xref: /openbmc/qemu/hw/ppc/spapr.c (revision 795c40b8)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
31 #include "hw/hw.h"
32 #include "qemu/log.h"
33 #include "hw/fw-path-provider.h"
34 #include "elf.h"
35 #include "net/net.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/migration.h"
42 #include "mmu-hash64.h"
43 #include "mmu-book3s-v3.h"
44 #include "qom/cpu.h"
45 
46 #include "hw/boards.h"
47 #include "hw/ppc/ppc.h"
48 #include "hw/loader.h"
49 
50 #include "hw/ppc/fdt.h"
51 #include "hw/ppc/spapr.h"
52 #include "hw/ppc/spapr_vio.h"
53 #include "hw/pci-host/spapr.h"
54 #include "hw/ppc/xics.h"
55 #include "hw/pci/msi.h"
56 
57 #include "hw/pci/pci.h"
58 #include "hw/scsi/scsi.h"
59 #include "hw/virtio/virtio-scsi.h"
60 
61 #include "exec/address-spaces.h"
62 #include "hw/usb.h"
63 #include "qemu/config-file.h"
64 #include "qemu/error-report.h"
65 #include "trace.h"
66 #include "hw/nmi.h"
67 #include "hw/intc/intc.h"
68 
69 #include "hw/compat.h"
70 #include "qemu/cutils.h"
71 #include "hw/ppc/spapr_cpu_core.h"
72 #include "qmp-commands.h"
73 
74 #include <libfdt.h>
75 
76 /* SLOF memory layout:
77  *
78  * SLOF raw image loaded at 0, copies its romfs right below the flat
79  * device-tree, then position SLOF itself 31M below that
80  *
81  * So we set FW_OVERHEAD to 40MB which should account for all of that
82  * and more
83  *
84  * We load our kernel at 4M, leaving space for SLOF initial image
85  */
86 #define FDT_MAX_SIZE            0x100000
87 #define RTAS_MAX_SIZE           0x10000
88 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
89 #define FW_MAX_SIZE             0x400000
90 #define FW_FILE_NAME            "slof.bin"
91 #define FW_OVERHEAD             0x2800000
92 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
93 
94 #define MIN_RMA_SLOF            128UL
95 
96 #define PHANDLE_XICP            0x00001111
97 
98 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
99 
100 static ICSState *spapr_ics_create(sPAPRMachineState *spapr,
101                                   const char *type_ics,
102                                   int nr_irqs, Error **errp)
103 {
104     Error *err = NULL, *local_err = NULL;
105     Object *obj;
106 
107     obj = object_new(type_ics);
108     object_property_add_child(OBJECT(spapr), "ics", obj, NULL);
109     object_property_add_const_link(obj, "xics", OBJECT(spapr), &error_abort);
110     object_property_set_int(obj, nr_irqs, "nr-irqs", &err);
111     object_property_set_bool(obj, true, "realized", &local_err);
112     error_propagate(&err, local_err);
113     if (err) {
114         error_propagate(errp, err);
115         return NULL;
116     }
117 
118     return ICS_SIMPLE(obj);
119 }
120 
121 static void xics_system_init(MachineState *machine, int nr_irqs, Error **errp)
122 {
123     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
124 
125     if (kvm_enabled()) {
126         Error *err = NULL;
127 
128         if (machine_kernel_irqchip_allowed(machine) &&
129             !xics_kvm_init(spapr, errp)) {
130             spapr->icp_type = TYPE_KVM_ICP;
131             spapr->ics = spapr_ics_create(spapr, TYPE_ICS_KVM, nr_irqs, &err);
132         }
133         if (machine_kernel_irqchip_required(machine) && !spapr->ics) {
134             error_reportf_err(err,
135                               "kernel_irqchip requested but unavailable: ");
136         } else {
137             error_free(err);
138         }
139     }
140 
141     if (!spapr->ics) {
142         xics_spapr_init(spapr, errp);
143         spapr->icp_type = TYPE_ICP;
144         spapr->ics = spapr_ics_create(spapr, TYPE_ICS_SIMPLE, nr_irqs, errp);
145     }
146 }
147 
148 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
149                                   int smt_threads)
150 {
151     int i, ret = 0;
152     uint32_t servers_prop[smt_threads];
153     uint32_t gservers_prop[smt_threads * 2];
154     int index = ppc_get_vcpu_dt_id(cpu);
155 
156     if (cpu->compat_pvr) {
157         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
158         if (ret < 0) {
159             return ret;
160         }
161     }
162 
163     /* Build interrupt servers and gservers properties */
164     for (i = 0; i < smt_threads; i++) {
165         servers_prop[i] = cpu_to_be32(index + i);
166         /* Hack, direct the group queues back to cpu 0 */
167         gservers_prop[i*2] = cpu_to_be32(index + i);
168         gservers_prop[i*2 + 1] = 0;
169     }
170     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
171                       servers_prop, sizeof(servers_prop));
172     if (ret < 0) {
173         return ret;
174     }
175     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
176                       gservers_prop, sizeof(gservers_prop));
177 
178     return ret;
179 }
180 
181 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
182 {
183     int ret = 0;
184     PowerPCCPU *cpu = POWERPC_CPU(cs);
185     int index = ppc_get_vcpu_dt_id(cpu);
186     uint32_t associativity[] = {cpu_to_be32(0x5),
187                                 cpu_to_be32(0x0),
188                                 cpu_to_be32(0x0),
189                                 cpu_to_be32(0x0),
190                                 cpu_to_be32(cs->numa_node),
191                                 cpu_to_be32(index)};
192 
193     /* Advertise NUMA via ibm,associativity */
194     if (nb_numa_nodes > 1) {
195         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
196                           sizeof(associativity));
197     }
198 
199     return ret;
200 }
201 
202 /* Populate the "ibm,pa-features" property */
203 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset,
204                                       bool legacy_guest)
205 {
206     uint8_t pa_features_206[] = { 6, 0,
207         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
208     uint8_t pa_features_207[] = { 24, 0,
209         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
210         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
211         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
212         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
213     uint8_t pa_features_300[] = { 66, 0,
214         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
215         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
216         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
217         /* 6: DS207 */
218         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
219         /* 16: Vector */
220         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
221         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
222         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
223         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
224         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
225         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
226         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
227         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
228         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
229         /* 42: PM, 44: PC RA, 46: SC vec'd */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
231         /* 48: SIMD, 50: QP BFP, 52: String */
232         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
233         /* 54: DecFP, 56: DecI, 58: SHA */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
235         /* 60: NM atomic, 62: RNG */
236         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
237     };
238     uint8_t *pa_features;
239     size_t pa_size;
240 
241     switch (POWERPC_MMU_VER(env->mmu_model)) {
242     case POWERPC_MMU_VER_2_06:
243         pa_features = pa_features_206;
244         pa_size = sizeof(pa_features_206);
245         break;
246     case POWERPC_MMU_VER_2_07:
247         pa_features = pa_features_207;
248         pa_size = sizeof(pa_features_207);
249         break;
250     case POWERPC_MMU_VER_3_00:
251         pa_features = pa_features_300;
252         pa_size = sizeof(pa_features_300);
253         break;
254     default:
255         return;
256     }
257 
258     if (env->ci_large_pages) {
259         /*
260          * Note: we keep CI large pages off by default because a 64K capable
261          * guest provisioned with large pages might otherwise try to map a qemu
262          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
263          * even if that qemu runs on a 4k host.
264          * We dd this bit back here if we are confident this is not an issue
265          */
266         pa_features[3] |= 0x20;
267     }
268     if (kvmppc_has_cap_htm() && pa_size > 24) {
269         pa_features[24] |= 0x80;    /* Transactional memory support */
270     }
271     if (legacy_guest && pa_size > 40) {
272         /* Workaround for broken kernels that attempt (guest) radix
273          * mode when they can't handle it, if they see the radix bit set
274          * in pa-features. So hide it from them. */
275         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
276     }
277 
278     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
279 }
280 
281 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
282 {
283     int ret = 0, offset, cpus_offset;
284     CPUState *cs;
285     char cpu_model[32];
286     int smt = kvmppc_smt_threads();
287     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
288 
289     CPU_FOREACH(cs) {
290         PowerPCCPU *cpu = POWERPC_CPU(cs);
291         CPUPPCState *env = &cpu->env;
292         DeviceClass *dc = DEVICE_GET_CLASS(cs);
293         int index = ppc_get_vcpu_dt_id(cpu);
294         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
295 
296         if ((index % smt) != 0) {
297             continue;
298         }
299 
300         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
301 
302         cpus_offset = fdt_path_offset(fdt, "/cpus");
303         if (cpus_offset < 0) {
304             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
305                                           "cpus");
306             if (cpus_offset < 0) {
307                 return cpus_offset;
308             }
309         }
310         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
311         if (offset < 0) {
312             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
313             if (offset < 0) {
314                 return offset;
315             }
316         }
317 
318         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
319                           pft_size_prop, sizeof(pft_size_prop));
320         if (ret < 0) {
321             return ret;
322         }
323 
324         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
325         if (ret < 0) {
326             return ret;
327         }
328 
329         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
330         if (ret < 0) {
331             return ret;
332         }
333 
334         spapr_populate_pa_features(env, fdt, offset,
335                                          spapr->cas_legacy_guest_workaround);
336     }
337     return ret;
338 }
339 
340 static hwaddr spapr_node0_size(void)
341 {
342     MachineState *machine = MACHINE(qdev_get_machine());
343 
344     if (nb_numa_nodes) {
345         int i;
346         for (i = 0; i < nb_numa_nodes; ++i) {
347             if (numa_info[i].node_mem) {
348                 return MIN(pow2floor(numa_info[i].node_mem),
349                            machine->ram_size);
350             }
351         }
352     }
353     return machine->ram_size;
354 }
355 
356 static void add_str(GString *s, const gchar *s1)
357 {
358     g_string_append_len(s, s1, strlen(s1) + 1);
359 }
360 
361 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
362                                        hwaddr size)
363 {
364     uint32_t associativity[] = {
365         cpu_to_be32(0x4), /* length */
366         cpu_to_be32(0x0), cpu_to_be32(0x0),
367         cpu_to_be32(0x0), cpu_to_be32(nodeid)
368     };
369     char mem_name[32];
370     uint64_t mem_reg_property[2];
371     int off;
372 
373     mem_reg_property[0] = cpu_to_be64(start);
374     mem_reg_property[1] = cpu_to_be64(size);
375 
376     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
377     off = fdt_add_subnode(fdt, 0, mem_name);
378     _FDT(off);
379     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
380     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
381                       sizeof(mem_reg_property))));
382     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
383                       sizeof(associativity))));
384     return off;
385 }
386 
387 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
388 {
389     MachineState *machine = MACHINE(spapr);
390     hwaddr mem_start, node_size;
391     int i, nb_nodes = nb_numa_nodes;
392     NodeInfo *nodes = numa_info;
393     NodeInfo ramnode;
394 
395     /* No NUMA nodes, assume there is just one node with whole RAM */
396     if (!nb_numa_nodes) {
397         nb_nodes = 1;
398         ramnode.node_mem = machine->ram_size;
399         nodes = &ramnode;
400     }
401 
402     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
403         if (!nodes[i].node_mem) {
404             continue;
405         }
406         if (mem_start >= machine->ram_size) {
407             node_size = 0;
408         } else {
409             node_size = nodes[i].node_mem;
410             if (node_size > machine->ram_size - mem_start) {
411                 node_size = machine->ram_size - mem_start;
412             }
413         }
414         if (!mem_start) {
415             /* ppc_spapr_init() checks for rma_size <= node0_size already */
416             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
417             mem_start += spapr->rma_size;
418             node_size -= spapr->rma_size;
419         }
420         for ( ; node_size; ) {
421             hwaddr sizetmp = pow2floor(node_size);
422 
423             /* mem_start != 0 here */
424             if (ctzl(mem_start) < ctzl(sizetmp)) {
425                 sizetmp = 1ULL << ctzl(mem_start);
426             }
427 
428             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
429             node_size -= sizetmp;
430             mem_start += sizetmp;
431         }
432     }
433 
434     return 0;
435 }
436 
437 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
438                                   sPAPRMachineState *spapr)
439 {
440     PowerPCCPU *cpu = POWERPC_CPU(cs);
441     CPUPPCState *env = &cpu->env;
442     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
443     int index = ppc_get_vcpu_dt_id(cpu);
444     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
445                        0xffffffff, 0xffffffff};
446     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
447         : SPAPR_TIMEBASE_FREQ;
448     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
449     uint32_t page_sizes_prop[64];
450     size_t page_sizes_prop_size;
451     uint32_t vcpus_per_socket = smp_threads * smp_cores;
452     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
453     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
454     sPAPRDRConnector *drc;
455     sPAPRDRConnectorClass *drck;
456     int drc_index;
457     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
458     int i;
459 
460     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
461     if (drc) {
462         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
463         drc_index = drck->get_index(drc);
464         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
465     }
466 
467     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
468     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
469 
470     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
471     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
472                            env->dcache_line_size)));
473     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
474                            env->dcache_line_size)));
475     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
476                            env->icache_line_size)));
477     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
478                            env->icache_line_size)));
479 
480     if (pcc->l1_dcache_size) {
481         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
482                                pcc->l1_dcache_size)));
483     } else {
484         error_report("Warning: Unknown L1 dcache size for cpu");
485     }
486     if (pcc->l1_icache_size) {
487         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
488                                pcc->l1_icache_size)));
489     } else {
490         error_report("Warning: Unknown L1 icache size for cpu");
491     }
492 
493     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
494     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
495     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
496     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
497     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
498     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
499 
500     if (env->spr_cb[SPR_PURR].oea_read) {
501         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
502     }
503 
504     if (env->mmu_model & POWERPC_MMU_1TSEG) {
505         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
506                           segs, sizeof(segs))));
507     }
508 
509     /* Advertise VMX/VSX (vector extensions) if available
510      *   0 / no property == no vector extensions
511      *   1               == VMX / Altivec available
512      *   2               == VSX available */
513     if (env->insns_flags & PPC_ALTIVEC) {
514         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
515 
516         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
517     }
518 
519     /* Advertise DFP (Decimal Floating Point) if available
520      *   0 / no property == no DFP
521      *   1               == DFP available */
522     if (env->insns_flags2 & PPC2_DFP) {
523         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
524     }
525 
526     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
527                                                   sizeof(page_sizes_prop));
528     if (page_sizes_prop_size) {
529         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
530                           page_sizes_prop, page_sizes_prop_size)));
531     }
532 
533     spapr_populate_pa_features(env, fdt, offset, false);
534 
535     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
536                            cs->cpu_index / vcpus_per_socket)));
537 
538     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
539                       pft_size_prop, sizeof(pft_size_prop))));
540 
541     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
542 
543     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
544 
545     if (pcc->radix_page_info) {
546         for (i = 0; i < pcc->radix_page_info->count; i++) {
547             radix_AP_encodings[i] =
548                 cpu_to_be32(pcc->radix_page_info->entries[i]);
549         }
550         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
551                           radix_AP_encodings,
552                           pcc->radix_page_info->count *
553                           sizeof(radix_AP_encodings[0]))));
554     }
555 }
556 
557 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
558 {
559     CPUState *cs;
560     int cpus_offset;
561     char *nodename;
562     int smt = kvmppc_smt_threads();
563 
564     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
565     _FDT(cpus_offset);
566     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
567     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
568 
569     /*
570      * We walk the CPUs in reverse order to ensure that CPU DT nodes
571      * created by fdt_add_subnode() end up in the right order in FDT
572      * for the guest kernel the enumerate the CPUs correctly.
