xref: /openbmc/qemu/hw/ppc/spapr.c (revision 4b615be540d8566f4b981245b4d5401163c57ced)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
46 #include "cpu-models.h"
47 #include "qom/cpu.h"
48 
49 #include "hw/boards.h"
50 #include "hw/ppc/ppc.h"
51 #include "hw/loader.h"
52 
53 #include "hw/ppc/fdt.h"
54 #include "hw/ppc/spapr.h"
55 #include "hw/ppc/spapr_vio.h"
56 #include "hw/pci-host/spapr.h"
57 #include "hw/ppc/xics.h"
58 #include "hw/pci/msi.h"
59 
60 #include "hw/pci/pci.h"
61 #include "hw/scsi/scsi.h"
62 #include "hw/virtio/virtio-scsi.h"
63 #include "hw/virtio/vhost-scsi-common.h"
64 
65 #include "exec/address-spaces.h"
66 #include "exec/ram_addr.h"
67 #include "hw/usb.h"
68 #include "qemu/config-file.h"
69 #include "qemu/error-report.h"
70 #include "trace.h"
71 #include "hw/nmi.h"
72 #include "hw/intc/intc.h"
73 
74 #include "hw/compat.h"
75 #include "qemu/cutils.h"
76 #include "hw/ppc/spapr_cpu_core.h"
77 #include "hw/mem/memory-device.h"
78 
79 #include <libfdt.h>
80 
81 /* SLOF memory layout:
82  *
83  * SLOF raw image loaded at 0, copies its romfs right below the flat
84  * device-tree, then position SLOF itself 31M below that
85  *
86  * So we set FW_OVERHEAD to 40MB which should account for all of that
87  * and more
88  *
89  * We load our kernel at 4M, leaving space for SLOF initial image
90  */
91 #define FDT_MAX_SIZE            0x100000
92 #define RTAS_MAX_SIZE           0x10000
93 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
94 #define FW_MAX_SIZE             0x400000
95 #define FW_FILE_NAME            "slof.bin"
96 #define FW_OVERHEAD             0x2800000
97 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
98 
99 #define MIN_RMA_SLOF            128UL
100 
101 #define PHANDLE_XICP            0x00001111
102 
103 /* These two functions implement the VCPU id numbering: one to compute them
104  * all and one to identify thread 0 of a VCORE. Any change to the first one
105  * is likely to have an impact on the second one, so let's keep them close.
106  */
107 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
108 {
109     assert(spapr->vsmt);
110     return
111         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
112 }
113 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
114                                       PowerPCCPU *cpu)
115 {
116     assert(spapr->vsmt);
117     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
118 }
119 
120 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
121                                   const char *type_ics,
122                                   int nr_irqs, Error **errp)
123 {
124     Error *local_err = NULL;
125     Object *obj;
126 
127     obj = object_new(type_ics);
128     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
129     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
130                                    &error_abort);
131     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
132     if (local_err) {
133         goto error;
134     }
135     object_property_set_bool(obj, true, "realized", &local_err);
136     if (local_err) {
137         goto error;
138     }
139 
140     return ICS_BASE(obj);
141 
142 error:
143     error_propagate(errp, local_err);
144     return NULL;
145 }
146 
147 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
148 {
149     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
150      * and newer QEMUs don't even have them. In both cases, we don't want
151      * to send anything on the wire.
152      */
153     return false;
154 }
155 
156 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
157     .name = "icp/server",
158     .version_id = 1,
159     .minimum_version_id = 1,
160     .needed = pre_2_10_vmstate_dummy_icp_needed,
161     .fields = (VMStateField[]) {
162         VMSTATE_UNUSED(4), /* uint32_t xirr */
163         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
164         VMSTATE_UNUSED(1), /* uint8_t mfrr */
165         VMSTATE_END_OF_LIST()
166     },
167 };
168 
169 static void pre_2_10_vmstate_register_dummy_icp(int i)
170 {
171     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
172                      (void *)(uintptr_t) i);
173 }
174 
175 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
176 {
177     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
178                        (void *)(uintptr_t) i);
179 }
180 
181 static int xics_max_server_number(sPAPRMachineState *spapr)
182 {
183     assert(spapr->vsmt);
184     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
185 }
186 
187 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
188 {
189     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
190     Error *local_err = NULL;
191 
192     if (kvm_enabled()) {
193         if (machine_kernel_irqchip_allowed(machine) &&
194             !xics_kvm_init(spapr, &local_err)) {
195             spapr->icp_type = TYPE_KVM_ICP;
196             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs,
197                                           &local_err);
198         }
199         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
200             error_prepend(&local_err,
201                           "kernel_irqchip requested but unavailable: ");
202             goto error;
203         }
204         error_free(local_err);
205         local_err = NULL;
206     }
207 
208     if (!spapr->ics) {
209         xics_spapr_init(spapr);
210         spapr->icp_type = TYPE_ICP;
211         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs,
212                                       &local_err);
213     }
214 
215 error:
216     error_propagate(errp, local_err);
217 }
218 
219 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
220                                   int smt_threads)
221 {
222     int i, ret = 0;
223     uint32_t servers_prop[smt_threads];
224     uint32_t gservers_prop[smt_threads * 2];
225     int index = spapr_get_vcpu_id(cpu);
226 
227     if (cpu->compat_pvr) {
228         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
229         if (ret < 0) {
230             return ret;
231         }
232     }
233 
234     /* Build interrupt servers and gservers properties */
235     for (i = 0; i < smt_threads; i++) {
236         servers_prop[i] = cpu_to_be32(index + i);
237         /* Hack, direct the group queues back to cpu 0 */
238         gservers_prop[i*2] = cpu_to_be32(index + i);
239         gservers_prop[i*2 + 1] = 0;
240     }
241     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
242                       servers_prop, sizeof(servers_prop));
243     if (ret < 0) {
244         return ret;
245     }
246     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
247                       gservers_prop, sizeof(gservers_prop));
248 
249     return ret;
250 }
251 
252 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
253 {
254     int index = spapr_get_vcpu_id(cpu);
255     uint32_t associativity[] = {cpu_to_be32(0x5),
256                                 cpu_to_be32(0x0),
257                                 cpu_to_be32(0x0),
258                                 cpu_to_be32(0x0),
259                                 cpu_to_be32(cpu->node_id),
260                                 cpu_to_be32(index)};
261 
262     /* Advertise NUMA via ibm,associativity */
263     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
264                           sizeof(associativity));
265 }
266 
267 /* Populate the "ibm,pa-features" property */
268 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
269                                        PowerPCCPU *cpu,
270                                        void *fdt, int offset,
271                                        bool legacy_guest)
272 {
273     uint8_t pa_features_206[] = { 6, 0,
274         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
275     uint8_t pa_features_207[] = { 24, 0,
276         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
277         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
278         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
279         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
280     uint8_t pa_features_300[] = { 66, 0,
281         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
282         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
283         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
284         /* 6: DS207 */
285         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
286         /* 16: Vector */
287         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
288         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
289         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
290         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
291         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
292         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
293         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
294         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
295         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
296         /* 42: PM, 44: PC RA, 46: SC vec'd */
297         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
298         /* 48: SIMD, 50: QP BFP, 52: String */
299         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
300         /* 54: DecFP, 56: DecI, 58: SHA */
301         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
302         /* 60: NM atomic, 62: RNG */
303         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
304     };
305     uint8_t *pa_features = NULL;
306     size_t pa_size;
307 
308     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
309         pa_features = pa_features_206;
310         pa_size = sizeof(pa_features_206);
311     }
312     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
313         pa_features = pa_features_207;
314         pa_size = sizeof(pa_features_207);
315     }
316     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
317         pa_features = pa_features_300;
318         pa_size = sizeof(pa_features_300);
319     }
320     if (!pa_features) {
321         return;
322     }
323 
324     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
325         /*
326          * Note: we keep CI large pages off by default because a 64K capable
327          * guest provisioned with large pages might otherwise try to map a qemu
328          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
329          * even if that qemu runs on a 4k host.
330          * We dd this bit back here if we are confident this is not an issue
331          */
332         pa_features[3] |= 0x20;
333     }
334     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
335         pa_features[24] |= 0x80;    /* Transactional memory support */
336     }
337     if (legacy_guest && pa_size > 40) {
338         /* Workaround for broken kernels that attempt (guest) radix
339          * mode when they can't handle it, if they see the radix bit set
340          * in pa-features. So hide it from them. */
341         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
342     }
343 
344     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
345 }
346 
347 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
348 {
349     int ret = 0, offset, cpus_offset;
350     CPUState *cs;
351     char cpu_model[32];
352     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
353 
354     CPU_FOREACH(cs) {
355         PowerPCCPU *cpu = POWERPC_CPU(cs);
356         DeviceClass *dc = DEVICE_GET_CLASS(cs);
357         int index = spapr_get_vcpu_id(cpu);
358         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
359 
360         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
361             continue;
362         }
363 
364         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
365 
366         cpus_offset = fdt_path_offset(fdt, "/cpus");
367         if (cpus_offset < 0) {
368             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
369             if (cpus_offset < 0) {
370                 return cpus_offset;
371             }
372         }
373         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
374         if (offset < 0) {
375             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
376             if (offset < 0) {
377                 return offset;
378             }
379         }
380 
381         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
382                           pft_size_prop, sizeof(pft_size_prop));
383         if (ret < 0) {
384             return ret;
385         }
386 
387         if (nb_numa_nodes > 1) {
388             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
389             if (ret < 0) {
390                 return ret;
391             }
392         }
393 
394         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
395         if (ret < 0) {
396             return ret;
397         }
398 
399         spapr_populate_pa_features(spapr, cpu, fdt, offset,
400                                    spapr->cas_legacy_guest_workaround);
401     }
402     return ret;
403 }
404 
405 static hwaddr spapr_node0_size(MachineState *machine)
406 {
407     if (nb_numa_nodes) {
408         int i;
409         for (i = 0; i < nb_numa_nodes; ++i) {
410             if (numa_info[i].node_mem) {
411                 return MIN(pow2floor(numa_info[i].node_mem),
412                            machine->ram_size);
413             }
414         }
415     }
416     return machine->ram_size;
417 }
418 
419 static void add_str(GString *s, const gchar *s1)
420 {
421     g_string_append_len(s, s1, strlen(s1) + 1);
422 }
423 
424 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
425                                        hwaddr size)
426 {
427     uint32_t associativity[] = {
428         cpu_to_be32(0x4), /* length */
429         cpu_to_be32(0x0), cpu_to_be32(0x0),
430         cpu_to_be32(0x0), cpu_to_be32(nodeid)
431     };
432     char mem_name[32];
433     uint64_t mem_reg_property[2];
434     int off;
435 
436     mem_reg_property[0] = cpu_to_be64(start);
437     mem_reg_property[1] = cpu_to_be64(size);
438 
439     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
440     off = fdt_add_subnode(fdt, 0, mem_name);
441     _FDT(off);
442     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
443     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
444                       sizeof(mem_reg_property))));
445     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
446                       sizeof(associativity))));
447     return off;
448 }
449 
450 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
451 {
452     MachineState *machine = MACHINE(spapr);
453     hwaddr mem_start, node_size;
454     int i, nb_nodes = nb_numa_nodes;
455     NodeInfo *nodes = numa_info;
456     NodeInfo ramnode;
457 
458     /* No NUMA nodes, assume there is just one node with whole RAM */
459     if (!nb_numa_nodes) {
460         nb_nodes = 1;
461         ramnode.node_mem = machine->ram_size;
462         nodes = &ramnode;
463     }
464 
465     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
466         if (!nodes[i].node_mem) {
467             continue;
468         }
469         if (mem_start >= machine->ram_size) {
470             node_size = 0;
471         } else {
472             node_size = nodes[i].node_mem;
473             if (node_size > machine->ram_size - mem_start) {
474                 node_size = machine->ram_size - mem_start;
475             }
476         }
477         if (!mem_start) {
478             /* spapr_machine_init() checks for rma_size <= node0_size
479              * already */
480             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
481             mem_start += spapr->rma_size;
482             node_size -= spapr->rma_size;
483         }
484         for ( ; node_size; ) {
485             hwaddr sizetmp = pow2floor(node_size);
486 
487             /* mem_start != 0 here */
488             if (ctzl(mem_start) < ctzl(sizetmp)) {
489                 sizetmp = 1ULL << ctzl(mem_start);
490             }
491 
492             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
493             node_size -= sizetmp;
494             mem_start += sizetmp;
495         }
496     }
497 
498     return 0;
499 }
500 
501 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
502                                   sPAPRMachineState *spapr)
503 {
504     PowerPCCPU *cpu = POWERPC_CPU(cs);
505     CPUPPCState *env = &cpu->env;
506     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
507     int index = spapr_get_vcpu_id(cpu);
508     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
509                        0xffffffff, 0xffffffff};
510     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
511         : SPAPR_TIMEBASE_FREQ;
512     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
513     uint32_t page_sizes_prop[64];
514     size_t page_sizes_prop_size;
515     uint32_t vcpus_per_socket = smp_threads * smp_cores;
516     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
517     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
518     sPAPRDRConnector *drc;
519     int drc_index;
520     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
521     int i;
522 
523     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
524     if (drc) {
525         drc_index = spapr_drc_index(drc);
526         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
527     }
528 
529     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
530     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
531 
532     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
533     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
534                            env->dcache_line_size)));
535     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
536                            env->dcache_line_size)));
537     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
538                            env->icache_line_size)));
539     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
540                            env->icache_line_size)));
541 
542     if (pcc->l1_dcache_size) {
543         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
544                                pcc->l1_dcache_size)));
545     } else {
546         warn_report("Unknown L1 dcache size for cpu");
547     }
548     if (pcc->l1_icache_size) {
549         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
550                                pcc->l1_icache_size)));
551     } else {
552         warn_report("Unknown L1 icache size for cpu");
553     }
554 
555     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
556     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
557     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
558     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
559     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
560     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
561 
562     if (env->spr_cb[SPR_PURR].oea_read) {
563         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
564     }
565 
566     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
567         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
568                           segs, sizeof(segs))));
569     }
570 
571     /* Advertise VSX (vector extensions) if available
572      *   1               == VMX / Altivec available
573      *   2               == VSX available
574      *
575      * Only CPUs for which we create core types in spapr_cpu_core.c
576      * are possible, and all of those have VMX */
577     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
578         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
579     } else {
580         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
581     }
582 
583     /* Advertise DFP (Decimal Floating Point) if available
584      *   0 / no property == no DFP
585      *   1               == DFP available */
586     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
587         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
588     }
589 
590     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
591                                                       sizeof(page_sizes_prop));
592     if (page_sizes_prop_size) {
593         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
594                           page_sizes_prop, page_sizes_prop_size)));
595     }
596 
597     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
598 
599     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
600                            cs->cpu_index / vcpus_per_socket)));
601 
602     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
603                       pft_size_prop, sizeof(pft_size_prop))));
604 
605     if (nb_numa_nodes > 1) {
606         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
607     }
608 
609     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
610 
611     if (pcc->radix_page_info) {
612         for (i = 0; i < pcc->radix_page_info->count; i++) {
613             radix_AP_encodings[i] =
614                 cpu_to_be32(pcc->radix_page_info->entries[i]);
615         }
616         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
617                           radix_AP_encodings,
618                           pcc->radix_page_info->count *
619                           sizeof(radix_AP_encodings[0]))));
620     }
621 }
622 
623 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
624 {
625     CPUState **rev;
626     CPUState *cs;
627     int n_cpus;
628     int cpus_offset;
629     char *nodename;
630     int i;
631 
632     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
633     _FDT(cpus_offset);
634     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
635     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
636 
637     /*
638      * We walk the CPUs in reverse order to ensure that CPU DT nodes
639      * created by fdt_add_subnode() end up in the right order in FDT
640      * for the guest kernel the enumerate the CPUs correctly.
