xref: /openbmc/qemu/hw/ppc/spapr.c (revision 5bc8d26de20ca6befe6390f381cb98120237e4be)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
31 #include "hw/hw.h"
32 #include "qemu/log.h"
33 #include "hw/fw-path-provider.h"
34 #include "elf.h"
35 #include "net/net.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/migration.h"
42 #include "mmu-hash64.h"
43 #include "mmu-book3s-v3.h"
44 #include "qom/cpu.h"
45 
46 #include "hw/boards.h"
47 #include "hw/ppc/ppc.h"
48 #include "hw/loader.h"
49 
50 #include "hw/ppc/fdt.h"
51 #include "hw/ppc/spapr.h"
52 #include "hw/ppc/spapr_vio.h"
53 #include "hw/pci-host/spapr.h"
54 #include "hw/ppc/xics.h"
55 #include "hw/pci/msi.h"
56 
57 #include "hw/pci/pci.h"
58 #include "hw/scsi/scsi.h"
59 #include "hw/virtio/virtio-scsi.h"
60 
61 #include "exec/address-spaces.h"
62 #include "hw/usb.h"
63 #include "qemu/config-file.h"
64 #include "qemu/error-report.h"
65 #include "trace.h"
66 #include "hw/nmi.h"
67 #include "hw/intc/intc.h"
68 
69 #include "hw/compat.h"
70 #include "qemu/cutils.h"
71 #include "hw/ppc/spapr_cpu_core.h"
72 #include "qmp-commands.h"
73 
74 #include <libfdt.h>
75 
76 /* SLOF memory layout:
77  *
78  * SLOF raw image loaded at 0, copies its romfs right below the flat
79  * device-tree, then position SLOF itself 31M below that
80  *
81  * So we set FW_OVERHEAD to 40MB which should account for all of that
82  * and more
83  *
84  * We load our kernel at 4M, leaving space for SLOF initial image
85  */
86 #define FDT_MAX_SIZE            0x100000
87 #define RTAS_MAX_SIZE           0x10000
88 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
89 #define FW_MAX_SIZE             0x400000
90 #define FW_FILE_NAME            "slof.bin"
91 #define FW_OVERHEAD             0x2800000
92 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
93 
94 #define MIN_RMA_SLOF            128UL
95 
96 #define PHANDLE_XICP            0x00001111
97 
98 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
99 
100 static int try_create_xics(sPAPRMachineState *spapr, const char *type_ics,
101                            const char *type_icp, int nr_servers,
102                            int nr_irqs, Error **errp)
103 {
104     XICSFabric *xi = XICS_FABRIC(spapr);
105     Error *err = NULL, *local_err = NULL;
106     ICSState *ics = NULL;
107 
108     ics = ICS_SIMPLE(object_new(type_ics));
109     object_property_add_child(OBJECT(spapr), "ics", OBJECT(ics), NULL);
110     object_property_set_int(OBJECT(ics), nr_irqs, "nr-irqs", &err);
111     object_property_add_const_link(OBJECT(ics), "xics", OBJECT(xi), NULL);
112     object_property_set_bool(OBJECT(ics), true, "realized", &local_err);
113     error_propagate(&err, local_err);
114     if (err) {
115         error_propagate(errp, err);
116         return -1;
117     }
118 
119     spapr->nr_servers = nr_servers;
120     spapr->ics = ics;
121     spapr->icp_type = type_icp;
122     return 0;
123 }
124 
125 static int xics_system_init(MachineState *machine,
126                             int nr_servers, int nr_irqs, Error **errp)
127 {
128     int rc = -1;
129 
130     if (kvm_enabled()) {
131         Error *err = NULL;
132 
133         if (machine_kernel_irqchip_allowed(machine) &&
134             !xics_kvm_init(SPAPR_MACHINE(machine), errp)) {
135             rc = try_create_xics(SPAPR_MACHINE(machine), TYPE_ICS_KVM,
136                                  TYPE_KVM_ICP, nr_servers, nr_irqs, &err);
137         }
138         if (machine_kernel_irqchip_required(machine) && rc < 0) {
139             error_reportf_err(err,
140                               "kernel_irqchip requested but unavailable: ");
141         } else {
142             error_free(err);
143         }
144     }
145 
146     if (rc < 0) {
147         xics_spapr_init(SPAPR_MACHINE(machine), errp);
148         rc = try_create_xics(SPAPR_MACHINE(machine), TYPE_ICS_SIMPLE,
149                                TYPE_ICP, nr_servers, nr_irqs, errp);
150     }
151 
152     return rc;
153 }
154 
155 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
156                                   int smt_threads)
157 {
158     int i, ret = 0;
159     uint32_t servers_prop[smt_threads];
160     uint32_t gservers_prop[smt_threads * 2];
161     int index = ppc_get_vcpu_dt_id(cpu);
162 
163     if (cpu->compat_pvr) {
164         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
165         if (ret < 0) {
166             return ret;
167         }
168     }
169 
170     /* Build interrupt servers and gservers properties */
171     for (i = 0; i < smt_threads; i++) {
172         servers_prop[i] = cpu_to_be32(index + i);
173         /* Hack, direct the group queues back to cpu 0 */
174         gservers_prop[i*2] = cpu_to_be32(index + i);
175         gservers_prop[i*2 + 1] = 0;
176     }
177     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
178                       servers_prop, sizeof(servers_prop));
179     if (ret < 0) {
180         return ret;
181     }
182     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
183                       gservers_prop, sizeof(gservers_prop));
184 
185     return ret;
186 }
187 
188 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
189 {
190     int ret = 0;
191     PowerPCCPU *cpu = POWERPC_CPU(cs);
192     int index = ppc_get_vcpu_dt_id(cpu);
193     uint32_t associativity[] = {cpu_to_be32(0x5),
194                                 cpu_to_be32(0x0),
195                                 cpu_to_be32(0x0),
196                                 cpu_to_be32(0x0),
197                                 cpu_to_be32(cs->numa_node),
198                                 cpu_to_be32(index)};
199 
200     /* Advertise NUMA via ibm,associativity */
201     if (nb_numa_nodes > 1) {
202         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
203                           sizeof(associativity));
204     }
205 
206     return ret;
207 }
208 
209 /* Populate the "ibm,pa-features" property */
210 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset,
211                                       bool legacy_guest)
212 {
213     uint8_t pa_features_206[] = { 6, 0,
214         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
215     uint8_t pa_features_207[] = { 24, 0,
216         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
217         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
218         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
219         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
220     uint8_t pa_features_300[] = { 66, 0,
221         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
222         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
223         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
224         /* 6: DS207 */
225         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
226         /* 16: Vector */
227         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
228         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
229         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 18 - 23 */
230         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
231         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
232         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
233         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
234         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
236         /* 42: PM, 44: PC RA, 46: SC vec'd */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
238         /* 48: SIMD, 50: QP BFP, 52: String */
239         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
240         /* 54: DecFP, 56: DecI, 58: SHA */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
242         /* 60: NM atomic, 62: RNG */
243         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
244     };
245     uint8_t *pa_features;
246     size_t pa_size;
247 
248     switch (POWERPC_MMU_VER(env->mmu_model)) {
249     case POWERPC_MMU_VER_2_06:
250         pa_features = pa_features_206;
251         pa_size = sizeof(pa_features_206);
252         break;
253     case POWERPC_MMU_VER_2_07:
254         pa_features = pa_features_207;
255         pa_size = sizeof(pa_features_207);
256         break;
257     case POWERPC_MMU_VER_3_00:
258         pa_features = pa_features_300;
259         pa_size = sizeof(pa_features_300);
260         break;
261     default:
262         return;
263     }
264 
265     if (env->ci_large_pages) {
266         /*
267          * Note: we keep CI large pages off by default because a 64K capable
268          * guest provisioned with large pages might otherwise try to map a qemu
269          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
270          * even if that qemu runs on a 4k host.
271          * We dd this bit back here if we are confident this is not an issue
272          */
273         pa_features[3] |= 0x20;
274     }
275     if (kvmppc_has_cap_htm() && pa_size > 24) {
276         pa_features[24] |= 0x80;    /* Transactional memory support */
277     }
278     if (legacy_guest && pa_size > 40) {
279         /* Workaround for broken kernels that attempt (guest) radix
280          * mode when they can't handle it, if they see the radix bit set
281          * in pa-features. So hide it from them. */
282         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
283     }
284 
285     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
286 }
287 
288 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
289 {
290     int ret = 0, offset, cpus_offset;
291     CPUState *cs;
292     char cpu_model[32];
293     int smt = kvmppc_smt_threads();
294     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
295 
296     CPU_FOREACH(cs) {
297         PowerPCCPU *cpu = POWERPC_CPU(cs);
298         CPUPPCState *env = &cpu->env;
299         DeviceClass *dc = DEVICE_GET_CLASS(cs);
300         int index = ppc_get_vcpu_dt_id(cpu);
301         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
302 
303         if ((index % smt) != 0) {
304             continue;
305         }
306 
307         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
308 
309         cpus_offset = fdt_path_offset(fdt, "/cpus");
310         if (cpus_offset < 0) {
311             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
312                                           "cpus");
313             if (cpus_offset < 0) {
314                 return cpus_offset;
315             }
316         }
317         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
318         if (offset < 0) {
319             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
320             if (offset < 0) {
321                 return offset;
322             }
323         }
324 
325         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
326                           pft_size_prop, sizeof(pft_size_prop));
327         if (ret < 0) {
328             return ret;
329         }
330 
331         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
332         if (ret < 0) {
333             return ret;
334         }
335 
336         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
337         if (ret < 0) {
338             return ret;
339         }
340 
341         spapr_populate_pa_features(env, fdt, offset,
342                                          spapr->cas_legacy_guest_workaround);
343     }
344     return ret;
345 }
346 
347 static hwaddr spapr_node0_size(void)
348 {
349     MachineState *machine = MACHINE(qdev_get_machine());
350 
351     if (nb_numa_nodes) {
352         int i;
353         for (i = 0; i < nb_numa_nodes; ++i) {
354             if (numa_info[i].node_mem) {
355                 return MIN(pow2floor(numa_info[i].node_mem),
356                            machine->ram_size);
357             }
358         }
359     }
360     return machine->ram_size;
361 }
362 
363 static void add_str(GString *s, const gchar *s1)
364 {
365     g_string_append_len(s, s1, strlen(s1) + 1);
366 }
367 
368 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
369                                        hwaddr size)
370 {
371     uint32_t associativity[] = {
372         cpu_to_be32(0x4), /* length */
373         cpu_to_be32(0x0), cpu_to_be32(0x0),
374         cpu_to_be32(0x0), cpu_to_be32(nodeid)
375     };
376     char mem_name[32];
377     uint64_t mem_reg_property[2];
378     int off;
379 
380     mem_reg_property[0] = cpu_to_be64(start);
381     mem_reg_property[1] = cpu_to_be64(size);
382 
383     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
384     off = fdt_add_subnode(fdt, 0, mem_name);
385     _FDT(off);
386     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
387     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
388                       sizeof(mem_reg_property))));
389     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
390                       sizeof(associativity))));
391     return off;
392 }
393 
394 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
395 {
396     MachineState *machine = MACHINE(spapr);
397     hwaddr mem_start, node_size;
398     int i, nb_nodes = nb_numa_nodes;
399     NodeInfo *nodes = numa_info;
400     NodeInfo ramnode;
401 
402     /* No NUMA nodes, assume there is just one node with whole RAM */
403     if (!nb_numa_nodes) {
404         nb_nodes = 1;
405         ramnode.node_mem = machine->ram_size;
406         nodes = &ramnode;
407     }
408 
409     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
410         if (!nodes[i].node_mem) {
411             continue;
412         }
413         if (mem_start >= machine->ram_size) {
414             node_size = 0;
415         } else {
416             node_size = nodes[i].node_mem;
417             if (node_size > machine->ram_size - mem_start) {
418                 node_size = machine->ram_size - mem_start;
419             }
420         }
421         if (!mem_start) {
422             /* ppc_spapr_init() checks for rma_size <= node0_size already */
423             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
424             mem_start += spapr->rma_size;
425             node_size -= spapr->rma_size;
426         }
427         for ( ; node_size; ) {
428             hwaddr sizetmp = pow2floor(node_size);
429 
430             /* mem_start != 0 here */
431             if (ctzl(mem_start) < ctzl(sizetmp)) {
432                 sizetmp = 1ULL << ctzl(mem_start);
433             }
434 
435             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
436             node_size -= sizetmp;
437             mem_start += sizetmp;
438         }
439     }
440 
441     return 0;
442 }
443 
444 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
445                                   sPAPRMachineState *spapr)
446 {
447     PowerPCCPU *cpu = POWERPC_CPU(cs);
448     CPUPPCState *env = &cpu->env;
449     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
450     int index = ppc_get_vcpu_dt_id(cpu);
451     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
452                        0xffffffff, 0xffffffff};
453     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
454         : SPAPR_TIMEBASE_FREQ;
455     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
456     uint32_t page_sizes_prop[64];
457     size_t page_sizes_prop_size;
458     uint32_t vcpus_per_socket = smp_threads * smp_cores;
459     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
460     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
461     sPAPRDRConnector *drc;
462     sPAPRDRConnectorClass *drck;
463     int drc_index;
464     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
465     int i;
466 
467     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
468     if (drc) {
469         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
470         drc_index = drck->get_index(drc);
471         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
472     }
473 
474     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
475     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
476 
477     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
478     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
479                            env->dcache_line_size)));
480     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
481                            env->dcache_line_size)));
482     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
483                            env->icache_line_size)));
484     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
485                            env->icache_line_size)));
486 
487     if (pcc->l1_dcache_size) {
488         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
489                                pcc->l1_dcache_size)));
490     } else {
491         error_report("Warning: Unknown L1 dcache size for cpu");
492     }
493     if (pcc->l1_icache_size) {
494         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
495                                pcc->l1_icache_size)));
496     } else {
497         error_report("Warning: Unknown L1 icache size for cpu");
498     }
499 
500     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
501     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
502     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
503     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
504     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
505     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
506 
507     if (env->spr_cb[SPR_PURR].oea_read) {
508         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
509     }
510 
511     if (env->mmu_model & POWERPC_MMU_1TSEG) {
512         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
513                           segs, sizeof(segs))));
514     }
515 
516     /* Advertise VMX/VSX (vector extensions) if available
517      *   0 / no property == no vector extensions
518      *   1               == VMX / Altivec available
519      *   2               == VSX available */
520     if (env->insns_flags & PPC_ALTIVEC) {
521         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
522 
523         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
524     }
525 
526     /* Advertise DFP (Decimal Floating Point) if available
527      *   0 / no property == no DFP
528      *   1               == DFP available */
529     if (env->insns_flags2 & PPC2_DFP) {
530         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
531     }
532 
533     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
534                                                   sizeof(page_sizes_prop));
535     if (page_sizes_prop_size) {
536         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
537                           page_sizes_prop, page_sizes_prop_size)));
538     }
539 
540     spapr_populate_pa_features(env, fdt, offset, false);
541 
542     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
543                            cs->cpu_index / vcpus_per_socket)));
544 
545     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
546                       pft_size_prop, sizeof(pft_size_prop))));
547 
548     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
549 
550     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
551 
552     if (pcc->radix_page_info) {
553         for (i = 0; i < pcc->radix_page_info->count; i++) {
554             radix_AP_encodings[i] =
555                 cpu_to_be32(pcc->radix_page_info->entries[i]);
556         }
557         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
558                           radix_AP_encodings,
559                           pcc->radix_page_info->count *
560                           sizeof(radix_AP_encodings[0]))));
561     }
562 }
563 
564 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
565 {
566     CPUState *cs;
567     int cpus_offset;
568     char *nodename;
569     int smt = kvmppc_smt_threads();
570 
571     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
572     _FDT(cpus_offset);
573     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
574     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
575 
576     /*
577      * We walk the CPUs in reverse order to ensure that CPU DT nodes
578      * created by fdt_add_subnode() end up in the right order in FDT
579      * for the guest kernel the enumerate the CPUs correctly.
