xref: /openbmc/qemu/hw/ppc/spapr.c (revision f7759e4331ed04b2128af36efd395e55e3076406)
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 "qom/cpu.h"
44 
45 #include "hw/boards.h"
46 #include "hw/ppc/ppc.h"
47 #include "hw/loader.h"
48 
49 #include "hw/ppc/fdt.h"
50 #include "hw/ppc/spapr.h"
51 #include "hw/ppc/spapr_vio.h"
52 #include "hw/pci-host/spapr.h"
53 #include "hw/ppc/xics.h"
54 #include "hw/pci/msi.h"
55 
56 #include "hw/pci/pci.h"
57 #include "hw/scsi/scsi.h"
58 #include "hw/virtio/virtio-scsi.h"
59 
60 #include "exec/address-spaces.h"
61 #include "hw/usb.h"
62 #include "qemu/config-file.h"
63 #include "qemu/error-report.h"
64 #include "trace.h"
65 #include "hw/nmi.h"
66 
67 #include "hw/compat.h"
68 #include "qemu/cutils.h"
69 #include "hw/ppc/spapr_cpu_core.h"
70 #include "qmp-commands.h"
71 
72 #include <libfdt.h>
73 
74 /* SLOF memory layout:
75  *
76  * SLOF raw image loaded at 0, copies its romfs right below the flat
77  * device-tree, then position SLOF itself 31M below that
78  *
79  * So we set FW_OVERHEAD to 40MB which should account for all of that
80  * and more
81  *
82  * We load our kernel at 4M, leaving space for SLOF initial image
83  */
84 #define FDT_MAX_SIZE            0x100000
85 #define RTAS_MAX_SIZE           0x10000
86 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
87 #define FW_MAX_SIZE             0x400000
88 #define FW_FILE_NAME            "slof.bin"
89 #define FW_OVERHEAD             0x2800000
90 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
91 
92 #define MIN_RMA_SLOF            128UL
93 
94 #define PHANDLE_XICP            0x00001111
95 
96 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
97 
98 static XICSState *try_create_xics(sPAPRMachineState *spapr,
99                                   const char *type, const char *type_ics,
100                                   const char *type_icp, int nr_servers,
101                                   int nr_irqs, Error **errp)
102 {
103     Error *err = NULL, *local_err = NULL;
104     XICSState *xics;
105     ICSState *ics = NULL;
106     int i;
107 
108     xics = XICS_COMMON(object_new(type));
109     qdev_set_parent_bus(DEVICE(xics), sysbus_get_default());
110     object_property_set_bool(OBJECT(xics), true, "realized", &err);
111     if (err) {
112         goto error;
113     }
114 
115     ics = ICS_SIMPLE(object_new(type_ics));
116     object_property_add_child(OBJECT(spapr), "ics", OBJECT(ics), NULL);
117     object_property_set_int(OBJECT(ics), nr_irqs, "nr-irqs", &err);
118     object_property_add_const_link(OBJECT(ics), "xics", OBJECT(xics), NULL);
119     object_property_set_bool(OBJECT(ics), true, "realized", &local_err);
120     error_propagate(&err, local_err);
121     if (err) {
122         goto error;
123     }
124     QLIST_INSERT_HEAD(&xics->ics, ics, list);
125 
126     xics->ss = g_malloc0(nr_servers * sizeof(ICPState));
127     xics->nr_servers = nr_servers;
128 
129     for (i = 0; i < nr_servers; i++) {
130         ICPState *icp = &xics->ss[i];
131 
132         object_initialize(icp, sizeof(*icp), type_icp);
133         object_property_add_child(OBJECT(xics), "icp[*]", OBJECT(icp), NULL);
134         object_property_add_const_link(OBJECT(icp), "xics", OBJECT(xics), NULL);
135         object_property_set_bool(OBJECT(icp), true, "realized", &err);
136         if (err) {
137             goto error;
138         }
139         object_unref(OBJECT(icp));
140     }
141 
142     spapr->ics = ics;
143     return xics;
144 
145 error:
146     error_propagate(errp, err);
147     if (ics) {
148         object_unparent(OBJECT(ics));
149     }
150     object_unparent(OBJECT(xics));
151     return NULL;
152 }
153 
154 static XICSState *xics_system_init(MachineState *machine,
155                                    int nr_servers, int nr_irqs, Error **errp)
156 {
157     XICSState *xics = NULL;
158 
159     if (kvm_enabled()) {
160         Error *err = NULL;
161 
162         if (machine_kernel_irqchip_allowed(machine)) {
163             xics = try_create_xics(SPAPR_MACHINE(machine),
164                                    TYPE_XICS_SPAPR_KVM, TYPE_ICS_KVM,
165                                    TYPE_KVM_ICP, nr_servers, nr_irqs, &err);
166         }
167         if (machine_kernel_irqchip_required(machine) && !xics) {
168             error_reportf_err(err,
169                               "kernel_irqchip requested but unavailable: ");
170         } else {
171             error_free(err);
172         }
173     }
174 
175     if (!xics) {
176         xics = try_create_xics(SPAPR_MACHINE(machine),
177                                TYPE_XICS_SPAPR, TYPE_ICS_SIMPLE,
178                                TYPE_ICP, nr_servers, nr_irqs, errp);
179     }
180 
181     return xics;
182 }
183 
184 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
185                                   int smt_threads)
186 {
187     int i, ret = 0;
188     uint32_t servers_prop[smt_threads];
189     uint32_t gservers_prop[smt_threads * 2];
190     int index = ppc_get_vcpu_dt_id(cpu);
191 
192     if (cpu->compat_pvr) {
193         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
194         if (ret < 0) {
195             return ret;
196         }
197     }
198 
199     /* Build interrupt servers and gservers properties */
200     for (i = 0; i < smt_threads; i++) {
201         servers_prop[i] = cpu_to_be32(index + i);
202         /* Hack, direct the group queues back to cpu 0 */
203         gservers_prop[i*2] = cpu_to_be32(index + i);
204         gservers_prop[i*2 + 1] = 0;
205     }
206     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
207                       servers_prop, sizeof(servers_prop));
208     if (ret < 0) {
209         return ret;
210     }
211     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
212                       gservers_prop, sizeof(gservers_prop));
213 
214     return ret;
215 }
216 
217 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
218 {
219     int ret = 0;
220     PowerPCCPU *cpu = POWERPC_CPU(cs);
221     int index = ppc_get_vcpu_dt_id(cpu);
222     uint32_t associativity[] = {cpu_to_be32(0x5),
223                                 cpu_to_be32(0x0),
224                                 cpu_to_be32(0x0),
225                                 cpu_to_be32(0x0),
226                                 cpu_to_be32(cs->numa_node),
227                                 cpu_to_be32(index)};
228 
229     /* Advertise NUMA via ibm,associativity */
230     if (nb_numa_nodes > 1) {
231         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
232                           sizeof(associativity));
233     }
234 
235     return ret;
236 }
237 
238 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
239 {
240     int ret = 0, offset, cpus_offset;
241     CPUState *cs;
242     char cpu_model[32];
243     int smt = kvmppc_smt_threads();
244     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
245 
246     CPU_FOREACH(cs) {
247         PowerPCCPU *cpu = POWERPC_CPU(cs);
248         DeviceClass *dc = DEVICE_GET_CLASS(cs);
249         int index = ppc_get_vcpu_dt_id(cpu);
250         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
251 
252         if ((index % smt) != 0) {
253             continue;
254         }
255 
256         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
257 
258         cpus_offset = fdt_path_offset(fdt, "/cpus");
259         if (cpus_offset < 0) {
260             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
261                                           "cpus");
262             if (cpus_offset < 0) {
263                 return cpus_offset;
264             }
265         }
266         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
267         if (offset < 0) {
268             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
269             if (offset < 0) {
270                 return offset;
271             }
272         }
273 
274         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
275                           pft_size_prop, sizeof(pft_size_prop));
276         if (ret < 0) {
277             return ret;
278         }
279 
280         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
281         if (ret < 0) {
282             return ret;
283         }
284 
285         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
286         if (ret < 0) {
287             return ret;
288         }
289     }
290     return ret;
291 }
292 
293 static hwaddr spapr_node0_size(void)
294 {
295     MachineState *machine = MACHINE(qdev_get_machine());
296 
297     if (nb_numa_nodes) {
298         int i;
299         for (i = 0; i < nb_numa_nodes; ++i) {
300             if (numa_info[i].node_mem) {
301                 return MIN(pow2floor(numa_info[i].node_mem),
302                            machine->ram_size);
303             }
304         }
305     }
306     return machine->ram_size;
307 }
308 
309 static void add_str(GString *s, const gchar *s1)
310 {
311     g_string_append_len(s, s1, strlen(s1) + 1);
312 }
313 
314 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
315                                        hwaddr size)
316 {
317     uint32_t associativity[] = {
318         cpu_to_be32(0x4), /* length */
319         cpu_to_be32(0x0), cpu_to_be32(0x0),
320         cpu_to_be32(0x0), cpu_to_be32(nodeid)
321     };
322     char mem_name[32];
323     uint64_t mem_reg_property[2];
324     int off;
325 
326     mem_reg_property[0] = cpu_to_be64(start);
327     mem_reg_property[1] = cpu_to_be64(size);
328 
329     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
330     off = fdt_add_subnode(fdt, 0, mem_name);
331     _FDT(off);
332     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
333     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
334                       sizeof(mem_reg_property))));
335     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
336                       sizeof(associativity))));
337     return off;
338 }
339 
340 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
341 {
342     MachineState *machine = MACHINE(spapr);
343     hwaddr mem_start, node_size;
344     int i, nb_nodes = nb_numa_nodes;
345     NodeInfo *nodes = numa_info;
346     NodeInfo ramnode;
347 
348     /* No NUMA nodes, assume there is just one node with whole RAM */
349     if (!nb_numa_nodes) {
350         nb_nodes = 1;
351         ramnode.node_mem = machine->ram_size;
352         nodes = &ramnode;
353     }
354 
355     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
356         if (!nodes[i].node_mem) {
357             continue;
358         }
359         if (mem_start >= machine->ram_size) {
360             node_size = 0;
361         } else {
362             node_size = nodes[i].node_mem;
363             if (node_size > machine->ram_size - mem_start) {
364                 node_size = machine->ram_size - mem_start;
365             }
366         }
367         if (!mem_start) {
368             /* ppc_spapr_init() checks for rma_size <= node0_size already */
369             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
370             mem_start += spapr->rma_size;
371             node_size -= spapr->rma_size;
372         }
373         for ( ; node_size; ) {
374             hwaddr sizetmp = pow2floor(node_size);
375 
376             /* mem_start != 0 here */
377             if (ctzl(mem_start) < ctzl(sizetmp)) {
378                 sizetmp = 1ULL << ctzl(mem_start);
379             }
380 
381             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
382             node_size -= sizetmp;
383             mem_start += sizetmp;
384         }
385     }
386 
387     return 0;
388 }
389 
390 /* Populate the "ibm,pa-features" property */
391 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset)
392 {
393     uint8_t pa_features_206[] = { 6, 0,
394         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
395     uint8_t pa_features_207[] = { 24, 0,
396         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
397         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
398         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
399         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
400     uint8_t *pa_features;
401     size_t pa_size;
402 
403     switch (env->mmu_model) {
404     case POWERPC_MMU_2_06:
405     case POWERPC_MMU_2_06a:
406         pa_features = pa_features_206;
407         pa_size = sizeof(pa_features_206);
408         break;
409     case POWERPC_MMU_2_07:
410     case POWERPC_MMU_2_07a:
411         pa_features = pa_features_207;
412         pa_size = sizeof(pa_features_207);
413         break;
414     default:
415         return;
416     }
417 
418     if (env->ci_large_pages) {
419         /*
420          * Note: we keep CI large pages off by default because a 64K capable
421          * guest provisioned with large pages might otherwise try to map a qemu
422          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
423          * even if that qemu runs on a 4k host.
