xref: /openbmc/qemu/hw/ppc/spapr.c (revision 5d721b78)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
31 #include "hw/hw.h"
32 #include "qemu/log.h"
33 #include "hw/fw-path-provider.h"
34 #include "elf.h"
35 #include "net/net.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
46 #include "qom/cpu.h"
47 
48 #include "hw/boards.h"
49 #include "hw/ppc/ppc.h"
50 #include "hw/loader.h"
51 
52 #include "hw/ppc/fdt.h"
53 #include "hw/ppc/spapr.h"
54 #include "hw/ppc/spapr_vio.h"
55 #include "hw/pci-host/spapr.h"
56 #include "hw/ppc/xics.h"
57 #include "hw/pci/msi.h"
58 
59 #include "hw/pci/pci.h"
60 #include "hw/scsi/scsi.h"
61 #include "hw/virtio/virtio-scsi.h"
62 #include "hw/virtio/vhost-scsi-common.h"
63 
64 #include "exec/address-spaces.h"
65 #include "hw/usb.h"
66 #include "qemu/config-file.h"
67 #include "qemu/error-report.h"
68 #include "trace.h"
69 #include "hw/nmi.h"
70 #include "hw/intc/intc.h"
71 
72 #include "hw/compat.h"
73 #include "qemu/cutils.h"
74 #include "hw/ppc/spapr_cpu_core.h"
75 #include "qmp-commands.h"
76 
77 #include <libfdt.h>
78 
79 /* SLOF memory layout:
80  *
81  * SLOF raw image loaded at 0, copies its romfs right below the flat
82  * device-tree, then position SLOF itself 31M below that
83  *
84  * So we set FW_OVERHEAD to 40MB which should account for all of that
85  * and more
86  *
87  * We load our kernel at 4M, leaving space for SLOF initial image
88  */
89 #define FDT_MAX_SIZE            0x100000
90 #define RTAS_MAX_SIZE           0x10000
91 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
92 #define FW_MAX_SIZE             0x400000
93 #define FW_FILE_NAME            "slof.bin"
94 #define FW_OVERHEAD             0x2800000
95 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
96 
97 #define MIN_RMA_SLOF            128UL
98 
99 #define PHANDLE_XICP            0x00001111
100 
101 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
102 
103 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
104                                   const char *type_ics,
105                                   int nr_irqs, Error **errp)
106 {
107     Error *local_err = NULL;
108     Object *obj;
109 
110     obj = object_new(type_ics);
111     object_property_add_child(OBJECT(spapr), "ics", obj, &error_abort);
112     object_property_add_const_link(obj, ICS_PROP_XICS, OBJECT(spapr),
113                                    &error_abort);
114     object_property_set_int(obj, nr_irqs, "nr-irqs", &local_err);
115     if (local_err) {
116         goto error;
117     }
118     object_property_set_bool(obj, true, "realized", &local_err);
119     if (local_err) {
120         goto error;
121     }
122 
123     return ICS_SIMPLE(obj);
124 
125 error:
126     error_propagate(errp, local_err);
127     return NULL;
128 }
129 
130 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
131 {
132     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
133      * and newer QEMUs don't even have them. In both cases, we don't want
134      * to send anything on the wire.
135      */
136     return false;
137 }
138 
139 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
140     .name = "icp/server",
141     .version_id = 1,
142     .minimum_version_id = 1,
143     .needed = pre_2_10_vmstate_dummy_icp_needed,
144     .fields = (VMStateField[]) {
145         VMSTATE_UNUSED(4), /* uint32_t xirr */
146         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
147         VMSTATE_UNUSED(1), /* uint8_t mfrr */
148         VMSTATE_END_OF_LIST()
149     },
150 };
151 
152 static void pre_2_10_vmstate_register_dummy_icp(int i)
153 {
154     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
155                      (void *)(uintptr_t) i);
156 }
157 
158 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
159 {
160     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
161                        (void *)(uintptr_t) i);
162 }
163 
164 static inline int xics_max_server_number(void)
165 {
166     return DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(), smp_threads);
167 }
168 
169 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
170 {
171     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
172     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
173 
174     if (kvm_enabled()) {
175         if (machine_kernel_irqchip_allowed(machine) &&
176             !xics_kvm_init(spapr, errp)) {
177             spapr->icp_type = TYPE_KVM_ICP;
178             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, errp);
179         }
180         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
181             error_prepend(errp, "kernel_irqchip requested but unavailable: ");
182             return;
183         }
184     }
185 
186     if (!spapr->ics) {
187         xics_spapr_init(spapr);
188         spapr->icp_type = TYPE_ICP;
189         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
190         if (!spapr->ics) {
191             return;
192         }
193     }
194 
195     if (smc->pre_2_10_has_unused_icps) {
196         int i;
197 
198         for (i = 0; i < xics_max_server_number(); i++) {
199             /* Dummy entries get deregistered when real ICPState objects
200              * are registered during CPU core hotplug.
201              */
202             pre_2_10_vmstate_register_dummy_icp(i);
203         }
204     }
205 }
206 
207 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
208                                   int smt_threads)
209 {
210     int i, ret = 0;
211     uint32_t servers_prop[smt_threads];
212     uint32_t gservers_prop[smt_threads * 2];
213     int index = ppc_get_vcpu_dt_id(cpu);
214 
215     if (cpu->compat_pvr) {
216         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
217         if (ret < 0) {
218             return ret;
219         }
220     }
221 
222     /* Build interrupt servers and gservers properties */
223     for (i = 0; i < smt_threads; i++) {
224         servers_prop[i] = cpu_to_be32(index + i);
225         /* Hack, direct the group queues back to cpu 0 */
226         gservers_prop[i*2] = cpu_to_be32(index + i);
227         gservers_prop[i*2 + 1] = 0;
228     }
229     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
230                       servers_prop, sizeof(servers_prop));
231     if (ret < 0) {
232         return ret;
233     }
234     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
235                       gservers_prop, sizeof(gservers_prop));
236 
237     return ret;
238 }
239 
240 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
241 {
242     int index = ppc_get_vcpu_dt_id(cpu);
243     uint32_t associativity[] = {cpu_to_be32(0x5),
244                                 cpu_to_be32(0x0),
245                                 cpu_to_be32(0x0),
246                                 cpu_to_be32(0x0),
247                                 cpu_to_be32(cpu->node_id),
248                                 cpu_to_be32(index)};
249 
250     /* Advertise NUMA via ibm,associativity */
251     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
252                           sizeof(associativity));
253 }
254 
255 /* Populate the "ibm,pa-features" property */
256 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset,
257                                       bool legacy_guest)
258 {
259     uint8_t pa_features_206[] = { 6, 0,
260         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
261     uint8_t pa_features_207[] = { 24, 0,
262         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
263         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
264         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
265         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
266     uint8_t pa_features_300[] = { 66, 0,
267         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
268         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
269         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
270         /* 6: DS207 */
271         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
272         /* 16: Vector */
273         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
274         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
275         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
276         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
277         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
278         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
279         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
280         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
281         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
282         /* 42: PM, 44: PC RA, 46: SC vec'd */
283         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
284         /* 48: SIMD, 50: QP BFP, 52: String */
285         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
286         /* 54: DecFP, 56: DecI, 58: SHA */
287         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
288         /* 60: NM atomic, 62: RNG */
289         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
290     };
291     uint8_t *pa_features;
292     size_t pa_size;
293 
294     switch (POWERPC_MMU_VER(env->mmu_model)) {
295     case POWERPC_MMU_VER_2_06:
296         pa_features = pa_features_206;
297         pa_size = sizeof(pa_features_206);
298         break;
299     case POWERPC_MMU_VER_2_07:
300         pa_features = pa_features_207;
301         pa_size = sizeof(pa_features_207);
302         break;
303     case POWERPC_MMU_VER_3_00:
304         pa_features = pa_features_300;
305         pa_size = sizeof(pa_features_300);
306         break;
307     default:
308         return;
309     }
310 
311     if (env->ci_large_pages) {
312         /*
313          * Note: we keep CI large pages off by default because a 64K capable
314          * guest provisioned with large pages might otherwise try to map a qemu
315          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
316          * even if that qemu runs on a 4k host.
317          * We dd this bit back here if we are confident this is not an issue
318          */
319         pa_features[3] |= 0x20;
320     }
321     if (kvmppc_has_cap_htm() && pa_size > 24) {
322         pa_features[24] |= 0x80;    /* Transactional memory support */
323     }
324     if (legacy_guest && pa_size > 40) {
325         /* Workaround for broken kernels that attempt (guest) radix
326          * mode when they can't handle it, if they see the radix bit set
327          * in pa-features. So hide it from them. */
328         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
329     }
330 
331     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
332 }
333 
334 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
335 {
336     int ret = 0, offset, cpus_offset;
337     CPUState *cs;
338     char cpu_model[32];
339     int smt = kvmppc_smt_threads();
340     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
341 
342     CPU_FOREACH(cs) {
343         PowerPCCPU *cpu = POWERPC_CPU(cs);
344         CPUPPCState *env = &cpu->env;
345         DeviceClass *dc = DEVICE_GET_CLASS(cs);
346         int index = ppc_get_vcpu_dt_id(cpu);
347         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
348 
349         if ((index % smt) != 0) {
350             continue;
351         }
352 
353         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
354 
355         cpus_offset = fdt_path_offset(fdt, "/cpus");
356         if (cpus_offset < 0) {
357             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
358                                           "cpus");
359             if (cpus_offset < 0) {
360                 return cpus_offset;
361             }
362         }
363         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
364         if (offset < 0) {
365             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
366             if (offset < 0) {
367                 return offset;
368             }
369         }
370 
371         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
372                           pft_size_prop, sizeof(pft_size_prop));
373         if (ret < 0) {
374             return ret;
375         }
376 
377         if (nb_numa_nodes > 1) {
378             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
379             if (ret < 0) {
380                 return ret;
381             }
382         }
383 
384         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
385         if (ret < 0) {
386             return ret;
387         }
388 
389         spapr_populate_pa_features(env, fdt, offset,
390                                          spapr->cas_legacy_guest_workaround);
391     }
392     return ret;
393 }
394 
395 static hwaddr spapr_node0_size(void)
396 {
397     MachineState *machine = MACHINE(qdev_get_machine());
398 
399     if (nb_numa_nodes) {
400         int i;
401         for (i = 0; i < nb_numa_nodes; ++i) {
402             if (numa_info[i].node_mem) {
403                 return MIN(pow2floor(numa_info[i].node_mem),
404                            machine->ram_size);
405             }
406         }
407     }
408     return machine->ram_size;
409 }
410 
411 static void add_str(GString *s, const gchar *s1)
412 {
413     g_string_append_len(s, s1, strlen(s1) + 1);
414 }
415 
416 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
417                                        hwaddr size)
418 {
419     uint32_t associativity[] = {
420         cpu_to_be32(0x4), /* length */
421         cpu_to_be32(0x0), cpu_to_be32(0x0),
422         cpu_to_be32(0x0), cpu_to_be32(nodeid)
423     };
424     char mem_name[32];
425     uint64_t mem_reg_property[2];
426     int off;
427 
428     mem_reg_property[0] = cpu_to_be64(start);
429     mem_reg_property[1] = cpu_to_be64(size);
430 
431     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
432     off = fdt_add_subnode(fdt, 0, mem_name);
433     _FDT(off);
434     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
435     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
436                       sizeof(mem_reg_property))));
437     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
438                       sizeof(associativity))));
439     return off;
440 }
441 
442 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
443 {
444     MachineState *machine = MACHINE(spapr);
445     hwaddr mem_start, node_size;
446     int i, nb_nodes = nb_numa_nodes;
447     NodeInfo *nodes = numa_info;
448     NodeInfo ramnode;
449 
450     /* No NUMA nodes, assume there is just one node with whole RAM */
451     if (!nb_numa_nodes) {
452         nb_nodes = 1;
453         ramnode.node_mem = machine->ram_size;
454         nodes = &ramnode;
455     }
456 
457     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
458         if (!nodes[i].node_mem) {
459             continue;
460         }
461         if (mem_start >= machine->ram_size) {
462             node_size = 0;
463         } else {
464             node_size = nodes[i].node_mem;
465             if (node_size > machine->ram_size - mem_start) {
466                 node_size = machine->ram_size - mem_start;
467             }
468         }
469         if (!mem_start) {
470             /* ppc_spapr_init() checks for rma_size <= node0_size already */
471             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
472             mem_start += spapr->rma_size;
473             node_size -= spapr->rma_size;
474         }
475         for ( ; node_size; ) {
476             hwaddr sizetmp = pow2floor(node_size);
477 
478             /* mem_start != 0 here */
479             if (ctzl(mem_start) < ctzl(sizetmp)) {
480                 sizetmp = 1ULL << ctzl(mem_start);
481             }
482 
483             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
484             node_size -= sizetmp;
485             mem_start += sizetmp;
486         }
487     }
488 
489     return 0;
490 }
491 
492 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
493                                   sPAPRMachineState *spapr)
494 {
495     PowerPCCPU *cpu = POWERPC_CPU(cs);
496     CPUPPCState *env = &cpu->env;
497     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
498     int index = ppc_get_vcpu_dt_id(cpu);
499     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
500                        0xffffffff, 0xffffffff};
501     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
502         : SPAPR_TIMEBASE_FREQ;
503     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
504     uint32_t page_sizes_prop[64];
505     size_t page_sizes_prop_size;
506     uint32_t vcpus_per_socket = smp_threads * smp_cores;
507     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
508     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
509     sPAPRDRConnector *drc;
510     int drc_index;
511     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
512     int i;
513 
514     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
515     if (drc) {
516         drc_index = spapr_drc_index(drc);
517         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
518     }
519 
520     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
521     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
522 
523     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
524     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
525                            env->dcache_line_size)));
526     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
527                            env->dcache_line_size)));
528     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
529                            env->icache_line_size)));
530     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
531                            env->icache_line_size)));
532 
533     if (pcc->l1_dcache_size) {
534         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
535                                pcc->l1_dcache_size)));
536     } else {
537         error_report("Warning: Unknown L1 dcache size for cpu");
538     }
539     if (pcc->l1_icache_size) {
540         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
541                                pcc->l1_icache_size)));
542     } else {
543         error_report("Warning: Unknown L1 icache size for cpu");
544     }
545 
546     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
547     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
548     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
549     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
550     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
551     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
552 
553     if (env->spr_cb[SPR_PURR].oea_read) {
554         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
555     }
556 
557     if (env->mmu_model & POWERPC_MMU_1TSEG) {
558         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
559                           segs, sizeof(segs))));
560     }
561 
562     /* Advertise VMX/VSX (vector extensions) if available
563      *   0 / no property == no vector extensions
564      *   1               == VMX / Altivec available
565      *   2               == VSX available */
566     if (env->insns_flags & PPC_ALTIVEC) {
567         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
568 
569         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
570     }
571 
572     /* Advertise DFP (Decimal Floating Point) if available
573      *   0 / no property == no DFP
574      *   1               == DFP available */
575     if (env->insns_flags2 & PPC2_DFP) {
576         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
577     }
578 
579     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
580                                                   sizeof(page_sizes_prop));
581     if (page_sizes_prop_size) {
582         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
583                           page_sizes_prop, page_sizes_prop_size)));
584     }
585 
586     spapr_populate_pa_features(env, fdt, offset, false);
587 
588     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
589                            cs->cpu_index / vcpus_per_socket)));
590 
591     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
592                       pft_size_prop, sizeof(pft_size_prop))));
593 
594     if (nb_numa_nodes > 1) {
595         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
596     }
597 
598     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
599 
600     if (pcc->radix_page_info) {
601         for (i = 0; i < pcc->radix_page_info->count; i++) {
602             radix_AP_encodings[i] =
603                 cpu_to_be32(pcc->radix_page_info->entries[i]);
604         }
605         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
606                           radix_AP_encodings,
607                           pcc->radix_page_info->count *
608                           sizeof(radix_AP_encodings[0]))));
609     }
610 }
611 
612 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
613 {
614     CPUState *cs;
615     int cpus_offset;
616     char *nodename;
617     int smt = kvmppc_smt_threads();
618 
619     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
620     _FDT(cpus_offset);
621     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
622     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
623 
624     /*
625      * We walk the CPUs in reverse order to ensure that CPU DT nodes
626      * created by fdt_add_subnode() end up in the right order in FDT
627      * for the guest kernel the enumerate the CPUs correctly.