573      */
574     CPU_FOREACH_REVERSE(cs) {
575         PowerPCCPU *cpu = POWERPC_CPU(cs);
576         int index = ppc_get_vcpu_dt_id(cpu);
577         DeviceClass *dc = DEVICE_GET_CLASS(cs);
578         int offset;
579 
580         if ((index % smt) != 0) {
581             continue;
582         }
583 
584         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
585         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
586         g_free(nodename);
587         _FDT(offset);
588         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
589     }
590 
591 }
592 
593 /*
594  * Adds ibm,dynamic-reconfiguration-memory node.
595  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
596  * of this device tree node.
597  */
598 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
599 {
600     MachineState *machine = MACHINE(spapr);
601     int ret, i, offset;
602     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
603     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
604     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
605     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
606                        memory_region_size(&spapr->hotplug_memory.mr)) /
607                        lmb_size;
608     uint32_t *int_buf, *cur_index, buf_len;
609     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
610 
611     /*
612      * Don't create the node if there is no hotpluggable memory
613      */
614     if (machine->ram_size == machine->maxram_size) {
615         return 0;
616     }
617 
618     /*
619      * Allocate enough buffer size to fit in ibm,dynamic-memory
620      * or ibm,associativity-lookup-arrays
621      */
622     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
623               * sizeof(uint32_t);
624     cur_index = int_buf = g_malloc0(buf_len);
625 
626     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
627 
628     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
629                     sizeof(prop_lmb_size));
630     if (ret < 0) {
631         goto out;
632     }
633 
634     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
635     if (ret < 0) {
636         goto out;
637     }
638 
639     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
640     if (ret < 0) {
641         goto out;
642     }
643 
644     /* ibm,dynamic-memory */
645     int_buf[0] = cpu_to_be32(nr_lmbs);
646     cur_index++;
647     for (i = 0; i < nr_lmbs; i++) {
648         uint64_t addr = i * lmb_size;
649         uint32_t *dynamic_memory = cur_index;
650 
651         if (i >= hotplug_lmb_start) {
652             sPAPRDRConnector *drc;
653             sPAPRDRConnectorClass *drck;
654 
655             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
656             g_assert(drc);
657             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
658 
659             dynamic_memory[0] = cpu_to_be32(addr >> 32);
660             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
661             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
662             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
663             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
664             if (memory_region_present(get_system_memory(), addr)) {
665                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
666             } else {
667                 dynamic_memory[5] = cpu_to_be32(0);
668             }
669         } else {
670             /*
671              * LMB information for RMA, boot time RAM and gap b/n RAM and
672              * hotplug memory region -- all these are marked as reserved
673              * and as having no valid DRC.
674              */
675             dynamic_memory[0] = cpu_to_be32(addr >> 32);
676             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
677             dynamic_memory[2] = cpu_to_be32(0);
678             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
679             dynamic_memory[4] = cpu_to_be32(-1);
680             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
681                                             SPAPR_LMB_FLAGS_DRC_INVALID);
682         }
683 
684         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
685     }
686     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
687     if (ret < 0) {
688         goto out;
689     }
690 
691     /* ibm,associativity-lookup-arrays */
692     cur_index = int_buf;
693     int_buf[0] = cpu_to_be32(nr_nodes);
694     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
695     cur_index += 2;
696     for (i = 0; i < nr_nodes; i++) {
697         uint32_t associativity[] = {
698             cpu_to_be32(0x0),
699             cpu_to_be32(0x0),
700             cpu_to_be32(0x0),
701             cpu_to_be32(i)
702         };
703         memcpy(cur_index, associativity, sizeof(associativity));
704         cur_index += 4;
705     }
706     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
707             (cur_index - int_buf) * sizeof(uint32_t));
708 out:
709     g_free(int_buf);
710     return ret;
711 }
712 
713 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
714                                 sPAPROptionVector *ov5_updates)
715 {
716     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
717     int ret = 0, offset;
718 
719     /* Generate ibm,dynamic-reconfiguration-memory node if required */
720     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
721         g_assert(smc->dr_lmb_enabled);
722         ret = spapr_populate_drconf_memory(spapr, fdt);
723         if (ret) {
724             goto out;
725         }
726     }
727 
728     offset = fdt_path_offset(fdt, "/chosen");
729     if (offset < 0) {
730         offset = fdt_add_subnode(fdt, 0, "chosen");
731         if (offset < 0) {
732             return offset;
733         }
734     }
735     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
736                                  "ibm,architecture-vec-5");
737 
738 out:
739     return ret;
740 }
741 
742 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
743                                  target_ulong addr, target_ulong size,
744                                  sPAPROptionVector *ov5_updates)
745 {
746     void *fdt, *fdt_skel;
747     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
748 
749     size -= sizeof(hdr);
750 
751     /* Create sceleton */
752     fdt_skel = g_malloc0(size);
753     _FDT((fdt_create(fdt_skel, size)));
754     _FDT((fdt_begin_node(fdt_skel, "")));
755     _FDT((fdt_end_node(fdt_skel)));
756     _FDT((fdt_finish(fdt_skel)));
757     fdt = g_malloc0(size);
758     _FDT((fdt_open_into(fdt_skel, fdt, size)));
759     g_free(fdt_skel);
760 
761     /* Fixup cpu nodes */
762     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
763 
764     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
765         return -1;
766     }
767 
768     /* Pack resulting tree */
769     _FDT((fdt_pack(fdt)));
770 
771     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
772         trace_spapr_cas_failed(size);
773         return -1;
774     }
775 
776     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
777     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
778     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
779     g_free(fdt);
780 
781     return 0;
782 }
783 
784 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
785 {
786     int rtas;
787     GString *hypertas = g_string_sized_new(256);
788     GString *qemu_hypertas = g_string_sized_new(256);
789     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
790     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
791         memory_region_size(&spapr->hotplug_memory.mr);
792     uint32_t lrdr_capacity[] = {
793         cpu_to_be32(max_hotplug_addr >> 32),
794         cpu_to_be32(max_hotplug_addr & 0xffffffff),
795         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
796         cpu_to_be32(max_cpus / smp_threads),
797     };
798 
799     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
800 
801     /* hypertas */
802     add_str(hypertas, "hcall-pft");
803     add_str(hypertas, "hcall-term");
804     add_str(hypertas, "hcall-dabr");
805     add_str(hypertas, "hcall-interrupt");
806     add_str(hypertas, "hcall-tce");
807     add_str(hypertas, "hcall-vio");
808     add_str(hypertas, "hcall-splpar");
809     add_str(hypertas, "hcall-bulk");
810     add_str(hypertas, "hcall-set-mode");
811     add_str(hypertas, "hcall-sprg0");
812     add_str(hypertas, "hcall-copy");
813     add_str(hypertas, "hcall-debug");
814     add_str(qemu_hypertas, "hcall-memop1");
815 
816     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
817         add_str(hypertas, "hcall-multi-tce");
818     }
819     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
820                      hypertas->str, hypertas->len));
821     g_string_free(hypertas, TRUE);
822     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
823                      qemu_hypertas->str, qemu_hypertas->len));
824     g_string_free(qemu_hypertas, TRUE);
825 
826     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
827                      refpoints, sizeof(refpoints)));
828 
829     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
830                           RTAS_ERROR_LOG_MAX));
831     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
832                           RTAS_EVENT_SCAN_RATE));
833 
834     if (msi_nonbroken) {
835         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
836     }
837 
838     /*
839      * According to PAPR, rtas ibm,os-term does not guarantee a return
840      * back to the guest cpu.
841      *
842      * While an additional ibm,extended-os-term property indicates
843      * that rtas call return will always occur. Set this property.