424          * We dd this bit back here if we are confident this is not an issue
425          */
426         pa_features[3] |= 0x20;
427     }
428     if (kvmppc_has_cap_htm() && pa_size > 24) {
429         pa_features[24] |= 0x80;    /* Transactional memory support */
430     }
431 
432     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
433 }
434 
435 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
436                                   sPAPRMachineState *spapr)
437 {
438     PowerPCCPU *cpu = POWERPC_CPU(cs);
439     CPUPPCState *env = &cpu->env;
440     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
441     int index = ppc_get_vcpu_dt_id(cpu);
442     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
443                        0xffffffff, 0xffffffff};
444     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
445         : SPAPR_TIMEBASE_FREQ;
446     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
447     uint32_t page_sizes_prop[64];
448     size_t page_sizes_prop_size;
449     uint32_t vcpus_per_socket = smp_threads * smp_cores;
450     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
451     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
452     sPAPRDRConnector *drc;
453     sPAPRDRConnectorClass *drck;
454     int drc_index;
455 
456     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
457     if (drc) {
458         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
459         drc_index = drck->get_index(drc);
460         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
461     }
462 
463     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
464     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
465 
466     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
467     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
468                            env->dcache_line_size)));
469     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
470                            env->dcache_line_size)));
471     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
472                            env->icache_line_size)));
473     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
474                            env->icache_line_size)));
475 
476     if (pcc->l1_dcache_size) {
477         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
478                                pcc->l1_dcache_size)));
479     } else {
480         error_report("Warning: Unknown L1 dcache size for cpu");
481     }
482     if (pcc->l1_icache_size) {
483         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
484                                pcc->l1_icache_size)));
485     } else {
486         error_report("Warning: Unknown L1 icache size for cpu");
487     }
488 
489     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
490     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
491     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
492     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
493     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
494     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
495 
496     if (env->spr_cb[SPR_PURR].oea_read) {
497         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
498     }
499 
500     if (env->mmu_model & POWERPC_MMU_1TSEG) {
501         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
502                           segs, sizeof(segs))));
503     }
504 
505     /* Advertise VMX/VSX (vector extensions) if available
506      *   0 / no property == no vector extensions
507      *   1               == VMX / Altivec available
508      *   2               == VSX available */
509     if (env->insns_flags & PPC_ALTIVEC) {
510         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
511 
512         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
513     }
514 
515     /* Advertise DFP (Decimal Floating Point) if available
516      *   0 / no property == no DFP
517      *   1               == DFP available */
518     if (env->insns_flags2 & PPC2_DFP) {
519         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
520     }
521 
522     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
523                                                   sizeof(page_sizes_prop));
524     if (page_sizes_prop_size) {
525         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
526                           page_sizes_prop, page_sizes_prop_size)));
527     }
528 
529     spapr_populate_pa_features(env, fdt, offset);
530 
531     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
532                            cs->cpu_index / vcpus_per_socket)));
533 
534     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
535                       pft_size_prop, sizeof(pft_size_prop))));
536 
537     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
538 
539     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
540 }
541 
542 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
543 {
544     CPUState *cs;
545     int cpus_offset;
546     char *nodename;
547     int smt = kvmppc_smt_threads();
548 
549     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
550     _FDT(cpus_offset);
551     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
552     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
553 
554     /*
555      * We walk the CPUs in reverse order to ensure that CPU DT nodes
556      * created by fdt_add_subnode() end up in the right order in FDT
557      * for the guest kernel the enumerate the CPUs correctly.
558      */
559     CPU_FOREACH_REVERSE(cs) {
560         PowerPCCPU *cpu = POWERPC_CPU(cs);
561         int index = ppc_get_vcpu_dt_id(cpu);
562         DeviceClass *dc = DEVICE_GET_CLASS(cs);
563         int offset;
564 
565         if ((index % smt) != 0) {
566             continue;
567         }
568 
569         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
570         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
571         g_free(nodename);
572         _FDT(offset);
573         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
574     }
575 
576 }
577 
578 /*
579  * Adds ibm,dynamic-reconfiguration-memory node.
580  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
581  * of this device tree node.
582  */
583 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
584 {
585     MachineState *machine = MACHINE(spapr);
586     int ret, i, offset;
587     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
588     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
589     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
590     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
591                        memory_region_size(&spapr->hotplug_memory.mr)) /
592                        lmb_size;
593     uint32_t *int_buf, *cur_index, buf_len;
594     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
595 
596     /*
597      * Don't create the node if there is no hotpluggable memory
598      */
599     if (machine->ram_size == machine->maxram_size) {
600         return 0;
601     }
602 
603     /*
604      * Allocate enough buffer size to fit in ibm,dynamic-memory
605      * or ibm,associativity-lookup-arrays
606      */
607     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
608               * sizeof(uint32_t);
609     cur_index = int_buf = g_malloc0(buf_len);
610 
611     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
612 
613     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
614                     sizeof(prop_lmb_size));
615     if (ret < 0) {
616         goto out;
617     }
618 
619     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
620     if (ret < 0) {
621         goto out;
622     }
623 
624     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
625     if (ret < 0) {
626         goto out;
627     }
628 
629     /* ibm,dynamic-memory */
630     int_buf[0] = cpu_to_be32(nr_lmbs);
631     cur_index++;
632     for (i = 0; i < nr_lmbs; i++) {
633         uint64_t addr = i * lmb_size;
634         uint32_t *dynamic_memory = cur_index;
635 
636         if (i >= hotplug_lmb_start) {
637             sPAPRDRConnector *drc;
638             sPAPRDRConnectorClass *drck;
639 
640             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
641             g_assert(drc);
642             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
643 
644             dynamic_memory[0] = cpu_to_be32(addr >> 32);
645             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
646             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
647             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
648             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
649             if (memory_region_present(get_system_memory(), addr)) {
650                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
651             } else {
652                 dynamic_memory[5] = cpu_to_be32(0);
653             }
654         } else {
655             /*
656              * LMB information for RMA, boot time RAM and gap b/n RAM and
657              * hotplug memory region -- all these are marked as reserved
658              * and as having no valid DRC.
659              */
660             dynamic_memory[0] = cpu_to_be32(addr >> 32);
661             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
662             dynamic_memory[2] = cpu_to_be32(0);
663             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
664             dynamic_memory[4] = cpu_to_be32(-1);
665             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
666                                             SPAPR_LMB_FLAGS_DRC_INVALID);
667         }
668 
669         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
670     }
671     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
672     if (ret < 0) {
673         goto out;
674     }
675 
676     /* ibm,associativity-lookup-arrays */
677     cur_index = int_buf;
678     int_buf[0] = cpu_to_be32(nr_nodes);
679     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
680     cur_index += 2;
681     for (i = 0; i < nr_nodes; i++) {
682         uint32_t associativity[] = {
683             cpu_to_be32(0x0),
684             cpu_to_be32(0x0),
685             cpu_to_be32(0x0),
686             cpu_to_be32(i)
687         };
688         memcpy(cur_index, associativity, sizeof(associativity));
689         cur_index += 4;
690     }
691     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
692             (cur_index - int_buf) * sizeof(uint32_t));
693 out:
694     g_free(int_buf);
695     return ret;
696 }
697 
698 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
699                                 sPAPROptionVector *ov5_updates)
700 {
701     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
702     int ret = 0, offset;
703 
704     /* Generate ibm,dynamic-reconfiguration-memory node if required */
705     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
706         g_assert(smc->dr_lmb_enabled);
707         ret = spapr_populate_drconf_memory(spapr, fdt);
708         if (ret) {
709             goto out;
710         }
711     }
712 
713     offset = fdt_path_offset(fdt, "/chosen");
714     if (offset < 0) {
715         offset = fdt_add_subnode(fdt, 0, "chosen");
716         if (offset < 0) {
717             return offset;
718         }
719     }
720     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
721                                  "ibm,architecture-vec-5");
722 
723 out:
724     return ret;
725 }
726 
727 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
728                                  target_ulong addr, target_ulong size,
729                                  sPAPROptionVector *ov5_updates)
730 {
731     void *fdt, *fdt_skel;
732     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
733 
734     size -= sizeof(hdr);
735 
736     /* Create sceleton */
737     fdt_skel = g_malloc0(size);
738     _FDT((fdt_create(fdt_skel, size)));
739     _FDT((fdt_begin_node(fdt_skel, "")));
740     _FDT((fdt_end_node(fdt_skel)));
741     _FDT((fdt_finish(fdt_skel)));
742     fdt = g_malloc0(size);
743     _FDT((fdt_open_into(fdt_skel, fdt, size)));
744     g_free(fdt_skel);
745 
746     /* Fixup cpu nodes */
747     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
748 
749     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
750         return -1;
751     }
752 
753     /* Pack resulting tree */
754     _FDT((fdt_pack(fdt)));
755 
756     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
757         trace_spapr_cas_failed(size);
758         return -1;
759     }
760 
761     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
762     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
763     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
764     g_free(fdt);
765 
766     return 0;
767 }
768 
769 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
770 {
771     int rtas;
772     GString *hypertas = g_string_sized_new(256);
773     GString *qemu_hypertas = g_string_sized_new(256);
774     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
775     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
776         memory_region_size(&spapr->hotplug_memory.mr);
777     uint32_t lrdr_capacity[] = {
778         cpu_to_be32(max_hotplug_addr >> 32),
779         cpu_to_be32(max_hotplug_addr & 0xffffffff),
780         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
781         cpu_to_be32(max_cpus / smp_threads),
782     };
783 
784     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
785 
786     /* hypertas */
787     add_str(hypertas, "hcall-pft");
788     add_str(hypertas, "hcall-term");
789     add_str(hypertas, "hcall-dabr");
790     add_str(hypertas, "hcall-interrupt");
791     add_str(hypertas, "hcall-tce");
792     add_str(hypertas, "hcall-vio");
793     add_str(hypertas, "hcall-splpar");
794     add_str(hypertas, "hcall-bulk");
795     add_str(hypertas, "hcall-set-mode");
796     add_str(hypertas, "hcall-sprg0");
797     add_str(hypertas, "hcall-copy");
798     add_str(hypertas, "hcall-debug");
799     add_str(qemu_hypertas, "hcall-memop1");
800 
801     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
802         add_str(hypertas, "hcall-multi-tce");
803     }
804     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
805                      hypertas->str, hypertas->len));
806     g_string_free(hypertas, TRUE);
807     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
808                      qemu_hypertas->str, qemu_hypertas->len));
809     g_string_free(qemu_hypertas, TRUE);
810 
811     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
812                      refpoints, sizeof(refpoints)));
813 
814     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
815                           RTAS_ERROR_LOG_MAX));
816     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
817                           RTAS_EVENT_SCAN_RATE));
818 
819     if (msi_nonbroken) {
820         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
821     }
822 
823     /*
824      * According to PAPR, rtas ibm,os-term does not guarantee a return
825      * back to the guest cpu.
826      *
827      * While an additional ibm,extended-os-term property indicates
828      * that rtas call return will always occur. Set this property.