580      */
581     CPU_FOREACH_REVERSE(cs) {
582         PowerPCCPU *cpu = POWERPC_CPU(cs);
583         int index = ppc_get_vcpu_dt_id(cpu);
584         DeviceClass *dc = DEVICE_GET_CLASS(cs);
585         int offset;
586 
587         if ((index % smt) != 0) {
588             continue;
589         }
590 
591         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
592         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
593         g_free(nodename);
594         _FDT(offset);
595         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
596     }
597 
598 }
599 
600 /*
601  * Adds ibm,dynamic-reconfiguration-memory node.
602  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
603  * of this device tree node.
604  */
605 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
606 {
607     MachineState *machine = MACHINE(spapr);
608     int ret, i, offset;
609     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
610     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
611     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
612     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
613                        memory_region_size(&spapr->hotplug_memory.mr)) /
614                        lmb_size;
615     uint32_t *int_buf, *cur_index, buf_len;
616     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
617 
618     /*
619      * Don't create the node if there is no hotpluggable memory
620      */
621     if (machine->ram_size == machine->maxram_size) {
622         return 0;
623     }
624 
625     /*
626      * Allocate enough buffer size to fit in ibm,dynamic-memory
627      * or ibm,associativity-lookup-arrays
628      */
629     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
630               * sizeof(uint32_t);
631     cur_index = int_buf = g_malloc0(buf_len);
632 
633     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
634 
635     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
636                     sizeof(prop_lmb_size));
637     if (ret < 0) {
638         goto out;
639     }
640 
641     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
642     if (ret < 0) {
643         goto out;
644     }
645 
646     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
647     if (ret < 0) {
648         goto out;
649     }
650 
651     /* ibm,dynamic-memory */
652     int_buf[0] = cpu_to_be32(nr_lmbs);
653     cur_index++;
654     for (i = 0; i < nr_lmbs; i++) {
655         uint64_t addr = i * lmb_size;
656         uint32_t *dynamic_memory = cur_index;
657 
658         if (i >= hotplug_lmb_start) {
659             sPAPRDRConnector *drc;
660             sPAPRDRConnectorClass *drck;
661 
662             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
663             g_assert(drc);
664             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
665 
666             dynamic_memory[0] = cpu_to_be32(addr >> 32);
667             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
668             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
669             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
670             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
671             if (memory_region_present(get_system_memory(), addr)) {
672                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
673             } else {
674                 dynamic_memory[5] = cpu_to_be32(0);
675             }
676         } else {
677             /*
678              * LMB information for RMA, boot time RAM and gap b/n RAM and
679              * hotplug memory region -- all these are marked as reserved
680              * and as having no valid DRC.
681              */
682             dynamic_memory[0] = cpu_to_be32(addr >> 32);
683             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
684             dynamic_memory[2] = cpu_to_be32(0);
685             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
686             dynamic_memory[4] = cpu_to_be32(-1);
687             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
688                                             SPAPR_LMB_FLAGS_DRC_INVALID);
689         }
690 
691         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
692     }
693     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
694     if (ret < 0) {
695         goto out;
696     }
697 
698     /* ibm,associativity-lookup-arrays */
699     cur_index = int_buf;
700     int_buf[0] = cpu_to_be32(nr_nodes);
701     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
702     cur_index += 2;
703     for (i = 0; i < nr_nodes; i++) {
704         uint32_t associativity[] = {
705             cpu_to_be32(0x0),
706             cpu_to_be32(0x0),
707             cpu_to_be32(0x0),
708             cpu_to_be32(i)
709         };
710         memcpy(cur_index, associativity, sizeof(associativity));
711         cur_index += 4;
712     }
713     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
714             (cur_index - int_buf) * sizeof(uint32_t));
715 out:
716     g_free(int_buf);
717     return ret;
718 }
719 
720 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
721                                 sPAPROptionVector *ov5_updates)
722 {
723     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
724     int ret = 0, offset;
725 
726     /* Generate ibm,dynamic-reconfiguration-memory node if required */
727     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
728         g_assert(smc->dr_lmb_enabled);
729         ret = spapr_populate_drconf_memory(spapr, fdt);
730         if (ret) {
731             goto out;
732         }
733     }
734 
735     offset = fdt_path_offset(fdt, "/chosen");
736     if (offset < 0) {
737         offset = fdt_add_subnode(fdt, 0, "chosen");
738         if (offset < 0) {
739             return offset;
740         }
741     }
742     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
743                                  "ibm,architecture-vec-5");
744 
745 out:
746     return ret;
747 }
748 
749 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
750                                  target_ulong addr, target_ulong size,
751                                  sPAPROptionVector *ov5_updates)
752 {
753     void *fdt, *fdt_skel;
754     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
755 
756     size -= sizeof(hdr);
757 
758     /* Create sceleton */
759     fdt_skel = g_malloc0(size);
760     _FDT((fdt_create(fdt_skel, size)));
761     _FDT((fdt_begin_node(fdt_skel, "")));
762     _FDT((fdt_end_node(fdt_skel)));
763     _FDT((fdt_finish(fdt_skel)));
764     fdt = g_malloc0(size);
765     _FDT((fdt_open_into(fdt_skel, fdt, size)));
766     g_free(fdt_skel);
767 
768     /* Fixup cpu nodes */
769     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
770 
771     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
772         return -1;
773     }
774 
775     /* Pack resulting tree */
776     _FDT((fdt_pack(fdt)));
777 
778     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
779         trace_spapr_cas_failed(size);
780         return -1;
781     }
782 
783     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
784     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
785     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
786     g_free(fdt);
787 
788     return 0;
789 }
790 
791 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
792 {
793     int rtas;
794     GString *hypertas = g_string_sized_new(256);
795     GString *qemu_hypertas = g_string_sized_new(256);
796     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
797     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
798         memory_region_size(&spapr->hotplug_memory.mr);
799     uint32_t lrdr_capacity[] = {
800         cpu_to_be32(max_hotplug_addr >> 32),
801         cpu_to_be32(max_hotplug_addr & 0xffffffff),
802         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
803         cpu_to_be32(max_cpus / smp_threads),
804     };
805 
806     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
807 
808     /* hypertas */
809     add_str(hypertas, "hcall-pft");
810     add_str(hypertas, "hcall-term");
811     add_str(hypertas, "hcall-dabr");
812     add_str(hypertas, "hcall-interrupt");
813     add_str(hypertas, "hcall-tce");
814     add_str(hypertas, "hcall-vio");
815     add_str(hypertas, "hcall-splpar");
816     add_str(hypertas, "hcall-bulk");
817     add_str(hypertas, "hcall-set-mode");
818     add_str(hypertas, "hcall-sprg0");
819     add_str(hypertas, "hcall-copy");
820     add_str(hypertas, "hcall-debug");
821     add_str(qemu_hypertas, "hcall-memop1");
822 
823     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
824         add_str(hypertas, "hcall-multi-tce");
825     }
826     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
827                      hypertas->str, hypertas->len));
828     g_string_free(hypertas, TRUE);
829     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
830                      qemu_hypertas->str, qemu_hypertas->len));
831     g_string_free(qemu_hypertas, TRUE);
832 
833     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
834                      refpoints, sizeof(refpoints)));
835 
836     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
837                           RTAS_ERROR_LOG_MAX));
838     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
839                           RTAS_EVENT_SCAN_RATE));
840 
841     if (msi_nonbroken) {
842         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
843     }
844 
845     /*
846      * According to PAPR, rtas ibm,os-term does not guarantee a return
847      * back to the guest cpu.
848      *
849      * While an additional ibm,extended-os-term property indicates
850      * that rtas call return will always occur. Set this property.
