xref: /openbmc/qemu/hw/ppc/spapr.c (revision 2192a9303d43ee5e1b2b65f5ed9a93922bcdd1df)
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     XICSFabric *xi = XICS_FABRIC(spapr);
104     Error *err = NULL, *local_err = NULL;
105     XICSState *xics;
106     ICSState *ics = NULL;
107     int i;
108 
109     xics = XICS_COMMON(object_new(type));
110     qdev_set_parent_bus(DEVICE(xics), sysbus_get_default());
111     object_property_set_bool(OBJECT(xics), true, "realized", &err);
112     if (err) {
113         goto error;
114     }
115 
116     ics = ICS_SIMPLE(object_new(type_ics));
117     qdev_set_parent_bus(DEVICE(ics), sysbus_get_default());
118     object_property_add_child(OBJECT(spapr), "ics", OBJECT(ics), NULL);
119     object_property_set_int(OBJECT(ics), nr_irqs, "nr-irqs", &err);
120     object_property_add_const_link(OBJECT(ics), "xics", OBJECT(xi), NULL);
121     object_property_set_bool(OBJECT(ics), true, "realized", &local_err);
122     error_propagate(&err, local_err);
123     if (err) {
124         goto error;
125     }
126 
127     spapr->icps = g_malloc0(nr_servers * sizeof(ICPState));
128     spapr->nr_servers = nr_servers;
129 
130     for (i = 0; i < nr_servers; i++) {
131         ICPState *icp = &spapr->icps[i];
132 
133         object_initialize(icp, sizeof(*icp), type_icp);
134         qdev_set_parent_bus(DEVICE(icp), sysbus_get_default());
135         object_property_add_child(OBJECT(spapr), "icp[*]", OBJECT(icp), NULL);
136         object_property_add_const_link(OBJECT(icp), "xics", OBJECT(xi), NULL);
137         object_property_set_bool(OBJECT(icp), true, "realized", &err);
138         if (err) {
139             goto error;
140         }
141         object_unref(OBJECT(icp));
142     }
143 
144     spapr->ics = ics;
145     return xics;
146 
147 error:
148     error_propagate(errp, err);
149     if (ics) {
150         object_unparent(OBJECT(ics));
151     }
152     object_unparent(OBJECT(xics));
153     return NULL;
154 }
155 
156 static XICSState *xics_system_init(MachineState *machine,
157                                    int nr_servers, int nr_irqs, Error **errp)
158 {
159     XICSState *xics = NULL;
160 
161     if (kvm_enabled()) {
162         Error *err = NULL;
163 
164         if (machine_kernel_irqchip_allowed(machine) &&
165             !xics_kvm_init(SPAPR_MACHINE(machine), errp)) {
166             xics = try_create_xics(SPAPR_MACHINE(machine),
167                                    TYPE_XICS_SPAPR_KVM, TYPE_ICS_KVM,
168                                    TYPE_KVM_ICP, nr_servers, nr_irqs, &err);
169         }
170         if (machine_kernel_irqchip_required(machine) && !xics) {
171             error_reportf_err(err,
172                               "kernel_irqchip requested but unavailable: ");
173         } else {
174             error_free(err);
175         }
176     }
177 
178     if (!xics) {
179         xics_spapr_init(SPAPR_MACHINE(machine), errp);
180         xics = try_create_xics(SPAPR_MACHINE(machine),
181                                TYPE_XICS_SPAPR, TYPE_ICS_SIMPLE,
182                                TYPE_ICP, nr_servers, nr_irqs, errp);
183     }
184 
185     return xics;
186 }
187 
188 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
189                                   int smt_threads)
190 {
191     int i, ret = 0;
192     uint32_t servers_prop[smt_threads];
193     uint32_t gservers_prop[smt_threads * 2];
194     int index = ppc_get_vcpu_dt_id(cpu);
195 
196     if (cpu->compat_pvr) {
197         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
198         if (ret < 0) {
199             return ret;
200         }
201     }
202 
203     /* Build interrupt servers and gservers properties */
204     for (i = 0; i < smt_threads; i++) {
205         servers_prop[i] = cpu_to_be32(index + i);
206         /* Hack, direct the group queues back to cpu 0 */
207         gservers_prop[i*2] = cpu_to_be32(index + i);
208         gservers_prop[i*2 + 1] = 0;
209     }
210     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
211                       servers_prop, sizeof(servers_prop));
212     if (ret < 0) {
213         return ret;
214     }
215     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
216                       gservers_prop, sizeof(gservers_prop));
217 
218     return ret;
219 }
220 
221 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
222 {
223     int ret = 0;
224     PowerPCCPU *cpu = POWERPC_CPU(cs);
225     int index = ppc_get_vcpu_dt_id(cpu);
226     uint32_t associativity[] = {cpu_to_be32(0x5),
227                                 cpu_to_be32(0x0),
228                                 cpu_to_be32(0x0),
229                                 cpu_to_be32(0x0),
230                                 cpu_to_be32(cs->numa_node),
231                                 cpu_to_be32(index)};
232 
233     /* Advertise NUMA via ibm,associativity */
234     if (nb_numa_nodes > 1) {
235         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
236                           sizeof(associativity));
237     }
238 
239     return ret;
240 }
241 
242 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
243 {
244     int ret = 0, offset, cpus_offset;
245     CPUState *cs;
246     char cpu_model[32];
247     int smt = kvmppc_smt_threads();
248     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
249 
250     CPU_FOREACH(cs) {
251         PowerPCCPU *cpu = POWERPC_CPU(cs);
252         DeviceClass *dc = DEVICE_GET_CLASS(cs);
253         int index = ppc_get_vcpu_dt_id(cpu);
254         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
255 
256         if ((index % smt) != 0) {
257             continue;
258         }
259 
260         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
261 
262         cpus_offset = fdt_path_offset(fdt, "/cpus");
263         if (cpus_offset < 0) {
264             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
265                                           "cpus");
266             if (cpus_offset < 0) {
267                 return cpus_offset;
268             }
269         }
270         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
271         if (offset < 0) {
272             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
273             if (offset < 0) {
274                 return offset;
275             }
276         }
277 
278         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
279                           pft_size_prop, sizeof(pft_size_prop));
280         if (ret < 0) {
281             return ret;
282         }
283 
284         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
285         if (ret < 0) {
286             return ret;
287         }
288 
289         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
290         if (ret < 0) {
291             return ret;
292         }
293     }
294     return ret;
295 }
296 
297 static hwaddr spapr_node0_size(void)
298 {
299     MachineState *machine = MACHINE(qdev_get_machine());
300 
301     if (nb_numa_nodes) {
302         int i;
303         for (i = 0; i < nb_numa_nodes; ++i) {
304             if (numa_info[i].node_mem) {
305                 return MIN(pow2floor(numa_info[i].node_mem),
306                            machine->ram_size);
307             }
308         }
309     }
310     return machine->ram_size;
311 }
312 
313 static void add_str(GString *s, const gchar *s1)
314 {
315     g_string_append_len(s, s1, strlen(s1) + 1);
316 }
317 
318 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
319                                        hwaddr size)
320 {
321     uint32_t associativity[] = {
322         cpu_to_be32(0x4), /* length */
323         cpu_to_be32(0x0), cpu_to_be32(0x0),
324         cpu_to_be32(0x0), cpu_to_be32(nodeid)
325     };
326     char mem_name[32];
327     uint64_t mem_reg_property[2];
328     int off;
329 
330     mem_reg_property[0] = cpu_to_be64(start);
331     mem_reg_property[1] = cpu_to_be64(size);
332 
333     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
334     off = fdt_add_subnode(fdt, 0, mem_name);
335     _FDT(off);
336     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
337     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
338                       sizeof(mem_reg_property))));
339     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
340                       sizeof(associativity))));
341     return off;
342 }
343 
344 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
345 {
346     MachineState *machine = MACHINE(spapr);
347     hwaddr mem_start, node_size;
348     int i, nb_nodes = nb_numa_nodes;
349     NodeInfo *nodes = numa_info;
350     NodeInfo ramnode;
351 
352     /* No NUMA nodes, assume there is just one node with whole RAM */
353     if (!nb_numa_nodes) {
354         nb_nodes = 1;
355         ramnode.node_mem = machine->ram_size;
356         nodes = &ramnode;
357     }
358 
359     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
360         if (!nodes[i].node_mem) {
361             continue;
362         }
363         if (mem_start >= machine->ram_size) {
364             node_size = 0;
365         } else {
366             node_size = nodes[i].node_mem;
367             if (node_size > machine->ram_size - mem_start) {
368                 node_size = machine->ram_size - mem_start;
369             }
370         }
371         if (!mem_start) {
372             /* ppc_spapr_init() checks for rma_size <= node0_size already */
373             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
374             mem_start += spapr->rma_size;
375             node_size -= spapr->rma_size;
376         }
377         for ( ; node_size; ) {
378             hwaddr sizetmp = pow2floor(node_size);
379 
380             /* mem_start != 0 here */
381             if (ctzl(mem_start) < ctzl(sizetmp)) {
382                 sizetmp = 1ULL << ctzl(mem_start);
383             }
384 
385             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
386             node_size -= sizetmp;
387             mem_start += sizetmp;
388         }
389     }
390 
391     return 0;
392 }
393 
394 /* Populate the "ibm,pa-features" property */
395 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset)
396 {
397     uint8_t pa_features_206[] = { 6, 0,
398         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
399     uint8_t pa_features_207[] = { 24, 0,
400         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
401         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
402         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
403         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
404     uint8_t *pa_features;
405     size_t pa_size;
406 
407     switch (env->mmu_model) {
408     case POWERPC_MMU_2_06:
409     case POWERPC_MMU_2_06a:
410         pa_features = pa_features_206;
411         pa_size = sizeof(pa_features_206);
412         break;
413     case POWERPC_MMU_2_07:
414     case POWERPC_MMU_2_07a:
415         pa_features = pa_features_207;
416         pa_size = sizeof(pa_features_207);
417         break;
418     default:
419         return;
420     }
421 
422     if (env->ci_large_pages) {
423         /*
424          * Note: we keep CI large pages off by default because a 64K capable
425          * guest provisioned with large pages might otherwise try to map a qemu
426          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
427          * even if that qemu runs on a 4k host.