628      */
629     CPU_FOREACH_REVERSE(cs) {
630         PowerPCCPU *cpu = POWERPC_CPU(cs);
631         int index = ppc_get_vcpu_dt_id(cpu);
632         DeviceClass *dc = DEVICE_GET_CLASS(cs);
633         int offset;
634 
635         if ((index % smt) != 0) {
636             continue;
637         }
638 
639         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
640         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
641         g_free(nodename);
642         _FDT(offset);
643         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
644     }
645 
646 }
647 
648 /*
649  * Adds ibm,dynamic-reconfiguration-memory node.
650  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
651  * of this device tree node.
652  */
653 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
654 {
655     MachineState *machine = MACHINE(spapr);
656     int ret, i, offset;
657     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
658     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
659     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
660     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
661                        memory_region_size(&spapr->hotplug_memory.mr)) /
662                        lmb_size;
663     uint32_t *int_buf, *cur_index, buf_len;
664     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
665 
666     /*
667      * Don't create the node if there is no hotpluggable memory
668      */
669     if (machine->ram_size == machine->maxram_size) {
670         return 0;
671     }
672 
673     /*
674      * Allocate enough buffer size to fit in ibm,dynamic-memory
675      * or ibm,associativity-lookup-arrays
676      */
677     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
678               * sizeof(uint32_t);
679     cur_index = int_buf = g_malloc0(buf_len);
680 
681     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
682 
683     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
684                     sizeof(prop_lmb_size));
685     if (ret < 0) {
686         goto out;
687     }
688 
689     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
690     if (ret < 0) {
691         goto out;
692     }
693 
694     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
695     if (ret < 0) {
696         goto out;
697     }
698 
699     /* ibm,dynamic-memory */
700     int_buf[0] = cpu_to_be32(nr_lmbs);
701     cur_index++;
702     for (i = 0; i < nr_lmbs; i++) {
703         uint64_t addr = i * lmb_size;
704         uint32_t *dynamic_memory = cur_index;
705 
706         if (i >= hotplug_lmb_start) {
707             sPAPRDRConnector *drc;
708 
709             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
710             g_assert(drc);
711 
712             dynamic_memory[0] = cpu_to_be32(addr >> 32);
713             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
714             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
715             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
716             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
717             if (memory_region_present(get_system_memory(), addr)) {
718                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
719             } else {
720                 dynamic_memory[5] = cpu_to_be32(0);
721             }
722         } else {
723             /*
724              * LMB information for RMA, boot time RAM and gap b/n RAM and
725              * hotplug memory region -- all these are marked as reserved
726              * and as having no valid DRC.
727              */
728             dynamic_memory[0] = cpu_to_be32(addr >> 32);
729             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
730             dynamic_memory[2] = cpu_to_be32(0);
731             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
732             dynamic_memory[4] = cpu_to_be32(-1);
733             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
734                                             SPAPR_LMB_FLAGS_DRC_INVALID);
735         }
736 
737         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
738     }
739     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
740     if (ret < 0) {
741         goto out;
742     }
743 
744     /* ibm,associativity-lookup-arrays */
745     cur_index = int_buf;
746     int_buf[0] = cpu_to_be32(nr_nodes);
747     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
748     cur_index += 2;
749     for (i = 0; i < nr_nodes; i++) {
750         uint32_t associativity[] = {
751             cpu_to_be32(0x0),
752             cpu_to_be32(0x0),
753             cpu_to_be32(0x0),
754             cpu_to_be32(i)
755         };
756         memcpy(cur_index, associativity, sizeof(associativity));
757         cur_index += 4;
758     }
759     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
760             (cur_index - int_buf) * sizeof(uint32_t));
761 out:
762     g_free(int_buf);
763     return ret;
764 }
765 
766 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
767                                 sPAPROptionVector *ov5_updates)
768 {
769     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
770     int ret = 0, offset;
771 
772     /* Generate ibm,dynamic-reconfiguration-memory node if required */
773     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
774         g_assert(smc->dr_lmb_enabled);
775         ret = spapr_populate_drconf_memory(spapr, fdt);
776         if (ret) {
777             goto out;
778         }
779     }
780 
781     offset = fdt_path_offset(fdt, "/chosen");
782     if (offset < 0) {
783         offset = fdt_add_subnode(fdt, 0, "chosen");
784         if (offset < 0) {
785             return offset;
786         }
787     }
788     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
789                                  "ibm,architecture-vec-5");
790 
791 out:
792     return ret;
793 }
794 
795 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
796                                  target_ulong addr, target_ulong size,
797                                  sPAPROptionVector *ov5_updates)
798 {
799     void *fdt, *fdt_skel;
800     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
801 
802     size -= sizeof(hdr);
803 
804     /* Create sceleton */
805     fdt_skel = g_malloc0(size);
806     _FDT((fdt_create(fdt_skel, size)));
807     _FDT((fdt_begin_node(fdt_skel, "")));
808     _FDT((fdt_end_node(fdt_skel)));
809     _FDT((fdt_finish(fdt_skel)));
810     fdt = g_malloc0(size);
811     _FDT((fdt_open_into(fdt_skel, fdt, size)));
812     g_free(fdt_skel);
813 
814     /* Fixup cpu nodes */
815     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
816 
817     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
818         return -1;
819     }
820 
821     /* Pack resulting tree */
822     _FDT((fdt_pack(fdt)));
823 
824     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
825         trace_spapr_cas_failed(size);
826         return -1;
827     }
828 
829     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
830     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
831     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
832     g_free(fdt);
833 
834     return 0;
835 }
836 
837 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
838 {
839     int rtas;
840     GString *hypertas = g_string_sized_new(256);
841     GString *qemu_hypertas = g_string_sized_new(256);
842     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
843     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
844         memory_region_size(&spapr->hotplug_memory.mr);
845     uint32_t lrdr_capacity[] = {
846         cpu_to_be32(max_hotplug_addr >> 32),
847         cpu_to_be32(max_hotplug_addr & 0xffffffff),
848         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
849         cpu_to_be32(max_cpus / smp_threads),
850     };
851 
852     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
853 
854     /* hypertas */
855     add_str(hypertas, "hcall-pft");
856     add_str(hypertas, "hcall-term");
857     add_str(hypertas, "hcall-dabr");
858     add_str(hypertas, "hcall-interrupt");
859     add_str(hypertas, "hcall-tce");
860     add_str(hypertas, "hcall-vio");
861     add_str(hypertas, "hcall-splpar");
862     add_str(hypertas, "hcall-bulk");
863     add_str(hypertas, "hcall-set-mode");
864     add_str(hypertas, "hcall-sprg0");
865     add_str(hypertas, "hcall-copy");
866     add_str(hypertas, "hcall-debug");
867     add_str(qemu_hypertas, "hcall-memop1");
868 
869     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
870         add_str(hypertas, "hcall-multi-tce");
871     }
872     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
873                      hypertas->str, hypertas->len));
874     g_string_free(hypertas, TRUE);
875     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
876                      qemu_hypertas->str, qemu_hypertas->len));
877     g_string_free(qemu_hypertas, TRUE);
878 
879     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
880                      refpoints, sizeof(refpoints)));
881 
882     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
883                           RTAS_ERROR_LOG_MAX));
884     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
885                           RTAS_EVENT_SCAN_RATE));
886 
887     if (msi_nonbroken) {
888         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
889     }
890 
891     /*
892      * According to PAPR, rtas ibm,os-term does not guarantee a return
893      * back to the guest cpu.
894      *
895      * While an additional ibm,extended-os-term property indicates
896      * that rtas call return will always occur. Set this property.
897      */
898     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
899 
900     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
901                      lrdr_capacity, sizeof(lrdr_capacity)));
902 
903     spapr_dt_rtas_tokens(fdt, rtas);
904 }
905 
906 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
907  * that the guest may request and thus the valid values for bytes 24..26 of
908  * option vector 5: */
909 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
910 {
911     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
912 
913     char val[2 * 3] = {
914         24, 0x00, /* Hash/Radix, filled in below. */
915         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
916         26, 0x40, /* Radix options: GTSE == yes. */
917     };
918 
919     if (kvm_enabled()) {
920         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
921             val[1] = 0x80; /* OV5_MMU_BOTH */
922         } else if (kvmppc_has_cap_mmu_radix()) {
923             val[1] = 0x40; /* OV5_MMU_RADIX_300 */
924         } else {
925             val[1] = 0x00; /* Hash */
926         }
927     } else {
928         if (first_ppc_cpu->env.mmu_model & POWERPC_MMU_V3) {
929             /* V3 MMU supports both hash and radix (with dynamic switching) */
930             val[1] = 0xC0;
931         } else {
932             /* Otherwise we can only do hash */
933             val[1] = 0x00;
934         }
935     }
936     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
937                      val, sizeof(val)));
938 }
939 
940 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
941 {
942     MachineState *machine = MACHINE(spapr);
943     int chosen;
944     const char *boot_device = machine->boot_order;
945     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
946     size_t cb = 0;
947     char *bootlist = get_boot_devices_list(&cb, true);
948 
949     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
950 
951     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
952     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
953                           spapr->initrd_base));
954     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
955                           spapr->initrd_base + spapr->initrd_size));
956 
957     if (spapr->kernel_size) {
958         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
959                               cpu_to_be64(spapr->kernel_size) };
960 
961         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
962                          &kprop, sizeof(kprop)));
963         if (spapr->kernel_le) {
964             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
965         }
966     }
967     if (boot_menu) {
968         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
969     }
970     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
971     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
972     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
973 
974     if (cb && bootlist) {
975         int i;
976 
977         for (i = 0; i < cb; i++) {
978             if (bootlist[i] == '\n') {
979                 bootlist[i] = ' ';
980             }
981         }
982         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
983     }
984 
985     if (boot_device && strlen(boot_device)) {
986         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
987     }
988 
989     if (!spapr->has_graphics && stdout_path) {
990         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
991     }
992 
993     spapr_dt_ov5_platform_support(fdt, chosen);
994 
995     g_free(stdout_path);
996     g_free(bootlist);
997 }
998 
999 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1000 {
1001     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1002      * KVM to work under pHyp with some guest co-operation */
1003     int hypervisor;
1004     uint8_t hypercall[16];
1005 
1006     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1007     /* indicate KVM hypercall interface */
1008     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1009     if (kvmppc_has_cap_fixup_hcalls()) {
1010         /*
1011          * Older KVM versions with older guest kernels were broken
1012          * with the magic page, don't allow the guest to map it.