844      */
845     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
846 
847     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
848                      lrdr_capacity, sizeof(lrdr_capacity)));
849 
850     spapr_dt_rtas_tokens(fdt, rtas);
851 }
852 
853 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
854  * that the guest may request and thus the valid values for bytes 24..26 of
855  * option vector 5: */
856 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
857 {
858     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
859 
860     char val[2 * 3] = {
861         24, 0x00, /* Hash/Radix, filled in below. */
862         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
863         26, 0x40, /* Radix options: GTSE == yes. */
864     };
865 
866     if (kvm_enabled()) {
867         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
868             val[1] = 0x80; /* OV5_MMU_BOTH */
869         } else if (kvmppc_has_cap_mmu_radix()) {
870             val[1] = 0x40; /* OV5_MMU_RADIX_300 */
871         } else {
872             val[1] = 0x00; /* Hash */
873         }
874     } else {
875         if (first_ppc_cpu->env.mmu_model & POWERPC_MMU_V3) {
876             /* V3 MMU supports both hash and radix (with dynamic switching) */
877             val[1] = 0xC0;
878         } else {
879             /* Otherwise we can only do hash */
880             val[1] = 0x00;
881         }
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     int smt = kvmppc_smt_threads();
981 
982     fdt = g_malloc0(FDT_MAX_SIZE);
983     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
984 
985     /* Root node */
986     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
987     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
988     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
989 
990     /*
991      * Add info to guest to indentify which host is it being run on
992      * and what is the uuid of the guest
993      */
994     if (kvmppc_get_host_model(&buf)) {
995         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
996         g_free(buf);
997     }
998     if (kvmppc_get_host_serial(&buf)) {
999         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1000         g_free(buf);
1001     }
1002 
1003     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1004 
1005     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1006     if (qemu_uuid_set) {
1007         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1008     }
1009     g_free(buf);
1010 
1011     if (qemu_get_vm_name()) {
1012         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1013                                 qemu_get_vm_name()));
1014     }
1015 
1016     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1017     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1018 
1019     /* /interrupt controller */
1020     spapr_dt_xics(DIV_ROUND_UP(max_cpus * smt, smp_threads), fdt, PHANDLE_XICP);
1021 
1022     ret = spapr_populate_memory(spapr, fdt);
1023     if (ret < 0) {
1024         error_report("couldn't setup memory nodes in fdt");
1025         exit(1);
1026     }
1027 
1028     /* /vdevice */
1029     spapr_dt_vdevice(spapr->vio_bus, fdt);
1030 
1031     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1032         ret = spapr_rng_populate_dt(fdt);
1033         if (ret < 0) {
1034             error_report("could not set up rng device in the fdt");
1035             exit(1);
1036         }
1037     }
1038 
1039     QLIST_FOREACH(phb, &spapr->phbs, list) {
1040         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1041         if (ret < 0) {
1042             error_report("couldn't setup PCI devices in fdt");
1043             exit(1);
1044         }
1045     }
1046 
1047     /* cpus */
1048     spapr_populate_cpus_dt_node(fdt, spapr);
1049 
1050     if (smc->dr_lmb_enabled) {
1051         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1052     }
1053 
1054     if (mc->has_hotpluggable_cpus) {
1055         int offset = fdt_path_offset(fdt, "/cpus");
1056         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1057                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1058         if (ret < 0) {
1059             error_report("Couldn't set up CPU DR device tree properties");
1060             exit(1);
1061         }
1062     }
1063 
1064     /* /event-sources */
1065     spapr_dt_events(spapr, fdt);
1066 
1067     /* /rtas */
1068     spapr_dt_rtas(spapr, fdt);
1069 
1070     /* /chosen */
1071     spapr_dt_chosen(spapr, fdt);
1072 
1073     /* /hypervisor */
1074     if (kvm_enabled()) {
1075         spapr_dt_hypervisor(spapr, fdt);
1076     }
1077 
1078     /* Build memory reserve map */
1079     if (spapr->kernel_size) {
1080         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1081     }
1082     if (spapr->initrd_size) {
1083         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1084     }
1085 
1086     /* ibm,client-architecture-support updates */
1087     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1088     if (ret < 0) {
1089         error_report("couldn't setup CAS properties fdt");
1090         exit(1);
1091     }
1092 
1093     return fdt;
1094 }
1095 
1096 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1097 {
1098     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1099 }
1100 
1101 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1102                                     PowerPCCPU *cpu)
1103 {
1104     CPUPPCState *env = &cpu->env;
1105 
1106     /* The TCG path should also be holding the BQL at this point */
1107     g_assert(qemu_mutex_iothread_locked());
1108 
1109     if (msr_pr) {
1110         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1111         env->gpr[3] = H_PRIVILEGE;
1112     } else {
1113         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1114     }
1115 }
1116 
1117 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1118 {
1119     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1120 
1121     return spapr->patb_entry;
1122 }
1123 
1124 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1125 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1126 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1127 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1128 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1129 
1130 /*
1131  * Get the fd to access the kernel htab, re-opening it if necessary
1132  */
1133 static int get_htab_fd(sPAPRMachineState *spapr)
1134 {
1135     if (spapr->htab_fd >= 0) {
1136         return spapr->htab_fd;
1137     }
1138 
1139     spapr->htab_fd = kvmppc_get_htab_fd(false);
1140     if (spapr->htab_fd < 0) {
1141         error_report("Unable to open fd for reading hash table from KVM: %s",
1142                      strerror(errno));
1143     }
1144 
1145     return spapr->htab_fd;
1146 }
1147 
1148 void close_htab_fd(sPAPRMachineState *spapr)
1149 {
1150     if (spapr->htab_fd >= 0) {
1151         close(spapr->htab_fd);
1152     }
1153     spapr->htab_fd = -1;
1154 }
1155 
1156 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1157 {
1158     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1159 
1160     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1161 }
1162 
1163 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1164                                                 hwaddr ptex, int n)
1165 {
1166     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1167     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1168 
1169     if (!spapr->htab) {
1170         /*
1171          * HTAB is controlled by KVM. Fetch into temporary buffer
1172          */
1173         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1174         kvmppc_read_hptes(hptes, ptex, n);
1175         return hptes;
1176     }
1177 
1178     /*
1179      * HTAB is controlled by QEMU. Just point to the internally
1180      * accessible PTEG.
1181      */
1182     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1183 }
1184 
1185 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1186                               const ppc_hash_pte64_t *hptes,
1187                               hwaddr ptex, int n)
1188 {
1189     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1190 
1191     if (!spapr->htab) {
1192         g_free((void *)hptes);
1193     }
1194 
1195     /* Nothing to do for qemu managed HPT */
1196 }
1197 
1198 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1199                              uint64_t pte0, uint64_t pte1)
1200 {
1201     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1202     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1203 
1204     if (!spapr->htab) {
1205         kvmppc_write_hpte(ptex, pte0, pte1);
1206     } else {
1207         stq_p(spapr->htab + offset, pte0);
1208         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1209     }
1210 }
1211 
1212 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1213 {
1214     int shift;
1215 
1216     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1217      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1218      * that's much more than is needed for Linux guests */
1219     shift = ctz64(pow2ceil(ramsize)) - 7;
1220     shift = MAX(shift, 18); /* Minimum architected size */
1221     shift = MIN(shift, 46); /* Maximum architected size */
1222     return shift;
1223 }
1224 
1225 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1226                                  Error **errp)
1227 {
1228     long rc;
1229 
1230     /* Clean up any HPT info from a previous boot */
1231     g_free(spapr->htab);
1232     spapr->htab = NULL;
1233     spapr->htab_shift = 0;
1234     close_htab_fd(spapr);
1235 
1236     rc = kvmppc_reset_htab(shift);
1237     if (rc < 0) {
1238         /* kernel-side HPT needed, but couldn't allocate one */
1239         error_setg_errno(errp, errno,
1240                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1241                          shift);
1242         /* This is almost certainly fatal, but if the caller really
1243          * wants to carry on with shift == 0, it's welcome to try */
1244     } else if (rc > 0) {
1245         /* kernel-side HPT allocated */
1246         if (rc != shift) {
1247             error_setg(errp,
1248                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1249                        shift, rc);
1250         }
1251 
1252         spapr->htab_shift = shift;
1253         spapr->htab = NULL;
1254     } else {
1255         /* kernel-side HPT not needed, allocate in userspace instead */
1256         size_t size = 1ULL << shift;
1257         int i;
1258 
1259         spapr->htab = qemu_memalign(size, size);
1260         if (!spapr->htab) {
1261             error_setg_errno(errp, errno,
1262                              "Could not allocate HPT of order %d", shift);
1263             return;
1264         }
1265 
1266         memset(spapr->htab, 0, size);
1267         spapr->htab_shift = shift;
1268 
1269         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1270             DIRTY_HPTE(HPTE(spapr->htab, i));
1271         }
1272     }
1273 }
1274 
1275 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1276 {
1277     spapr_reallocate_hpt(spapr,
1278                      spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size),
1279                      &error_fatal);
1280     if (spapr->vrma_adjust) {
1281         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1282                                           spapr->htab_shift);
1283     }
1284     /* We're setting up a hash table, so that means we're not radix */
1285     spapr->patb_entry = 0;
1286 }
1287 
1288 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1289 {
1290     bool matched = false;
1291 
1292     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1293         matched = true;
1294     }
1295 
1296     if (!matched) {
1297         error_report("Device %s is not supported by this machine yet.",
1298                      qdev_fw_name(DEVICE(sbdev)));
1299         exit(1);
1300     }
1301 }
1302 
1303 static void ppc_spapr_reset(void)
1304 {
1305     MachineState *machine = MACHINE(qdev_get_machine());
1306     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1307     PowerPCCPU *first_ppc_cpu;
1308     uint32_t rtas_limit;
1309     hwaddr rtas_addr, fdt_addr;
1310     void *fdt;
1311     int rc;
1312 
1313     /* Check for unknown sysbus devices */
1314     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1315 
1316     if (kvm_enabled() && kvmppc_has_cap_mmu_radix()) {
1317         /* If using KVM with radix mode available, VCPUs can be started
1318          * without a HPT because KVM will start them in radix mode.
1319          * Set the GR bit in PATB so that we know there is no HPT. */
1320         spapr->patb_entry = PATBE1_GR;
1321     } else {
1322         spapr->patb_entry = 0;
1323         spapr_setup_hpt_and_vrma(spapr);
1324     }
1325 
1326     qemu_devices_reset();
1327 
1328     /*
1329      * We place the device tree and RTAS just below either the top of the RMA,
1330      * or just below 2GB, whichever is lowere, so that it can be
1331      * processed with 32-bit real mode code if necessary
1332      */
1333     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1334     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1335     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1336 
1337     /* if this reset wasn't generated by CAS, we should reset our
1338      * negotiated options and start from scratch */
1339     if (!spapr->cas_reboot) {
1340         spapr_ovec_cleanup(spapr->ov5_cas);
1341         spapr->ov5_cas = spapr_ovec_new();
1342     }
1343 
1344     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1345 
1346     spapr_load_rtas(spapr, fdt, rtas_addr);
1347 
1348     rc = fdt_pack(fdt);
1349 
1350     /* Should only fail if we've built a corrupted tree */
1351     assert(rc == 0);
1352 
1353     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1354         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1355                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1356         exit(1);
1357     }
1358 
1359     /* Load the fdt */
1360     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1361     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1362     g_free(fdt);
1363 
1364     /* Set up the entry state */
1365     first_ppc_cpu = POWERPC_CPU(first_cpu);
1366     first_ppc_cpu->env.gpr[3] = fdt_addr;
1367     first_ppc_cpu->env.gpr[5] = 0;
1368     first_cpu->halted = 0;
1369     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1370 
1371     spapr->cas_reboot = false;
1372 }
1373 
1374 static void spapr_create_nvram(sPAPRMachineState *spapr)
1375 {
1376     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1377     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1378 
1379     if (dinfo) {
1380         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1381                             &error_fatal);
1382     }
1383 
1384     qdev_init_nofail(dev);
1385 
1386     spapr->nvram = (struct sPAPRNVRAM *)dev;
1387 }
1388 
1389 static void spapr_rtc_create(sPAPRMachineState *spapr)
1390 {
1391     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1392     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1393                               &error_fatal);
1394     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1395                               &error_fatal);
1396     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1397                               "date", &error_fatal);
1398 }
1399 
1400 /* Returns whether we want to use VGA or not */
1401 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1402 {
1403     switch (vga_interface_type) {
1404     case VGA_NONE:
1405         return false;
1406     case VGA_DEVICE:
1407         return true;
1408     case VGA_STD:
1409     case VGA_VIRTIO:
1410         return pci_vga_init(pci_bus) != NULL;
1411     default:
1412         error_setg(errp,
1413                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1414         return false;
1415     }
1416 }
1417 
1418 static int spapr_post_load(void *opaque, int version_id)
1419 {
1420     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1421     int err = 0;
1422 
1423     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1424         CPUState *cs;
1425         CPU_FOREACH(cs) {
1426             PowerPCCPU *cpu = POWERPC_CPU(cs);
1427             icp_resend(ICP(cpu->intc));
1428         }
1429     }
1430 
1431     /* In earlier versions, there was no separate qdev for the PAPR
1432      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1433      * So when migrating from those versions, poke the incoming offset
1434      * value into the RTC device */
1435     if (version_id < 3) {
1436         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1437     }
1438 
1439     return err;
1440 }
1441 
1442 static bool version_before_3(void *opaque, int version_id)
1443 {
1444     return version_id < 3;
1445 }
1446 
1447 static bool spapr_ov5_cas_needed(void *opaque)
1448 {
1449     sPAPRMachineState *spapr = opaque;
1450     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1451     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1452     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1453     bool cas_needed;
1454 
1455     /* Prior to the introduction of sPAPROptionVector, we had two option
1456      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1457      * Both of these options encode machine topology into the device-tree
1458      * in such a way that the now-booted OS should still be able to interact
1459      * appropriately with QEMU regardless of what options were actually
1460      * negotiatied on the source side.