829      */
830     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
831 
832     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
833                      lrdr_capacity, sizeof(lrdr_capacity)));
834 
835     spapr_dt_rtas_tokens(fdt, rtas);
836 }
837 
838 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
839 {
840     MachineState *machine = MACHINE(spapr);
841     int chosen;
842     const char *boot_device = machine->boot_order;
843     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
844     size_t cb = 0;
845     char *bootlist = get_boot_devices_list(&cb, true);
846 
847     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
848 
849     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
850     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
851                           spapr->initrd_base));
852     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
853                           spapr->initrd_base + spapr->initrd_size));
854 
855     if (spapr->kernel_size) {
856         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
857                               cpu_to_be64(spapr->kernel_size) };
858 
859         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
860                          &kprop, sizeof(kprop)));
861         if (spapr->kernel_le) {
862             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
863         }
864     }
865     if (boot_menu) {
866         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
867     }
868     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
869     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
870     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
871 
872     if (cb && bootlist) {
873         int i;
874 
875         for (i = 0; i < cb; i++) {
876             if (bootlist[i] == '\n') {
877                 bootlist[i] = ' ';
878             }
879         }
880         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
881     }
882 
883     if (boot_device && strlen(boot_device)) {
884         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
885     }
886 
887     if (!spapr->has_graphics && stdout_path) {
888         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
889     }
890 
891     g_free(stdout_path);
892     g_free(bootlist);
893 }
894 
895 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
896 {
897     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
898      * KVM to work under pHyp with some guest co-operation */
899     int hypervisor;
900     uint8_t hypercall[16];
901 
902     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
903     /* indicate KVM hypercall interface */
904     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
905     if (kvmppc_has_cap_fixup_hcalls()) {
906         /*
907          * Older KVM versions with older guest kernels were broken
908          * with the magic page, don't allow the guest to map it.
909          */
910         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
911                                   sizeof(hypercall))) {
912             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
913                              hypercall, sizeof(hypercall)));
914         }
915     }
916 }
917 
918 static void *spapr_build_fdt(sPAPRMachineState *spapr,
919                              hwaddr rtas_addr,
920                              hwaddr rtas_size)
921 {
922     MachineState *machine = MACHINE(qdev_get_machine());
923     MachineClass *mc = MACHINE_GET_CLASS(machine);
924     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
925     int ret;
926     void *fdt;
927     sPAPRPHBState *phb;
928     char *buf;
929 
930     fdt = g_malloc0(FDT_MAX_SIZE);
931     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
932 
933     /* Root node */
934     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
935     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
936     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
937 
938     /*
939      * Add info to guest to indentify which host is it being run on
940      * and what is the uuid of the guest
941      */
942     if (kvmppc_get_host_model(&buf)) {
943         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
944         g_free(buf);
945     }
946     if (kvmppc_get_host_serial(&buf)) {
947         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
948         g_free(buf);
949     }
950 
951     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
952 
953     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
954     if (qemu_uuid_set) {
955         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
956     }
957     g_free(buf);
958 
959     if (qemu_get_vm_name()) {
960         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
961                                 qemu_get_vm_name()));
962     }
963 
964     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
965     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
966 
967     /* /interrupt controller */
968     spapr_dt_xics(spapr->xics, fdt, PHANDLE_XICP);
969 
970     ret = spapr_populate_memory(spapr, fdt);
971     if (ret < 0) {
972         error_report("couldn't setup memory nodes in fdt");
973         exit(1);
974     }
975 
976     /* /vdevice */
977     spapr_dt_vdevice(spapr->vio_bus, fdt);
978 
979     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
980         ret = spapr_rng_populate_dt(fdt);
981         if (ret < 0) {
982             error_report("could not set up rng device in the fdt");
983             exit(1);
984         }
985     }
986 
987     QLIST_FOREACH(phb, &spapr->phbs, list) {
988         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
989         if (ret < 0) {
990             error_report("couldn't setup PCI devices in fdt");
991             exit(1);
992         }
993     }
994 
995     /* cpus */
996     spapr_populate_cpus_dt_node(fdt, spapr);
997 
998     if (smc->dr_lmb_enabled) {
999         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1000     }
1001 
1002     if (mc->has_hotpluggable_cpus) {
1003         int offset = fdt_path_offset(fdt, "/cpus");
1004         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1005                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1006         if (ret < 0) {
1007             error_report("Couldn't set up CPU DR device tree properties");
1008             exit(1);
1009         }
1010     }
1011 
1012     /* /event-sources */
1013     spapr_dt_events(spapr, fdt);
1014 
1015     /* /rtas */
1016     spapr_dt_rtas(spapr, fdt);
1017 
1018     /* /chosen */
1019     spapr_dt_chosen(spapr, fdt);
1020 
1021     /* /hypervisor */
1022     if (kvm_enabled()) {
1023         spapr_dt_hypervisor(spapr, fdt);
1024     }
1025 
1026     /* Build memory reserve map */
1027     if (spapr->kernel_size) {
1028         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1029     }
1030     if (spapr->initrd_size) {
1031         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1032     }
1033 
1034     /* ibm,client-architecture-support updates */
1035     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1036     if (ret < 0) {
1037         error_report("couldn't setup CAS properties fdt");
1038         exit(1);
1039     }
1040 
1041     return fdt;
1042 }
1043 
1044 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1045 {
1046     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1047 }
1048 
1049 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1050                                     PowerPCCPU *cpu)
1051 {
1052     CPUPPCState *env = &cpu->env;
1053 
1054     /* The TCG path should also be holding the BQL at this point */
1055     g_assert(qemu_mutex_iothread_locked());
1056 
1057     if (msr_pr) {
1058         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1059         env->gpr[3] = H_PRIVILEGE;
1060     } else {
1061         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1062     }
1063 }
1064 
1065 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1066 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1067 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1068 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1069 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1070 
1071 /*
1072  * Get the fd to access the kernel htab, re-opening it if necessary
1073  */
1074 static int get_htab_fd(sPAPRMachineState *spapr)
1075 {
1076     if (spapr->htab_fd >= 0) {
1077         return spapr->htab_fd;
1078     }
1079 
1080     spapr->htab_fd = kvmppc_get_htab_fd(false);
1081     if (spapr->htab_fd < 0) {
1082         error_report("Unable to open fd for reading hash table from KVM: %s",
1083                      strerror(errno));
1084     }
1085 
1086     return spapr->htab_fd;
1087 }
1088 
1089 static void close_htab_fd(sPAPRMachineState *spapr)
1090 {
1091     if (spapr->htab_fd >= 0) {
1092         close(spapr->htab_fd);
1093     }
1094     spapr->htab_fd = -1;
1095 }
1096 
1097 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1098 {
1099     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1100 
1101     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1102 }
1103 
1104 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1105                                                 hwaddr ptex, int n)
1106 {
1107     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1108     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1109 
1110     if (!spapr->htab) {
1111         /*
1112          * HTAB is controlled by KVM. Fetch into temporary buffer
1113          */
1114         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1115         kvmppc_read_hptes(hptes, ptex, n);
1116         return hptes;
1117     }
1118 
1119     /*
1120      * HTAB is controlled by QEMU. Just point to the internally
1121      * accessible PTEG.
1122      */
1123     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1124 }
1125 
1126 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1127                               const ppc_hash_pte64_t *hptes,
1128                               hwaddr ptex, int n)
1129 {
1130     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1131 
1132     if (!spapr->htab) {
1133         g_free((void *)hptes);
1134     }
1135 
1136     /* Nothing to do for qemu managed HPT */
1137 }
1138 
1139 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1140                              uint64_t pte0, uint64_t pte1)
1141 {
1142     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1143     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1144 
1145     if (!spapr->htab) {
1146         kvmppc_write_hpte(ptex, pte0, pte1);
1147     } else {
1148         stq_p(spapr->htab + offset, pte0);
1149         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1150     }
1151 }
1152 
1153 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1154 {
1155     int shift;
1156 
1157     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1158      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1159      * that's much more than is needed for Linux guests */
1160     shift = ctz64(pow2ceil(ramsize)) - 7;
1161     shift = MAX(shift, 18); /* Minimum architected size */
1162     shift = MIN(shift, 46); /* Maximum architected size */
1163     return shift;
1164 }
1165 
1166 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1167                                  Error **errp)
1168 {
1169     long rc;
1170 
1171     /* Clean up any HPT info from a previous boot */
1172     g_free(spapr->htab);
1173     spapr->htab = NULL;
1174     spapr->htab_shift = 0;
1175     close_htab_fd(spapr);
1176 
1177     rc = kvmppc_reset_htab(shift);
1178     if (rc < 0) {
1179         /* kernel-side HPT needed, but couldn't allocate one */
1180         error_setg_errno(errp, errno,
1181                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1182                          shift);
1183         /* This is almost certainly fatal, but if the caller really
1184          * wants to carry on with shift == 0, it's welcome to try */
1185     } else if (rc > 0) {
1186         /* kernel-side HPT allocated */
1187         if (rc != shift) {
1188             error_setg(errp,
1189                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1190                        shift, rc);
1191         }
1192 
1193         spapr->htab_shift = shift;
1194         spapr->htab = NULL;
1195     } else {
1196         /* kernel-side HPT not needed, allocate in userspace instead */
1197         size_t size = 1ULL << shift;
1198         int i;
1199 
1200         spapr->htab = qemu_memalign(size, size);
1201         if (!spapr->htab) {
1202             error_setg_errno(errp, errno,
1203                              "Could not allocate HPT of order %d", shift);
1204             return;
1205         }
1206 
1207         memset(spapr->htab, 0, size);
1208         spapr->htab_shift = shift;
1209 
1210         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1211             DIRTY_HPTE(HPTE(spapr->htab, i));
1212         }
1213     }
1214 }
1215 
1216 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1217 {
1218     bool matched = false;
1219 
1220     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1221         matched = true;
1222     }
1223 
1224     if (!matched) {
1225         error_report("Device %s is not supported by this machine yet.",
1226                      qdev_fw_name(DEVICE(sbdev)));
1227         exit(1);
1228     }
1229 }
1230 
1231 static void ppc_spapr_reset(void)
1232 {
1233     MachineState *machine = MACHINE(qdev_get_machine());
1234     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1235     PowerPCCPU *first_ppc_cpu;
1236     uint32_t rtas_limit;
1237     hwaddr rtas_addr, fdt_addr;
1238     void *fdt;
1239     int rc;
1240 
1241     /* Check for unknown sysbus devices */
1242     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1243 
1244     /* Allocate and/or reset the hash page table */
1245     spapr_reallocate_hpt(spapr,
1246                          spapr_hpt_shift_for_ramsize(machine->maxram_size),
1247                          &error_fatal);
1248 
1249     /* Update the RMA size if necessary */
1250     if (spapr->vrma_adjust) {
1251         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1252                                           spapr->htab_shift);
1253     }
1254 
1255     qemu_devices_reset();
1256 
1257     /*
1258      * We place the device tree and RTAS just below either the top of the RMA,
1259      * or just below 2GB, whichever is lowere, so that it can be
1260      * processed with 32-bit real mode code if necessary
1261      */
1262     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1263     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1264     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1265 
1266     /* if this reset wasn't generated by CAS, we should reset our
1267      * negotiated options and start from scratch */
1268     if (!spapr->cas_reboot) {
1269         spapr_ovec_cleanup(spapr->ov5_cas);
1270         spapr->ov5_cas = spapr_ovec_new();
1271     }
1272 
1273     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1274 
1275     spapr_load_rtas(spapr, fdt, rtas_addr);
1276 
1277     rc = fdt_pack(fdt);
1278 
1279     /* Should only fail if we've built a corrupted tree */
1280     assert(rc == 0);
1281 
1282     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1283         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1284                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1285         exit(1);
1286     }
1287 
1288     /* Load the fdt */
1289     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1290     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1291     g_free(fdt);
1292 
1293     /* Set up the entry state */
1294     first_ppc_cpu = POWERPC_CPU(first_cpu);
1295     first_ppc_cpu->env.gpr[3] = fdt_addr;
1296     first_ppc_cpu->env.gpr[5] = 0;
1297     first_cpu->halted = 0;
1298     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1299 
1300     spapr->cas_reboot = false;
1301 }
1302 
1303 static void spapr_create_nvram(sPAPRMachineState *spapr)
1304 {
1305     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1306     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1307 
1308     if (dinfo) {
1309         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1310                             &error_fatal);
1311     }
1312 
1313     qdev_init_nofail(dev);
1314 
1315     spapr->nvram = (struct sPAPRNVRAM *)dev;
1316 }
1317 
1318 static void spapr_rtc_create(sPAPRMachineState *spapr)
1319 {
1320     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1321 
1322     qdev_init_nofail(dev);
1323     spapr->rtc = dev;
1324 
1325     object_property_add_alias(qdev_get_machine(), "rtc-time",
1326                               OBJECT(spapr->rtc), "date", NULL);
1327 }
1328 
1329 /* Returns whether we want to use VGA or not */
1330 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1331 {
1332     switch (vga_interface_type) {
1333     case VGA_NONE:
1334         return false;
1335     case VGA_DEVICE:
1336         return true;
1337     case VGA_STD:
1338     case VGA_VIRTIO:
1339         return pci_vga_init(pci_bus) != NULL;
1340     default:
1341         error_setg(errp,
1342                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1343         return false;
1344     }
1345 }
1346 
1347 static int spapr_post_load(void *opaque, int version_id)
1348 {
1349     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1350     int err = 0;
1351 
1352     /* In earlier versions, there was no separate qdev for the PAPR
1353      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1354      * So when migrating from those versions, poke the incoming offset
1355      * value into the RTC device */
1356     if (version_id < 3) {
1357         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1358     }
1359 
1360     return err;
1361 }
1362 
1363 static bool version_before_3(void *opaque, int version_id)
1364 {
1365     return version_id < 3;
1366 }
1367 
1368 static bool spapr_ov5_cas_needed(void *opaque)
1369 {
1370     sPAPRMachineState *spapr = opaque;
1371     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1372     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1373     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1374     bool cas_needed;
1375 
1376     /* Prior to the introduction of sPAPROptionVector, we had two option
1377      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1378      * Both of these options encode machine topology into the device-tree
1379      * in such a way that the now-booted OS should still be able to interact
1380      * appropriately with QEMU regardless of what options were actually
1381      * negotiatied on the source side.