1013          */
1014         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1015                                   sizeof(hypercall))) {
1016             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1017                              hypercall, sizeof(hypercall)));
1018         }
1019     }
1020 }
1021 
1022 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1023                              hwaddr rtas_addr,
1024                              hwaddr rtas_size)
1025 {
1026     MachineState *machine = MACHINE(qdev_get_machine());
1027     MachineClass *mc = MACHINE_GET_CLASS(machine);
1028     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1029     int ret;
1030     void *fdt;
1031     sPAPRPHBState *phb;
1032     char *buf;
1033 
1034     fdt = g_malloc0(FDT_MAX_SIZE);
1035     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1036 
1037     /* Root node */
1038     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1039     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1040     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1041 
1042     /*
1043      * Add info to guest to indentify which host is it being run on
1044      * and what is the uuid of the guest
1045      */
1046     if (kvmppc_get_host_model(&buf)) {
1047         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1048         g_free(buf);
1049     }
1050     if (kvmppc_get_host_serial(&buf)) {
1051         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1052         g_free(buf);
1053     }
1054 
1055     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1056 
1057     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1058     if (qemu_uuid_set) {
1059         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1060     }
1061     g_free(buf);
1062 
1063     if (qemu_get_vm_name()) {
1064         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1065                                 qemu_get_vm_name()));
1066     }
1067 
1068     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1069     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1070 
1071     /* /interrupt controller */
1072     spapr_dt_xics(xics_max_server_number(), fdt, PHANDLE_XICP);
1073 
1074     ret = spapr_populate_memory(spapr, fdt);
1075     if (ret < 0) {
1076         error_report("couldn't setup memory nodes in fdt");
1077         exit(1);
1078     }
1079 
1080     /* /vdevice */
1081     spapr_dt_vdevice(spapr->vio_bus, fdt);
1082 
1083     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1084         ret = spapr_rng_populate_dt(fdt);
1085         if (ret < 0) {
1086             error_report("could not set up rng device in the fdt");
1087             exit(1);
1088         }
1089     }
1090 
1091     QLIST_FOREACH(phb, &spapr->phbs, list) {
1092         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1093         if (ret < 0) {
1094             error_report("couldn't setup PCI devices in fdt");
1095             exit(1);
1096         }
1097     }
1098 
1099     /* cpus */
1100     spapr_populate_cpus_dt_node(fdt, spapr);
1101 
1102     if (smc->dr_lmb_enabled) {
1103         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1104     }
1105 
1106     if (mc->has_hotpluggable_cpus) {
1107         int offset = fdt_path_offset(fdt, "/cpus");
1108         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1109                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1110         if (ret < 0) {
1111             error_report("Couldn't set up CPU DR device tree properties");
1112             exit(1);
1113         }
1114     }
1115 
1116     /* /event-sources */
1117     spapr_dt_events(spapr, fdt);
1118 
1119     /* /rtas */
1120     spapr_dt_rtas(spapr, fdt);
1121 
1122     /* /chosen */
1123     spapr_dt_chosen(spapr, fdt);
1124 
1125     /* /hypervisor */
1126     if (kvm_enabled()) {
1127         spapr_dt_hypervisor(spapr, fdt);
1128     }
1129 
1130     /* Build memory reserve map */
1131     if (spapr->kernel_size) {
1132         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1133     }
1134     if (spapr->initrd_size) {
1135         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1136     }
1137 
1138     /* ibm,client-architecture-support updates */
1139     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1140     if (ret < 0) {
1141         error_report("couldn't setup CAS properties fdt");
1142         exit(1);
1143     }
1144 
1145     return fdt;
1146 }
1147 
1148 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1149 {
1150     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1151 }
1152 
1153 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1154                                     PowerPCCPU *cpu)
1155 {
1156     CPUPPCState *env = &cpu->env;
1157 
1158     /* The TCG path should also be holding the BQL at this point */
1159     g_assert(qemu_mutex_iothread_locked());
1160 
1161     if (msr_pr) {
1162         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1163         env->gpr[3] = H_PRIVILEGE;
1164     } else {
1165         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1166     }
1167 }
1168 
1169 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1170 {
1171     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1172 
1173     return spapr->patb_entry;
1174 }
1175 
1176 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1177 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1178 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1179 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1180 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1181 
1182 /*
1183  * Get the fd to access the kernel htab, re-opening it if necessary
1184  */
1185 static int get_htab_fd(sPAPRMachineState *spapr)
1186 {
1187     if (spapr->htab_fd >= 0) {
1188         return spapr->htab_fd;
1189     }
1190 
1191     spapr->htab_fd = kvmppc_get_htab_fd(false);
1192     if (spapr->htab_fd < 0) {
1193         error_report("Unable to open fd for reading hash table from KVM: %s",
1194                      strerror(errno));
1195     }
1196 
1197     return spapr->htab_fd;
1198 }
1199 
1200 void close_htab_fd(sPAPRMachineState *spapr)
1201 {
1202     if (spapr->htab_fd >= 0) {
1203         close(spapr->htab_fd);
1204     }
1205     spapr->htab_fd = -1;
1206 }
1207 
1208 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1209 {
1210     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1211 
1212     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1213 }
1214 
1215 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1216                                                 hwaddr ptex, int n)
1217 {
1218     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1219     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1220 
1221     if (!spapr->htab) {
1222         /*
1223          * HTAB is controlled by KVM. Fetch into temporary buffer
1224          */
1225         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1226         kvmppc_read_hptes(hptes, ptex, n);
1227         return hptes;
1228     }
1229 
1230     /*
1231      * HTAB is controlled by QEMU. Just point to the internally
1232      * accessible PTEG.
1233      */
1234     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1235 }
1236 
1237 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1238                               const ppc_hash_pte64_t *hptes,
1239                               hwaddr ptex, int n)
1240 {
1241     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1242 
1243     if (!spapr->htab) {
1244         g_free((void *)hptes);
1245     }
1246 
1247     /* Nothing to do for qemu managed HPT */
1248 }
1249 
1250 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1251                              uint64_t pte0, uint64_t pte1)
1252 {
1253     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1254     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1255 
1256     if (!spapr->htab) {
1257         kvmppc_write_hpte(ptex, pte0, pte1);
1258     } else {
1259         stq_p(spapr->htab + offset, pte0);
1260         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1261     }
1262 }
1263 
1264 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1265 {
1266     int shift;
1267 
1268     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1269      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1270      * that's much more than is needed for Linux guests */
1271     shift = ctz64(pow2ceil(ramsize)) - 7;
1272     shift = MAX(shift, 18); /* Minimum architected size */
1273     shift = MIN(shift, 46); /* Maximum architected size */
1274     return shift;
1275 }
1276 
1277 void spapr_free_hpt(sPAPRMachineState *spapr)
1278 {
1279     g_free(spapr->htab);
1280     spapr->htab = NULL;
1281     spapr->htab_shift = 0;
1282     close_htab_fd(spapr);
1283 }
1284 
1285 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1286                                  Error **errp)
1287 {
1288     long rc;
1289 
1290     /* Clean up any HPT info from a previous boot */
1291     spapr_free_hpt(spapr);
1292 
1293     rc = kvmppc_reset_htab(shift);
1294     if (rc < 0) {
1295         /* kernel-side HPT needed, but couldn't allocate one */
1296         error_setg_errno(errp, errno,
1297                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1298                          shift);
1299         /* This is almost certainly fatal, but if the caller really
1300          * wants to carry on with shift == 0, it's welcome to try */
1301     } else if (rc > 0) {
1302         /* kernel-side HPT allocated */
1303         if (rc != shift) {
1304             error_setg(errp,
1305                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1306                        shift, rc);
1307         }
1308 
1309         spapr->htab_shift = shift;
1310         spapr->htab = NULL;
1311     } else {
1312         /* kernel-side HPT not needed, allocate in userspace instead */
1313         size_t size = 1ULL << shift;
1314         int i;
1315 
1316         spapr->htab = qemu_memalign(size, size);
1317         if (!spapr->htab) {
1318             error_setg_errno(errp, errno,
1319                              "Could not allocate HPT of order %d", shift);
1320             return;
1321         }
1322 
1323         memset(spapr->htab, 0, size);
1324         spapr->htab_shift = shift;
1325 
1326         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1327             DIRTY_HPTE(HPTE(spapr->htab, i));
1328         }
1329     }
1330 }
1331 
1332 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1333 {
1334     spapr_reallocate_hpt(spapr,
1335                      spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size),
1336                      &error_fatal);
1337     if (spapr->vrma_adjust) {
1338         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1339                                           spapr->htab_shift);
1340     }
1341     /* We're setting up a hash table, so that means we're not radix */
1342     spapr->patb_entry = 0;
1343 }
1344 
1345 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1346 {
1347     bool matched = false;
1348 
1349     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1350         matched = true;
1351     }
1352 
1353     if (!matched) {
1354         error_report("Device %s is not supported by this machine yet.",
1355                      qdev_fw_name(DEVICE(sbdev)));
1356         exit(1);
1357     }
1358 }
1359 
1360 static void ppc_spapr_reset(void)
1361 {
1362     MachineState *machine = MACHINE(qdev_get_machine());
1363     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1364     PowerPCCPU *first_ppc_cpu;
1365     uint32_t rtas_limit;
1366     hwaddr rtas_addr, fdt_addr;
1367     void *fdt;
1368     int rc;
1369 
1370     /* Check for unknown sysbus devices */
1371     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1372 
1373     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1374         /* If using KVM with radix mode available, VCPUs can be started
1375          * without a HPT because KVM will start them in radix mode.
1376          * Set the GR bit in PATB so that we know there is no HPT. */
1377         spapr->patb_entry = PATBE1_GR;
1378     } else {
1379         spapr_setup_hpt_and_vrma(spapr);
1380     }
1381 
1382     qemu_devices_reset();
1383 
1384     /*
1385      * We place the device tree and RTAS just below either the top of the RMA,
1386      * or just below 2GB, whichever is lowere, so that it can be
1387      * processed with 32-bit real mode code if necessary
1388      */
1389     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1390     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1391     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1392 
1393     /* if this reset wasn't generated by CAS, we should reset our
1394      * negotiated options and start from scratch */
1395     if (!spapr->cas_reboot) {
1396         spapr_ovec_cleanup(spapr->ov5_cas);
1397         spapr->ov5_cas = spapr_ovec_new();
1398 
1399         ppc_set_compat_all(spapr->max_compat_pvr, &error_fatal);
1400     }
1401 
1402     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1403 
1404     spapr_load_rtas(spapr, fdt, rtas_addr);
1405 
1406     rc = fdt_pack(fdt);
1407 
1408     /* Should only fail if we've built a corrupted tree */
1409     assert(rc == 0);
1410 
1411     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1412         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1413                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1414         exit(1);
1415     }
1416 
1417     /* Load the fdt */
1418     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1419     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1420     g_free(fdt);
1421 
1422     /* Set up the entry state */
1423     first_ppc_cpu = POWERPC_CPU(first_cpu);
1424     first_ppc_cpu->env.gpr[3] = fdt_addr;
1425     first_ppc_cpu->env.gpr[5] = 0;
1426     first_cpu->halted = 0;
1427     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1428 
1429     spapr->cas_reboot = false;
1430 }
1431 
1432 static void spapr_create_nvram(sPAPRMachineState *spapr)
1433 {
1434     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1435     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1436 
1437     if (dinfo) {
1438         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1439                             &error_fatal);
1440     }
1441 
1442     qdev_init_nofail(dev);
1443 
1444     spapr->nvram = (struct sPAPRNVRAM *)dev;
1445 }
1446 
1447 static void spapr_rtc_create(sPAPRMachineState *spapr)
1448 {
1449     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1450     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1451                               &error_fatal);
1452     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1453                               &error_fatal);
1454     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1455                               "date", &error_fatal);
1456 }
1457 
1458 /* Returns whether we want to use VGA or not */
1459 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1460 {
1461     switch (vga_interface_type) {
1462     case VGA_NONE:
1463         return false;
1464     case VGA_DEVICE:
1465         return true;
1466     case VGA_STD:
1467     case VGA_VIRTIO:
1468         return pci_vga_init(pci_bus) != NULL;
1469     default:
1470         error_setg(errp,
1471                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1472         return false;
1473     }
1474 }
1475 
1476 static int spapr_post_load(void *opaque, int version_id)
1477 {
1478     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1479     int err = 0;
1480 
1481     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1482         CPUState *cs;
1483         CPU_FOREACH(cs) {
1484             PowerPCCPU *cpu = POWERPC_CPU(cs);
1485             icp_resend(ICP(cpu->intc));
1486         }
1487     }
1488 
1489     /* In earlier versions, there was no separate qdev for the PAPR
1490      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1491      * So when migrating from those versions, poke the incoming offset
1492      * value into the RTC device */
1493     if (version_id < 3) {
1494         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1495     }
1496 
1497     if (spapr->patb_entry) {
1498         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1499         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1500         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1501 
1502         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1503         if (err) {
1504             error_report("Process table config unsupported by the host");
1505             return -EINVAL;
1506         }
1507     }
1508 
1509     return err;
1510 }
1511 
1512 static bool version_before_3(void *opaque, int version_id)
1513 {
1514     return version_id < 3;
1515 }
1516 
1517 static bool spapr_ov5_cas_needed(void *opaque)
1518 {
1519     sPAPRMachineState *spapr = opaque;
1520     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1521     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1522     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1523     bool cas_needed;
1524 
1525     /* Prior to the introduction of sPAPROptionVector, we had two option
1526      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1527      * Both of these options encode machine topology into the device-tree
1528      * in such a way that the now-booted OS should still be able to interact
1529      * appropriately with QEMU regardless of what options were actually
1530      * negotiatied on the source side.