1461      *
1462      * As such, we can avoid migrating the CAS-negotiated options if these
1463      * are the only options available on the current machine/platform.
1464      * Since these are the only options available for pseries-2.7 and
1465      * earlier, this allows us to maintain old->new/new->old migration
1466      * compatibility.
1467      *
1468      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1469      * via default pseries-2.8 machines and explicit command-line parameters.
1470      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1471      * of the actual CAS-negotiated values to continue working properly. For
1472      * example, availability of memory unplug depends on knowing whether
1473      * OV5_HP_EVT was negotiated via CAS.
1474      *
1475      * Thus, for any cases where the set of available CAS-negotiatable
1476      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1477      * include the CAS-negotiated options in the migration stream.
1478      */
1479     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1480     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1481 
1482     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1483      * the mask itself since in the future it's possible "legacy" bits may be
1484      * removed via machine options, which could generate a false positive
1485      * that breaks migration.
1486      */
1487     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1488     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1489 
1490     spapr_ovec_cleanup(ov5_mask);
1491     spapr_ovec_cleanup(ov5_legacy);
1492     spapr_ovec_cleanup(ov5_removed);
1493 
1494     return cas_needed;
1495 }
1496 
1497 static const VMStateDescription vmstate_spapr_ov5_cas = {
1498     .name = "spapr_option_vector_ov5_cas",
1499     .version_id = 1,
1500     .minimum_version_id = 1,
1501     .needed = spapr_ov5_cas_needed,
1502     .fields = (VMStateField[]) {
1503         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1504                                  vmstate_spapr_ovec, sPAPROptionVector),
1505         VMSTATE_END_OF_LIST()
1506     },
1507 };
1508 
1509 static bool spapr_patb_entry_needed(void *opaque)
1510 {
1511     sPAPRMachineState *spapr = opaque;
1512 
1513     return !!spapr->patb_entry;
1514 }
1515 
1516 static const VMStateDescription vmstate_spapr_patb_entry = {
1517     .name = "spapr_patb_entry",
1518     .version_id = 1,
1519     .minimum_version_id = 1,
1520     .needed = spapr_patb_entry_needed,
1521     .fields = (VMStateField[]) {
1522         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1523         VMSTATE_END_OF_LIST()
1524     },
1525 };
1526 
1527 static const VMStateDescription vmstate_spapr = {
1528     .name = "spapr",
1529     .version_id = 3,
1530     .minimum_version_id = 1,
1531     .post_load = spapr_post_load,
1532     .fields = (VMStateField[]) {
1533         /* used to be @next_irq */
1534         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1535 
1536         /* RTC offset */
1537         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1538 
1539         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1540         VMSTATE_END_OF_LIST()
1541     },
1542     .subsections = (const VMStateDescription*[]) {
1543         &vmstate_spapr_ov5_cas,
1544         &vmstate_spapr_patb_entry,
1545         NULL
1546     }
1547 };
1548 
1549 static int htab_save_setup(QEMUFile *f, void *opaque)
1550 {
1551     sPAPRMachineState *spapr = opaque;
1552 
1553     /* "Iteration" header */
1554     qemu_put_be32(f, spapr->htab_shift);
1555 
1556     if (spapr->htab) {
1557         spapr->htab_save_index = 0;
1558         spapr->htab_first_pass = true;
1559     } else {
1560         assert(kvm_enabled());
1561     }
1562 
1563 
1564     return 0;
1565 }
1566 
1567 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1568                                  int64_t max_ns)
1569 {
1570     bool has_timeout = max_ns != -1;
1571     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1572     int index = spapr->htab_save_index;
1573     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1574 
1575     assert(spapr->htab_first_pass);
1576 
1577     do {
1578         int chunkstart;
1579 
1580         /* Consume invalid HPTEs */
1581         while ((index < htabslots)
1582                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1583             CLEAN_HPTE(HPTE(spapr->htab, index));
1584             index++;
1585         }
1586 
1587         /* Consume valid HPTEs */
1588         chunkstart = index;
1589         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1590                && HPTE_VALID(HPTE(spapr->htab, index))) {
1591             CLEAN_HPTE(HPTE(spapr->htab, index));
1592             index++;
1593         }
1594 
1595         if (index > chunkstart) {
1596             int n_valid = index - chunkstart;
1597 
1598             qemu_put_be32(f, chunkstart);
1599             qemu_put_be16(f, n_valid);
1600             qemu_put_be16(f, 0);
1601             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1602                             HASH_PTE_SIZE_64 * n_valid);
1603 
1604             if (has_timeout &&
1605                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1606                 break;
1607             }
1608         }
1609     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1610 
1611     if (index >= htabslots) {
1612         assert(index == htabslots);
1613         index = 0;
1614         spapr->htab_first_pass = false;
1615     }
1616     spapr->htab_save_index = index;
1617 }
1618 
1619 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1620                                 int64_t max_ns)
1621 {
1622     bool final = max_ns < 0;
1623     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1624     int examined = 0, sent = 0;
1625     int index = spapr->htab_save_index;
1626     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1627 
1628     assert(!spapr->htab_first_pass);
1629 
1630     do {
1631         int chunkstart, invalidstart;
1632 
1633         /* Consume non-dirty HPTEs */
1634         while ((index < htabslots)
1635                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1636             index++;
1637             examined++;
1638         }
1639 
1640         chunkstart = index;
1641         /* Consume valid dirty HPTEs */
1642         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1643                && HPTE_DIRTY(HPTE(spapr->htab, index))
1644                && HPTE_VALID(HPTE(spapr->htab, index))) {
1645             CLEAN_HPTE(HPTE(spapr->htab, index));
1646             index++;
1647             examined++;
1648         }
1649 
1650         invalidstart = index;
1651         /* Consume invalid dirty HPTEs */
1652         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1653                && HPTE_DIRTY(HPTE(spapr->htab, index))
1654                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1655             CLEAN_HPTE(HPTE(spapr->htab, index));
1656             index++;
1657             examined++;
1658         }
1659 
1660         if (index > chunkstart) {
1661             int n_valid = invalidstart - chunkstart;
1662             int n_invalid = index - invalidstart;
1663 
1664             qemu_put_be32(f, chunkstart);
1665             qemu_put_be16(f, n_valid);
1666             qemu_put_be16(f, n_invalid);
1667             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1668                             HASH_PTE_SIZE_64 * n_valid);
1669             sent += index - chunkstart;
1670 
1671             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1672                 break;
1673             }
1674         }
1675 
1676         if (examined >= htabslots) {
1677             break;
1678         }
1679 
1680         if (index >= htabslots) {
1681             assert(index == htabslots);
1682             index = 0;
1683         }
1684     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1685 
1686     if (index >= htabslots) {
1687         assert(index == htabslots);
1688         index = 0;
1689     }
1690 
1691     spapr->htab_save_index = index;
1692 
1693     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1694 }
1695 
1696 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1697 #define MAX_KVM_BUF_SIZE    2048
1698 
1699 static int htab_save_iterate(QEMUFile *f, void *opaque)
1700 {
1701     sPAPRMachineState *spapr = opaque;
1702     int fd;
1703     int rc = 0;
1704 
1705     /* Iteration header */
1706     qemu_put_be32(f, 0);
1707 
1708     if (!spapr->htab) {
1709         assert(kvm_enabled());
1710 
1711         fd = get_htab_fd(spapr);
1712         if (fd < 0) {
1713             return fd;
1714         }
1715 
1716         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1717         if (rc < 0) {
1718             return rc;
1719         }
1720     } else  if (spapr->htab_first_pass) {
1721         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1722     } else {
1723         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1724     }
1725 
1726     /* End marker */
1727     qemu_put_be32(f, 0);
1728     qemu_put_be16(f, 0);
1729     qemu_put_be16(f, 0);
1730 
1731     return rc;
1732 }
1733 
1734 static int htab_save_complete(QEMUFile *f, void *opaque)
1735 {
1736     sPAPRMachineState *spapr = opaque;
1737     int fd;
1738 
1739     /* Iteration header */
1740     qemu_put_be32(f, 0);
1741 
1742     if (!spapr->htab) {
1743         int rc;
1744 
1745         assert(kvm_enabled());
1746 
1747         fd = get_htab_fd(spapr);
1748         if (fd < 0) {
1749             return fd;
1750         }
1751 
1752         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1753         if (rc < 0) {
1754             return rc;
1755         }
1756     } else {
1757         if (spapr->htab_first_pass) {
1758             htab_save_first_pass(f, spapr, -1);
1759         }
1760         htab_save_later_pass(f, spapr, -1);
1761     }
1762 
1763     /* End marker */
1764     qemu_put_be32(f, 0);
1765     qemu_put_be16(f, 0);
1766     qemu_put_be16(f, 0);
1767 
1768     return 0;
1769 }
1770 
1771 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1772 {
1773     sPAPRMachineState *spapr = opaque;
1774     uint32_t section_hdr;
1775     int fd = -1;
1776 
1777     if (version_id < 1 || version_id > 1) {
1778         error_report("htab_load() bad version");
1779         return -EINVAL;
1780     }
1781 
1782     section_hdr = qemu_get_be32(f);
1783 
1784     if (section_hdr) {
1785         Error *local_err = NULL;
1786 
1787         /* First section gives the htab size */
1788         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1789         if (local_err) {
1790             error_report_err(local_err);
1791             return -EINVAL;
1792         }
1793         return 0;
1794     }
1795 
1796     if (!spapr->htab) {
1797         assert(kvm_enabled());
1798 
1799         fd = kvmppc_get_htab_fd(true);
1800         if (fd < 0) {
1801             error_report("Unable to open fd to restore KVM hash table: %s",
1802                          strerror(errno));
1803         }
1804     }
1805 
1806     while (true) {
1807         uint32_t index;
1808         uint16_t n_valid, n_invalid;
1809 
1810         index = qemu_get_be32(f);
1811         n_valid = qemu_get_be16(f);
1812         n_invalid = qemu_get_be16(f);
1813 
1814         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1815             /* End of Stream */
1816             break;
1817         }
1818 
1819         if ((index + n_valid + n_invalid) >
1820             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1821             /* Bad index in stream */
1822             error_report(
1823                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1824                 index, n_valid, n_invalid, spapr->htab_shift);
1825             return -EINVAL;
1826         }
1827 
1828         if (spapr->htab) {
1829             if (n_valid) {
1830                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1831                                 HASH_PTE_SIZE_64 * n_valid);
1832             }
1833             if (n_invalid) {
1834                 memset(HPTE(spapr->htab, index + n_valid), 0,
1835                        HASH_PTE_SIZE_64 * n_invalid);
1836             }
1837         } else {
1838             int rc;
1839 
1840             assert(fd >= 0);
1841 
1842             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1843             if (rc < 0) {
1844                 return rc;
1845             }
1846         }
1847     }
1848 
1849     if (!spapr->htab) {
1850         assert(fd >= 0);
1851         close(fd);
1852     }
1853 
1854     return 0;
1855 }
1856 
1857 static void htab_cleanup(void *opaque)
1858 {
1859     sPAPRMachineState *spapr = opaque;
1860 
1861     close_htab_fd(spapr);
1862 }
1863 
1864 static SaveVMHandlers savevm_htab_handlers = {
1865     .save_live_setup = htab_save_setup,
1866     .save_live_iterate = htab_save_iterate,
1867     .save_live_complete_precopy = htab_save_complete,
1868     .cleanup = htab_cleanup,
1869     .load_state = htab_load,
1870 };
1871 
1872 static void spapr_boot_set(void *opaque, const char *boot_device,
1873                            Error **errp)
1874 {
1875     MachineState *machine = MACHINE(qdev_get_machine());
1876     machine->boot_order = g_strdup(boot_device);
1877 }
1878 
1879 /*
1880  * Reset routine for LMB DR devices.
