr *mmio64,
3188                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3189 {
3190     /*
3191      * New-style PHB window placement.
3192      *
3193      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3194      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3195      * windows.
3196      *
3197      * Some guest kernels can't work with MMIO windows above 1<<46
3198      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3199      *
3200      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3201      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3202      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3203      * 1TiB 64-bit MMIO windows for each PHB.
3204      */
3205     const uint64_t base_buid = 0x800000020000000ULL;
3206 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3207                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3208     int i;
3209 
3210     /* Sanity check natural alignments */
3211     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3212     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3213     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3214     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3215     /* Sanity check bounds */
3216     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3217                       SPAPR_PCI_MEM32_WIN_SIZE);
3218     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3219                       SPAPR_PCI_MEM64_WIN_SIZE);
3220 
3221     if (index >= SPAPR_MAX_PHBS) {
3222         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3223                    SPAPR_MAX_PHBS - 1);
3224         return;
3225     }
3226 
3227     *buid = base_buid + index;
3228     for (i = 0; i < n_dma; ++i) {
3229         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3230     }
3231 
3232     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3233     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3234     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3235 }
3236 
3237 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3238 {
3239     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3240 
3241     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3242 }
3243 
3244 static void spapr_ics_resend(XICSFabric *dev)
3245 {
3246     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3247 
3248     ics_resend(spapr->ics);
3249 }
3250 
3251 static ICPState *spapr_icp_get(XICSFabric *xi, int cpu_dt_id)
3252 {
3253     PowerPCCPU *cpu = ppc_get_vcpu_by_dt_id(cpu_dt_id);
3254 
3255     return cpu ? ICP(cpu->intc) : NULL;
3256 }
3257 
3258 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3259                                  Monitor *mon)
3260 {
3261     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3262     CPUState *cs;
3263 
3264     CPU_FOREACH(cs) {
3265         PowerPCCPU *cpu = POWERPC_CPU(cs);
3266 
3267         icp_pic_print_info(ICP(cpu->intc), mon);
3268     }
3269 
3270     ics_pic_print_info(spapr->ics, mon);
3271 }
3272 
3273 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3274 {
3275     MachineClass *mc = MACHINE_CLASS(oc);
3276     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3277     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3278     NMIClass *nc = NMI_CLASS(oc);
3279     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3280     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3281     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3282     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3283 
3284     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3285 
3286     /*
3287      * We set up the default / latest behaviour here.  The class_init
3288      * functions for the specific versioned machine types can override
3289      * these details for backwards compatibility
3290      */
3291     mc->init = ppc_spapr_init;
3292     mc->reset = ppc_spapr_reset;
3293     mc->block_default_type = IF_SCSI;
3294     mc->max_cpus = 1024;
3295     mc->no_parallel = 1;
3296     mc->default_boot_order = "";
3297     mc->default_ram_size = 512 * M_BYTE;
3298     mc->kvm_type = spapr_kvm_type;
3299     mc->has_dynamic_sysbus = true;
3300     mc->pci_allow_0_address = true;
3301     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3302     hc->pre_plug = spapr_machine_device_pre_plug;
3303     hc->plug = spapr_machine_device_plug;
3304     hc->unplug = spapr_machine_device_unplug;
3305     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3306     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3307     hc->unplug_request = spapr_machine_device_unplug_request;
3308 
3309     smc->dr_lmb_enabled = true;
3310     smc->tcg_default_cpu = "POWER8";