1531      *
1532      * As such, we can avoid migrating the CAS-negotiated options if these
1533      * are the only options available on the current machine/platform.
1534      * Since these are the only options available for pseries-2.7 and
1535      * earlier, this allows us to maintain old->new/new->old migration
1536      * compatibility.
1537      *
1538      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1539      * via default pseries-2.8 machines and explicit command-line parameters.
1540      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1541      * of the actual CAS-negotiated values to continue working properly. For
1542      * example, availability of memory unplug depends on knowing whether
1543      * OV5_HP_EVT was negotiated via CAS.
1544      *
1545      * Thus, for any cases where the set of available CAS-negotiatable
1546      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1547      * include the CAS-negotiated options in the migration stream.
1548      */
1549     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1550     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1551 
1552     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1553      * the mask itself since in the future it's possible "legacy" bits may be
1554      * removed via machine options, which could generate a false positive
1555      * that breaks migration.
1556      */
1557     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1558     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1559 
1560     spapr_ovec_cleanup(ov5_mask);
1561     spapr_ovec_cleanup(ov5_legacy);
1562     spapr_ovec_cleanup(ov5_removed);
1563 
1564     return cas_needed;
1565 }
1566 
1567 static const VMStateDescription vmstate_spapr_ov5_cas = {
1568     .name = "spapr_option_vector_ov5_cas",
1569     .version_id = 1,
1570     .minimum_version_id = 1,
1571     .needed = spapr_ov5_cas_needed,
1572     .fields = (VMStateField[]) {
1573         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1574                                  vmstate_spapr_ovec, sPAPROptionVector),
1575         VMSTATE_END_OF_LIST()
1576     },
1577 };
1578 
1579 static bool spapr_patb_entry_needed(void *opaque)
1580 {
1581     sPAPRMachineState *spapr = opaque;
1582 
1583     return !!spapr->patb_entry;
1584 }
1585 
1586 static const VMStateDescription vmstate_spapr_patb_entry = {
1587     .name = "spapr_patb_entry",
1588     .version_id = 1,
1589     .minimum_version_id = 1,
1590     .needed = spapr_patb_entry_needed,
1591     .fields = (VMStateField[]) {
1592         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1593         VMSTATE_END_OF_LIST()
1594     },
1595 };
1596 
1597 static const VMStateDescription vmstate_spapr = {
1598     .name = "spapr",
1599     .version_id = 3,
1600     .minimum_version_id = 1,
1601     .post_load = spapr_post_load,
1602     .fields = (VMStateField[]) {
1603         /* used to be @next_irq */
1604         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1605 
1606         /* RTC offset */
1607         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1608 
1609         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1610         VMSTATE_END_OF_LIST()
1611     },
1612     .subsections = (const VMStateDescription*[]) {
1613         &vmstate_spapr_ov5_cas,
1614         &vmstate_spapr_patb_entry,
1615         NULL
1616     }
1617 };
1618 
1619 static int htab_save_setup(QEMUFile *f, void *opaque)
1620 {
1621     sPAPRMachineState *spapr = opaque;
1622 
1623     /* "Iteration" header */
1624     if (!spapr->htab_shift) {
1625         qemu_put_be32(f, -1);
1626     } else {
1627         qemu_put_be32(f, spapr->htab_shift);
1628     }
1629 
1630     if (spapr->htab) {
1631         spapr->htab_save_index = 0;
1632         spapr->htab_first_pass = true;
1633     } else {
1634         if (spapr->htab_shift) {
1635             assert(kvm_enabled());
1636         }
1637     }
1638 
1639 
1640     return 0;
1641 }
1642 
1643 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1644                                  int64_t max_ns)
1645 {
1646     bool has_timeout = max_ns != -1;
1647     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1648     int index = spapr->htab_save_index;
1649     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1650 
1651     assert(spapr->htab_first_pass);
1652 
1653     do {
1654         int chunkstart;
1655 
1656         /* Consume invalid HPTEs */
1657         while ((index < htabslots)
1658                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1659             CLEAN_HPTE(HPTE(spapr->htab, index));
1660             index++;
1661         }
1662 
1663         /* Consume valid HPTEs */
1664         chunkstart = index;
1665         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1666                && HPTE_VALID(HPTE(spapr->htab, index))) {
1667             CLEAN_HPTE(HPTE(spapr->htab, index));
1668             index++;
1669         }
1670 
1671         if (index > chunkstart) {
1672             int n_valid = index - chunkstart;
1673 
1674             qemu_put_be32(f, chunkstart);
1675             qemu_put_be16(f, n_valid);
1676             qemu_put_be16(f, 0);
1677             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1678                             HASH_PTE_SIZE_64 * n_valid);
1679 
1680             if (has_timeout &&
1681                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1682                 break;
1683             }
1684         }
1685     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1686 
1687     if (index >= htabslots) {
1688         assert(index == htabslots);
1689         index = 0;
1690         spapr->htab_first_pass = false;
1691     }
1692     spapr->htab_save_index = index;
1693 }
1694 
1695 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1696                                 int64_t max_ns)
1697 {
1698     bool final = max_ns < 0;
1699     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1700     int examined = 0, sent = 0;
1701     int index = spapr->htab_save_index;
1702     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1703 
1704     assert(!spapr->htab_first_pass);
1705 
1706     do {
1707         int chunkstart, invalidstart;
1708 
1709         /* Consume non-dirty HPTEs */
1710         while ((index < htabslots)
1711                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1712             index++;
1713             examined++;
1714         }
1715 
1716         chunkstart = index;
1717         /* Consume valid dirty HPTEs */
1718         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1719                && HPTE_DIRTY(HPTE(spapr->htab, index))
1720                && HPTE_VALID(HPTE(spapr->htab, index))) {
1721             CLEAN_HPTE(HPTE(spapr->htab, index));
1722             index++;
1723             examined++;
1724         }
1725 
1726         invalidstart = index;
1727         /* Consume invalid dirty HPTEs */
1728         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1729                && HPTE_DIRTY(HPTE(spapr->htab, index))
1730                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1731             CLEAN_HPTE(HPTE(spapr->htab, index));
1732             index++;
1733             examined++;
1734         }
1735 
1736         if (index > chunkstart) {
1737             int n_valid = invalidstart - chunkstart;
1738             int n_invalid = index - invalidstart;
1739 
1740             qemu_put_be32(f, chunkstart);
1741             qemu_put_be16(f, n_valid);
1742             qemu_put_be16(f, n_invalid);
1743             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1744                             HASH_PTE_SIZE_64 * n_valid);
1745             sent += index - chunkstart;
1746 
1747             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1748                 break;
1749             }
1750         }
1751 
1752         if (examined >= htabslots) {
1753             break;
1754         }
1755 
1756         if (index >= htabslots) {
1757             assert(index == htabslots);
1758             index = 0;
1759         }
1760     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1761 
1762     if (index >= htabslots) {
1763         assert(index == htabslots);
1764         index = 0;
1765     }
1766 
1767     spapr->htab_save_index = index;
1768 
1769     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1770 }
1771 
1772 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1773 #define MAX_KVM_BUF_SIZE    2048
1774 
1775 static int htab_save_iterate(QEMUFile *f, void *opaque)
1776 {
1777     sPAPRMachineState *spapr = opaque;
1778     int fd;
1779     int rc = 0;
1780 
1781     /* Iteration header */
1782     if (!spapr->htab_shift) {
1783         qemu_put_be32(f, -1);
1784         return 0;
1785     } else {
1786         qemu_put_be32(f, 0);
1787     }
1788 
1789     if (!spapr->htab) {
1790         assert(kvm_enabled());
1791 
1792         fd = get_htab_fd(spapr);
1793         if (fd < 0) {
1794             return fd;
1795         }
1796 
1797         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1798         if (rc < 0) {
1799             return rc;
1800         }
1801     } else  if (spapr->htab_first_pass) {
1802         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1803     } else {
1804         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1805     }
1806 
1807     /* End marker */
1808     qemu_put_be32(f, 0);
1809     qemu_put_be16(f, 0);
1810     qemu_put_be16(f, 0);
1811 
1812     return rc;
1813 }
1814 
1815 static int htab_save_complete(QEMUFile *f, void *opaque)
1816 {
1817     sPAPRMachineState *spapr = opaque;
1818     int fd;
1819 
1820     /* Iteration header */
1821     if (!spapr->htab_shift) {
1822         qemu_put_be32(f, -1);
1823         return 0;
1824     } else {
1825         qemu_put_be32(f, 0);
1826     }
1827 
1828     if (!spapr->htab) {
1829         int rc;
1830 
1831         assert(kvm_enabled());
1832 
1833         fd = get_htab_fd(spapr);
1834         if (fd < 0) {
1835             return fd;
1836         }
1837 
1838         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1839         if (rc < 0) {
1840             return rc;
1841         }
1842     } else {
1843         if (spapr->htab_first_pass) {
1844             htab_save_first_pass(f, spapr, -1);
1845         }
1846         htab_save_later_pass(f, spapr, -1);
1847     }
1848 
1849     /* End marker */
1850     qemu_put_be32(f, 0);
1851     qemu_put_be16(f, 0);
1852     qemu_put_be16(f, 0);
1853 
1854     return 0;
1855 }
1856 
1857 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1858 {
1859     sPAPRMachineState *spapr = opaque;
1860     uint32_t section_hdr;
1861     int fd = -1;
1862 
1863     if (version_id < 1 || version_id > 1) {
1864         error_report("htab_load() bad version");
1865         return -EINVAL;
1866     }
1867 
1868     section_hdr = qemu_get_be32(f);
1869 
1870     if (section_hdr == -1) {
1871         spapr_free_hpt(spapr);
1872         return 0;
1873     }
1874 
1875     if (section_hdr) {
1876         Error *local_err = NULL;
1877 
1878         /* First section gives the htab size */
1879         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1880         if (local_err) {
1881             error_report_err(local_err);
1882             return -EINVAL;
1883         }
1884         return 0;
1885     }
1886 
1887     if (!spapr->htab) {
1888         assert(kvm_enabled());
1889 
1890         fd = kvmppc_get_htab_fd(true);
1891         if (fd < 0) {
1892             error_report("Unable to open fd to restore KVM hash table: %s",
1893                          strerror(errno));
1894         }
1895     }
1896 
1897     while (true) {
1898         uint32_t index;
1899         uint16_t n_valid, n_invalid;
1900 
1901         index = qemu_get_be32(f);
1902         n_valid = qemu_get_be16(f);
1903         n_invalid = qemu_get_be16(f);
1904 
1905         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1906             /* End of Stream */
1907             break;
1908         }
1909 
1910         if ((index + n_valid + n_invalid) >
1911             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1912             /* Bad index in stream */
1913             error_report(
1914                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1915                 index, n_valid, n_invalid, spapr->htab_shift);
1916             return -EINVAL;
1917         }
1918 
1919         if (spapr->htab) {
1920             if (n_valid) {
1921                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1922                                 HASH_PTE_SIZE_64 * n_valid);
1923             }
1924             if (n_invalid) {
1925                 memset(HPTE(spapr->htab, index + n_valid), 0,
1926                        HASH_PTE_SIZE_64 * n_invalid);
1927             }
1928         } else {
1929             int rc;
1930 
1931             assert(fd >= 0);
1932 
1933             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1934             if (rc < 0) {
1935                 return rc;
1936             }
1937         }
1938     }
1939 
1940     if (!spapr->htab) {
1941         assert(fd >= 0);
1942         close(fd);
1943     }
1944 
1945     return 0;
1946 }
1947 
1948 static void htab_save_cleanup(void *opaque)
1949 {
1950     sPAPRMachineState *spapr = opaque;
1951 
1952     close_htab_fd(spapr);
1953 }
1954 
1955 static SaveVMHandlers savevm_htab_handlers = {
1956     .save_setup = htab_save_setup,
1957     .save_live_iterate = htab_save_iterate,
1958     .save_live_complete_precopy = htab_save_complete,
1959     .save_cleanup = htab_save_cleanup,
1960     .load_state = htab_load,
1961 };
1962 
1963 static void spapr_boot_set(void *opaque, const char *boot_device,
1964                            Error **errp)
1965 {
1966     MachineState *machine = MACHINE(qdev_get_machine());
1967     machine->boot_order = g_strdup(boot_device);
1968 }
1969 
1970 /*
1971  * Reset routine for LMB DR devices.
