xref: /openbmc/qemu/hw/ppc/spapr.c (revision 397d30e9)
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/kvm.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(const char *type, int nr_servers,
99                                   int nr_irqs, Error **errp)
100 {
101     Error *err = NULL;
102     DeviceState *dev;
103 
104     dev = qdev_create(NULL, type);
105     qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
106     qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
107     object_property_set_bool(OBJECT(dev), true, "realized", &err);
108     if (err) {
109         error_propagate(errp, err);
110         object_unparent(OBJECT(dev));
111         return NULL;
112     }
113     return XICS_COMMON(dev);
114 }
115 
116 static XICSState *xics_system_init(MachineState *machine,
117                                    int nr_servers, int nr_irqs, Error **errp)
118 {
119     XICSState *xics = NULL;
120 
121     if (kvm_enabled()) {
122         Error *err = NULL;
123 
124         if (machine_kernel_irqchip_allowed(machine)) {
125             xics = try_create_xics(TYPE_XICS_SPAPR_KVM, nr_servers, nr_irqs,
126                                    &err);
127         }
128         if (machine_kernel_irqchip_required(machine) && !xics) {
129             error_reportf_err(err,
130                               "kernel_irqchip requested but unavailable: ");
131         } else {
132             error_free(err);
133         }
134     }
135 
136     if (!xics) {
137         xics = try_create_xics(TYPE_XICS_SPAPR, nr_servers, nr_irqs, errp);
138     }
139 
140     return xics;
141 }
142 
143 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
144                                   int smt_threads)
145 {
146     int i, ret = 0;
147     uint32_t servers_prop[smt_threads];
148     uint32_t gservers_prop[smt_threads * 2];
149     int index = ppc_get_vcpu_dt_id(cpu);
150 
151     if (cpu->cpu_version) {
152         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
153         if (ret < 0) {
154             return ret;
155         }
156     }
157 
158     /* Build interrupt servers and gservers properties */
159     for (i = 0; i < smt_threads; i++) {
160         servers_prop[i] = cpu_to_be32(index + i);
161         /* Hack, direct the group queues back to cpu 0 */
162         gservers_prop[i*2] = cpu_to_be32(index + i);
163         gservers_prop[i*2 + 1] = 0;
164     }
165     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
166                       servers_prop, sizeof(servers_prop));
167     if (ret < 0) {
168         return ret;
169     }
170     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
171                       gservers_prop, sizeof(gservers_prop));
172 
173     return ret;
174 }
175 
176 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
177 {
178     int ret = 0;
179     PowerPCCPU *cpu = POWERPC_CPU(cs);
180     int index = ppc_get_vcpu_dt_id(cpu);
181     uint32_t associativity[] = {cpu_to_be32(0x5),
182                                 cpu_to_be32(0x0),
183                                 cpu_to_be32(0x0),
184                                 cpu_to_be32(0x0),
185                                 cpu_to_be32(cs->numa_node),
186                                 cpu_to_be32(index)};
187 
188     /* Advertise NUMA via ibm,associativity */
189     if (nb_numa_nodes > 1) {
190         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
191                           sizeof(associativity));
192     }
193 
194     return ret;
195 }
196 
197 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
198 {
199     int ret = 0, offset, cpus_offset;
200     CPUState *cs;
201     char cpu_model[32];
202     int smt = kvmppc_smt_threads();
203     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
204 
205     CPU_FOREACH(cs) {
206         PowerPCCPU *cpu = POWERPC_CPU(cs);
207         DeviceClass *dc = DEVICE_GET_CLASS(cs);
208         int index = ppc_get_vcpu_dt_id(cpu);
209 
210         if ((index % smt) != 0) {
211             continue;
212         }
213 
214         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
215 
216         cpus_offset = fdt_path_offset(fdt, "/cpus");
217         if (cpus_offset < 0) {
218             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
219                                           "cpus");
220             if (cpus_offset < 0) {
221                 return cpus_offset;
222             }
223         }
224         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
225         if (offset < 0) {
226             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
227             if (offset < 0) {
228                 return offset;
229             }
230         }
231 
232         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
233                           pft_size_prop, sizeof(pft_size_prop));
234         if (ret < 0) {
235             return ret;
236         }
237 
238         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
239         if (ret < 0) {
240             return ret;
241         }
242 
243         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
244                                      ppc_get_compat_smt_threads(cpu));
245         if (ret < 0) {
246             return ret;
247         }
248     }
249     return ret;
250 }
251 
252 static hwaddr spapr_node0_size(void)
253 {
254     MachineState *machine = MACHINE(qdev_get_machine());
255 
256     if (nb_numa_nodes) {
257         int i;
258         for (i = 0; i < nb_numa_nodes; ++i) {
259             if (numa_info[i].node_mem) {
260                 return MIN(pow2floor(numa_info[i].node_mem),
261                            machine->ram_size);
262             }
263         }
264     }
265     return machine->ram_size;
266 }
267 
268 static void add_str(GString *s, const gchar *s1)
269 {
270     g_string_append_len(s, s1, strlen(s1) + 1);
271 }
272 
273 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
274                                        hwaddr size)
275 {
276     uint32_t associativity[] = {
277         cpu_to_be32(0x4), /* length */
278         cpu_to_be32(0x0), cpu_to_be32(0x0),
279         cpu_to_be32(0x0), cpu_to_be32(nodeid)
280     };
281     char mem_name[32];
282     uint64_t mem_reg_property[2];
283     int off;
284 
285     mem_reg_property[0] = cpu_to_be64(start);
286     mem_reg_property[1] = cpu_to_be64(size);
287 
288     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
289     off = fdt_add_subnode(fdt, 0, mem_name);
290     _FDT(off);
291     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
292     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
293                       sizeof(mem_reg_property))));
294     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
295                       sizeof(associativity))));
296     return off;
297 }
298 
299 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
300 {
301     MachineState *machine = MACHINE(spapr);
302     hwaddr mem_start, node_size;
303     int i, nb_nodes = nb_numa_nodes;
304     NodeInfo *nodes = numa_info;
305     NodeInfo ramnode;
306 
307     /* No NUMA nodes, assume there is just one node with whole RAM */
308     if (!nb_numa_nodes) {
309         nb_nodes = 1;
310         ramnode.node_mem = machine->ram_size;
311         nodes = &ramnode;
312     }
313 
314     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
315         if (!nodes[i].node_mem) {
316             continue;
317         }
318         if (mem_start >= machine->ram_size) {
319             node_size = 0;
320         } else {
321             node_size = nodes[i].node_mem;
322             if (node_size > machine->ram_size - mem_start) {
323                 node_size = machine->ram_size - mem_start;
324             }
325         }
326         if (!mem_start) {
327             /* ppc_spapr_init() checks for rma_size <= node0_size already */
328             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
329             mem_start += spapr->rma_size;
330             node_size -= spapr->rma_size;
331         }
332         for ( ; node_size; ) {
333             hwaddr sizetmp = pow2floor(node_size);
334 
335             /* mem_start != 0 here */
336             if (ctzl(mem_start) < ctzl(sizetmp)) {
337                 sizetmp = 1ULL << ctzl(mem_start);
338             }
339 
340             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
341             node_size -= sizetmp;
342             mem_start += sizetmp;
343         }
344     }
345 
346     return 0;
347 }
348 
349 /* Populate the "ibm,pa-features" property */
350 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset)
351 {
352     uint8_t pa_features_206[] = { 6, 0,
353         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
354     uint8_t pa_features_207[] = { 24, 0,
355         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
356         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
357         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
358         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
359     uint8_t *pa_features;
360     size_t pa_size;
361 
362     switch (env->mmu_model) {
363     case POWERPC_MMU_2_06:
364     case POWERPC_MMU_2_06a:
365         pa_features = pa_features_206;
366         pa_size = sizeof(pa_features_206);
367         break;
368     case POWERPC_MMU_2_07:
369     case POWERPC_MMU_2_07a:
370         pa_features = pa_features_207;
371         pa_size = sizeof(pa_features_207);
372         break;
373     default:
374         return;
375     }
376 
377     if (env->ci_large_pages) {
378         /*
379          * Note: we keep CI large pages off by default because a 64K capable
380          * guest provisioned with large pages might otherwise try to map a qemu
381          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
382          * even if that qemu runs on a 4k host.
383          * We dd this bit back here if we are confident this is not an issue
384          */
385         pa_features[3] |= 0x20;
386     }
387     if (kvmppc_has_cap_htm() && pa_size > 24) {
388         pa_features[24] |= 0x80;    /* Transactional memory support */
389     }
390 
391     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
392 }
393 
394 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
395                                   sPAPRMachineState *spapr)
396 {
397     PowerPCCPU *cpu = POWERPC_CPU(cs);
398     CPUPPCState *env = &cpu->env;
399     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
400     int index = ppc_get_vcpu_dt_id(cpu);
401     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
402                        0xffffffff, 0xffffffff};
403     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
404         : SPAPR_TIMEBASE_FREQ;
405     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
406     uint32_t page_sizes_prop[64];
407     size_t page_sizes_prop_size;
408     uint32_t vcpus_per_socket = smp_threads * smp_cores;
409     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
410     sPAPRDRConnector *drc;
411     sPAPRDRConnectorClass *drck;
412     int drc_index;
413 
414     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
415     if (drc) {
416         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
417         drc_index = drck->get_index(drc);
418         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
419     }
420 
421     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
422     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
423 
424     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
425     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
426                            env->dcache_line_size)));
427     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
428                            env->dcache_line_size)));
429     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
430                            env->icache_line_size)));
431     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
432                            env->icache_line_size)));
433 
434     if (pcc->l1_dcache_size) {
435         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
436                                pcc->l1_dcache_size)));
437     } else {
438         error_report("Warning: Unknown L1 dcache size for cpu");
439     }
440     if (pcc->l1_icache_size) {
441         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
442                                pcc->l1_icache_size)));
443     } else {
444         error_report("Warning: Unknown L1 icache size for cpu");
445     }
446 
447     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
448     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
449     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
450     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
451     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
452     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
453 
454     if (env->spr_cb[SPR_PURR].oea_read) {
455         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
456     }
457 
458     if (env->mmu_model & POWERPC_MMU_1TSEG) {
459         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
460                           segs, sizeof(segs))));
461     }
462 
463     /* Advertise VMX/VSX (vector extensions) if available
464      *   0 / no property == no vector extensions
465      *   1               == VMX / Altivec available
466      *   2               == VSX available */
467     if (env->insns_flags & PPC_ALTIVEC) {
468         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
469 
470         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
471     }
472 
473     /* Advertise DFP (Decimal Floating Point) if available
474      *   0 / no property == no DFP
475      *   1               == DFP available */
476     if (env->insns_flags2 & PPC2_DFP) {
477         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
478     }
479 
480     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
481                                                   sizeof(page_sizes_prop));
482     if (page_sizes_prop_size) {
483         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
484                           page_sizes_prop, page_sizes_prop_size)));
485     }
486 
487     spapr_populate_pa_features(env, fdt, offset);
488 
489     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
490                            cs->cpu_index / vcpus_per_socket)));
491 
492     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
493                       pft_size_prop, sizeof(pft_size_prop))));
494 
495     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
496 
497     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
498                                 ppc_get_compat_smt_threads(cpu)));
499 }
500 
501 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
502 {
503     CPUState *cs;
504     int cpus_offset;
505     char *nodename;
506     int smt = kvmppc_smt_threads();
507 
508     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
509     _FDT(cpus_offset);
510     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
511     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
512 
513     /*
514      * We walk the CPUs in reverse order to ensure that CPU DT nodes
515      * created by fdt_add_subnode() end up in the right order in FDT
516      * for the guest kernel the enumerate the CPUs correctly.