641      *
642      * The CPU list cannot be traversed in reverse order, so we need
643      * to do extra work.
644      */
645     n_cpus = 0;
646     rev = NULL;
647     CPU_FOREACH(cs) {
648         rev = g_renew(CPUState *, rev, n_cpus + 1);
649         rev[n_cpus++] = cs;
650     }
651 
652     for (i = n_cpus - 1; i >= 0; i--) {
653         CPUState *cs = rev[i];
654         PowerPCCPU *cpu = POWERPC_CPU(cs);
655         int index = spapr_get_vcpu_id(cpu);
656         DeviceClass *dc = DEVICE_GET_CLASS(cs);
657         int offset;
658 
659         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
660             continue;
661         }
662 
663         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
664         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
665         g_free(nodename);
666         _FDT(offset);
667         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
668     }
669 
670 }
671 
672 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
673 {
674     MemoryDeviceInfoList *info;
675 
676     for (info = list; info; info = info->next) {
677         MemoryDeviceInfo *value = info->value;
678 
679         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
680             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
681 
682             if (addr >= pcdimm_info->addr &&
683                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
684                 return pcdimm_info->node;
685             }
686         }
687     }
688 
689     return -1;
690 }
691 
692 struct sPAPRDrconfCellV2 {
693      uint32_t seq_lmbs;
694      uint64_t base_addr;
695      uint32_t drc_index;
696      uint32_t aa_index;
697      uint32_t flags;
698 } QEMU_PACKED;
699 
700 typedef struct DrconfCellQueue {
701     struct sPAPRDrconfCellV2 cell;
702     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
703 } DrconfCellQueue;
704 
705 static DrconfCellQueue *
706 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
707                       uint32_t drc_index, uint32_t aa_index,
708                       uint32_t flags)
709 {
710     DrconfCellQueue *elem;
711 
712     elem = g_malloc0(sizeof(*elem));
713     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
714     elem->cell.base_addr = cpu_to_be64(base_addr);
715     elem->cell.drc_index = cpu_to_be32(drc_index);
716     elem->cell.aa_index = cpu_to_be32(aa_index);
717     elem->cell.flags = cpu_to_be32(flags);
718 
719     return elem;
720 }
721 
722 /* ibm,dynamic-memory-v2 */
723 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
724                                    int offset, MemoryDeviceInfoList *dimms)
725 {
726     MachineState *machine = MACHINE(spapr);
727     uint8_t *int_buf, *cur_index, buf_len;
728     int ret;
729     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
730     uint64_t addr, cur_addr, size;
731     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
732     uint64_t mem_end = machine->device_memory->base +
733                        memory_region_size(&machine->device_memory->mr);
734     uint32_t node, nr_entries = 0;
735     sPAPRDRConnector *drc;
736     DrconfCellQueue *elem, *next;
737     MemoryDeviceInfoList *info;
738     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
739         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
740 
741     /* Entry to cover RAM and the gap area */
742     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
743                                  SPAPR_LMB_FLAGS_RESERVED |
744                                  SPAPR_LMB_FLAGS_DRC_INVALID);
745     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
746     nr_entries++;
747 
748     cur_addr = machine->device_memory->base;
749     for (info = dimms; info; info = info->next) {
750         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
751 
752         addr = di->addr;
753         size = di->size;
754         node = di->node;
755 
756         /* Entry for hot-pluggable area */
757         if (cur_addr < addr) {
758             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
759             g_assert(drc);
760             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
761                                          cur_addr, spapr_drc_index(drc), -1, 0);
762             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
763             nr_entries++;
764         }
765 
766         /* Entry for DIMM */
767         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
768         g_assert(drc);
769         elem = spapr_get_drconf_cell(size / lmb_size, addr,
770                                      spapr_drc_index(drc), node,
771                                      SPAPR_LMB_FLAGS_ASSIGNED);
772         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
773         nr_entries++;
774         cur_addr = addr + size;
775     }
776 
777     /* Entry for remaining hotpluggable area */
778     if (cur_addr < mem_end) {
779         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
780         g_assert(drc);
781         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
782                                      cur_addr, spapr_drc_index(drc), -1, 0);
783         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
784         nr_entries++;
785     }
786 
787     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
788     int_buf = cur_index = g_malloc0(buf_len);
789     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
790     cur_index += sizeof(nr_entries);
791 
792     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
793         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
794         cur_index += sizeof(elem->cell);
795         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
796         g_free(elem);
797     }
798 
799     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
800     g_free(int_buf);
801     if (ret < 0) {
802         return -1;
803     }
804     return 0;
805 }
806 
807 /* ibm,dynamic-memory */
808 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
809                                    int offset, MemoryDeviceInfoList *dimms)
810 {
811     MachineState *machine = MACHINE(spapr);
812     int i, ret;
813     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
814     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
815     uint32_t nr_lmbs = (machine->device_memory->base +
816                        memory_region_size(&machine->device_memory->mr)) /
817                        lmb_size;
818     uint32_t *int_buf, *cur_index, buf_len;
819 
820     /*
821      * Allocate enough buffer size to fit in ibm,dynamic-memory
822      */
823     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
824     cur_index = int_buf = g_malloc0(buf_len);
825     int_buf[0] = cpu_to_be32(nr_lmbs);
826     cur_index++;
827     for (i = 0; i < nr_lmbs; i++) {
828         uint64_t addr = i * lmb_size;
829         uint32_t *dynamic_memory = cur_index;
830 
831         if (i >= device_lmb_start) {
832             sPAPRDRConnector *drc;
833 
834             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
835             g_assert(drc);
836 
837             dynamic_memory[0] = cpu_to_be32(addr >> 32);
838             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
839             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
840             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
841             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
842             if (memory_region_present(get_system_memory(), addr)) {
843                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
844             } else {
845                 dynamic_memory[5] = cpu_to_be32(0);
846             }
847         } else {
848             /*
849              * LMB information for RMA, boot time RAM and gap b/n RAM and
850              * device memory region -- all these are marked as reserved
851              * and as having no valid DRC.
852              */
853             dynamic_memory[0] = cpu_to_be32(addr >> 32);
854             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
855             dynamic_memory[2] = cpu_to_be32(0);
856             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
857             dynamic_memory[4] = cpu_to_be32(-1);
858             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
859                                             SPAPR_LMB_FLAGS_DRC_INVALID);
860         }
861 
862         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
863     }
864     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
865     g_free(int_buf);
866     if (ret < 0) {
867         return -1;
868     }
869     return 0;
870 }
871 
872 /*
873  * Adds ibm,dynamic-reconfiguration-memory node.
874  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
875  * of this device tree node.
876  */
877 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
878 {
879     MachineState *machine = MACHINE(spapr);
880     int ret, i, offset;
881     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
882     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
883     uint32_t *int_buf, *cur_index, buf_len;
884     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
885     MemoryDeviceInfoList *dimms = NULL;
886 
887     /*
888      * Don't create the node if there is no device memory
889      */
890     if (machine->ram_size == machine->maxram_size) {
891         return 0;
892     }
893 
894     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
895 
896     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
897                     sizeof(prop_lmb_size));
898     if (ret < 0) {
899         return ret;
900     }
901 
902     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
903     if (ret < 0) {
904         return ret;
905     }
906 
907     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
908     if (ret < 0) {
909         return ret;
910     }
911 
912     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
913     dimms = qmp_memory_device_list();
914     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
915         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
916     } else {
917         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
918     }
919     qapi_free_MemoryDeviceInfoList(dimms);
920 
921     if (ret < 0) {
922         return ret;
923     }
924 
925     /* ibm,associativity-lookup-arrays */
926     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
927     cur_index = int_buf = g_malloc0(buf_len);
928 
929     cur_index = int_buf;
930     int_buf[0] = cpu_to_be32(nr_nodes);
931     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
932     cur_index += 2;
933     for (i = 0; i < nr_nodes; i++) {
934         uint32_t associativity[] = {
935             cpu_to_be32(0x0),
936             cpu_to_be32(0x0),
937             cpu_to_be32(0x0),
938             cpu_to_be32(i)
939         };
940         memcpy(cur_index, associativity, sizeof(associativity));
941         cur_index += 4;
942     }
943     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
944             (cur_index - int_buf) * sizeof(uint32_t));
945     g_free(int_buf);
946 
947     return ret;
948 }
949 
950 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
951                                 sPAPROptionVector *ov5_updates)
952 {
953     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
954     int ret = 0, offset;
955 
956     /* Generate ibm,dynamic-reconfiguration-memory node if required */
957     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
958         g_assert(smc->dr_lmb_enabled);
959         ret = spapr_populate_drconf_memory(spapr, fdt);
960         if (ret) {
961             goto out;
962         }
963     }
964 
965     offset = fdt_path_offset(fdt, "/chosen");
966     if (offset < 0) {
967         offset = fdt_add_subnode(fdt, 0, "chosen");
968         if (offset < 0) {
969             return offset;
970         }
971     }
972     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
973                                  "ibm,architecture-vec-5");
974 
975 out:
976     return ret;
977 }
978 
979 static bool spapr_hotplugged_dev_before_cas(void)
980 {
981     Object *drc_container, *obj;
982     ObjectProperty *prop;
983     ObjectPropertyIterator iter;
984 
985     drc_container = container_get(object_get_root(), "/dr-connector");
986     object_property_iter_init(&iter, drc_container);
987     while ((prop = object_property_iter_next(&iter))) {
988         if (!strstart(prop->type, "link<", NULL)) {
989             continue;
990         }
991         obj = object_property_get_link(drc_container, prop->name, NULL);
992         if (spapr_drc_needed(obj)) {
993             return true;
994         }
995     }
996     return false;
997 }
998 
999 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
1000                                  target_ulong addr, target_ulong size,
1001                                  sPAPROptionVector *ov5_updates)
1002 {
1003     void *fdt, *fdt_skel;
1004     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
1005 
1006     if (spapr_hotplugged_dev_before_cas()) {
1007         return 1;
1008     }
1009 
1010     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
1011         error_report("SLOF provided an unexpected CAS buffer size "
1012                      TARGET_FMT_lu " (min: %zu, max: %u)",
1013                      size, sizeof(hdr), FW_MAX_SIZE);
1014         exit(EXIT_FAILURE);
1015     }
1016 
1017     size -= sizeof(hdr);
1018 
1019     /* Create skeleton */
1020     fdt_skel = g_malloc0(size);
1021     _FDT((fdt_create(fdt_skel, size)));
1022     _FDT((fdt_finish_reservemap(fdt_skel)));
1023     _FDT((fdt_begin_node(fdt_skel, "")));
1024     _FDT((fdt_end_node(fdt_skel)));
1025     _FDT((fdt_finish(fdt_skel)));
1026     fdt = g_malloc0(size);
1027     _FDT((fdt_open_into(fdt_skel, fdt, size)));
1028     g_free(fdt_skel);
1029 
1030     /* Fixup cpu nodes */
1031     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
1032 
1033     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
1034         return -1;
1035     }
1036 
1037     /* Pack resulting tree */
1038     _FDT((fdt_pack(fdt)));
1039 
1040     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1041         trace_spapr_cas_failed(size);
1042         return -1;
1043     }
1044 
1045     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1046     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1047     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1048     g_free(fdt);
1049 
1050     return 0;
1051 }
1052 
1053 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1054 {
1055     int rtas;
1056     GString *hypertas = g_string_sized_new(256);
1057     GString *qemu_hypertas = g_string_sized_new(256);
1058     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1059     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1060         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1061     uint32_t lrdr_capacity[] = {
1062         cpu_to_be32(max_device_addr >> 32),
1063         cpu_to_be32(max_device_addr & 0xffffffff),
1064         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1065         cpu_to_be32(max_cpus / smp_threads),
1066     };
1067     uint32_t maxdomains[] = {
1068         cpu_to_be32(4),
1069         cpu_to_be32(0),
1070         cpu_to_be32(0),
1071         cpu_to_be32(0),
1072         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
1073     };
1074 
1075     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1076 
1077     /* hypertas */
1078     add_str(hypertas, "hcall-pft");
1079     add_str(hypertas, "hcall-term");
1080     add_str(hypertas, "hcall-dabr");
1081     add_str(hypertas, "hcall-interrupt");
1082     add_str(hypertas, "hcall-tce");
1083     add_str(hypertas, "hcall-vio");
1084     add_str(hypertas, "hcall-splpar");
1085     add_str(hypertas, "hcall-bulk");
1086     add_str(hypertas, "hcall-set-mode");
1087     add_str(hypertas, "hcall-sprg0");
1088     add_str(hypertas, "hcall-copy");
1089     add_str(hypertas, "hcall-debug");
1090     add_str(qemu_hypertas, "hcall-memop1");
1091 
1092     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1093         add_str(hypertas, "hcall-multi-tce");
1094     }
1095 
1096     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1097         add_str(hypertas, "hcall-hpt-resize");
1098     }
1099 
1100     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1101                      hypertas->str, hypertas->len));
1102     g_string_free(hypertas, TRUE);
1103     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1104                      qemu_hypertas->str, qemu_hypertas->len));
1105     g_string_free(qemu_hypertas, TRUE);
1106 
1107     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1108                      refpoints, sizeof(refpoints)));
1109 
1110     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1111                      maxdomains, sizeof(maxdomains)));
1112 
1113     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1114                           RTAS_ERROR_LOG_MAX));
1115     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1116                           RTAS_EVENT_SCAN_RATE));
1117 
1118     g_assert(msi_nonbroken);
1119     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1120 
1121     /*
1122      * According to PAPR, rtas ibm,os-term does not guarantee a return
1123      * back to the guest cpu.
1124      *
1125      * While an additional ibm,extended-os-term property indicates
1126      * that rtas call return will always occur. Set this property.