851      */
852     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
853 
854     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
855                      lrdr_capacity, sizeof(lrdr_capacity)));
856 
857     spapr_dt_rtas_tokens(fdt, rtas);
858 }
859 
860 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
861  * that the guest may request and thus the valid values for bytes 24..26 of
862  * option vector 5: */
863 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
864 {
865     char val[2 * 3] = {
866         24, 0x00, /* Hash/Radix, filled in below. */
867         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
868         26, 0x40, /* Radix options: GTSE == yes. */
869     };
870 
871     if (kvm_enabled()) {
872         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
873             val[1] = 0x80; /* OV5_MMU_BOTH */
874         } else if (kvmppc_has_cap_mmu_radix()) {
875             val[1] = 0x40; /* OV5_MMU_RADIX_300 */
876         } else {
877             val[1] = 0x00; /* Hash */
878         }
879     } else {
880         /* TODO: TCG case, hash */
881         val[1] = 0x00;
882     }
883     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
884                      val, sizeof(val)));
885 }
886 
887 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
888 {
889     MachineState *machine = MACHINE(spapr);
890     int chosen;
891     const char *boot_device = machine->boot_order;
892     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
893     size_t cb = 0;
894     char *bootlist = get_boot_devices_list(&cb, true);
895 
896     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
897 
898     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
899     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
900                           spapr->initrd_base));
901     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
902                           spapr->initrd_base + spapr->initrd_size));
903 
904     if (spapr->kernel_size) {
905         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
906                               cpu_to_be64(spapr->kernel_size) };
907 
908         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
909                          &kprop, sizeof(kprop)));
910         if (spapr->kernel_le) {
911             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
912         }
913     }
914     if (boot_menu) {
915         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
916     }
917     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
918     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
919     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
920 
921     if (cb && bootlist) {
922         int i;
923 
924         for (i = 0; i < cb; i++) {
925             if (bootlist[i] == '\n') {
926                 bootlist[i] = ' ';
927             }
928         }
929         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
930     }
931 
932     if (boot_device && strlen(boot_device)) {
933         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
934     }
935 
936     if (!spapr->has_graphics && stdout_path) {
937         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
938     }
939 
940     spapr_dt_ov5_platform_support(fdt, chosen);
941 
942     g_free(stdout_path);
943     g_free(bootlist);
944 }
945 
946 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
947 {
948     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
949      * KVM to work under pHyp with some guest co-operation */
950     int hypervisor;
951     uint8_t hypercall[16];
952 
953     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
954     /* indicate KVM hypercall interface */
955     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
956     if (kvmppc_has_cap_fixup_hcalls()) {
957         /*
958          * Older KVM versions with older guest kernels were broken
959          * with the magic page, don't allow the guest to map it.
960          */
961         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
962                                   sizeof(hypercall))) {
963             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
964                              hypercall, sizeof(hypercall)));
965         }
966     }
967 }
968 
969 static void *spapr_build_fdt(sPAPRMachineState *spapr,
970                              hwaddr rtas_addr,
971                              hwaddr rtas_size)
972 {
973     MachineState *machine = MACHINE(qdev_get_machine());
974     MachineClass *mc = MACHINE_GET_CLASS(machine);
975     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
976     int ret;
977     void *fdt;
978     sPAPRPHBState *phb;
979     char *buf;
980 
981     fdt = g_malloc0(FDT_MAX_SIZE);
982     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
983 
984     /* Root node */
985     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
986     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
987     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
988 
989     /*
990      * Add info to guest to indentify which host is it being run on
991      * and what is the uuid of the guest
992      */
993     if (kvmppc_get_host_model(&buf)) {
994         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
995         g_free(buf);
996     }
997     if (kvmppc_get_host_serial(&buf)) {
998         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
999         g_free(buf);
1000     }
1001 
1002     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1003 
1004     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1005     if (qemu_uuid_set) {
1006         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1007     }
1008     g_free(buf);
1009 
1010     if (qemu_get_vm_name()) {
1011         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1012                                 qemu_get_vm_name()));
1013     }
1014 
1015     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1016     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1017 
1018     /* /interrupt controller */
1019     spapr_dt_xics(spapr->nr_servers, fdt, PHANDLE_XICP);
1020 
1021     ret = spapr_populate_memory(spapr, fdt);
1022     if (ret < 0) {
1023         error_report("couldn't setup memory nodes in fdt");
1024         exit(1);
1025     }
1026 
1027     /* /vdevice */
1028     spapr_dt_vdevice(spapr->vio_bus, fdt);
1029 
1030     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1031         ret = spapr_rng_populate_dt(fdt);
1032         if (ret < 0) {
1033             error_report("could not set up rng device in the fdt");
1034             exit(1);
1035         }
1036     }
1037 
1038     QLIST_FOREACH(phb, &spapr->phbs, list) {
1039         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1040         if (ret < 0) {
1041             error_report("couldn't setup PCI devices in fdt");
1042             exit(1);
1043         }
1044     }
1045 
1046     /* cpus */
1047     spapr_populate_cpus_dt_node(fdt, spapr);
1048 
1049     if (smc->dr_lmb_enabled) {
1050         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1051     }
1052 
1053     if (mc->has_hotpluggable_cpus) {
1054         int offset = fdt_path_offset(fdt, "/cpus");
1055         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1056                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1057         if (ret < 0) {
1058             error_report("Couldn't set up CPU DR device tree properties");
1059             exit(1);
1060         }
1061     }
1062 
1063     /* /event-sources */
1064     spapr_dt_events(spapr, fdt);
1065 
1066     /* /rtas */
1067     spapr_dt_rtas(spapr, fdt);
1068 
1069     /* /chosen */
1070     spapr_dt_chosen(spapr, fdt);
1071 
1072     /* /hypervisor */
1073     if (kvm_enabled()) {
1074         spapr_dt_hypervisor(spapr, fdt);
1075     }
1076 
1077     /* Build memory reserve map */
1078     if (spapr->kernel_size) {
1079         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1080     }
1081     if (spapr->initrd_size) {
1082         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1083     }
1084 
1085     /* ibm,client-architecture-support updates */
1086     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1087     if (ret < 0) {
1088         error_report("couldn't setup CAS properties fdt");
1089         exit(1);
1090     }
1091 
1092     return fdt;
1093 }
1094 
1095 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1096 {
1097     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1098 }
1099 
1100 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1101                                     PowerPCCPU *cpu)
1102 {
1103     CPUPPCState *env = &cpu->env;
1104 
1105     /* The TCG path should also be holding the BQL at this point */
1106     g_assert(qemu_mutex_iothread_locked());
1107 
1108     if (msr_pr) {
1109         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1110         env->gpr[3] = H_PRIVILEGE;
1111     } else {
1112         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1113     }
1114 }
1115 
1116 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1117 {
1118     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1119 
1120     return spapr->patb_entry;
1121 }
1122 
1123 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1124 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1125 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1126 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1127 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1128 
1129 /*
1130  * Get the fd to access the kernel htab, re-opening it if necessary
1131  */
1132 static int get_htab_fd(sPAPRMachineState *spapr)
1133 {
1134     if (spapr->htab_fd >= 0) {
1135         return spapr->htab_fd;
1136     }
1137 
1138     spapr->htab_fd = kvmppc_get_htab_fd(false);
1139     if (spapr->htab_fd < 0) {
1140         error_report("Unable to open fd for reading hash table from KVM: %s",
1141                      strerror(errno));
1142     }
1143 
1144     return spapr->htab_fd;
1145 }
1146 
1147 void close_htab_fd(sPAPRMachineState *spapr)
1148 {
1149     if (spapr->htab_fd >= 0) {
1150         close(spapr->htab_fd);
1151     }
1152     spapr->htab_fd = -1;
1153 }
1154 
1155 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1156 {
1157     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1158 
1159     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1160 }
1161 
1162 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1163                                                 hwaddr ptex, int n)
1164 {
1165     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1166     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1167 
1168     if (!spapr->htab) {
1169         /*
1170          * HTAB is controlled by KVM. Fetch into temporary buffer
1171          */
1172         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1173         kvmppc_read_hptes(hptes, ptex, n);
1174         return hptes;
1175     }
1176 
1177     /*
1178      * HTAB is controlled by QEMU. Just point to the internally
1179      * accessible PTEG.
1180      */
1181     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1182 }
1183 
1184 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1185                               const ppc_hash_pte64_t *hptes,
1186                               hwaddr ptex, int n)
1187 {
1188     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1189 
1190     if (!spapr->htab) {
1191         g_free((void *)hptes);
1192     }
1193 
1194     /* Nothing to do for qemu managed HPT */
1195 }
1196 
1197 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1198                              uint64_t pte0, uint64_t pte1)
1199 {
1200     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1201     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1202 
1203     if (!spapr->htab) {
1204         kvmppc_write_hpte(ptex, pte0, pte1);
1205     } else {
1206         stq_p(spapr->htab + offset, pte0);
1207         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1208     }
1209 }
1210 
1211 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1212 {
1213     int shift;
1214 
1215     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1216      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1217      * that's much more than is needed for Linux guests */
1218     shift = ctz64(pow2ceil(ramsize)) - 7;
1219     shift = MAX(shift, 18); /* Minimum architected size */
1220     shift = MIN(shift, 46); /* Maximum architected size */
1221     return shift;
1222 }
1223 
1224 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1225                                  Error **errp)
1226 {
1227     long rc;
1228 
1229     /* Clean up any HPT info from a previous boot */
1230     g_free(spapr->htab);
1231     spapr->htab = NULL;
1232     spapr->htab_shift = 0;
1233     close_htab_fd(spapr);
1234 
1235     rc = kvmppc_reset_htab(shift);
1236     if (rc < 0) {
1237         /* kernel-side HPT needed, but couldn't allocate one */
1238         error_setg_errno(errp, errno,
1239                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1240                          shift);
1241         /* This is almost certainly fatal, but if the caller really
1242          * wants to carry on with shift == 0, it's welcome to try */
1243     } else if (rc > 0) {
1244         /* kernel-side HPT allocated */
1245         if (rc != shift) {
1246             error_setg(errp,
1247                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1248                        shift, rc);
1249         }
1250 
1251         spapr->htab_shift = shift;
1252         spapr->htab = NULL;
1253     } else {
1254         /* kernel-side HPT not needed, allocate in userspace instead */
1255         size_t size = 1ULL << shift;
1256         int i;
1257 
1258         spapr->htab = qemu_memalign(size, size);
1259         if (!spapr->htab) {
1260             error_setg_errno(errp, errno,
1261                              "Could not allocate HPT of order %d", shift);
1262             return;
1263         }
1264 
1265         memset(spapr->htab, 0, size);
1266         spapr->htab_shift = shift;
1267 
1268         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1269             DIRTY_HPTE(HPTE(spapr->htab, i));
1270         }
1271     }
1272 }
1273 
1274 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1275 {
1276     spapr_reallocate_hpt(spapr,
1277                      spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size),
1278                      &error_fatal);
1279     if (spapr->vrma_adjust) {
1280         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1281                                           spapr->htab_shift);
1282     }
1283     /* We're setting up a hash table, so that means we're not radix */
1284     spapr->patb_entry = 0;
1285 }
1286 
1287 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1288 {
1289     bool matched = false;
1290 
1291     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1292         matched = true;
1293     }
1294 
1295     if (!matched) {
1296         error_report("Device %s is not supported by this machine yet.",
1297                      qdev_fw_name(DEVICE(sbdev)));
1298         exit(1);
1299     }
1300 }
1301 
1302 static void ppc_spapr_reset(void)
1303 {
1304     MachineState *machine = MACHINE(qdev_get_machine());
1305     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1306     PowerPCCPU *first_ppc_cpu;
1307     uint32_t rtas_limit;
1308     hwaddr rtas_addr, fdt_addr;
1309     void *fdt;
1310     int rc;
1311 
1312     /* Check for unknown sysbus devices */
1313     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1314 
1315     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1316         /* If using KVM with radix mode available, VCPUs can be started
1317          * without a HPT because KVM will start them in radix mode.