517      */
518     CPU_FOREACH_REVERSE(cs) {
519         PowerPCCPU *cpu = POWERPC_CPU(cs);
520         int index = ppc_get_vcpu_dt_id(cpu);
521         DeviceClass *dc = DEVICE_GET_CLASS(cs);
522         int offset;
523 
524         if ((index % smt) != 0) {
525             continue;
526         }
527 
528         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
529         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
530         g_free(nodename);
531         _FDT(offset);
532         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
533     }
534 
535 }
536 
537 /*
538  * Adds ibm,dynamic-reconfiguration-memory node.
539  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
540  * of this device tree node.
541  */
542 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
543 {
544     MachineState *machine = MACHINE(spapr);
545     int ret, i, offset;
546     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
547     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
548     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
549     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
550                        memory_region_size(&spapr->hotplug_memory.mr)) /
551                        lmb_size;
552     uint32_t *int_buf, *cur_index, buf_len;
553     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
554 
555     /*
556      * Don't create the node if there is no hotpluggable memory
557      */
558     if (machine->ram_size == machine->maxram_size) {
559         return 0;
560     }
561 
562     /*
563      * Allocate enough buffer size to fit in ibm,dynamic-memory
564      * or ibm,associativity-lookup-arrays
565      */
566     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
567               * sizeof(uint32_t);
568     cur_index = int_buf = g_malloc0(buf_len);
569 
570     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
571 
572     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
573                     sizeof(prop_lmb_size));
574     if (ret < 0) {
575         goto out;
576     }
577 
578     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
579     if (ret < 0) {
580         goto out;
581     }
582 
583     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
584     if (ret < 0) {
585         goto out;
586     }
587 
588     /* ibm,dynamic-memory */
589     int_buf[0] = cpu_to_be32(nr_lmbs);
590     cur_index++;
591     for (i = 0; i < nr_lmbs; i++) {
592         uint64_t addr = i * lmb_size;
593         uint32_t *dynamic_memory = cur_index;
594 
595         if (i >= hotplug_lmb_start) {
596             sPAPRDRConnector *drc;
597             sPAPRDRConnectorClass *drck;
598 
599             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
600             g_assert(drc);
601             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
602 
603             dynamic_memory[0] = cpu_to_be32(addr >> 32);
604             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
605             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
606             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
607             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
608             if (memory_region_present(get_system_memory(), addr)) {
609                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
610             } else {
611                 dynamic_memory[5] = cpu_to_be32(0);
612             }
613         } else {
614             /*
615              * LMB information for RMA, boot time RAM and gap b/n RAM and
616              * hotplug memory region -- all these are marked as reserved
617              * and as having no valid DRC.
618              */
619             dynamic_memory[0] = cpu_to_be32(addr >> 32);
620             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
621             dynamic_memory[2] = cpu_to_be32(0);
622             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
623             dynamic_memory[4] = cpu_to_be32(-1);
624             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
625                                             SPAPR_LMB_FLAGS_DRC_INVALID);
626         }
627 
628         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
629     }
630     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
631     if (ret < 0) {
632         goto out;
633     }
634 
635     /* ibm,associativity-lookup-arrays */
636     cur_index = int_buf;
637     int_buf[0] = cpu_to_be32(nr_nodes);
638     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
639     cur_index += 2;
640     for (i = 0; i < nr_nodes; i++) {
641         uint32_t associativity[] = {
642             cpu_to_be32(0x0),
643             cpu_to_be32(0x0),
644             cpu_to_be32(0x0),
645             cpu_to_be32(i)
646         };
647         memcpy(cur_index, associativity, sizeof(associativity));
648         cur_index += 4;
649     }
650     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
651             (cur_index - int_buf) * sizeof(uint32_t));
652 out:
653     g_free(int_buf);
654     return ret;
655 }
656 
657 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
658                                 sPAPROptionVector *ov5_updates)
659 {
660     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
661     int ret = 0, offset;
662 
663     /* Generate ibm,dynamic-reconfiguration-memory node if required */
664     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
665         g_assert(smc->dr_lmb_enabled);
666         ret = spapr_populate_drconf_memory(spapr, fdt);
667         if (ret) {
668             goto out;
669         }
670     }
671 
672     offset = fdt_path_offset(fdt, "/chosen");
673     if (offset < 0) {
674         offset = fdt_add_subnode(fdt, 0, "chosen");
675         if (offset < 0) {
676             return offset;
677         }
678     }
679     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
680                                  "ibm,architecture-vec-5");
681 
682 out:
683     return ret;
684 }
685 
686 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
687                                  target_ulong addr, target_ulong size,
688                                  bool cpu_update,
689                                  sPAPROptionVector *ov5_updates)
690 {
691     void *fdt, *fdt_skel;
692     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
693 
694     size -= sizeof(hdr);
695 
696     /* Create sceleton */
697     fdt_skel = g_malloc0(size);
698     _FDT((fdt_create(fdt_skel, size)));
699     _FDT((fdt_begin_node(fdt_skel, "")));
700     _FDT((fdt_end_node(fdt_skel)));
701     _FDT((fdt_finish(fdt_skel)));
702     fdt = g_malloc0(size);
703     _FDT((fdt_open_into(fdt_skel, fdt, size)));
704     g_free(fdt_skel);
705 
706     /* Fixup cpu nodes */
707     if (cpu_update) {
708         _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
709     }
710 
711     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
712         return -1;
713     }
714 
715     /* Pack resulting tree */
716     _FDT((fdt_pack(fdt)));
717 
718     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
719         trace_spapr_cas_failed(size);
720         return -1;
721     }
722 
723     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
724     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
725     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
726     g_free(fdt);
727 
728     return 0;
729 }
730 
731 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
732 {
733     int rtas;
734     GString *hypertas = g_string_sized_new(256);
735     GString *qemu_hypertas = g_string_sized_new(256);
736     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
737     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
738         memory_region_size(&spapr->hotplug_memory.mr);
739     uint32_t lrdr_capacity[] = {
740         cpu_to_be32(max_hotplug_addr >> 32),
741         cpu_to_be32(max_hotplug_addr & 0xffffffff),
742         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
743         cpu_to_be32(max_cpus / smp_threads),
744     };
745 
746     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
747 
748     /* hypertas */
749     add_str(hypertas, "hcall-pft");
750     add_str(hypertas, "hcall-term");
751     add_str(hypertas, "hcall-dabr");
752     add_str(hypertas, "hcall-interrupt");
753     add_str(hypertas, "hcall-tce");
754     add_str(hypertas, "hcall-vio");
755     add_str(hypertas, "hcall-splpar");
756     add_str(hypertas, "hcall-bulk");
757     add_str(hypertas, "hcall-set-mode");
758     add_str(hypertas, "hcall-sprg0");
759     add_str(hypertas, "hcall-copy");
760     add_str(hypertas, "hcall-debug");
761     add_str(qemu_hypertas, "hcall-memop1");
762 
763     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
764         add_str(hypertas, "hcall-multi-tce");
765     }
766     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
767                      hypertas->str, hypertas->len));
768     g_string_free(hypertas, TRUE);
769     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
770                      qemu_hypertas->str, qemu_hypertas->len));
771     g_string_free(qemu_hypertas, TRUE);
772 
773     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
774                      refpoints, sizeof(refpoints)));
775 
776     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
777                           RTAS_ERROR_LOG_MAX));
778     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
779                           RTAS_EVENT_SCAN_RATE));
780 
781     if (msi_nonbroken) {
782         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
783     }
784 
785     /*
786      * According to PAPR, rtas ibm,os-term does not guarantee a return
787      * back to the guest cpu.
788      *
789      * While an additional ibm,extended-os-term property indicates
790      * that rtas call return will always occur. Set this property.