428          * We dd this bit back here if we are confident this is not an issue
429          */
430         pa_features[3] |= 0x20;
431     }
432     if (kvmppc_has_cap_htm() && pa_size > 24) {
433         pa_features[24] |= 0x80;    /* Transactional memory support */
434     }
435 
436     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
437 }
438 
439 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
440                                   sPAPRMachineState *spapr)
441 {
442     PowerPCCPU *cpu = POWERPC_CPU(cs);
443     CPUPPCState *env = &cpu->env;
444     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
445     int index = ppc_get_vcpu_dt_id(cpu);
446     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
447                        0xffffffff, 0xffffffff};
448     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
449         : SPAPR_TIMEBASE_FREQ;
450     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
451     uint32_t page_sizes_prop[64];
452     size_t page_sizes_prop_size;
453     uint32_t vcpus_per_socket = smp_threads * smp_cores;
454     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
455     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
456     sPAPRDRConnector *drc;
457     sPAPRDRConnectorClass *drck;
458     int drc_index;
459 
460     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
461     if (drc) {
462         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
463         drc_index = drck->get_index(drc);
464         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
465     }
466 
467     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
468     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
469 
470     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
471     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
472                            env->dcache_line_size)));
473     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
474                            env->dcache_line_size)));
475     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
476                            env->icache_line_size)));
477     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
478                            env->icache_line_size)));
479 
480     if (pcc->l1_dcache_size) {
481         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
482                                pcc->l1_dcache_size)));
483     } else {
484         error_report("Warning: Unknown L1 dcache size for cpu");
485     }
486     if (pcc->l1_icache_size) {
487         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
488                                pcc->l1_icache_size)));
489     } else {
490         error_report("Warning: Unknown L1 icache size for cpu");
491     }
492 
493     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
494     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
495     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
496     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
497     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
498     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
499 
500     if (env->spr_cb[SPR_PURR].oea_read) {
501         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
502     }
503 
504     if (env->mmu_model & POWERPC_MMU_1TSEG) {
505         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
506                           segs, sizeof(segs))));
507     }
508 
509     /* Advertise VMX/VSX (vector extensions) if available
510      *   0 / no property == no vector extensions
511      *   1               == VMX / Altivec available
512      *   2               == VSX available */
513     if (env->insns_flags & PPC_ALTIVEC) {
514         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
515 
516         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
517     }
518 
519     /* Advertise DFP (Decimal Floating Point) if available
520      *   0 / no property == no DFP
521      *   1               == DFP available */
522     if (env->insns_flags2 & PPC2_DFP) {
523         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
524     }
525 
526     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
527                                                   sizeof(page_sizes_prop));
528     if (page_sizes_prop_size) {
529         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
530                           page_sizes_prop, page_sizes_prop_size)));
531     }
532 
533     spapr_populate_pa_features(env, fdt, offset);
534 
535     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
536                            cs->cpu_index / vcpus_per_socket)));
537 
538     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
539                       pft_size_prop, sizeof(pft_size_prop))));
540 
541     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
542 
543     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
544 }
545 
546 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
547 {
548     CPUState *cs;
549     int cpus_offset;
550     char *nodename;
551     int smt = kvmppc_smt_threads();
552 
553     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
554     _FDT(cpus_offset);
555     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
556     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
557 
558     /*
559      * We walk the CPUs in reverse order to ensure that CPU DT nodes
560      * created by fdt_add_subnode() end up in the right order in FDT
561      * for the guest kernel the enumerate the CPUs correctly.
562      */
563     CPU_FOREACH_REVERSE(cs) {
564         PowerPCCPU *cpu = POWERPC_CPU(cs);
565         int index = ppc_get_vcpu_dt_id(cpu);
566         DeviceClass *dc = DEVICE_GET_CLASS(cs);
567         int offset;
568 
569         if ((index % smt) != 0) {
570             continue;
571         }
572 
573         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
574         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
575         g_free(nodename);
576         _FDT(offset);
577         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
578     }
579 
580 }
581 
582 /*
583  * Adds ibm,dynamic-reconfiguration-memory node.
584  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
585  * of this device tree node.
586  */
587 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
588 {
589     MachineState *machine = MACHINE(spapr);
590     int ret, i, offset;
591     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
592     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
593     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
594     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
595                        memory_region_size(&spapr->hotplug_memory.mr)) /
596                        lmb_size;
597     uint32_t *int_buf, *cur_index, buf_len;
598     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
599 
600     /*
601      * Don't create the node if there is no hotpluggable memory
602      */
603     if (machine->ram_size == machine->maxram_size) {
604         return 0;
605     }
606 
607     /*
608      * Allocate enough buffer size to fit in ibm,dynamic-memory
609      * or ibm,associativity-lookup-arrays
610      */
611     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
612               * sizeof(uint32_t);
613     cur_index = int_buf = g_malloc0(buf_len);
614 
615     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
616 
617     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
618                     sizeof(prop_lmb_size));
619     if (ret < 0) {
620         goto out;
621     }
622 
623     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
624     if (ret < 0) {
625         goto out;
626     }
627 
628     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
629     if (ret < 0) {
630         goto out;
631     }
632 
633     /* ibm,dynamic-memory */
634     int_buf[0] = cpu_to_be32(nr_lmbs);
635     cur_index++;
636     for (i = 0; i < nr_lmbs; i++) {
637         uint64_t addr = i * lmb_size;
638         uint32_t *dynamic_memory = cur_index;
639 
640         if (i >= hotplug_lmb_start) {
641             sPAPRDRConnector *drc;
642             sPAPRDRConnectorClass *drck;
643 
644             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
645             g_assert(drc);
646             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
647 
648             dynamic_memory[0] = cpu_to_be32(addr >> 32);
649             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
650             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
651             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
652             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
653             if (memory_region_present(get_system_memory(), addr)) {
654                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
655             } else {
656                 dynamic_memory[5] = cpu_to_be32(0);
657             }
658         } else {
659             /*
660              * LMB information for RMA, boot time RAM and gap b/n RAM and
661              * hotplug memory region -- all these are marked as reserved
662              * and as having no valid DRC.
663              */
664             dynamic_memory[0] = cpu_to_be32(addr >> 32);
665             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
666             dynamic_memory[2] = cpu_to_be32(0);
667             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
668             dynamic_memory[4] = cpu_to_be32(-1);
669             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
670                                             SPAPR_LMB_FLAGS_DRC_INVALID);
671         }
672 
673         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
674     }
675     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
676     if (ret < 0) {
677         goto out;
678     }
679 
680     /* ibm,associativity-lookup-arrays */
681     cur_index = int_buf;
682     int_buf[0] = cpu_to_be32(nr_nodes);
683     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
684     cur_index += 2;
685     for (i = 0; i < nr_nodes; i++) {
686         uint32_t associativity[] = {
687             cpu_to_be32(0x0),
688             cpu_to_be32(0x0),
689             cpu_to_be32(0x0),
690             cpu_to_be32(i)
691         };
692         memcpy(cur_index, associativity, sizeof(associativity));
693         cur_index += 4;
694     }
695     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
696             (cur_index - int_buf) * sizeof(uint32_t));
697 out:
698     g_free(int_buf);
699     return ret;
700 }
701 
702 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
703                                 sPAPROptionVector *ov5_updates)
704 {
705     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
706     int ret = 0, offset;
707 
708     /* Generate ibm,dynamic-reconfiguration-memory node if required */
709     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
710         g_assert(smc->dr_lmb_enabled);
711         ret = spapr_populate_drconf_memory(spapr, fdt);
712         if (ret) {
713             goto out;
714         }
715     }
716 
717     offset = fdt_path_offset(fdt, "/chosen");
718     if (offset < 0) {
719         offset = fdt_add_subnode(fdt, 0, "chosen");
720         if (offset < 0) {
721             return offset;
722         }
723     }
724     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
725                                  "ibm,architecture-vec-5");
726 
727 out:
728     return ret;
729 }
730 
731 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
732                                  target_ulong addr, target_ulong size,
733                                  sPAPROptionVector *ov5_updates)
734 {
735     void *fdt, *fdt_skel;
736     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
737 
738     size -= sizeof(hdr);
739 
740     /* Create sceleton */
741     fdt_skel = g_malloc0(size);
742     _FDT((fdt_create(fdt_skel, size)));
743     _FDT((fdt_begin_node(fdt_skel, "")));
744     _FDT((fdt_end_node(fdt_skel)));
745     _FDT((fdt_finish(fdt_skel)));
746     fdt = g_malloc0(size);
747     _FDT((fdt_open_into(fdt_skel, fdt, size)));
748     g_free(fdt_skel);
749 
750     /* Fixup cpu nodes */
751     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
752 
753     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
754         return -1;
755     }
756 
757     /* Pack resulting tree */
758     _FDT((fdt_pack(fdt)));
759 
760     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
761         trace_spapr_cas_failed(size);
762         return -1;
763     }
764 
765     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
766     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
767     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
768     g_free(fdt);
769 
770     return 0;
771 }
772 
773 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
774 {
775     int rtas;
776     GString *hypertas = g_string_sized_new(256);
777     GString *qemu_hypertas = g_string_sized_new(256);
778     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
779     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
780         memory_region_size(&spapr->hotplug_memory.mr);
781     uint32_t lrdr_capacity[] = {
782         cpu_to_be32(max_hotplug_addr >> 32),
783         cpu_to_be32(max_hotplug_addr & 0xffffffff),
784         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
785         cpu_to_be32(max_cpus / smp_threads),
786     };
787 
788     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
789 
790     /* hypertas */
791     add_str(hypertas, "hcall-pft");
792     add_str(hypertas, "hcall-term");
793     add_str(hypertas, "hcall-dabr");
794     add_str(hypertas, "hcall-interrupt");
795     add_str(hypertas, "hcall-tce");
796     add_str(hypertas, "hcall-vio");
797     add_str(hypertas, "hcall-splpar");
798     add_str(hypertas, "hcall-bulk");
799     add_str(hypertas, "hcall-set-mode");
800     add_str(hypertas, "hcall-sprg0");
801     add_str(hypertas, "hcall-copy");
802     add_str(hypertas, "hcall-debug");
803     add_str(qemu_hypertas, "hcall-memop1");
804 
805     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
806         add_str(hypertas, "hcall-multi-tce");
807     }
808     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
809                      hypertas->str, hypertas->len));
810     g_string_free(hypertas, TRUE);
811     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
812                      qemu_hypertas->str, qemu_hypertas->len));
813     g_string_free(qemu_hypertas, TRUE);
814 
815     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
816                      refpoints, sizeof(refpoints)));
817 
818     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
819                           RTAS_ERROR_LOG_MAX));
820     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
821                           RTAS_EVENT_SCAN_RATE));
822 
823     if (msi_nonbroken) {
824         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
825     }
826 
827     /*
828      * According to PAPR, rtas ibm,os-term does not guarantee a return
829      * back to the guest cpu.
830      *
831      * While an additional ibm,extended-os-term property indicates
832      * that rtas call return will always occur. Set this property.
833      */
834     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
835 
836     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
837                      lrdr_capacity, sizeof(lrdr_capacity)));
838 
839     spapr_dt_rtas_tokens(fdt, rtas);
840 }
841 
842 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
843 {
844     MachineState *machine = MACHINE(spapr);
845     int chosen;
846     const char *boot_device = machine->boot_order;
847     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
848     size_t cb = 0;
849     char *bootlist = get_boot_devices_list(&cb, true);
850 
851     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
852 
853     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
854     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
855                           spapr->initrd_base));
856     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
857                           spapr->initrd_base + spapr->initrd_size));
858 
859     if (spapr->kernel_size) {
860         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
861                               cpu_to_be64(spapr->kernel_size) };
862 
863         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
864                          &kprop, sizeof(kprop)));
865         if (spapr->kernel_le) {
866             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
867         }
868     }
869     if (boot_menu) {
870         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
871     }
872     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
873     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
874     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
875 
876     if (cb && bootlist) {
877         int i;
878 
879         for (i = 0; i < cb; i++) {
880             if (bootlist[i] == '\n') {
881                 bootlist[i] = ' ';
882             }
883         }
884         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
885     }
886 
887     if (boot_device && strlen(boot_device)) {
888         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
889     }
890 
891     if (!spapr->has_graphics && stdout_path) {
892         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
893     }
894 
895     g_free(stdout_path);
896     g_free(bootlist);
897 }
898 
899 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
900 {
901     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
902      * KVM to work under pHyp with some guest co-operation */
903     int hypervisor;
904     uint8_t hypercall[16];
905 
906     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
907     /* indicate KVM hypercall interface */
908     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
909     if (kvmppc_has_cap_fixup_hcalls()) {
910         /*
911          * Older KVM versions with older guest kernels were broken
912          * with the magic page, don't allow the guest to map it.