1972  *
1973  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1974  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1975  * when it walks all its children devices. LMB devices reset occurs
1976  * as part of spapr_ppc_reset().
1977  */
1978 static void spapr_drc_reset(void *opaque)
1979 {
1980     sPAPRDRConnector *drc = opaque;
1981     DeviceState *d = DEVICE(drc);
1982 
1983     if (d) {
1984         device_reset(d);
1985     }
1986 }
1987 
1988 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1989 {
1990     MachineState *machine = MACHINE(spapr);
1991     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1992     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1993     int i;
1994 
1995     for (i = 0; i < nr_lmbs; i++) {
1996         sPAPRDRConnector *drc;
1997         uint64_t addr;
1998 
1999         addr = i * lmb_size + spapr->hotplug_memory.base;
2000         drc = spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2001                                      addr/lmb_size);
2002         qemu_register_reset(spapr_drc_reset, drc);
2003     }
2004 }
2005 
2006 /*
2007  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2008  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2009  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2010  */
2011 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2012 {
2013     int i;
2014 
2015     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2016         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2017                    " is not aligned to %llu MiB",
2018                    machine->ram_size,
2019                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2020         return;
2021     }
2022 
2023     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2024         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2025                    " is not aligned to %llu MiB",
2026                    machine->ram_size,
2027                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2028         return;
2029     }
2030 
2031     for (i = 0; i < nb_numa_nodes; i++) {
2032         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2033             error_setg(errp,
2034                        "Node %d memory size 0x%" PRIx64
2035                        " is not aligned to %llu MiB",
2036                        i, numa_info[i].node_mem,
2037                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2038             return;
2039         }
2040     }
2041 }
2042 
2043 /* find cpu slot in machine->possible_cpus by core_id */
2044 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2045 {
2046     int index = id / smp_threads;
2047 
2048     if (index >= ms->possible_cpus->len) {
2049         return NULL;
2050     }
2051     if (idx) {
2052         *idx = index;
2053     }
2054     return &ms->possible_cpus->cpus[index];
2055 }
2056 
2057 static void spapr_init_cpus(sPAPRMachineState *spapr)
2058 {
2059     MachineState *machine = MACHINE(spapr);
2060     MachineClass *mc = MACHINE_GET_CLASS(machine);
2061     char *type = spapr_get_cpu_core_type(machine->cpu_model);
2062     int smt = kvmppc_smt_threads();
2063     const CPUArchIdList *possible_cpus;
2064     int boot_cores_nr = smp_cpus / smp_threads;
2065     int i;
2066 
2067     if (!type) {
2068         error_report("Unable to find sPAPR CPU Core definition");
2069         exit(1);
2070     }
2071 
2072     possible_cpus = mc->possible_cpu_arch_ids(machine);
2073     if (mc->has_hotpluggable_cpus) {
2074         if (smp_cpus % smp_threads) {
2075             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2076                          smp_cpus, smp_threads);
2077             exit(1);
2078         }
2079         if (max_cpus % smp_threads) {
2080             error_report("max_cpus (%u) must be multiple of threads (%u)",
2081                          max_cpus, smp_threads);
2082             exit(1);
2083         }
2084     } else {
2085         if (max_cpus != smp_cpus) {
2086             error_report("This machine version does not support CPU hotplug");
2087             exit(1);
2088         }
2089         boot_cores_nr = possible_cpus->len;
2090     }
2091 
2092     for (i = 0; i < possible_cpus->len; i++) {
2093         int core_id = i * smp_threads;
2094 
2095         if (mc->has_hotpluggable_cpus) {
2096             sPAPRDRConnector *drc =
2097                 spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2098                                        (core_id / smp_threads) * smt);
2099 
2100             qemu_register_reset(spapr_drc_reset, drc);
2101         }
2102 
2103         if (i < boot_cores_nr) {
2104             Object *core  = object_new(type);
2105             int nr_threads = smp_threads;
2106 
2107             /* Handle the partially filled core for older machine types */
2108             if ((i + 1) * smp_threads >= smp_cpus) {
2109                 nr_threads = smp_cpus - i * smp_threads;
2110             }
2111 
2112             object_property_set_int(core, nr_threads, "nr-threads",
2113                                     &error_fatal);
2114             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2115                                     &error_fatal);
2116             object_property_set_bool(core, true, "realized", &error_fatal);
2117         }
2118     }
2119     g_free(type);
2120 }
2121 
2122 /* pSeries LPAR / sPAPR hardware init */
2123 static void ppc_spapr_init(MachineState *machine)
2124 {
2125     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2126     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2127     const char *kernel_filename = machine->kernel_filename;
2128     const char *initrd_filename = machine->initrd_filename;
2129     PCIHostState *phb;
2130     int i;
2131     MemoryRegion *sysmem = get_system_memory();
2132     MemoryRegion *ram = g_new(MemoryRegion, 1);
2133     MemoryRegion *rma_region;
2134     void *rma = NULL;
2135     hwaddr rma_alloc_size;
2136     hwaddr node0_size = spapr_node0_size();
2137     long load_limit, fw_size;
2138     char *filename;
2139 
2140     msi_nonbroken = true;
2141 
2142     QLIST_INIT(&spapr->phbs);
2143     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2144 
2145     /* Allocate RMA if necessary */
2146     rma_alloc_size = kvmppc_alloc_rma(&rma);
2147 
2148     if (rma_alloc_size == -1) {
2149         error_report("Unable to create RMA");
2150         exit(1);
2151     }
2152 
2153     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2154         spapr->rma_size = rma_alloc_size;
2155     } else {
2156         spapr->rma_size = node0_size;
2157 
2158         /* With KVM, we don't actually know whether KVM supports an
2159          * unbounded RMA (PR KVM) or is limited by the hash table size
2160          * (HV KVM using VRMA), so we always assume the latter
2161          *
2162          * In that case, we also limit the initial allocations for RTAS
2163          * etc... to 256M since we have no way to know what the VRMA size
2164          * is going to be as it depends on the size of the hash table
2165          * isn't determined yet.
2166          */
2167         if (kvm_enabled()) {
2168             spapr->vrma_adjust = 1;
2169             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2170         }
2171 
2172         /* Actually we don't support unbounded RMA anymore since we
2173          * added proper emulation of HV mode. The max we can get is
2174          * 16G which also happens to be what we configure for PAPR
2175          * mode so make sure we don't do anything bigger than that
2176          */
2177         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2178     }
2179 
2180     if (spapr->rma_size > node0_size) {
2181         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2182                      spapr->rma_size);
2183         exit(1);
2184     }
2185 
2186     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2187     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2188 
2189     /* Set up Interrupt Controller before we create the VCPUs */
2190     xics_system_init(machine, XICS_IRQS_SPAPR, &error_fatal);
2191 
2192     /* Set up containers for ibm,client-set-architecture negotiated options */
2193     spapr->ov5 = spapr_ovec_new();
2194     spapr->ov5_cas = spapr_ovec_new();
2195 
2196     if (smc->dr_lmb_enabled) {
2197         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2198         spapr_validate_node_memory(machine, &error_fatal);
2199     }
2200 
2201     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2202     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2203         /* KVM and TCG always allow GTSE with radix... */
2204         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2205     }
2206     /* ... but not with hash (currently). */
2207 
2208     /* advertise support for dedicated HP event source to guests */
2209     if (spapr->use_hotplug_event_source) {
2210         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2211     }
2212 
2213     /* init CPUs */
2214     if (machine->cpu_model == NULL) {
2215         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2216     }
2217 
2218     spapr_cpu_parse_features(spapr);
2219 
2220     spapr_init_cpus(spapr);
2221 
2222     if (kvm_enabled()) {
2223         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2224         kvmppc_enable_logical_ci_hcalls();
2225         kvmppc_enable_set_mode_hcall();
2226 
2227         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2228         kvmppc_enable_clear_ref_mod_hcalls();
2229     }
2230 
2231     /* allocate RAM */
2232     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2233                                          machine->ram_size);
2234     memory_region_add_subregion(sysmem, 0, ram);
2235 
2236     if (rma_alloc_size && rma) {
2237         rma_region = g_new(MemoryRegion, 1);
2238         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2239                                    rma_alloc_size, rma);
2240         vmstate_register_ram_global(rma_region);
2241         memory_region_add_subregion(sysmem, 0, rma_region);
2242     }
2243 
2244     /* initialize hotplug memory address space */
2245     if (machine->ram_size < machine->maxram_size) {
2246         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2247         /*
2248          * Limit the number of hotpluggable memory slots to half the number
2249          * slots that KVM supports, leaving the other half for PCI and other
2250          * devices. However ensure that number of slots doesn't drop below 32.
2251          */
2252         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2253                            SPAPR_MAX_RAM_SLOTS;
2254 
2255         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2256             max_memslots = SPAPR_MAX_RAM_SLOTS;
2257         }
2258         if (machine->ram_slots > max_memslots) {
2259             error_report("Specified number of memory slots %"
2260                          PRIu64" exceeds max supported %d",
2261                          machine->ram_slots, max_memslots);
2262             exit(1);
2263         }
2264 
2265         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2266                                               SPAPR_HOTPLUG_MEM_ALIGN);
2267         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2268                            "hotplug-memory", hotplug_mem_size);
2269         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2270                                     &spapr->hotplug_memory.mr);
2271     }
2272 
2273     if (smc->dr_lmb_enabled) {
2274         spapr_create_lmb_dr_connectors(spapr);
2275     }
2276 
2277     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2278     if (!filename) {
2279         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2280         exit(1);
2281     }
2282     spapr->rtas_size = get_image_size(filename);
2283     if (spapr->rtas_size < 0) {
2284         error_report("Could not get size of LPAR rtas '%s'", filename);
2285         exit(1);
2286     }
2287     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2288     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2289         error_report("Could not load LPAR rtas '%s'", filename);
2290         exit(1);
2291     }
2292     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2293         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2294                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2295         exit(1);
2296     }
2297     g_free(filename);
2298 
2299     /* Set up RTAS event infrastructure */
2300     spapr_events_init(spapr);
2301 
2302     /* Set up the RTC RTAS interfaces */
2303     spapr_rtc_create(spapr);
2304 
2305     /* Set up VIO bus */
2306     spapr->vio_bus = spapr_vio_bus_init();
2307 
2308     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2309         if (serial_hds[i]) {
2310             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2311         }
2312     }
2313 
2314     /* We always have at least the nvram device on VIO */
2315     spapr_create_nvram(spapr);
2316 
2317     /* Set up PCI */
2318     spapr_pci_rtas_init();
2319 
2320     phb = spapr_create_phb(spapr, 0);
2321 
2322     for (i = 0; i < nb_nics; i++) {
2323         NICInfo *nd = &nd_table[i];
2324 
2325         if (!nd->model) {
2326             nd->model = g_strdup("ibmveth");
2327         }
2328 
2329         if (strcmp(nd->model, "ibmveth") == 0) {
2330             spapr_vlan_create(spapr->vio_bus, nd);
2331         } else {
2332             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2333         }
2334     }
2335 
2336     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2337         spapr_vscsi_create(spapr->vio_bus);
2338     }
2339 
2340     /* Graphics */
2341     if (spapr_vga_init(phb->bus, &error_fatal)) {
2342         spapr->has_graphics = true;
2343         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2344     }
2345 
2346     if (machine->usb) {
2347         if (smc->use_ohci_by_default) {
2348             pci_create_simple(phb->bus, -1, "pci-ohci");
2349         } else {
2350             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2351         }
2352 
2353         if (spapr->has_graphics) {
2354             USBBus *usb_bus = usb_bus_find(-1);
2355 
2356             usb_create_simple(usb_bus, "usb-kbd");
2357             usb_create_simple(usb_bus, "usb-mouse");
2358         }
2359     }
2360 
2361     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2362         error_report(
2363             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2364             MIN_RMA_SLOF);
2365         exit(1);
2366     }
2367 
2368     if (kernel_filename) {
2369         uint64_t lowaddr = 0;
2370 
2371         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2372                                       NULL, NULL, &lowaddr, NULL, 1,
2373                                       PPC_ELF_MACHINE, 0, 0);
2374         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2375             spapr->kernel_size = load_elf(kernel_filename,
2376                                           translate_kernel_address, NULL, NULL,
2377                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2378                                           0, 0);
2379             spapr->kernel_le = spapr->kernel_size > 0;
2380         }
2381         if (spapr->kernel_size < 0) {
2382             error_report("error loading %s: %s", kernel_filename,
2383                          load_elf_strerror(spapr->kernel_size));
2384             exit(1);
2385         }
2386 
2387         /* load initrd */
2388         if (initrd_filename) {
2389             /* Try to locate the initrd in the gap between the kernel
2390              * and the firmware. Add a bit of space just in case
2391              */
2392             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2393                                   + 0x1ffff) & ~0xffff;
2394             spapr->initrd_size = load_image_targphys(initrd_filename,
2395                                                      spapr->initrd_base,
2396                                                      load_limit
2397                                                      - spapr->initrd_base);
2398             if (spapr->initrd_size < 0) {
2399                 error_report("could not load initial ram disk '%s'",
2400                              initrd_filename);
2401                 exit(1);
2402             }
2403         }
2404     }
2405 
2406     if (bios_name == NULL) {
2407         bios_name = FW_FILE_NAME;
2408     }
2409     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2410     if (!filename) {
2411         error_report("Could not find LPAR firmware '%s'", bios_name);
2412         exit(1);
2413     }
2414     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2415     if (fw_size <= 0) {
2416         error_report("Could not load LPAR firmware '%s'", filename);
2417         exit(1);
2418     }
2419     g_free(filename);
2420 
2421     /* FIXME: Should register things through the MachineState's qdev
2422      * interface, this is a legacy from the sPAPREnvironment structure
2423      * which predated MachineState but had a similar function */
2424     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2425     register_savevm_live(NULL, "spapr/htab", -1, 1,
2426                          &savevm_htab_handlers, spapr);
2427 
2428     qemu_register_boot_set(spapr_boot_set, spapr);
2429 
2430     if (kvm_enabled()) {
2431         /* to stop and start vmclock */
2432         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2433                                          &spapr->tb);
2434 
2435         kvmppc_spapr_enable_inkernel_multitce();
2436     }
2437 }
2438 
2439 static int spapr_kvm_type(const char *vm_type)
2440 {
2441     if (!vm_type) {
2442         return 0;
2443     }
2444 
2445     if (!strcmp(vm_type, "HV")) {
2446         return 1;
2447     }
2448 
2449     if (!strcmp(vm_type, "PR")) {
2450         return 2;
2451     }
2452 
2453     error_report("Unknown kvm-type specified '%s'", vm_type);
2454     exit(1);
2455 }
2456 
2457 /*
2458  * Implementation of an interface to adjust firmware path
2459  * for the bootindex property handling.