1318          * Set the GR bit in PATB so that we know there is no HPT. */
1319         spapr->patb_entry = PATBE1_GR;
1320     } else {
1321         spapr->patb_entry = 0;
1322         spapr_setup_hpt_and_vrma(spapr);
1323     }
1324 
1325     qemu_devices_reset();
1326 
1327     /*
1328      * We place the device tree and RTAS just below either the top of the RMA,
1329      * or just below 2GB, whichever is lowere, so that it can be
1330      * processed with 32-bit real mode code if necessary
1331      */
1332     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1333     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1334     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1335 
1336     /* if this reset wasn't generated by CAS, we should reset our
1337      * negotiated options and start from scratch */
1338     if (!spapr->cas_reboot) {
1339         spapr_ovec_cleanup(spapr->ov5_cas);
1340         spapr->ov5_cas = spapr_ovec_new();
1341     }
1342 
1343     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1344 
1345     spapr_load_rtas(spapr, fdt, rtas_addr);
1346 
1347     rc = fdt_pack(fdt);
1348 
1349     /* Should only fail if we've built a corrupted tree */
1350     assert(rc == 0);
1351 
1352     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1353         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1354                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1355         exit(1);
1356     }
1357 
1358     /* Load the fdt */
1359     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1360     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1361     g_free(fdt);
1362 
1363     /* Set up the entry state */
1364     first_ppc_cpu = POWERPC_CPU(first_cpu);
1365     first_ppc_cpu->env.gpr[3] = fdt_addr;
1366     first_ppc_cpu->env.gpr[5] = 0;
1367     first_cpu->halted = 0;
1368     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1369 
1370     spapr->cas_reboot = false;
1371 }
1372 
1373 static void spapr_create_nvram(sPAPRMachineState *spapr)
1374 {
1375     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1376     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1377 
1378     if (dinfo) {
1379         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1380                             &error_fatal);
1381     }
1382 
1383     qdev_init_nofail(dev);
1384 
1385     spapr->nvram = (struct sPAPRNVRAM *)dev;
1386 }
1387 
1388 static void spapr_rtc_create(sPAPRMachineState *spapr)
1389 {
1390     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1391     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1392                               &error_fatal);
1393     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1394                               &error_fatal);
1395     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1396                               "date", &error_fatal);
1397 }
1398 
1399 /* Returns whether we want to use VGA or not */
1400 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1401 {
1402     switch (vga_interface_type) {
1403     case VGA_NONE:
1404         return false;
1405     case VGA_DEVICE:
1406         return true;
1407     case VGA_STD:
1408     case VGA_VIRTIO:
1409         return pci_vga_init(pci_bus) != NULL;
1410     default:
1411         error_setg(errp,
1412                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1413         return false;
1414     }
1415 }
1416 
1417 static int spapr_post_load(void *opaque, int version_id)
1418 {
1419     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1420     int err = 0;
1421 
1422     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1423         CPUState *cs;
1424         CPU_FOREACH(cs) {
1425             PowerPCCPU *cpu = POWERPC_CPU(cs);
1426             icp_resend(ICP(cpu->intc));
1427         }
1428     }
1429 
1430     /* In earlier versions, there was no separate qdev for the PAPR
1431      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1432      * So when migrating from those versions, poke the incoming offset
1433      * value into the RTC device */
1434     if (version_id < 3) {
1435         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1436     }
1437 
1438     return err;
1439 }
1440 
1441 static bool version_before_3(void *opaque, int version_id)
1442 {
1443     return version_id < 3;
1444 }
1445 
1446 static bool spapr_ov5_cas_needed(void *opaque)
1447 {
1448     sPAPRMachineState *spapr = opaque;
1449     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1450     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1451     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1452     bool cas_needed;
1453 
1454     /* Prior to the introduction of sPAPROptionVector, we had two option
1455      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1456      * Both of these options encode machine topology into the device-tree
1457      * in such a way that the now-booted OS should still be able to interact
1458      * appropriately with QEMU regardless of what options were actually
1459      * negotiatied on the source side.
1460      *
1461      * As such, we can avoid migrating the CAS-negotiated options if these
1462      * are the only options available on the current machine/platform.
1463      * Since these are the only options available for pseries-2.7 and
1464      * earlier, this allows us to maintain old->new/new->old migration
1465      * compatibility.
1466      *
1467      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1468      * via default pseries-2.8 machines and explicit command-line parameters.
1469      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1470      * of the actual CAS-negotiated values to continue working properly. For
1471      * example, availability of memory unplug depends on knowing whether
1472      * OV5_HP_EVT was negotiated via CAS.
1473      *
1474      * Thus, for any cases where the set of available CAS-negotiatable
1475      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1476      * include the CAS-negotiated options in the migration stream.
1477      */
1478     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1479     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1480 
1481     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1482      * the mask itself since in the future it's possible "legacy" bits may be
1483      * removed via machine options, which could generate a false positive
1484      * that breaks migration.
1485      */
1486     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1487     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1488 
1489     spapr_ovec_cleanup(ov5_mask);
1490     spapr_ovec_cleanup(ov5_legacy);
1491     spapr_ovec_cleanup(ov5_removed);
1492 
1493     return cas_needed;
1494 }
1495 
1496 static const VMStateDescription vmstate_spapr_ov5_cas = {
1497     .name = "spapr_option_vector_ov5_cas",
1498     .version_id = 1,
1499     .minimum_version_id = 1,
1500     .needed = spapr_ov5_cas_needed,
1501     .fields = (VMStateField[]) {
1502         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1503                                  vmstate_spapr_ovec, sPAPROptionVector),
1504         VMSTATE_END_OF_LIST()
1505     },
1506 };
1507 
1508 static bool spapr_patb_entry_needed(void *opaque)
1509 {
1510     sPAPRMachineState *spapr = opaque;
1511 
1512     return !!spapr->patb_entry;
1513 }
1514 
1515 static const VMStateDescription vmstate_spapr_patb_entry = {
1516     .name = "spapr_patb_entry",
1517     .version_id = 1,
1518     .minimum_version_id = 1,
1519     .needed = spapr_patb_entry_needed,
1520     .fields = (VMStateField[]) {
1521         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1522         VMSTATE_END_OF_LIST()
1523     },
1524 };
1525 
1526 static const VMStateDescription vmstate_spapr = {
1527     .name = "spapr",
1528     .version_id = 3,
1529     .minimum_version_id = 1,
1530     .post_load = spapr_post_load,
1531     .fields = (VMStateField[]) {
1532         /* used to be @next_irq */
1533         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1534 
1535         /* RTC offset */
1536         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1537 
1538         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1539         VMSTATE_END_OF_LIST()
1540     },
1541     .subsections = (const VMStateDescription*[]) {
1542         &vmstate_spapr_ov5_cas,
1543         &vmstate_spapr_patb_entry,
1544         NULL
1545     }
1546 };
1547 
1548 static int htab_save_setup(QEMUFile *f, void *opaque)
1549 {
1550     sPAPRMachineState *spapr = opaque;
1551 
1552     /* "Iteration" header */
1553     qemu_put_be32(f, spapr->htab_shift);
1554 
1555     if (spapr->htab) {
1556         spapr->htab_save_index = 0;
1557         spapr->htab_first_pass = true;
1558     } else {
1559         assert(kvm_enabled());
1560     }
1561 
1562 
1563     return 0;
1564 }
1565 
1566 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1567                                  int64_t max_ns)
1568 {
1569     bool has_timeout = max_ns != -1;
1570     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1571     int index = spapr->htab_save_index;
1572     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1573 
1574     assert(spapr->htab_first_pass);
1575 
1576     do {
1577         int chunkstart;
1578 
1579         /* Consume invalid HPTEs */
1580         while ((index < htabslots)
1581                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1582             CLEAN_HPTE(HPTE(spapr->htab, index));
1583             index++;
1584         }
1585 
1586         /* Consume valid HPTEs */
1587         chunkstart = index;
1588         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1589                && HPTE_VALID(HPTE(spapr->htab, index))) {
1590             CLEAN_HPTE(HPTE(spapr->htab, index));
1591             index++;
1592         }
1593 
1594         if (index > chunkstart) {
1595             int n_valid = index - chunkstart;
1596 
1597             qemu_put_be32(f, chunkstart);
1598             qemu_put_be16(f, n_valid);
1599             qemu_put_be16(f, 0);
1600             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1601                             HASH_PTE_SIZE_64 * n_valid);
1602 
1603             if (has_timeout &&
1604                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1605                 break;
1606             }
1607         }
1608     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1609 
1610     if (index >= htabslots) {
1611         assert(index == htabslots);
1612         index = 0;
1613         spapr->htab_first_pass = false;
1614     }
1615     spapr->htab_save_index = index;
1616 }
1617 
1618 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1619                                 int64_t max_ns)
1620 {
1621     bool final = max_ns < 0;
1622     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1623     int examined = 0, sent = 0;
1624     int index = spapr->htab_save_index;
1625     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1626 
1627     assert(!spapr->htab_first_pass);
1628 
1629     do {
1630         int chunkstart, invalidstart;
1631 
1632         /* Consume non-dirty HPTEs */
1633         while ((index < htabslots)
1634                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1635             index++;
1636             examined++;
1637         }
1638 
1639         chunkstart = index;
1640         /* Consume valid dirty HPTEs */
1641         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1642                && HPTE_DIRTY(HPTE(spapr->htab, index))
1643                && HPTE_VALID(HPTE(spapr->htab, index))) {
1644             CLEAN_HPTE(HPTE(spapr->htab, index));
1645             index++;
1646             examined++;
1647         }
1648 
1649         invalidstart = index;
1650         /* Consume invalid dirty HPTEs */
1651         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1652                && HPTE_DIRTY(HPTE(spapr->htab, index))
1653                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1654             CLEAN_HPTE(HPTE(spapr->htab, index));
1655             index++;
1656             examined++;
1657         }
1658 
1659         if (index > chunkstart) {
1660             int n_valid = invalidstart - chunkstart;
1661             int n_invalid = index - invalidstart;
1662 
1663             qemu_put_be32(f, chunkstart);
1664             qemu_put_be16(f, n_valid);
1665             qemu_put_be16(f, n_invalid);
1666             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1667                             HASH_PTE_SIZE_64 * n_valid);
1668             sent += index - chunkstart;
1669 
1670             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1671                 break;
1672             }
1673         }
1674 
1675         if (examined >= htabslots) {
1676             break;
1677         }
1678 
1679         if (index >= htabslots) {
1680             assert(index == htabslots);
1681             index = 0;
1682         }
1683     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1684 
1685     if (index >= htabslots) {
1686         assert(index == htabslots);
1687         index = 0;
1688     }
1689 
1690     spapr->htab_save_index = index;
1691 
1692     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1693 }
1694 
1695 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1696 #define MAX_KVM_BUF_SIZE    2048
1697 
1698 static int htab_save_iterate(QEMUFile *f, void *opaque)
1699 {
1700     sPAPRMachineState *spapr = opaque;
1701     int fd;
1702     int rc = 0;
1703 
1704     /* Iteration header */
1705     qemu_put_be32(f, 0);
1706 
1707     if (!spapr->htab) {
1708         assert(kvm_enabled());
1709 
1710         fd = get_htab_fd(spapr);
1711         if (fd < 0) {
1712             return fd;
1713         }
1714 
1715         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1716         if (rc < 0) {
1717             return rc;
1718         }
1719     } else  if (spapr->htab_first_pass) {
1720         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1721     } else {
1722         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1723     }
1724 
1725     /* End marker */
1726     qemu_put_be32(f, 0);
1727     qemu_put_be16(f, 0);
1728     qemu_put_be16(f, 0);
1729 
1730     return rc;
1731 }
1732 
1733 static int htab_save_complete(QEMUFile *f, void *opaque)
1734 {
1735     sPAPRMachineState *spapr = opaque;
1736     int fd;
1737 
1738     /* Iteration header */
1739     qemu_put_be32(f, 0);
1740 
1741     if (!spapr->htab) {
1742         int rc;
1743 
1744         assert(kvm_enabled());
1745 
1746         fd = get_htab_fd(spapr);
1747         if (fd < 0) {
1748             return fd;
1749         }
1750 
1751         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1752         if (rc < 0) {
1753             return rc;
1754         }
1755     } else {
1756         if (spapr->htab_first_pass) {
1757             htab_save_first_pass(f, spapr, -1);
1758         }
1759         htab_save_later_pass(f, spapr, -1);
1760     }
1761 
1762     /* End marker */
1763     qemu_put_be32(f, 0);
1764     qemu_put_be16(f, 0);
1765     qemu_put_be16(f, 0);
1766 
1767     return 0;
1768 }
1769 
1770 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1771 {
1772     sPAPRMachineState *spapr = opaque;
1773     uint32_t section_hdr;
1774     int fd = -1;
1775 
1776     if (version_id < 1 || version_id > 1) {
1777         error_report("htab_load() bad version");
1778         return -EINVAL;
1779     }
1780 
1781     section_hdr = qemu_get_be32(f);
1782 
1783     if (section_hdr) {
1784         Error *local_err = NULL;
1785 
1786         /* First section gives the htab size */
1787         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1788         if (local_err) {
1789             error_report_err(local_err);
1790             return -EINVAL;
1791         }
1792         return 0;
1793     }
1794 
1795     if (!spapr->htab) {
1796         assert(kvm_enabled());
1797 
1798         fd = kvmppc_get_htab_fd(true);
1799         if (fd < 0) {
1800             error_report("Unable to open fd to restore KVM hash table: %s",
1801                          strerror(errno));
1802         }
1803     }
1804 
1805     while (true) {
1806         uint32_t index;
1807         uint16_t n_valid, n_invalid;
1808 
1809         index = qemu_get_be32(f);
1810         n_valid = qemu_get_be16(f);
1811         n_invalid = qemu_get_be16(f);
1812 
1813         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1814             /* End of Stream */
1815             break;
1816         }
1817 
1818         if ((index + n_valid + n_invalid) >
1819             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1820             /* Bad index in stream */
1821             error_report(
1822                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1823                 index, n_valid, n_invalid, spapr->htab_shift);
1824             return -EINVAL;
1825         }
1826 
1827         if (spapr->htab) {
1828             if (n_valid) {
1829                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1830                                 HASH_PTE_SIZE_64 * n_valid);
1831             }
1832             if (n_invalid) {
1833                 memset(HPTE(spapr->htab, index + n_valid), 0,
1834                        HASH_PTE_SIZE_64 * n_invalid);
1835             }
1836         } else {
1837             int rc;
1838 
1839             assert(fd >= 0);
1840 
1841             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1842             if (rc < 0) {
1843                 return rc;
1844             }
1845         }
1846     }
1847 
1848     if (!spapr->htab) {
1849         assert(fd >= 0);
1850         close(fd);
1851     }
1852 
1853     return 0;
1854 }
1855 
1856 static void htab_cleanup(void *opaque)
1857 {
1858     sPAPRMachineState *spapr = opaque;
1859 
1860     close_htab_fd(spapr);
1861 }
1862 
1863 static SaveVMHandlers savevm_htab_handlers = {
1864     .save_live_setup = htab_save_setup,
1865     .save_live_iterate = htab_save_iterate,
1866     .save_live_complete_precopy = htab_save_complete,
1867     .cleanup = htab_cleanup,
1868     .load_state = htab_load,
1869 };
1870 
1871 static void spapr_boot_set(void *opaque, const char *boot_device,
1872                            Error **errp)
1873 {
1874     MachineState *machine = MACHINE(qdev_get_machine());
1875     machine->boot_order = g_strdup(boot_device);
1876 }
1877 
1878 /*
1879  * Reset routine for LMB DR devices.