913          */
914         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
915                                   sizeof(hypercall))) {
916             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
917                              hypercall, sizeof(hypercall)));
918         }
919     }
920 }
921 
922 static void *spapr_build_fdt(sPAPRMachineState *spapr,
923                              hwaddr rtas_addr,
924                              hwaddr rtas_size)
925 {
926     MachineState *machine = MACHINE(qdev_get_machine());
927     MachineClass *mc = MACHINE_GET_CLASS(machine);
928     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
929     int ret;
930     void *fdt;
931     sPAPRPHBState *phb;
932     char *buf;
933 
934     fdt = g_malloc0(FDT_MAX_SIZE);
935     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
936 
937     /* Root node */
938     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
939     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
940     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
941 
942     /*
943      * Add info to guest to indentify which host is it being run on
944      * and what is the uuid of the guest
945      */
946     if (kvmppc_get_host_model(&buf)) {
947         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
948         g_free(buf);
949     }
950     if (kvmppc_get_host_serial(&buf)) {
951         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
952         g_free(buf);
953     }
954 
955     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
956 
957     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
958     if (qemu_uuid_set) {
959         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
960     }
961     g_free(buf);
962 
963     if (qemu_get_vm_name()) {
964         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
965                                 qemu_get_vm_name()));
966     }
967 
968     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
969     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
970 
971     /* /interrupt controller */
972     spapr_dt_xics(spapr->nr_servers, fdt, PHANDLE_XICP);
973 
974     ret = spapr_populate_memory(spapr, fdt);
975     if (ret < 0) {
976         error_report("couldn't setup memory nodes in fdt");
977         exit(1);
978     }
979 
980     /* /vdevice */
981     spapr_dt_vdevice(spapr->vio_bus, fdt);
982 
983     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
984         ret = spapr_rng_populate_dt(fdt);
985         if (ret < 0) {
986             error_report("could not set up rng device in the fdt");
987             exit(1);
988         }
989     }
990 
991     QLIST_FOREACH(phb, &spapr->phbs, list) {
992         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
993         if (ret < 0) {
994             error_report("couldn't setup PCI devices in fdt");
995             exit(1);
996         }
997     }
998 
999     /* cpus */
1000     spapr_populate_cpus_dt_node(fdt, spapr);
1001 
1002     if (smc->dr_lmb_enabled) {
1003         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1004     }
1005 
1006     if (mc->has_hotpluggable_cpus) {
1007         int offset = fdt_path_offset(fdt, "/cpus");
1008         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1009                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1010         if (ret < 0) {
1011             error_report("Couldn't set up CPU DR device tree properties");
1012             exit(1);
1013         }
1014     }
1015 
1016     /* /event-sources */
1017     spapr_dt_events(spapr, fdt);
1018 
1019     /* /rtas */
1020     spapr_dt_rtas(spapr, fdt);
1021 
1022     /* /chosen */
1023     spapr_dt_chosen(spapr, fdt);
1024 
1025     /* /hypervisor */
1026     if (kvm_enabled()) {
1027         spapr_dt_hypervisor(spapr, fdt);
1028     }
1029 
1030     /* Build memory reserve map */
1031     if (spapr->kernel_size) {
1032         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1033     }
1034     if (spapr->initrd_size) {
1035         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1036     }
1037 
1038     /* ibm,client-architecture-support updates */
1039     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1040     if (ret < 0) {
1041         error_report("couldn't setup CAS properties fdt");
1042         exit(1);
1043     }
1044 
1045     return fdt;
1046 }
1047 
1048 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1049 {
1050     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1051 }
1052 
1053 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1054                                     PowerPCCPU *cpu)
1055 {
1056     CPUPPCState *env = &cpu->env;
1057 
1058     /* The TCG path should also be holding the BQL at this point */
1059     g_assert(qemu_mutex_iothread_locked());
1060 
1061     if (msr_pr) {
1062         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1063         env->gpr[3] = H_PRIVILEGE;
1064     } else {
1065         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1066     }
1067 }
1068 
1069 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1070 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1071 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1072 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1073 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1074 
1075 /*
1076  * Get the fd to access the kernel htab, re-opening it if necessary
1077  */
1078 static int get_htab_fd(sPAPRMachineState *spapr)
1079 {
1080     if (spapr->htab_fd >= 0) {
1081         return spapr->htab_fd;
1082     }
1083 
1084     spapr->htab_fd = kvmppc_get_htab_fd(false);
1085     if (spapr->htab_fd < 0) {
1086         error_report("Unable to open fd for reading hash table from KVM: %s",
1087                      strerror(errno));
1088     }
1089 
1090     return spapr->htab_fd;
1091 }
1092 
1093 static void close_htab_fd(sPAPRMachineState *spapr)
1094 {
1095     if (spapr->htab_fd >= 0) {
1096         close(spapr->htab_fd);
1097     }
1098     spapr->htab_fd = -1;
1099 }
1100 
1101 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1102 {
1103     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1104 
1105     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1106 }
1107 
1108 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1109                                                 hwaddr ptex, int n)
1110 {
1111     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1112     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1113 
1114     if (!spapr->htab) {
1115         /*
1116          * HTAB is controlled by KVM. Fetch into temporary buffer
1117          */
1118         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1119         kvmppc_read_hptes(hptes, ptex, n);
1120         return hptes;
1121     }
1122 
1123     /*
1124      * HTAB is controlled by QEMU. Just point to the internally
1125      * accessible PTEG.
1126      */
1127     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1128 }
1129 
1130 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1131                               const ppc_hash_pte64_t *hptes,
1132                               hwaddr ptex, int n)
1133 {
1134     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1135 
1136     if (!spapr->htab) {
1137         g_free((void *)hptes);
1138     }
1139 
1140     /* Nothing to do for qemu managed HPT */
1141 }
1142 
1143 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1144                              uint64_t pte0, uint64_t pte1)
1145 {
1146     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1147     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1148 
1149     if (!spapr->htab) {
1150         kvmppc_write_hpte(ptex, pte0, pte1);
1151     } else {
1152         stq_p(spapr->htab + offset, pte0);
1153         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1154     }
1155 }
1156 
1157 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1158 {
1159     int shift;
1160 
1161     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1162      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1163      * that's much more than is needed for Linux guests */
1164     shift = ctz64(pow2ceil(ramsize)) - 7;
1165     shift = MAX(shift, 18); /* Minimum architected size */
1166     shift = MIN(shift, 46); /* Maximum architected size */
1167     return shift;
1168 }
1169 
1170 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1171                                  Error **errp)
1172 {
1173     long rc;
1174 
1175     /* Clean up any HPT info from a previous boot */
1176     g_free(spapr->htab);
1177     spapr->htab = NULL;
1178     spapr->htab_shift = 0;
1179     close_htab_fd(spapr);
1180 
1181     rc = kvmppc_reset_htab(shift);
1182     if (rc < 0) {
1183         /* kernel-side HPT needed, but couldn't allocate one */
1184         error_setg_errno(errp, errno,
1185                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1186                          shift);
1187         /* This is almost certainly fatal, but if the caller really
1188          * wants to carry on with shift == 0, it's welcome to try */
1189     } else if (rc > 0) {
1190         /* kernel-side HPT allocated */
1191         if (rc != shift) {
1192             error_setg(errp,
1193                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1194                        shift, rc);
1195         }
1196 
1197         spapr->htab_shift = shift;
1198         spapr->htab = NULL;
1199     } else {
1200         /* kernel-side HPT not needed, allocate in userspace instead */
1201         size_t size = 1ULL << shift;
1202         int i;
1203 
1204         spapr->htab = qemu_memalign(size, size);
1205         if (!spapr->htab) {
1206             error_setg_errno(errp, errno,
1207                              "Could not allocate HPT of order %d", shift);
1208             return;
1209         }
1210 
1211         memset(spapr->htab, 0, size);
1212         spapr->htab_shift = shift;
1213 
1214         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1215             DIRTY_HPTE(HPTE(spapr->htab, i));
1216         }
1217     }
1218 }
1219 
1220 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1221 {
1222     bool matched = false;
1223 
1224     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1225         matched = true;
1226     }
1227 
1228     if (!matched) {
1229         error_report("Device %s is not supported by this machine yet.",
1230                      qdev_fw_name(DEVICE(sbdev)));
1231         exit(1);
1232     }
1233 }
1234 
1235 static void ppc_spapr_reset(void)
1236 {
1237     MachineState *machine = MACHINE(qdev_get_machine());
1238     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1239     PowerPCCPU *first_ppc_cpu;
1240     uint32_t rtas_limit;
1241     hwaddr rtas_addr, fdt_addr;
1242     void *fdt;
1243     int rc;
1244 
1245     /* Check for unknown sysbus devices */
1246     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1247 
1248     /* Allocate and/or reset the hash page table */
1249     spapr_reallocate_hpt(spapr,
1250                          spapr_hpt_shift_for_ramsize(machine->maxram_size),
1251                          &error_fatal);
1252 
1253     /* Update the RMA size if necessary */
1254     if (spapr->vrma_adjust) {
1255         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1256                                           spapr->htab_shift);
1257     }
1258 
1259     qemu_devices_reset();
1260 
1261     /*
1262      * We place the device tree and RTAS just below either the top of the RMA,
1263      * or just below 2GB, whichever is lowere, so that it can be
1264      * processed with 32-bit real mode code if necessary
1265      */
1266     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1267     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1268     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1269 
1270     /* if this reset wasn't generated by CAS, we should reset our
1271      * negotiated options and start from scratch */
1272     if (!spapr->cas_reboot) {
1273         spapr_ovec_cleanup(spapr->ov5_cas);
1274         spapr->ov5_cas = spapr_ovec_new();
1275     }
1276 
1277     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1278 
1279     spapr_load_rtas(spapr, fdt, rtas_addr);
1280 
1281     rc = fdt_pack(fdt);
1282 
1283     /* Should only fail if we've built a corrupted tree */
1284     assert(rc == 0);
1285 
1286     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1287         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1288                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1289         exit(1);
1290     }
1291 
1292     /* Load the fdt */
1293     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1294     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1295     g_free(fdt);
1296 
1297     /* Set up the entry state */
1298     first_ppc_cpu = POWERPC_CPU(first_cpu);
1299     first_ppc_cpu->env.gpr[3] = fdt_addr;
1300     first_ppc_cpu->env.gpr[5] = 0;
1301     first_cpu->halted = 0;
1302     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1303 
1304     spapr->cas_reboot = false;
1305 }
1306 
1307 static void spapr_create_nvram(sPAPRMachineState *spapr)
1308 {
1309     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1310     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1311 
1312     if (dinfo) {
1313         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1314                             &error_fatal);
1315     }
1316 
1317     qdev_init_nofail(dev);
1318 
1319     spapr->nvram = (struct sPAPRNVRAM *)dev;
1320 }
1321 
1322 static void spapr_rtc_create(sPAPRMachineState *spapr)
1323 {
1324     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1325 
1326     qdev_init_nofail(dev);
1327     spapr->rtc = dev;
1328 
1329     object_property_add_alias(qdev_get_machine(), "rtc-time",
1330                               OBJECT(spapr->rtc), "date", NULL);
1331 }
1332 
1333 /* Returns whether we want to use VGA or not */
1334 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1335 {
1336     switch (vga_interface_type) {
1337     case VGA_NONE:
1338         return false;
1339     case VGA_DEVICE:
1340         return true;
1341     case VGA_STD:
1342     case VGA_VIRTIO:
1343         return pci_vga_init(pci_bus) != NULL;
1344     default:
1345         error_setg(errp,
1346                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1347         return false;
1348     }
1349 }
1350 
1351 static int spapr_post_load(void *opaque, int version_id)
1352 {
1353     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1354     int err = 0;
1355 
1356     /* In earlier versions, there was no separate qdev for the PAPR
1357      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1358      * So when migrating from those versions, poke the incoming offset
1359      * value into the RTC device */
1360     if (version_id < 3) {
1361         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1362     }
1363 
1364     return err;
1365 }
1366 
1367 static bool version_before_3(void *opaque, int version_id)
1368 {
1369     return version_id < 3;
1370 }
1371 
1372 static bool spapr_ov5_cas_needed(void *opaque)
1373 {
1374     sPAPRMachineState *spapr = opaque;
1375     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1376     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1377     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1378     bool cas_needed;
1379 
1380     /* Prior to the introduction of sPAPROptionVector, we had two option
1381      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1382      * Both of these options encode machine topology into the device-tree
1383      * in such a way that the now-booted OS should still be able to interact
1384      * appropriately with QEMU regardless of what options were actually
1385      * negotiatied on the source side.