3311     mc->has_hotpluggable_cpus = true;
3312     fwc->get_dev_path = spapr_get_fw_dev_path;
3313     nc->nmi_monitor_handler = spapr_nmi;
3314     smc->phb_placement = spapr_phb_placement;
3315     vhc->hypercall = emulate_spapr_hypercall;
3316     vhc->hpt_mask = spapr_hpt_mask;
3317     vhc->map_hptes = spapr_map_hptes;
3318     vhc->unmap_hptes = spapr_unmap_hptes;
3319     vhc->store_hpte = spapr_store_hpte;
3320     vhc->get_patbe = spapr_get_patbe;
3321     xic->ics_get = spapr_ics_get;
3322     xic->ics_resend = spapr_ics_resend;
3323     xic->icp_get = spapr_icp_get;
3324     ispc->print_info = spapr_pic_print_info;
3325     /* Force NUMA node memory size to be a multiple of
3326      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3327      * in which LMBs are represented and hot-added
3328      */
3329     mc->numa_mem_align_shift = 28;
3330 }
3331 
3332 static const TypeInfo spapr_machine_info = {
3333     .name          = TYPE_SPAPR_MACHINE,
3334     .parent        = TYPE_MACHINE,
3335     .abstract      = true,
3336     .instance_size = sizeof(sPAPRMachineState),
3337     .instance_init = spapr_machine_initfn,
3338     .instance_finalize = spapr_machine_finalizefn,
3339     .class_size    = sizeof(sPAPRMachineClass),
3340     .class_init    = spapr_machine_class_init,
3341     .interfaces = (InterfaceInfo[]) {
3342         { TYPE_FW_PATH_PROVIDER },
3343         { TYPE_NMI },
3344         { TYPE_HOTPLUG_HANDLER },
3345         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3346         { TYPE_XICS_FABRIC },
3347         { TYPE_INTERRUPT_STATS_PROVIDER },
3348         { }
3349     },
3350 };
3351 
3352 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3353     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3354                                                     void *data)      \
3355     {                                                                \
3356         MachineClass *mc = MACHINE_CLASS(oc);                        \
3357         spapr_machine_##suffix##_class_options(mc);                  \
3358         if (latest) {                                                \
3359             mc->alias = "pseries";                                   \
3360             mc->is_default = 1;                                      \
3361         }                                                            \
3362     }                                                                \
3363     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3364     {                                                                \
3365         MachineState *machine = MACHINE(obj);                        \
3366         spapr_machine_##suffix##_instance_options(machine);          \
3367     }                                                                \
3368     static const TypeInfo spapr_machine_##suffix##_info = {          \
3369         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3370         .parent = TYPE_SPAPR_MACHINE,                                \
3371         .class_init = spapr_machine_##suffix##_class_init,           \
3372         .instance_init = spapr_machine_##suffix##_instance_init,     \
3373     };                                                               \
3374     static void spapr_machine_register_##suffix(void)                \
3375     {                                                                \
3376         type_register(&spapr_machine_##suffix##_info);               \
3377     }                                                                \
3378     type_init(spapr_machine_register_##suffix)
3379 
3380 /*
3381  * pseries-2.10
3382  */
3383 static void spapr_machine_2_10_instance_options(MachineState *machine)
3384 {
3385 }
3386 
3387 static void spapr_machine_2_10_class_options(MachineClass *mc)
3388 {
3389     /* Defaults for the latest behaviour inherited from the base class */
3390 }
3391 
3392 DEFINE_SPAPR_MACHINE(2_10, "2.10", true);
3393 
3394 /*
3395  * pseries-2.9
3396  */
3397 #define SPAPR_COMPAT_2_9                                               \
3398     HW_COMPAT_2_9                                                      \
3399     {                                                                  \
3400         .driver = TYPE_POWERPC_CPU,                                    \
3401         .property = "pre-2.10-migration",                              \
3402         .value    = "on",                                              \
3403     },                                                                 \
3404 
3405 static void spapr_machine_2_9_instance_options(MachineState *machine)
3406 {
3407     spapr_machine_2_10_instance_options(machine);
3408 }