1880  *
1881  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1882  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1883  * when it walks all its children devices. LMB devices reset occurs
1884  * as part of spapr_ppc_reset().
1885  */
1886 static void spapr_drc_reset(void *opaque)
1887 {
1888     sPAPRDRConnector *drc = opaque;
1889     DeviceState *d = DEVICE(drc);
1890 
1891     if (d) {
1892         device_reset(d);
1893     }
1894 }
1895 
1896 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1897 {
1898     MachineState *machine = MACHINE(spapr);
1899     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1900     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1901     int i;
1902 
1903     for (i = 0; i < nr_lmbs; i++) {
1904         sPAPRDRConnector *drc;
1905         uint64_t addr;
1906 
1907         addr = i * lmb_size + spapr->hotplug_memory.base;
1908         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1909                                      addr/lmb_size);
1910         qemu_register_reset(spapr_drc_reset, drc);
1911     }
1912 }
1913 
1914 /*
1915  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1916  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1917  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1918  */
1919 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1920 {
1921     int i;
1922 
1923     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1924         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1925                    " is not aligned to %llu MiB",
1926                    machine->ram_size,
1927                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1928         return;
1929     }
1930 
1931     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1932         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1933                    " is not aligned to %llu MiB",
1934                    machine->ram_size,
1935                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1936         return;
1937     }
1938 
1939     for (i = 0; i < nb_numa_nodes; i++) {
1940         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1941             error_setg(errp,
1942                        "Node %d memory size 0x%" PRIx64
1943                        " is not aligned to %llu MiB",
1944                        i, numa_info[i].node_mem,
1945                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1946             return;
1947         }
1948     }
1949 }
1950 
1951 /* find cpu slot in machine->possible_cpus by core_id */
1952 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1953 {
1954     int index = id / smp_threads;
1955 
1956     if (index >= ms->possible_cpus->len) {
1957         return NULL;
1958     }
1959     if (idx) {
1960         *idx = index;
1961     }
1962     return &ms->possible_cpus->cpus[index];
1963 }
1964 
1965 static void spapr_init_cpus(sPAPRMachineState *spapr)
1966 {
1967     MachineState *machine = MACHINE(spapr);
1968     MachineClass *mc = MACHINE_GET_CLASS(machine);
1969     char *type = spapr_get_cpu_core_type(machine->cpu_model);
1970     int smt = kvmppc_smt_threads();
1971     const CPUArchIdList *possible_cpus;
1972     int boot_cores_nr = smp_cpus / smp_threads;
1973     int i;
1974 
1975     if (!type) {
1976         error_report("Unable to find sPAPR CPU Core definition");
1977         exit(1);
1978     }
1979 
1980     possible_cpus = mc->possible_cpu_arch_ids(machine);
1981     if (mc->has_hotpluggable_cpus) {
1982         if (smp_cpus % smp_threads) {
1983             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1984                          smp_cpus, smp_threads);
1985             exit(1);
1986         }
1987         if (max_cpus % smp_threads) {
1988             error_report("max_cpus (%u) must be multiple of threads (%u)",
1989                          max_cpus, smp_threads);
1990             exit(1);
1991         }
1992     } else {
1993         if (max_cpus != smp_cpus) {
1994             error_report("This machine version does not support CPU hotplug");
1995             exit(1);
1996         }
1997         boot_cores_nr = possible_cpus->len;
1998     }
1999 
2000     for (i = 0; i < possible_cpus->len; i++) {
2001         int core_id = i * smp_threads;
2002 
2003         if (mc->has_hotpluggable_cpus) {
2004             sPAPRDRConnector *drc =
2005                 spapr_dr_connector_new(OBJECT(spapr),
2006                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
2007                                        (core_id / smp_threads) * smt);
2008 
2009             qemu_register_reset(spapr_drc_reset, drc);
2010         }
2011 
2012         if (i < boot_cores_nr) {
2013             Object *core  = object_new(type);
2014             int nr_threads = smp_threads;
2015 
2016             /* Handle the partially filled core for older machine types */
2017             if ((i + 1) * smp_threads >= smp_cpus) {
2018                 nr_threads = smp_cpus - i * smp_threads;
2019             }
2020 
2021             object_property_set_int(core, nr_threads, "nr-threads",
2022                                     &error_fatal);
2023             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2024                                     &error_fatal);
2025             object_property_set_bool(core, true, "realized", &error_fatal);
2026         }
2027     }
2028     g_free(type);
2029 }
2030 
2031 /* pSeries LPAR / sPAPR hardware init */
2032 static void ppc_spapr_init(MachineState *machine)
2033 {
2034     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2035     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2036     const char *kernel_filename = machine->kernel_filename;
2037     const char *initrd_filename = machine->initrd_filename;
2038     PCIHostState *phb;
2039     int i;
2040     MemoryRegion *sysmem = get_system_memory();
2041     MemoryRegion *ram = g_new(MemoryRegion, 1);
2042     MemoryRegion *rma_region;
2043     void *rma = NULL;
2044     hwaddr rma_alloc_size;
2045     hwaddr node0_size = spapr_node0_size();
2046     long load_limit, fw_size;
2047     char *filename;
2048     int smt = kvmppc_smt_threads();
2049 
2050     msi_nonbroken = true;
2051 
2052     QLIST_INIT(&spapr->phbs);
2053 
2054     /* Allocate RMA if necessary */
2055     rma_alloc_size = kvmppc_alloc_rma(&rma);
2056 
2057     if (rma_alloc_size == -1) {
2058         error_report("Unable to create RMA");
2059         exit(1);
2060     }
2061 
2062     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
2063         spapr->rma_size = rma_alloc_size;
2064     } else {
2065         spapr->rma_size = node0_size;
2066 
2067         /* With KVM, we don't actually know whether KVM supports an
2068          * unbounded RMA (PR KVM) or is limited by the hash table size
2069          * (HV KVM using VRMA), so we always assume the latter
2070          *
2071          * In that case, we also limit the initial allocations for RTAS
2072          * etc... to 256M since we have no way to know what the VRMA size
2073          * is going to be as it depends on the size of the hash table
2074          * isn't determined yet.
2075          */
2076         if (kvm_enabled()) {
2077             spapr->vrma_adjust = 1;
2078             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2079         }
2080 
2081         /* Actually we don't support unbounded RMA anymore since we
2082          * added proper emulation of HV mode. The max we can get is
2083          * 16G which also happens to be what we configure for PAPR
2084          * mode so make sure we don't do anything bigger than that
2085          */
2086         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2087     }
2088 
2089     if (spapr->rma_size > node0_size) {
2090         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2091                      spapr->rma_size);
2092         exit(1);
2093     }
2094 
2095     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2096     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2097 
2098     /* Set up Interrupt Controller before we create the VCPUs */
2099     xics_system_init(machine, DIV_ROUND_UP(max_cpus * smt, smp_threads),
2100                      XICS_IRQS_SPAPR, &error_fatal);
2101 
2102     /* Set up containers for ibm,client-set-architecture negotiated options */
2103     spapr->ov5 = spapr_ovec_new();
2104     spapr->ov5_cas = spapr_ovec_new();
2105 
2106     if (smc->dr_lmb_enabled) {
2107         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2108         spapr_validate_node_memory(machine, &error_fatal);
2109     }
2110 
2111     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2112     if (kvmppc_has_cap_mmu_radix()) {
2113         /* KVM always allows GTSE with radix... */
2114         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2115     }
2116     /* ... but not with hash (currently). */
2117 
2118     /* advertise support for dedicated HP event source to guests */
2119     if (spapr->use_hotplug_event_source) {
2120         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2121     }
2122 
2123     /* init CPUs */
2124     if (machine->cpu_model == NULL) {
2125         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2126     }
2127 
2128     ppc_cpu_parse_features(machine->cpu_model);
2129 
2130     spapr_init_cpus(spapr);
2131 
2132     if (kvm_enabled()) {
2133         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2134         kvmppc_enable_logical_ci_hcalls();
2135         kvmppc_enable_set_mode_hcall();
2136 
2137         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2138         kvmppc_enable_clear_ref_mod_hcalls();
2139     }
2140 
2141     /* allocate RAM */
2142     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2143                                          machine->ram_size);
2144     memory_region_add_subregion(sysmem, 0, ram);
2145 
2146     if (rma_alloc_size && rma) {
2147         rma_region = g_new(MemoryRegion, 1);
2148         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2149                                    rma_alloc_size, rma);
2150         vmstate_register_ram_global(rma_region);
2151         memory_region_add_subregion(sysmem, 0, rma_region);
2152     }
2153 
2154     /* initialize hotplug memory address space */
2155     if (machine->ram_size < machine->maxram_size) {
2156         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2157         /*
2158          * Limit the number of hotpluggable memory slots to half the number
2159          * slots that KVM supports, leaving the other half for PCI and other
2160          * devices. However ensure that number of slots doesn't drop below 32.