1386      *
1387      * As such, we can avoid migrating the CAS-negotiated options if these
1388      * are the only options available on the current machine/platform.
1389      * Since these are the only options available for pseries-2.7 and
1390      * earlier, this allows us to maintain old->new/new->old migration
1391      * compatibility.
1392      *
1393      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1394      * via default pseries-2.8 machines and explicit command-line parameters.
1395      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1396      * of the actual CAS-negotiated values to continue working properly. For
1397      * example, availability of memory unplug depends on knowing whether
1398      * OV5_HP_EVT was negotiated via CAS.
1399      *
1400      * Thus, for any cases where the set of available CAS-negotiatable
1401      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1402      * include the CAS-negotiated options in the migration stream.
1403      */
1404     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1405     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1406 
1407     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1408      * the mask itself since in the future it's possible "legacy" bits may be
1409      * removed via machine options, which could generate a false positive
1410      * that breaks migration.
1411      */
1412     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1413     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1414 
1415     spapr_ovec_cleanup(ov5_mask);
1416     spapr_ovec_cleanup(ov5_legacy);
1417     spapr_ovec_cleanup(ov5_removed);
1418 
1419     return cas_needed;
1420 }
1421 
1422 static const VMStateDescription vmstate_spapr_ov5_cas = {
1423     .name = "spapr_option_vector_ov5_cas",
1424     .version_id = 1,
1425     .minimum_version_id = 1,
1426     .needed = spapr_ov5_cas_needed,
1427     .fields = (VMStateField[]) {
1428         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1429                                  vmstate_spapr_ovec, sPAPROptionVector),
1430         VMSTATE_END_OF_LIST()
1431     },
1432 };
1433 
1434 static const VMStateDescription vmstate_spapr = {
1435     .name = "spapr",
1436     .version_id = 3,
1437     .minimum_version_id = 1,
1438     .post_load = spapr_post_load,
1439     .fields = (VMStateField[]) {
1440         /* used to be @next_irq */
1441         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1442 
1443         /* RTC offset */
1444         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1445 
1446         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1447         VMSTATE_END_OF_LIST()
1448     },
1449     .subsections = (const VMStateDescription*[]) {
1450         &vmstate_spapr_ov5_cas,
1451         NULL
1452     }
1453 };
1454 
1455 static int htab_save_setup(QEMUFile *f, void *opaque)
1456 {
1457     sPAPRMachineState *spapr = opaque;
1458 
1459     /* "Iteration" header */
1460     qemu_put_be32(f, spapr->htab_shift);
1461 
1462     if (spapr->htab) {
1463         spapr->htab_save_index = 0;
1464         spapr->htab_first_pass = true;
1465     } else {
1466         assert(kvm_enabled());
1467     }
1468 
1469 
1470     return 0;
1471 }
1472 
1473 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1474                                  int64_t max_ns)
1475 {
1476     bool has_timeout = max_ns != -1;
1477     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1478     int index = spapr->htab_save_index;
1479     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1480 
1481     assert(spapr->htab_first_pass);
1482 
1483     do {
1484         int chunkstart;
1485 
1486         /* Consume invalid HPTEs */
1487         while ((index < htabslots)
1488                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1489             index++;
1490             CLEAN_HPTE(HPTE(spapr->htab, index));
1491         }
1492 
1493         /* Consume valid HPTEs */
1494         chunkstart = index;
1495         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1496                && HPTE_VALID(HPTE(spapr->htab, index))) {
1497             index++;
1498             CLEAN_HPTE(HPTE(spapr->htab, index));
1499         }
1500 
1501         if (index > chunkstart) {
1502             int n_valid = index - chunkstart;
1503 
1504             qemu_put_be32(f, chunkstart);
1505             qemu_put_be16(f, n_valid);
1506             qemu_put_be16(f, 0);
1507             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1508                             HASH_PTE_SIZE_64 * n_valid);
1509 
1510             if (has_timeout &&
1511                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1512                 break;
1513             }
1514         }
1515     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1516 
1517     if (index >= htabslots) {
1518         assert(index == htabslots);
1519         index = 0;
1520         spapr->htab_first_pass = false;
1521     }
1522     spapr->htab_save_index = index;
1523 }
1524 
1525 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1526                                 int64_t max_ns)
1527 {
1528     bool final = max_ns < 0;
1529     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1530     int examined = 0, sent = 0;
1531     int index = spapr->htab_save_index;
1532     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1533 
1534     assert(!spapr->htab_first_pass);
1535 
1536     do {
1537         int chunkstart, invalidstart;
1538 
1539         /* Consume non-dirty HPTEs */
1540         while ((index < htabslots)
1541                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1542             index++;
1543             examined++;
1544         }
1545 
1546         chunkstart = index;
1547         /* Consume valid dirty HPTEs */
1548         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1549                && HPTE_DIRTY(HPTE(spapr->htab, index))
1550                && HPTE_VALID(HPTE(spapr->htab, index))) {
1551             CLEAN_HPTE(HPTE(spapr->htab, index));
1552             index++;
1553             examined++;
1554         }
1555 
1556         invalidstart = index;
1557         /* Consume invalid dirty HPTEs */
1558         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1559                && HPTE_DIRTY(HPTE(spapr->htab, index))
1560                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1561             CLEAN_HPTE(HPTE(spapr->htab, index));
1562             index++;
1563             examined++;
1564         }
1565 
1566         if (index > chunkstart) {
1567             int n_valid = invalidstart - chunkstart;
1568             int n_invalid = index - invalidstart;
1569 
1570             qemu_put_be32(f, chunkstart);
1571             qemu_put_be16(f, n_valid);
1572             qemu_put_be16(f, n_invalid);
1573             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1574                             HASH_PTE_SIZE_64 * n_valid);
1575             sent += index - chunkstart;
1576 
1577             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1578                 break;
1579             }
1580         }
1581 
1582         if (examined >= htabslots) {
1583             break;
1584         }
1585 
1586         if (index >= htabslots) {
1587             assert(index == htabslots);
1588             index = 0;
1589         }
1590     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1591 
1592     if (index >= htabslots) {
1593         assert(index == htabslots);
1594         index = 0;
1595     }
1596 
1597     spapr->htab_save_index = index;
1598 
1599     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1600 }
1601 
1602 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1603 #define MAX_KVM_BUF_SIZE    2048
1604 
1605 static int htab_save_iterate(QEMUFile *f, void *opaque)
1606 {
1607     sPAPRMachineState *spapr = opaque;
1608     int fd;
1609     int rc = 0;
1610 
1611     /* Iteration header */
1612     qemu_put_be32(f, 0);
1613 
1614     if (!spapr->htab) {
1615         assert(kvm_enabled());
1616 
1617         fd = get_htab_fd(spapr);
1618         if (fd < 0) {
1619             return fd;
1620         }
1621 
1622         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1623         if (rc < 0) {
1624             return rc;
1625         }
1626     } else  if (spapr->htab_first_pass) {
1627         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1628     } else {
1629         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1630     }
1631 
1632     /* End marker */
1633     qemu_put_be32(f, 0);
1634     qemu_put_be16(f, 0);
1635     qemu_put_be16(f, 0);
1636 
1637     return rc;
1638 }
1639 
1640 static int htab_save_complete(QEMUFile *f, void *opaque)
1641 {
1642     sPAPRMachineState *spapr = opaque;
1643     int fd;
1644 
1645     /* Iteration header */
1646     qemu_put_be32(f, 0);
1647 
1648     if (!spapr->htab) {
1649         int rc;
1650 
1651         assert(kvm_enabled());
1652 
1653         fd = get_htab_fd(spapr);
1654         if (fd < 0) {
1655             return fd;
1656         }
1657 
1658         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1659         if (rc < 0) {
1660             return rc;
1661         }
1662     } else {
1663         if (spapr->htab_first_pass) {
1664             htab_save_first_pass(f, spapr, -1);
1665         }
1666         htab_save_later_pass(f, spapr, -1);
1667     }
1668 
1669     /* End marker */
1670     qemu_put_be32(f, 0);
1671     qemu_put_be16(f, 0);
1672     qemu_put_be16(f, 0);
1673 
1674     return 0;
1675 }
1676 
1677 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1678 {
1679     sPAPRMachineState *spapr = opaque;
1680     uint32_t section_hdr;
1681     int fd = -1;
1682 
1683     if (version_id < 1 || version_id > 1) {
1684         error_report("htab_load() bad version");
1685         return -EINVAL;
1686     }
1687 
1688     section_hdr = qemu_get_be32(f);
1689 
1690     if (section_hdr) {
1691         Error *local_err = NULL;
1692 
1693         /* First section gives the htab size */
1694         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1695         if (local_err) {
1696             error_report_err(local_err);
1697             return -EINVAL;
1698         }
1699         return 0;
1700     }
1701 
1702     if (!spapr->htab) {
1703         assert(kvm_enabled());
1704 
1705         fd = kvmppc_get_htab_fd(true);
1706         if (fd < 0) {
1707             error_report("Unable to open fd to restore KVM hash table: %s",
1708                          strerror(errno));
1709         }
1710     }
1711 
1712     while (true) {
1713         uint32_t index;
1714         uint16_t n_valid, n_invalid;
1715 
1716         index = qemu_get_be32(f);
1717         n_valid = qemu_get_be16(f);
1718         n_invalid = qemu_get_be16(f);
1719 
1720         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1721             /* End of Stream */
1722             break;
1723         }
1724 
1725         if ((index + n_valid + n_invalid) >
1726             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1727             /* Bad index in stream */
1728             error_report(
1729                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1730                 index, n_valid, n_invalid, spapr->htab_shift);
1731             return -EINVAL;
1732         }
1733 
1734         if (spapr->htab) {
1735             if (n_valid) {
1736                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1737                                 HASH_PTE_SIZE_64 * n_valid);
1738             }
1739             if (n_invalid) {
1740                 memset(HPTE(spapr->htab, index + n_valid), 0,
1741                        HASH_PTE_SIZE_64 * n_invalid);
1742             }
1743         } else {
1744             int rc;
1745 
1746             assert(fd >= 0);
1747 
1748             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1749             if (rc < 0) {
1750                 return rc;
1751             }
1752         }
1753     }
1754 
1755     if (!spapr->htab) {
1756         assert(fd >= 0);
1757         close(fd);
1758     }
1759 
1760     return 0;
1761 }
1762 
1763 static void htab_cleanup(void *opaque)
1764 {
1765     sPAPRMachineState *spapr = opaque;
1766 
1767     close_htab_fd(spapr);
1768 }
1769 
1770 static SaveVMHandlers savevm_htab_handlers = {
1771     .save_live_setup = htab_save_setup,
1772     .save_live_iterate = htab_save_iterate,
1773     .save_live_complete_precopy = htab_save_complete,
1774     .cleanup = htab_cleanup,
1775     .load_state = htab_load,
1776 };
1777 
1778 static void spapr_boot_set(void *opaque, const char *boot_device,
1779                            Error **errp)
1780 {
1781     MachineState *machine = MACHINE(qdev_get_machine());
1782     machine->boot_order = g_strdup(boot_device);
1783 }
1784 
1785 /*
1786  * Reset routine for LMB DR devices.