1881  *
1882  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1883  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1884  * when it walks all its children devices. LMB devices reset occurs
1885  * as part of spapr_ppc_reset().
1886  */
1887 static void spapr_drc_reset(void *opaque)
1888 {
1889     sPAPRDRConnector *drc = opaque;
1890     DeviceState *d = DEVICE(drc);
1891 
1892     if (d) {
1893         device_reset(d);
1894     }
1895 }
1896 
1897 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1898 {
1899     MachineState *machine = MACHINE(spapr);
1900     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1901     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1902     int i;
1903 
1904     for (i = 0; i < nr_lmbs; i++) {
1905         sPAPRDRConnector *drc;
1906         uint64_t addr;
1907 
1908         addr = i * lmb_size + spapr->hotplug_memory.base;
1909         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1910                                      addr/lmb_size);
1911         qemu_register_reset(spapr_drc_reset, drc);
1912     }
1913 }
1914 
1915 /*
1916  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1917  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1918  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1919  */
1920 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1921 {
1922     int i;
1923 
1924     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1925         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1926                    " is not aligned to %llu MiB",
1927                    machine->ram_size,
1928                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1929         return;
1930     }
1931 
1932     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1933         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1934                    " is not aligned to %llu MiB",
1935                    machine->ram_size,
1936                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1937         return;
1938     }
1939 
1940     for (i = 0; i < nb_numa_nodes; i++) {
1941         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1942             error_setg(errp,
1943                        "Node %d memory size 0x%" PRIx64
1944                        " is not aligned to %llu MiB",
1945                        i, numa_info[i].node_mem,
1946                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1947             return;
1948         }
1949     }
1950 }
1951 
1952 /* find cpu slot in machine->possible_cpus by core_id */
1953 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1954 {
1955     int index = id / smp_threads;
1956 
1957     if (index >= ms->possible_cpus->len) {
1958         return NULL;
1959     }
1960     if (idx) {
1961         *idx = index;
1962     }
1963     return &ms->possible_cpus->cpus[index];
1964 }
1965 
1966 static void spapr_init_cpus(sPAPRMachineState *spapr)
1967 {
1968     MachineState *machine = MACHINE(spapr);
1969     MachineClass *mc = MACHINE_GET_CLASS(machine);
1970     char *type = spapr_get_cpu_core_type(machine->cpu_model);
1971     int smt = kvmppc_smt_threads();
1972     const CPUArchIdList *possible_cpus;
1973     int boot_cores_nr = smp_cpus / smp_threads;
1974     int i;
1975 
1976     if (!type) {
1977         error_report("Unable to find sPAPR CPU Core definition");
1978         exit(1);
1979     }
1980 
1981     possible_cpus = mc->possible_cpu_arch_ids(machine);
1982     if (mc->has_hotpluggable_cpus) {
1983         if (smp_cpus % smp_threads) {
1984             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1985                          smp_cpus, smp_threads);
1986             exit(1);
1987         }
1988         if (max_cpus % smp_threads) {
1989             error_report("max_cpus (%u) must be multiple of threads (%u)",
1990                          max_cpus, smp_threads);
1991             exit(1);
1992         }
1993     } else {
1994         if (max_cpus != smp_cpus) {
1995             error_report("This machine version does not support CPU hotplug");
1996             exit(1);
1997         }
1998         boot_cores_nr = possible_cpus->len;
1999     }
2000 
2001     for (i = 0; i < possible_cpus->len; i++) {
2002         int core_id = i * smp_threads;
2003 
2004         if (mc->has_hotpluggable_cpus) {
2005             sPAPRDRConnector *drc =
2006                 spapr_dr_connector_new(OBJECT(spapr),
2007                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
2008                                        (core_id / smp_threads) * smt);
2009 
2010             qemu_register_reset(spapr_drc_reset, drc);
2011         }
2012 
2013         if (i < boot_cores_nr) {
2014             Object *core  = object_new(type);
2015             int nr_threads = smp_threads;
2016 
2017             /* Handle the partially filled core for older machine types */
2018             if ((i + 1) * smp_threads >= smp_cpus) {
2019                 nr_threads = smp_cpus - i * smp_threads;
2020             }
2021 
2022             object_property_set_int(core, nr_threads, "nr-threads",
2023                                     &error_fatal);
2024             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2025                                     &error_fatal);
2026             object_property_set_bool(core, true, "realized", &error_fatal);
2027         }
2028     }
2029     g_free(type);
2030 }
2031 
2032 /* pSeries LPAR / sPAPR hardware init */
2033 static void ppc_spapr_init(MachineState *machine)
2034 {
2035     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2036     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2037     const char *kernel_filename = machine->kernel_filename;
2038     const char *initrd_filename = machine->initrd_filename;
2039     PCIHostState *phb;
2040     int i;
2041     MemoryRegion *sysmem = get_system_memory();
2042     MemoryRegion *ram = g_new(MemoryRegion, 1);
2043     MemoryRegion *rma_region;
2044     void *rma = NULL;
2045     hwaddr rma_alloc_size;
2046     hwaddr node0_size = spapr_node0_size();
2047     long load_limit, fw_size;
2048     char *filename;
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, XICS_IRQS_SPAPR, &error_fatal);
2100 
2101     /* Set up containers for ibm,client-set-architecture negotiated options */
2102     spapr->ov5 = spapr_ovec_new();
2103     spapr->ov5_cas = spapr_ovec_new();
2104 
2105     if (smc->dr_lmb_enabled) {
2106         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2107         spapr_validate_node_memory(machine, &error_fatal);
2108     }
2109 
2110     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2111     if (!kvm_enabled() || kvmppc_has_cap_mmu_radix()) {
2112         /* KVM and TCG always allow GTSE with radix... */
2113         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2114     }
2115     /* ... but not with hash (currently). */
2116 
2117     /* advertise support for dedicated HP event source to guests */
2118     if (spapr->use_hotplug_event_source) {
2119         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2120     }
2121 
2122     /* init CPUs */
2123     if (machine->cpu_model == NULL) {
2124         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2125     }
2126 
2127     ppc_cpu_parse_features(machine->cpu_model);
2128 
2129     spapr_init_cpus(spapr);
2130 
2131     if (kvm_enabled()) {
2132         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2133         kvmppc_enable_logical_ci_hcalls();
2134         kvmppc_enable_set_mode_hcall();
2135 
2136         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2137         kvmppc_enable_clear_ref_mod_hcalls();
2138     }
2139 
2140     /* allocate RAM */
2141     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2142                                          machine->ram_size);
2143     memory_region_add_subregion(sysmem, 0, ram);
2144 
2145     if (rma_alloc_size && rma) {
2146         rma_region = g_new(MemoryRegion, 1);
2147         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2148                                    rma_alloc_size, rma);
2149         vmstate_register_ram_global(rma_region);
2150         memory_region_add_subregion(sysmem, 0, rma_region);
2151     }
2152 
2153     /* initialize hotplug memory address space */
2154     if (machine->ram_size < machine->maxram_size) {
2155         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2156         /*
2157          * Limit the number of hotpluggable memory slots to half the number
2158          * slots that KVM supports, leaving the other half for PCI and other
2159          * devices. However ensure that number of slots doesn't drop below 32.
2160          */
2161         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2162                            SPAPR_MAX_RAM_SLOTS;
2163 
2164         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2165             max_memslots = SPAPR_MAX_RAM_SLOTS;
2166         }
2167         if (machine->ram_slots > max_memslots) {
2168             error_report("Specified number of memory slots %"
2169                          PRIu64" exceeds max supported %d",
2170                          machine->ram_slots, max_memslots);
2171             exit(1);
2172         }
2173 
2174         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2175                                               SPAPR_HOTPLUG_MEM_ALIGN);
2176         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2177                            "hotplug-memory", hotplug_mem_size);
2178         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2179                                     &spapr->hotplug_memory.mr);
2180     }
2181 
2182     if (smc->dr_lmb_enabled) {
2183         spapr_create_lmb_dr_connectors(spapr);
2184     }
2185 
2186     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2187     if (!filename) {
2188         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2189         exit(1);
2190     }
2191     spapr->rtas_size = get_image_size(filename);
2192     if (spapr->rtas_size < 0) {
2193         error_report("Could not get size of LPAR rtas '%s'", filename);
2194         exit(1);
2195     }
2196     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2197     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2198         error_report("Could not load LPAR rtas '%s'", filename);
2199         exit(1);
2200     }
2201     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2202         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2203                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2204         exit(1);
2205     }
2206     g_free(filename);
2207 
2208     /* Set up RTAS event infrastructure */
2209     spapr_events_init(spapr);
2210 
2211     /* Set up the RTC RTAS interfaces */
2212     spapr_rtc_create(spapr);
2213 
2214     /* Set up VIO bus */
2215     spapr->vio_bus = spapr_vio_bus_init();
2216 
2217     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2218         if (serial_hds[i]) {
2219             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2220         }
2221     }
2222 
2223     /* We always have at least the nvram device on VIO */
2224     spapr_create_nvram(spapr);
2225 
2226     /* Set up PCI */
2227     spapr_pci_rtas_init();
2228 
2229     phb = spapr_create_phb(spapr, 0);
2230 
2231     for (i = 0; i < nb_nics; i++) {
2232         NICInfo *nd = &nd_table[i];
2233 
2234         if (!nd->model) {
2235             nd->model = g_strdup("ibmveth");
2236         }
2237 
2238         if (strcmp(nd->model, "ibmveth") == 0) {
2239             spapr_vlan_create(spapr->vio_bus, nd);
2240         } else {
2241             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2242         }
2243     }
2244 
2245     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2246         spapr_vscsi_create(spapr->vio_bus);
2247     }
2248 
2249     /* Graphics */
2250     if (spapr_vga_init(phb->bus, &error_fatal)) {
2251         spapr->has_graphics = true;
2252         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2253     }
2254 
2255     if (machine->usb) {
2256         if (smc->use_ohci_by_default) {
2257             pci_create_simple(phb->bus, -1, "pci-ohci");
2258         } else {
2259             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2260         }
2261 
2262         if (spapr->has_graphics) {
2263             USBBus *usb_bus = usb_bus_find(-1);
2264 
2265             usb_create_simple(usb_bus, "usb-kbd");
2266             usb_create_simple(usb_bus, "usb-mouse");
2267         }
2268     }
2269 
2270     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2271         error_report(
2272             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2273             MIN_RMA_SLOF);
2274         exit(1);
2275     }
2276 
2277     if (kernel_filename) {
2278         uint64_t lowaddr = 0;
2279 
2280         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2281                                       NULL, NULL, &lowaddr, NULL, 1,
2282                                       PPC_ELF_MACHINE, 0, 0);
2283         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2284             spapr->kernel_size = load_elf(kernel_filename,
2285                                           translate_kernel_address, NULL, NULL,
2286                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2287                                           0, 0);
2288             spapr->kernel_le = spapr->kernel_size > 0;
2289         }
2290         if (spapr->kernel_size < 0) {
2291             error_report("error loading %s: %s", kernel_filename,
2292                          load_elf_strerror(spapr->kernel_size));
2293             exit(1);
2294         }
2295 
2296         /* load initrd */
2297         if (initrd_filename) {
2298             /* Try to locate the initrd in the gap between the kernel
2299              * and the firmware. Add a bit of space just in case
2300              */
2301             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2302                                   + 0x1ffff) & ~0xffff;
2303             spapr->initrd_size = load_image_targphys(initrd_filename,
2304                                                      spapr->initrd_base,
2305                                                      load_limit
2306                                                      - spapr->initrd_base);
2307             if (spapr->initrd_size < 0) {
2308                 error_report("could not load initial ram disk '%s'",
2309                              initrd_filename);
2310                 exit(1);
2311             }
2312         }
2313     }
2314 
2315     if (bios_name == NULL) {
2316         bios_name = FW_FILE_NAME;
2317     }
2318     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2319     if (!filename) {
2320         error_report("Could not find LPAR firmware '%s'", bios_name);
2321         exit(1);
2322     }
2323     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2324     if (fw_size <= 0) {
2325         error_report("Could not load LPAR firmware '%s'", filename);
2326         exit(1);
2327     }
2328     g_free(filename);
2329 
2330     /* FIXME: Should register things through the MachineState's qdev
2331      * interface, this is a legacy from the sPAPREnvironment structure
2332      * which predated MachineState but had a similar function */
2333     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2334     register_savevm_live(NULL, "spapr/htab", -1, 1,
2335                          &savevm_htab_handlers, spapr);
2336 
2337     /* used by RTAS */
2338     QTAILQ_INIT(&spapr->ccs_list);
2339     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2340 
2341     qemu_register_boot_set(spapr_boot_set, spapr);
2342 
2343     if (kvm_enabled()) {
2344         /* to stop and start vmclock */
2345         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2346                                          &spapr->tb);
2347 
2348         kvmppc_spapr_enable_inkernel_multitce();
2349     }
2350 }
2351 
2352 static int spapr_kvm_type(const char *vm_type)
2353 {
2354     if (!vm_type) {
2355         return 0;
2356     }
2357 
2358     if (!strcmp(vm_type, "HV")) {
2359         return 1;
2360     }
2361 
2362     if (!strcmp(vm_type, "PR")) {
2363         return 2;
2364     }
2365 
2366     error_report("Unknown kvm-type specified '%s'", vm_type);
2367     exit(1);
2368 }
2369 
2370 /*
2371  * Implementation of an interface to adjust firmware path
2372  * for the bootindex property handling.