791      */
792     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
793 
794     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
795                      lrdr_capacity, sizeof(lrdr_capacity)));
796 
797     spapr_dt_rtas_tokens(fdt, rtas);
798 }
799 
800 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
801 {
802     MachineState *machine = MACHINE(spapr);
803     int chosen;
804     const char *boot_device = machine->boot_order;
805     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
806     size_t cb = 0;
807     char *bootlist = get_boot_devices_list(&cb, true);
808 
809     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
810 
811     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
812     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
813                           spapr->initrd_base));
814     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
815                           spapr->initrd_base + spapr->initrd_size));
816 
817     if (spapr->kernel_size) {
818         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
819                               cpu_to_be64(spapr->kernel_size) };
820 
821         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
822                          &kprop, sizeof(kprop)));
823         if (spapr->kernel_le) {
824             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
825         }
826     }
827     if (boot_menu) {
828         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
829     }
830     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
831     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
832     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
833 
834     if (cb && bootlist) {
835         int i;
836 
837         for (i = 0; i < cb; i++) {
838             if (bootlist[i] == '\n') {
839                 bootlist[i] = ' ';
840             }
841         }
842         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
843     }
844 
845     if (boot_device && strlen(boot_device)) {
846         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
847     }
848 
849     if (!spapr->has_graphics && stdout_path) {
850         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
851     }
852 
853     g_free(stdout_path);
854     g_free(bootlist);
855 }
856 
857 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
858 {
859     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
860      * KVM to work under pHyp with some guest co-operation */
861     int hypervisor;
862     uint8_t hypercall[16];
863 
864     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
865     /* indicate KVM hypercall interface */
866     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
867     if (kvmppc_has_cap_fixup_hcalls()) {
868         /*
869          * Older KVM versions with older guest kernels were broken
870          * with the magic page, don't allow the guest to map it.
871          */
872         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
873                                   sizeof(hypercall))) {
874             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
875                              hypercall, sizeof(hypercall)));
876         }
877     }
878 }
879 
880 static void *spapr_build_fdt(sPAPRMachineState *spapr,
881                              hwaddr rtas_addr,
882                              hwaddr rtas_size)
883 {
884     MachineState *machine = MACHINE(qdev_get_machine());
885     MachineClass *mc = MACHINE_GET_CLASS(machine);
886     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
887     int ret;
888     void *fdt;
889     sPAPRPHBState *phb;
890     char *buf;
891 
892     fdt = g_malloc0(FDT_MAX_SIZE);
893     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
894 
895     /* Root node */
896     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
897     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
898     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
899 
900     /*
901      * Add info to guest to indentify which host is it being run on
902      * and what is the uuid of the guest
903      */
904     if (kvmppc_get_host_model(&buf)) {
905         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
906         g_free(buf);
907     }
908     if (kvmppc_get_host_serial(&buf)) {
909         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
910         g_free(buf);
911     }
912 
913     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
914 
915     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
916     if (qemu_uuid_set) {
917         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
918     }
919     g_free(buf);
920 
921     if (qemu_get_vm_name()) {
922         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
923                                 qemu_get_vm_name()));
924     }
925 
926     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
927     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
928 
929     /* /interrupt controller */
930     spapr_dt_xics(spapr->xics, fdt, PHANDLE_XICP);
931 
932     ret = spapr_populate_memory(spapr, fdt);
933     if (ret < 0) {
934         error_report("couldn't setup memory nodes in fdt");
935         exit(1);
936     }
937 
938     /* /vdevice */
939     spapr_dt_vdevice(spapr->vio_bus, fdt);
940 
941     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
942         ret = spapr_rng_populate_dt(fdt);
943         if (ret < 0) {
944             error_report("could not set up rng device in the fdt");
945             exit(1);
946         }
947     }
948 
949     QLIST_FOREACH(phb, &spapr->phbs, list) {
950         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
951         if (ret < 0) {
952             error_report("couldn't setup PCI devices in fdt");
953             exit(1);
954         }
955     }
956 
957     /* cpus */
958     spapr_populate_cpus_dt_node(fdt, spapr);
959 
960     if (smc->dr_lmb_enabled) {
961         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
962     }
963 
964     if (mc->query_hotpluggable_cpus) {
965         int offset = fdt_path_offset(fdt, "/cpus");
966         ret = spapr_drc_populate_dt(fdt, offset, NULL,
967                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
968         if (ret < 0) {
969             error_report("Couldn't set up CPU DR device tree properties");
970             exit(1);
971         }
972     }
973 
974     /* /event-sources */
975     spapr_dt_events(spapr, fdt);
976 
977     /* /rtas */
978     spapr_dt_rtas(spapr, fdt);
979 
980     /* /chosen */
981     spapr_dt_chosen(spapr, fdt);
982 
983     /* /hypervisor */
984     if (kvm_enabled()) {
985         spapr_dt_hypervisor(spapr, fdt);
986     }
987 
988     /* Build memory reserve map */
989     if (spapr->kernel_size) {
990         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
991     }
992     if (spapr->initrd_size) {
993         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
994     }
995 
996     /* ibm,client-architecture-support updates */
997     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
998     if (ret < 0) {
999         error_report("couldn't setup CAS properties fdt");
1000         exit(1);
1001     }
1002 
1003     return fdt;
1004 }
1005 
1006 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1007 {
1008     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1009 }
1010 
1011 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
1012 {
1013     CPUPPCState *env = &cpu->env;
1014 
1015     if (msr_pr) {
1016         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1017         env->gpr[3] = H_PRIVILEGE;
1018     } else {
1019         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1020     }
1021 }
1022 
1023 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1024 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1025 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1026 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1027 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1028 
1029 /*
1030  * Get the fd to access the kernel htab, re-opening it if necessary
1031  */
1032 static int get_htab_fd(sPAPRMachineState *spapr)
1033 {
1034     if (spapr->htab_fd >= 0) {
1035         return spapr->htab_fd;
1036     }
1037 
1038     spapr->htab_fd = kvmppc_get_htab_fd(false);
1039     if (spapr->htab_fd < 0) {
1040         error_report("Unable to open fd for reading hash table from KVM: %s",
1041                      strerror(errno));
1042     }
1043 
1044     return spapr->htab_fd;
1045 }
1046 
1047 static void close_htab_fd(sPAPRMachineState *spapr)
1048 {
1049     if (spapr->htab_fd >= 0) {
1050         close(spapr->htab_fd);
1051     }
1052     spapr->htab_fd = -1;
1053 }
1054 
1055 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1056 {
1057     int shift;
1058 
1059     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1060      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1061      * that's much more than is needed for Linux guests */
1062     shift = ctz64(pow2ceil(ramsize)) - 7;
1063     shift = MAX(shift, 18); /* Minimum architected size */
1064     shift = MIN(shift, 46); /* Maximum architected size */
1065     return shift;
1066 }
1067 
1068 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1069                                  Error **errp)
1070 {
1071     long rc;
1072 
1073     /* Clean up any HPT info from a previous boot */
1074     g_free(spapr->htab);
1075     spapr->htab = NULL;
1076     spapr->htab_shift = 0;
1077     close_htab_fd(spapr);
1078 
1079     rc = kvmppc_reset_htab(shift);
1080     if (rc < 0) {
1081         /* kernel-side HPT needed, but couldn't allocate one */
1082         error_setg_errno(errp, errno,
1083                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1084                          shift);
1085         /* This is almost certainly fatal, but if the caller really
1086          * wants to carry on with shift == 0, it's welcome to try */
1087     } else if (rc > 0) {
1088         /* kernel-side HPT allocated */
1089         if (rc != shift) {
1090             error_setg(errp,
1091                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1092                        shift, rc);
1093         }
1094 
1095         spapr->htab_shift = shift;
1096         spapr->htab = NULL;
1097     } else {
1098         /* kernel-side HPT not needed, allocate in userspace instead */
1099         size_t size = 1ULL << shift;
1100         int i;
1101 
1102         spapr->htab = qemu_memalign(size, size);
1103         if (!spapr->htab) {
1104             error_setg_errno(errp, errno,
1105                              "Could not allocate HPT of order %d", shift);
1106             return;
1107         }
1108 
1109         memset(spapr->htab, 0, size);
1110         spapr->htab_shift = shift;
1111 
1112         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1113             DIRTY_HPTE(HPTE(spapr->htab, i));
1114         }
1115     }
1116 }
1117 
1118 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1119 {
1120     bool matched = false;
1121 
1122     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1123         matched = true;
1124     }
1125 
1126     if (!matched) {
1127         error_report("Device %s is not supported by this machine yet.",
1128                      qdev_fw_name(DEVICE(sbdev)));
1129         exit(1);
1130     }
1131 }
1132 
1133 static void ppc_spapr_reset(void)
1134 {
1135     MachineState *machine = MACHINE(qdev_get_machine());
1136     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1137     PowerPCCPU *first_ppc_cpu;
1138     uint32_t rtas_limit;
1139     hwaddr rtas_addr, fdt_addr;
1140     void *fdt;
1141     int rc;
1142 
1143     /* Check for unknown sysbus devices */
1144     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1145 
1146     /* Allocate and/or reset the hash page table */
1147     spapr_reallocate_hpt(spapr,
1148                          spapr_hpt_shift_for_ramsize(machine->maxram_size),
1149                          &error_fatal);
1150 
1151     /* Update the RMA size if necessary */
1152     if (spapr->vrma_adjust) {
1153         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1154                                           spapr->htab_shift);
1155     }
1156 
1157     qemu_devices_reset();
1158 
1159     /*
1160      * We place the device tree and RTAS just below either the top of the RMA,
1161      * or just below 2GB, whichever is lowere, so that it can be
1162      * processed with 32-bit real mode code if necessary
1163      */
1164     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1165     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1166     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1167 
1168     /* if this reset wasn't generated by CAS, we should reset our
1169      * negotiated options and start from scratch */
1170     if (!spapr->cas_reboot) {
1171         spapr_ovec_cleanup(spapr->ov5_cas);
1172         spapr->ov5_cas = spapr_ovec_new();
1173     }
1174 
1175     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1176 
1177     spapr_load_rtas(spapr, fdt, rtas_addr);
1178 
1179     rc = fdt_pack(fdt);
1180 
1181     /* Should only fail if we've built a corrupted tree */
1182     assert(rc == 0);
1183 