1382      *
1383      * As such, we can avoid migrating the CAS-negotiated options if these
1384      * are the only options available on the current machine/platform.
1385      * Since these are the only options available for pseries-2.7 and
1386      * earlier, this allows us to maintain old->new/new->old migration
1387      * compatibility.
1388      *
1389      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1390      * via default pseries-2.8 machines and explicit command-line parameters.
1391      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1392      * of the actual CAS-negotiated values to continue working properly. For
1393      * example, availability of memory unplug depends on knowing whether
1394      * OV5_HP_EVT was negotiated via CAS.
1395      *
1396      * Thus, for any cases where the set of available CAS-negotiatable
1397      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1398      * include the CAS-negotiated options in the migration stream.
1399      */
1400     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1401     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1402 
1403     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1404      * the mask itself since in the future it's possible "legacy" bits may be
1405      * removed via machine options, which could generate a false positive
1406      * that breaks migration.
1407      */
1408     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1409     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1410 
1411     spapr_ovec_cleanup(ov5_mask);
1412     spapr_ovec_cleanup(ov5_legacy);
1413     spapr_ovec_cleanup(ov5_removed);
1414 
1415     return cas_needed;
1416 }
1417 
1418 static const VMStateDescription vmstate_spapr_ov5_cas = {
1419     .name = "spapr_option_vector_ov5_cas",
1420     .version_id = 1,
1421     .minimum_version_id = 1,
1422     .needed = spapr_ov5_cas_needed,
1423     .fields = (VMStateField[]) {
1424         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1425                                  vmstate_spapr_ovec, sPAPROptionVector),
1426         VMSTATE_END_OF_LIST()
1427     },
1428 };
1429 
1430 static const VMStateDescription vmstate_spapr = {
1431     .name = "spapr",
1432     .version_id = 3,
1433     .minimum_version_id = 1,
1434     .post_load = spapr_post_load,
1435     .fields = (VMStateField[]) {
1436         /* used to be @next_irq */
1437         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1438 
1439         /* RTC offset */
1440         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1441 
1442         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1443         VMSTATE_END_OF_LIST()
1444     },
1445     .subsections = (const VMStateDescription*[]) {
1446         &vmstate_spapr_ov5_cas,
1447         NULL
1448     }
1449 };
1450 
1451 static int htab_save_setup(QEMUFile *f, void *opaque)
1452 {
1453     sPAPRMachineState *spapr = opaque;
1454 
1455     /* "Iteration" header */
1456     qemu_put_be32(f, spapr->htab_shift);
1457 
1458     if (spapr->htab) {
1459         spapr->htab_save_index = 0;
1460         spapr->htab_first_pass = true;
1461     } else {
1462         assert(kvm_enabled());
1463     }
1464 
1465 
1466     return 0;
1467 }
1468 
1469 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1470                                  int64_t max_ns)
1471 {
1472     bool has_timeout = max_ns != -1;
1473     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1474     int index = spapr->htab_save_index;
1475     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1476 
1477     assert(spapr->htab_first_pass);
1478 
1479     do {
1480         int chunkstart;
1481 
1482         /* Consume invalid HPTEs */
1483         while ((index < htabslots)
1484                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1485             index++;
1486             CLEAN_HPTE(HPTE(spapr->htab, index));
1487         }
1488 
1489         /* Consume valid HPTEs */
1490         chunkstart = index;
1491         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1492                && HPTE_VALID(HPTE(spapr->htab, index))) {
1493             index++;
1494             CLEAN_HPTE(HPTE(spapr->htab, index));
1495         }
1496 
1497         if (index > chunkstart) {
1498             int n_valid = index - chunkstart;
1499 
1500             qemu_put_be32(f, chunkstart);
1501             qemu_put_be16(f, n_valid);
1502             qemu_put_be16(f, 0);
1503             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1504                             HASH_PTE_SIZE_64 * n_valid);
1505 
1506             if (has_timeout &&
1507                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1508                 break;
1509             }
1510         }
1511     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1512 
1513     if (index >= htabslots) {
1514         assert(index == htabslots);
1515         index = 0;
1516         spapr->htab_first_pass = false;
1517     }
1518     spapr->htab_save_index = index;
1519 }
1520 
1521 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1522                                 int64_t max_ns)
1523 {
1524     bool final = max_ns < 0;
1525     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1526     int examined = 0, sent = 0;
1527     int index = spapr->htab_save_index;
1528     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1529 
1530     assert(!spapr->htab_first_pass);
1531 
1532     do {
1533         int chunkstart, invalidstart;
1534 
1535         /* Consume non-dirty HPTEs */
1536         while ((index < htabslots)
1537                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1538             index++;
1539             examined++;
1540         }
1541 
1542         chunkstart = index;
1543         /* Consume valid dirty HPTEs */
1544         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1545                && HPTE_DIRTY(HPTE(spapr->htab, index))
1546                && HPTE_VALID(HPTE(spapr->htab, index))) {
1547             CLEAN_HPTE(HPTE(spapr->htab, index));
1548             index++;
1549             examined++;
1550         }
1551 
1552         invalidstart = index;
1553         /* Consume invalid dirty HPTEs */
1554         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1555                && HPTE_DIRTY(HPTE(spapr->htab, index))
1556                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1557             CLEAN_HPTE(HPTE(spapr->htab, index));
1558             index++;
1559             examined++;
1560         }
1561 
1562         if (index > chunkstart) {
1563             int n_valid = invalidstart - chunkstart;
1564             int n_invalid = index - invalidstart;
1565 
1566             qemu_put_be32(f, chunkstart);
1567             qemu_put_be16(f, n_valid);
1568             qemu_put_be16(f, n_invalid);
1569             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1570                             HASH_PTE_SIZE_64 * n_valid);
1571             sent += index - chunkstart;
1572 
1573             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1574                 break;
1575             }
1576         }
1577 
1578         if (examined >= htabslots) {
1579             break;
1580         }
1581 
1582         if (index >= htabslots) {
1583             assert(index == htabslots);
1584             index = 0;
1585         }
1586     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1587 
1588     if (index >= htabslots) {
1589         assert(index == htabslots);
1590         index = 0;
1591     }
1592 
1593     spapr->htab_save_index = index;
1594 
1595     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1596 }
1597 
1598 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1599 #define MAX_KVM_BUF_SIZE    2048
1600 
1601 static int htab_save_iterate(QEMUFile *f, void *opaque)
1602 {
1603     sPAPRMachineState *spapr = opaque;
1604     int fd;
1605     int rc = 0;
1606 
1607     /* Iteration header */
1608     qemu_put_be32(f, 0);
1609 
1610     if (!spapr->htab) {
1611         assert(kvm_enabled());
1612 
1613         fd = get_htab_fd(spapr);
1614         if (fd < 0) {
1615             return fd;
1616         }
1617 
1618         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1619         if (rc < 0) {
1620             return rc;
1621         }
1622     } else  if (spapr->htab_first_pass) {
1623         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1624     } else {
1625         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1626     }
1627 
1628     /* End marker */
1629     qemu_put_be32(f, 0);
1630     qemu_put_be16(f, 0);
1631     qemu_put_be16(f, 0);
1632 
1633     return rc;
1634 }
1635 
1636 static int htab_save_complete(QEMUFile *f, void *opaque)
1637 {
1638     sPAPRMachineState *spapr = opaque;
1639     int fd;
1640 
1641     /* Iteration header */
1642     qemu_put_be32(f, 0);
1643 
1644     if (!spapr->htab) {
1645         int rc;
1646 
1647         assert(kvm_enabled());
1648 
1649         fd = get_htab_fd(spapr);
1650         if (fd < 0) {
1651             return fd;
1652         }
1653 
1654         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1655         if (rc < 0) {
1656             return rc;
1657         }
1658     } else {
1659         if (spapr->htab_first_pass) {
1660             htab_save_first_pass(f, spapr, -1);
1661         }
1662         htab_save_later_pass(f, spapr, -1);
1663     }
1664 
1665     /* End marker */
1666     qemu_put_be32(f, 0);
1667     qemu_put_be16(f, 0);
1668     qemu_put_be16(f, 0);
1669 
1670     return 0;
1671 }
1672 
1673 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1674 {
1675     sPAPRMachineState *spapr = opaque;
1676     uint32_t section_hdr;
1677     int fd = -1;
1678 
1679     if (version_id < 1 || version_id > 1) {
1680         error_report("htab_load() bad version");
1681         return -EINVAL;
1682     }
1683 
1684     section_hdr = qemu_get_be32(f);
1685 
1686     if (section_hdr) {
1687         Error *local_err = NULL;
1688 
1689         /* First section gives the htab size */
1690         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1691         if (local_err) {
1692             error_report_err(local_err);
1693             return -EINVAL;
1694         }
1695         return 0;
1696     }
1697 
1698     if (!spapr->htab) {
1699         assert(kvm_enabled());
1700 
1701         fd = kvmppc_get_htab_fd(true);
1702         if (fd < 0) {
1703             error_report("Unable to open fd to restore KVM hash table: %s",
1704                          strerror(errno));
1705         }
1706     }
1707 
1708     while (true) {
1709         uint32_t index;
1710         uint16_t n_valid, n_invalid;
1711 
1712         index = qemu_get_be32(f);
1713         n_valid = qemu_get_be16(f);
1714         n_invalid = qemu_get_be16(f);
1715 
1716         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1717             /* End of Stream */
1718             break;
1719         }
1720 
1721         if ((index + n_valid + n_invalid) >
1722             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1723             /* Bad index in stream */
1724             error_report(
1725                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1726                 index, n_valid, n_invalid, spapr->htab_shift);
1727             return -EINVAL;
1728         }
1729 
1730         if (spapr->htab) {
1731             if (n_valid) {
1732                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1733                                 HASH_PTE_SIZE_64 * n_valid);
1734             }
1735             if (n_invalid) {
1736                 memset(HPTE(spapr->htab, index + n_valid), 0,
1737                        HASH_PTE_SIZE_64 * n_invalid);
1738             }
1739         } else {
1740             int rc;
1741 
1742             assert(fd >= 0);
1743 
1744             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1745             if (rc < 0) {
1746                 return rc;
1747             }
1748         }
1749     }
1750 
1751     if (!spapr->htab) {
1752         assert(fd >= 0);
1753         close(fd);
1754     }
1755 
1756     return 0;
1757 }
1758 
1759 static void htab_cleanup(void *opaque)
1760 {
1761     sPAPRMachineState *spapr = opaque;
1762 
1763     close_htab_fd(spapr);
1764 }
1765 
1766 static SaveVMHandlers savevm_htab_handlers = {
1767     .save_live_setup = htab_save_setup,
1768     .save_live_iterate = htab_save_iterate,
1769     .save_live_complete_precopy = htab_save_complete,
1770     .cleanup = htab_cleanup,
1771     .load_state = htab_load,
1772 };
1773 
1774 static void spapr_boot_set(void *opaque, const char *boot_device,
1775                            Error **errp)
1776 {
1777     MachineState *machine = MACHINE(qdev_get_machine());
1778     machine->boot_order = g_strdup(boot_device);
1779 }
1780 
1781 /*
1782  * Reset routine for LMB DR devices.