1787  *
1788  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1789  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1790  * when it walks all its children devices. LMB devices reset occurs
1791  * as part of spapr_ppc_reset().
1792  */
1793 static void spapr_drc_reset(void *opaque)
1794 {
1795     sPAPRDRConnector *drc = opaque;
1796     DeviceState *d = DEVICE(drc);
1797 
1798     if (d) {
1799         device_reset(d);
1800     }
1801 }
1802 
1803 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1804 {
1805     MachineState *machine = MACHINE(spapr);
1806     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1807     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1808     int i;
1809 
1810     for (i = 0; i < nr_lmbs; i++) {
1811         sPAPRDRConnector *drc;
1812         uint64_t addr;
1813 
1814         addr = i * lmb_size + spapr->hotplug_memory.base;
1815         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1816                                      addr/lmb_size);
1817         qemu_register_reset(spapr_drc_reset, drc);
1818     }
1819 }
1820 
1821 /*
1822  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1823  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1824  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1825  */
1826 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1827 {
1828     int i;
1829 
1830     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1831         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1832                    " is not aligned to %llu MiB",
1833                    machine->ram_size,
1834                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1835         return;
1836     }
1837 
1838     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1839         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1840                    " is not aligned to %llu MiB",
1841                    machine->ram_size,
1842                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1843         return;
1844     }
1845 
1846     for (i = 0; i < nb_numa_nodes; i++) {
1847         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1848             error_setg(errp,
1849                        "Node %d memory size 0x%" PRIx64
1850                        " is not aligned to %llu MiB",
1851                        i, numa_info[i].node_mem,
1852                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1853             return;
1854         }
1855     }
1856 }
1857 
1858 /* find cpu slot in machine->possible_cpus by core_id */
1859 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1860 {
1861     int index = id / smp_threads;
1862 
1863     if (index >= ms->possible_cpus->len) {
1864         return NULL;
1865     }
1866     if (idx) {
1867         *idx = index;
1868     }
1869     return &ms->possible_cpus->cpus[index];
1870 }
1871 
1872 static void spapr_init_cpus(sPAPRMachineState *spapr)
1873 {
1874     MachineState *machine = MACHINE(spapr);
1875     MachineClass *mc = MACHINE_GET_CLASS(machine);
1876     char *type = spapr_get_cpu_core_type(machine->cpu_model);
1877     int smt = kvmppc_smt_threads();
1878     const CPUArchIdList *possible_cpus;
1879     int boot_cores_nr = smp_cpus / smp_threads;
1880     int i;
1881 
1882     if (!type) {
1883         error_report("Unable to find sPAPR CPU Core definition");
1884         exit(1);
1885     }
1886 
1887     possible_cpus = mc->possible_cpu_arch_ids(machine);
1888     if (mc->has_hotpluggable_cpus) {
1889         if (smp_cpus % smp_threads) {
1890             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1891                          smp_cpus, smp_threads);
1892             exit(1);
1893         }
1894         if (max_cpus % smp_threads) {
1895             error_report("max_cpus (%u) must be multiple of threads (%u)",
1896                          max_cpus, smp_threads);
1897             exit(1);
1898         }
1899     } else {
1900         if (max_cpus != smp_cpus) {
1901             error_report("This machine version does not support CPU hotplug");
1902             exit(1);
1903         }
1904         boot_cores_nr = possible_cpus->len;
1905     }
1906 
1907     for (i = 0; i < possible_cpus->len; i++) {
1908         int core_id = i * smp_threads;
1909 
1910         if (mc->has_hotpluggable_cpus) {
1911             sPAPRDRConnector *drc =
1912                 spapr_dr_connector_new(OBJECT(spapr),
1913                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
1914                                        (core_id / smp_threads) * smt);
1915 
1916             qemu_register_reset(spapr_drc_reset, drc);
1917         }
1918 
1919         if (i < boot_cores_nr) {
1920             Object *core  = object_new(type);
1921             int nr_threads = smp_threads;
1922 
1923             /* Handle the partially filled core for older machine types */
1924             if ((i + 1) * smp_threads >= smp_cpus) {
1925                 nr_threads = smp_cpus - i * smp_threads;
1926             }
1927 
1928             object_property_set_int(core, nr_threads, "nr-threads",
1929                                     &error_fatal);
1930             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1931                                     &error_fatal);
1932             object_property_set_bool(core, true, "realized", &error_fatal);
1933         }
1934     }
1935     g_free(type);
1936 }
1937 
1938 /* pSeries LPAR / sPAPR hardware init */
1939 static void ppc_spapr_init(MachineState *machine)
1940 {
1941     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1942     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1943     const char *kernel_filename = machine->kernel_filename;
1944     const char *initrd_filename = machine->initrd_filename;
1945     PCIHostState *phb;
1946     int i;
1947     MemoryRegion *sysmem = get_system_memory();
1948     MemoryRegion *ram = g_new(MemoryRegion, 1);
1949     MemoryRegion *rma_region;
1950     void *rma = NULL;
1951     hwaddr rma_alloc_size;
1952     hwaddr node0_size = spapr_node0_size();
1953     long load_limit, fw_size;
1954     char *filename;
1955     int smt = kvmppc_smt_threads();
1956 
1957     msi_nonbroken = true;
1958 
1959     QLIST_INIT(&spapr->phbs);
1960 
1961     /* Allocate RMA if necessary */
1962     rma_alloc_size = kvmppc_alloc_rma(&rma);
1963 
1964     if (rma_alloc_size == -1) {
1965         error_report("Unable to create RMA");
1966         exit(1);
1967     }
1968 
1969     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1970         spapr->rma_size = rma_alloc_size;
1971     } else {
1972         spapr->rma_size = node0_size;
1973 
1974         /* With KVM, we don't actually know whether KVM supports an
1975          * unbounded RMA (PR KVM) or is limited by the hash table size
1976          * (HV KVM using VRMA), so we always assume the latter
1977          *
1978          * In that case, we also limit the initial allocations for RTAS
1979          * etc... to 256M since we have no way to know what the VRMA size
1980          * is going to be as it depends on the size of the hash table
1981          * isn't determined yet.
1982          */
1983         if (kvm_enabled()) {
1984             spapr->vrma_adjust = 1;
1985             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1986         }
1987 
1988         /* Actually we don't support unbounded RMA anymore since we
1989          * added proper emulation of HV mode. The max we can get is
1990          * 16G which also happens to be what we configure for PAPR
1991          * mode so make sure we don't do anything bigger than that
1992          */
1993         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1994     }
1995 
1996     if (spapr->rma_size > node0_size) {
1997         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1998                      spapr->rma_size);
1999         exit(1);
2000     }
2001 
2002     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2003     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2004 
2005     /* Set up Interrupt Controller before we create the VCPUs */
2006     spapr->xics = xics_system_init(machine,
2007                                    DIV_ROUND_UP(max_cpus * smt, smp_threads),
2008                                    XICS_IRQS_SPAPR, &error_fatal);
2009 
2010     /* Set up containers for ibm,client-set-architecture negotiated options */
2011     spapr->ov5 = spapr_ovec_new();
2012     spapr->ov5_cas = spapr_ovec_new();
2013 
2014     if (smc->dr_lmb_enabled) {
2015         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2016         spapr_validate_node_memory(machine, &error_fatal);
2017     }
2018 
2019     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2020 
2021     /* advertise support for dedicated HP event source to guests */
2022     if (spapr->use_hotplug_event_source) {
2023         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2024     }
2025 
2026     /* init CPUs */
2027     if (machine->cpu_model == NULL) {
2028         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2029     }
2030 
2031     ppc_cpu_parse_features(machine->cpu_model);
2032 
2033     spapr_init_cpus(spapr);
2034 
2035     if (kvm_enabled()) {
2036         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2037         kvmppc_enable_logical_ci_hcalls();
2038         kvmppc_enable_set_mode_hcall();
2039 
2040         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2041         kvmppc_enable_clear_ref_mod_hcalls();
2042     }
2043 
2044     /* allocate RAM */
2045     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2046                                          machine->ram_size);
2047     memory_region_add_subregion(sysmem, 0, ram);
2048 
2049     if (rma_alloc_size && rma) {
2050         rma_region = g_new(MemoryRegion, 1);
2051         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2052                                    rma_alloc_size, rma);
2053         vmstate_register_ram_global(rma_region);
2054         memory_region_add_subregion(sysmem, 0, rma_region);
2055     }
2056 
2057     /* initialize hotplug memory address space */
2058     if (machine->ram_size < machine->maxram_size) {
2059         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2060         /*
2061          * Limit the number of hotpluggable memory slots to half the number
2062          * slots that KVM supports, leaving the other half for PCI and other
2063          * devices. However ensure that number of slots doesn't drop below 32.