1184     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1185         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1186                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1187         exit(1);
1188     }
1189 
1190     /* Load the fdt */
1191     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1192     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1193     g_free(fdt);
1194 
1195     /* Set up the entry state */
1196     first_ppc_cpu = POWERPC_CPU(first_cpu);
1197     first_ppc_cpu->env.gpr[3] = fdt_addr;
1198     first_ppc_cpu->env.gpr[5] = 0;
1199     first_cpu->halted = 0;
1200     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1201 
1202     spapr->cas_reboot = false;
1203 }
1204 
1205 static void spapr_create_nvram(sPAPRMachineState *spapr)
1206 {
1207     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1208     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1209 
1210     if (dinfo) {
1211         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1212                             &error_fatal);
1213     }
1214 
1215     qdev_init_nofail(dev);
1216 
1217     spapr->nvram = (struct sPAPRNVRAM *)dev;
1218 }
1219 
1220 static void spapr_rtc_create(sPAPRMachineState *spapr)
1221 {
1222     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1223 
1224     qdev_init_nofail(dev);
1225     spapr->rtc = dev;
1226 
1227     object_property_add_alias(qdev_get_machine(), "rtc-time",
1228                               OBJECT(spapr->rtc), "date", NULL);
1229 }
1230 
1231 /* Returns whether we want to use VGA or not */
1232 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1233 {
1234     switch (vga_interface_type) {
1235     case VGA_NONE:
1236         return false;
1237     case VGA_DEVICE:
1238         return true;
1239     case VGA_STD:
1240     case VGA_VIRTIO:
1241         return pci_vga_init(pci_bus) != NULL;
1242     default:
1243         error_setg(errp,
1244                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1245         return false;
1246     }
1247 }
1248 
1249 static int spapr_post_load(void *opaque, int version_id)
1250 {
1251     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1252     int err = 0;
1253 
1254     /* In earlier versions, there was no separate qdev for the PAPR
1255      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1256      * So when migrating from those versions, poke the incoming offset
1257      * value into the RTC device */
1258     if (version_id < 3) {
1259         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1260     }
1261 
1262     return err;
1263 }
1264 
1265 static bool version_before_3(void *opaque, int version_id)
1266 {
1267     return version_id < 3;
1268 }
1269 
1270 static const VMStateDescription vmstate_spapr = {
1271     .name = "spapr",
1272     .version_id = 3,
1273     .minimum_version_id = 1,
1274     .post_load = spapr_post_load,
1275     .fields = (VMStateField[]) {
1276         /* used to be @next_irq */
1277         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1278 
1279         /* RTC offset */
1280         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1281 
1282         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1283         VMSTATE_END_OF_LIST()
1284     },
1285 };
1286 
1287 static int htab_save_setup(QEMUFile *f, void *opaque)
1288 {
1289     sPAPRMachineState *spapr = opaque;
1290 
1291     /* "Iteration" header */
1292     qemu_put_be32(f, spapr->htab_shift);
1293 
1294     if (spapr->htab) {
1295         spapr->htab_save_index = 0;
1296         spapr->htab_first_pass = true;
1297     } else {
1298         assert(kvm_enabled());
1299     }
1300 
1301 
1302     return 0;
1303 }
1304 
1305 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1306                                  int64_t max_ns)
1307 {
1308     bool has_timeout = max_ns != -1;
1309     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1310     int index = spapr->htab_save_index;
1311     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1312 
1313     assert(spapr->htab_first_pass);
1314 
1315     do {
1316         int chunkstart;
1317 
1318         /* Consume invalid HPTEs */
1319         while ((index < htabslots)
1320                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1321             index++;
1322             CLEAN_HPTE(HPTE(spapr->htab, index));
1323         }
1324 
1325         /* Consume valid HPTEs */
1326         chunkstart = index;
1327         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1328                && HPTE_VALID(HPTE(spapr->htab, index))) {
1329             index++;
1330             CLEAN_HPTE(HPTE(spapr->htab, index));
1331         }
1332 
1333         if (index > chunkstart) {
1334             int n_valid = index - chunkstart;
1335 
1336             qemu_put_be32(f, chunkstart);
1337             qemu_put_be16(f, n_valid);
1338             qemu_put_be16(f, 0);
1339             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1340                             HASH_PTE_SIZE_64 * n_valid);
1341 
1342             if (has_timeout &&
1343                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1344                 break;
1345             }
1346         }
1347     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1348 
1349     if (index >= htabslots) {
1350         assert(index == htabslots);
1351         index = 0;
1352         spapr->htab_first_pass = false;
1353     }
1354     spapr->htab_save_index = index;
1355 }
1356 
1357 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1358                                 int64_t max_ns)
1359 {
1360     bool final = max_ns < 0;
1361     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1362     int examined = 0, sent = 0;
1363     int index = spapr->htab_save_index;
1364     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1365 
1366     assert(!spapr->htab_first_pass);
1367 
1368     do {
1369         int chunkstart, invalidstart;
1370 
1371         /* Consume non-dirty HPTEs */
1372         while ((index < htabslots)
1373                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1374             index++;
1375             examined++;
1376         }
1377 
1378         chunkstart = index;
1379         /* Consume valid dirty HPTEs */
1380         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1381                && HPTE_DIRTY(HPTE(spapr->htab, index))
1382                && HPTE_VALID(HPTE(spapr->htab, index))) {
1383             CLEAN_HPTE(HPTE(spapr->htab, index));
1384             index++;
1385             examined++;
1386         }
1387 
1388         invalidstart = index;
1389         /* Consume invalid dirty HPTEs */
1390         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1391                && HPTE_DIRTY(HPTE(spapr->htab, index))
1392                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1393             CLEAN_HPTE(HPTE(spapr->htab, index));
1394             index++;
1395             examined++;
1396         }
1397 
1398         if (index > chunkstart) {
1399             int n_valid = invalidstart - chunkstart;
1400             int n_invalid = index - invalidstart;
1401 
1402             qemu_put_be32(f, chunkstart);
1403             qemu_put_be16(f, n_valid);
1404             qemu_put_be16(f, n_invalid);
1405             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1406                             HASH_PTE_SIZE_64 * n_valid);
1407             sent += index - chunkstart;
1408 
1409             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1410                 break;
1411             }
1412         }
1413 
1414         if (examined >= htabslots) {
1415             break;
1416         }
1417 
1418         if (index >= htabslots) {
1419             assert(index == htabslots);
1420             index = 0;
1421         }
1422     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1423 
1424     if (index >= htabslots) {
1425         assert(index == htabslots);
1426         index = 0;
1427     }
1428 
1429     spapr->htab_save_index = index;
1430 
1431     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1432 }
1433 
1434 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1435 #define MAX_KVM_BUF_SIZE    2048
1436 
1437 static int htab_save_iterate(QEMUFile *f, void *opaque)
1438 {
1439     sPAPRMachineState *spapr = opaque;
1440     int fd;
1441     int rc = 0;
1442 
1443     /* Iteration header */
1444     qemu_put_be32(f, 0);
1445 
1446     if (!spapr->htab) {
1447         assert(kvm_enabled());
1448 
1449         fd = get_htab_fd(spapr);
1450         if (fd < 0) {
1451             return fd;
1452         }
1453 
1454         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1455         if (rc < 0) {
1456             return rc;
1457         }
1458     } else  if (spapr->htab_first_pass) {
1459         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1460     } else {
1461         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1462     }
1463 
1464     /* End marker */
1465     qemu_put_be32(f, 0);
1466     qemu_put_be16(f, 0);
1467     qemu_put_be16(f, 0);
1468 
1469     return rc;
1470 }
1471 
1472 static int htab_save_complete(QEMUFile *f, void *opaque)
1473 {
1474     sPAPRMachineState *spapr = opaque;
1475     int fd;
1476 
1477     /* Iteration header */
1478     qemu_put_be32(f, 0);
1479 
1480     if (!spapr->htab) {
1481         int rc;
1482 
1483         assert(kvm_enabled());
1484 
1485         fd = get_htab_fd(spapr);
1486         if (fd < 0) {
1487             return fd;
1488         }
1489 
1490         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1491         if (rc < 0) {
1492             return rc;
1493         }
1494     } else {
1495         if (spapr->htab_first_pass) {
1496             htab_save_first_pass(f, spapr, -1);
1497         }
1498         htab_save_later_pass(f, spapr, -1);
1499     }
1500 
1501     /* End marker */
1502     qemu_put_be32(f, 0);
1503     qemu_put_be16(f, 0);
1504     qemu_put_be16(f, 0);
1505 
1506     return 0;
1507 }
1508 
1509 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1510 {
1511     sPAPRMachineState *spapr = opaque;
1512     uint32_t section_hdr;
1513     int fd = -1;
1514 
1515     if (version_id < 1 || version_id > 1) {
1516         error_report("htab_load() bad version");
1517         return -EINVAL;
1518     }
1519 
1520     section_hdr = qemu_get_be32(f);
1521 
1522     if (section_hdr) {
1523         Error *local_err = NULL;
1524 
1525         /* First section gives the htab size */
1526         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1527         if (local_err) {
1528             error_report_err(local_err);
1529             return -EINVAL;
1530         }
1531         return 0;
1532     }
1533 
1534     if (!spapr->htab) {
1535         assert(kvm_enabled());
1536 
1537         fd = kvmppc_get_htab_fd(true);
1538         if (fd < 0) {
1539             error_report("Unable to open fd to restore KVM hash table: %s",
1540                          strerror(errno));
1541         }
1542     }
1543 
1544     while (true) {
1545         uint32_t index;
1546         uint16_t n_valid, n_invalid;
1547 
1548         index = qemu_get_be32(f);
1549         n_valid = qemu_get_be16(f);
1550         n_invalid = qemu_get_be16(f);
1551 
1552         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1553             /* End of Stream */
1554             break;
1555         }
1556 
1557         if ((index + n_valid + n_invalid) >
1558             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1559             /* Bad index in stream */
1560             error_report(
1561                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1562                 index, n_valid, n_invalid, spapr->htab_shift);
1563             return -EINVAL;
1564         }
1565 
1566         if (spapr->htab) {
1567             if (n_valid) {
1568                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1569                                 HASH_PTE_SIZE_64 * n_valid);
1570             }
1571             if (n_invalid) {
1572                 memset(HPTE(spapr->htab, index + n_valid), 0,
1573                        HASH_PTE_SIZE_64 * n_invalid);
1574             }
1575         } else {
1576             int rc;
1577 
1578             assert(fd >= 0);
1579 
1580             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1581             if (rc < 0) {
1582                 return rc;
1583             }
1584         }
1585     }
1586 
1587     if (!spapr->htab) {
1588         assert(fd >= 0);
1589         close(fd);
1590     }
1591 
1592     return 0;
1593 }
1594 
1595 static void htab_cleanup(void *opaque)
1596 {
1597     sPAPRMachineState *spapr = opaque;
1598 
1599     close_htab_fd(spapr);
1600 }
1601 
1602 static SaveVMHandlers savevm_htab_handlers = {
1603     .save_live_setup = htab_save_setup,
1604     .save_live_iterate = htab_save_iterate,
1605     .save_live_complete_precopy = htab_save_complete,
1606     .cleanup = htab_cleanup,
1607     .load_state = htab_load,
1608 };
1609 
1610 static void spapr_boot_set(void *opaque, const char *boot_device,
1611                            Error **errp)
1612 {
1613     MachineState *machine = MACHINE(qdev_get_machine());
1614     machine->boot_order = g_strdup(boot_device);
1615 }
1616 
1617 /*
1618  * Reset routine for LMB DR devices.