3409 
3410 static void spapr_machine_2_9_class_options(MachineClass *mc)
3411 {
3412     spapr_machine_2_10_class_options(mc);
3413     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3414     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
3415 }
3416 
3417 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3418 
3419 /*
3420  * pseries-2.8
3421  */
3422 #define SPAPR_COMPAT_2_8                                        \
3423     HW_COMPAT_2_8                                               \
3424     {                                                           \
3425         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3426         .property = "pcie-extended-configuration-space",        \
3427         .value    = "off",                                      \
3428     },
3429 
3430 static void spapr_machine_2_8_instance_options(MachineState *machine)
3431 {
3432     spapr_machine_2_9_instance_options(machine);
3433 }
3434 
3435 static void spapr_machine_2_8_class_options(MachineClass *mc)
3436 {
3437     spapr_machine_2_9_class_options(mc);
3438     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3439     mc->numa_mem_align_shift = 23;
3440 }
3441 
3442 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3443 
3444 /*
3445  * pseries-2.7
3446  */
3447 #define SPAPR_COMPAT_2_7                            \
3448     HW_COMPAT_2_7                                   \
3449     {                                               \
3450         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3451         .property = "mem_win_size",                 \
3452         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3453     },                                              \
3454     {                                               \
3455         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3456         .property = "mem64_win_size",               \
3457         .value    = "0",                            \
3458     },                                              \
3459     {                                               \
3460         .driver = TYPE_POWERPC_CPU,                 \
3461         .property = "pre-2.8-migration",            \
3462         .value    = "on",                           \
3463     },                                              \
3464     {                                               \
3465         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3466         .property = "pre-2.8-migration",            \
3467         .value    = "on",                           \
3468     },
3469 
3470 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3471                               uint64_t *buid, hwaddr *pio,
3472                               hwaddr *mmio32, hwaddr *mmio64,
3473                               unsigned n_dma, uint32_t *liobns, Error **errp)
3474 {
3475     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3476     const uint64_t base_buid = 0x800000020000000ULL;
3477     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3478     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3479     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3480     const uint32_t max_index = 255;
3481     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3482 
3483     uint64_t ram_top = MACHINE(spapr)->ram_size;
3484     hwaddr phb0_base, phb_base;
3485     int i;
3486 
3487     /* Do we have hotpluggable memory? */
3488     if (MACHINE(spapr)->maxram_size > ram_top) {
3489         /* Can't just use maxram_size, because there may be an
3490          * alignment gap between normal and hotpluggable memory
3491          * regions */
3492         ram_top = spapr->hotplug_memory.base +
3493             memory_region_size(&spapr->hotplug_memory.mr);
3494     }
3495 
3496     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3497 
3498     if (index > max_index) {
3499         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3500                    max_index);
3501         return;
3502     }
3503 
3504     *buid = base_buid + index;
3505     for (i = 0; i < n_dma; ++i) {
3506         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3507     }
3508 
3509     phb_base = phb0_base + index * phb_spacing;
3510     *pio = phb_base + pio_offset;
3511     *mmio32 = phb_base + mmio_offset;
3512     /*
3513      * We don't set the 64-bit MMIO window, relying on the PHB's
3514      * fallback behaviour of automatically splitting a large "32-bit"
3515      * window into contiguous 32-bit and 64-bit windows
3516      */
3517 }
3518 
3519 static void spapr_machine_2_7_instance_options(MachineState *machine)
3520 {
3521     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3522 