2161          */
2162         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2163                            SPAPR_MAX_RAM_SLOTS;
2164 
2165         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2166             max_memslots = SPAPR_MAX_RAM_SLOTS;
2167         }
2168         if (machine->ram_slots > max_memslots) {
2169             error_report("Specified number of memory slots %"
2170                          PRIu64" exceeds max supported %d",
2171                          machine->ram_slots, max_memslots);
2172             exit(1);
2173         }
2174 
2175         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2176                                               SPAPR_HOTPLUG_MEM_ALIGN);
2177         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2178                            "hotplug-memory", hotplug_mem_size);
2179         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2180                                     &spapr->hotplug_memory.mr);
2181     }
2182 
2183     if (smc->dr_lmb_enabled) {
2184         spapr_create_lmb_dr_connectors(spapr);
2185     }
2186 
2187     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2188     if (!filename) {
2189         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2190         exit(1);
2191     }
2192     spapr->rtas_size = get_image_size(filename);
2193     if (spapr->rtas_size < 0) {
2194         error_report("Could not get size of LPAR rtas '%s'", filename);
2195         exit(1);
2196     }
2197     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2198     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2199         error_report("Could not load LPAR rtas '%s'", filename);
2200         exit(1);
2201     }
2202     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2203         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2204                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2205         exit(1);
2206     }
2207     g_free(filename);
2208 
2209     /* Set up RTAS event infrastructure */
2210     spapr_events_init(spapr);
2211 
2212     /* Set up the RTC RTAS interfaces */
2213     spapr_rtc_create(spapr);
2214 
2215     /* Set up VIO bus */
2216     spapr->vio_bus = spapr_vio_bus_init();
2217 
2218     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2219         if (serial_hds[i]) {
2220             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2221         }
2222     }
2223 
2224     /* We always have at least the nvram device on VIO */
2225     spapr_create_nvram(spapr);
2226 
2227     /* Set up PCI */
2228     spapr_pci_rtas_init();
2229 
2230     phb = spapr_create_phb(spapr, 0);
2231 
2232     for (i = 0; i < nb_nics; i++) {
2233         NICInfo *nd = &nd_table[i];
2234 
2235         if (!nd->model) {
2236             nd->model = g_strdup("ibmveth");
2237         }
2238 
2239         if (strcmp(nd->model, "ibmveth") == 0) {
2240             spapr_vlan_create(spapr->vio_bus, nd);
2241         } else {
2242             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2243         }
2244     }
2245 
2246     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2247         spapr_vscsi_create(spapr->vio_bus);
2248     }
2249 
2250     /* Graphics */
2251     if (spapr_vga_init(phb->bus, &error_fatal)) {
2252         spapr->has_graphics = true;
2253         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2254     }
2255 
2256     if (machine->usb) {
2257         if (smc->use_ohci_by_default) {
2258             pci_create_simple(phb->bus, -1, "pci-ohci");
2259         } else {
2260             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2261         }
2262 
2263         if (spapr->has_graphics) {
2264             USBBus *usb_bus = usb_bus_find(-1);
2265 
2266             usb_create_simple(usb_bus, "usb-kbd");
2267             usb_create_simple(usb_bus, "usb-mouse");
2268         }
2269     }
2270 
2271     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2272         error_report(
2273             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2274             MIN_RMA_SLOF);
2275         exit(1);
2276     }
2277 
2278     if (kernel_filename) {
2279         uint64_t lowaddr = 0;
2280 
2281         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2282                                       NULL, NULL, &lowaddr, NULL, 1,
2283                                       PPC_ELF_MACHINE, 0, 0);
2284         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2285             spapr->kernel_size = load_elf(kernel_filename,
2286                                           translate_kernel_address, NULL, NULL,
2287                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2288                                           0, 0);
2289             spapr->kernel_le = spapr->kernel_size > 0;
2290         }
2291         if (spapr->kernel_size < 0) {
2292             error_report("error loading %s: %s", kernel_filename,
2293                          load_elf_strerror(spapr->kernel_size));
2294             exit(1);
2295         }
2296 
2297         /* load initrd */
2298         if (initrd_filename) {
2299             /* Try to locate the initrd in the gap between the kernel
2300              * and the firmware. Add a bit of space just in case
2301              */
2302             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2303                                   + 0x1ffff) & ~0xffff;
2304             spapr->initrd_size = load_image_targphys(initrd_filename,
2305                                                      spapr->initrd_base,
2306                                                      load_limit
2307                                                      - spapr->initrd_base);
2308             if (spapr->initrd_size < 0) {
2309                 error_report("could not load initial ram disk '%s'",
2310                              initrd_filename);
2311                 exit(1);
2312             }
2313         }
2314     }
2315 
2316     if (bios_name == NULL) {
2317         bios_name = FW_FILE_NAME;
2318     }
2319     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2320     if (!filename) {
2321         error_report("Could not find LPAR firmware '%s'", bios_name);
2322         exit(1);
2323     }
2324     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2325     if (fw_size <= 0) {
2326         error_report("Could not load LPAR firmware '%s'", filename);
2327         exit(1);
2328     }
2329     g_free(filename);
2330 
2331     /* FIXME: Should register things through the MachineState's qdev
2332      * interface, this is a legacy from the sPAPREnvironment structure
2333      * which predated MachineState but had a similar function */
2334     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2335     register_savevm_live(NULL, "spapr/htab", -1, 1,
2336                          &savevm_htab_handlers, spapr);
2337 
2338     /* used by RTAS */
2339     QTAILQ_INIT(&spapr->ccs_list);
2340     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2341 
2342     qemu_register_boot_set(spapr_boot_set, spapr);
2343 
2344     if (kvm_enabled()) {
2345         /* to stop and start vmclock */
2346         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2347                                          &spapr->tb);
2348 
2349         kvmppc_spapr_enable_inkernel_multitce();
2350     }
2351 }
2352 
2353 static int spapr_kvm_type(const char *vm_type)
2354 {
2355     if (!vm_type) {
2356         return 0;
2357     }
2358 
2359     if (!strcmp(vm_type, "HV")) {
2360         return 1;
2361     }
2362 
2363     if (!strcmp(vm_type, "PR")) {
2364         return 2;
2365     }
2366 
2367     error_report("Unknown kvm-type specified '%s'", vm_type);
2368     exit(1);
2369 }
2370 
2371 /*
2372  * Implementation of an interface to adjust firmware path
2373  * for the bootindex property handling.
2374  */
2375 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2376                                    DeviceState *dev)
2377 {
2378 #define CAST(type, obj, name) \
2379     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2380     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2381     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2382 
2383     if (d) {
2384         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2385         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2386         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2387 
2388         if (spapr) {
2389             /*
2390              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2391              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2392              * in the top 16 bits of the 64-bit LUN
2393              */
2394             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2395             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2396                                    (uint64_t)id << 48);
2397         } else if (virtio) {
2398             /*
2399              * We use SRP luns of the form 01000000 | (target << 8) | lun
2400              * in the top 32 bits of the 64-bit LUN
2401              * Note: the quote above is from SLOF and it is wrong,
2402              * the actual binding is:
2403              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2404              */
2405             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2406             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2407                                    (uint64_t)id << 32);
2408         } else if (usb) {
2409             /*
2410              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2411              * in the top 32 bits of the 64-bit LUN
2412              */
2413             unsigned usb_port = atoi(usb->port->path);
2414             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2415             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2416                                    (uint64_t)id << 32);
2417         }
2418     }
2419 
2420     /*
2421      * SLOF probes the USB devices, and if it recognizes that the device is a
2422      * storage device, it changes its name to "storage" instead of "usb-host",
2423      * and additionally adds a child node for the SCSI LUN, so the correct
2424      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2425      */
2426     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2427         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2428         if (usb_host_dev_is_scsi_storage(usbdev)) {
2429             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2430         }
2431     }
2432 
2433     if (phb) {
2434         /* Replace "pci" with "pci@800000020000000" */
2435         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2436     }
2437 
2438     return NULL;
2439 }
2440 
2441 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2442 {
2443     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2444 
2445     return g_strdup(spapr->kvm_type);
2446 }
2447 
2448 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2449 {
2450     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2451 
2452     g_free(spapr->kvm_type);
2453     spapr->kvm_type = g_strdup(value);
2454 }
2455 
2456 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2457 {
2458     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2459 
2460     return spapr->use_hotplug_event_source;
2461 }
2462 
2463 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2464                                             Error **errp)
2465 {
2466     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2467 
2468     spapr->use_hotplug_event_source = value;
2469 }
2470 
2471 static void spapr_machine_initfn(Object *obj)
2472 {
2473     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2474 
2475     spapr->htab_fd = -1;
2476     spapr->use_hotplug_event_source = true;
2477     object_property_add_str(obj, "kvm-type",
2478                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2479     object_property_set_description(obj, "kvm-type",
2480                                     "Specifies the KVM virtualization mode (HV, PR)",
2481                                     NULL);
2482     object_property_add_bool(obj, "modern-hotplug-events",
2483                             spapr_get_modern_hotplug_events,
2484                             spapr_set_modern_hotplug_events,
2485                             NULL);
2486     object_property_set_description(obj, "modern-hotplug-events",
2487                                     "Use dedicated hotplug event mechanism in"
2488                                     " place of standard EPOW events when possible"
2489                                     " (required for memory hot-unplug support)",
2490                                     NULL);
2491 }
2492 
2493 static void spapr_machine_finalizefn(Object *obj)
2494 {
2495     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2496 
2497     g_free(spapr->kvm_type);
2498 }
2499 
2500 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2501 {
2502     cpu_synchronize_state(cs);
2503     ppc_cpu_do_system_reset(cs);
2504 }
2505 
2506 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2507 {
2508     CPUState *cs;
2509 
2510     CPU_FOREACH(cs) {
2511         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2512     }
2513 }
2514 
2515 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2516                            uint32_t node, bool dedicated_hp_event_source,
2517                            Error **errp)
2518 {
2519     sPAPRDRConnector *drc;
2520     sPAPRDRConnectorClass *drck;
2521     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2522     int i, fdt_offset, fdt_size;
2523     void *fdt;
2524     uint64_t addr = addr_start;
2525 
2526     for (i = 0; i < nr_lmbs; i++) {
2527         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2528                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2529         g_assert(drc);
2530 
2531         fdt = create_device_tree(&fdt_size);
2532         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2533                                                 SPAPR_MEMORY_BLOCK_SIZE);
2534 
2535         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2536         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2537         addr += SPAPR_MEMORY_BLOCK_SIZE;
2538         if (!dev->hotplugged) {
2539             /* guests expect coldplugged LMBs to be pre-allocated */
2540             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2541             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2542         }
2543     }
2544     /* send hotplug notification to the
2545      * guest only in case of hotplugged memory
2546      */
2547     if (dev->hotplugged) {
2548         if (dedicated_hp_event_source) {
2549             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2550                     addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2551             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2552             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2553                                                    nr_lmbs,
2554                                                    drck->get_index(drc));
2555         } else {
2556             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2557                                            nr_lmbs);
2558         }
2559     }
2560 }
2561 
2562 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2563                               uint32_t node, Error **errp)
2564 {
2565     Error *local_err = NULL;
2566     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2567     PCDIMMDevice *dimm = PC_DIMM(dev);
2568     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2569     MemoryRegion *mr = ddc->get_memory_region(dimm);
2570     uint64_t align = memory_region_get_alignment(mr);
2571     uint64_t size = memory_region_size(mr);
2572     uint64_t addr;
2573     char *mem_dev;
2574 
2575     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2576         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2577                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2578         goto out;
2579     }
2580 
2581     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2582     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2583         error_setg(&local_err, "Memory backend has bad page size. "
2584                    "Use 'memory-backend-file' with correct mem-path.");
2585         goto out;
2586     }
2587 
2588     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2589     if (local_err) {
2590         goto out;
2591     }
2592 
2593     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2594     if (local_err) {
2595         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2596         goto out;
2597     }
2598 
2599     spapr_add_lmbs(dev, addr, size, node,
2600                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2601                    &error_abort);
2602 
2603 out:
2604     error_propagate(errp, local_err);
2605 }
2606 
2607 typedef struct sPAPRDIMMState {
2608     uint32_t nr_lmbs;
2609 } sPAPRDIMMState;
2610 
2611 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2612 {
2613     sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2614     HotplugHandler *hotplug_ctrl;
2615 
2616     if (--ds->nr_lmbs) {
2617         return;
2618     }
2619 
2620     g_free(ds);
2621 
2622     /*
2623      * Now that all the LMBs have been removed by the guest, call the
2624      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2625      */
2626     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2627     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2628 }
2629 
2630 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2631                            Error **errp)
2632 {
2633     sPAPRDRConnector *drc;
2634     sPAPRDRConnectorClass *drck;
2635     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2636     int i;
2637     sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2638     uint64_t addr = addr_start;
2639 
2640     ds->nr_lmbs = nr_lmbs;
2641     for (i = 0; i < nr_lmbs; i++) {
2642         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2643                 addr / SPAPR_MEMORY_BLOCK_SIZE);
2644         g_assert(drc);
2645 
2646         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2647         drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2648         addr += SPAPR_MEMORY_BLOCK_SIZE;
2649     }
2650 
2651     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2652                                    addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2653     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2654     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2655                                               nr_lmbs,
2656                                               drck->get_index(drc));
2657 }
2658 
2659 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2660                                 Error **errp)
2661 {
2662     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2663     PCDIMMDevice *dimm = PC_DIMM(dev);
2664     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2665     MemoryRegion *mr = ddc->get_memory_region(dimm);
2666 
2667     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2668     object_unparent(OBJECT(dev));
2669 }
2670 
2671 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2672                                         DeviceState *dev, Error **errp)
2673 {
2674     Error *local_err = NULL;
2675     PCDIMMDevice *dimm = PC_DIMM(dev);
2676     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2677     MemoryRegion *mr = ddc->get_memory_region(dimm);
2678     uint64_t size = memory_region_size(mr);
2679     uint64_t addr;
2680 
2681     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2682     if (local_err) {
2683         goto out;
2684     }
2685 
2686     spapr_del_lmbs(dev, addr, size, &error_abort);
2687 out:
2688     error_propagate(errp, local_err);
2689 }
2690 
2691 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2692                                     sPAPRMachineState *spapr)
2693 {
2694     PowerPCCPU *cpu = POWERPC_CPU(cs);
2695     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2696     int id = ppc_get_vcpu_dt_id(cpu);
2697     void *fdt;
2698     int offset, fdt_size;
2699     char *nodename;
2700 
2701     fdt = create_device_tree(&fdt_size);
2702     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2703     offset = fdt_add_subnode(fdt, 0, nodename);
2704 
2705     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2706     g_free(nodename);
2707 
2708     *fdt_offset = offset;
2709     return fdt;
2710 }
2711 
2712 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2713                               Error **errp)
2714 {
2715     MachineState *ms = MACHINE(qdev_get_machine());
2716     CPUCore *cc = CPU_CORE(dev);
2717     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2718 
2719     core_slot->cpu = NULL;
2720     object_unparent(OBJECT(dev));
2721 }
2722 
2723 static void spapr_core_release(DeviceState *dev, void *opaque)
2724 {
2725     HotplugHandler *hotplug_ctrl;
2726 
2727     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2728     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2729 }
2730 
2731 static
2732 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2733                                Error **errp)
2734 {
2735     int index;
2736     sPAPRDRConnector *drc;
2737     sPAPRDRConnectorClass *drck;
2738     Error *local_err = NULL;
2739     CPUCore *cc = CPU_CORE(dev);
2740     int smt = kvmppc_smt_threads();
2741 
2742     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2743         error_setg(errp, "Unable to find CPU core with core-id: %d",
2744                    cc->core_id);
2745         return;
2746     }
2747     if (index == 0) {
2748         error_setg(errp, "Boot CPU core may not be unplugged");
2749         return;
2750     }
2751 
2752     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2753     g_assert(drc);
2754 
2755     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2756     drck->detach(drc, dev, spapr_core_release, NULL, &local_err);
2757     if (local_err) {
2758         error_propagate(errp, local_err);
2759         return;
2760     }
2761 
2762     spapr_hotplug_req_remove_by_index(drc);
2763 }
2764 
2765 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2766                             Error **errp)
2767 {
2768     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2769     MachineClass *mc = MACHINE_GET_CLASS(spapr);
2770     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2771     CPUCore *cc = CPU_CORE(dev);
2772     CPUState *cs = CPU(core->threads);
2773     sPAPRDRConnector *drc;
2774     Error *local_err = NULL;
2775     void *fdt = NULL;
2776     int fdt_offset = 0;
2777     int smt = kvmppc_smt_threads();
2778     CPUArchId *core_slot;
2779     int index;
2780 
2781     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2782     if (!core_slot) {
2783         error_setg(errp, "Unable to find CPU core with core-id: %d",
2784                    cc->core_id);
2785         return;
2786     }
2787     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2788 
2789     g_assert(drc || !mc->has_hotpluggable_cpus);
2790 
2791     /*
2792      * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2793      * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2794      */
2795     if (dev->hotplugged) {
2796         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2797     }
2798 
2799     if (drc) {
2800         sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2801         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
2802         if (local_err) {
2803             g_free(fdt);
2804             error_propagate(errp, local_err);
2805             return;
2806         }
2807     }
2808 
2809     if (dev->hotplugged) {
2810         /*
2811          * Send hotplug notification interrupt to the guest only in case
2812          * of hotplugged CPUs.