1619  *
1620  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1621  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1622  * when it walks all its children devices. LMB devices reset occurs
1623  * as part of spapr_ppc_reset().
1624  */
1625 static void spapr_drc_reset(void *opaque)
1626 {
1627     sPAPRDRConnector *drc = opaque;
1628     DeviceState *d = DEVICE(drc);
1629 
1630     if (d) {
1631         device_reset(d);
1632     }
1633 }
1634 
1635 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1636 {
1637     MachineState *machine = MACHINE(spapr);
1638     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1639     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1640     int i;
1641 
1642     for (i = 0; i < nr_lmbs; i++) {
1643         sPAPRDRConnector *drc;
1644         uint64_t addr;
1645 
1646         addr = i * lmb_size + spapr->hotplug_memory.base;
1647         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1648                                      addr/lmb_size);
1649         qemu_register_reset(spapr_drc_reset, drc);
1650     }
1651 }
1652 
1653 /*
1654  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1655  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1656  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1657  */
1658 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1659 {
1660     int i;
1661 
1662     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1663         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1664                    " is not aligned to %llu MiB",
1665                    machine->ram_size,
1666                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1667         return;
1668     }
1669 
1670     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1671         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1672                    " is not aligned to %llu MiB",
1673                    machine->ram_size,
1674                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1675         return;
1676     }
1677 
1678     for (i = 0; i < nb_numa_nodes; i++) {
1679         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1680             error_setg(errp,
1681                        "Node %d memory size 0x%" PRIx64
1682                        " is not aligned to %llu MiB",
1683                        i, numa_info[i].node_mem,
1684                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1685             return;
1686         }
1687     }
1688 }
1689 
1690 /* pSeries LPAR / sPAPR hardware init */
1691 static void ppc_spapr_init(MachineState *machine)
1692 {
1693     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1694     MachineClass *mc = MACHINE_GET_CLASS(machine);
1695     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1696     const char *kernel_filename = machine->kernel_filename;
1697     const char *initrd_filename = machine->initrd_filename;
1698     PCIHostState *phb;
1699     int i;
1700     MemoryRegion *sysmem = get_system_memory();
1701     MemoryRegion *ram = g_new(MemoryRegion, 1);
1702     MemoryRegion *rma_region;
1703     void *rma = NULL;
1704     hwaddr rma_alloc_size;
1705     hwaddr node0_size = spapr_node0_size();
1706     long load_limit, fw_size;
1707     char *filename;
1708     int smt = kvmppc_smt_threads();
1709     int spapr_cores = smp_cpus / smp_threads;
1710     int spapr_max_cores = max_cpus / smp_threads;
1711 
1712     if (mc->query_hotpluggable_cpus) {
1713         if (smp_cpus % smp_threads) {
1714             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1715                          smp_cpus, smp_threads);
1716             exit(1);
1717         }
1718         if (max_cpus % smp_threads) {
1719             error_report("max_cpus (%u) must be multiple of threads (%u)",
1720                          max_cpus, smp_threads);
1721             exit(1);
1722         }
1723     }
1724 
1725     msi_nonbroken = true;
1726 
1727     QLIST_INIT(&spapr->phbs);
1728 
1729     cpu_ppc_hypercall = emulate_spapr_hypercall;
1730 
1731     /* Allocate RMA if necessary */
1732     rma_alloc_size = kvmppc_alloc_rma(&rma);
1733 
1734     if (rma_alloc_size == -1) {
1735         error_report("Unable to create RMA");
1736         exit(1);
1737     }
1738 
1739     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1740         spapr->rma_size = rma_alloc_size;
1741     } else {
1742         spapr->rma_size = node0_size;
1743 
1744         /* With KVM, we don't actually know whether KVM supports an
1745          * unbounded RMA (PR KVM) or is limited by the hash table size
1746          * (HV KVM using VRMA), so we always assume the latter
1747          *
1748          * In that case, we also limit the initial allocations for RTAS
1749          * etc... to 256M since we have no way to know what the VRMA size
1750          * is going to be as it depends on the size of the hash table
1751          * isn't determined yet.
1752          */
1753         if (kvm_enabled()) {
1754             spapr->vrma_adjust = 1;
1755             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1756         }
1757 
1758         /* Actually we don't support unbounded RMA anymore since we
1759          * added proper emulation of HV mode. The max we can get is
1760          * 16G which also happens to be what we configure for PAPR
1761          * mode so make sure we don't do anything bigger than that
1762          */
1763         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1764     }
1765 
1766     if (spapr->rma_size > node0_size) {
1767         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1768                      spapr->rma_size);
1769         exit(1);
1770     }
1771 
1772     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1773     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1774 
1775     /* Set up Interrupt Controller before we create the VCPUs */
1776     spapr->xics = xics_system_init(machine,
1777                                    DIV_ROUND_UP(max_cpus * smt, smp_threads),
1778                                    XICS_IRQS_SPAPR, &error_fatal);
1779 
1780     /* Set up containers for ibm,client-set-architecture negotiated options */
1781     spapr->ov5 = spapr_ovec_new();
1782     spapr->ov5_cas = spapr_ovec_new();
1783 
1784     if (smc->dr_lmb_enabled) {
1785         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
1786         spapr_validate_node_memory(machine, &error_fatal);
1787     }
1788 
1789     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
1790 
1791     /* advertise support for dedicated HP event source to guests */
1792     if (spapr->use_hotplug_event_source) {
1793         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
1794     }
1795 
1796     /* init CPUs */
1797     if (machine->cpu_model == NULL) {
1798         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
1799     }
1800 
1801     ppc_cpu_parse_features(machine->cpu_model);
1802 
1803     if (mc->query_hotpluggable_cpus) {
1804         char *type = spapr_get_cpu_core_type(machine->cpu_model);
1805 
1806         if (type == NULL) {
1807             error_report("Unable to find sPAPR CPU Core definition");
1808             exit(1);
1809         }
1810 
1811         spapr->cores = g_new0(Object *, spapr_max_cores);
1812         for (i = 0; i < spapr_max_cores; i++) {
1813             int core_id = i * smp_threads;
1814             sPAPRDRConnector *drc =
1815                 spapr_dr_connector_new(OBJECT(spapr),
1816                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
1817                                        (core_id / smp_threads) * smt);
1818 
1819             qemu_register_reset(spapr_drc_reset, drc);
1820 
1821             if (i < spapr_cores) {
1822                 Object *core  = object_new(type);
1823                 object_property_set_int(core, smp_threads, "nr-threads",
1824                                         &error_fatal);
1825                 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1826                                         &error_fatal);
1827                 object_property_set_bool(core, true, "realized", &error_fatal);
1828             }
1829         }
1830         g_free(type);
1831     } else {
1832         for (i = 0; i < smp_cpus; i++) {
1833             PowerPCCPU *cpu = cpu_ppc_init(machine->cpu_model);
1834             if (cpu == NULL) {
1835                 error_report("Unable to find PowerPC CPU definition");
1836                 exit(1);
1837             }
1838             spapr_cpu_init(spapr, cpu, &error_fatal);
1839        }
1840     }
1841 
1842     if (kvm_enabled()) {
1843         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1844         kvmppc_enable_logical_ci_hcalls();
1845         kvmppc_enable_set_mode_hcall();
1846 
1847         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
1848         kvmppc_enable_clear_ref_mod_hcalls();
1849     }
1850 
1851     /* allocate RAM */
1852     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1853                                          machine->ram_size);
1854     memory_region_add_subregion(sysmem, 0, ram);
1855 
1856     if (rma_alloc_size && rma) {
1857         rma_region = g_new(MemoryRegion, 1);
1858         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1859                                    rma_alloc_size, rma);
1860         vmstate_register_ram_global(rma_region);
1861         memory_region_add_subregion(sysmem, 0, rma_region);
1862     }
1863 
1864     /* initialize hotplug memory address space */
1865     if (machine->ram_size < machine->maxram_size) {
1866         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
1867         /*
1868          * Limit the number of hotpluggable memory slots to half the number
1869          * slots that KVM supports, leaving the other half for PCI and other
1870          * devices. However ensure that number of slots doesn't drop below 32.