3523     spapr_machine_2_8_instance_options(machine);
3524     spapr->use_hotplug_event_source = false;
3525 }
3526 
3527 static void spapr_machine_2_7_class_options(MachineClass *mc)
3528 {
3529     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3530 
3531     spapr_machine_2_8_class_options(mc);
3532     smc->tcg_default_cpu = "POWER7";
3533     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3534     smc->phb_placement = phb_placement_2_7;
3535 }
3536 
3537 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3538 
3539 /*
3540  * pseries-2.6
3541  */
3542 #define SPAPR_COMPAT_2_6 \
3543     HW_COMPAT_2_6 \
3544     { \
3545         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3546         .property = "ddw",\
3547         .value    = stringify(off),\
3548     },
3549 
3550 static void spapr_machine_2_6_instance_options(MachineState *machine)
3551 {
3552     spapr_machine_2_7_instance_options(machine);
3553 }
3554 
3555 static void spapr_machine_2_6_class_options(MachineClass *mc)
3556 {
3557     spapr_machine_2_7_class_options(mc);
3558     mc->has_hotpluggable_cpus = false;
3559     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3560 }
3561 
3562 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3563 
3564 /*
3565  * pseries-2.5
3566  */
3567 #define SPAPR_COMPAT_2_5 \
3568     HW_COMPAT_2_5 \
3569     { \
3570         .driver   = "spapr-vlan", \
3571         .property = "use-rx-buffer-pools", \
3572         .value    = "off", \
3573     },
3574 
3575 static void spapr_machine_2_5_instance_options(MachineState *machine)
3576 {
3577     spapr_machine_2_6_instance_options(machine);
3578 }
3579 
3580 static void spapr_machine_2_5_class_options(MachineClass *mc)
3581 {
3582     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3583 
3584     spapr_machine_2_6_class_options(mc);
3585     smc->use_ohci_by_default = true;
3586     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3587 }
3588 
3589 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3590 
3591 /*
3592  * pseries-2.4
3593  */
3594 #define SPAPR_COMPAT_2_4 \
3595         HW_COMPAT_2_4
3596 
3597 static void spapr_machine_2_4_instance_options(MachineState *machine)
3598 {
3599     spapr_machine_2_5_instance_options(machine);
3600 }
3601 
3602 static void spapr_machine_2_4_class_options(MachineClass *mc)
3603 {
3604     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3605 
3606     spapr_machine_2_5_class_options(mc);
3607     smc->dr_lmb_enabled = false;
3608     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3609 }
3610 
3611 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3612 
3613 /*
3614  * pseries-2.3
3615  */
3616 #define SPAPR_COMPAT_2_3 \
3617         HW_COMPAT_2_3 \
3618         {\
3619             .driver   = "spapr-pci-host-bridge",\
3620             .property = "dynamic-reconfiguration",\
3621             .value    = "off",\
3622         },
3623 
3624 static void spapr_machine_2_3_instance_options(MachineState *machine)
3625 {
3626     spapr_machine_2_4_instance_options(machine);
3627 }
3628 
3629 static void spapr_machine_2_3_class_options(MachineClass *mc)
3630 {
3631     spapr_machine_2_4_class_options(mc);
3632     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3633 }
3634 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3635 
3636 /*
3637  * pseries-2.2
3638  */
3639 
3640 #define SPAPR_COMPAT_2_2 \
3641         HW_COMPAT_2_2 \
3642         {\
3643             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3644             .property = "mem_win_size",\
3645             .value    = "0x20000000",\
3646         },
3647 
3648 static void spapr_machine_2_2_instance_options(MachineState *machine)
3649 {
3650     spapr_machine_2_3_instance_options(machine);
3651     machine->suppress_vmdesc = true;
3652 }
3653 
3654 static void spapr_machine_2_2_class_options(MachineClass *mc)
3655 {
3656     spapr_machine_2_3_class_options(mc);
3657     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3658 }
3659 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3660 
3661 /*
3662  * pseries-2.1
3663  */
3664 #define SPAPR_COMPAT_2_1 \
3665         HW_COMPAT_2_1
3666 
3667 static void spapr_machine_2_1_instance_options(MachineState *machine)
3668 {
3669     spapr_machine_2_2_instance_options(machine);
3670 }
3671 
3672 static void spapr_machine_2_1_class_options(MachineClass *mc)
3673 {
3674     spapr_machine_2_2_class_options(mc);
3675     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3676 }
3677 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3678 
3679 static void spapr_machine_register_types(void)
3680 {
3681     type_register_static(&spapr_machine_info);
3682 }
3683 
3684 type_init(spapr_machine_register_types)
3685