2813          */
2814         spapr_hotplug_req_add_by_index(drc);
2815     } else {
2816         /*
2817          * Set the right DRC states for cold plugged CPU.
2818          */
2819         if (drc) {
2820             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2821             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2822             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2823         }
2824     }
2825     core_slot->cpu = OBJECT(dev);
2826 }
2827 
2828 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2829                                 Error **errp)
2830 {
2831     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2832     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2833     Error *local_err = NULL;
2834     CPUCore *cc = CPU_CORE(dev);
2835     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2836     const char *type = object_get_typename(OBJECT(dev));
2837     CPUArchId *core_slot;
2838     int index;
2839 
2840     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2841         error_setg(&local_err, "CPU hotplug not supported for this machine");
2842         goto out;
2843     }
2844 
2845     if (strcmp(base_core_type, type)) {
2846         error_setg(&local_err, "CPU core type should be %s", base_core_type);
2847         goto out;
2848     }
2849 
2850     if (cc->core_id % smp_threads) {
2851         error_setg(&local_err, "invalid core id %d", cc->core_id);
2852         goto out;
2853     }
2854 
2855     if (cc->nr_threads != smp_threads) {
2856         error_setg(errp, "invalid nr-threads %d, must be %d",
2857                    cc->nr_threads, smp_threads);
2858         return;
2859     }
2860 
2861     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2862     if (!core_slot) {
2863         error_setg(&local_err, "core id %d out of range", cc->core_id);
2864         goto out;
2865     }
2866 
2867     if (core_slot->cpu) {
2868         error_setg(&local_err, "core %d already populated", cc->core_id);
2869         goto out;
2870     }
2871 
2872 out:
2873     g_free(base_core_type);
2874     error_propagate(errp, local_err);
2875 }
2876 
2877 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2878                                       DeviceState *dev, Error **errp)
2879 {
2880     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2881 
2882     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2883         int node;
2884 
2885         if (!smc->dr_lmb_enabled) {
2886             error_setg(errp, "Memory hotplug not supported for this machine");
2887             return;
2888         }
2889         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2890         if (*errp) {
2891             return;
2892         }
2893         if (node < 0 || node >= MAX_NODES) {
2894             error_setg(errp, "Invaild node %d", node);
2895             return;
2896         }
2897 
2898         /*
2899          * Currently PowerPC kernel doesn't allow hot-adding memory to
2900          * memory-less node, but instead will silently add the memory
2901          * to the first node that has some memory. This causes two
2902          * unexpected behaviours for the user.
2903          *
2904          * - Memory gets hotplugged to a different node than what the user
2905          *   specified.
2906          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2907          *   to memory-less node, a reboot will set things accordingly
2908          *   and the previously hotplugged memory now ends in the right node.
2909          *   This appears as if some memory moved from one node to another.
2910          *
2911          * So until kernel starts supporting memory hotplug to memory-less
2912          * nodes, just prevent such attempts upfront in QEMU.
2913          */
2914         if (nb_numa_nodes && !numa_info[node].node_mem) {
2915             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2916                        node);
2917             return;
2918         }
2919 
2920         spapr_memory_plug(hotplug_dev, dev, node, errp);
2921     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2922         spapr_core_plug(hotplug_dev, dev, errp);
2923     }
2924 }
2925 
2926 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2927                                       DeviceState *dev, Error **errp)
2928 {
2929     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2930     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2931 
2932     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2933         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2934             spapr_memory_unplug(hotplug_dev, dev, errp);
2935         } else {
2936             error_setg(errp, "Memory hot unplug not supported for this guest");
2937         }
2938     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2939         if (!mc->has_hotpluggable_cpus) {
2940             error_setg(errp, "CPU hot unplug not supported on this machine");
2941             return;
2942         }
2943         spapr_core_unplug(hotplug_dev, dev, errp);
2944     }
2945 }
2946 
2947 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2948                                                 DeviceState *dev, Error **errp)
2949 {
2950     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2951     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2952 
2953     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2954         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2955             spapr_memory_unplug_request(hotplug_dev, dev, errp);
2956         } else {
2957             /* NOTE: this means there is a window after guest reset, prior to
2958              * CAS negotiation, where unplug requests will fail due to the
2959              * capability not being detected yet. This is a bit different than
2960              * the case with PCI unplug, where the events will be queued and
2961              * eventually handled by the guest after boot
2962              */
2963             error_setg(errp, "Memory hot unplug not supported for this guest");
2964         }
2965     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2966         if (!mc->has_hotpluggable_cpus) {
2967             error_setg(errp, "CPU hot unplug not supported on this machine");
2968             return;
2969         }
2970         spapr_core_unplug_request(hotplug_dev, dev, errp);
2971     }
2972 }
2973 
2974 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2975                                           DeviceState *dev, Error **errp)
2976 {
2977     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2978         spapr_core_pre_plug(hotplug_dev, dev, errp);
2979     }
2980 }
2981 
2982 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2983                                                  DeviceState *dev)
2984 {
2985     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2986         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2987         return HOTPLUG_HANDLER(machine);
2988     }
2989     return NULL;
2990 }
2991 
2992 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2993 {
2994     /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2995      * socket means much for the paravirtualized PAPR platform) */
2996     return cpu_index / smp_threads / smp_cores;
2997 }
2998 
2999 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3000 {
3001     int i;
3002     int spapr_max_cores = max_cpus / smp_threads;
3003     MachineClass *mc = MACHINE_GET_CLASS(machine);
3004 
3005     if (!mc->has_hotpluggable_cpus) {
3006         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3007     }
3008     if (machine->possible_cpus) {
3009         assert(machine->possible_cpus->len == spapr_max_cores);
3010         return machine->possible_cpus;
3011     }
3012 
3013     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3014                              sizeof(CPUArchId) * spapr_max_cores);
3015     machine->possible_cpus->len = spapr_max_cores;
3016     for (i = 0; i < machine->possible_cpus->len; i++) {
3017         int core_id = i * smp_threads;
3018 
3019         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3020         machine->possible_cpus->cpus[i].arch_id = core_id;
3021         machine->possible_cpus->cpus[i].props.has_core_id = true;
3022         machine->possible_cpus->cpus[i].props.core_id = core_id;
3023         /* TODO: add 'has_node/node' here to describe
3024            to which node core belongs */
3025     }
3026     return machine->possible_cpus;
3027 }
3028 
3029 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3030                                 uint64_t *buid, hwaddr *pio,
3031                                 hwaddr *mmio32, hwaddr *mmio64,
3032                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3033 {
3034     /*
3035      * New-style PHB window placement.
3036      *
3037      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3038      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3039      * windows.
3040      *
3041      * Some guest kernels can't work with MMIO windows above 1<<46
3042      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3043      *
3044      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3045      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3046      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3047      * 1TiB 64-bit MMIO windows for each PHB.