1871          */
1872         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
1873                            SPAPR_MAX_RAM_SLOTS;
1874 
1875         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
1876             max_memslots = SPAPR_MAX_RAM_SLOTS;
1877         }
1878         if (machine->ram_slots > max_memslots) {
1879             error_report("Specified number of memory slots %"
1880                          PRIu64" exceeds max supported %d",
1881                          machine->ram_slots, max_memslots);
1882             exit(1);
1883         }
1884 
1885         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
1886                                               SPAPR_HOTPLUG_MEM_ALIGN);
1887         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
1888                            "hotplug-memory", hotplug_mem_size);
1889         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
1890                                     &spapr->hotplug_memory.mr);
1891     }
1892 
1893     if (smc->dr_lmb_enabled) {
1894         spapr_create_lmb_dr_connectors(spapr);
1895     }
1896 
1897     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1898     if (!filename) {
1899         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1900         exit(1);
1901     }
1902     spapr->rtas_size = get_image_size(filename);
1903     if (spapr->rtas_size < 0) {
1904         error_report("Could not get size of LPAR rtas '%s'", filename);
1905         exit(1);
1906     }
1907     spapr->rtas_blob = g_malloc(spapr->rtas_size);
1908     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1909         error_report("Could not load LPAR rtas '%s'", filename);
1910         exit(1);
1911     }
1912     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1913         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1914                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1915         exit(1);
1916     }
1917     g_free(filename);
1918 
1919     /* Set up RTAS event infrastructure */
1920     spapr_events_init(spapr);
1921 
1922     /* Set up the RTC RTAS interfaces */
1923     spapr_rtc_create(spapr);
1924 
1925     /* Set up VIO bus */
1926     spapr->vio_bus = spapr_vio_bus_init();
1927 
1928     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1929         if (serial_hds[i]) {
1930             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1931         }
1932     }
1933 
1934     /* We always have at least the nvram device on VIO */
1935     spapr_create_nvram(spapr);
1936 
1937     /* Set up PCI */
1938     spapr_pci_rtas_init();
1939 
1940     phb = spapr_create_phb(spapr, 0);
1941 
1942     for (i = 0; i < nb_nics; i++) {
1943         NICInfo *nd = &nd_table[i];
1944 
1945         if (!nd->model) {
1946             nd->model = g_strdup("ibmveth");
1947         }
1948 
1949         if (strcmp(nd->model, "ibmveth") == 0) {
1950             spapr_vlan_create(spapr->vio_bus, nd);
1951         } else {
1952             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1953         }
1954     }
1955 
1956     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1957         spapr_vscsi_create(spapr->vio_bus);
1958     }
1959 
1960     /* Graphics */
1961     if (spapr_vga_init(phb->bus, &error_fatal)) {
1962         spapr->has_graphics = true;
1963         machine->usb |= defaults_enabled() && !machine->usb_disabled;
1964     }
1965 
1966     if (machine->usb) {
1967         if (smc->use_ohci_by_default) {
1968             pci_create_simple(phb->bus, -1, "pci-ohci");
1969         } else {
1970             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
1971         }
1972 
1973         if (spapr->has_graphics) {
1974             USBBus *usb_bus = usb_bus_find(-1);
1975 
1976             usb_create_simple(usb_bus, "usb-kbd");
1977             usb_create_simple(usb_bus, "usb-mouse");
1978         }
1979     }
1980 
1981     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1982         error_report(
1983             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
1984             MIN_RMA_SLOF);
1985         exit(1);
1986     }
1987 
1988     if (kernel_filename) {
1989         uint64_t lowaddr = 0;
1990 
1991         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
1992                                       NULL, NULL, &lowaddr, NULL, 1,
1993                                       PPC_ELF_MACHINE, 0, 0);
1994         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1995             spapr->kernel_size = load_elf(kernel_filename,
1996                                           translate_kernel_address, NULL, NULL,
1997                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
1998                                           0, 0);
1999             spapr->kernel_le = spapr->kernel_size > 0;
2000         }
2001         if (spapr->kernel_size < 0) {
2002             error_report("error loading %s: %s", kernel_filename,
2003                          load_elf_strerror(spapr->kernel_size));
2004             exit(1);
2005         }
2006 
2007         /* load initrd */
2008         if (initrd_filename) {
2009             /* Try to locate the initrd in the gap between the kernel
2010              * and the firmware. Add a bit of space just in case
2011              */
2012             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2013                                   + 0x1ffff) & ~0xffff;
2014             spapr->initrd_size = load_image_targphys(initrd_filename,
2015                                                      spapr->initrd_base,
2016                                                      load_limit
2017                                                      - spapr->initrd_base);
2018             if (spapr->initrd_size < 0) {
2019                 error_report("could not load initial ram disk '%s'",
2020                              initrd_filename);
2021                 exit(1);
2022             }
2023         }
2024     }
2025 
2026     if (bios_name == NULL) {
2027         bios_name = FW_FILE_NAME;
2028     }
2029     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2030     if (!filename) {
2031         error_report("Could not find LPAR firmware '%s'", bios_name);
2032         exit(1);
2033     }
2034     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2035     if (fw_size <= 0) {
2036         error_report("Could not load LPAR firmware '%s'", filename);
2037         exit(1);
2038     }
2039     g_free(filename);
2040 
2041     /* FIXME: Should register things through the MachineState's qdev
2042      * interface, this is a legacy from the sPAPREnvironment structure
2043      * which predated MachineState but had a similar function */
2044     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2045     register_savevm_live(NULL, "spapr/htab", -1, 1,
2046                          &savevm_htab_handlers, spapr);
2047 
2048     /* used by RTAS */
2049     QTAILQ_INIT(&spapr->ccs_list);
2050     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2051 
2052     qemu_register_boot_set(spapr_boot_set, spapr);
2053 }
2054 
2055 static int spapr_kvm_type(const char *vm_type)
2056 {
2057     if (!vm_type) {
2058         return 0;
2059     }
2060 
2061     if (!strcmp(vm_type, "HV")) {
2062         return 1;
2063     }
2064 
2065     if (!strcmp(vm_type, "PR")) {
2066         return 2;
2067     }
2068 
2069     error_report("Unknown kvm-type specified '%s'", vm_type);
2070     exit(1);
2071 }
2072 
2073 /*
2074  * Implementation of an interface to adjust firmware path
2075  * for the bootindex property handling.
2076  */
2077 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2078                                    DeviceState *dev)
2079 {
2080 #define CAST(type, obj, name) \
2081     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2082     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2083     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2084 
2085     if (d) {
2086         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2087         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2088         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2089 
2090         if (spapr) {
2091             /*
2092              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2093              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2094              * in the top 16 bits of the 64-bit LUN
2095              */
2096             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2097             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2098                                    (uint64_t)id << 48);
2099         } else if (virtio) {
2100             /*
2101              * We use SRP luns of the form 01000000 | (target << 8) | lun
2102              * in the top 32 bits of the 64-bit LUN
2103              * Note: the quote above is from SLOF and it is wrong,
2104              * the actual binding is:
2105              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2106              */
2107             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2108             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2109                                    (uint64_t)id << 32);
2110         } else if (usb) {
2111             /*
2112              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2113              * in the top 32 bits of the 64-bit LUN
2114              */
2115             unsigned usb_port = atoi(usb->port->path);
2116             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2117             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2118                                    (uint64_t)id << 32);
2119         }
2120     }
2121 
2122     if (phb) {
2123         /* Replace "pci" with "pci@800000020000000" */
2124         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2125     }
2126 
2127     return NULL;
2128 }
2129 
2130 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2131 {
2132     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2133 
2134     return g_strdup(spapr->kvm_type);
2135 }
2136 
2137 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2138 {
2139     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2140 
2141     g_free(spapr->kvm_type);
2142     spapr->kvm_type = g_strdup(value);
2143 }
2144 
2145 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2146 {
2147     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2148 
2149     return spapr->use_hotplug_event_source;
2150 }
2151 
2152 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2153                                             Error **errp)
2154 {
2155     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2156 
2157     spapr->use_hotplug_event_source = value;
2158 }
2159 
2160 static void spapr_machine_initfn(Object *obj)
2161 {
2162     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2163 
2164     spapr->htab_fd = -1;
2165     spapr->use_hotplug_event_source = true;
2166     object_property_add_str(obj, "kvm-type",
2167                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2168     object_property_set_description(obj, "kvm-type",
2169                                     "Specifies the KVM virtualization mode (HV, PR)",
2170                                     NULL);
2171     object_property_add_bool(obj, "modern-hotplug-events",
2172                             spapr_get_modern_hotplug_events,
2173                             spapr_set_modern_hotplug_events,
2174                             NULL);
2175     object_property_set_description(obj, "modern-hotplug-events",
2176                                     "Use dedicated hotplug event mechanism in"
2177                                     " place of standard EPOW events when possible"
2178                                     " (required for memory hot-unplug support)",
2179                                     NULL);
2180 }
2181 
2182 static void spapr_machine_finalizefn(Object *obj)
2183 {
2184     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2185 
2186     g_free(spapr->kvm_type);
2187 }
2188 
2189 static void ppc_cpu_do_nmi_on_cpu(CPUState *cs, run_on_cpu_data arg)
2190 {
2191     cpu_synchronize_state(cs);
2192     ppc_cpu_do_system_reset(cs);
2193 }
2194 
2195 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2196 {
2197     CPUState *cs;
2198 
2199     CPU_FOREACH(cs) {
2200         async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, RUN_ON_CPU_NULL);
2201     }
2202 }
2203 
2204 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2205                            uint32_t node, bool dedicated_hp_event_source,
2206                            Error **errp)
2207 {
2208     sPAPRDRConnector *drc;
2209     sPAPRDRConnectorClass *drck;
2210     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2211     int i, fdt_offset, fdt_size;
2212     void *fdt;
2213     uint64_t addr = addr_start;
2214 
2215     for (i = 0; i < nr_lmbs; i++) {
2216         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2217                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2218         g_assert(drc);
2219 
2220         fdt = create_device_tree(&fdt_size);
2221         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2222                                                 SPAPR_MEMORY_BLOCK_SIZE);
2223 