1783  *
1784  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1785  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1786  * when it walks all its children devices. LMB devices reset occurs
1787  * as part of spapr_ppc_reset().
1788  */
1789 static void spapr_drc_reset(void *opaque)
1790 {
1791     sPAPRDRConnector *drc = opaque;
1792     DeviceState *d = DEVICE(drc);
1793 
1794     if (d) {
1795         device_reset(d);
1796     }
1797 }
1798 
1799 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1800 {
1801     MachineState *machine = MACHINE(spapr);
1802     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1803     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1804     int i;
1805 
1806     for (i = 0; i < nr_lmbs; i++) {
1807         sPAPRDRConnector *drc;
1808         uint64_t addr;
1809 
1810         addr = i * lmb_size + spapr->hotplug_memory.base;
1811         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1812                                      addr/lmb_size);
1813         qemu_register_reset(spapr_drc_reset, drc);
1814     }
1815 }
1816 
1817 /*
1818  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1819  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1820  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1821  */
1822 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1823 {
1824     int i;
1825 
1826     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1827         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1828                    " is not aligned to %llu MiB",
1829                    machine->ram_size,
1830                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1831         return;
1832     }
1833 
1834     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1835         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1836                    " is not aligned to %llu MiB",
1837                    machine->ram_size,
1838                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1839         return;
1840     }
1841 
1842     for (i = 0; i < nb_numa_nodes; i++) {
1843         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1844             error_setg(errp,
1845                        "Node %d memory size 0x%" PRIx64
1846                        " is not aligned to %llu MiB",
1847                        i, numa_info[i].node_mem,
1848                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1849             return;
1850         }
1851     }
1852 }
1853 
1854 /* find cpu slot in machine->possible_cpus by core_id */
1855 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1856 {
1857     int index = id / smp_threads;
1858 
1859     if (index >= ms->possible_cpus->len) {
1860         return NULL;
1861     }
1862     if (idx) {
1863         *idx = index;
1864     }
1865     return &ms->possible_cpus->cpus[index];
1866 }
1867 
1868 static void spapr_init_cpus(sPAPRMachineState *spapr)
1869 {
1870     MachineState *machine = MACHINE(spapr);
1871     MachineClass *mc = MACHINE_GET_CLASS(machine);
1872     char *type = spapr_get_cpu_core_type(machine->cpu_model);
1873     int smt = kvmppc_smt_threads();
1874     const CPUArchIdList *possible_cpus;
1875     int boot_cores_nr = smp_cpus / smp_threads;
1876     int i;
1877 
1878     if (!type) {
1879         error_report("Unable to find sPAPR CPU Core definition");
1880         exit(1);
1881     }
1882 
1883     possible_cpus = mc->possible_cpu_arch_ids(machine);
1884     if (mc->has_hotpluggable_cpus) {
1885         if (smp_cpus % smp_threads) {
1886             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1887                          smp_cpus, smp_threads);
1888             exit(1);
1889         }
1890         if (max_cpus % smp_threads) {
1891             error_report("max_cpus (%u) must be multiple of threads (%u)",
1892                          max_cpus, smp_threads);
1893             exit(1);
1894         }
1895     } else {
1896         if (max_cpus != smp_cpus) {
1897             error_report("This machine version does not support CPU hotplug");
1898             exit(1);
1899         }
1900         boot_cores_nr = possible_cpus->len;
1901     }
1902 
1903     for (i = 0; i < possible_cpus->len; i++) {
1904         int core_id = i * smp_threads;
1905 
1906         if (mc->has_hotpluggable_cpus) {
1907             sPAPRDRConnector *drc =
1908                 spapr_dr_connector_new(OBJECT(spapr),
1909                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
1910                                        (core_id / smp_threads) * smt);
1911 
1912             qemu_register_reset(spapr_drc_reset, drc);
1913         }
1914 
1915         if (i < boot_cores_nr) {
1916             Object *core  = object_new(type);
1917             int nr_threads = smp_threads;
1918 
1919             /* Handle the partially filled core for older machine types */
1920             if ((i + 1) * smp_threads >= smp_cpus) {
1921                 nr_threads = smp_cpus - i * smp_threads;
1922             }
1923 
1924             object_property_set_int(core, nr_threads, "nr-threads",
1925                                     &error_fatal);
1926             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1927                                     &error_fatal);
1928             object_property_set_bool(core, true, "realized", &error_fatal);
1929         }
1930     }
1931     g_free(type);
1932 }
1933 
1934 /* pSeries LPAR / sPAPR hardware init */
1935 static void ppc_spapr_init(MachineState *machine)
1936 {
1937     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1938     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1939     const char *kernel_filename = machine->kernel_filename;
1940     const char *initrd_filename = machine->initrd_filename;
1941     PCIHostState *phb;
1942     int i;
1943     MemoryRegion *sysmem = get_system_memory();
1944     MemoryRegion *ram = g_new(MemoryRegion, 1);
1945     MemoryRegion *rma_region;
1946     void *rma = NULL;
1947     hwaddr rma_alloc_size;
1948     hwaddr node0_size = spapr_node0_size();
1949     long load_limit, fw_size;
1950     char *filename;
1951     int smt = kvmppc_smt_threads();
1952 
1953     msi_nonbroken = true;
1954 
1955     QLIST_INIT(&spapr->phbs);
1956 
1957     /* Allocate RMA if necessary */
1958     rma_alloc_size = kvmppc_alloc_rma(&rma);
1959 
1960     if (rma_alloc_size == -1) {
1961         error_report("Unable to create RMA");
1962         exit(1);
1963     }
1964 
1965     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1966         spapr->rma_size = rma_alloc_size;
1967     } else {
1968         spapr->rma_size = node0_size;
1969 
1970         /* With KVM, we don't actually know whether KVM supports an
1971          * unbounded RMA (PR KVM) or is limited by the hash table size
1972          * (HV KVM using VRMA), so we always assume the latter
1973          *
1974          * In that case, we also limit the initial allocations for RTAS
1975          * etc... to 256M since we have no way to know what the VRMA size
1976          * is going to be as it depends on the size of the hash table
1977          * isn't determined yet.
1978          */
1979         if (kvm_enabled()) {
1980             spapr->vrma_adjust = 1;
1981             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1982         }
1983 
1984         /* Actually we don't support unbounded RMA anymore since we
1985          * added proper emulation of HV mode. The max we can get is
1986          * 16G which also happens to be what we configure for PAPR
1987          * mode so make sure we don't do anything bigger than that
1988          */
1989         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1990     }
1991 
1992     if (spapr->rma_size > node0_size) {
1993         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1994                      spapr->rma_size);
1995         exit(1);
1996     }
1997 
1998     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1999     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2000 
2001     /* Set up Interrupt Controller before we create the VCPUs */
2002     spapr->xics = xics_system_init(machine,
2003                                    DIV_ROUND_UP(max_cpus * smt, smp_threads),
2004                                    XICS_IRQS_SPAPR, &error_fatal);
2005 
2006     /* Set up containers for ibm,client-set-architecture negotiated options */
2007     spapr->ov5 = spapr_ovec_new();
2008     spapr->ov5_cas = spapr_ovec_new();
2009 
2010     if (smc->dr_lmb_enabled) {
2011         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2012         spapr_validate_node_memory(machine, &error_fatal);
2013     }
2014 
2015     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2016 
2017     /* advertise support for dedicated HP event source to guests */
2018     if (spapr->use_hotplug_event_source) {
2019         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2020     }
2021 
2022     /* init CPUs */
2023     if (machine->cpu_model == NULL) {
2024         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2025     }
2026 
2027     ppc_cpu_parse_features(machine->cpu_model);
2028 
2029     spapr_init_cpus(spapr);
2030 
2031     if (kvm_enabled()) {
2032         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2033         kvmppc_enable_logical_ci_hcalls();
2034         kvmppc_enable_set_mode_hcall();
2035 
2036         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2037         kvmppc_enable_clear_ref_mod_hcalls();
2038     }
2039 
2040     /* allocate RAM */
2041     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2042                                          machine->ram_size);
2043     memory_region_add_subregion(sysmem, 0, ram);
2044 
2045     if (rma_alloc_size && rma) {
2046         rma_region = g_new(MemoryRegion, 1);
2047         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2048                                    rma_alloc_size, rma);
2049         vmstate_register_ram_global(rma_region);
2050         memory_region_add_subregion(sysmem, 0, rma_region);
2051     }
2052 
2053     /* initialize hotplug memory address space */
2054     if (machine->ram_size < machine->maxram_size) {
2055         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2056         /*
2057          * Limit the number of hotpluggable memory slots to half the number
2058          * slots that KVM supports, leaving the other half for PCI and other
2059          * devices. However ensure that number of slots doesn't drop below 32.