3048      */
3049     const uint64_t base_buid = 0x800000020000000ULL;
3050 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3051                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3052     int i;
3053 
3054     /* Sanity check natural alignments */
3055     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3056     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3057     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3058     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3059     /* Sanity check bounds */
3060     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3061                       SPAPR_PCI_MEM32_WIN_SIZE);
3062     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3063                       SPAPR_PCI_MEM64_WIN_SIZE);
3064 
3065     if (index >= SPAPR_MAX_PHBS) {
3066         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3067                    SPAPR_MAX_PHBS - 1);
3068         return;
3069     }
3070 
3071     *buid = base_buid + index;
3072     for (i = 0; i < n_dma; ++i) {
3073         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3074     }
3075 
3076     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3077     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3078     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3079 }
3080 
3081 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3082 {
3083     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3084 
3085     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3086 }
3087 
3088 static void spapr_ics_resend(XICSFabric *dev)
3089 {
3090     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3091 
3092     ics_resend(spapr->ics);
3093 }
3094 
3095 static ICPState *spapr_icp_get(XICSFabric *xi, int cpu_dt_id)
3096 {
3097     PowerPCCPU *cpu = ppc_get_vcpu_by_dt_id(cpu_dt_id);
3098 
3099     return cpu ? ICP(cpu->intc) : NULL;
3100 }
3101 
3102 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3103                                  Monitor *mon)
3104 {
3105     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3106     CPUState *cs;
3107 
3108     CPU_FOREACH(cs) {
3109         PowerPCCPU *cpu = POWERPC_CPU(cs);
3110 
3111         icp_pic_print_info(ICP(cpu->intc), mon);
3112     }
3113 
3114     ics_pic_print_info(spapr->ics, mon);
3115 }
3116 
3117 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3118 {
3119     MachineClass *mc = MACHINE_CLASS(oc);
3120     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3121     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3122     NMIClass *nc = NMI_CLASS(oc);
3123     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3124     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3125     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3126     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3127 
3128     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3129 
3130     /*
3131      * We set up the default / latest behaviour here.  The class_init
3132      * functions for the specific versioned machine types can override
3133      * these details for backwards compatibility
3134      */
3135     mc->init = ppc_spapr_init;
3136     mc->reset = ppc_spapr_reset;
3137     mc->block_default_type = IF_SCSI;
3138     mc->max_cpus = 1024;
3139     mc->no_parallel = 1;
3140     mc->default_boot_order = "";
3141     mc->default_ram_size = 512 * M_BYTE;
3142     mc->kvm_type = spapr_kvm_type;
3143     mc->has_dynamic_sysbus = true;
3144     mc->pci_allow_0_address = true;
3145     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3146     hc->pre_plug = spapr_machine_device_pre_plug;
3147     hc->plug = spapr_machine_device_plug;
3148     hc->unplug = spapr_machine_device_unplug;
3149     mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
3150     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3151     hc->unplug_request = spapr_machine_device_unplug_request;
3152 
3153     smc->dr_lmb_enabled = true;
3154     smc->tcg_default_cpu = "POWER8";
3155     mc->has_hotpluggable_cpus = true;
3156     fwc->get_dev_path = spapr_get_fw_dev_path;
3157     nc->nmi_monitor_handler = spapr_nmi;
3158     smc->phb_placement = spapr_phb_placement;
3159     vhc->hypercall = emulate_spapr_hypercall;
3160     vhc->hpt_mask = spapr_hpt_mask;
3161     vhc->map_hptes = spapr_map_hptes;
3162     vhc->unmap_hptes = spapr_unmap_hptes;
3163     vhc->store_hpte = spapr_store_hpte;
3164     vhc->get_patbe = spapr_get_patbe;
3165     xic->ics_get = spapr_ics_get;
3166     xic->ics_resend = spapr_ics_resend;
3167     xic->icp_get = spapr_icp_get;
3168     ispc->print_info = spapr_pic_print_info;
3169     /* Force NUMA node memory size to be a multiple of
3170      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3171      * in which LMBs are represented and hot-added
3172      */
3173     mc->numa_mem_align_shift = 28;
3174 }
3175 
3176 static const TypeInfo spapr_machine_info = {
3177     .name          = TYPE_SPAPR_MACHINE,
3178     .parent        = TYPE_MACHINE,
3179     .abstract      = true,
3180     .instance_size = sizeof(sPAPRMachineState),
3181     .instance_init = spapr_machine_initfn,
3182     .instance_finalize = spapr_machine_finalizefn,
3183     .class_size    = sizeof(sPAPRMachineClass),
3184     .class_init    = spapr_machine_class_init,
3185     .interfaces = (InterfaceInfo[]) {
3186         { TYPE_FW_PATH_PROVIDER },
3187         { TYPE_NMI },
3188         { TYPE_HOTPLUG_HANDLER },
3189         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3190         { TYPE_XICS_FABRIC },
3191         { TYPE_INTERRUPT_STATS_PROVIDER },
3192         { }
3193     },
3194 };
3195 
3196 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3197     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3198                                                     void *data)      \
3199     {                                                                \
3200         MachineClass *mc = MACHINE_CLASS(oc);                        \
3201         spapr_machine_##suffix##_class_options(mc);                  \
3202         if (latest) {                                                \
3203             mc->alias = "pseries";                                   \
3204             mc->is_default = 1;                                      \
3205         }                                                            \
3206     }                                                                \
3207     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3208     {                                                                \
3209         MachineState *machine = MACHINE(obj);                        \
3210         spapr_machine_##suffix##_instance_options(machine);          \
3211     }                                                                \
3212     static const TypeInfo spapr_machine_##suffix##_info = {          \
3213         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3214         .parent = TYPE_SPAPR_MACHINE,                                \
3215         .class_init = spapr_machine_##suffix##_class_init,           \
3216         .instance_init = spapr_machine_##suffix##_instance_init,     \
3217     };                                                               \
3218     static void spapr_machine_register_##suffix(void)                \
3219     {                                                                \
3220         type_register(&spapr_machine_##suffix##_info);               \
3221     }                                                                \
3222     type_init(spapr_machine_register_##suffix)
3223 
3224 /*
3225  * pseries-2.10
3226  */
3227 static void spapr_machine_2_10_instance_options(MachineState *machine)
3228 {
3229 }
3230 
3231 static void spapr_machine_2_10_class_options(MachineClass *mc)
3232 {
3233     /* Defaults for the latest behaviour inherited from the base class */
3234 }
3235 
3236 DEFINE_SPAPR_MACHINE(2_10, "2.10", true);
3237 
3238 /*
3239  * pseries-2.9
3240  */
3241 #define SPAPR_COMPAT_2_9                                               \
3242     HW_COMPAT_2_9
3243 
3244 static void spapr_machine_2_9_instance_options(MachineState *machine)
3245 {
3246     spapr_machine_2_10_instance_options(machine);
3247 }
3248 
3249 static void spapr_machine_2_9_class_options(MachineClass *mc)
3250 {
3251     spapr_machine_2_10_class_options(mc);
3252     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3253 }
3254 
3255 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3256 
3257 /*
3258  * pseries-2.8
3259  */
3260 #define SPAPR_COMPAT_2_8                                        \
3261     HW_COMPAT_2_8                                               \
3262     {                                                           \
3263         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3264         .property = "pcie-extended-configuration-space",        \
3265         .value    = "off",                                      \
3266     },
3267 
3268 static void spapr_machine_2_8_instance_options(MachineState *machine)
3269 {
3270     spapr_machine_2_9_instance_options(machine);
3271 }
3272 
3273 static void spapr_machine_2_8_class_options(MachineClass *mc)
3274 {
3275     spapr_machine_2_9_class_options(mc);
3276     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3277     mc->numa_mem_align_shift = 23;
3278 }
3279 
3280 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3281 
3282 /*
3283  * pseries-2.7
3284  */
3285 #define SPAPR_COMPAT_2_7                            \
3286     HW_COMPAT_2_7                                   \
3287     {                                               \
3288         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3289         .property = "mem_win_size",                 \
3290         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3291     },                                              \
3292     {                                               \
3293         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3294         .property = "mem64_win_size",               \
3295         .value    = "0",                            \
3296     },                                              \
3297     {                                               \
3298         .driver = TYPE_POWERPC_CPU,                 \
3299         .property = "pre-2.8-migration",            \
3300         .value    = "on",                           \
3301     },                                              \
3302     {                                               \
3303         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3304         .property = "pre-2.8-migration",            \
3305         .value    = "on",                           \
3306     },
3307 
3308 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3309                               uint64_t *buid, hwaddr *pio,
3310                               hwaddr *mmio32, hwaddr *mmio64,
3311                               unsigned n_dma, uint32_t *liobns, Error **errp)
3312 {
3313     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3314     const uint64_t base_buid = 0x800000020000000ULL;
3315     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3316     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3317     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3318     const uint32_t max_index = 255;
3319     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3320 
3321     uint64_t ram_top = MACHINE(spapr)->ram_size;
3322     hwaddr phb0_base, phb_base;
3323     int i;
3324 
3325     /* Do we have hotpluggable memory? */
3326     if (MACHINE(spapr)->maxram_size > ram_top) {
3327         /* Can't just use maxram_size, because there may be an
3328          * alignment gap between normal and hotpluggable memory
3329          * regions */
3330         ram_top = spapr->hotplug_memory.base +
3331             memory_region_size(&spapr->hotplug_memory.mr);
3332     }
3333 
3334     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3335 
3336     if (index > max_index) {
3337         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3338                    max_index);
3339         return;
3340     }
3341 
3342     *buid = base_buid + index;
3343     for (i = 0; i < n_dma; ++i) {
3344         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3345     }
3346 
3347     phb_base = phb0_base + index * phb_spacing;
3348     *pio = phb_base + pio_offset;
3349     *mmio32 = phb_base + mmio_offset;
3350     /*
3351      * We don't set the 64-bit MMIO window, relying on the PHB's
3352      * fallback behaviour of automatically splitting a large "32-bit"
3353      * window into contiguous 32-bit and 64-bit windows
3354      */
3355 }
3356 
3357 static void spapr_machine_2_7_instance_options(MachineState *machine)
3358 {
3359     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3360 
3361     spapr_machine_2_8_instance_options(machine);
3362     spapr->use_hotplug_event_source = false;
3363 }
3364 
3365 static void spapr_machine_2_7_class_options(MachineClass *mc)
3366 {
3367     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3368 
3369     spapr_machine_2_8_class_options(mc);
3370     smc->tcg_default_cpu = "POWER7";
3371     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3372     smc->phb_placement = phb_placement_2_7;
3373 }
3374 
3375 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3376 
3377 /*
3378  * pseries-2.6
3379  */
3380 #define SPAPR_COMPAT_2_6 \
3381     HW_COMPAT_2_6 \
3382     { \
3383         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3384         .property = "ddw",\
3385         .value    = stringify(off),\
3386     },
3387 
3388 static void spapr_machine_2_6_instance_options(MachineState *machine)
3389 {
3390     spapr_machine_2_7_instance_options(machine);
3391 }
3392 
3393 static void spapr_machine_2_6_class_options(MachineClass *mc)
3394 {
3395     spapr_machine_2_7_class_options(mc);
3396     mc->has_hotpluggable_cpus = false;
3397     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3398 }
3399 
3400 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3401 
3402 /*
3403  * pseries-2.5
3404  */
3405 #define SPAPR_COMPAT_2_5 \
3406     HW_COMPAT_2_5 \
3407     { \
3408         .driver   = "spapr-vlan", \
3409         .property = "use-rx-buffer-pools", \
3410         .value    = "off", \
3411     },
3412 
3413 static void spapr_machine_2_5_instance_options(MachineState *machine)
3414 {
3415     spapr_machine_2_6_instance_options(machine);
3416 }
3417 
3418 static void spapr_machine_2_5_class_options(MachineClass *mc)
3419 {
3420     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3421 
3422     spapr_machine_2_6_class_options(mc);
3423     smc->use_ohci_by_default = true;
3424     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3425 }
3426 
3427 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3428 
3429 /*
3430  * pseries-2.4
3431  */
3432 #define SPAPR_COMPAT_2_4 \
3433         HW_COMPAT_2_4
3434 
3435 static void spapr_machine_2_4_instance_options(MachineState *machine)
3436 {
3437     spapr_machine_2_5_instance_options(machine);
3438 }
3439 
3440 static void spapr_machine_2_4_class_options(MachineClass *mc)
3441 {
3442     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3443 
3444     spapr_machine_2_5_class_options(mc);
3445     smc->dr_lmb_enabled = false;
3446     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3447 }
3448 
3449 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3450 
3451 /*
3452  * pseries-2.3
3453  */
3454 #define SPAPR_COMPAT_2_3 \
3455         HW_COMPAT_2_3 \
3456         {\
3457             .driver   = "spapr-pci-host-bridge",\
3458             .property = "dynamic-reconfiguration",\
3459             .value    = "off",\
3460         },
3461 
3462 static void spapr_machine_2_3_instance_options(MachineState *machine)
3463 {
3464     spapr_machine_2_4_instance_options(machine);
3465     savevm_skip_section_footers();
3466     global_state_set_optional();
3467     savevm_skip_configuration();
3468 }
3469 
3470 static void spapr_machine_2_3_class_options(MachineClass *mc)
3471 {
3472     spapr_machine_2_4_class_options(mc);
3473     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3474 }
3475 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3476 
3477 /*
3478  * pseries-2.2
3479  */
3480 
3481 #define SPAPR_COMPAT_2_2 \
3482         HW_COMPAT_2_2 \
3483         {\
3484             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3485             .property = "mem_win_size",\
3486             .value    = "0x20000000",\
3487         },
3488 
3489 static void spapr_machine_2_2_instance_options(MachineState *machine)
3490 {
3491     spapr_machine_2_3_instance_options(machine);
3492     machine->suppress_vmdesc = true;
3493 }
3494 
3495 static void spapr_machine_2_2_class_options(MachineClass *mc)
3496 {
3497     spapr_machine_2_3_class_options(mc);
3498     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3499 }
3500 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3501 
3502 /*
3503  * pseries-2.1
3504  */
3505 #define SPAPR_COMPAT_2_1 \
3506         HW_COMPAT_2_1
3507 
3508 static void spapr_machine_2_1_instance_options(MachineState *machine)
3509 {
3510     spapr_machine_2_2_instance_options(machine);
3511 }
3512 
3513 static void spapr_machine_2_1_class_options(MachineClass *mc)
3514 {
3515     spapr_machine_2_2_class_options(mc);
3516     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3517 }
3518 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3519 
3520 static void spapr_machine_register_types(void)
3521 {
3522     type_register_static(&spapr_machine_info);
3523 }
3524 
3525 type_init(spapr_machine_register_types)
3526