2224         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2225         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2226         addr += SPAPR_MEMORY_BLOCK_SIZE;
2227     }
2228     /* send hotplug notification to the
2229      * guest only in case of hotplugged memory
2230      */
2231     if (dev->hotplugged) {
2232         if (dedicated_hp_event_source) {
2233             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2234                     addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2235             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2236             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2237                                                    nr_lmbs,
2238                                                    drck->get_index(drc));
2239         } else {
2240             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2241                                            nr_lmbs);
2242         }
2243     }
2244 }
2245 
2246 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2247                               uint32_t node, Error **errp)
2248 {
2249     Error *local_err = NULL;
2250     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2251     PCDIMMDevice *dimm = PC_DIMM(dev);
2252     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2253     MemoryRegion *mr = ddc->get_memory_region(dimm);
2254     uint64_t align = memory_region_get_alignment(mr);
2255     uint64_t size = memory_region_size(mr);
2256     uint64_t addr;
2257 
2258     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2259         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2260                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2261         goto out;
2262     }
2263 
2264     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2265     if (local_err) {
2266         goto out;
2267     }
2268 
2269     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2270     if (local_err) {
2271         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2272         goto out;
2273     }
2274 
2275     spapr_add_lmbs(dev, addr, size, node,
2276                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2277                    &error_abort);
2278 
2279 out:
2280     error_propagate(errp, local_err);
2281 }
2282 
2283 typedef struct sPAPRDIMMState {
2284     uint32_t nr_lmbs;
2285 } sPAPRDIMMState;
2286 
2287 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2288 {
2289     sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2290     HotplugHandler *hotplug_ctrl;
2291 
2292     if (--ds->nr_lmbs) {
2293         return;
2294     }
2295 
2296     g_free(ds);
2297 
2298     /*
2299      * Now that all the LMBs have been removed by the guest, call the
2300      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2301      */
2302     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2303     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2304 }
2305 
2306 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2307                            Error **errp)
2308 {
2309     sPAPRDRConnector *drc;
2310     sPAPRDRConnectorClass *drck;
2311     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2312     int i;
2313     sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2314     uint64_t addr = addr_start;
2315 
2316     ds->nr_lmbs = nr_lmbs;
2317     for (i = 0; i < nr_lmbs; i++) {
2318         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2319                 addr / SPAPR_MEMORY_BLOCK_SIZE);
2320         g_assert(drc);
2321 
2322         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2323         drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2324         addr += SPAPR_MEMORY_BLOCK_SIZE;
2325     }
2326 
2327     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2328                                    addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2329     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2330     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2331                                               nr_lmbs,
2332                                               drck->get_index(drc));
2333 }
2334 
2335 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2336                                 Error **errp)
2337 {
2338     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2339     PCDIMMDevice *dimm = PC_DIMM(dev);
2340     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2341     MemoryRegion *mr = ddc->get_memory_region(dimm);
2342 
2343     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2344     object_unparent(OBJECT(dev));
2345 }
2346 
2347 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2348                                         DeviceState *dev, Error **errp)
2349 {
2350     Error *local_err = NULL;
2351     PCDIMMDevice *dimm = PC_DIMM(dev);
2352     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2353     MemoryRegion *mr = ddc->get_memory_region(dimm);
2354     uint64_t size = memory_region_size(mr);
2355     uint64_t addr;
2356 
2357     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2358     if (local_err) {
2359         goto out;
2360     }
2361 
2362     spapr_del_lmbs(dev, addr, size, &error_abort);
2363 out:
2364     error_propagate(errp, local_err);
2365 }
2366 
2367 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2368                                     sPAPRMachineState *spapr)
2369 {
2370     PowerPCCPU *cpu = POWERPC_CPU(cs);
2371     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2372     int id = ppc_get_vcpu_dt_id(cpu);
2373     void *fdt;
2374     int offset, fdt_size;
2375     char *nodename;
2376 
2377     fdt = create_device_tree(&fdt_size);
2378     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2379     offset = fdt_add_subnode(fdt, 0, nodename);
2380 
2381     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2382     g_free(nodename);
2383 
2384     *fdt_offset = offset;
2385     return fdt;
2386 }
2387 
2388 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2389                                       DeviceState *dev, Error **errp)
2390 {
2391     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2392 
2393     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2394         int node;
2395 
2396         if (!smc->dr_lmb_enabled) {
2397             error_setg(errp, "Memory hotplug not supported for this machine");
2398             return;
2399         }
2400         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2401         if (*errp) {
2402             return;
2403         }
2404         if (node < 0 || node >= MAX_NODES) {
2405             error_setg(errp, "Invaild node %d", node);
2406             return;
2407         }
2408 
2409         /*
2410          * Currently PowerPC kernel doesn't allow hot-adding memory to
2411          * memory-less node, but instead will silently add the memory
2412          * to the first node that has some memory. This causes two
2413          * unexpected behaviours for the user.
2414          *
2415          * - Memory gets hotplugged to a different node than what the user
2416          *   specified.
2417          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2418          *   to memory-less node, a reboot will set things accordingly
2419          *   and the previously hotplugged memory now ends in the right node.
2420          *   This appears as if some memory moved from one node to another.
2421          *
2422          * So until kernel starts supporting memory hotplug to memory-less
2423          * nodes, just prevent such attempts upfront in QEMU.
2424          */
2425         if (nb_numa_nodes && !numa_info[node].node_mem) {
2426             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2427                        node);
2428             return;
2429         }
2430 
2431         spapr_memory_plug(hotplug_dev, dev, node, errp);
2432     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2433         spapr_core_plug(hotplug_dev, dev, errp);
2434     }
2435 }
2436 
2437 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2438                                       DeviceState *dev, Error **errp)
2439 {
2440     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2441     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2442 
2443     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2444         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2445             spapr_memory_unplug(hotplug_dev, dev, errp);
2446         } else {
2447             error_setg(errp, "Memory hot unplug not supported for this guest");
2448         }
2449     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2450         if (!mc->query_hotpluggable_cpus) {
2451             error_setg(errp, "CPU hot unplug not supported on this machine");
2452             return;
2453         }
2454         spapr_core_unplug(hotplug_dev, dev, errp);
2455     }
2456 }
2457 
2458 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2459                                                 DeviceState *dev, Error **errp)
2460 {
2461     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2462     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2463 
2464     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2465         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2466             spapr_memory_unplug_request(hotplug_dev, dev, errp);
2467         } else {
2468             /* NOTE: this means there is a window after guest reset, prior to
2469              * CAS negotiation, where unplug requests will fail due to the
2470              * capability not being detected yet. This is a bit different than
2471              * the case with PCI unplug, where the events will be queued and
2472              * eventually handled by the guest after boot
2473              */
2474             error_setg(errp, "Memory hot unplug not supported for this guest");
2475         }
2476     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2477         if (!mc->query_hotpluggable_cpus) {
2478             error_setg(errp, "CPU hot unplug not supported on this machine");
2479             return;
2480         }
2481         spapr_core_unplug(hotplug_dev, dev, errp);
2482     }
2483 }
2484 
2485 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2486                                           DeviceState *dev, Error **errp)
2487 {
2488     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2489         spapr_core_pre_plug(hotplug_dev, dev, errp);
2490     }
2491 }
2492 
2493 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2494                                                  DeviceState *dev)
2495 {
2496     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2497         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2498         return HOTPLUG_HANDLER(machine);
2499     }
2500     return NULL;
2501 }
2502 
2503 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2504 {
2505     /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2506      * socket means much for the paravirtualized PAPR platform) */
2507     return cpu_index / smp_threads / smp_cores;
2508 }
2509 
2510 static HotpluggableCPUList *spapr_query_hotpluggable_cpus(MachineState *machine)
2511 {
2512     int i;
2513     HotpluggableCPUList *head = NULL;
2514     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2515     int spapr_max_cores = max_cpus / smp_threads;
2516 
2517     for (i = 0; i < spapr_max_cores; i++) {
2518         HotpluggableCPUList *list_item = g_new0(typeof(*list_item), 1);
2519         HotpluggableCPU *cpu_item = g_new0(typeof(*cpu_item), 1);
2520         CpuInstanceProperties *cpu_props = g_new0(typeof(*cpu_props), 1);
2521 
2522         cpu_item->type = spapr_get_cpu_core_type(machine->cpu_model);
2523         cpu_item->vcpus_count = smp_threads;
2524         cpu_props->has_core_id = true;
2525         cpu_props->core_id = i * smp_threads;
2526         /* TODO: add 'has_node/node' here to describe
2527            to which node core belongs */
2528 
2529         cpu_item->props = cpu_props;
2530         if (spapr->cores[i]) {
2531             cpu_item->has_qom_path = true;
2532             cpu_item->qom_path = object_get_canonical_path(spapr->cores[i]);
2533         }
2534         list_item->value = cpu_item;
2535         list_item->next = head;
2536         head = list_item;
2537     }
2538     return head;
2539 }
2540 
2541 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
2542                                 uint64_t *buid, hwaddr *pio,
2543                                 hwaddr *mmio32, hwaddr *mmio64,
2544                                 unsigned n_dma, uint32_t *liobns, Error **errp)
2545 {
2546     /*
2547      * New-style PHB window placement.
2548      *
2549      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
2550      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
2551      * windows.
2552      *
2553      * Some guest kernels can't work with MMIO windows above 1<<46
2554      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
2555      *
2556      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
2557      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
2558      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
2559      * 1TiB 64-bit MMIO windows for each PHB.