2064          */
2065         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2066                            SPAPR_MAX_RAM_SLOTS;
2067 
2068         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2069             max_memslots = SPAPR_MAX_RAM_SLOTS;
2070         }
2071         if (machine->ram_slots > max_memslots) {
2072             error_report("Specified number of memory slots %"
2073                          PRIu64" exceeds max supported %d",
2074                          machine->ram_slots, max_memslots);
2075             exit(1);
2076         }
2077 
2078         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2079                                               SPAPR_HOTPLUG_MEM_ALIGN);
2080         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2081                            "hotplug-memory", hotplug_mem_size);
2082         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2083                                     &spapr->hotplug_memory.mr);
2084     }
2085 
2086     if (smc->dr_lmb_enabled) {
2087         spapr_create_lmb_dr_connectors(spapr);
2088     }
2089 
2090     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2091     if (!filename) {
2092         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2093         exit(1);
2094     }
2095     spapr->rtas_size = get_image_size(filename);
2096     if (spapr->rtas_size < 0) {
2097         error_report("Could not get size of LPAR rtas '%s'", filename);
2098         exit(1);
2099     }
2100     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2101     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2102         error_report("Could not load LPAR rtas '%s'", filename);
2103         exit(1);
2104     }
2105     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2106         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2107                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2108         exit(1);
2109     }
2110     g_free(filename);
2111 
2112     /* Set up RTAS event infrastructure */
2113     spapr_events_init(spapr);
2114 
2115     /* Set up the RTC RTAS interfaces */
2116     spapr_rtc_create(spapr);
2117 
2118     /* Set up VIO bus */
2119     spapr->vio_bus = spapr_vio_bus_init();
2120 
2121     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2122         if (serial_hds[i]) {
2123             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2124         }
2125     }
2126 
2127     /* We always have at least the nvram device on VIO */
2128     spapr_create_nvram(spapr);
2129 
2130     /* Set up PCI */
2131     spapr_pci_rtas_init();
2132 
2133     phb = spapr_create_phb(spapr, 0);
2134 
2135     for (i = 0; i < nb_nics; i++) {
2136         NICInfo *nd = &nd_table[i];
2137 
2138         if (!nd->model) {
2139             nd->model = g_strdup("ibmveth");
2140         }
2141 
2142         if (strcmp(nd->model, "ibmveth") == 0) {
2143             spapr_vlan_create(spapr->vio_bus, nd);
2144         } else {
2145             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2146         }
2147     }
2148 
2149     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2150         spapr_vscsi_create(spapr->vio_bus);
2151     }
2152 
2153     /* Graphics */
2154     if (spapr_vga_init(phb->bus, &error_fatal)) {
2155         spapr->has_graphics = true;
2156         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2157     }
2158 
2159     if (machine->usb) {
2160         if (smc->use_ohci_by_default) {
2161             pci_create_simple(phb->bus, -1, "pci-ohci");
2162         } else {
2163             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2164         }
2165 
2166         if (spapr->has_graphics) {
2167             USBBus *usb_bus = usb_bus_find(-1);
2168 
2169             usb_create_simple(usb_bus, "usb-kbd");
2170             usb_create_simple(usb_bus, "usb-mouse");
2171         }
2172     }
2173 
2174     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2175         error_report(
2176             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2177             MIN_RMA_SLOF);
2178         exit(1);
2179     }
2180 
2181     if (kernel_filename) {
2182         uint64_t lowaddr = 0;
2183 
2184         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2185                                       NULL, NULL, &lowaddr, NULL, 1,
2186                                       PPC_ELF_MACHINE, 0, 0);
2187         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2188             spapr->kernel_size = load_elf(kernel_filename,
2189                                           translate_kernel_address, NULL, NULL,
2190                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2191                                           0, 0);
2192             spapr->kernel_le = spapr->kernel_size > 0;
2193         }
2194         if (spapr->kernel_size < 0) {
2195             error_report("error loading %s: %s", kernel_filename,
2196                          load_elf_strerror(spapr->kernel_size));
2197             exit(1);
2198         }
2199 
2200         /* load initrd */
2201         if (initrd_filename) {
2202             /* Try to locate the initrd in the gap between the kernel
2203              * and the firmware. Add a bit of space just in case
2204              */
2205             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2206                                   + 0x1ffff) & ~0xffff;
2207             spapr->initrd_size = load_image_targphys(initrd_filename,
2208                                                      spapr->initrd_base,
2209                                                      load_limit
2210                                                      - spapr->initrd_base);
2211             if (spapr->initrd_size < 0) {
2212                 error_report("could not load initial ram disk '%s'",
2213                              initrd_filename);
2214                 exit(1);
2215             }
2216         }
2217     }
2218 
2219     if (bios_name == NULL) {
2220         bios_name = FW_FILE_NAME;
2221     }
2222     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2223     if (!filename) {
2224         error_report("Could not find LPAR firmware '%s'", bios_name);
2225         exit(1);
2226     }
2227     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2228     if (fw_size <= 0) {
2229         error_report("Could not load LPAR firmware '%s'", filename);
2230         exit(1);
2231     }
2232     g_free(filename);
2233 
2234     /* FIXME: Should register things through the MachineState's qdev
2235      * interface, this is a legacy from the sPAPREnvironment structure
2236      * which predated MachineState but had a similar function */
2237     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2238     register_savevm_live(NULL, "spapr/htab", -1, 1,
2239                          &savevm_htab_handlers, spapr);
2240 
2241     /* used by RTAS */
2242     QTAILQ_INIT(&spapr->ccs_list);
2243     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2244 
2245     qemu_register_boot_set(spapr_boot_set, spapr);
2246 
2247     /* to stop and start vmclock */
2248     if (kvm_enabled()) {
2249         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2250                                          &spapr->tb);
2251     }
2252 }
2253 
2254 static int spapr_kvm_type(const char *vm_type)
2255 {
2256     if (!vm_type) {
2257         return 0;
2258     }
2259 
2260     if (!strcmp(vm_type, "HV")) {
2261         return 1;
2262     }
2263 
2264     if (!strcmp(vm_type, "PR")) {
2265         return 2;
2266     }
2267 
2268     error_report("Unknown kvm-type specified '%s'", vm_type);
2269     exit(1);
2270 }
2271 
2272 /*
2273  * Implementation of an interface to adjust firmware path
2274  * for the bootindex property handling.
2275  */
2276 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2277                                    DeviceState *dev)
2278 {
2279 #define CAST(type, obj, name) \
2280     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2281     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2282     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2283 
2284     if (d) {
2285         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2286         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2287         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2288 
2289         if (spapr) {
2290             /*
2291              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2292              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2293              * in the top 16 bits of the 64-bit LUN
2294              */
2295             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2296             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2297                                    (uint64_t)id << 48);
2298         } else if (virtio) {
2299             /*
2300              * We use SRP luns of the form 01000000 | (target << 8) | lun
2301              * in the top 32 bits of the 64-bit LUN
2302              * Note: the quote above is from SLOF and it is wrong,
2303              * the actual binding is:
2304              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2305              */
2306             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2307             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2308                                    (uint64_t)id << 32);
2309         } else if (usb) {
2310             /*
2311              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2312              * in the top 32 bits of the 64-bit LUN
2313              */
2314             unsigned usb_port = atoi(usb->port->path);
2315             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2316             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2317                                    (uint64_t)id << 32);
2318         }
2319     }
2320 
2321     /*
2322      * SLOF probes the USB devices, and if it recognizes that the device is a
2323      * storage device, it changes its name to "storage" instead of "usb-host",
2324      * and additionally adds a child node for the SCSI LUN, so the correct
2325      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2326      */
2327     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2328         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2329         if (usb_host_dev_is_scsi_storage(usbdev)) {
2330             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2331         }
2332     }
2333 
2334     if (phb) {
2335         /* Replace "pci" with "pci@800000020000000" */
2336         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2337     }
2338 
2339     return NULL;
2340 }
2341 
2342 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2343 {
2344     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2345 
2346     return g_strdup(spapr->kvm_type);
2347 }
2348 
2349 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2350 {
2351     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2352 
2353     g_free(spapr->kvm_type);
2354     spapr->kvm_type = g_strdup(value);
2355 }
2356 
2357 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2358 {
2359     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2360 
2361     return spapr->use_hotplug_event_source;
2362 }
2363 
2364 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2365                                             Error **errp)
2366 {
2367     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2368 
2369     spapr->use_hotplug_event_source = value;
2370 }
2371 
2372 static void spapr_machine_initfn(Object *obj)
2373 {
2374     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2375 
2376     spapr->htab_fd = -1;
2377     spapr->use_hotplug_event_source = true;
2378     object_property_add_str(obj, "kvm-type",
2379                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2380     object_property_set_description(obj, "kvm-type",
2381                                     "Specifies the KVM virtualization mode (HV, PR)",
2382                                     NULL);
2383     object_property_add_bool(obj, "modern-hotplug-events",
2384                             spapr_get_modern_hotplug_events,
2385                             spapr_set_modern_hotplug_events,
2386                             NULL);
2387     object_property_set_description(obj, "modern-hotplug-events",
2388                                     "Use dedicated hotplug event mechanism in"
2389                                     " place of standard EPOW events when possible"
2390                                     " (required for memory hot-unplug support)",
2391                                     NULL);
2392 }
2393 
2394 static void spapr_machine_finalizefn(Object *obj)
2395 {
2396     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2397 
2398     g_free(spapr->kvm_type);
2399 }
2400 
2401 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2402 {
2403     cpu_synchronize_state(cs);
2404     ppc_cpu_do_system_reset(cs);
2405 }
2406 
2407 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2408 {
2409     CPUState *cs;
2410 
2411     CPU_FOREACH(cs) {
2412         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2413     }
2414 }
2415 
2416 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2417                            uint32_t node, bool dedicated_hp_event_source,
2418                            Error **errp)
2419 {
2420     sPAPRDRConnector *drc;
2421     sPAPRDRConnectorClass *drck;
2422     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2423     int i, fdt_offset, fdt_size;
2424     void *fdt;
2425     uint64_t addr = addr_start;
2426 
2427     for (i = 0; i < nr_lmbs; i++) {
2428         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2429                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2430         g_assert(drc);
2431 
2432         fdt = create_device_tree(&fdt_size);
2433         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2434                                                 SPAPR_MEMORY_BLOCK_SIZE);
2435 
2436         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2437         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2438         addr += SPAPR_MEMORY_BLOCK_SIZE;
2439         if (!dev->hotplugged) {
2440             /* guests expect coldplugged LMBs to be pre-allocated */
2441             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2442             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2443         }
2444     }
2445     /* send hotplug notification to the
2446      * guest only in case of hotplugged memory
2447      */
2448     if (dev->hotplugged) {
2449         if (dedicated_hp_event_source) {
2450             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2451                     addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2452             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2453             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2454                                                    nr_lmbs,
2455                                                    drck->get_index(drc));
2456         } else {
2457             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2458                                            nr_lmbs);
2459         }
2460     }
2461 }
2462 
2463 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2464                               uint32_t node, Error **errp)
2465 {
2466     Error *local_err = NULL;
2467     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2468     PCDIMMDevice *dimm = PC_DIMM(dev);
2469     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2470     MemoryRegion *mr = ddc->get_memory_region(dimm);
2471     uint64_t align = memory_region_get_alignment(mr);
2472     uint64_t size = memory_region_size(mr);
2473     uint64_t addr;
2474     char *mem_dev;
2475 
2476     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2477         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2478                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2479         goto out;
2480     }
2481 
2482     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2483     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2484         error_setg(&local_err, "Memory backend has bad page size. "
2485                    "Use 'memory-backend-file' with correct mem-path.");
2486         goto out;
2487     }
2488 
2489     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2490     if (local_err) {
2491         goto out;
2492     }
2493 
2494     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2495     if (local_err) {
2496         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2497         goto out;
2498     }
2499 
2500     spapr_add_lmbs(dev, addr, size, node,
2501                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2502                    &error_abort);
2503 
2504 out:
2505     error_propagate(errp, local_err);
2506 }
2507 
2508 typedef struct sPAPRDIMMState {
2509     uint32_t nr_lmbs;
2510 } sPAPRDIMMState;
2511 
2512 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2513 {
2514     sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2515     HotplugHandler *hotplug_ctrl;
2516 
2517     if (--ds->nr_lmbs) {
2518         return;
2519     }
2520 
2521     g_free(ds);
2522 
2523     /*
2524      * Now that all the LMBs have been removed by the guest, call the
2525      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2526      */
2527     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2528     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2529 }
2530 
2531 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2532                            Error **errp)
2533 {
2534     sPAPRDRConnector *drc;
2535     sPAPRDRConnectorClass *drck;
2536     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2537     int i;
2538     sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2539     uint64_t addr = addr_start;
2540 
2541     ds->nr_lmbs = nr_lmbs;
2542     for (i = 0; i < nr_lmbs; i++) {
2543         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2544                 addr / SPAPR_MEMORY_BLOCK_SIZE);
2545         g_assert(drc);
2546 
2547         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2548         drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2549         addr += SPAPR_MEMORY_BLOCK_SIZE;
2550     }
2551 
2552     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2553                                    addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2554     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2555     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2556                                               nr_lmbs,
2557                                               drck->get_index(drc));
2558 }
2559 
2560 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2561                                 Error **errp)
2562 {
2563     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2564     PCDIMMDevice *dimm = PC_DIMM(dev);
2565     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2566     MemoryRegion *mr = ddc->get_memory_region(dimm);
2567 
2568     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2569     object_unparent(OBJECT(dev));
2570 }
2571 
2572 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2573                                         DeviceState *dev, Error **errp)
2574 {
2575     Error *local_err = NULL;
2576     PCDIMMDevice *dimm = PC_DIMM(dev);
2577     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2578     MemoryRegion *mr = ddc->get_memory_region(dimm);
2579     uint64_t size = memory_region_size(mr);
2580     uint64_t addr;
2581 
2582     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2583     if (local_err) {
2584         goto out;
2585     }
2586 
2587     spapr_del_lmbs(dev, addr, size, &error_abort);
2588 out:
2589     error_propagate(errp, local_err);
2590 }
2591 
2592 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2593                                     sPAPRMachineState *spapr)
2594 {
2595     PowerPCCPU *cpu = POWERPC_CPU(cs);
2596     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2597     int id = ppc_get_vcpu_dt_id(cpu);
2598     void *fdt;
2599     int offset, fdt_size;
2600     char *nodename;
2601 
2602     fdt = create_device_tree(&fdt_size);
2603     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2604     offset = fdt_add_subnode(fdt, 0, nodename);
2605 
2606     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2607     g_free(nodename);
2608 
2609     *fdt_offset = offset;
2610     return fdt;
2611 }
2612 
2613 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2614                               Error **errp)
2615 {
2616     MachineState *ms = MACHINE(qdev_get_machine());
2617     CPUCore *cc = CPU_CORE(dev);
2618     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2619 
2620     core_slot->cpu = NULL;
2621     object_unparent(OBJECT(dev));
2622 }
2623 
2624 static void spapr_core_release(DeviceState *dev, void *opaque)
2625 {
2626     HotplugHandler *hotplug_ctrl;
2627 
2628     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2629     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2630 }
2631 
2632 static
2633 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2634                                Error **errp)
2635 {
2636     int index;
2637     sPAPRDRConnector *drc;
2638     sPAPRDRConnectorClass *drck;
2639     Error *local_err = NULL;
2640     CPUCore *cc = CPU_CORE(dev);
2641     int smt = kvmppc_smt_threads();
2642 
2643     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2644         error_setg(errp, "Unable to find CPU core with core-id: %d",
2645                    cc->core_id);
2646         return;
2647     }
2648     if (index == 0) {
2649         error_setg(errp, "Boot CPU core may not be unplugged");
2650         return;
2651     }
2652 
2653     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2654     g_assert(drc);
2655 
2656     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2657     drck->detach(drc, dev, spapr_core_release, NULL, &local_err);
2658     if (local_err) {
2659         error_propagate(errp, local_err);
2660         return;
2661     }
2662 
2663     spapr_hotplug_req_remove_by_index(drc);
2664 }
2665 
2666 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2667                             Error **errp)
2668 {
2669     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2670     MachineClass *mc = MACHINE_GET_CLASS(spapr);
2671     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2672     CPUCore *cc = CPU_CORE(dev);
2673     CPUState *cs = CPU(core->threads);
2674     sPAPRDRConnector *drc;
2675     Error *local_err = NULL;
2676     void *fdt = NULL;
2677     int fdt_offset = 0;
2678     int smt = kvmppc_smt_threads();
2679     CPUArchId *core_slot;
2680     int index;
2681 
2682     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2683     if (!core_slot) {
2684         error_setg(errp, "Unable to find CPU core with core-id: %d",
2685                    cc->core_id);
2686         return;
2687     }
2688     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2689 
2690     g_assert(drc || !mc->has_hotpluggable_cpus);
2691 
2692     /*
2693      * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2694      * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2695      */
2696     if (dev->hotplugged) {
2697         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2698     }
2699 
2700     if (drc) {
2701         sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2702         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
2703         if (local_err) {
2704             g_free(fdt);
2705             error_propagate(errp, local_err);
2706             return;
2707         }
2708     }
2709 
2710     if (dev->hotplugged) {
2711         /*
2712          * Send hotplug notification interrupt to the guest only in case
2713          * of hotplugged CPUs.