2460  */
2461 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2462                                    DeviceState *dev)
2463 {
2464 #define CAST(type, obj, name) \
2465     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2466     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2467     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2468     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2469 
2470     if (d) {
2471         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2472         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2473         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2474 
2475         if (spapr) {
2476             /*
2477              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2478              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2479              * in the top 16 bits of the 64-bit LUN
2480              */
2481             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2482             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2483                                    (uint64_t)id << 48);
2484         } else if (virtio) {
2485             /*
2486              * We use SRP luns of the form 01000000 | (target << 8) | lun
2487              * in the top 32 bits of the 64-bit LUN
2488              * Note: the quote above is from SLOF and it is wrong,
2489              * the actual binding is:
2490              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2491              */
2492             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2493             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2494                                    (uint64_t)id << 32);
2495         } else if (usb) {
2496             /*
2497              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2498              * in the top 32 bits of the 64-bit LUN
2499              */
2500             unsigned usb_port = atoi(usb->port->path);
2501             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2502             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2503                                    (uint64_t)id << 32);
2504         }
2505     }
2506 
2507     /*
2508      * SLOF probes the USB devices, and if it recognizes that the device is a
2509      * storage device, it changes its name to "storage" instead of "usb-host",
2510      * and additionally adds a child node for the SCSI LUN, so the correct
2511      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2512      */
2513     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2514         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2515         if (usb_host_dev_is_scsi_storage(usbdev)) {
2516             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2517         }
2518     }
2519 
2520     if (phb) {
2521         /* Replace "pci" with "pci@800000020000000" */
2522         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2523     }
2524 
2525     if (vsc) {
2526         /* Same logic as virtio above */
2527         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2528         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2529     }
2530 
2531     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2532         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2533         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2534         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2535     }
2536 
2537     return NULL;
2538 }
2539 
2540 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2541 {
2542     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2543 
2544     return g_strdup(spapr->kvm_type);
2545 }
2546 
2547 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2548 {
2549     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2550 
2551     g_free(spapr->kvm_type);
2552     spapr->kvm_type = g_strdup(value);
2553 }
2554 
2555 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2556 {
2557     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2558 
2559     return spapr->use_hotplug_event_source;
2560 }
2561 
2562 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2563                                             Error **errp)
2564 {
2565     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2566 
2567     spapr->use_hotplug_event_source = value;
2568 }
2569 
2570 static void spapr_machine_initfn(Object *obj)
2571 {
2572     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2573 
2574     spapr->htab_fd = -1;
2575     spapr->use_hotplug_event_source = true;
2576     object_property_add_str(obj, "kvm-type",
2577                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2578     object_property_set_description(obj, "kvm-type",
2579                                     "Specifies the KVM virtualization mode (HV, PR)",
2580                                     NULL);
2581     object_property_add_bool(obj, "modern-hotplug-events",
2582                             spapr_get_modern_hotplug_events,
2583                             spapr_set_modern_hotplug_events,
2584                             NULL);
2585     object_property_set_description(obj, "modern-hotplug-events",
2586                                     "Use dedicated hotplug event mechanism in"
2587                                     " place of standard EPOW events when possible"
2588                                     " (required for memory hot-unplug support)",
2589                                     NULL);
2590 
2591     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
2592                             "Maximum permitted CPU compatibility mode",
2593                             &error_fatal);
2594 }
2595 
2596 static void spapr_machine_finalizefn(Object *obj)
2597 {
2598     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2599 
2600     g_free(spapr->kvm_type);
2601 }
2602 
2603 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2604 {
2605     cpu_synchronize_state(cs);
2606     ppc_cpu_do_system_reset(cs);
2607 }
2608 
2609 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2610 {
2611     CPUState *cs;
2612 
2613     CPU_FOREACH(cs) {
2614         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2615     }
2616 }
2617 
2618 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2619                            uint32_t node, bool dedicated_hp_event_source,
2620                            Error **errp)
2621 {
2622     sPAPRDRConnector *drc;
2623     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2624     int i, fdt_offset, fdt_size;
2625     void *fdt;
2626     uint64_t addr = addr_start;
2627 
2628     for (i = 0; i < nr_lmbs; i++) {
2629         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2630                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2631         g_assert(drc);
2632 
2633         fdt = create_device_tree(&fdt_size);
2634         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2635                                                 SPAPR_MEMORY_BLOCK_SIZE);
2636 
2637         spapr_drc_attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2638         addr += SPAPR_MEMORY_BLOCK_SIZE;
2639     }
2640     /* send hotplug notification to the
2641      * guest only in case of hotplugged memory
2642      */
2643     if (dev->hotplugged) {
2644         if (dedicated_hp_event_source) {
2645             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2646                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2647             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2648                                                    nr_lmbs,
2649                                                    spapr_drc_index(drc));
2650         } else {
2651             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2652                                            nr_lmbs);
2653         }
2654     }
2655 }
2656 
2657 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2658                               uint32_t node, Error **errp)
2659 {
2660     Error *local_err = NULL;
2661     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2662     PCDIMMDevice *dimm = PC_DIMM(dev);
2663     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2664     MemoryRegion *mr = ddc->get_memory_region(dimm);
2665     uint64_t align = memory_region_get_alignment(mr);
2666     uint64_t size = memory_region_size(mr);
2667     uint64_t addr;
2668 
2669     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2670     if (local_err) {
2671         goto out;
2672     }
2673 
2674     addr = object_property_get_uint(OBJECT(dimm),
2675                                     PC_DIMM_ADDR_PROP, &local_err);
2676     if (local_err) {
2677         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2678         goto out;
2679     }
2680 
2681     spapr_add_lmbs(dev, addr, size, node,
2682                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2683                    &error_abort);
2684 
2685 out:
2686     error_propagate(errp, local_err);
2687 }
2688 
2689 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2690                                   Error **errp)
2691 {
2692     PCDIMMDevice *dimm = PC_DIMM(dev);
2693     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2694     MemoryRegion *mr = ddc->get_memory_region(dimm);
2695     uint64_t size = memory_region_size(mr);
2696     char *mem_dev;
2697 
2698     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2699         error_setg(errp, "Hotplugged memory size must be a multiple of "
2700                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
2701         return;
2702     }
2703 
2704     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2705     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2706         error_setg(errp, "Memory backend has bad page size. "
2707                    "Use 'memory-backend-file' with correct mem-path.");
2708         goto out;
2709     }
2710 
2711 out:
2712     g_free(mem_dev);
2713 }
2714 
2715 struct sPAPRDIMMState {
2716     PCDIMMDevice *dimm;
2717     uint32_t nr_lmbs;
2718     QTAILQ_ENTRY(sPAPRDIMMState) next;
2719 };
2720 
2721 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
2722                                                        PCDIMMDevice *dimm)
2723 {
2724     sPAPRDIMMState *dimm_state = NULL;
2725 
2726     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
2727         if (dimm_state->dimm == dimm) {
2728             break;
2729         }
2730     }
2731     return dimm_state;
2732 }
2733 
2734 static void spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
2735                                            sPAPRDIMMState *dimm_state)
2736 {
2737     g_assert(!spapr_pending_dimm_unplugs_find(spapr, dimm_state->dimm));
2738     QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, dimm_state, next);
2739 }
2740 
2741 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
2742                                               sPAPRDIMMState *dimm_state)
2743 {
2744     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
2745     g_free(dimm_state);
2746 }
2747 
2748 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
2749                                                         PCDIMMDevice *dimm)
2750 {
2751     sPAPRDRConnector *drc;
2752     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2753     MemoryRegion *mr = ddc->get_memory_region(dimm);
2754     uint64_t size = memory_region_size(mr);
2755     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2756     uint32_t avail_lmbs = 0;
2757     uint64_t addr_start, addr;
2758     int i;
2759     sPAPRDIMMState *ds;
2760 
2761     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2762                                          &error_abort);
2763 
2764     addr = addr_start;
2765     for (i = 0; i < nr_lmbs; i++) {
2766         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2767                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2768         g_assert(drc);
2769         if (drc->dev) {
2770             avail_lmbs++;
2771         }
2772         addr += SPAPR_MEMORY_BLOCK_SIZE;
2773     }
2774 
2775     ds = g_malloc0(sizeof(sPAPRDIMMState));
2776     ds->nr_lmbs = avail_lmbs;
2777     ds->dimm = dimm;
2778     spapr_pending_dimm_unplugs_add(ms, ds);
2779     return ds;
2780 }
2781 
2782 /* Callback to be called during DRC release. */
2783 void spapr_lmb_release(DeviceState *dev)
2784 {
2785     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
2786     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
2787     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
2788 
2789     /* This information will get lost if a migration occurs
2790      * during the unplug process. In this case recover it. */
2791     if (ds == NULL) {
2792         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
2793         /* The DRC being examined by the caller at least must be counted */
2794         g_assert(ds->nr_lmbs);
2795     }
2796 
2797     if (--ds->nr_lmbs) {
2798         return;
2799     }
2800 
2801     spapr_pending_dimm_unplugs_remove(spapr, ds);
2802 
2803     /*
2804      * Now that all the LMBs have been removed by the guest, call the
2805      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2806      */
2807     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2808 }
2809 
2810 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2811                                 Error **errp)
2812 {
2813     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2814     PCDIMMDevice *dimm = PC_DIMM(dev);
2815     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2816     MemoryRegion *mr = ddc->get_memory_region(dimm);
2817 
2818     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2819     object_unparent(OBJECT(dev));
2820 }
2821 
2822 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2823                                         DeviceState *dev, Error **errp)
2824 {
2825     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
2826     Error *local_err = NULL;
2827     PCDIMMDevice *dimm = PC_DIMM(dev);
2828     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2829     MemoryRegion *mr = ddc->get_memory_region(dimm);
2830     uint64_t size = memory_region_size(mr);
2831     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2832     uint64_t addr_start, addr;
2833     int i;
2834     sPAPRDRConnector *drc;
2835     sPAPRDIMMState *ds;
2836 
2837     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
2838                                          &local_err);
2839     if (local_err) {
2840         goto out;
2841     }
2842 
2843     ds = g_malloc0(sizeof(sPAPRDIMMState));
2844     ds->nr_lmbs = nr_lmbs;
2845     ds->dimm = dimm;
2846     spapr_pending_dimm_unplugs_add(spapr, ds);
2847 
2848     addr = addr_start;
2849     for (i = 0; i < nr_lmbs; i++) {
2850         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2851                               addr / SPAPR_MEMORY_BLOCK_SIZE);
2852         g_assert(drc);
2853 
2854         spapr_drc_detach(drc, dev, errp);
2855         addr += SPAPR_MEMORY_BLOCK_SIZE;
2856     }
2857 
2858     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
2859                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2860     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2861                                               nr_lmbs, spapr_drc_index(drc));
2862 out:
2863     error_propagate(errp, local_err);
2864 }
2865 
2866 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2867                                     sPAPRMachineState *spapr)
2868 {
2869     PowerPCCPU *cpu = POWERPC_CPU(cs);
2870     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2871     int id = ppc_get_vcpu_dt_id(cpu);
2872     void *fdt;
2873     int offset, fdt_size;
2874     char *nodename;
2875 
2876     fdt = create_device_tree(&fdt_size);
2877     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2878     offset = fdt_add_subnode(fdt, 0, nodename);
2879 
2880     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2881     g_free(nodename);
2882 
2883     *fdt_offset = offset;
2884     return fdt;
2885 }
2886 
2887 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2888                               Error **errp)
2889 {
2890     MachineState *ms = MACHINE(qdev_get_machine());
2891     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
2892     CPUCore *cc = CPU_CORE(dev);
2893     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2894 
2895     if (smc->pre_2_10_has_unused_icps) {
2896         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
2897         sPAPRCPUCoreClass *scc = SPAPR_CPU_CORE_GET_CLASS(OBJECT(cc));
2898         const char *typename = object_class_get_name(scc->cpu_class);
2899         size_t size = object_type_get_instance_size(typename);
2900         int i;
2901 
2902         for (i = 0; i < cc->nr_threads; i++) {
2903             CPUState *cs = CPU(sc->threads + i * size);
2904 
2905             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
2906         }
2907     }
2908 
2909     assert(core_slot);
2910     core_slot->cpu = NULL;
2911     object_unparent(OBJECT(dev));
2912 }
2913 
2914 /* Callback to be called during DRC release. */
2915 void spapr_core_release(DeviceState *dev)
2916 {
2917     HotplugHandler *hotplug_ctrl;
2918 
2919     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2920     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2921 }
2922 
2923 static
2924 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2925                                Error **errp)
2926 {
2927     int index;
2928     sPAPRDRConnector *drc;
2929     Error *local_err = NULL;
2930     CPUCore *cc = CPU_CORE(dev);
2931     int smt = kvmppc_smt_threads();
2932 
2933     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2934         error_setg(errp, "Unable to find CPU core with core-id: %d",
2935                    cc->core_id);
2936         return;
2937     }
2938     if (index == 0) {
2939         error_setg(errp, "Boot CPU core may not be unplugged");
2940         return;
2941     }
2942 
2943     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
2944     g_assert(drc);
2945 
2946     spapr_drc_detach(drc, dev, &local_err);
2947     if (local_err) {
2948         error_propagate(errp, local_err);
2949         return;
2950     }
2951 
2952     spapr_hotplug_req_remove_by_index(drc);
2953 }
2954 
2955 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2956                             Error **errp)
2957 {
2958     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2959     MachineClass *mc = MACHINE_GET_CLASS(spapr);
2960     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2961     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2962     CPUCore *cc = CPU_CORE(dev);
2963     CPUState *cs = CPU(core->threads);
2964     sPAPRDRConnector *drc;
2965     Error *local_err = NULL;
2966     void *fdt = NULL;
2967     int fdt_offset = 0;
2968     int smt = kvmppc_smt_threads();
2969     CPUArchId *core_slot;
2970     int index;
2971 
2972     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2973     if (!core_slot) {
2974         error_setg(errp, "Unable to find CPU core with core-id: %d",
2975                    cc->core_id);
2976         return;
2977     }
2978     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index * smt);
2979 
2980     g_assert(drc || !mc->has_hotpluggable_cpus);
2981 
2982     /*
2983      * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2984      * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2985      */
2986     if (dev->hotplugged) {
2987         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2988     }
2989 
2990     if (drc) {
2991         spapr_drc_attach(drc, dev, fdt, fdt_offset, !dev->hotplugged,
2992                          &local_err);
2993         if (local_err) {
2994             g_free(fdt);
2995             error_propagate(errp, local_err);
2996             return;
2997         }
2998     }
2999 
3000     if (dev->hotplugged) {
3001         /*
3002          * Send hotplug notification interrupt to the guest only in case
3003          * of hotplugged CPUs.