2060          */
2061         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2062                            SPAPR_MAX_RAM_SLOTS;
2063 
2064         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2065             max_memslots = SPAPR_MAX_RAM_SLOTS;
2066         }
2067         if (machine->ram_slots > max_memslots) {
2068             error_report("Specified number of memory slots %"
2069                          PRIu64" exceeds max supported %d",
2070                          machine->ram_slots, max_memslots);
2071             exit(1);
2072         }
2073 
2074         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2075                                               SPAPR_HOTPLUG_MEM_ALIGN);
2076         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2077                            "hotplug-memory", hotplug_mem_size);
2078         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2079                                     &spapr->hotplug_memory.mr);
2080     }
2081 
2082     if (smc->dr_lmb_enabled) {
2083         spapr_create_lmb_dr_connectors(spapr);
2084     }
2085 
2086     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2087     if (!filename) {
2088         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2089         exit(1);
2090     }
2091     spapr->rtas_size = get_image_size(filename);
2092     if (spapr->rtas_size < 0) {
2093         error_report("Could not get size of LPAR rtas '%s'", filename);
2094         exit(1);
2095     }
2096     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2097     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2098         error_report("Could not load LPAR rtas '%s'", filename);
2099         exit(1);
2100     }
2101     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2102         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2103                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2104         exit(1);
2105     }
2106     g_free(filename);
2107 
2108     /* Set up RTAS event infrastructure */
2109     spapr_events_init(spapr);
2110 
2111     /* Set up the RTC RTAS interfaces */
2112     spapr_rtc_create(spapr);
2113 
2114     /* Set up VIO bus */
2115     spapr->vio_bus = spapr_vio_bus_init();
2116 
2117     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2118         if (serial_hds[i]) {
2119             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2120         }
2121     }
2122 
2123     /* We always have at least the nvram device on VIO */
2124     spapr_create_nvram(spapr);
2125 
2126     /* Set up PCI */
2127     spapr_pci_rtas_init();
2128 
2129     phb = spapr_create_phb(spapr, 0);
2130 
2131     for (i = 0; i < nb_nics; i++) {
2132         NICInfo *nd = &nd_table[i];
2133 
2134         if (!nd->model) {
2135             nd->model = g_strdup("ibmveth");
2136         }
2137 
2138         if (strcmp(nd->model, "ibmveth") == 0) {
2139             spapr_vlan_create(spapr->vio_bus, nd);
2140         } else {
2141             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2142         }
2143     }
2144 
2145     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2146         spapr_vscsi_create(spapr->vio_bus);
2147     }
2148 
2149     /* Graphics */
2150     if (spapr_vga_init(phb->bus, &error_fatal)) {
2151         spapr->has_graphics = true;
2152         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2153     }
2154 
2155     if (machine->usb) {
2156         if (smc->use_ohci_by_default) {
2157             pci_create_simple(phb->bus, -1, "pci-ohci");
2158         } else {
2159             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2160         }
2161 
2162         if (spapr->has_graphics) {
2163             USBBus *usb_bus = usb_bus_find(-1);
2164 
2165             usb_create_simple(usb_bus, "usb-kbd");
2166             usb_create_simple(usb_bus, "usb-mouse");
2167         }
2168     }
2169 
2170     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2171         error_report(
2172             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2173             MIN_RMA_SLOF);
2174         exit(1);
2175     }
2176 
2177     if (kernel_filename) {
2178         uint64_t lowaddr = 0;
2179 
2180         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2181                                       NULL, NULL, &lowaddr, NULL, 1,
2182                                       PPC_ELF_MACHINE, 0, 0);
2183         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2184             spapr->kernel_size = load_elf(kernel_filename,
2185                                           translate_kernel_address, NULL, NULL,
2186                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2187                                           0, 0);
2188             spapr->kernel_le = spapr->kernel_size > 0;
2189         }
2190         if (spapr->kernel_size < 0) {
2191             error_report("error loading %s: %s", kernel_filename,
2192                          load_elf_strerror(spapr->kernel_size));
2193             exit(1);
2194         }
2195 
2196         /* load initrd */
2197         if (initrd_filename) {
2198             /* Try to locate the initrd in the gap between the kernel
2199              * and the firmware. Add a bit of space just in case
2200              */
2201             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2202                                   + 0x1ffff) & ~0xffff;
2203             spapr->initrd_size = load_image_targphys(initrd_filename,
2204                                                      spapr->initrd_base,
2205                                                      load_limit
2206                                                      - spapr->initrd_base);
2207             if (spapr->initrd_size < 0) {
2208                 error_report("could not load initial ram disk '%s'",
2209                              initrd_filename);
2210                 exit(1);
2211             }
2212         }
2213     }
2214 
2215     if (bios_name == NULL) {
2216         bios_name = FW_FILE_NAME;
2217     }
2218     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2219     if (!filename) {
2220         error_report("Could not find LPAR firmware '%s'", bios_name);
2221         exit(1);
2222     }
2223     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2224     if (fw_size <= 0) {
2225         error_report("Could not load LPAR firmware '%s'", filename);
2226         exit(1);
2227     }
2228     g_free(filename);
2229 
2230     /* FIXME: Should register things through the MachineState's qdev
2231      * interface, this is a legacy from the sPAPREnvironment structure
2232      * which predated MachineState but had a similar function */
2233     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2234     register_savevm_live(NULL, "spapr/htab", -1, 1,
2235                          &savevm_htab_handlers, spapr);
2236 
2237     /* used by RTAS */
2238     QTAILQ_INIT(&spapr->ccs_list);
2239     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2240 
2241     qemu_register_boot_set(spapr_boot_set, spapr);
2242 
2243     /* to stop and start vmclock */
2244     if (kvm_enabled()) {
2245         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2246                                          &spapr->tb);
2247     }
2248 }
2249 
2250 static int spapr_kvm_type(const char *vm_type)
2251 {
2252     if (!vm_type) {
2253         return 0;
2254     }
2255 
2256     if (!strcmp(vm_type, "HV")) {
2257         return 1;
2258     }
2259 
2260     if (!strcmp(vm_type, "PR")) {
2261         return 2;
2262     }
2263 
2264     error_report("Unknown kvm-type specified '%s'", vm_type);
2265     exit(1);
2266 }
2267 
2268 /*
2269  * Implementation of an interface to adjust firmware path
2270  * for the bootindex property handling.
2271  */
2272 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2273                                    DeviceState *dev)
2274 {
2275 #define CAST(type, obj, name) \
2276     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2277     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2278     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2279 
2280     if (d) {
2281         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2282         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2283         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2284 
2285         if (spapr) {
2286             /*
2287              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2288              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2289              * in the top 16 bits of the 64-bit LUN
2290              */
2291             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2292             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2293                                    (uint64_t)id << 48);
2294         } else if (virtio) {
2295             /*
2296              * We use SRP luns of the form 01000000 | (target << 8) | lun
2297              * in the top 32 bits of the 64-bit LUN
2298              * Note: the quote above is from SLOF and it is wrong,
2299              * the actual binding is:
2300              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2301              */
2302             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2303             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2304                                    (uint64_t)id << 32);
2305         } else if (usb) {
2306             /*
2307              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2308              * in the top 32 bits of the 64-bit LUN
2309              */
2310             unsigned usb_port = atoi(usb->port->path);
2311             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2312             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2313                                    (uint64_t)id << 32);
2314         }
2315     }
2316 
2317     /*
2318      * SLOF probes the USB devices, and if it recognizes that the device is a
2319      * storage device, it changes its name to "storage" instead of "usb-host",
2320      * and additionally adds a child node for the SCSI LUN, so the correct
2321      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2322      */
2323     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2324         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2325         if (usb_host_dev_is_scsi_storage(usbdev)) {
2326             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2327         }
2328     }
2329 
2330     if (phb) {
2331         /* Replace "pci" with "pci@800000020000000" */
2332         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2333     }
2334 
2335     return NULL;
2336 }
2337 
2338 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2339 {
2340     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2341 
2342     return g_strdup(spapr->kvm_type);
2343 }
2344 
2345 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2346 {
2347     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2348 
2349     g_free(spapr->kvm_type);
2350     spapr->kvm_type = g_strdup(value);
2351 }
2352 
2353 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2354 {
2355     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2356 
2357     return spapr->use_hotplug_event_source;
2358 }
2359 
2360 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2361                                             Error **errp)
2362 {
2363     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2364 
2365     spapr->use_hotplug_event_source = value;
2366 }
2367 
2368 static void spapr_machine_initfn(Object *obj)
2369 {
2370     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2371 
2372     spapr->htab_fd = -1;
2373     spapr->use_hotplug_event_source = true;
2374     object_property_add_str(obj, "kvm-type",
2375                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2376     object_property_set_description(obj, "kvm-type",
2377                                     "Specifies the KVM virtualization mode (HV, PR)",
2378                                     NULL);
2379     object_property_add_bool(obj, "modern-hotplug-events",
2380                             spapr_get_modern_hotplug_events,
2381                             spapr_set_modern_hotplug_events,
2382                             NULL);
2383     object_property_set_description(obj, "modern-hotplug-events",
2384                                     "Use dedicated hotplug event mechanism in"
2385                                     " place of standard EPOW events when possible"
2386                                     " (required for memory hot-unplug support)",
2387                                     NULL);
2388 }
2389 
2390 static void spapr_machine_finalizefn(Object *obj)
2391 {
2392     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2393 
2394     g_free(spapr->kvm_type);
2395 }
2396 
2397 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2398 {
2399     cpu_synchronize_state(cs);
2400     ppc_cpu_do_system_reset(cs);
2401 }
2402 
2403 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2404 {
2405     CPUState *cs;
2406 
2407     CPU_FOREACH(cs) {
2408         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2409     }
2410 }
2411 
2412 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2413                            uint32_t node, bool dedicated_hp_event_source,
2414                            Error **errp)
2415 {
2416     sPAPRDRConnector *drc;
2417     sPAPRDRConnectorClass *drck;
2418     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2419     int i, fdt_offset, fdt_size;
2420     void *fdt;
2421     uint64_t addr = addr_start;
2422 
2423     for (i = 0; i < nr_lmbs; i++) {
2424         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2425                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2426         g_assert(drc);
2427 
2428         fdt = create_device_tree(&fdt_size);
2429         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2430                                                 SPAPR_MEMORY_BLOCK_SIZE);
2431 
2432         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2433         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2434         addr += SPAPR_MEMORY_BLOCK_SIZE;
2435         if (!dev->hotplugged) {
2436             /* guests expect coldplugged LMBs to be pre-allocated */
2437             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2438             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2439         }
2440     }
2441     /* send hotplug notification to the
2442      * guest only in case of hotplugged memory
2443      */
2444     if (dev->hotplugged) {
2445         if (dedicated_hp_event_source) {
2446             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2447                     addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2448             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2449             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2450                                                    nr_lmbs,
2451                                                    drck->get_index(drc));
2452         } else {
2453             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2454                                            nr_lmbs);
2455         }
2456     }
2457 }
2458 
2459 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2460                               uint32_t node, Error **errp)
2461 {
2462     Error *local_err = NULL;
2463     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2464     PCDIMMDevice *dimm = PC_DIMM(dev);
2465     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2466     MemoryRegion *mr = ddc->get_memory_region(dimm);
2467     uint64_t align = memory_region_get_alignment(mr);
2468     uint64_t size = memory_region_size(mr);
2469     uint64_t addr;
2470     char *mem_dev;
2471 
2472     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2473         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2474                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2475         goto out;
2476     }
2477 
2478     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2479     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2480         error_setg(&local_err, "Memory backend has bad page size. "
2481                    "Use 'memory-backend-file' with correct mem-path.");
2482         goto out;
2483     }
2484 
2485     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2486     if (local_err) {
2487         goto out;