2373  */
2374 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2375                                    DeviceState *dev)
2376 {
2377 #define CAST(type, obj, name) \
2378     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2379     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2380     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2381 
2382     if (d) {
2383         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2384         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2385         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2386 
2387         if (spapr) {
2388             /*
2389              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2390              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2391              * in the top 16 bits of the 64-bit LUN
2392              */
2393             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2394             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2395                                    (uint64_t)id << 48);
2396         } else if (virtio) {
2397             /*
2398              * We use SRP luns of the form 01000000 | (target << 8) | lun
2399              * in the top 32 bits of the 64-bit LUN
2400              * Note: the quote above is from SLOF and it is wrong,
2401              * the actual binding is:
2402              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2403              */
2404             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2405             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2406                                    (uint64_t)id << 32);
2407         } else if (usb) {
2408             /*
2409              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2410              * in the top 32 bits of the 64-bit LUN
2411              */
2412             unsigned usb_port = atoi(usb->port->path);
2413             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2414             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2415                                    (uint64_t)id << 32);
2416         }
2417     }
2418 
2419     /*
2420      * SLOF probes the USB devices, and if it recognizes that the device is a
2421      * storage device, it changes its name to "storage" instead of "usb-host",
2422      * and additionally adds a child node for the SCSI LUN, so the correct
2423      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2424      */
2425     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2426         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2427         if (usb_host_dev_is_scsi_storage(usbdev)) {
2428             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2429         }
2430     }
2431 
2432     if (phb) {
2433         /* Replace "pci" with "pci@800000020000000" */
2434         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2435     }
2436 
2437     return NULL;
2438 }
2439 
2440 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2441 {
2442     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2443 
2444     return g_strdup(spapr->kvm_type);
2445 }
2446 
2447 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2448 {
2449     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2450 
2451     g_free(spapr->kvm_type);
2452     spapr->kvm_type = g_strdup(value);
2453 }
2454 
2455 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2456 {
2457     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2458 
2459     return spapr->use_hotplug_event_source;
2460 }
2461 
2462 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2463                                             Error **errp)
2464 {
2465     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2466 
2467     spapr->use_hotplug_event_source = value;
2468 }
2469 
2470 static void spapr_machine_initfn(Object *obj)
2471 {
2472     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2473 
2474     spapr->htab_fd = -1;
2475     spapr->use_hotplug_event_source = true;
2476     object_property_add_str(obj, "kvm-type",
2477                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2478     object_property_set_description(obj, "kvm-type",
2479                                     "Specifies the KVM virtualization mode (HV, PR)",
2480                                     NULL);
2481     object_property_add_bool(obj, "modern-hotplug-events",
2482                             spapr_get_modern_hotplug_events,
2483                             spapr_set_modern_hotplug_events,
2484                             NULL);
2485     object_property_set_description(obj, "modern-hotplug-events",
2486                                     "Use dedicated hotplug event mechanism in"
2487                                     " place of standard EPOW events when possible"
2488                                     " (required for memory hot-unplug support)",
2489                                     NULL);
2490 }
2491 
2492 static void spapr_machine_finalizefn(Object *obj)
2493 {
2494     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2495 
2496     g_free(spapr->kvm_type);
2497 }
2498 
2499 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2500 {
2501     cpu_synchronize_state(cs);
2502     ppc_cpu_do_system_reset(cs);
2503 }
2504 
2505 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2506 {
2507     CPUState *cs;
2508 
2509     CPU_FOREACH(cs) {
2510         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2511     }
2512 }
2513 
2514 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2515                            uint32_t node, bool dedicated_hp_event_source,
2516                            Error **errp)
2517 {
2518     sPAPRDRConnector *drc;
2519     sPAPRDRConnectorClass *drck;
2520     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2521     int i, fdt_offset, fdt_size;
2522     void *fdt;
2523     uint64_t addr = addr_start;
2524 
2525     for (i = 0; i < nr_lmbs; i++) {
2526         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2527                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2528         g_assert(drc);
2529 
2530         fdt = create_device_tree(&fdt_size);
2531         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2532                                                 SPAPR_MEMORY_BLOCK_SIZE);
2533 
2534         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2535         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2536         addr += SPAPR_MEMORY_BLOCK_SIZE;
2537         if (!dev->hotplugged) {
2538             /* guests expect coldplugged LMBs to be pre-allocated */
2539             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2540             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2541         }
2542     }
2543     /* send hotplug notification to the
2544      * guest only in case of hotplugged memory
2545      */
2546     if (dev->hotplugged) {
2547         if (dedicated_hp_event_source) {
2548             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2549                     addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2550             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2551             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2552                                                    nr_lmbs,
2553                                                    drck->get_index(drc));
2554         } else {
2555             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2556                                            nr_lmbs);
2557         }
2558     }
2559 }
2560 
2561 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2562                               uint32_t node, Error **errp)
2563 {
2564     Error *local_err = NULL;
2565     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2566     PCDIMMDevice *dimm = PC_DIMM(dev);
2567     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2568     MemoryRegion *mr = ddc->get_memory_region(dimm);
2569     uint64_t align = memory_region_get_alignment(mr);
2570     uint64_t size = memory_region_size(mr);
2571     uint64_t addr;
2572     char *mem_dev;
2573 
2574     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2575         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2576                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2577         goto out;
2578     }
2579 
2580     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2581     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2582         error_setg(&local_err, "Memory backend has bad page size. "
2583                    "Use 'memory-backend-file' with correct mem-path.");
2584         goto out;
2585     }
2586 
2587     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2588     if (local_err) {
2589         goto out;
2590     }
2591 
2592     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2593     if (local_err) {
2594         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2595         goto out;
2596     }
2597 
2598     spapr_add_lmbs(dev, addr, size, node,
2599                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2600                    &error_abort);
2601 
2602 out:
2603     error_propagate(errp, local_err);
2604 }
2605 
2606 typedef struct sPAPRDIMMState {
2607     uint32_t nr_lmbs;
2608 } sPAPRDIMMState;
2609 
2610 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2611 {
2612     sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2613     HotplugHandler *hotplug_ctrl;
2614 
2615     if (--ds->nr_lmbs) {
2616         return;
2617     }
2618 
2619     g_free(ds);
2620 
2621     /*
2622      * Now that all the LMBs have been removed by the guest, call the
2623      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2624      */
2625     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2626     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2627 }
2628 
2629 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2630                            Error **errp)
2631 {
2632     sPAPRDRConnector *drc;
2633     sPAPRDRConnectorClass *drck;
2634     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2635     int i;
2636     sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2637     uint64_t addr = addr_start;
2638 
2639     ds->nr_lmbs = nr_lmbs;
2640     for (i = 0; i < nr_lmbs; i++) {
2641         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2642                 addr / SPAPR_MEMORY_BLOCK_SIZE);
2643         g_assert(drc);
2644 
2645         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2646         drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2647         addr += SPAPR_MEMORY_BLOCK_SIZE;
2648     }
2649 
2650     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2651                                    addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2652     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2653     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2654                                               nr_lmbs,
2655                                               drck->get_index(drc));
2656 }
2657 
2658 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2659                                 Error **errp)
2660 {
2661     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2662     PCDIMMDevice *dimm = PC_DIMM(dev);
2663     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2664     MemoryRegion *mr = ddc->get_memory_region(dimm);
2665 
2666     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2667     object_unparent(OBJECT(dev));
2668 }
2669 
2670 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2671                                         DeviceState *dev, Error **errp)
2672 {
2673     Error *local_err = NULL;
2674     PCDIMMDevice *dimm = PC_DIMM(dev);
2675     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2676     MemoryRegion *mr = ddc->get_memory_region(dimm);
2677     uint64_t size = memory_region_size(mr);
2678     uint64_t addr;
2679 
2680     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2681     if (local_err) {
2682         goto out;
2683     }
2684 
2685     spapr_del_lmbs(dev, addr, size, &error_abort);
2686 out:
2687     error_propagate(errp, local_err);
2688 }
2689 
2690 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2691                                     sPAPRMachineState *spapr)
2692 {
2693     PowerPCCPU *cpu = POWERPC_CPU(cs);
2694     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2695     int id = ppc_get_vcpu_dt_id(cpu);
2696     void *fdt;
2697     int offset, fdt_size;
2698     char *nodename;
2699 
2700     fdt = create_device_tree(&fdt_size);
2701     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2702     offset = fdt_add_subnode(fdt, 0, nodename);
2703 
2704     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2705     g_free(nodename);
2706 
2707     *fdt_offset = offset;
2708     return fdt;
2709 }
2710 
2711 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2712                               Error **errp)
2713 {
2714     MachineState *ms = MACHINE(qdev_get_machine());
2715     CPUCore *cc = CPU_CORE(dev);
2716     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2717 
2718     core_slot->cpu = NULL;
2719     object_unparent(OBJECT(dev));
2720 }
2721 
2722 static void spapr_core_release(DeviceState *dev, void *opaque)
2723 {
2724     HotplugHandler *hotplug_ctrl;
2725 
2726     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2727     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2728 }
2729 
2730 static
2731 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2732                                Error **errp)
2733 {
2734     int index;
2735     sPAPRDRConnector *drc;
2736     sPAPRDRConnectorClass *drck;
2737     Error *local_err = NULL;
2738     CPUCore *cc = CPU_CORE(dev);
2739     int smt = kvmppc_smt_threads();
2740 
2741     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2742         error_setg(errp, "Unable to find CPU core with core-id: %d",
2743                    cc->core_id);
2744         return;
2745     }
2746     if (index == 0) {
2747         error_setg(errp, "Boot CPU core may not be unplugged");
2748         return;
2749     }
2750 
2751     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2752     g_assert(drc);
2753 
2754     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2755     drck->detach(drc, dev, spapr_core_release, NULL, &local_err);
2756     if (local_err) {
2757         error_propagate(errp, local_err);
2758         return;
2759     }
2760 
2761     spapr_hotplug_req_remove_by_index(drc);
2762 }
2763 
2764 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2765                             Error **errp)
2766 {
2767     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2768     MachineClass *mc = MACHINE_GET_CLASS(spapr);
2769     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2770     CPUCore *cc = CPU_CORE(dev);
2771     CPUState *cs = CPU(core->threads);
2772     sPAPRDRConnector *drc;
2773     Error *local_err = NULL;
2774     void *fdt = NULL;
2775     int fdt_offset = 0;
2776     int smt = kvmppc_smt_threads();
2777     CPUArchId *core_slot;
2778     int index;
2779 
2780     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2781     if (!core_slot) {
2782         error_setg(errp, "Unable to find CPU core with core-id: %d",
2783                    cc->core_id);
2784         return;
2785     }
2786     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2787 
2788     g_assert(drc || !mc->has_hotpluggable_cpus);
2789 
2790     /*
2791      * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2792      * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2793      */
2794     if (dev->hotplugged) {
2795         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2796     }
2797 
2798     if (drc) {
2799         sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2800         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
2801         if (local_err) {
2802             g_free(fdt);
2803             error_propagate(errp, local_err);
2804             return;
2805         }
2806     }
2807 
2808     if (dev->hotplugged) {
2809         /*
2810          * Send hotplug notification interrupt to the guest only in case
2811          * of hotplugged CPUs.