3004          */
3005         spapr_hotplug_req_add_by_index(drc);
3006     }
3007     core_slot->cpu = OBJECT(dev);
3008 
3009     if (smc->pre_2_10_has_unused_icps) {
3010         sPAPRCPUCoreClass *scc = SPAPR_CPU_CORE_GET_CLASS(OBJECT(cc));
3011         const char *typename = object_class_get_name(scc->cpu_class);
3012         size_t size = object_type_get_instance_size(typename);
3013         int i;
3014 
3015         for (i = 0; i < cc->nr_threads; i++) {
3016             sPAPRCPUCore *sc = SPAPR_CPU_CORE(dev);
3017             void *obj = sc->threads + i * size;
3018 
3019             cs = CPU(obj);
3020             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3021         }
3022     }
3023 }
3024 
3025 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3026                                 Error **errp)
3027 {
3028     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3029     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3030     Error *local_err = NULL;
3031     CPUCore *cc = CPU_CORE(dev);
3032     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
3033     const char *type = object_get_typename(OBJECT(dev));
3034     CPUArchId *core_slot;
3035     int index;
3036 
3037     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3038         error_setg(&local_err, "CPU hotplug not supported for this machine");
3039         goto out;
3040     }
3041 
3042     if (strcmp(base_core_type, type)) {
3043         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3044         goto out;
3045     }
3046 
3047     if (cc->core_id % smp_threads) {
3048         error_setg(&local_err, "invalid core id %d", cc->core_id);
3049         goto out;
3050     }
3051 
3052     /*
3053      * In general we should have homogeneous threads-per-core, but old
3054      * (pre hotplug support) machine types allow the last core to have
3055      * reduced threads as a compatibility hack for when we allowed
3056      * total vcpus not a multiple of threads-per-core.
3057      */
3058     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3059         error_setg(errp, "invalid nr-threads %d, must be %d",
3060                    cc->nr_threads, smp_threads);
3061         return;
3062     }
3063 
3064     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3065     if (!core_slot) {
3066         error_setg(&local_err, "core id %d out of range", cc->core_id);
3067         goto out;
3068     }
3069 
3070     if (core_slot->cpu) {
3071         error_setg(&local_err, "core %d already populated", cc->core_id);
3072         goto out;
3073     }
3074 
3075     numa_cpu_pre_plug(core_slot, dev, &local_err);
3076 
3077 out:
3078     g_free(base_core_type);
3079     error_propagate(errp, local_err);
3080 }
3081 
3082 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3083                                       DeviceState *dev, Error **errp)
3084 {
3085     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
3086 
3087     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3088         int node;
3089 
3090         if (!smc->dr_lmb_enabled) {
3091             error_setg(errp, "Memory hotplug not supported for this machine");
3092             return;
3093         }
3094         node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
3095         if (*errp) {
3096             return;
3097         }
3098         if (node < 0 || node >= MAX_NODES) {
3099             error_setg(errp, "Invaild node %d", node);
3100             return;
3101         }
3102 
3103         /*
3104          * Currently PowerPC kernel doesn't allow hot-adding memory to
3105          * memory-less node, but instead will silently add the memory
3106          * to the first node that has some memory. This causes two
3107          * unexpected behaviours for the user.
3108          *
3109          * - Memory gets hotplugged to a different node than what the user
3110          *   specified.
3111          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
3112          *   to memory-less node, a reboot will set things accordingly
3113          *   and the previously hotplugged memory now ends in the right node.
3114          *   This appears as if some memory moved from one node to another.
3115          *
3116          * So until kernel starts supporting memory hotplug to memory-less
3117          * nodes, just prevent such attempts upfront in QEMU.
3118          */
3119         if (nb_numa_nodes && !numa_info[node].node_mem) {
3120             error_setg(errp, "Can't hotplug memory to memory-less node %d",
3121                        node);
3122             return;
3123         }
3124 
3125         spapr_memory_plug(hotplug_dev, dev, node, errp);
3126     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3127         spapr_core_plug(hotplug_dev, dev, errp);
3128     }
3129 }
3130 
3131 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3132                                       DeviceState *dev, Error **errp)
3133 {
3134     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
3135     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
3136 
3137     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3138         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3139             spapr_memory_unplug(hotplug_dev, dev, errp);
3140         } else {
3141             error_setg(errp, "Memory hot unplug not supported for this guest");
3142         }
3143     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3144         if (!mc->has_hotpluggable_cpus) {
3145             error_setg(errp, "CPU hot unplug not supported on this machine");
3146             return;
3147         }
3148         spapr_core_unplug(hotplug_dev, dev, errp);
3149     }
3150 }
3151 
3152 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3153                                                 DeviceState *dev, Error **errp)
3154 {
3155     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
3156     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
3157 
3158     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3159         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3160             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3161         } else {
3162             /* NOTE: this means there is a window after guest reset, prior to
3163              * CAS negotiation, where unplug requests will fail due to the
3164              * capability not being detected yet. This is a bit different than
3165              * the case with PCI unplug, where the events will be queued and
3166              * eventually handled by the guest after boot
3167              */
3168             error_setg(errp, "Memory hot unplug not supported for this guest");
3169         }
3170     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3171         if (!mc->has_hotpluggable_cpus) {
3172             error_setg(errp, "CPU hot unplug not supported on this machine");
3173             return;
3174         }
3175         spapr_core_unplug_request(hotplug_dev, dev, errp);
3176     }
3177 }
3178 
3179 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3180                                           DeviceState *dev, Error **errp)
3181 {
3182     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3183         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3184     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3185         spapr_core_pre_plug(hotplug_dev, dev, errp);
3186     }
3187 }
3188 
3189 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3190                                                  DeviceState *dev)
3191 {
3192     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3193         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3194         return HOTPLUG_HANDLER(machine);
3195     }
3196     return NULL;
3197 }
3198 
3199 static CpuInstanceProperties
3200 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3201 {
3202     CPUArchId *core_slot;
3203     MachineClass *mc = MACHINE_GET_CLASS(machine);
3204 
3205     /* make sure possible_cpu are intialized */
3206     mc->possible_cpu_arch_ids(machine);
3207     /* get CPU core slot containing thread that matches cpu_index */
3208     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3209     assert(core_slot);
3210     return core_slot->props;
3211 }
3212 
3213 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3214 {
3215     int i;
3216     int spapr_max_cores = max_cpus / smp_threads;
3217     MachineClass *mc = MACHINE_GET_CLASS(machine);
3218 
3219     if (!mc->has_hotpluggable_cpus) {
3220         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3221     }
3222     if (machine->possible_cpus) {
3223         assert(machine->possible_cpus->len == spapr_max_cores);
3224         return machine->possible_cpus;
3225     }
3226 
3227     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3228                              sizeof(CPUArchId) * spapr_max_cores);
3229     machine->possible_cpus->len = spapr_max_cores;
3230     for (i = 0; i < machine->possible_cpus->len; i++) {
3231         int core_id = i * smp_threads;
3232 
3233         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3234         machine->possible_cpus->cpus[i].arch_id = core_id;
3235         machine->possible_cpus->cpus[i].props.has_core_id = true;
3236         machine->possible_cpus->cpus[i].props.core_id = core_id;
3237 
3238         /* default distribution of CPUs over NUMA nodes */
3239         if (nb_numa_nodes) {
3240             /* preset values but do not enable them i.e. 'has_node_id = false',
3241              * numa init code will enable them later if manual mapping wasn't
3242              * present on CLI */
3243             machine->possible_cpus->cpus[i].props.node_id =
3244                 core_id / smp_threads / smp_cores % nb_numa_nodes;
3245         }
3246     }
3247     return machine->possible_cpus;
3248 }
3249 
3250 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3251                                 uint64_t *buid, hwaddr *pio,
3252                                 hwaddr *mmio32, hwaddr *mmio64,
3253                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3254 {
3255     /*
3256      * New-style PHB window placement.
3257      *
3258      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3259      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3260      * windows.
3261      *
3262      * Some guest kernels can't work with MMIO windows above 1<<46
3263      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3264      *
3265      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3266      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3267      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3268      * 1TiB 64-bit MMIO windows for each PHB.