1127      */
1128     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1129 
1130     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1131                      lrdr_capacity, sizeof(lrdr_capacity)));
1132 
1133     spapr_dt_rtas_tokens(fdt, rtas);
1134 }
1135 
1136 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1137  * that the guest may request and thus the valid values for bytes 24..26 of
1138  * option vector 5: */
1139 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1140 {
1141     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1142 
1143     char val[2 * 4] = {
1144         23, 0x00, /* Xive mode, filled in below. */
1145         24, 0x00, /* Hash/Radix, filled in below. */
1146         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1147         26, 0x40, /* Radix options: GTSE == yes. */
1148     };
1149 
1150     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1151                           first_ppc_cpu->compat_pvr)) {
1152         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1153         val[3] = 0x00; /* Hash */
1154     } else if (kvm_enabled()) {
1155         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1156             val[3] = 0x80; /* OV5_MMU_BOTH */
1157         } else if (kvmppc_has_cap_mmu_radix()) {
1158             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1159         } else {
1160             val[3] = 0x00; /* Hash */
1161         }
1162     } else {
1163         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1164         val[3] = 0xC0;
1165     }
1166     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1167                      val, sizeof(val)));
1168 }
1169 
1170 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1171 {
1172     MachineState *machine = MACHINE(spapr);
1173     int chosen;
1174     const char *boot_device = machine->boot_order;
1175     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1176     size_t cb = 0;
1177     char *bootlist = get_boot_devices_list(&cb);
1178 
1179     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1180 
1181     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1182     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1183                           spapr->initrd_base));
1184     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1185                           spapr->initrd_base + spapr->initrd_size));
1186 
1187     if (spapr->kernel_size) {
1188         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1189                               cpu_to_be64(spapr->kernel_size) };
1190 
1191         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1192                          &kprop, sizeof(kprop)));
1193         if (spapr->kernel_le) {
1194             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1195         }
1196     }
1197     if (boot_menu) {
1198         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1199     }
1200     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1201     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1202     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1203 
1204     if (cb && bootlist) {
1205         int i;
1206 
1207         for (i = 0; i < cb; i++) {
1208             if (bootlist[i] == '\n') {
1209                 bootlist[i] = ' ';
1210             }
1211         }
1212         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1213     }
1214 
1215     if (boot_device && strlen(boot_device)) {
1216         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1217     }
1218 
1219     if (!spapr->has_graphics && stdout_path) {
1220         /*
1221          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1222          * kernel. New platforms should only use the "stdout-path" property. Set
1223          * the new property and continue using older property to remain
1224          * compatible with the existing firmware.
1225          */
1226         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1227         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1228     }
1229 
1230     spapr_dt_ov5_platform_support(fdt, chosen);
1231 
1232     g_free(stdout_path);
1233     g_free(bootlist);
1234 }
1235 
1236 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1237 {
1238     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1239      * KVM to work under pHyp with some guest co-operation */
1240     int hypervisor;
1241     uint8_t hypercall[16];
1242 
1243     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1244     /* indicate KVM hypercall interface */
1245     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1246     if (kvmppc_has_cap_fixup_hcalls()) {
1247         /*
1248          * Older KVM versions with older guest kernels were broken
1249          * with the magic page, don't allow the guest to map it.
1250          */
1251         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1252                                   sizeof(hypercall))) {
1253             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1254                              hypercall, sizeof(hypercall)));
1255         }
1256     }
1257 }
1258 
1259 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1260                              hwaddr rtas_addr,
1261                              hwaddr rtas_size)
1262 {
1263     MachineState *machine = MACHINE(spapr);
1264     MachineClass *mc = MACHINE_GET_CLASS(machine);
1265     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1266     int ret;
1267     void *fdt;
1268     sPAPRPHBState *phb;
1269     char *buf;
1270 
1271     fdt = g_malloc0(FDT_MAX_SIZE);
1272     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1273 
1274     /* Root node */
1275     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1276     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1277     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1278 
1279     /*
1280      * Add info to guest to indentify which host is it being run on
1281      * and what is the uuid of the guest
1282      */
1283     if (kvmppc_get_host_model(&buf)) {
1284         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1285         g_free(buf);
1286     }
1287     if (kvmppc_get_host_serial(&buf)) {
1288         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1289         g_free(buf);
1290     }
1291 
1292     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1293 
1294     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1295     if (qemu_uuid_set) {
1296         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1297     }
1298     g_free(buf);
1299 
1300     if (qemu_get_vm_name()) {
1301         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1302                                 qemu_get_vm_name()));
1303     }
1304 
1305     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1306     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1307 
1308     /* /interrupt controller */
1309     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1310 
1311     ret = spapr_populate_memory(spapr, fdt);
1312     if (ret < 0) {
1313         error_report("couldn't setup memory nodes in fdt");
1314         exit(1);
1315     }
1316 
1317     /* /vdevice */
1318     spapr_dt_vdevice(spapr->vio_bus, fdt);
1319 
1320     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1321         ret = spapr_rng_populate_dt(fdt);
1322         if (ret < 0) {
1323             error_report("could not set up rng device in the fdt");
1324             exit(1);
1325         }
1326     }
1327 
1328     QLIST_FOREACH(phb, &spapr->phbs, list) {
1329         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1330         if (ret < 0) {
1331             error_report("couldn't setup PCI devices in fdt");
1332             exit(1);
1333         }
1334     }
1335 
1336     /* cpus */
1337     spapr_populate_cpus_dt_node(fdt, spapr);
1338 
1339     if (smc->dr_lmb_enabled) {
1340         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1341     }
1342 
1343     if (mc->has_hotpluggable_cpus) {
1344         int offset = fdt_path_offset(fdt, "/cpus");
1345         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1346                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1347         if (ret < 0) {
1348             error_report("Couldn't set up CPU DR device tree properties");
1349             exit(1);
1350         }
1351     }
1352 
1353     /* /event-sources */
1354     spapr_dt_events(spapr, fdt);
1355 
1356     /* /rtas */
1357     spapr_dt_rtas(spapr, fdt);
1358 
1359     /* /chosen */
1360     spapr_dt_chosen(spapr, fdt);
1361 
1362     /* /hypervisor */
1363     if (kvm_enabled()) {
1364         spapr_dt_hypervisor(spapr, fdt);
1365     }
1366 
1367     /* Build memory reserve map */
1368     if (spapr->kernel_size) {
1369         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1370     }
1371     if (spapr->initrd_size) {
1372         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1373     }
1374 
1375     /* ibm,client-architecture-support updates */
1376     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1377     if (ret < 0) {
1378         error_report("couldn't setup CAS properties fdt");
1379         exit(1);
1380     }
1381 
1382     return fdt;
1383 }
1384 
1385 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1386 {
1387     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1388 }
1389 
1390 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1391                                     PowerPCCPU *cpu)
1392 {
1393     CPUPPCState *env = &cpu->env;
1394 
1395     /* The TCG path should also be holding the BQL at this point */
1396     g_assert(qemu_mutex_iothread_locked());
1397 
1398     if (msr_pr) {
1399         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1400         env->gpr[3] = H_PRIVILEGE;
1401     } else {
1402         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1403     }
1404 }
1405 
1406 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1407 {
1408     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1409 
1410     return spapr->patb_entry;
1411 }
1412 
1413 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1414 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1415 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1416 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1417 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1418 
1419 /*
1420  * Get the fd to access the kernel htab, re-opening it if necessary
1421  */
1422 static int get_htab_fd(sPAPRMachineState *spapr)
1423 {
1424     Error *local_err = NULL;
1425 
1426     if (spapr->htab_fd >= 0) {
1427         return spapr->htab_fd;
1428     }
1429 
1430     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1431     if (spapr->htab_fd < 0) {
1432         error_report_err(local_err);
1433     }
1434 
1435     return spapr->htab_fd;
1436 }
1437 
1438 void close_htab_fd(sPAPRMachineState *spapr)
1439 {
1440     if (spapr->htab_fd >= 0) {
1441         close(spapr->htab_fd);
1442     }
1443     spapr->htab_fd = -1;
1444 }
1445 
1446 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1447 {
1448     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1449 
1450     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1451 }
1452 
1453 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1454 {
1455     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1456 
1457     assert(kvm_enabled());
1458 
1459     if (!spapr->htab) {
1460         return 0;
1461     }
1462 
1463     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1464 }
1465 
1466 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1467                                                 hwaddr ptex, int n)
1468 {
1469     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1470     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1471 
1472     if (!spapr->htab) {
1473         /*
1474          * HTAB is controlled by KVM. Fetch into temporary buffer
1475          */
1476         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1477         kvmppc_read_hptes(hptes, ptex, n);
1478         return hptes;
1479     }
1480 
1481     /*
1482      * HTAB is controlled by QEMU. Just point to the internally
1483      * accessible PTEG.
1484      */
1485     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1486 }
1487 
1488 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1489                               const ppc_hash_pte64_t *hptes,
1490                               hwaddr ptex, int n)
1491 {
1492     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1493 
1494     if (!spapr->htab) {
1495         g_free((void *)hptes);
1496     }
1497 
1498     /* Nothing to do for qemu managed HPT */
1499 }
1500 
1501 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1502                              uint64_t pte0, uint64_t pte1)
1503 {
1504     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1505     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1506 
1507     if (!spapr->htab) {
1508         kvmppc_write_hpte(ptex, pte0, pte1);
1509     } else {
1510         stq_p(spapr->htab + offset, pte0);
1511         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1512     }
1513 }
1514 
1515 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1516 {
1517     int shift;
1518 
1519     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1520      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1521      * that's much more than is needed for Linux guests */
1522     shift = ctz64(pow2ceil(ramsize)) - 7;
1523     shift = MAX(shift, 18); /* Minimum architected size */
1524     shift = MIN(shift, 46); /* Maximum architected size */
1525     return shift;
1526 }
1527 
1528 void spapr_free_hpt(sPAPRMachineState *spapr)
1529 {
1530     g_free(spapr->htab);
1531     spapr->htab = NULL;
1532     spapr->htab_shift = 0;
1533     close_htab_fd(spapr);
1534 }
1535 
1536 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1537                           Error **errp)
1538 {
1539     long rc;
1540 
1541     /* Clean up any HPT info from a previous boot */
1542     spapr_free_hpt(spapr);
1543 
1544     rc = kvmppc_reset_htab(shift);
1545     if (rc < 0) {
1546         /* kernel-side HPT needed, but couldn't allocate one */
1547         error_setg_errno(errp, errno,
1548                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1549                          shift);
1550         /* This is almost certainly fatal, but if the caller really
1551          * wants to carry on with shift == 0, it's welcome to try */
1552     } else if (rc > 0) {
1553         /* kernel-side HPT allocated */
1554         if (rc != shift) {
1555             error_setg(errp,
1556                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1557                        shift, rc);
1558         }
1559 
1560         spapr->htab_shift = shift;
1561         spapr->htab = NULL;
1562     } else {
1563         /* kernel-side HPT not needed, allocate in userspace instead */
1564         size_t size = 1ULL << shift;
1565         int i;
1566 
1567         spapr->htab = qemu_memalign(size, size);
1568         if (!spapr->htab) {
1569             error_setg_errno(errp, errno,
1570                              "Could not allocate HPT of order %d", shift);
1571             return;
1572         }
1573 
1574         memset(spapr->htab, 0, size);
1575         spapr->htab_shift = shift;
1576 
1577         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1578             DIRTY_HPTE(HPTE(spapr->htab, i));
1579         }
1580     }
1581     /* We're setting up a hash table, so that means we're not radix */
1582     spapr->patb_entry = 0;
1583 }
1584 
1585 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1586 {
1587     int hpt_shift;
1588 
1589     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1590         || (spapr->cas_reboot
1591             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1592         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1593     } else {
1594         uint64_t current_ram_size;
1595 
1596         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1597         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1598     }
1599     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1600 
1601     if (spapr->vrma_adjust) {
1602         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1603                                           spapr->htab_shift);
1604     }
1605 }
1606 
1607 static int spapr_reset_drcs(Object *child, void *opaque)
1608 {
1609     sPAPRDRConnector *drc =
1610         (sPAPRDRConnector *) object_dynamic_cast(child,
1611                                                  TYPE_SPAPR_DR_CONNECTOR);
1612 
1613     if (drc) {
1614         spapr_drc_reset(drc);
1615     }
1616 
1617     return 0;
1618 }
1619 
1620 static void spapr_machine_reset(void)
1621 {
1622     MachineState *machine = MACHINE(qdev_get_machine());
1623     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1624     PowerPCCPU *first_ppc_cpu;
1625     uint32_t rtas_limit;
1626     hwaddr rtas_addr, fdt_addr;
1627     void *fdt;
1628     int rc;
1629 
1630     spapr_caps_apply(spapr);
1631 
1632     first_ppc_cpu = POWERPC_CPU(first_cpu);
1633     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1634         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1635                               spapr->max_compat_pvr)) {
1636         /* If using KVM with radix mode available, VCPUs can be started
1637          * without a HPT because KVM will start them in radix mode.
1638          * Set the GR bit in PATB so that we know there is no HPT. */
1639         spapr->patb_entry = PATBE1_GR;
1640     } else {
1641         spapr_setup_hpt_and_vrma(spapr);
1642     }
1643 
1644     /* if this reset wasn't generated by CAS, we should reset our
1645      * negotiated options and start from scratch */
1646     if (!spapr->cas_reboot) {
1647         spapr_ovec_cleanup(spapr->ov5_cas);
1648         spapr->ov5_cas = spapr_ovec_new();
1649 
1650         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1651     }
1652 
1653     qemu_devices_reset();
1654 
1655     /* DRC reset may cause a device to be unplugged. This will cause troubles
1656      * if this device is used by another device (eg, a running vhost backend
1657      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1658      * situations, we reset DRCs after all devices have been reset.
1659      */
1660     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1661 
1662     spapr_clear_pending_events(spapr);
1663 
1664     /*
1665      * We place the device tree and RTAS just below either the top of the RMA,
1666      * or just below 2GB, whichever is lowere, so that it can be
1667      * processed with 32-bit real mode code if necessary
1668      */
1669     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1670     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1671     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1672 
1673     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1674 
1675     spapr_load_rtas(spapr, fdt, rtas_addr);
1676 
1677     rc = fdt_pack(fdt);
1678 
1679     /* Should only fail if we've built a corrupted tree */
1680     assert(rc == 0);
1681 
1682     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1683         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1684                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1685         exit(1);
1686     }
1687 
1688     /* Load the fdt */
1689     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1690     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1691     g_free(fdt);
1692 
1693     /* Set up the entry state */
1694     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1695     first_ppc_cpu->env.gpr[5] = 0;
1696 
1697     spapr->cas_reboot = false;
1698 }
1699 
1700 static void spapr_create_nvram(sPAPRMachineState *spapr)
1701 {
1702     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1703     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1704 
1705     if (dinfo) {
1706         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1707                             &error_fatal);
1708     }
1709 
1710     qdev_init_nofail(dev);
1711 
1712     spapr->nvram = (struct sPAPRNVRAM *)dev;
1713 }
1714 
1715 static void spapr_rtc_create(sPAPRMachineState *spapr)
1716 {
1717     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1718     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1719                               &error_fatal);
1720     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1721                               &error_fatal);
1722     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1723                               "date", &error_fatal);
1724 }
1725 
1726 /* Returns whether we want to use VGA or not */
1727 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1728 {
1729     switch (vga_interface_type) {
1730     case VGA_NONE:
1731         return false;
1732     case VGA_DEVICE:
1733         return true;
1734     case VGA_STD:
1735     case VGA_VIRTIO:
1736         return pci_vga_init(pci_bus) != NULL;
1737     default:
1738         error_setg(errp,
1739                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1740         return false;
1741     }
1742 }
1743 
1744 static int spapr_pre_load(void *opaque)
1745 {
1746     int rc;
1747 
1748     rc = spapr_caps_pre_load(opaque);
1749     if (rc) {
1750         return rc;
1751     }
1752 
1753     return 0;
1754 }
1755 
1756 static int spapr_post_load(void *opaque, int version_id)
1757 {
1758     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1759     int err = 0;
1760 
1761     err = spapr_caps_post_migration(spapr);
1762     if (err) {
1763         return err;
1764     }
1765 
1766     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1767         CPUState *cs;
1768         CPU_FOREACH(cs) {
1769             PowerPCCPU *cpu = POWERPC_CPU(cs);
1770             icp_resend(ICP(cpu->intc));
1771         }
1772     }
1773 
1774     /* In earlier versions, there was no separate qdev for the PAPR
1775      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1776      * So when migrating from those versions, poke the incoming offset
1777      * value into the RTC device */
1778     if (version_id < 3) {
1779         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1780     }
1781 
1782     if (kvm_enabled() && spapr->patb_entry) {
1783         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1784         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1785         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1786 
1787         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1788         if (err) {
1789             error_report("Process table config unsupported by the host");
1790             return -EINVAL;
1791         }
1792     }
1793 
1794     return err;
1795 }
1796 
1797 static int spapr_pre_save(void *opaque)
1798 {
1799     int rc;
1800 
1801     rc = spapr_caps_pre_save(opaque);
1802     if (rc) {
1803         return rc;
1804     }
1805 
1806     return 0;
1807 }
1808 
1809 static bool version_before_3(void *opaque, int version_id)
1810 {
1811     return version_id < 3;
1812 }
1813 
1814 static bool spapr_pending_events_needed(void *opaque)
1815 {
1816     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1817     return !QTAILQ_EMPTY(&spapr->pending_events);
1818 }
1819 
1820 static const VMStateDescription vmstate_spapr_event_entry = {
1821     .name = "spapr_event_log_entry",
1822     .version_id = 1,
1823     .minimum_version_id = 1,
1824     .fields = (VMStateField[]) {
1825         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1826         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1827         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1828                                      NULL, extended_length),
1829         VMSTATE_END_OF_LIST()
1830     },
1831 };
1832 
1833 static const VMStateDescription vmstate_spapr_pending_events = {
1834     .name = "spapr_pending_events",
1835     .version_id = 1,
1836     .minimum_version_id = 1,
1837     .needed = spapr_pending_events_needed,
1838     .fields = (VMStateField[]) {
1839         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1840                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1841         VMSTATE_END_OF_LIST()
1842     },
1843 };
1844 
1845 static bool spapr_ov5_cas_needed(void *opaque)
1846 {
1847     sPAPRMachineState *spapr = opaque;
1848     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1849     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1850     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1851     bool cas_needed;
1852 
1853     /* Prior to the introduction of sPAPROptionVector, we had two option
1854      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1855      * Both of these options encode machine topology into the device-tree
1856      * in such a way that the now-booted OS should still be able to interact
1857      * appropriately with QEMU regardless of what options were actually
1858      * negotiatied on the source side.