2812          */
2813         spapr_hotplug_req_add_by_index(drc);
2814     } else {
2815         /*
2816          * Set the right DRC states for cold plugged CPU.
2817          */
2818         if (drc) {
2819             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2820             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2821             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2822         }
2823     }
2824     core_slot->cpu = OBJECT(dev);
2825 }
2826 
2827 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2828                                 Error **errp)
2829 {
2830     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2831     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2832     Error *local_err = NULL;
2833     CPUCore *cc = CPU_CORE(dev);
2834     sPAPRCPUCore *sc = SPAPR_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 node_id;
2839     int index;
2840 
2841     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2842         error_setg(&local_err, "CPU hotplug not supported for this machine");
2843         goto out;
2844     }
2845 
2846     if (strcmp(base_core_type, type)) {
2847         error_setg(&local_err, "CPU core type should be %s", base_core_type);
2848         goto out;
2849     }
2850 
2851     if (cc->core_id % smp_threads) {
2852         error_setg(&local_err, "invalid core id %d", cc->core_id);
2853         goto out;
2854     }
2855 
2856     if (cc->nr_threads != smp_threads) {
2857         error_setg(errp, "invalid nr-threads %d, must be %d",
2858                    cc->nr_threads, smp_threads);
2859         return;
2860     }
2861 
2862     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2863     if (!core_slot) {
2864         error_setg(&local_err, "core id %d out of range", cc->core_id);
2865         goto out;
2866     }
2867 
2868     if (core_slot->cpu) {
2869         error_setg(&local_err, "core %d already populated", cc->core_id);
2870         goto out;
2871     }
2872 
2873     node_id = core_slot->props.node_id;
2874     if (!core_slot->props.has_node_id) {
2875         /* by default CPUState::numa_node was 0 if it's not set via CLI
2876          * keep it this way for now but in future we probably should
2877          * refuse to start up with incomplete numa mapping */
2878         node_id = 0;
2879     }
2880     if (sc->node_id == CPU_UNSET_NUMA_NODE_ID) {
2881         sc->node_id = node_id;
2882     } else if (sc->node_id != node_id) {
2883         error_setg(&local_err, "node-id %d must match numa node specified"
2884             "with -numa option for cpu-index %d", sc->node_id, cc->core_id);
2885         goto out;
2886     }
2887 
2888 out:
2889     g_free(base_core_type);
2890     error_propagate(errp, local_err);
2891 }
2892 
2893 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2894                                       DeviceState *dev, Error **errp)
2895 {
2896     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2897 
2898     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2899         int node;
2900 
2901         if (!smc->dr_lmb_enabled) {
2902             error_setg(errp, "Memory hotplug not supported for this machine");
2903             return;
2904         }
2905         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2906         if (*errp) {
2907             return;
2908         }
2909         if (node < 0 || node >= MAX_NODES) {
2910             error_setg(errp, "Invaild node %d", node);
2911             return;
2912         }
2913 
2914         /*
2915          * Currently PowerPC kernel doesn't allow hot-adding memory to
2916          * memory-less node, but instead will silently add the memory
2917          * to the first node that has some memory. This causes two
2918          * unexpected behaviours for the user.
2919          *
2920          * - Memory gets hotplugged to a different node than what the user
2921          *   specified.
2922          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2923          *   to memory-less node, a reboot will set things accordingly
2924          *   and the previously hotplugged memory now ends in the right node.
2925          *   This appears as if some memory moved from one node to another.
2926          *
2927          * So until kernel starts supporting memory hotplug to memory-less
2928          * nodes, just prevent such attempts upfront in QEMU.
2929          */
2930         if (nb_numa_nodes && !numa_info[node].node_mem) {
2931             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2932                        node);
2933             return;
2934         }
2935 
2936         spapr_memory_plug(hotplug_dev, dev, node, errp);
2937     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2938         spapr_core_plug(hotplug_dev, dev, errp);
2939     }
2940 }
2941 
2942 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2943                                       DeviceState *dev, Error **errp)
2944 {
2945     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2946     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2947 
2948     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2949         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2950             spapr_memory_unplug(hotplug_dev, dev, errp);
2951         } else {
2952             error_setg(errp, "Memory hot unplug not supported for this guest");
2953         }
2954     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2955         if (!mc->has_hotpluggable_cpus) {
2956             error_setg(errp, "CPU hot unplug not supported on this machine");
2957             return;
2958         }
2959         spapr_core_unplug(hotplug_dev, dev, errp);
2960     }
2961 }
2962 
2963 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2964                                                 DeviceState *dev, Error **errp)
2965 {
2966     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2967     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2968 
2969     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2970         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2971             spapr_memory_unplug_request(hotplug_dev, dev, errp);
2972         } else {
2973             /* NOTE: this means there is a window after guest reset, prior to
2974              * CAS negotiation, where unplug requests will fail due to the
2975              * capability not being detected yet. This is a bit different than
2976              * the case with PCI unplug, where the events will be queued and
2977              * eventually handled by the guest after boot
2978              */
2979             error_setg(errp, "Memory hot unplug not supported for this guest");
2980         }
2981     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2982         if (!mc->has_hotpluggable_cpus) {
2983             error_setg(errp, "CPU hot unplug not supported on this machine");
2984             return;
2985         }
2986         spapr_core_unplug_request(hotplug_dev, dev, errp);
2987     }
2988 }
2989 
2990 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2991                                           DeviceState *dev, Error **errp)
2992 {
2993     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2994         spapr_core_pre_plug(hotplug_dev, dev, errp);
2995     }
2996 }
2997 
2998 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2999                                                  DeviceState *dev)
3000 {
3001     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3002         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3003         return HOTPLUG_HANDLER(machine);
3004     }
3005     return NULL;
3006 }
3007 
3008 static CpuInstanceProperties
3009 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3010 {
3011     CPUArchId *core_slot;
3012     MachineClass *mc = MACHINE_GET_CLASS(machine);
3013 
3014     /* make sure possible_cpu are intialized */
3015     mc->possible_cpu_arch_ids(machine);
3016     /* get CPU core slot containing thread that matches cpu_index */
3017     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3018     assert(core_slot);
3019     return core_slot->props;
3020 }
3021 
3022 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3023 {
3024     int i;
3025     int spapr_max_cores = max_cpus / smp_threads;
3026     MachineClass *mc = MACHINE_GET_CLASS(machine);
3027 
3028     if (!mc->has_hotpluggable_cpus) {
3029         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3030     }
3031     if (machine->possible_cpus) {
3032         assert(machine->possible_cpus->len == spapr_max_cores);
3033         return machine->possible_cpus;
3034     }
3035 
3036     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3037                              sizeof(CPUArchId) * spapr_max_cores);
3038     machine->possible_cpus->len = spapr_max_cores;
3039     for (i = 0; i < machine->possible_cpus->len; i++) {
3040         int core_id = i * smp_threads;
3041 
3042         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3043         machine->possible_cpus->cpus[i].arch_id = core_id;
3044         machine->possible_cpus->cpus[i].props.has_core_id = true;
3045         machine->possible_cpus->cpus[i].props.core_id = core_id;
3046 
3047         /* default distribution of CPUs over NUMA nodes */
3048         if (nb_numa_nodes) {
3049             /* preset values but do not enable them i.e. 'has_node_id = false',
3050              * numa init code will enable them later if manual mapping wasn't
3051              * present on CLI */
3052             machine->possible_cpus->cpus[i].props.node_id =
3053                 core_id / smp_threads / smp_cores % nb_numa_nodes;
3054         }
3055     }
3056     return machine->possible_cpus;
3057 }
3058 
3059 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3060                                 uint64_t *buid, hwaddr *pio,
3061                                 hwaddr *mmio32, hwaddr *mmio64,
3062                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3063 {
3064     /*
3065      * New-style PHB window placement.
3066      *
3067      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3068      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3069      * windows.
3070      *
3071      * Some guest kernels can't work with MMIO windows above 1<<46
3072      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3073      *
3074      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3075      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3076      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3077      * 1TiB 64-bit MMIO windows for each PHB.