3269      */
3270     const uint64_t base_buid = 0x800000020000000ULL;
3271 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3272                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3273     int i;
3274 
3275     /* Sanity check natural alignments */
3276     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3277     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3278     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3279     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3280     /* Sanity check bounds */
3281     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3282                       SPAPR_PCI_MEM32_WIN_SIZE);
3283     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3284                       SPAPR_PCI_MEM64_WIN_SIZE);
3285 
3286     if (index >= SPAPR_MAX_PHBS) {
3287         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3288                    SPAPR_MAX_PHBS - 1);
3289         return;
3290     }
3291 
3292     *buid = base_buid + index;
3293     for (i = 0; i < n_dma; ++i) {
3294         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3295     }
3296 
3297     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3298     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3299     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3300 }
3301 
3302 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3303 {
3304     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3305 
3306     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3307 }
3308 
3309 static void spapr_ics_resend(XICSFabric *dev)
3310 {
3311     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3312 
3313     ics_resend(spapr->ics);
3314 }
3315 
3316 static ICPState *spapr_icp_get(XICSFabric *xi, int cpu_dt_id)
3317 {
3318     PowerPCCPU *cpu = ppc_get_vcpu_by_dt_id(cpu_dt_id);
3319 
3320     return cpu ? ICP(cpu->intc) : NULL;
3321 }
3322 
3323 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3324                                  Monitor *mon)
3325 {
3326     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3327     CPUState *cs;
3328 
3329     CPU_FOREACH(cs) {
3330         PowerPCCPU *cpu = POWERPC_CPU(cs);
3331 
3332         icp_pic_print_info(ICP(cpu->intc), mon);
3333     }
3334 
3335     ics_pic_print_info(spapr->ics, mon);
3336 }
3337 
3338 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3339 {
3340     MachineClass *mc = MACHINE_CLASS(oc);
3341     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3342     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3343     NMIClass *nc = NMI_CLASS(oc);
3344     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3345     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3346     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3347     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3348 
3349     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3350 
3351     /*
3352      * We set up the default / latest behaviour here.  The class_init
3353      * functions for the specific versioned machine types can override
3354      * these details for backwards compatibility
3355      */
3356     mc->init = ppc_spapr_init;
3357     mc->reset = ppc_spapr_reset;
3358     mc->block_default_type = IF_SCSI;
3359     mc->max_cpus = 1024;
3360     mc->no_parallel = 1;
3361     mc->default_boot_order = "";
3362     mc->default_ram_size = 512 * M_BYTE;
3363     mc->kvm_type = spapr_kvm_type;
3364     mc->has_dynamic_sysbus = true;
3365     mc->pci_allow_0_address = true;
3366     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3367     hc->pre_plug = spapr_machine_device_pre_plug;
3368     hc->plug = spapr_machine_device_plug;
3369     hc->unplug = spapr_machine_device_unplug;
3370     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3371     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3372     hc->unplug_request = spapr_machine_device_unplug_request;
3373 
3374     smc->dr_lmb_enabled = true;
3375     smc->tcg_default_cpu = "POWER8";
3376     mc->has_hotpluggable_cpus = true;
3377     fwc->get_dev_path = spapr_get_fw_dev_path;
3378     nc->nmi_monitor_handler = spapr_nmi;
3379     smc->phb_placement = spapr_phb_placement;
3380     vhc->hypercall = emulate_spapr_hypercall;
3381     vhc->hpt_mask = spapr_hpt_mask;
3382     vhc->map_hptes = spapr_map_hptes;
3383     vhc->unmap_hptes = spapr_unmap_hptes;
3384     vhc->store_hpte = spapr_store_hpte;
3385     vhc->get_patbe = spapr_get_patbe;
3386     xic->ics_get = spapr_ics_get;
3387     xic->ics_resend = spapr_ics_resend;
3388     xic->icp_get = spapr_icp_get;
3389     ispc->print_info = spapr_pic_print_info;
3390     /* Force NUMA node memory size to be a multiple of
3391      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3392      * in which LMBs are represented and hot-added
3393      */
3394     mc->numa_mem_align_shift = 28;
3395 }
3396 
3397 static const TypeInfo spapr_machine_info = {
3398     .name          = TYPE_SPAPR_MACHINE,
3399     .parent        = TYPE_MACHINE,
3400     .abstract      = true,
3401     .instance_size = sizeof(sPAPRMachineState),
3402     .instance_init = spapr_machine_initfn,
3403     .instance_finalize = spapr_machine_finalizefn,
3404     .class_size    = sizeof(sPAPRMachineClass),
3405     .class_init    = spapr_machine_class_init,
3406     .interfaces = (InterfaceInfo[]) {
3407         { TYPE_FW_PATH_PROVIDER },
3408         { TYPE_NMI },
3409         { TYPE_HOTPLUG_HANDLER },
3410         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3411         { TYPE_XICS_FABRIC },
3412         { TYPE_INTERRUPT_STATS_PROVIDER },
3413         { }
3414     },
3415 };
3416 
3417 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3418     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3419                                                     void *data)      \
3420     {                                                                \
3421         MachineClass *mc = MACHINE_CLASS(oc);                        \
3422         spapr_machine_##suffix##_class_options(mc);                  \
3423         if (latest) {                                                \
3424             mc->alias = "pseries";                                   \
3425             mc->is_default = 1;                                      \
3426         }                                                            \
3427     }                                                                \
3428     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3429     {                                                                \
3430         MachineState *machine = MACHINE(obj);                        \
3431         spapr_machine_##suffix##_instance_options(machine);          \
3432     }                                                                \
3433     static const TypeInfo spapr_machine_##suffix##_info = {          \
3434         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3435         .parent = TYPE_SPAPR_MACHINE,                                \
3436         .class_init = spapr_machine_##suffix##_class_init,           \
3437         .instance_init = spapr_machine_##suffix##_instance_init,     \
3438     };                                                               \
3439     static void spapr_machine_register_##suffix(void)                \
3440     {                                                                \
3441         type_register(&spapr_machine_##suffix##_info);               \
3442     }                                                                \
3443     type_init(spapr_machine_register_##suffix)
3444 
3445 /*
3446  * pseries-2.10
3447  */
3448 static void spapr_machine_2_10_instance_options(MachineState *machine)
3449 {
3450 }
3451 
3452 static void spapr_machine_2_10_class_options(MachineClass *mc)
3453 {
3454     /* Defaults for the latest behaviour inherited from the base class */
3455 }
3456 
3457 DEFINE_SPAPR_MACHINE(2_10, "2.10", true);
3458 
3459 /*
3460  * pseries-2.9
3461  */
3462 #define SPAPR_COMPAT_2_9                                               \
3463     HW_COMPAT_2_9                                                      \
3464     {                                                                  \
3465         .driver = TYPE_POWERPC_CPU,                                    \
3466         .property = "pre-2.10-migration",                              \
3467         .value    = "on",                                              \
3468     },                                                                 \
3469 
3470 static void spapr_machine_2_9_instance_options(MachineState *machine)
3471 {
3472     spapr_machine_2_10_instance_options(machine);
3473 }
3474 
3475 static void spapr_machine_2_9_class_options(MachineClass *mc)
3476 {
3477     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3478 
3479     spapr_machine_2_10_class_options(mc);
3480     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3481     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
3482     smc->pre_2_10_has_unused_icps = true;
3483 }
3484 
3485 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3486 
3487 /*
3488  * pseries-2.8
3489  */
3490 #define SPAPR_COMPAT_2_8                                        \
3491     HW_COMPAT_2_8                                               \
3492     {                                                           \
3493         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3494         .property = "pcie-extended-configuration-space",        \
3495         .value    = "off",                                      \
3496     },
3497 
3498 static void spapr_machine_2_8_instance_options(MachineState *machine)
3499 {
3500     spapr_machine_2_9_instance_options(machine);
3501 }
3502 
3503 static void spapr_machine_2_8_class_options(MachineClass *mc)
3504 {
3505     spapr_machine_2_9_class_options(mc);
3506     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3507     mc->numa_mem_align_shift = 23;
3508 }
3509 
3510 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3511 
3512 /*
3513  * pseries-2.7
3514  */
3515 #define SPAPR_COMPAT_2_7                            \
3516     HW_COMPAT_2_7                                   \
3517     {                                               \
3518         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3519         .property = "mem_win_size",                 \
3520         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3521     },                                              \
3522     {                                               \
3523         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3524         .property = "mem64_win_size",               \
3525         .value    = "0",                            \
3526     },                                              \
3527     {                                               \
3528         .driver = TYPE_POWERPC_CPU,                 \
3529         .property = "pre-2.8-migration",            \
3530         .value    = "on",                           \
3531     },                                              \
3532     {                                               \
3533         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3534         .property = "pre-2.8-migration",            \
3535         .value    = "on",                           \
3536     },
3537 
3538 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3539                               uint64_t *buid, hwaddr *pio,
3540                               hwaddr *mmio32, hwaddr *mmio64,
3541                               unsigned n_dma, uint32_t *liobns, Error **errp)
3542 {
3543     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3544     const uint64_t base_buid = 0x800000020000000ULL;
3545     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3546     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3547     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3548     const uint32_t max_index = 255;
3549     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3550 
3551     uint64_t ram_top = MACHINE(spapr)->ram_size;
3552     hwaddr phb0_base, phb_base;
3553     int i;
3554 
3555     /* Do we have hotpluggable memory? */
3556     if (MACHINE(spapr)->maxram_size > ram_top) {
3557         /* Can't just use maxram_size, because there may be an
3558          * alignment gap between normal and hotpluggable memory
3559          * regions */
3560         ram_top = spapr->hotplug_memory.base +
3561             memory_region_size(&spapr->hotplug_memory.mr);
3562     }
3563 
3564     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3565 
3566     if (index > max_index) {
3567         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3568                    max_index);
3569         return;
3570     }
3571 
3572     *buid = base_buid + index;
3573     for (i = 0; i < n_dma; ++i) {
3574         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3575     }
3576 
3577     phb_base = phb0_base + index * phb_spacing;
3578     *pio = phb_base + pio_offset;
3579     *mmio32 = phb_base + mmio_offset;
3580     /*
3581      * We don't set the 64-bit MMIO window, relying on the PHB's
3582      * fallback behaviour of automatically splitting a large "32-bit"
3583      * window into contiguous 32-bit and 64-bit windows
3584      */
3585 }
3586 
3587 static void spapr_machine_2_7_instance_options(MachineState *machine)
3588 {
3589     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3590 
3591     spapr_machine_2_8_instance_options(machine);
3592     spapr->use_hotplug_event_source = false;
3593 }
3594 
3595 static void spapr_machine_2_7_class_options(MachineClass *mc)
3596 {
3597     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3598 
3599     spapr_machine_2_8_class_options(mc);
3600     smc->tcg_default_cpu = "POWER7";
3601     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3602     smc->phb_placement = phb_placement_2_7;
3603 }
3604 
3605 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3606 
3607 /*
3608  * pseries-2.6
3609  */
3610 #define SPAPR_COMPAT_2_6 \
3611     HW_COMPAT_2_6 \
3612     { \
3613         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3614         .property = "ddw",\
3615         .value    = stringify(off),\
3616     },
3617 
3618 static void spapr_machine_2_6_instance_options(MachineState *machine)
3619 {
3620     spapr_machine_2_7_instance_options(machine);
3621 }
3622 
3623 static void spapr_machine_2_6_class_options(MachineClass *mc)
3624 {
3625     spapr_machine_2_7_class_options(mc);
3626     mc->has_hotpluggable_cpus = false;
3627     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3628 }
3629 
3630 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3631 
3632 /*
3633  * pseries-2.5
3634  */
3635 #define SPAPR_COMPAT_2_5 \
3636     HW_COMPAT_2_5 \
3637     { \
3638         .driver   = "spapr-vlan", \
3639         .property = "use-rx-buffer-pools", \
3640         .value    = "off", \
3641     },
3642 
3643 static void spapr_machine_2_5_instance_options(MachineState *machine)
3644 {
3645     spapr_machine_2_6_instance_options(machine);
3646 }
3647 
3648 static void spapr_machine_2_5_class_options(MachineClass *mc)
3649 {
3650     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3651 
3652     spapr_machine_2_6_class_options(mc);
3653     smc->use_ohci_by_default = true;
3654     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3655 }
3656 
3657 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3658 
3659 /*
3660  * pseries-2.4
3661  */
3662 #define SPAPR_COMPAT_2_4 \
3663         HW_COMPAT_2_4
3664 
3665 static void spapr_machine_2_4_instance_options(MachineState *machine)
3666 {
3667     spapr_machine_2_5_instance_options(machine);
3668 }
3669 
3670 static void spapr_machine_2_4_class_options(MachineClass *mc)
3671 {
3672     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3673 
3674     spapr_machine_2_5_class_options(mc);
3675     smc->dr_lmb_enabled = false;
3676     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3677 }
3678 
3679 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3680 
3681 /*
3682  * pseries-2.3
3683  */
3684 #define SPAPR_COMPAT_2_3 \
3685         HW_COMPAT_2_3 \
3686         {\
3687             .driver   = "spapr-pci-host-bridge",\
3688             .property = "dynamic-reconfiguration",\
3689             .value    = "off",\
3690         },
3691 
3692 static void spapr_machine_2_3_instance_options(MachineState *machine)
3693 {
3694     spapr_machine_2_4_instance_options(machine);
3695 }
3696 
3697 static void spapr_machine_2_3_class_options(MachineClass *mc)
3698 {
3699     spapr_machine_2_4_class_options(mc);
3700     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3701 }
3702 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3703 
3704 /*
3705  * pseries-2.2
3706  */
3707 
3708 #define SPAPR_COMPAT_2_2 \
3709         HW_COMPAT_2_2 \
3710         {\
3711             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3712             .property = "mem_win_size",\
3713             .value    = "0x20000000",\
3714         },
3715 
3716 static void spapr_machine_2_2_instance_options(MachineState *machine)
3717 {
3718     spapr_machine_2_3_instance_options(machine);
3719     machine->suppress_vmdesc = true;
3720 }
3721 
3722 static void spapr_machine_2_2_class_options(MachineClass *mc)
3723 {
3724     spapr_machine_2_3_class_options(mc);
3725     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3726 }
3727 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3728 
3729 /*
3730  * pseries-2.1
3731  */
3732 #define SPAPR_COMPAT_2_1 \
3733         HW_COMPAT_2_1
3734 
3735 static void spapr_machine_2_1_instance_options(MachineState *machine)
3736 {
3737     spapr_machine_2_2_instance_options(machine);
3738 }
3739 
3740 static void spapr_machine_2_1_class_options(MachineClass *mc)
3741 {
3742     spapr_machine_2_2_class_options(mc);
3743     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3744 }
3745 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3746 
3747 static void spapr_machine_register_types(void)
3748 {
3749     type_register_static(&spapr_machine_info);
3750 }
3751 
3752 type_init(spapr_machine_register_types)
3753