1859      *
1860      * As such, we can avoid migrating the CAS-negotiated options if these
1861      * are the only options available on the current machine/platform.
1862      * Since these are the only options available for pseries-2.7 and
1863      * earlier, this allows us to maintain old->new/new->old migration
1864      * compatibility.
1865      *
1866      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1867      * via default pseries-2.8 machines and explicit command-line parameters.
1868      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1869      * of the actual CAS-negotiated values to continue working properly. For
1870      * example, availability of memory unplug depends on knowing whether
1871      * OV5_HP_EVT was negotiated via CAS.
1872      *
1873      * Thus, for any cases where the set of available CAS-negotiatable
1874      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1875      * include the CAS-negotiated options in the migration stream, unless
1876      * if they affect boot time behaviour only.
1877      */
1878     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1879     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1880     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1881 
1882     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1883      * the mask itself since in the future it's possible "legacy" bits may be
1884      * removed via machine options, which could generate a false positive
1885      * that breaks migration.
1886      */
1887     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1888     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1889 
1890     spapr_ovec_cleanup(ov5_mask);
1891     spapr_ovec_cleanup(ov5_legacy);
1892     spapr_ovec_cleanup(ov5_removed);
1893 
1894     return cas_needed;
1895 }
1896 
1897 static const VMStateDescription vmstate_spapr_ov5_cas = {
1898     .name = "spapr_option_vector_ov5_cas",
1899     .version_id = 1,
1900     .minimum_version_id = 1,
1901     .needed = spapr_ov5_cas_needed,
1902     .fields = (VMStateField[]) {
1903         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1904                                  vmstate_spapr_ovec, sPAPROptionVector),
1905         VMSTATE_END_OF_LIST()
1906     },
1907 };
1908 
1909 static bool spapr_patb_entry_needed(void *opaque)
1910 {
1911     sPAPRMachineState *spapr = opaque;
1912 
1913     return !!spapr->patb_entry;
1914 }
1915 
1916 static const VMStateDescription vmstate_spapr_patb_entry = {
1917     .name = "spapr_patb_entry",
1918     .version_id = 1,
1919     .minimum_version_id = 1,
1920     .needed = spapr_patb_entry_needed,
1921     .fields = (VMStateField[]) {
1922         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1923         VMSTATE_END_OF_LIST()
1924     },
1925 };
1926 
1927 static const VMStateDescription vmstate_spapr = {
1928     .name = "spapr",
1929     .version_id = 3,
1930     .minimum_version_id = 1,
1931     .pre_load = spapr_pre_load,
1932     .post_load = spapr_post_load,
1933     .pre_save = spapr_pre_save,
1934     .fields = (VMStateField[]) {
1935         /* used to be @next_irq */
1936         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1937 
1938         /* RTC offset */
1939         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1940 
1941         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1942         VMSTATE_END_OF_LIST()
1943     },
1944     .subsections = (const VMStateDescription*[]) {
1945         &vmstate_spapr_ov5_cas,
1946         &vmstate_spapr_patb_entry,
1947         &vmstate_spapr_pending_events,
1948         &vmstate_spapr_cap_htm,
1949         &vmstate_spapr_cap_vsx,
1950         &vmstate_spapr_cap_dfp,
1951         &vmstate_spapr_cap_cfpc,
1952         &vmstate_spapr_cap_sbbc,
1953         &vmstate_spapr_cap_ibs,
1954         NULL
1955     }
1956 };
1957 
1958 static int htab_save_setup(QEMUFile *f, void *opaque)
1959 {
1960     sPAPRMachineState *spapr = opaque;
1961 
1962     /* "Iteration" header */
1963     if (!spapr->htab_shift) {
1964         qemu_put_be32(f, -1);
1965     } else {
1966         qemu_put_be32(f, spapr->htab_shift);
1967     }
1968 
1969     if (spapr->htab) {
1970         spapr->htab_save_index = 0;
1971         spapr->htab_first_pass = true;
1972     } else {
1973         if (spapr->htab_shift) {
1974             assert(kvm_enabled());
1975         }
1976     }
1977 
1978 
1979     return 0;
1980 }
1981 
1982 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1983                             int chunkstart, int n_valid, int n_invalid)
1984 {
1985     qemu_put_be32(f, chunkstart);
1986     qemu_put_be16(f, n_valid);
1987     qemu_put_be16(f, n_invalid);
1988     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1989                     HASH_PTE_SIZE_64 * n_valid);
1990 }
1991 
1992 static void htab_save_end_marker(QEMUFile *f)
1993 {
1994     qemu_put_be32(f, 0);
1995     qemu_put_be16(f, 0);
1996     qemu_put_be16(f, 0);
1997 }
1998 
1999 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
2000                                  int64_t max_ns)
2001 {
2002     bool has_timeout = max_ns != -1;
2003     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2004     int index = spapr->htab_save_index;
2005     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2006 
2007     assert(spapr->htab_first_pass);
2008 
2009     do {
2010         int chunkstart;
2011 
2012         /* Consume invalid HPTEs */
2013         while ((index < htabslots)
2014                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2015             CLEAN_HPTE(HPTE(spapr->htab, index));
2016             index++;
2017         }
2018 
2019         /* Consume valid HPTEs */
2020         chunkstart = index;
2021         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2022                && HPTE_VALID(HPTE(spapr->htab, index))) {
2023             CLEAN_HPTE(HPTE(spapr->htab, index));
2024             index++;
2025         }
2026 
2027         if (index > chunkstart) {
2028             int n_valid = index - chunkstart;
2029 
2030             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2031 
2032             if (has_timeout &&
2033                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2034                 break;
2035             }
2036         }
2037     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2038 
2039     if (index >= htabslots) {
2040         assert(index == htabslots);
2041         index = 0;
2042         spapr->htab_first_pass = false;
2043     }
2044     spapr->htab_save_index = index;
2045 }
2046 
2047 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2048                                 int64_t max_ns)
2049 {
2050     bool final = max_ns < 0;
2051     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2052     int examined = 0, sent = 0;
2053     int index = spapr->htab_save_index;
2054     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2055 
2056     assert(!spapr->htab_first_pass);
2057 
2058     do {
2059         int chunkstart, invalidstart;
2060 
2061         /* Consume non-dirty HPTEs */
2062         while ((index < htabslots)
2063                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2064             index++;
2065             examined++;
2066         }
2067 
2068         chunkstart = index;
2069         /* Consume valid dirty HPTEs */
2070         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2071                && HPTE_DIRTY(HPTE(spapr->htab, index))
2072                && HPTE_VALID(HPTE(spapr->htab, index))) {
2073             CLEAN_HPTE(HPTE(spapr->htab, index));
2074             index++;
2075             examined++;
2076         }
2077 
2078         invalidstart = index;
2079         /* Consume invalid dirty HPTEs */
2080         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2081                && HPTE_DIRTY(HPTE(spapr->htab, index))
2082                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2083             CLEAN_HPTE(HPTE(spapr->htab, index));
2084             index++;
2085             examined++;
2086         }
2087 
2088         if (index > chunkstart) {
2089             int n_valid = invalidstart - chunkstart;
2090             int n_invalid = index - invalidstart;
2091 
2092             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2093             sent += index - chunkstart;
2094 
2095             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2096                 break;
2097             }
2098         }
2099 
2100         if (examined >= htabslots) {
2101             break;
2102         }
2103 
2104         if (index >= htabslots) {
2105             assert(index == htabslots);
2106             index = 0;
2107         }
2108     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2109 
2110     if (index >= htabslots) {
2111         assert(index == htabslots);
2112         index = 0;
2113     }
2114 
2115     spapr->htab_save_index = index;
2116 
2117     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2118 }
2119 
2120 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2121 #define MAX_KVM_BUF_SIZE    2048
2122 
2123 static int htab_save_iterate(QEMUFile *f, void *opaque)
2124 {
2125     sPAPRMachineState *spapr = opaque;
2126     int fd;
2127     int rc = 0;
2128 
2129     /* Iteration header */
2130     if (!spapr->htab_shift) {
2131         qemu_put_be32(f, -1);
2132         return 1;
2133     } else {
2134         qemu_put_be32(f, 0);
2135     }
2136 
2137     if (!spapr->htab) {
2138         assert(kvm_enabled());
2139 
2140         fd = get_htab_fd(spapr);
2141         if (fd < 0) {
2142             return fd;
2143         }
2144 
2145         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2146         if (rc < 0) {
2147             return rc;
2148         }
2149     } else  if (spapr->htab_first_pass) {
2150         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2151     } else {
2152         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2153     }
2154 
2155     htab_save_end_marker(f);
2156 
2157     return rc;
2158 }
2159 
2160 static int htab_save_complete(QEMUFile *f, void *opaque)
2161 {
2162     sPAPRMachineState *spapr = opaque;
2163     int fd;
2164 
2165     /* Iteration header */
2166     if (!spapr->htab_shift) {
2167         qemu_put_be32(f, -1);
2168         return 0;
2169     } else {
2170         qemu_put_be32(f, 0);
2171     }
2172 
2173     if (!spapr->htab) {
2174         int rc;
2175 
2176         assert(kvm_enabled());
2177 
2178         fd = get_htab_fd(spapr);
2179         if (fd < 0) {
2180             return fd;
2181         }
2182 
2183         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2184         if (rc < 0) {
2185             return rc;
2186         }
2187     } else {
2188         if (spapr->htab_first_pass) {
2189             htab_save_first_pass(f, spapr, -1);
2190         }
2191         htab_save_later_pass(f, spapr, -1);
2192     }
2193 
2194     /* End marker */
2195     htab_save_end_marker(f);
2196 
2197     return 0;
2198 }
2199 
2200 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2201 {
2202     sPAPRMachineState *spapr = opaque;
2203     uint32_t section_hdr;
2204     int fd = -1;
2205     Error *local_err = NULL;
2206 
2207     if (version_id < 1 || version_id > 1) {
2208         error_report("htab_load() bad version");
2209         return -EINVAL;
2210     }
2211 
2212     section_hdr = qemu_get_be32(f);
2213 
2214     if (section_hdr == -1) {
2215         spapr_free_hpt(spapr);
2216         return 0;
2217     }
2218 
2219     if (section_hdr) {
2220         /* First section gives the htab size */
2221         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2222         if (local_err) {
2223             error_report_err(local_err);
2224             return -EINVAL;
2225         }
2226         return 0;
2227     }
2228 
2229     if (!spapr->htab) {
2230         assert(kvm_enabled());
2231 
2232         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2233         if (fd < 0) {
2234             error_report_err(local_err);
2235             return fd;
2236         }
2237     }
2238 
2239     while (true) {
2240         uint32_t index;
2241         uint16_t n_valid, n_invalid;
2242 
2243         index = qemu_get_be32(f);
2244         n_valid = qemu_get_be16(f);
2245         n_invalid = qemu_get_be16(f);
2246 
2247         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2248             /* End of Stream */
2249             break;
2250         }
2251 
2252         if ((index + n_valid + n_invalid) >
2253             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2254             /* Bad index in stream */
2255             error_report(
2256                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2257                 index, n_valid, n_invalid, spapr->htab_shift);
2258             return -EINVAL;
2259         }
2260 
2261         if (spapr->htab) {
2262             if (n_valid) {
2263                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2264                                 HASH_PTE_SIZE_64 * n_valid);
2265             }
2266             if (n_invalid) {
2267                 memset(HPTE(spapr->htab, index + n_valid), 0,
2268                        HASH_PTE_SIZE_64 * n_invalid);
2269             }
2270         } else {
2271             int rc;
2272 
2273             assert(fd >= 0);
2274 
2275             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2276             if (rc < 0) {
2277                 return rc;
2278             }
2279         }
2280     }
2281 
2282     if (!spapr->htab) {
2283         assert(fd >= 0);
2284         close(fd);
2285     }
2286 
2287     return 0;
2288 }
2289 
2290 static void htab_save_cleanup(void *opaque)
2291 {
2292     sPAPRMachineState *spapr = opaque;
2293 
2294     close_htab_fd(spapr);
2295 }
2296 
2297 static SaveVMHandlers savevm_htab_handlers = {
2298     .save_setup = htab_save_setup,
2299     .save_live_iterate = htab_save_iterate,
2300     .save_live_complete_precopy = htab_save_complete,
2301     .save_cleanup = htab_save_cleanup,
2302     .load_state = htab_load,
2303 };
2304 
2305 static void spapr_boot_set(void *opaque, const char *boot_device,
2306                            Error **errp)
2307 {
2308     MachineState *machine = MACHINE(opaque);
2309     machine->boot_order = g_strdup(boot_device);
2310 }
2311 
2312 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2313 {
2314     MachineState *machine = MACHINE(spapr);
2315     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2316     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2317     int i;
2318 
2319     for (i = 0; i < nr_lmbs; i++) {
2320         uint64_t addr;
2321 
2322         addr = i * lmb_size + machine->device_memory->base;
2323         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2324                                addr / lmb_size);
2325     }
2326 }
2327 
2328 /*
2329  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2330  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2331  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2332  */
2333 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2334 {
2335     int i;
2336 
2337     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2338         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2339                    " is not aligned to %" PRIu64 " MiB",
2340                    machine->ram_size,
2341                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2342         return;
2343     }
2344 
2345     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2346         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2347                    " is not aligned to %" PRIu64 " MiB",
2348                    machine->ram_size,
2349                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2350         return;
2351     }
2352 
2353     for (i = 0; i < nb_numa_nodes; i++) {
2354         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2355             error_setg(errp,
2356                        "Node %d memory size 0x%" PRIx64
2357                        " is not aligned to %" PRIu64 " MiB",
2358                        i, numa_info[i].node_mem,
2359                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2360             return;
2361         }
2362     }
2363 }
2364 
2365 /* find cpu slot in machine->possible_cpus by core_id */
2366 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2367 {
2368     int index = id / smp_threads;
2369 
2370     if (index >= ms->possible_cpus->len) {
2371         return NULL;
2372     }
2373     if (idx) {
2374         *idx = index;
2375     }
2376     return &ms->possible_cpus->cpus[index];
2377 }
2378 
2379 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2380 {
2381     Error *local_err = NULL;
2382     bool vsmt_user = !!spapr->vsmt;
2383     int kvm_smt = kvmppc_smt_threads();
2384     int ret;
2385 
2386     if (!kvm_enabled() && (smp_threads > 1)) {
2387         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2388                      "on a pseries machine");
2389         goto out;
2390     }
2391     if (!is_power_of_2(smp_threads)) {
2392         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2393                      "machine because it must be a power of 2", smp_threads);
2394         goto out;
2395     }
2396 
2397     /* Detemine the VSMT mode to use: */
2398     if (vsmt_user) {
2399         if (spapr->vsmt < smp_threads) {
2400             error_setg(&local_err, "Cannot support VSMT mode %d"
2401                          " because it must be >= threads/core (%d)",
2402                          spapr->vsmt, smp_threads);
2403             goto out;
2404         }
2405         /* In this case, spapr->vsmt has been set by the command line */
2406     } else {
2407         /*
2408          * Default VSMT value is tricky, because we need it to be as
2409          * consistent as possible (for migration), but this requires
2410          * changing it for at least some existing cases.  We pick 8 as
2411          * the value that we'd get with KVM on POWER8, the
2412          * overwhelmingly common case in production systems.