2488     }
2489 
2490     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2491     if (local_err) {
2492         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2493         goto out;
2494     }
2495 
2496     spapr_add_lmbs(dev, addr, size, node,
2497                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2498                    &error_abort);
2499 
2500 out:
2501     error_propagate(errp, local_err);
2502 }
2503 
2504 typedef struct sPAPRDIMMState {
2505     uint32_t nr_lmbs;
2506 } sPAPRDIMMState;
2507 
2508 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2509 {
2510     sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2511     HotplugHandler *hotplug_ctrl;
2512 
2513     if (--ds->nr_lmbs) {
2514         return;
2515     }
2516 
2517     g_free(ds);
2518 
2519     /*
2520      * Now that all the LMBs have been removed by the guest, call the
2521      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2522      */
2523     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2524     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2525 }
2526 
2527 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2528                            Error **errp)
2529 {
2530     sPAPRDRConnector *drc;
2531     sPAPRDRConnectorClass *drck;
2532     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2533     int i;
2534     sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2535     uint64_t addr = addr_start;
2536 
2537     ds->nr_lmbs = nr_lmbs;
2538     for (i = 0; i < nr_lmbs; i++) {
2539         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2540                 addr / SPAPR_MEMORY_BLOCK_SIZE);
2541         g_assert(drc);
2542 
2543         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2544         drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2545         addr += SPAPR_MEMORY_BLOCK_SIZE;
2546     }
2547 
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_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2552                                               nr_lmbs,
2553                                               drck->get_index(drc));
2554 }
2555 
2556 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2557                                 Error **errp)
2558 {
2559     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2560     PCDIMMDevice *dimm = PC_DIMM(dev);
2561     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2562     MemoryRegion *mr = ddc->get_memory_region(dimm);
2563 
2564     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2565     object_unparent(OBJECT(dev));
2566 }
2567 
2568 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2569                                         DeviceState *dev, Error **errp)
2570 {
2571     Error *local_err = NULL;
2572     PCDIMMDevice *dimm = PC_DIMM(dev);
2573     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2574     MemoryRegion *mr = ddc->get_memory_region(dimm);
2575     uint64_t size = memory_region_size(mr);
2576     uint64_t addr;
2577 
2578     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2579     if (local_err) {
2580         goto out;
2581     }
2582 
2583     spapr_del_lmbs(dev, addr, size, &error_abort);
2584 out:
2585     error_propagate(errp, local_err);
2586 }
2587 
2588 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2589                                     sPAPRMachineState *spapr)
2590 {
2591     PowerPCCPU *cpu = POWERPC_CPU(cs);
2592     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2593     int id = ppc_get_vcpu_dt_id(cpu);
2594     void *fdt;
2595     int offset, fdt_size;
2596     char *nodename;
2597 
2598     fdt = create_device_tree(&fdt_size);
2599     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2600     offset = fdt_add_subnode(fdt, 0, nodename);
2601 
2602     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2603     g_free(nodename);
2604 
2605     *fdt_offset = offset;
2606     return fdt;
2607 }
2608 
2609 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2610                               Error **errp)
2611 {
2612     MachineState *ms = MACHINE(qdev_get_machine());
2613     CPUCore *cc = CPU_CORE(dev);
2614     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2615 
2616     core_slot->cpu = NULL;
2617     object_unparent(OBJECT(dev));
2618 }
2619 
2620 static void spapr_core_release(DeviceState *dev, void *opaque)
2621 {
2622     HotplugHandler *hotplug_ctrl;
2623 
2624     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2625     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2626 }
2627 
2628 static
2629 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2630                                Error **errp)
2631 {
2632     int index;
2633     sPAPRDRConnector *drc;
2634     sPAPRDRConnectorClass *drck;
2635     Error *local_err = NULL;
2636     CPUCore *cc = CPU_CORE(dev);
2637     int smt = kvmppc_smt_threads();
2638 
2639     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2640         error_setg(errp, "Unable to find CPU core with core-id: %d",
2641                    cc->core_id);
2642         return;
2643     }
2644     if (index == 0) {
2645         error_setg(errp, "Boot CPU core may not be unplugged");
2646         return;
2647     }
2648 
2649     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2650     g_assert(drc);
2651 
2652     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2653     drck->detach(drc, dev, spapr_core_release, NULL, &local_err);
2654     if (local_err) {
2655         error_propagate(errp, local_err);
2656         return;
2657     }
2658 
2659     spapr_hotplug_req_remove_by_index(drc);
2660 }
2661 
2662 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2663                             Error **errp)
2664 {
2665     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2666     MachineClass *mc = MACHINE_GET_CLASS(spapr);
2667     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2668     CPUCore *cc = CPU_CORE(dev);
2669     CPUState *cs = CPU(core->threads);
2670     sPAPRDRConnector *drc;
2671     Error *local_err = NULL;
2672     void *fdt = NULL;
2673     int fdt_offset = 0;
2674     int smt = kvmppc_smt_threads();
2675     CPUArchId *core_slot;
2676     int index;
2677 
2678     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2679     if (!core_slot) {
2680         error_setg(errp, "Unable to find CPU core with core-id: %d",
2681                    cc->core_id);
2682         return;
2683     }
2684     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2685 
2686     g_assert(drc || !mc->has_hotpluggable_cpus);
2687 
2688     /*
2689      * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2690      * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2691      */
2692     if (dev->hotplugged) {
2693         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2694     }
2695 
2696     if (drc) {
2697         sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2698         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
2699         if (local_err) {
2700             g_free(fdt);
2701             error_propagate(errp, local_err);
2702             return;
2703         }
2704     }
2705 
2706     if (dev->hotplugged) {
2707         /*
2708          * Send hotplug notification interrupt to the guest only in case
2709          * of hotplugged CPUs.
2710          */
2711         spapr_hotplug_req_add_by_index(drc);
2712     } else {
2713         /*
2714          * Set the right DRC states for cold plugged CPU.
2715          */
2716         if (drc) {
2717             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2718             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2719             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2720         }
2721     }
2722     core_slot->cpu = OBJECT(dev);
2723 }
2724 
2725 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2726                                 Error **errp)
2727 {
2728     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2729     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2730     Error *local_err = NULL;
2731     CPUCore *cc = CPU_CORE(dev);
2732     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2733     const char *type = object_get_typename(OBJECT(dev));
2734     CPUArchId *core_slot;
2735     int index;
2736 
2737     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2738         error_setg(&local_err, "CPU hotplug not supported for this machine");
2739         goto out;
2740     }
2741 
2742     if (strcmp(base_core_type, type)) {
2743         error_setg(&local_err, "CPU core type should be %s", base_core_type);
2744         goto out;
2745     }
2746 
2747     if (cc->core_id % smp_threads) {
2748         error_setg(&local_err, "invalid core id %d", cc->core_id);
2749         goto out;
2750     }
2751 
2752     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2753     if (!core_slot) {
2754         error_setg(&local_err, "core id %d out of range", cc->core_id);
2755         goto out;
2756     }
2757 
2758     if (core_slot->cpu) {
2759         error_setg(&local_err, "core %d already populated", cc->core_id);
2760         goto out;
2761     }
2762 
2763 out:
2764     g_free(base_core_type);
2765     error_propagate(errp, local_err);
2766 }
2767 
2768 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2769                                       DeviceState *dev, Error **errp)
2770 {
2771     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2772 
2773     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2774         int node;
2775 
2776         if (!smc->dr_lmb_enabled) {
2777             error_setg(errp, "Memory hotplug not supported for this machine");
2778             return;
2779         }
2780         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2781         if (*errp) {
2782             return;
2783         }
2784         if (node < 0 || node >= MAX_NODES) {
2785             error_setg(errp, "Invaild node %d", node);
2786             return;
2787         }
2788 
2789         /*
2790          * Currently PowerPC kernel doesn't allow hot-adding memory to
2791          * memory-less node, but instead will silently add the memory
2792          * to the first node that has some memory. This causes two
2793          * unexpected behaviours for the user.
2794          *
2795          * - Memory gets hotplugged to a different node than what the user
2796          *   specified.
2797          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2798          *   to memory-less node, a reboot will set things accordingly
2799          *   and the previously hotplugged memory now ends in the right node.
2800          *   This appears as if some memory moved from one node to another.
2801          *
2802          * So until kernel starts supporting memory hotplug to memory-less
2803          * nodes, just prevent such attempts upfront in QEMU.
2804          */
2805         if (nb_numa_nodes && !numa_info[node].node_mem) {
2806             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2807                        node);
2808             return;
2809         }
2810 
2811         spapr_memory_plug(hotplug_dev, dev, node, errp);
2812     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2813         spapr_core_plug(hotplug_dev, dev, errp);
2814     }
2815 }
2816 
2817 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2818                                       DeviceState *dev, Error **errp)
2819 {
2820     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2821     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2822 
2823     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2824         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2825             spapr_memory_unplug(hotplug_dev, dev, errp);
2826         } else {
2827             error_setg(errp, "Memory hot unplug not supported for this guest");
2828         }
2829     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2830         if (!mc->has_hotpluggable_cpus) {
2831             error_setg(errp, "CPU hot unplug not supported on this machine");
2832             return;
2833         }
2834         spapr_core_unplug(hotplug_dev, dev, errp);
2835     }
2836 }
2837 
2838 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2839                                                 DeviceState *dev, Error **errp)
2840 {
2841     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2842     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2843 
2844     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2845         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2846             spapr_memory_unplug_request(hotplug_dev, dev, errp);
2847         } else {
2848             /* NOTE: this means there is a window after guest reset, prior to
2849              * CAS negotiation, where unplug requests will fail due to the
2850              * capability not being detected yet. This is a bit different than
2851              * the case with PCI unplug, where the events will be queued and
2852              * eventually handled by the guest after boot
2853              */
2854             error_setg(errp, "Memory hot unplug not supported for this guest");
2855         }
2856     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2857         if (!mc->has_hotpluggable_cpus) {
2858             error_setg(errp, "CPU hot unplug not supported on this machine");
2859             return;
2860         }
2861         spapr_core_unplug_request(hotplug_dev, dev, errp);
2862     }
2863 }
2864 
2865 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2866                                           DeviceState *dev, Error **errp)
2867 {
2868     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2869         spapr_core_pre_plug(hotplug_dev, dev, errp);
2870     }
2871 }
2872 
2873 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2874                                                  DeviceState *dev)
2875 {
2876     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2877         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2878         return HOTPLUG_HANDLER(machine);
2879     }
2880     return NULL;
2881 }
2882 
2883 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2884 {
2885     /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2886      * socket means much for the paravirtualized PAPR platform) */
2887     return cpu_index / smp_threads / smp_cores;
2888 }
2889 
2890 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
2891 {
2892     int i;
2893     int spapr_max_cores = max_cpus / smp_threads;
2894     MachineClass *mc = MACHINE_GET_CLASS(machine);
2895 
2896     if (!mc->has_hotpluggable_cpus) {
2897         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
2898     }
2899     if (machine->possible_cpus) {
2900         assert(machine->possible_cpus->len == spapr_max_cores);
2901         return machine->possible_cpus;
2902     }
2903 
2904     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2905                              sizeof(CPUArchId) * spapr_max_cores);
2906     machine->possible_cpus->len = spapr_max_cores;
2907     for (i = 0; i < machine->possible_cpus->len; i++) {
2908         int core_id = i * smp_threads;
2909 
2910         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
2911         machine->possible_cpus->cpus[i].arch_id = core_id;
2912         machine->possible_cpus->cpus[i].props.has_core_id = true;
2913         machine->possible_cpus->cpus[i].props.core_id = core_id;
2914         /* TODO: add 'has_node/node' here to describe
2915            to which node core belongs */
2916     }
2917     return machine->possible_cpus;
2918 }
2919 
2920 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
2921                                 uint64_t *buid, hwaddr *pio,
2922                                 hwaddr *mmio32, hwaddr *mmio64,
2923                                 unsigned n_dma, uint32_t *liobns, Error **errp)
2924 {
2925     /*
2926      * New-style PHB window placement.
2927      *
2928      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
2929      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
2930      * windows.
2931      *
2932      * Some guest kernels can't work with MMIO windows above 1<<46
2933      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
2934      *
2935      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
2936      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
2937      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
2938      * 1TiB 64-bit MMIO windows for each PHB.