2714          */
2715         spapr_hotplug_req_add_by_index(drc);
2716     } else {
2717         /*
2718          * Set the right DRC states for cold plugged CPU.
2719          */
2720         if (drc) {
2721             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2722             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2723             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2724         }
2725     }
2726     core_slot->cpu = OBJECT(dev);
2727 }
2728 
2729 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2730                                 Error **errp)
2731 {
2732     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2733     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2734     Error *local_err = NULL;
2735     CPUCore *cc = CPU_CORE(dev);
2736     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2737     const char *type = object_get_typename(OBJECT(dev));
2738     CPUArchId *core_slot;
2739     int index;
2740 
2741     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2742         error_setg(&local_err, "CPU hotplug not supported for this machine");
2743         goto out;
2744     }
2745 
2746     if (strcmp(base_core_type, type)) {
2747         error_setg(&local_err, "CPU core type should be %s", base_core_type);
2748         goto out;
2749     }
2750 
2751     if (cc->core_id % smp_threads) {
2752         error_setg(&local_err, "invalid core id %d", cc->core_id);
2753         goto out;
2754     }
2755 
2756     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2757     if (!core_slot) {
2758         error_setg(&local_err, "core id %d out of range", cc->core_id);
2759         goto out;
2760     }
2761 
2762     if (core_slot->cpu) {
2763         error_setg(&local_err, "core %d already populated", cc->core_id);
2764         goto out;
2765     }
2766 
2767 out:
2768     g_free(base_core_type);
2769     error_propagate(errp, local_err);
2770 }
2771 
2772 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2773                                       DeviceState *dev, Error **errp)
2774 {
2775     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2776 
2777     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2778         int node;
2779 
2780         if (!smc->dr_lmb_enabled) {
2781             error_setg(errp, "Memory hotplug not supported for this machine");
2782             return;
2783         }
2784         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2785         if (*errp) {
2786             return;
2787         }
2788         if (node < 0 || node >= MAX_NODES) {
2789             error_setg(errp, "Invaild node %d", node);
2790             return;
2791         }
2792 
2793         /*
2794          * Currently PowerPC kernel doesn't allow hot-adding memory to
2795          * memory-less node, but instead will silently add the memory
2796          * to the first node that has some memory. This causes two
2797          * unexpected behaviours for the user.
2798          *
2799          * - Memory gets hotplugged to a different node than what the user
2800          *   specified.
2801          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2802          *   to memory-less node, a reboot will set things accordingly
2803          *   and the previously hotplugged memory now ends in the right node.
2804          *   This appears as if some memory moved from one node to another.
2805          *
2806          * So until kernel starts supporting memory hotplug to memory-less
2807          * nodes, just prevent such attempts upfront in QEMU.
2808          */
2809         if (nb_numa_nodes && !numa_info[node].node_mem) {
2810             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2811                        node);
2812             return;
2813         }
2814 
2815         spapr_memory_plug(hotplug_dev, dev, node, errp);
2816     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2817         spapr_core_plug(hotplug_dev, dev, errp);
2818     }
2819 }
2820 
2821 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2822                                       DeviceState *dev, Error **errp)
2823 {
2824     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2825     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2826 
2827     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2828         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2829             spapr_memory_unplug(hotplug_dev, dev, errp);
2830         } else {
2831             error_setg(errp, "Memory hot unplug not supported for this guest");
2832         }
2833     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2834         if (!mc->has_hotpluggable_cpus) {
2835             error_setg(errp, "CPU hot unplug not supported on this machine");
2836             return;
2837         }
2838         spapr_core_unplug(hotplug_dev, dev, errp);
2839     }
2840 }
2841 
2842 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2843                                                 DeviceState *dev, Error **errp)
2844 {
2845     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2846     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2847 
2848     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2849         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2850             spapr_memory_unplug_request(hotplug_dev, dev, errp);
2851         } else {
2852             /* NOTE: this means there is a window after guest reset, prior to
2853              * CAS negotiation, where unplug requests will fail due to the
2854              * capability not being detected yet. This is a bit different than
2855              * the case with PCI unplug, where the events will be queued and
2856              * eventually handled by the guest after boot
2857              */
2858             error_setg(errp, "Memory hot unplug not supported for this guest");
2859         }
2860     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2861         if (!mc->has_hotpluggable_cpus) {
2862             error_setg(errp, "CPU hot unplug not supported on this machine");
2863             return;
2864         }
2865         spapr_core_unplug_request(hotplug_dev, dev, errp);
2866     }
2867 }
2868 
2869 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2870                                           DeviceState *dev, Error **errp)
2871 {
2872     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2873         spapr_core_pre_plug(hotplug_dev, dev, errp);
2874     }
2875 }
2876 
2877 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2878                                                  DeviceState *dev)
2879 {
2880     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2881         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2882         return HOTPLUG_HANDLER(machine);
2883     }
2884     return NULL;
2885 }
2886 
2887 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2888 {
2889     /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2890      * socket means much for the paravirtualized PAPR platform) */
2891     return cpu_index / smp_threads / smp_cores;
2892 }
2893 
2894 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
2895 {
2896     int i;
2897     int spapr_max_cores = max_cpus / smp_threads;
2898     MachineClass *mc = MACHINE_GET_CLASS(machine);
2899 
2900     if (!mc->has_hotpluggable_cpus) {
2901         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
2902     }
2903     if (machine->possible_cpus) {
2904         assert(machine->possible_cpus->len == spapr_max_cores);
2905         return machine->possible_cpus;
2906     }
2907 
2908     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2909                              sizeof(CPUArchId) * spapr_max_cores);
2910     machine->possible_cpus->len = spapr_max_cores;
2911     for (i = 0; i < machine->possible_cpus->len; i++) {
2912         int core_id = i * smp_threads;
2913 
2914         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
2915         machine->possible_cpus->cpus[i].arch_id = core_id;
2916         machine->possible_cpus->cpus[i].props.has_core_id = true;
2917         machine->possible_cpus->cpus[i].props.core_id = core_id;
2918         /* TODO: add 'has_node/node' here to describe
2919            to which node core belongs */
2920     }
2921     return machine->possible_cpus;
2922 }
2923 
2924 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
2925                                 uint64_t *buid, hwaddr *pio,
2926                                 hwaddr *mmio32, hwaddr *mmio64,
2927                                 unsigned n_dma, uint32_t *liobns, Error **errp)
2928 {
2929     /*
2930      * New-style PHB window placement.
2931      *
2932      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
2933      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
2934      * windows.
2935      *
2936      * Some guest kernels can't work with MMIO windows above 1<<46
2937      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
2938      *
2939      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
2940      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
2941      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
2942      * 1TiB 64-bit MMIO windows for each PHB.