3078      */
3079     const uint64_t base_buid = 0x800000020000000ULL;
3080 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3081                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3082     int i;
3083 
3084     /* Sanity check natural alignments */
3085     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3086     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3087     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3088     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3089     /* Sanity check bounds */
3090     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3091                       SPAPR_PCI_MEM32_WIN_SIZE);
3092     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3093                       SPAPR_PCI_MEM64_WIN_SIZE);
3094 
3095     if (index >= SPAPR_MAX_PHBS) {
3096         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3097                    SPAPR_MAX_PHBS - 1);
3098         return;
3099     }
3100 
3101     *buid = base_buid + index;
3102     for (i = 0; i < n_dma; ++i) {
3103         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3104     }
3105 
3106     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3107     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3108     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3109 }
3110 
3111 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3112 {
3113     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3114 
3115     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3116 }
3117 
3118 static void spapr_ics_resend(XICSFabric *dev)
3119 {
3120     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3121 
3122     ics_resend(spapr->ics);
3123 }
3124 
3125 static ICPState *spapr_icp_get(XICSFabric *xi, int cpu_dt_id)
3126 {
3127     PowerPCCPU *cpu = ppc_get_vcpu_by_dt_id(cpu_dt_id);
3128 
3129     return cpu ? ICP(cpu->intc) : NULL;
3130 }
3131 
3132 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3133                                  Monitor *mon)
3134 {
3135     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3136     CPUState *cs;
3137 
3138     CPU_FOREACH(cs) {
3139         PowerPCCPU *cpu = POWERPC_CPU(cs);
3140 
3141         icp_pic_print_info(ICP(cpu->intc), mon);
3142     }
3143 
3144     ics_pic_print_info(spapr->ics, mon);
3145 }
3146 
3147 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3148 {
3149     MachineClass *mc = MACHINE_CLASS(oc);
3150     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3151     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3152     NMIClass *nc = NMI_CLASS(oc);
3153     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3154     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3155     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3156     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3157 
3158     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3159 
3160     /*
3161      * We set up the default / latest behaviour here.  The class_init
3162      * functions for the specific versioned machine types can override
3163      * these details for backwards compatibility
3164      */
3165     mc->init = ppc_spapr_init;
3166     mc->reset = ppc_spapr_reset;
3167     mc->block_default_type = IF_SCSI;
3168     mc->max_cpus = 1024;
3169     mc->no_parallel = 1;
3170     mc->default_boot_order = "";
3171     mc->default_ram_size = 512 * M_BYTE;
3172     mc->kvm_type = spapr_kvm_type;
3173     mc->has_dynamic_sysbus = true;
3174     mc->pci_allow_0_address = true;
3175     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3176     hc->pre_plug = spapr_machine_device_pre_plug;
3177     hc->plug = spapr_machine_device_plug;
3178     hc->unplug = spapr_machine_device_unplug;
3179     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3180     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3181     hc->unplug_request = spapr_machine_device_unplug_request;
3182 
3183     smc->dr_lmb_enabled = true;
3184     smc->tcg_default_cpu = "POWER8";
3185     mc->has_hotpluggable_cpus = true;
3186     fwc->get_dev_path = spapr_get_fw_dev_path;
3187     nc->nmi_monitor_handler = spapr_nmi;
3188     smc->phb_placement = spapr_phb_placement;
3189     vhc->hypercall = emulate_spapr_hypercall;
3190     vhc->hpt_mask = spapr_hpt_mask;
3191     vhc->map_hptes = spapr_map_hptes;
3192     vhc->unmap_hptes = spapr_unmap_hptes;
3193     vhc->store_hpte = spapr_store_hpte;
3194     vhc->get_patbe = spapr_get_patbe;
3195     xic->ics_get = spapr_ics_get;
3196     xic->ics_resend = spapr_ics_resend;
3197     xic->icp_get = spapr_icp_get;
3198     ispc->print_info = spapr_pic_print_info;
3199     /* Force NUMA node memory size to be a multiple of
3200      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3201      * in which LMBs are represented and hot-added
3202      */
3203     mc->numa_mem_align_shift = 28;
3204 }
3205 
3206 static const TypeInfo spapr_machine_info = {
3207     .name          = TYPE_SPAPR_MACHINE,
3208     .parent        = TYPE_MACHINE,
3209     .abstract      = true,
3210     .instance_size = sizeof(sPAPRMachineState),
3211     .instance_init = spapr_machine_initfn,
3212     .instance_finalize = spapr_machine_finalizefn,
3213     .class_size    = sizeof(sPAPRMachineClass),
3214     .class_init    = spapr_machine_class_init,
3215     .interfaces = (InterfaceInfo[]) {
3216         { TYPE_FW_PATH_PROVIDER },
3217         { TYPE_NMI },
3218         { TYPE_HOTPLUG_HANDLER },
3219         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3220         { TYPE_XICS_FABRIC },
3221         { TYPE_INTERRUPT_STATS_PROVIDER },
3222         { }
3223     },
3224 };
3225 
3226 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3227     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3228                                                     void *data)      \
3229     {                                                                \
3230         MachineClass *mc = MACHINE_CLASS(oc);                        \
3231         spapr_machine_##suffix##_class_options(mc);                  \
3232         if (latest) {                                                \
3233             mc->alias = "pseries";                                   \
3234             mc->is_default = 1;                                      \
3235         }                                                            \
3236     }                                                                \
3237     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3238     {                                                                \
3239         MachineState *machine = MACHINE(obj);                        \
3240         spapr_machine_##suffix##_instance_options(machine);          \
3241     }                                                                \
3242     static const TypeInfo spapr_machine_##suffix##_info = {          \
3243         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3244         .parent = TYPE_SPAPR_MACHINE,                                \
3245         .class_init = spapr_machine_##suffix##_class_init,           \
3246         .instance_init = spapr_machine_##suffix##_instance_init,     \
3247     };                                                               \
3248     static void spapr_machine_register_##suffix(void)                \
3249     {                                                                \
3250         type_register(&spapr_machine_##suffix##_info);               \
3251     }                                                                \
3252     type_init(spapr_machine_register_##suffix)
3253 
3254 /*
3255  * pseries-2.10
3256  */
3257 static void spapr_machine_2_10_instance_options(MachineState *machine)
3258 {
3259 }
3260 
3261 static void spapr_machine_2_10_class_options(MachineClass *mc)
3262 {
3263     /* Defaults for the latest behaviour inherited from the base class */
3264 }
3265 
3266 DEFINE_SPAPR_MACHINE(2_10, "2.10", true);
3267 
3268 /*
3269  * pseries-2.9
3270  */
3271 #define SPAPR_COMPAT_2_9                                               \
3272     HW_COMPAT_2_9
3273 
3274 static void spapr_machine_2_9_instance_options(MachineState *machine)
3275 {
3276     spapr_machine_2_10_instance_options(machine);
3277 }
3278 
3279 static void spapr_machine_2_9_class_options(MachineClass *mc)
3280 {
3281     spapr_machine_2_10_class_options(mc);
3282     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
3283     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
3284 }
3285 
3286 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
3287 
3288 /*
3289  * pseries-2.8
3290  */
3291 #define SPAPR_COMPAT_2_8                                        \
3292     HW_COMPAT_2_8                                               \
3293     {                                                           \
3294         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
3295         .property = "pcie-extended-configuration-space",        \
3296         .value    = "off",                                      \
3297     },
3298 
3299 static void spapr_machine_2_8_instance_options(MachineState *machine)
3300 {
3301     spapr_machine_2_9_instance_options(machine);
3302 }
3303 
3304 static void spapr_machine_2_8_class_options(MachineClass *mc)
3305 {
3306     spapr_machine_2_9_class_options(mc);
3307     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3308     mc->numa_mem_align_shift = 23;
3309 }
3310 
3311 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3312 
3313 /*
3314  * pseries-2.7
3315  */
3316 #define SPAPR_COMPAT_2_7                            \
3317     HW_COMPAT_2_7                                   \
3318     {                                               \
3319         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3320         .property = "mem_win_size",                 \
3321         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3322     },                                              \
3323     {                                               \
3324         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3325         .property = "mem64_win_size",               \
3326         .value    = "0",                            \
3327     },                                              \
3328     {                                               \
3329         .driver = TYPE_POWERPC_CPU,                 \
3330         .property = "pre-2.8-migration",            \
3331         .value    = "on",                           \
3332     },                                              \
3333     {                                               \
3334         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3335         .property = "pre-2.8-migration",            \
3336         .value    = "on",                           \
3337     },
3338 
3339 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3340                               uint64_t *buid, hwaddr *pio,
3341                               hwaddr *mmio32, hwaddr *mmio64,
3342                               unsigned n_dma, uint32_t *liobns, Error **errp)
3343 {
3344     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3345     const uint64_t base_buid = 0x800000020000000ULL;
3346     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3347     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3348     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3349     const uint32_t max_index = 255;
3350     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3351 
3352     uint64_t ram_top = MACHINE(spapr)->ram_size;
3353     hwaddr phb0_base, phb_base;
3354     int i;
3355 
3356     /* Do we have hotpluggable memory? */
3357     if (MACHINE(spapr)->maxram_size > ram_top) {
3358         /* Can't just use maxram_size, because there may be an
3359          * alignment gap between normal and hotpluggable memory
3360          * regions */
3361         ram_top = spapr->hotplug_memory.base +
3362             memory_region_size(&spapr->hotplug_memory.mr);
3363     }
3364 
3365     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3366 
3367     if (index > max_index) {
3368         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3369                    max_index);
3370         return;
3371     }
3372 
3373     *buid = base_buid + index;
3374     for (i = 0; i < n_dma; ++i) {
3375         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3376     }
3377 
3378     phb_base = phb0_base + index * phb_spacing;
3379     *pio = phb_base + pio_offset;
3380     *mmio32 = phb_base + mmio_offset;
3381     /*
3382      * We don't set the 64-bit MMIO window, relying on the PHB's
3383      * fallback behaviour of automatically splitting a large "32-bit"
3384      * window into contiguous 32-bit and 64-bit windows
3385      */
3386 }
3387 
3388 static void spapr_machine_2_7_instance_options(MachineState *machine)
3389 {
3390     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3391 
3392     spapr_machine_2_8_instance_options(machine);
3393     spapr->use_hotplug_event_source = false;
3394 }
3395 
3396 static void spapr_machine_2_7_class_options(MachineClass *mc)
3397 {
3398     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3399 
3400     spapr_machine_2_8_class_options(mc);
3401     smc->tcg_default_cpu = "POWER7";
3402     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3403     smc->phb_placement = phb_placement_2_7;
3404 }
3405 
3406 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3407 
3408 /*
3409  * pseries-2.6
3410  */
3411 #define SPAPR_COMPAT_2_6 \
3412     HW_COMPAT_2_6 \
3413     { \
3414         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3415         .property = "ddw",\
3416         .value    = stringify(off),\
3417     },
3418 
3419 static void spapr_machine_2_6_instance_options(MachineState *machine)
3420 {
3421     spapr_machine_2_7_instance_options(machine);
3422 }
3423 
3424 static void spapr_machine_2_6_class_options(MachineClass *mc)
3425 {
3426     spapr_machine_2_7_class_options(mc);
3427     mc->has_hotpluggable_cpus = false;
3428     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3429 }
3430 
3431 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3432 
3433 /*
3434  * pseries-2.5
3435  */
3436 #define SPAPR_COMPAT_2_5 \
3437     HW_COMPAT_2_5 \
3438     { \
3439         .driver   = "spapr-vlan", \
3440         .property = "use-rx-buffer-pools", \
3441         .value    = "off", \
3442     },
3443 
3444 static void spapr_machine_2_5_instance_options(MachineState *machine)
3445 {
3446     spapr_machine_2_6_instance_options(machine);
3447 }
3448 
3449 static void spapr_machine_2_5_class_options(MachineClass *mc)
3450 {
3451     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3452 
3453     spapr_machine_2_6_class_options(mc);
3454     smc->use_ohci_by_default = true;
3455     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3456 }
3457 
3458 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3459 
3460 /*
3461  * pseries-2.4
3462  */
3463 #define SPAPR_COMPAT_2_4 \
3464         HW_COMPAT_2_4
3465 
3466 static void spapr_machine_2_4_instance_options(MachineState *machine)
3467 {
3468     spapr_machine_2_5_instance_options(machine);
3469 }
3470 
3471 static void spapr_machine_2_4_class_options(MachineClass *mc)
3472 {
3473     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3474 
3475     spapr_machine_2_5_class_options(mc);
3476     smc->dr_lmb_enabled = false;
3477     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3478 }
3479 
3480 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3481 
3482 /*
3483  * pseries-2.3
3484  */
3485 #define SPAPR_COMPAT_2_3 \
3486         HW_COMPAT_2_3 \
3487         {\
3488             .driver   = "spapr-pci-host-bridge",\
3489             .property = "dynamic-reconfiguration",\
3490             .value    = "off",\
3491         },
3492 
3493 static void spapr_machine_2_3_instance_options(MachineState *machine)
3494 {
3495     spapr_machine_2_4_instance_options(machine);
3496     savevm_skip_section_footers();
3497     global_state_set_optional();
3498     savevm_skip_configuration();
3499 }
3500 
3501 static void spapr_machine_2_3_class_options(MachineClass *mc)
3502 {
3503     spapr_machine_2_4_class_options(mc);
3504     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3505 }
3506 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3507 
3508 /*
3509  * pseries-2.2
3510  */
3511 
3512 #define SPAPR_COMPAT_2_2 \
3513         HW_COMPAT_2_2 \
3514         {\
3515             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3516             .property = "mem_win_size",\
3517             .value    = "0x20000000",\
3518         },
3519 
3520 static void spapr_machine_2_2_instance_options(MachineState *machine)
3521 {
3522     spapr_machine_2_3_instance_options(machine);
3523     machine->suppress_vmdesc = true;
3524 }
3525 
3526 static void spapr_machine_2_2_class_options(MachineClass *mc)
3527 {
3528     spapr_machine_2_3_class_options(mc);
3529     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3530 }
3531 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3532 
3533 /*
3534  * pseries-2.1
3535  */
3536 #define SPAPR_COMPAT_2_1 \
3537         HW_COMPAT_2_1
3538 
3539 static void spapr_machine_2_1_instance_options(MachineState *machine)
3540 {
3541     spapr_machine_2_2_instance_options(machine);
3542 }
3543 
3544 static void spapr_machine_2_1_class_options(MachineClass *mc)
3545 {
3546     spapr_machine_2_2_class_options(mc);
3547     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3548 }
3549 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3550 
3551 static void spapr_machine_register_types(void)
3552 {
3553     type_register_static(&spapr_machine_info);
3554 }
3555 
3556 type_init(spapr_machine_register_types)
3557