2560      */
2561     const uint64_t base_buid = 0x800000020000000ULL;
2562     const int max_phbs =
2563         (SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / SPAPR_PCI_MEM64_WIN_SIZE - 1;
2564     int i;
2565 
2566     /* Sanity check natural alignments */
2567     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2568     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2569     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
2570     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
2571     /* Sanity check bounds */
2572     QEMU_BUILD_BUG_ON((max_phbs * SPAPR_PCI_IO_WIN_SIZE) > SPAPR_PCI_MEM32_WIN_SIZE);
2573     QEMU_BUILD_BUG_ON((max_phbs * SPAPR_PCI_MEM32_WIN_SIZE) > SPAPR_PCI_MEM64_WIN_SIZE);
2574 
2575     if (index >= max_phbs) {
2576         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
2577                    max_phbs - 1);
2578         return;
2579     }
2580 
2581     *buid = base_buid + index;
2582     for (i = 0; i < n_dma; ++i) {
2583         liobns[i] = SPAPR_PCI_LIOBN(index, i);
2584     }
2585 
2586     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
2587     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
2588     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
2589 }
2590 
2591 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2592 {
2593     MachineClass *mc = MACHINE_CLASS(oc);
2594     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2595     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2596     NMIClass *nc = NMI_CLASS(oc);
2597     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2598 
2599     mc->desc = "pSeries Logical Partition (PAPR compliant)";
2600 
2601     /*
2602      * We set up the default / latest behaviour here.  The class_init
2603      * functions for the specific versioned machine types can override
2604      * these details for backwards compatibility
2605      */
2606     mc->init = ppc_spapr_init;
2607     mc->reset = ppc_spapr_reset;
2608     mc->block_default_type = IF_SCSI;
2609     mc->max_cpus = 255;
2610     mc->no_parallel = 1;
2611     mc->default_boot_order = "";
2612     mc->default_ram_size = 512 * M_BYTE;
2613     mc->kvm_type = spapr_kvm_type;
2614     mc->has_dynamic_sysbus = true;
2615     mc->pci_allow_0_address = true;
2616     mc->get_hotplug_handler = spapr_get_hotplug_handler;
2617     hc->pre_plug = spapr_machine_device_pre_plug;
2618     hc->plug = spapr_machine_device_plug;
2619     hc->unplug = spapr_machine_device_unplug;
2620     mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
2621     hc->unplug_request = spapr_machine_device_unplug_request;
2622 
2623     smc->dr_lmb_enabled = true;
2624     smc->tcg_default_cpu = "POWER8";
2625     mc->query_hotpluggable_cpus = spapr_query_hotpluggable_cpus;
2626     fwc->get_dev_path = spapr_get_fw_dev_path;
2627     nc->nmi_monitor_handler = spapr_nmi;
2628     smc->phb_placement = spapr_phb_placement;
2629 }
2630 
2631 static const TypeInfo spapr_machine_info = {
2632     .name          = TYPE_SPAPR_MACHINE,
2633     .parent        = TYPE_MACHINE,
2634     .abstract      = true,
2635     .instance_size = sizeof(sPAPRMachineState),
2636     .instance_init = spapr_machine_initfn,
2637     .instance_finalize = spapr_machine_finalizefn,
2638     .class_size    = sizeof(sPAPRMachineClass),
2639     .class_init    = spapr_machine_class_init,
2640     .interfaces = (InterfaceInfo[]) {
2641         { TYPE_FW_PATH_PROVIDER },
2642         { TYPE_NMI },
2643         { TYPE_HOTPLUG_HANDLER },
2644         { }
2645     },
2646 };
2647 
2648 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
2649     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
2650                                                     void *data)      \
2651     {                                                                \
2652         MachineClass *mc = MACHINE_CLASS(oc);                        \
2653         spapr_machine_##suffix##_class_options(mc);                  \
2654         if (latest) {                                                \
2655             mc->alias = "pseries";                                   \
2656             mc->is_default = 1;                                      \
2657         }                                                            \
2658     }                                                                \
2659     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
2660     {                                                                \
2661         MachineState *machine = MACHINE(obj);                        \
2662         spapr_machine_##suffix##_instance_options(machine);          \
2663     }                                                                \
2664     static const TypeInfo spapr_machine_##suffix##_info = {          \
2665         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
2666         .parent = TYPE_SPAPR_MACHINE,                                \
2667         .class_init = spapr_machine_##suffix##_class_init,           \
2668         .instance_init = spapr_machine_##suffix##_instance_init,     \
2669     };                                                               \
2670     static void spapr_machine_register_##suffix(void)                \
2671     {                                                                \
2672         type_register(&spapr_machine_##suffix##_info);               \
2673     }                                                                \
2674     type_init(spapr_machine_register_##suffix)
2675 
2676 /*
2677  * pseries-2.8
2678  */
2679 static void spapr_machine_2_8_instance_options(MachineState *machine)
2680 {
2681 }
2682 
2683 static void spapr_machine_2_8_class_options(MachineClass *mc)
2684 {
2685     /* Defaults for the latest behaviour inherited from the base class */
2686 }
2687 
2688 DEFINE_SPAPR_MACHINE(2_8, "2.8", true);
2689 
2690 /*
2691  * pseries-2.7
2692  */
2693 #define SPAPR_COMPAT_2_7                            \
2694     HW_COMPAT_2_7                                   \
2695     {                                               \
2696         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
2697         .property = "mem_win_size",                 \
2698         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
2699     },                                              \
2700     {                                               \
2701         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
2702         .property = "mem64_win_size",               \
2703         .value    = "0",                            \
2704     },
2705 
2706 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
2707                               uint64_t *buid, hwaddr *pio,
2708                               hwaddr *mmio32, hwaddr *mmio64,
2709                               unsigned n_dma, uint32_t *liobns, Error **errp)
2710 {
2711     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
2712     const uint64_t base_buid = 0x800000020000000ULL;
2713     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
2714     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
2715     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
2716     const uint32_t max_index = 255;
2717     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
2718 
2719     uint64_t ram_top = MACHINE(spapr)->ram_size;
2720     hwaddr phb0_base, phb_base;
2721     int i;
2722 
2723     /* Do we have hotpluggable memory? */
2724     if (MACHINE(spapr)->maxram_size > ram_top) {
2725         /* Can't just use maxram_size, because there may be an
2726          * alignment gap between normal and hotpluggable memory
2727          * regions */
2728         ram_top = spapr->hotplug_memory.base +
2729             memory_region_size(&spapr->hotplug_memory.mr);
2730     }
2731 
2732     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
2733 
2734     if (index > max_index) {
2735         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
2736                    max_index);
2737         return;
2738     }
2739 
2740     *buid = base_buid + index;
2741     for (i = 0; i < n_dma; ++i) {
2742         liobns[i] = SPAPR_PCI_LIOBN(index, i);
2743     }
2744 
2745     phb_base = phb0_base + index * phb_spacing;
2746     *pio = phb_base + pio_offset;
2747     *mmio32 = phb_base + mmio_offset;
2748     /*
2749      * We don't set the 64-bit MMIO window, relying on the PHB's
2750      * fallback behaviour of automatically splitting a large "32-bit"
2751      * window into contiguous 32-bit and 64-bit windows
2752      */
2753 }
2754 
2755 static void spapr_machine_2_7_instance_options(MachineState *machine)
2756 {
2757     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2758 
2759     spapr_machine_2_8_instance_options(machine);
2760     spapr->use_hotplug_event_source = false;
2761 }
2762 
2763 static void spapr_machine_2_7_class_options(MachineClass *mc)
2764 {
2765     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2766 
2767     spapr_machine_2_8_class_options(mc);
2768     smc->tcg_default_cpu = "POWER7";
2769     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
2770     smc->phb_placement = phb_placement_2_7;
2771 }
2772 
2773 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
2774 
2775 /*
2776  * pseries-2.6
2777  */
2778 #define SPAPR_COMPAT_2_6 \
2779     HW_COMPAT_2_6 \
2780     { \
2781         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2782         .property = "ddw",\
2783         .value    = stringify(off),\
2784     },
2785 
2786 static void spapr_machine_2_6_instance_options(MachineState *machine)
2787 {
2788     spapr_machine_2_7_instance_options(machine);
2789 }
2790 
2791 static void spapr_machine_2_6_class_options(MachineClass *mc)
2792 {
2793     spapr_machine_2_7_class_options(mc);
2794     mc->query_hotpluggable_cpus = NULL;
2795     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
2796 }
2797 
2798 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
2799 
2800 /*
2801  * pseries-2.5
2802  */
2803 #define SPAPR_COMPAT_2_5 \
2804     HW_COMPAT_2_5 \
2805     { \
2806         .driver   = "spapr-vlan", \
2807         .property = "use-rx-buffer-pools", \
2808         .value    = "off", \
2809     },
2810 
2811 static void spapr_machine_2_5_instance_options(MachineState *machine)
2812 {
2813     spapr_machine_2_6_instance_options(machine);
2814 }
2815 
2816 static void spapr_machine_2_5_class_options(MachineClass *mc)
2817 {
2818     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2819 
2820     spapr_machine_2_6_class_options(mc);
2821     smc->use_ohci_by_default = true;
2822     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
2823 }
2824 
2825 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
2826 
2827 /*
2828  * pseries-2.4
2829  */
2830 #define SPAPR_COMPAT_2_4 \
2831         HW_COMPAT_2_4
2832 
2833 static void spapr_machine_2_4_instance_options(MachineState *machine)
2834 {
2835     spapr_machine_2_5_instance_options(machine);
2836 }
2837 
2838 static void spapr_machine_2_4_class_options(MachineClass *mc)
2839 {
2840     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2841 
2842     spapr_machine_2_5_class_options(mc);
2843     smc->dr_lmb_enabled = false;
2844     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
2845 }
2846 
2847 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
2848 
2849 /*
2850  * pseries-2.3
2851  */
2852 #define SPAPR_COMPAT_2_3 \
2853         HW_COMPAT_2_3 \
2854         {\
2855             .driver   = "spapr-pci-host-bridge",\
2856             .property = "dynamic-reconfiguration",\
2857             .value    = "off",\
2858         },
2859 
2860 static void spapr_machine_2_3_instance_options(MachineState *machine)
2861 {
2862     spapr_machine_2_4_instance_options(machine);
2863     savevm_skip_section_footers();
2864     global_state_set_optional();
2865     savevm_skip_configuration();
2866 }
2867 
2868 static void spapr_machine_2_3_class_options(MachineClass *mc)
2869 {
2870     spapr_machine_2_4_class_options(mc);
2871     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
2872 }
2873 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
2874 
2875 /*
2876  * pseries-2.2
2877  */
2878 
2879 #define SPAPR_COMPAT_2_2 \
2880         HW_COMPAT_2_2 \
2881         {\
2882             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2883             .property = "mem_win_size",\
2884             .value    = "0x20000000",\
2885         },
2886 
2887 static void spapr_machine_2_2_instance_options(MachineState *machine)
2888 {
2889     spapr_machine_2_3_instance_options(machine);
2890     machine->suppress_vmdesc = true;
2891 }
2892 
2893 static void spapr_machine_2_2_class_options(MachineClass *mc)
2894 {
2895     spapr_machine_2_3_class_options(mc);
2896     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
2897 }
2898 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
2899 
2900 /*
2901  * pseries-2.1
2902  */
2903 #define SPAPR_COMPAT_2_1 \
2904         HW_COMPAT_2_1
2905 
2906 static void spapr_machine_2_1_instance_options(MachineState *machine)
2907 {
2908     spapr_machine_2_2_instance_options(machine);
2909 }
2910 
2911 static void spapr_machine_2_1_class_options(MachineClass *mc)
2912 {
2913     spapr_machine_2_2_class_options(mc);
2914     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
2915 }
2916 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
2917 
2918 static void spapr_machine_register_types(void)
2919 {
2920     type_register_static(&spapr_machine_info);
2921 }
2922 
2923 type_init(spapr_machine_register_types)
2924