2413          */
2414         spapr->vsmt = MAX(8, smp_threads);
2415     }
2416 
2417     /* KVM: If necessary, set the SMT mode: */
2418     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2419         ret = kvmppc_set_smt_threads(spapr->vsmt);
2420         if (ret) {
2421             /* Looks like KVM isn't able to change VSMT mode */
2422             error_setg(&local_err,
2423                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2424                        spapr->vsmt, ret);
2425             /* We can live with that if the default one is big enough
2426              * for the number of threads, and a submultiple of the one
2427              * we want.  In this case we'll waste some vcpu ids, but
2428              * behaviour will be correct */
2429             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2430                 warn_report_err(local_err);
2431                 local_err = NULL;
2432                 goto out;
2433             } else {
2434                 if (!vsmt_user) {
2435                     error_append_hint(&local_err,
2436                                       "On PPC, a VM with %d threads/core"
2437                                       " on a host with %d threads/core"
2438                                       " requires the use of VSMT mode %d.\n",
2439                                       smp_threads, kvm_smt, spapr->vsmt);
2440                 }
2441                 kvmppc_hint_smt_possible(&local_err);
2442                 goto out;
2443             }
2444         }
2445     }
2446     /* else TCG: nothing to do currently */
2447 out:
2448     error_propagate(errp, local_err);
2449 }
2450 
2451 static void spapr_init_cpus(sPAPRMachineState *spapr)
2452 {
2453     MachineState *machine = MACHINE(spapr);
2454     MachineClass *mc = MACHINE_GET_CLASS(machine);
2455     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2456     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2457     const CPUArchIdList *possible_cpus;
2458     int boot_cores_nr = smp_cpus / smp_threads;
2459     int i;
2460 
2461     possible_cpus = mc->possible_cpu_arch_ids(machine);
2462     if (mc->has_hotpluggable_cpus) {
2463         if (smp_cpus % smp_threads) {
2464             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2465                          smp_cpus, smp_threads);
2466             exit(1);
2467         }
2468         if (max_cpus % smp_threads) {
2469             error_report("max_cpus (%u) must be multiple of threads (%u)",
2470                          max_cpus, smp_threads);
2471             exit(1);
2472         }
2473     } else {
2474         if (max_cpus != smp_cpus) {
2475             error_report("This machine version does not support CPU hotplug");
2476             exit(1);
2477         }
2478         boot_cores_nr = possible_cpus->len;
2479     }
2480 
2481     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2482      * call xics_max_server_number() or spapr_vcpu_id().
2483      */
2484     spapr_set_vsmt_mode(spapr, &error_fatal);
2485 
2486     if (smc->pre_2_10_has_unused_icps) {
2487         int i;
2488 
2489         for (i = 0; i < xics_max_server_number(spapr); i++) {
2490             /* Dummy entries get deregistered when real ICPState objects
2491              * are registered during CPU core hotplug.
2492              */
2493             pre_2_10_vmstate_register_dummy_icp(i);
2494         }
2495     }
2496 
2497     for (i = 0; i < possible_cpus->len; i++) {
2498         int core_id = i * smp_threads;
2499 
2500         if (mc->has_hotpluggable_cpus) {
2501             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2502                                    spapr_vcpu_id(spapr, core_id));
2503         }
2504 
2505         if (i < boot_cores_nr) {
2506             Object *core  = object_new(type);
2507             int nr_threads = smp_threads;
2508 
2509             /* Handle the partially filled core for older machine types */
2510             if ((i + 1) * smp_threads >= smp_cpus) {
2511                 nr_threads = smp_cpus - i * smp_threads;
2512             }
2513 
2514             object_property_set_int(core, nr_threads, "nr-threads",
2515                                     &error_fatal);
2516             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2517                                     &error_fatal);
2518             object_property_set_bool(core, true, "realized", &error_fatal);
2519         }
2520     }
2521 }
2522 
2523 /* pSeries LPAR / sPAPR hardware init */
2524 static void spapr_machine_init(MachineState *machine)
2525 {
2526     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2527     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2528     const char *kernel_filename = machine->kernel_filename;
2529     const char *initrd_filename = machine->initrd_filename;
2530     PCIHostState *phb;
2531     int i;
2532     MemoryRegion *sysmem = get_system_memory();
2533     MemoryRegion *ram = g_new(MemoryRegion, 1);
2534     hwaddr node0_size = spapr_node0_size(machine);
2535     long load_limit, fw_size;
2536     char *filename;
2537     Error *resize_hpt_err = NULL;
2538 
2539     msi_nonbroken = true;
2540 
2541     QLIST_INIT(&spapr->phbs);
2542     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2543 
2544     /* Determine capabilities to run with */
2545     spapr_caps_init(spapr);
2546 
2547     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2548     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2549         /*
2550          * If the user explicitly requested a mode we should either
2551          * supply it, or fail completely (which we do below).  But if
2552          * it's not set explicitly, we reset our mode to something
2553          * that works
2554          */
2555         if (resize_hpt_err) {
2556             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2557             error_free(resize_hpt_err);
2558             resize_hpt_err = NULL;
2559         } else {
2560             spapr->resize_hpt = smc->resize_hpt_default;
2561         }
2562     }
2563 
2564     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2565 
2566     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2567         /*
2568          * User requested HPT resize, but this host can't supply it.  Bail out
2569          */
2570         error_report_err(resize_hpt_err);
2571         exit(1);
2572     }
2573 
2574     spapr->rma_size = node0_size;
2575 
2576     /* With KVM, we don't actually know whether KVM supports an
2577      * unbounded RMA (PR KVM) or is limited by the hash table size
2578      * (HV KVM using VRMA), so we always assume the latter
2579      *
2580      * In that case, we also limit the initial allocations for RTAS
2581      * etc... to 256M since we have no way to know what the VRMA size
2582      * is going to be as it depends on the size of the hash table
2583      * which isn't determined yet.
2584      */
2585     if (kvm_enabled()) {
2586         spapr->vrma_adjust = 1;
2587         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2588     }
2589 
2590     /* Actually we don't support unbounded RMA anymore since we added
2591      * proper emulation of HV mode. The max we can get is 16G which
2592      * also happens to be what we configure for PAPR mode so make sure
2593      * we don't do anything bigger than that
2594      */
2595     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2596 
2597     if (spapr->rma_size > node0_size) {
2598         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2599                      spapr->rma_size);
2600         exit(1);
2601     }
2602 
2603     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2604     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2605 
2606     /* Set up Interrupt Controller before we create the VCPUs */
2607     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2608 
2609     /* Set up containers for ibm,client-architecture-support negotiated options
2610      */
2611     spapr->ov5 = spapr_ovec_new();
2612     spapr->ov5_cas = spapr_ovec_new();
2613 
2614     if (smc->dr_lmb_enabled) {
2615         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2616         spapr_validate_node_memory(machine, &error_fatal);
2617     }
2618 
2619     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2620 
2621     /* advertise support for dedicated HP event source to guests */
2622     if (spapr->use_hotplug_event_source) {
2623         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2624     }
2625 
2626     /* advertise support for HPT resizing */
2627     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2628         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2629     }
2630 
2631     /* advertise support for ibm,dyamic-memory-v2 */
2632     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2633 
2634     /* init CPUs */
2635     spapr_init_cpus(spapr);
2636 
2637     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2638         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2639                               spapr->max_compat_pvr)) {
2640         /* KVM and TCG always allow GTSE with radix... */
2641         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2642     }
2643     /* ... but not with hash (currently). */
2644 
2645     if (kvm_enabled()) {
2646         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2647         kvmppc_enable_logical_ci_hcalls();
2648         kvmppc_enable_set_mode_hcall();
2649 
2650         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2651         kvmppc_enable_clear_ref_mod_hcalls();
2652     }
2653 
2654     /* allocate RAM */
2655     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2656                                          machine->ram_size);
2657     memory_region_add_subregion(sysmem, 0, ram);
2658 
2659     /* always allocate the device memory information */
2660     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2661 
2662     /* initialize hotplug memory address space */
2663     if (machine->ram_size < machine->maxram_size) {
2664         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2665         /*
2666          * Limit the number of hotpluggable memory slots to half the number
2667          * slots that KVM supports, leaving the other half for PCI and other
2668          * devices. However ensure that number of slots doesn't drop below 32.
2669          */
2670         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2671                            SPAPR_MAX_RAM_SLOTS;
2672 
2673         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2674             max_memslots = SPAPR_MAX_RAM_SLOTS;
2675         }
2676         if (machine->ram_slots > max_memslots) {
2677             error_report("Specified number of memory slots %"
2678                          PRIu64" exceeds max supported %d",
2679                          machine->ram_slots, max_memslots);
2680             exit(1);
2681         }
2682 
2683         machine->device_memory->base = ROUND_UP(machine->ram_size,
2684                                                 SPAPR_DEVICE_MEM_ALIGN);
2685         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2686                            "device-memory", device_mem_size);
2687         memory_region_add_subregion(sysmem, machine->device_memory->base,
2688                                     &machine->device_memory->mr);
2689     }
2690 
2691     if (smc->dr_lmb_enabled) {
2692         spapr_create_lmb_dr_connectors(spapr);
2693     }
2694 
2695     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2696     if (!filename) {
2697         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2698         exit(1);
2699     }
2700     spapr->rtas_size = get_image_size(filename);
2701     if (spapr->rtas_size < 0) {
2702         error_report("Could not get size of LPAR rtas '%s'", filename);
2703         exit(1);
2704     }
2705     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2706     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2707         error_report("Could not load LPAR rtas '%s'", filename);
2708         exit(1);
2709     }
2710     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2711         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2712                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2713         exit(1);
2714     }
2715     g_free(filename);
2716 
2717     /* Set up RTAS event infrastructure */
2718     spapr_events_init(spapr);
2719 
2720     /* Set up the RTC RTAS interfaces */
2721     spapr_rtc_create(spapr);
2722 
2723     /* Set up VIO bus */
2724     spapr->vio_bus = spapr_vio_bus_init();
2725 
2726     for (i = 0; i < serial_max_hds(); i++) {
2727         if (serial_hd(i)) {
2728             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2729         }
2730     }
2731 
2732     /* We always have at least the nvram device on VIO */
2733     spapr_create_nvram(spapr);
2734 
2735     /* Set up PCI */
2736     spapr_pci_rtas_init();
2737 
2738     phb = spapr_create_phb(spapr, 0);
2739 
2740     for (i = 0; i < nb_nics; i++) {
2741         NICInfo *nd = &nd_table[i];
2742 
2743         if (!nd->model) {
2744             nd->model = g_strdup("spapr-vlan");
2745         }
2746 
2747         if (g_str_equal(nd->model, "spapr-vlan") ||
2748             g_str_equal(nd->model, "ibmveth")) {
2749             spapr_vlan_create(spapr->vio_bus, nd);
2750         } else {
2751             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2752         }
2753     }
2754 
2755     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2756         spapr_vscsi_create(spapr->vio_bus);
2757     }
2758 
2759     /* Graphics */
2760     if (spapr_vga_init(phb->bus, &error_fatal)) {
2761         spapr->has_graphics = true;
2762         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2763     }
2764 
2765     if (machine->usb) {
2766         if (smc->use_ohci_by_default) {
2767             pci_create_simple(phb->bus, -1, "pci-ohci");
2768         } else {
2769             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2770         }
2771 
2772         if (spapr->has_graphics) {
2773             USBBus *usb_bus = usb_bus_find(-1);
2774 
2775             usb_create_simple(usb_bus, "usb-kbd");
2776             usb_create_simple(usb_bus, "usb-mouse");
2777         }
2778     }
2779 
2780     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2781         error_report(
2782             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2783             MIN_RMA_SLOF);
2784         exit(1);
2785     }
2786 
2787     if (kernel_filename) {
2788         uint64_t lowaddr = 0;
2789 
2790         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2791                                       NULL, NULL, &lowaddr, NULL, 1,
2792                                       PPC_ELF_MACHINE, 0, 0);
2793         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2794             spapr->kernel_size = load_elf(kernel_filename,
2795                                           translate_kernel_address, NULL, NULL,
2796                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2797                                           0, 0);
2798             spapr->kernel_le = spapr->kernel_size > 0;
2799         }
2800         if (spapr->kernel_size < 0) {
2801             error_report("error loading %s: %s", kernel_filename,
2802                          load_elf_strerror(spapr->kernel_size));
2803             exit(1);
2804         }
2805 
2806         /* load initrd */
2807         if (initrd_filename) {
2808             /* Try to locate the initrd in the gap between the kernel
2809              * and the firmware. Add a bit of space just in case
2810              */
2811             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2812                                   + 0x1ffff) & ~0xffff;
2813             spapr->initrd_size = load_image_targphys(initrd_filename,
2814                                                      spapr->initrd_base,
2815                                                      load_limit
2816                                                      - spapr->initrd_base);
2817             if (spapr->initrd_size < 0) {
2818                 error_report("could not load initial ram disk '%s'",
2819                              initrd_filename);
2820                 exit(1);
2821             }
2822         }
2823     }
2824 
2825     if (bios_name == NULL) {
2826         bios_name = FW_FILE_NAME;
2827     }
2828     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2829     if (!filename) {
2830         error_report("Could not find LPAR firmware '%s'", bios_name);
2831         exit(1);
2832     }
2833     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2834     if (fw_size <= 0) {
2835         error_report("Could not load LPAR firmware '%s'", filename);
2836         exit(1);
2837     }
2838     g_free(filename);
2839 
2840     /* FIXME: Should register things through the MachineState's qdev
2841      * interface, this is a legacy from the sPAPREnvironment structure
2842      * which predated MachineState but had a similar function */
2843     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2844     register_savevm_live(NULL, "spapr/htab", -1, 1,
2845                          &savevm_htab_handlers, spapr);
2846 
2847     qemu_register_boot_set(spapr_boot_set, spapr);
2848 
2849     if (kvm_enabled()) {
2850         /* to stop and start vmclock */
2851         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2852                                          &spapr->tb);
2853 
2854         kvmppc_spapr_enable_inkernel_multitce();
2855     }
2856 }
2857 
2858 static int spapr_kvm_type(const char *vm_type)
2859 {
2860     if (!vm_type) {
2861         return 0;
2862     }
2863 
2864     if (!strcmp(vm_type, "HV")) {
2865         return 1;
2866     }
2867 
2868     if (!strcmp(vm_type, "PR")) {
2869         return 2;
2870     }
2871 
2872     error_report("Unknown kvm-type specified '%s'", vm_type);
2873     exit(1);
2874 }
2875 
2876 /*
2877  * Implementation of an interface to adjust firmware path
2878  * for the bootindex property handling.