2939      */
2940     const uint64_t base_buid = 0x800000020000000ULL;
2941 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
2942                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
2943     int i;
2944 
2945     /* Sanity check natural alignments */
2946     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2947     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2948     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
2949     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
2950     /* Sanity check bounds */
2951     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
2952                       SPAPR_PCI_MEM32_WIN_SIZE);
2953     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
2954                       SPAPR_PCI_MEM64_WIN_SIZE);
2955 
2956     if (index >= SPAPR_MAX_PHBS) {
2957         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
2958                    SPAPR_MAX_PHBS - 1);
2959         return;
2960     }
2961 
2962     *buid = base_buid + index;
2963     for (i = 0; i < n_dma; ++i) {
2964         liobns[i] = SPAPR_PCI_LIOBN(index, i);
2965     }
2966 
2967     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
2968     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
2969     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
2970 }
2971 
2972 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
2973 {
2974     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2975 
2976     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
2977 }
2978 
2979 static void spapr_ics_resend(XICSFabric *dev)
2980 {
2981     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2982 
2983     ics_resend(spapr->ics);
2984 }
2985 
2986 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2987 {
2988     MachineClass *mc = MACHINE_CLASS(oc);
2989     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2990     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2991     NMIClass *nc = NMI_CLASS(oc);
2992     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2993     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
2994     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
2995 
2996     mc->desc = "pSeries Logical Partition (PAPR compliant)";
2997 
2998     /*
2999      * We set up the default / latest behaviour here.  The class_init
3000      * functions for the specific versioned machine types can override
3001      * these details for backwards compatibility
3002      */
3003     mc->init = ppc_spapr_init;
3004     mc->reset = ppc_spapr_reset;
3005     mc->block_default_type = IF_SCSI;
3006     mc->max_cpus = 1024;
3007     mc->no_parallel = 1;
3008     mc->default_boot_order = "";
3009     mc->default_ram_size = 512 * M_BYTE;
3010     mc->kvm_type = spapr_kvm_type;
3011     mc->has_dynamic_sysbus = true;
3012     mc->pci_allow_0_address = true;
3013     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3014     hc->pre_plug = spapr_machine_device_pre_plug;
3015     hc->plug = spapr_machine_device_plug;
3016     hc->unplug = spapr_machine_device_unplug;
3017     mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
3018     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3019     hc->unplug_request = spapr_machine_device_unplug_request;
3020 
3021     smc->dr_lmb_enabled = true;
3022     smc->tcg_default_cpu = "POWER8";
3023     mc->has_hotpluggable_cpus = true;
3024     fwc->get_dev_path = spapr_get_fw_dev_path;
3025     nc->nmi_monitor_handler = spapr_nmi;
3026     smc->phb_placement = spapr_phb_placement;
3027     vhc->hypercall = emulate_spapr_hypercall;
3028     vhc->hpt_mask = spapr_hpt_mask;
3029     vhc->map_hptes = spapr_map_hptes;
3030     vhc->unmap_hptes = spapr_unmap_hptes;
3031     vhc->store_hpte = spapr_store_hpte;
3032     xic->ics_get = spapr_ics_get;
3033     xic->ics_resend = spapr_ics_resend;
3034 }
3035 
3036 static const TypeInfo spapr_machine_info = {
3037     .name          = TYPE_SPAPR_MACHINE,
3038     .parent        = TYPE_MACHINE,
3039     .abstract      = true,
3040     .instance_size = sizeof(sPAPRMachineState),
3041     .instance_init = spapr_machine_initfn,
3042     .instance_finalize = spapr_machine_finalizefn,
3043     .class_size    = sizeof(sPAPRMachineClass),
3044     .class_init    = spapr_machine_class_init,
3045     .interfaces = (InterfaceInfo[]) {
3046         { TYPE_FW_PATH_PROVIDER },
3047         { TYPE_NMI },
3048         { TYPE_HOTPLUG_HANDLER },
3049         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3050         { TYPE_XICS_FABRIC },
3051         { }
3052     },
3053 };
3054 
3055 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3056     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3057                                                     void *data)      \
3058     {                                                                \
3059         MachineClass *mc = MACHINE_CLASS(oc);                        \
3060         spapr_machine_##suffix##_class_options(mc);                  \
3061         if (latest) {                                                \
3062             mc->alias = "pseries";                                   \
3063             mc->is_default = 1;                                      \
3064         }                                                            \
3065     }                                                                \
3066     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3067     {                                                                \
3068         MachineState *machine = MACHINE(obj);                        \
3069         spapr_machine_##suffix##_instance_options(machine);          \
3070     }                                                                \
3071     static const TypeInfo spapr_machine_##suffix##_info = {          \
3072         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3073         .parent = TYPE_SPAPR_MACHINE,                                \
3074         .class_init = spapr_machine_##suffix##_class_init,           \
3075         .instance_init = spapr_machine_##suffix##_instance_init,     \
3076     };                                                               \
3077     static void spapr_machine_register_##suffix(void)                \
3078     {                                                                \
3079         type_register(&spapr_machine_##suffix##_info);               \
3080     }                                                                \
3081     type_init(spapr_machine_register_##suffix)
3082 
3083 /*
3084  * pseries-2.9
3085  */
3086 static void spapr_machine_2_9_instance_options(MachineState *machine)
3087 {
3088 }
3089 
3090 static void spapr_machine_2_9_class_options(MachineClass *mc)
3091 {
3092     /* Defaults for the latest behaviour inherited from the base class */
3093 }
3094 
3095 DEFINE_SPAPR_MACHINE(2_9, "2.9", true);
3096 
3097 /*
3098  * pseries-2.8
3099  */
3100 #define SPAPR_COMPAT_2_8                            \
3101     HW_COMPAT_2_8
3102 
3103 static void spapr_machine_2_8_instance_options(MachineState *machine)
3104 {
3105     spapr_machine_2_9_instance_options(machine);
3106 }
3107 
3108 static void spapr_machine_2_8_class_options(MachineClass *mc)
3109 {
3110     spapr_machine_2_9_class_options(mc);
3111     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3112 }
3113 
3114 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3115 
3116 /*
3117  * pseries-2.7
3118  */
3119 #define SPAPR_COMPAT_2_7                            \
3120     HW_COMPAT_2_7                                   \
3121     {                                               \
3122         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3123         .property = "mem_win_size",                 \
3124         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3125     },                                              \
3126     {                                               \
3127         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3128         .property = "mem64_win_size",               \
3129         .value    = "0",                            \
3130     },                                              \
3131     {                                               \
3132         .driver = TYPE_POWERPC_CPU,                 \
3133         .property = "pre-2.8-migration",            \
3134         .value    = "on",                           \
3135     },                                              \
3136     {                                               \
3137         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3138         .property = "pre-2.8-migration",            \
3139         .value    = "on",                           \
3140     },
3141 
3142 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3143                               uint64_t *buid, hwaddr *pio,
3144                               hwaddr *mmio32, hwaddr *mmio64,
3145                               unsigned n_dma, uint32_t *liobns, Error **errp)
3146 {
3147     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3148     const uint64_t base_buid = 0x800000020000000ULL;
3149     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3150     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3151     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3152     const uint32_t max_index = 255;
3153     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3154 
3155     uint64_t ram_top = MACHINE(spapr)->ram_size;
3156     hwaddr phb0_base, phb_base;
3157     int i;
3158 
3159     /* Do we have hotpluggable memory? */
3160     if (MACHINE(spapr)->maxram_size > ram_top) {
3161         /* Can't just use maxram_size, because there may be an
3162          * alignment gap between normal and hotpluggable memory
3163          * regions */
3164         ram_top = spapr->hotplug_memory.base +
3165             memory_region_size(&spapr->hotplug_memory.mr);
3166     }
3167 
3168     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3169 
3170     if (index > max_index) {
3171         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3172                    max_index);
3173         return;
3174     }
3175 
3176     *buid = base_buid + index;
3177     for (i = 0; i < n_dma; ++i) {
3178         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3179     }
3180 
3181     phb_base = phb0_base + index * phb_spacing;
3182     *pio = phb_base + pio_offset;
3183     *mmio32 = phb_base + mmio_offset;
3184     /*
3185      * We don't set the 64-bit MMIO window, relying on the PHB's
3186      * fallback behaviour of automatically splitting a large "32-bit"
3187      * window into contiguous 32-bit and 64-bit windows
3188      */
3189 }
3190 
3191 static void spapr_machine_2_7_instance_options(MachineState *machine)
3192 {
3193     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3194 
3195     spapr_machine_2_8_instance_options(machine);
3196     spapr->use_hotplug_event_source = false;
3197 }
3198 
3199 static void spapr_machine_2_7_class_options(MachineClass *mc)
3200 {
3201     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3202 
3203     spapr_machine_2_8_class_options(mc);
3204     smc->tcg_default_cpu = "POWER7";
3205     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3206     smc->phb_placement = phb_placement_2_7;
3207 }
3208 
3209 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3210 
3211 /*
3212  * pseries-2.6
3213  */
3214 #define SPAPR_COMPAT_2_6 \
3215     HW_COMPAT_2_6 \
3216     { \
3217         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3218         .property = "ddw",\
3219         .value    = stringify(off),\
3220     },
3221 
3222 static void spapr_machine_2_6_instance_options(MachineState *machine)
3223 {
3224     spapr_machine_2_7_instance_options(machine);
3225 }
3226 
3227 static void spapr_machine_2_6_class_options(MachineClass *mc)
3228 {
3229     spapr_machine_2_7_class_options(mc);
3230     mc->has_hotpluggable_cpus = false;
3231     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3232 }
3233 
3234 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3235 
3236 /*
3237  * pseries-2.5
3238  */
3239 #define SPAPR_COMPAT_2_5 \
3240     HW_COMPAT_2_5 \
3241     { \
3242         .driver   = "spapr-vlan", \
3243         .property = "use-rx-buffer-pools", \
3244         .value    = "off", \
3245     },
3246 
3247 static void spapr_machine_2_5_instance_options(MachineState *machine)
3248 {
3249     spapr_machine_2_6_instance_options(machine);
3250 }
3251 
3252 static void spapr_machine_2_5_class_options(MachineClass *mc)
3253 {
3254     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3255 
3256     spapr_machine_2_6_class_options(mc);
3257     smc->use_ohci_by_default = true;
3258     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3259 }
3260 
3261 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3262 
3263 /*
3264  * pseries-2.4
3265  */
3266 #define SPAPR_COMPAT_2_4 \
3267         HW_COMPAT_2_4
3268 
3269 static void spapr_machine_2_4_instance_options(MachineState *machine)
3270 {
3271     spapr_machine_2_5_instance_options(machine);
3272 }
3273 
3274 static void spapr_machine_2_4_class_options(MachineClass *mc)
3275 {
3276     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3277 
3278     spapr_machine_2_5_class_options(mc);
3279     smc->dr_lmb_enabled = false;
3280     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3281 }
3282 
3283 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3284 
3285 /*
3286  * pseries-2.3
3287  */
3288 #define SPAPR_COMPAT_2_3 \
3289         HW_COMPAT_2_3 \
3290         {\
3291             .driver   = "spapr-pci-host-bridge",\
3292             .property = "dynamic-reconfiguration",\
3293             .value    = "off",\
3294         },
3295 
3296 static void spapr_machine_2_3_instance_options(MachineState *machine)
3297 {
3298     spapr_machine_2_4_instance_options(machine);
3299     savevm_skip_section_footers();
3300     global_state_set_optional();
3301     savevm_skip_configuration();
3302 }
3303 
3304 static void spapr_machine_2_3_class_options(MachineClass *mc)
3305 {
3306     spapr_machine_2_4_class_options(mc);
3307     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3308 }
3309 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3310 
3311 /*
3312  * pseries-2.2
3313  */
3314 
3315 #define SPAPR_COMPAT_2_2 \
3316         HW_COMPAT_2_2 \
3317         {\
3318             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3319             .property = "mem_win_size",\
3320             .value    = "0x20000000",\
3321         },
3322 
3323 static void spapr_machine_2_2_instance_options(MachineState *machine)
3324 {
3325     spapr_machine_2_3_instance_options(machine);
3326     machine->suppress_vmdesc = true;
3327 }
3328 
3329 static void spapr_machine_2_2_class_options(MachineClass *mc)
3330 {
3331     spapr_machine_2_3_class_options(mc);
3332     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3333 }
3334 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3335 
3336 /*
3337  * pseries-2.1
3338  */
3339 #define SPAPR_COMPAT_2_1 \
3340         HW_COMPAT_2_1
3341 
3342 static void spapr_machine_2_1_instance_options(MachineState *machine)
3343 {
3344     spapr_machine_2_2_instance_options(machine);
3345 }
3346 
3347 static void spapr_machine_2_1_class_options(MachineClass *mc)
3348 {
3349     spapr_machine_2_2_class_options(mc);
3350     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3351 }
3352 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3353 
3354 static void spapr_machine_register_types(void)
3355 {
3356     type_register_static(&spapr_machine_info);
3357 }
3358 
3359 type_init(spapr_machine_register_types)
3360