2943      */
2944     const uint64_t base_buid = 0x800000020000000ULL;
2945 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
2946                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
2947     int i;
2948 
2949     /* Sanity check natural alignments */
2950     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2951     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2952     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
2953     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
2954     /* Sanity check bounds */
2955     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
2956                       SPAPR_PCI_MEM32_WIN_SIZE);
2957     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
2958                       SPAPR_PCI_MEM64_WIN_SIZE);
2959 
2960     if (index >= SPAPR_MAX_PHBS) {
2961         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
2962                    SPAPR_MAX_PHBS - 1);
2963         return;
2964     }
2965 
2966     *buid = base_buid + index;
2967     for (i = 0; i < n_dma; ++i) {
2968         liobns[i] = SPAPR_PCI_LIOBN(index, i);
2969     }
2970 
2971     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
2972     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
2973     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
2974 }
2975 
2976 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
2977 {
2978     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2979 
2980     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
2981 }
2982 
2983 static void spapr_ics_resend(XICSFabric *dev)
2984 {
2985     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2986 
2987     ics_resend(spapr->ics);
2988 }
2989 
2990 static ICPState *spapr_icp_get(XICSFabric *xi, int server)
2991 {
2992     sPAPRMachineState *spapr = SPAPR_MACHINE(xi);
2993 
2994     return (server < spapr->nr_servers) ? &spapr->icps[server] : NULL;
2995 }
2996 
2997 static void spapr_icp_resend(XICSFabric *xi)
2998 {
2999     sPAPRMachineState *spapr = SPAPR_MACHINE(xi);
3000     int i;
3001 
3002     for (i = 0; i < spapr->nr_servers; i++) {
3003         icp_resend(&spapr->icps[i]);
3004     }
3005 }
3006 
3007 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3008 {
3009     MachineClass *mc = MACHINE_CLASS(oc);
3010     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3011     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3012     NMIClass *nc = NMI_CLASS(oc);
3013     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3014     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3015     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3016 
3017     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3018 
3019     /*
3020      * We set up the default / latest behaviour here.  The class_init
3021      * functions for the specific versioned machine types can override
3022      * these details for backwards compatibility
3023      */
3024     mc->init = ppc_spapr_init;
3025     mc->reset = ppc_spapr_reset;
3026     mc->block_default_type = IF_SCSI;
3027     mc->max_cpus = 1024;
3028     mc->no_parallel = 1;
3029     mc->default_boot_order = "";
3030     mc->default_ram_size = 512 * M_BYTE;
3031     mc->kvm_type = spapr_kvm_type;
3032     mc->has_dynamic_sysbus = true;
3033     mc->pci_allow_0_address = true;
3034     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3035     hc->pre_plug = spapr_machine_device_pre_plug;
3036     hc->plug = spapr_machine_device_plug;
3037     hc->unplug = spapr_machine_device_unplug;
3038     mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
3039     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3040     hc->unplug_request = spapr_machine_device_unplug_request;
3041 
3042     smc->dr_lmb_enabled = true;
3043     smc->tcg_default_cpu = "POWER8";
3044     mc->has_hotpluggable_cpus = true;
3045     fwc->get_dev_path = spapr_get_fw_dev_path;
3046     nc->nmi_monitor_handler = spapr_nmi;
3047     smc->phb_placement = spapr_phb_placement;
3048     vhc->hypercall = emulate_spapr_hypercall;
3049     vhc->hpt_mask = spapr_hpt_mask;
3050     vhc->map_hptes = spapr_map_hptes;
3051     vhc->unmap_hptes = spapr_unmap_hptes;
3052     vhc->store_hpte = spapr_store_hpte;
3053     xic->ics_get = spapr_ics_get;
3054     xic->ics_resend = spapr_ics_resend;
3055     xic->icp_get = spapr_icp_get;
3056     xic->icp_resend = spapr_icp_resend;
3057 }
3058 
3059 static const TypeInfo spapr_machine_info = {
3060     .name          = TYPE_SPAPR_MACHINE,
3061     .parent        = TYPE_MACHINE,
3062     .abstract      = true,
3063     .instance_size = sizeof(sPAPRMachineState),
3064     .instance_init = spapr_machine_initfn,
3065     .instance_finalize = spapr_machine_finalizefn,
3066     .class_size    = sizeof(sPAPRMachineClass),
3067     .class_init    = spapr_machine_class_init,
3068     .interfaces = (InterfaceInfo[]) {
3069         { TYPE_FW_PATH_PROVIDER },
3070         { TYPE_NMI },
3071         { TYPE_HOTPLUG_HANDLER },
3072         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3073         { TYPE_XICS_FABRIC },
3074         { }
3075     },
3076 };
3077 
3078 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3079     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3080                                                     void *data)      \
3081     {                                                                \
3082         MachineClass *mc = MACHINE_CLASS(oc);                        \
3083         spapr_machine_##suffix##_class_options(mc);                  \
3084         if (latest) {                                                \
3085             mc->alias = "pseries";                                   \
3086             mc->is_default = 1;                                      \
3087         }                                                            \
3088     }                                                                \
3089     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3090     {                                                                \
3091         MachineState *machine = MACHINE(obj);                        \
3092         spapr_machine_##suffix##_instance_options(machine);          \
3093     }                                                                \
3094     static const TypeInfo spapr_machine_##suffix##_info = {          \
3095         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3096         .parent = TYPE_SPAPR_MACHINE,                                \
3097         .class_init = spapr_machine_##suffix##_class_init,           \
3098         .instance_init = spapr_machine_##suffix##_instance_init,     \
3099     };                                                               \
3100     static void spapr_machine_register_##suffix(void)                \
3101     {                                                                \
3102         type_register(&spapr_machine_##suffix##_info);               \
3103     }                                                                \
3104     type_init(spapr_machine_register_##suffix)
3105 
3106 /*
3107  * pseries-2.9
3108  */
3109 static void spapr_machine_2_9_instance_options(MachineState *machine)
3110 {
3111 }
3112 
3113 static void spapr_machine_2_9_class_options(MachineClass *mc)
3114 {
3115     /* Defaults for the latest behaviour inherited from the base class */
3116 }
3117 
3118 DEFINE_SPAPR_MACHINE(2_9, "2.9", true);
3119 
3120 /*
3121  * pseries-2.8
3122  */
3123 #define SPAPR_COMPAT_2_8                            \
3124     HW_COMPAT_2_8
3125 
3126 static void spapr_machine_2_8_instance_options(MachineState *machine)
3127 {
3128     spapr_machine_2_9_instance_options(machine);
3129 }
3130 
3131 static void spapr_machine_2_8_class_options(MachineClass *mc)
3132 {
3133     spapr_machine_2_9_class_options(mc);
3134     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3135 }
3136 
3137 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3138 
3139 /*
3140  * pseries-2.7
3141  */
3142 #define SPAPR_COMPAT_2_7                            \
3143     HW_COMPAT_2_7                                   \
3144     {                                               \
3145         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3146         .property = "mem_win_size",                 \
3147         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3148     },                                              \
3149     {                                               \
3150         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3151         .property = "mem64_win_size",               \
3152         .value    = "0",                            \
3153     },                                              \
3154     {                                               \
3155         .driver = TYPE_POWERPC_CPU,                 \
3156         .property = "pre-2.8-migration",            \
3157         .value    = "on",                           \
3158     },                                              \
3159     {                                               \
3160         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3161         .property = "pre-2.8-migration",            \
3162         .value    = "on",                           \
3163     },
3164 
3165 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3166                               uint64_t *buid, hwaddr *pio,
3167                               hwaddr *mmio32, hwaddr *mmio64,
3168                               unsigned n_dma, uint32_t *liobns, Error **errp)
3169 {
3170     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3171     const uint64_t base_buid = 0x800000020000000ULL;
3172     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3173     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3174     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3175     const uint32_t max_index = 255;
3176     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3177 
3178     uint64_t ram_top = MACHINE(spapr)->ram_size;
3179     hwaddr phb0_base, phb_base;
3180     int i;
3181 
3182     /* Do we have hotpluggable memory? */
3183     if (MACHINE(spapr)->maxram_size > ram_top) {
3184         /* Can't just use maxram_size, because there may be an
3185          * alignment gap between normal and hotpluggable memory
3186          * regions */
3187         ram_top = spapr->hotplug_memory.base +
3188             memory_region_size(&spapr->hotplug_memory.mr);
3189     }
3190 
3191     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3192 
3193     if (index > max_index) {
3194         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3195                    max_index);
3196         return;
3197     }
3198 
3199     *buid = base_buid + index;
3200     for (i = 0; i < n_dma; ++i) {
3201         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3202     }
3203 
3204     phb_base = phb0_base + index * phb_spacing;
3205     *pio = phb_base + pio_offset;
3206     *mmio32 = phb_base + mmio_offset;
3207     /*
3208      * We don't set the 64-bit MMIO window, relying on the PHB's
3209      * fallback behaviour of automatically splitting a large "32-bit"
3210      * window into contiguous 32-bit and 64-bit windows
3211      */
3212 }
3213 
3214 static void spapr_machine_2_7_instance_options(MachineState *machine)
3215 {
3216     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3217 
3218     spapr_machine_2_8_instance_options(machine);
3219     spapr->use_hotplug_event_source = false;
3220 }
3221 
3222 static void spapr_machine_2_7_class_options(MachineClass *mc)
3223 {
3224     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3225 
3226     spapr_machine_2_8_class_options(mc);
3227     smc->tcg_default_cpu = "POWER7";
3228     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3229     smc->phb_placement = phb_placement_2_7;
3230 }
3231 
3232 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3233 
3234 /*
3235  * pseries-2.6
3236  */
3237 #define SPAPR_COMPAT_2_6 \
3238     HW_COMPAT_2_6 \
3239     { \
3240         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3241         .property = "ddw",\
3242         .value    = stringify(off),\
3243     },
3244 
3245 static void spapr_machine_2_6_instance_options(MachineState *machine)
3246 {
3247     spapr_machine_2_7_instance_options(machine);
3248 }
3249 
3250 static void spapr_machine_2_6_class_options(MachineClass *mc)
3251 {
3252     spapr_machine_2_7_class_options(mc);
3253     mc->has_hotpluggable_cpus = false;
3254     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3255 }
3256 
3257 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3258 
3259 /*
3260  * pseries-2.5
3261  */
3262 #define SPAPR_COMPAT_2_5 \
3263     HW_COMPAT_2_5 \
3264     { \
3265         .driver   = "spapr-vlan", \
3266         .property = "use-rx-buffer-pools", \
3267         .value    = "off", \
3268     },
3269 
3270 static void spapr_machine_2_5_instance_options(MachineState *machine)
3271 {
3272     spapr_machine_2_6_instance_options(machine);
3273 }
3274 
3275 static void spapr_machine_2_5_class_options(MachineClass *mc)
3276 {
3277     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3278 
3279     spapr_machine_2_6_class_options(mc);
3280     smc->use_ohci_by_default = true;
3281     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3282 }
3283 
3284 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3285 
3286 /*
3287  * pseries-2.4
3288  */
3289 #define SPAPR_COMPAT_2_4 \
3290         HW_COMPAT_2_4
3291 
3292 static void spapr_machine_2_4_instance_options(MachineState *machine)
3293 {
3294     spapr_machine_2_5_instance_options(machine);
3295 }
3296 
3297 static void spapr_machine_2_4_class_options(MachineClass *mc)
3298 {
3299     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3300 
3301     spapr_machine_2_5_class_options(mc);
3302     smc->dr_lmb_enabled = false;
3303     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3304 }
3305 
3306 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3307 
3308 /*
3309  * pseries-2.3
3310  */
3311 #define SPAPR_COMPAT_2_3 \
3312         HW_COMPAT_2_3 \
3313         {\
3314             .driver   = "spapr-pci-host-bridge",\
3315             .property = "dynamic-reconfiguration",\
3316             .value    = "off",\
3317         },
3318 
3319 static void spapr_machine_2_3_instance_options(MachineState *machine)
3320 {
3321     spapr_machine_2_4_instance_options(machine);
3322     savevm_skip_section_footers();
3323     global_state_set_optional();
3324     savevm_skip_configuration();
3325 }
3326 
3327 static void spapr_machine_2_3_class_options(MachineClass *mc)
3328 {
3329     spapr_machine_2_4_class_options(mc);
3330     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3331 }
3332 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3333 
3334 /*
3335  * pseries-2.2
3336  */
3337 
3338 #define SPAPR_COMPAT_2_2 \
3339         HW_COMPAT_2_2 \
3340         {\
3341             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3342             .property = "mem_win_size",\
3343             .value    = "0x20000000",\
3344         },
3345 
3346 static void spapr_machine_2_2_instance_options(MachineState *machine)
3347 {
3348     spapr_machine_2_3_instance_options(machine);
3349     machine->suppress_vmdesc = true;
3350 }
3351 
3352 static void spapr_machine_2_2_class_options(MachineClass *mc)
3353 {
3354     spapr_machine_2_3_class_options(mc);
3355     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3356 }
3357 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3358 
3359 /*
3360  * pseries-2.1
3361  */
3362 #define SPAPR_COMPAT_2_1 \
3363         HW_COMPAT_2_1
3364 
3365 static void spapr_machine_2_1_instance_options(MachineState *machine)
3366 {
3367     spapr_machine_2_2_instance_options(machine);
3368 }
3369 
3370 static void spapr_machine_2_1_class_options(MachineClass *mc)
3371 {
3372     spapr_machine_2_2_class_options(mc);
3373     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3374 }
3375 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3376 
3377 static void spapr_machine_register_types(void)
3378 {
3379     type_register_static(&spapr_machine_info);
3380 }
3381 
3382 type_init(spapr_machine_register_types)
3383