2879  */
2880 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2881                                    DeviceState *dev)
2882 {
2883 #define CAST(type, obj, name) \
2884     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2885     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2886     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2887     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2888 
2889     if (d) {
2890         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2891         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2892         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2893 
2894         if (spapr) {
2895             /*
2896              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2897              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2898              * in the top 16 bits of the 64-bit LUN
2899              */
2900             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2901             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2902                                    (uint64_t)id << 48);
2903         } else if (virtio) {
2904             /*
2905              * We use SRP luns of the form 01000000 | (target << 8) | lun
2906              * in the top 32 bits of the 64-bit LUN
2907              * Note: the quote above is from SLOF and it is wrong,
2908              * the actual binding is:
2909              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2910              */
2911             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2912             if (d->lun >= 256) {
2913                 /* Use the LUN "flat space addressing method" */
2914                 id |= 0x4000;
2915             }
2916             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2917                                    (uint64_t)id << 32);
2918         } else if (usb) {
2919             /*
2920              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2921              * in the top 32 bits of the 64-bit LUN
2922              */
2923             unsigned usb_port = atoi(usb->port->path);
2924             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2925             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2926                                    (uint64_t)id << 32);
2927         }
2928     }
2929 
2930     /*
2931      * SLOF probes the USB devices, and if it recognizes that the device is a
2932      * storage device, it changes its name to "storage" instead of "usb-host",
2933      * and additionally adds a child node for the SCSI LUN, so the correct
2934      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2935      */
2936     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2937         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2938         if (usb_host_dev_is_scsi_storage(usbdev)) {
2939             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2940         }
2941     }
2942 
2943     if (phb) {
2944         /* Replace "pci" with "pci@800000020000000" */
2945         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2946     }
2947 
2948     if (vsc) {
2949         /* Same logic as virtio above */
2950         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2951         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2952     }
2953 
2954     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2955         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2956         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2957         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2958     }
2959 
2960     return NULL;
2961 }
2962 
2963 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2964 {
2965     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2966 
2967     return g_strdup(spapr->kvm_type);
2968 }
2969 
2970 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2971 {
2972     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2973 
2974     g_free(spapr->kvm_type);
2975     spapr->kvm_type = g_strdup(value);
2976 }
2977 
2978 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2979 {
2980     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2981 
2982     return spapr->use_hotplug_event_source;
2983 }
2984 
2985 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2986                                             Error **errp)
2987 {
2988     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2989 
2990     spapr->use_hotplug_event_source = value;
2991 }
2992 
2993 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2994 {
2995     return true;
2996 }
2997 
2998 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2999 {
3000     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3001 
3002     switch (spapr->resize_hpt) {
3003     case SPAPR_RESIZE_HPT_DEFAULT:
3004         return g_strdup("default");
3005     case SPAPR_RESIZE_HPT_DISABLED:
3006         return g_strdup("disabled");
3007     case SPAPR_RESIZE_HPT_ENABLED:
3008         return g_strdup("enabled");
3009     case SPAPR_RESIZE_HPT_REQUIRED:
3010         return g_strdup("required");
3011     }
3012     g_assert_not_reached();
3013 }
3014 
3015 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3016 {
3017     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3018 
3019     if (strcmp(value, "default") == 0) {
3020         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3021     } else if (strcmp(value, "disabled") == 0) {
3022         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3023     } else if (strcmp(value, "enabled") == 0) {
3024         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3025     } else if (strcmp(value, "required") == 0) {
3026         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3027     } else {
3028         error_setg(errp, "Bad value for \"resize-hpt\" property");
3029     }
3030 }
3031 
3032 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3033                                    void *opaque, Error **errp)
3034 {
3035     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3036 }
3037 
3038 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3039                                    void *opaque, Error **errp)
3040 {
3041     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3042 }
3043 
3044 static void spapr_instance_init(Object *obj)
3045 {
3046     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3047 
3048     spapr->htab_fd = -1;
3049     spapr->use_hotplug_event_source = true;
3050     object_property_add_str(obj, "kvm-type",
3051                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3052     object_property_set_description(obj, "kvm-type",
3053                                     "Specifies the KVM virtualization mode (HV, PR)",
3054                                     NULL);
3055     object_property_add_bool(obj, "modern-hotplug-events",
3056                             spapr_get_modern_hotplug_events,
3057                             spapr_set_modern_hotplug_events,
3058                             NULL);
3059     object_property_set_description(obj, "modern-hotplug-events",
3060                                     "Use dedicated hotplug event mechanism in"
3061                                     " place of standard EPOW events when possible"
3062                                     " (required for memory hot-unplug support)",
3063                                     NULL);
3064     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3065                             "Maximum permitted CPU compatibility mode",
3066                             &error_fatal);
3067 
3068     object_property_add_str(obj, "resize-hpt",
3069                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3070     object_property_set_description(obj, "resize-hpt",
3071                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3072                                     NULL);
3073     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3074                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3075     object_property_set_description(obj, "vsmt",
3076                                     "Virtual SMT: KVM behaves as if this were"
3077                                     " the host's SMT mode", &error_abort);
3078     object_property_add_bool(obj, "vfio-no-msix-emulation",
3079                              spapr_get_msix_emulation, NULL, NULL);
3080 }
3081 
3082 static void spapr_machine_finalizefn(Object *obj)
3083 {
3084     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3085 
3086     g_free(spapr->kvm_type);
3087 }
3088 
3089 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3090 {
3091     cpu_synchronize_state(cs);
3092     ppc_cpu_do_system_reset(cs);
3093 }
3094 
3095 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3096 {
3097     CPUState *cs;
3098 
3099     CPU_FOREACH(cs) {
3100         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3101     }
3102 }
3103 
3104 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3105                            uint32_t node, bool dedicated_hp_event_source,
3106                            Error **errp)
3107 {
3108     sPAPRDRConnector *drc;
3109     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3110     int i, fdt_offset, fdt_size;
3111     void *fdt;
3112     uint64_t addr = addr_start;
3113     bool hotplugged = spapr_drc_hotplugged(dev);
3114     Error *local_err = NULL;
3115 
3116     for (i = 0; i < nr_lmbs; i++) {
3117         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3118                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3119         g_assert(drc);
3120 
3121         fdt = create_device_tree(&fdt_size);
3122         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3123                                                 SPAPR_MEMORY_BLOCK_SIZE);
3124 
3125         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3126         if (local_err) {
3127             while (addr > addr_start) {
3128                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3129                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3130                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3131                 spapr_drc_detach(drc);
3132             }
3133             g_free(fdt);
3134             error_propagate(errp, local_err);
3135             return;
3136         }
3137         if (!hotplugged) {
3138             spapr_drc_reset(drc);
3139         }
3140         addr += SPAPR_MEMORY_BLOCK_SIZE;
3141     }
3142     /* send hotplug notification to the
3143      * guest only in case of hotplugged memory
3144      */
3145     if (hotplugged) {
3146         if (dedicated_hp_event_source) {
3147             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3148                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3149             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3150                                                    nr_lmbs,
3151                                                    spapr_drc_index(drc));
3152         } else {
3153             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3154                                            nr_lmbs);
3155         }
3156     }
3157 }
3158 
3159 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3160                               Error **errp)
3161 {
3162     Error *local_err = NULL;
3163     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3164     PCDIMMDevice *dimm = PC_DIMM(dev);
3165     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3166     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3167     uint64_t size, addr;
3168     uint32_t node;
3169 
3170     size = memory_region_size(mr);
3171 
3172     pc_dimm_plug(dev, MACHINE(ms), &local_err);
3173     if (local_err) {
3174         goto out;
3175     }
3176 
3177     addr = object_property_get_uint(OBJECT(dimm),
3178                                     PC_DIMM_ADDR_PROP, &local_err);
3179     if (local_err) {
3180         goto out_unplug;
3181     }
3182 
3183     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3184                                     &error_abort);
3185     spapr_add_lmbs(dev, addr, size, node,
3186                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3187                    &local_err);
3188     if (local_err) {
3189         goto out_unplug;
3190     }
3191 
3192     return;
3193 
3194 out_unplug:
3195     pc_dimm_unplug(dev, MACHINE(ms));
3196 out:
3197     error_propagate(errp, local_err);
3198 }
3199 
3200 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3201                                   Error **errp)
3202 {
3203     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3204     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3205     PCDIMMDevice *dimm = PC_DIMM(dev);
3206     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3207     Error *local_err = NULL;
3208     MemoryRegion *mr;
3209     uint64_t size;
3210     Object *memdev;
3211     hwaddr pagesize;
3212 
3213     if (!smc->dr_lmb_enabled) {
3214         error_setg(errp, "Memory hotplug not supported for this machine");
3215         return;
3216     }
3217 
3218     mr = ddc->get_memory_region(dimm, errp);
3219     if (!mr) {
3220         return;
3221     }
3222     size = memory_region_size(mr);
3223 
3224     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3225         error_setg(errp, "Hotplugged memory size must be a multiple of "
3226                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3227         return;
3228     }
3229 
3230     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3231                                       &error_abort);
3232     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3233     spapr_check_pagesize(spapr, pagesize, &local_err);
3234     if (local_err) {
3235         error_propagate(errp, local_err);
3236         return;
3237     }
3238 
3239     pc_dimm_pre_plug(dev, MACHINE(hotplug_dev), NULL, errp);
3240 }
3241 
3242 struct sPAPRDIMMState {
3243     PCDIMMDevice *dimm;
3244     uint32_t nr_lmbs;
3245     QTAILQ_ENTRY(sPAPRDIMMState) next;
3246 };
3247 
3248 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3249                                                        PCDIMMDevice *dimm)
3250 {
3251     sPAPRDIMMState *dimm_state = NULL;
3252 
3253     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3254         if (dimm_state->dimm == dimm) {
3255             break;
3256         }
3257     }
3258     return dimm_state;
3259 }
3260 
3261 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3262                                                       uint32_t nr_lmbs,
3263                                                       PCDIMMDevice *dimm)
3264 {
3265     sPAPRDIMMState *ds = NULL;
3266 
3267     /*
3268      * If this request is for a DIMM whose removal had failed earlier
3269      * (due to guest's refusal to remove the LMBs), we would have this
3270      * dimm already in the pending_dimm_unplugs list. In that
3271      * case don't add again.
3272      */
3273     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3274     if (!ds) {
3275         ds = g_malloc0(sizeof(sPAPRDIMMState));
3276         ds->nr_lmbs = nr_lmbs;
3277         ds->dimm = dimm;
3278         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3279     }
3280     return ds;
3281 }
3282 
3283 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3284                                               sPAPRDIMMState *dimm_state)
3285 {
3286     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3287     g_free(dimm_state);
3288 }
3289 
3290 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3291                                                         PCDIMMDevice *dimm)
3292 {
3293     sPAPRDRConnector *drc;
3294     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3295     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3296     uint64_t size = memory_region_size(mr);
3297     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3298     uint32_t avail_lmbs = 0;
3299     uint64_t addr_start, addr;
3300     int i;
3301 
3302     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3303                                          &error_abort);
3304 
3305     addr = addr_start;
3306     for (i = 0; i < nr_lmbs; i++) {
3307         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3308                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3309         g_assert(drc);
3310         if (drc->dev) {
3311             avail_lmbs++;
3312         }
3313         addr += SPAPR_MEMORY_BLOCK_SIZE;
3314     }
3315 
3316     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3317 }
3318 
3319 /* Callback to be called during DRC release. */
3320 void spapr_lmb_release(DeviceState *dev)
3321 {
3322     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3323     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3324     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3325 
3326     /* This information will get lost if a migration occurs
3327      * during the unplug process. In this case recover it. */
3328     if (ds == NULL) {
3329         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3330         g_assert(ds);
3331         /* The DRC being examined by the caller at least must be counted */
3332         g_assert(ds->nr_lmbs);
3333     }
3334 
3335     if (--ds->nr_lmbs) {
3336         return;
3337     }
3338 
3339     /*
3340      * Now that all the LMBs have been removed by the guest, call the
3341      * unplug handler chain. This can never fail.
3342      */
3343     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3344 }
3345 
3346 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3347 {
3348     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3349     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3350 
3351     pc_dimm_unplug(dev, MACHINE(hotplug_dev));
3352     object_unparent(OBJECT(dev));
3353     spapr_pending_dimm_unplugs_remove(spapr, ds);
3354 }
3355 
3356 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3357                                         DeviceState *dev, Error **errp)
3358 {
3359     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3360     Error *local_err = NULL;
3361     PCDIMMDevice *dimm = PC_DIMM(dev);
3362     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3363     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3364     uint32_t nr_lmbs;
3365     uint64_t size, addr_start, addr;
3366     int i;
3367     sPAPRDRConnector *drc;
3368 
3369     size = memory_region_size(mr);
3370     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3371 
3372     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3373                                          &local_err);
3374     if (local_err) {
3375         goto out;
3376     }
3377 
3378     /*
3379      * An existing pending dimm state for this DIMM means that there is an
3380      * unplug operation in progress, waiting for the spapr_lmb_release
3381      * callback to complete the job (BQL can't cover that far). In this case,
3382      * bail out to avoid detaching DRCs that were already released.
3383      */
3384     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3385         error_setg(&local_err,
3386                    "Memory unplug already in progress for device %s",
3387                    dev->id);
3388         goto out;
3389     }
3390 
3391     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3392 
3393     addr = addr_start;
3394     for (i = 0; i < nr_lmbs; i++) {
3395         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3396                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3397         g_assert(drc);
3398 
3399         spapr_drc_detach(drc);
3400         addr += SPAPR_MEMORY_BLOCK_SIZE;
3401     }
3402 
3403     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3404                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3405     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3406                                               nr_lmbs, spapr_drc_index(drc));
3407 out:
3408     error_propagate(errp, local_err);
3409 }
3410 
3411 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3412                                            sPAPRMachineState *spapr)
3413 {
3414     PowerPCCPU *cpu = POWERPC_CPU(cs);
3415     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3416     int id = spapr_get_vcpu_id(cpu);
3417     void *fdt;
3418     int offset, fdt_size;
3419     char *nodename;
3420 
3421     fdt = create_device_tree(&fdt_size);
3422     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3423     offset = fdt_add_subnode(fdt, 0, nodename);
3424 
3425     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3426     g_free(nodename);
3427 
3428     *fdt_offset = offset;
3429     return fdt;
3430 }
3431 
3432 /* Callback to be called during DRC release. */
3433 void spapr_core_release(DeviceState *dev)
3434 {
3435     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3436 
3437     /* Call the unplug handler chain. This can never fail. */
3438     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3439 }
3440 
3441 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3442 {
3443     MachineState *ms = MACHINE(hotplug_dev);
3444     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3445     CPUCore *cc = CPU_CORE(dev);
3446     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3447 
3448     if (smc->pre_2_10_has_unused_icps) {
3449         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3450         int i;
3451 
3452         for (i = 0; i < cc->nr_threads; i++) {
3453             CPUState *cs = CPU(sc->threads[i]);
3454 
3455             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3456         }
3457     }
3458 
3459     assert(core_slot);
3460     core_slot->cpu = NULL;
3461     object_unparent(OBJECT(dev));
3462 }
3463 
3464 static
3465 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3466                                Error **errp)
3467 {
3468     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3469     int index;
3470     sPAPRDRConnector *drc;
3471     CPUCore *cc = CPU_CORE(dev);
3472 
3473     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3474         error_setg(errp, "Unable to find CPU core with core-id: %d",
3475                    cc->core_id);
3476         return;
3477     }
3478     if (index == 0) {
3479         error_setg(errp, "Boot CPU core may not be unplugged");
3480         return;
3481     }
3482 
3483     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3484                           spapr_vcpu_id(spapr, cc->core_id));
3485     g_assert(drc);
3486 
3487     spapr_drc_detach(drc);
3488 
3489     spapr_hotplug_req_remove_by_index(drc);
3490 }
3491 
3492 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3493                             Error **errp)
3494 {
3495     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3496     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3497     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3498     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3499     CPUCore *cc = CPU_CORE(dev);
3500     CPUState *cs = CPU(core->threads[0]);
3501     sPAPRDRConnector *drc;
3502     Error *local_err = NULL;
3503     CPUArchId *core_slot;
3504     int index;
3505     bool hotplugged = spapr_drc_hotplugged(dev);
3506 
3507     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3508     if (!core_slot) {
3509         error_setg(errp, "Unable to find CPU core with core-id: %d",
3510                    cc->core_id);
3511         return;
3512     }
3513     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3514                           spapr_vcpu_id(spapr, cc->core_id));
3515 
3516     g_assert(drc || !mc->has_hotpluggable_cpus);
3517 
3518     if (drc) {
3519         void *fdt;
3520         int fdt_offset;
3521 
3522         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3523 
3524         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3525         if (local_err) {
3526             g_free(fdt);
3527             error_propagate(errp, local_err);
3528             return;
3529         }
3530 
3531         if (hotplugged) {
3532             /*
3533              * Send hotplug notification interrupt to the guest only
3534              * in case of hotplugged CPUs.
3535              */
3536             spapr_hotplug_req_add_by_index(drc);
3537         } else {
3538             spapr_drc_reset(drc);
3539         }
3540     }
3541 
3542     core_slot->cpu = OBJECT(dev);
3543 
3544     if (smc->pre_2_10_has_unused_icps) {
3545         int i;
3546 
3547         for (i = 0; i < cc->nr_threads; i++) {
3548             cs = CPU(core->threads[i]);
3549             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3550         }
3551     }
3552 }
3553 
3554 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3555                                 Error **errp)
3556 {
3557     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3558     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3559     Error *local_err = NULL;
3560     CPUCore *cc = CPU_CORE(dev);
3561     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3562     const char *type = object_get_typename(OBJECT(dev));
3563     CPUArchId *core_slot;
3564     int index;
3565 
3566     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3567         error_setg(&local_err, "CPU hotplug not supported for this machine");
3568         goto out;
3569     }
3570 
3571     if (strcmp(base_core_type, type)) {
3572         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3573         goto out;
3574     }
3575 
3576     if (cc->core_id % smp_threads) {
3577         error_setg(&local_err, "invalid core id %d", cc->core_id);
3578         goto out;
3579     }
3580 
3581     /*
3582      * In general we should have homogeneous threads-per-core, but old
3583      * (pre hotplug support) machine types allow the last core to have
3584      * reduced threads as a compatibility hack for when we allowed
3585      * total vcpus not a multiple of threads-per-core.
3586      */
3587     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3588         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3589                    cc->nr_threads, smp_threads);
3590         goto out;
3591     }
3592 
3593     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3594     if (!core_slot) {
3595         error_setg(&local_err, "core id %d out of range", cc->core_id);
3596         goto out;
3597     }
3598 
3599     if (core_slot->cpu) {
3600         error_setg(&local_err, "core %d already populated", cc->core_id);
3601         goto out;
3602     }
3603 
3604     numa_cpu_pre_plug(core_slot, dev, &local_err);
3605 
3606 out:
3607     error_propagate(errp, local_err);
3608 }
3609 
3610 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3611                                       DeviceState *dev, Error **errp)
3612 {
3613     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3614         spapr_memory_plug(hotplug_dev, dev, errp);
3615     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3616         spapr_core_plug(hotplug_dev, dev, errp);
3617     }
3618 }
3619 
3620 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3621                                         DeviceState *dev, Error **errp)
3622 {
3623     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3624         spapr_memory_unplug(hotplug_dev, dev);
3625     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3626         spapr_core_unplug(hotplug_dev, dev);
3627     }
3628 }
3629 
3630 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3631                                                 DeviceState *dev, Error **errp)
3632 {
3633     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3634     MachineClass *mc = MACHINE_GET_CLASS(sms);
3635 
3636     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3637         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3638             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3639         } else {
3640             /* NOTE: this means there is a window after guest reset, prior to
3641              * CAS negotiation, where unplug requests will fail due to the
3642              * capability not being detected yet. This is a bit different than
3643              * the case with PCI unplug, where the events will be queued and
3644              * eventually handled by the guest after boot
3645              */
3646             error_setg(errp, "Memory hot unplug not supported for this guest");
3647         }
3648     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3649         if (!mc->has_hotpluggable_cpus) {
3650             error_setg(errp, "CPU hot unplug not supported on this machine");
3651             return;
3652         }
3653         spapr_core_unplug_request(hotplug_dev, dev, errp);
3654     }
3655 }
3656 
3657 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3658                                           DeviceState *dev, Error **errp)
3659 {
3660     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3661         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3662     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3663         spapr_core_pre_plug(hotplug_dev, dev, errp);
3664     }
3665 }
3666 
3667 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3668                                                  DeviceState *dev)
3669 {
3670     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3671         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3672         return HOTPLUG_HANDLER(machine);
3673     }
3674     return NULL;
3675 }
3676 
3677 static CpuInstanceProperties
3678 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3679 {
3680     CPUArchId *core_slot;
3681     MachineClass *mc = MACHINE_GET_CLASS(machine);
3682 
3683     /* make sure possible_cpu are intialized */
3684     mc->possible_cpu_arch_ids(machine);
3685     /* get CPU core slot containing thread that matches cpu_index */
3686     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3687     assert(core_slot);
3688     return core_slot->props;
3689 }
3690 
3691 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3692 {
3693     return idx / smp_cores % nb_numa_nodes;
3694 }
3695 
3696 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3697 {
3698     int i;
3699     const char *core_type;
3700     int spapr_max_cores = max_cpus / smp_threads;
3701     MachineClass *mc = MACHINE_GET_CLASS(machine);
3702 
3703     if (!mc->has_hotpluggable_cpus) {
3704         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3705     }
3706     if (machine->possible_cpus) {
3707         assert(machine->possible_cpus->len == spapr_max_cores);
3708         return machine->possible_cpus;
3709     }
3710 
3711     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3712     if (!core_type) {
3713         error_report("Unable to find sPAPR CPU Core definition");
3714         exit(1);
3715     }
3716 
3717     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3718                              sizeof(CPUArchId) * spapr_max_cores);
3719     machine->possible_cpus->len = spapr_max_cores;
3720     for (i = 0; i < machine->possible_cpus->len; i++) {
3721         int core_id = i * smp_threads;
3722 
3723         machine->possible_cpus->cpus[i].type = core_type;
3724         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3725         machine->possible_cpus->cpus[i].arch_id = core_id;
3726         machine->possible_cpus->cpus[i].props.has_core_id = true;
3727         machine->possible_cpus->cpus[i].props.core_id = core_id;
3728     }
3729     return machine->possible_cpus;
3730 }
3731 
3732 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3733                                 uint64_t *buid, hwaddr *pio,
3734                                 hwaddr *mmio32, hwaddr *mmio64,
3735                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3736 {
3737     /*
3738      * New-style PHB window placement.
3739      *
3740      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3741      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3742      * windows.
3743      *
3744      * Some guest kernels can't work with MMIO windows above 1<<46
3745      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3746      *
3747      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3748      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3749      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3750      * 1TiB 64-bit MMIO windows for each PHB.
3751      */
3752     const uint64_t base_buid = 0x800000020000000ULL;
3753 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3754                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3755     int i;
3756 
3757     /* Sanity check natural alignments */
3758     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3759     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3760     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3761     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3762     /* Sanity check bounds */
3763     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3764                       SPAPR_PCI_MEM32_WIN_SIZE);
3765     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3766                       SPAPR_PCI_MEM64_WIN_SIZE);
3767 
3768     if (index >= SPAPR_MAX_PHBS) {
3769         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3770                    SPAPR_MAX_PHBS - 1);
3771         return;
3772     }
3773 
3774     *buid = base_buid + index;
3775     for (i = 0; i < n_dma; ++i) {
3776         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3777     }
3778 
3779     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3780     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3781     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3782 }
3783 
3784 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3785 {
3786     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3787 
3788     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3789 }
3790 
3791 static void spapr_ics_resend(XICSFabric *dev)
3792 {
3793     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3794 
3795     ics_resend(spapr->ics);
3796 }
3797 
3798 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3799 {
3800     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3801 
3802     return cpu ? ICP(cpu->intc) : NULL;
3803 }
3804 
3805 #define ICS_IRQ_FREE(ics, srcno)   \
3806     (!((ics)->irqs[(srcno)].flags & (XICS_FLAGS_IRQ_MASK)))
3807 
3808 static int ics_find_free_block(ICSState *ics, int num, int alignnum)
3809 {
3810     int first, i;
3811 
3812     for (first = 0; first < ics->nr_irqs; first += alignnum) {
3813         if (num > (ics->nr_irqs - first)) {
3814             return -1;
3815         }
3816         for (i = first; i < first + num; ++i) {
3817             if (!ICS_IRQ_FREE(ics, i)) {
3818                 break;
3819             }
3820         }
3821         if (i == (first + num)) {
3822             return first;
3823         }
3824     }
3825 
3826     return -1;
3827 }
3828 
3829 int spapr_irq_find(sPAPRMachineState *spapr, int num, bool align, Error **errp)
3830 {
3831     ICSState *ics = spapr->ics;
3832     int first = -1;
3833 
3834     assert(ics);
3835 
3836     /*
3837      * MSIMesage::data is used for storing VIRQ so
3838      * it has to be aligned to num to support multiple
3839      * MSI vectors. MSI-X is not affected by this.
3840      * The hint is used for the first IRQ, the rest should
3841      * be allocated continuously.
3842      */
3843     if (align) {
3844         assert((num == 1) || (num == 2) || (num == 4) ||
3845                (num == 8) || (num == 16) || (num == 32));
3846         first = ics_find_free_block(ics, num, num);
3847     } else {
3848         first = ics_find_free_block(ics, num, 1);
3849     }
3850 
3851     if (first < 0) {
3852         error_setg(errp, "can't find a free %d-IRQ block", num);
3853         return -1;
3854     }
3855 
3856     return first + ics->offset;
3857 }
3858 
3859 int spapr_irq_claim(sPAPRMachineState *spapr, int irq, bool lsi, Error **errp)
3860 {
3861     ICSState *ics = spapr->ics;
3862 
3863     assert(ics);
3864 
3865     if (!ics_valid_irq(ics, irq)) {
3866         error_setg(errp, "IRQ %d is invalid", irq);
3867         return -1;
3868     }
3869 
3870     if (!ICS_IRQ_FREE(ics, irq - ics->offset)) {
3871         error_setg(errp, "IRQ %d is not free", irq);
3872         return -1;
3873     }
3874 
3875     ics_set_irq_type(ics, irq - ics->offset, lsi);
3876     return 0;
3877 }
3878 
3879 void spapr_irq_free(sPAPRMachineState *spapr, int irq, int num)
3880 {
3881     ICSState *ics = spapr->ics;
3882     int srcno = irq - ics->offset;
3883     int i;
3884 
3885     if (ics_valid_irq(ics, irq)) {
3886         trace_spapr_irq_free(0, irq, num);
3887         for (i = srcno; i < srcno + num; ++i) {
3888             if (ICS_IRQ_FREE(ics, i)) {
3889                 trace_spapr_irq_free_warn(0, i + ics->offset);
3890             }
3891             memset(&ics->irqs[i], 0, sizeof(ICSIRQState));
3892         }
3893     }
3894 }
3895 
3896 qemu_irq spapr_qirq(sPAPRMachineState *spapr, int irq)
3897 {
3898     ICSState *ics = spapr->ics;
3899 
3900     if (ics_valid_irq(ics, irq)) {
3901         return ics->qirqs[irq - ics->offset];
3902     }
3903 
3904     return NULL;
3905 }
3906 
3907 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3908                                  Monitor *mon)
3909 {
3910     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3911     CPUState *cs;
3912 
3913     CPU_FOREACH(cs) {
3914         PowerPCCPU *cpu = POWERPC_CPU(cs);
3915 
3916         icp_pic_print_info(ICP(cpu->intc), mon);
3917     }
3918 
3919     ics_pic_print_info(spapr->ics, mon);
3920 }
3921 
3922 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3923 {
3924     return cpu->vcpu_id;
3925 }
3926 
3927 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3928 {
3929     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3930     int vcpu_id;
3931 
3932     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3933 
3934     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3935         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3936         error_append_hint(errp, "Adjust the number of cpus to %d "
3937                           "or try to raise the number of threads per core\n",
3938                           vcpu_id * smp_threads / spapr->vsmt);
3939         return;
3940     }
3941 
3942     cpu->vcpu_id = vcpu_id;
3943 }
3944 
3945 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3946 {
3947     CPUState *cs;
3948 
3949     CPU_FOREACH(cs) {
3950         PowerPCCPU *cpu = POWERPC_CPU(cs);
3951 
3952         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3953             return cpu;
3954         }
3955     }
3956 
3957     return NULL;
3958 }
3959 
3960 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3961 {
3962     MachineClass *mc = MACHINE_CLASS(oc);
3963     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3964     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3965     NMIClass *nc = NMI_CLASS(oc);
3966     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3967     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3968     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3969     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3970 
3971     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3972     mc->ignore_boot_device_suffixes = true;
3973 
3974     /*
3975      * We set up the default / latest behaviour here.  The class_init
3976      * functions for the specific versioned machine types can override
3977      * these details for backwards compatibility
3978      */
3979     mc->init = spapr_machine_init;
3980     mc->reset = spapr_machine_reset;
3981     mc->block_default_type = IF_SCSI;
3982     mc->max_cpus = 1024;
3983     mc->no_parallel = 1;
3984     mc->default_boot_order = "";
3985     mc->default_ram_size = 512 * MiB;
3986     mc->default_display = "std";
3987     mc->kvm_type = spapr_kvm_type;
3988     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3989     mc->pci_allow_0_address = true;
3990     assert(!mc->get_hotplug_handler);
3991     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3992     hc->pre_plug = spapr_machine_device_pre_plug;
3993     hc->plug = spapr_machine_device_plug;
3994     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3995     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3996     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3997     hc->unplug_request = spapr_machine_device_unplug_request;
3998     hc->unplug = spapr_machine_device_unplug;
3999 
4000     smc->dr_lmb_enabled = true;
4001     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4002     mc->has_hotpluggable_cpus = true;
4003     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4004     fwc->get_dev_path = spapr_get_fw_dev_path;
4005     nc->nmi_monitor_handler = spapr_nmi;
4006     smc->phb_placement = spapr_phb_placement;
4007     vhc->hypercall = emulate_spapr_hypercall;
4008     vhc->hpt_mask = spapr_hpt_mask;
4009     vhc->map_hptes = spapr_map_hptes;
4010     vhc->unmap_hptes = spapr_unmap_hptes;
4011     vhc->store_hpte = spapr_store_hpte;
4012     vhc->get_patbe = spapr_get_patbe;
4013     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4014     xic->ics_get = spapr_ics_get;
4015     xic->ics_resend = spapr_ics_resend;
4016     xic->icp_get = spapr_icp_get;
4017     ispc->print_info = spapr_pic_print_info;
4018     /* Force NUMA node memory size to be a multiple of
4019      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4020      * in which LMBs are represented and hot-added
4021      */
4022     mc->numa_mem_align_shift = 28;
4023 
4024     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4025     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4026     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4027     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4028     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4029     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4030     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4031     spapr_caps_add_properties(smc, &error_abort);
4032 }
4033 
4034 static const TypeInfo spapr_machine_info = {
4035     .name          = TYPE_SPAPR_MACHINE,
4036     .parent        = TYPE_MACHINE,
4037     .abstract      = true,
4038     .instance_size = sizeof(sPAPRMachineState),
4039     .instance_init = spapr_instance_init,
4040     .instance_finalize = spapr_machine_finalizefn,
4041     .class_size    = sizeof(sPAPRMachineClass),
4042     .class_init    = spapr_machine_class_init,
4043     .interfaces = (InterfaceInfo[]) {
4044         { TYPE_FW_PATH_PROVIDER },
4045         { TYPE_NMI },
4046         { TYPE_HOTPLUG_HANDLER },
4047         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4048         { TYPE_XICS_FABRIC },
4049         { TYPE_INTERRUPT_STATS_PROVIDER },
4050         { }
4051     },
4052 };
4053 
4054 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4055     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4056                                                     void *data)      \
4057     {                                                                \
4058         MachineClass *mc = MACHINE_CLASS(oc);                        \
4059         spapr_machine_##suffix##_class_options(mc);                  \
4060         if (latest) {                                                \
4061             mc->alias = "pseries";                                   \
4062             mc->is_default = 1;                                      \
4063         }                                                            \
4064     }                                                                \
4065     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
4066     {                                                                \
4067         MachineState *machine = MACHINE(obj);                        \
4068         spapr_machine_##suffix##_instance_options(machine);          \
4069     }                                                                \
4070     static const TypeInfo spapr_machine_##suffix##_info = {          \
4071         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4072         .parent = TYPE_SPAPR_MACHINE,                                \
4073         .class_init = spapr_machine_##suffix##_class_init,           \
4074         .instance_init = spapr_machine_##suffix##_instance_init,     \
4075     };                                                               \
4076     static void spapr_machine_register_##suffix(void)                \
4077     {                                                                \
4078         type_register(&spapr_machine_##suffix##_info);               \
4079     }                                                                \
4080     type_init(spapr_machine_register_##suffix)
4081 
4082 /*
4083  * pseries-3.0
4084  */
4085 static void spapr_machine_3_0_instance_options(MachineState *machine)
4086 {
4087 }
4088 
4089 static void spapr_machine_3_0_class_options(MachineClass *mc)
4090 {
4091     /* Defaults for the latest behaviour inherited from the base class */
4092 }
4093 
4094 DEFINE_SPAPR_MACHINE(3_0, "3.0", true);
4095 
4096 /*
4097  * pseries-2.12
4098  */
4099 #define SPAPR_COMPAT_2_12                                              \
4100     HW_COMPAT_2_12                                                     \
4101     {                                                                  \
4102         .driver = TYPE_POWERPC_CPU,                                    \
4103         .property = "pre-3.0-migration",                               \
4104         .value    = "on",                                              \
4105     },                                                                 \
4106     {                                                                  \
4107         .driver = TYPE_SPAPR_CPU_CORE,                                 \
4108         .property = "pre-3.0-migration",                               \
4109         .value    = "on",                                              \
4110     },
4111 
4112 static void spapr_machine_2_12_instance_options(MachineState *machine)
4113 {
4114     spapr_machine_3_0_instance_options(machine);
4115 }
4116 
4117 static void spapr_machine_2_12_class_options(MachineClass *mc)
4118 {
4119     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4120 
4121     spapr_machine_3_0_class_options(mc);
4122     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4123 
4124     /* We depend on kvm_enabled() to choose a default value for the
4125      * hpt-max-page-size capability. Of course we can't do it here
4126      * because this is too early and the HW accelerator isn't initialzed
4127      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4128      */
4129     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4130 }
4131 
4132 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4133 
4134 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
4135 {
4136     spapr_machine_2_12_instance_options(machine);
4137 }
4138 
4139 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4140 {
4141     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4142 
4143     spapr_machine_2_12_class_options(mc);
4144     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4145     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4146     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4147 }
4148 
4149 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4150 
4151 /*
4152  * pseries-2.11
4153  */
4154 #define SPAPR_COMPAT_2_11                                              \
4155     HW_COMPAT_2_11
4156 
4157 static void spapr_machine_2_11_instance_options(MachineState *machine)
4158 {
4159     spapr_machine_2_12_instance_options(machine);
4160 }
4161 
4162 static void spapr_machine_2_11_class_options(MachineClass *mc)
4163 {
4164     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4165 
4166     spapr_machine_2_12_class_options(mc);
4167     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4168     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4169 }
4170 
4171 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4172 
4173 /*
4174  * pseries-2.10
4175  */
4176 #define SPAPR_COMPAT_2_10                                              \
4177     HW_COMPAT_2_10
4178 
4179 static void spapr_machine_2_10_instance_options(MachineState *machine)
4180 {
4181     spapr_machine_2_11_instance_options(machine);
4182 }
4183 
4184 static void spapr_machine_2_10_class_options(MachineClass *mc)
4185 {
4186     spapr_machine_2_11_class_options(mc);
4187     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4188 }
4189 
4190 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4191 
4192 /*
4193  * pseries-2.9
4194  */
4195 #define SPAPR_COMPAT_2_9                                               \
4196     HW_COMPAT_2_9                                                      \
4197     {                                                                  \
4198         .driver = TYPE_POWERPC_CPU,                                    \
4199         .property = "pre-2.10-migration",                              \
4200         .value    = "on",                                              \
4201     },                                                                 \
4202 
4203 static void spapr_machine_2_9_instance_options(MachineState *machine)
4204 {
4205     spapr_machine_2_10_instance_options(machine);
4206 }
4207 
4208 static void spapr_machine_2_9_class_options(MachineClass *mc)
4209 {
4210     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4211 
4212     spapr_machine_2_10_class_options(mc);
4213     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4214     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4215     smc->pre_2_10_has_unused_icps = true;
4216     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4217 }
4218 
4219 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4220 
4221 /*
4222  * pseries-2.8
4223  */
4224 #define SPAPR_COMPAT_2_8                                        \
4225     HW_COMPAT_2_8                                               \
4226     {                                                           \
4227         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4228         .property = "pcie-extended-configuration-space",        \
4229         .value    = "off",                                      \
4230     },
4231 
4232 static void spapr_machine_2_8_instance_options(MachineState *machine)
4233 {
4234     spapr_machine_2_9_instance_options(machine);
4235 }
4236 
4237 static void spapr_machine_2_8_class_options(MachineClass *mc)
4238 {
4239     spapr_machine_2_9_class_options(mc);
4240     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4241     mc->numa_mem_align_shift = 23;
4242 }
4243 
4244 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4245 
4246 /*
4247  * pseries-2.7
4248  */
4249 #define SPAPR_COMPAT_2_7                            \
4250     HW_COMPAT_2_7                                   \
4251     {                                               \
4252         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4253         .property = "mem_win_size",                 \
4254         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4255     },                                              \
4256     {                                               \
4257         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4258         .property = "mem64_win_size",               \
4259         .value    = "0",                            \
4260     },                                              \
4261     {                                               \
4262         .driver = TYPE_POWERPC_CPU,                 \
4263         .property = "pre-2.8-migration",            \
4264         .value    = "on",                           \
4265     },                                              \
4266     {                                               \
4267         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4268         .property = "pre-2.8-migration",            \
4269         .value    = "on",                           \
4270     },
4271 
4272 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4273                               uint64_t *buid, hwaddr *pio,
4274                               hwaddr *mmio32, hwaddr *mmio64,
4275                               unsigned n_dma, uint32_t *liobns, Error **errp)
4276 {
4277     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4278     const uint64_t base_buid = 0x800000020000000ULL;
4279     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4280     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4281     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4282     const uint32_t max_index = 255;
4283     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4284 
4285     uint64_t ram_top = MACHINE(spapr)->ram_size;
4286     hwaddr phb0_base, phb_base;
4287     int i;
4288 
4289     /* Do we have device memory? */
4290     if (MACHINE(spapr)->maxram_size > ram_top) {
4291         /* Can't just use maxram_size, because there may be an
4292          * alignment gap between normal and device memory regions
4293          */
4294         ram_top = MACHINE(spapr)->device_memory->base +
4295             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4296     }
4297 
4298     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4299 
4300     if (index > max_index) {
4301         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4302                    max_index);
4303         return;
4304     }
4305 
4306     *buid = base_buid + index;
4307     for (i = 0; i < n_dma; ++i) {
4308         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4309     }
4310 
4311     phb_base = phb0_base + index * phb_spacing;
4312     *pio = phb_base + pio_offset;
4313     *mmio32 = phb_base + mmio_offset;
4314     /*
4315      * We don't set the 64-bit MMIO window, relying on the PHB's
4316      * fallback behaviour of automatically splitting a large "32-bit"
4317      * window into contiguous 32-bit and 64-bit windows
4318      */
4319 }
4320 
4321 static void spapr_machine_2_7_instance_options(MachineState *machine)
4322 {
4323     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4324 
4325     spapr_machine_2_8_instance_options(machine);
4326     spapr->use_hotplug_event_source = false;
4327 }
4328 
4329 static void spapr_machine_2_7_class_options(MachineClass *mc)
4330 {
4331     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4332 
4333     spapr_machine_2_8_class_options(mc);
4334     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4335     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4336     smc->phb_placement = phb_placement_2_7;
4337 }
4338 
4339 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4340 
4341 /*
4342  * pseries-2.6
4343  */
4344 #define SPAPR_COMPAT_2_6 \
4345     HW_COMPAT_2_6 \
4346     { \
4347         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4348         .property = "ddw",\
4349         .value    = stringify(off),\
4350     },
4351 
4352 static void spapr_machine_2_6_instance_options(MachineState *machine)
4353 {
4354     spapr_machine_2_7_instance_options(machine);
4355 }
4356 
4357 static void spapr_machine_2_6_class_options(MachineClass *mc)
4358 {
4359     spapr_machine_2_7_class_options(mc);
4360     mc->has_hotpluggable_cpus = false;
4361     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4362 }
4363 
4364 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4365 
4366 /*
4367  * pseries-2.5
4368  */
4369 #define SPAPR_COMPAT_2_5 \
4370     HW_COMPAT_2_5 \
4371     { \
4372         .driver   = "spapr-vlan", \
4373         .property = "use-rx-buffer-pools", \
4374         .value    = "off", \
4375     },
4376 
4377 static void spapr_machine_2_5_instance_options(MachineState *machine)
4378 {
4379     spapr_machine_2_6_instance_options(machine);
4380 }
4381 
4382 static void spapr_machine_2_5_class_options(MachineClass *mc)
4383 {
4384     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4385 
4386     spapr_machine_2_6_class_options(mc);
4387     smc->use_ohci_by_default = true;
4388     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4389 }
4390 
4391 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4392 
4393 /*
4394  * pseries-2.4
4395  */
4396 #define SPAPR_COMPAT_2_4 \
4397         HW_COMPAT_2_4
4398 
4399 static void spapr_machine_2_4_instance_options(MachineState *machine)
4400 {
4401     spapr_machine_2_5_instance_options(machine);
4402 }
4403 
4404 static void spapr_machine_2_4_class_options(MachineClass *mc)
4405 {
4406     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4407 
4408     spapr_machine_2_5_class_options(mc);
4409     smc->dr_lmb_enabled = false;
4410     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4411 }
4412 
4413 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4414 
4415 /*
4416  * pseries-2.3
4417  */
4418 #define SPAPR_COMPAT_2_3 \
4419         HW_COMPAT_2_3 \
4420         {\
4421             .driver   = "spapr-pci-host-bridge",\
4422             .property = "dynamic-reconfiguration",\
4423             .value    = "off",\
4424         },
4425 
4426 static void spapr_machine_2_3_instance_options(MachineState *machine)
4427 {
4428     spapr_machine_2_4_instance_options(machine);
4429 }
4430 
4431 static void spapr_machine_2_3_class_options(MachineClass *mc)
4432 {
4433     spapr_machine_2_4_class_options(mc);
4434     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4435 }
4436 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4437 
4438 /*
4439  * pseries-2.2
4440  */
4441 
4442 #define SPAPR_COMPAT_2_2 \
4443         HW_COMPAT_2_2 \
4444         {\
4445             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4446             .property = "mem_win_size",\
4447             .value    = "0x20000000",\
4448         },
4449 
4450 static void spapr_machine_2_2_instance_options(MachineState *machine)
4451 {
4452     spapr_machine_2_3_instance_options(machine);
4453     machine->suppress_vmdesc = true;
4454 }
4455 
4456 static void spapr_machine_2_2_class_options(MachineClass *mc)
4457 {
4458     spapr_machine_2_3_class_options(mc);
4459     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4460 }
4461 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4462 
4463 /*
4464  * pseries-2.1
4465  */
4466 #define SPAPR_COMPAT_2_1 \
4467         HW_COMPAT_2_1
4468 
4469 static void spapr_machine_2_1_instance_options(MachineState *machine)
4470 {
4471     spapr_machine_2_2_instance_options(machine);
4472 }
4473 
4474 static void spapr_machine_2_1_class_options(MachineClass *mc)
4475 {
4476     spapr_machine_2_2_class_options(mc);
4477     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4478 }
4479 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4480 
4481 static void spapr_machine_register_types(void)
4482 {
4483     type_register_static(&spapr_machine_info);
4484 }
4485 
4486 type_init(spapr_machine_register_types)
4487