xref: /openbmc/qemu/hw/ppc/spapr.c (revision b45c03f5)
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 "sysemu/sysemu.h"
28 #include "sysemu/numa.h"
29 #include "hw/hw.h"
30 #include "hw/fw-path-provider.h"
31 #include "elf.h"
32 #include "net/net.h"
33 #include "sysemu/block-backend.h"
34 #include "sysemu/cpus.h"
35 #include "sysemu/kvm.h"
36 #include "kvm_ppc.h"
37 #include "migration/migration.h"
38 #include "mmu-hash64.h"
39 #include "qom/cpu.h"
40 
41 #include "hw/boards.h"
42 #include "hw/ppc/ppc.h"
43 #include "hw/loader.h"
44 
45 #include "hw/ppc/spapr.h"
46 #include "hw/ppc/spapr_vio.h"
47 #include "hw/pci-host/spapr.h"
48 #include "hw/ppc/xics.h"
49 #include "hw/pci/msi.h"
50 
51 #include "hw/pci/pci.h"
52 #include "hw/scsi/scsi.h"
53 #include "hw/virtio/virtio-scsi.h"
54 
55 #include "exec/address-spaces.h"
56 #include "hw/usb.h"
57 #include "qemu/config-file.h"
58 #include "qemu/error-report.h"
59 #include "trace.h"
60 #include "hw/nmi.h"
61 
62 #include "hw/compat.h"
63 
64 #include <libfdt.h>
65 
66 /* SLOF memory layout:
67  *
68  * SLOF raw image loaded at 0, copies its romfs right below the flat
69  * device-tree, then position SLOF itself 31M below that
70  *
71  * So we set FW_OVERHEAD to 40MB which should account for all of that
72  * and more
73  *
74  * We load our kernel at 4M, leaving space for SLOF initial image
75  */
76 #define FDT_MAX_SIZE            0x40000
77 #define RTAS_MAX_SIZE           0x10000
78 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
79 #define FW_MAX_SIZE             0x400000
80 #define FW_FILE_NAME            "slof.bin"
81 #define FW_OVERHEAD             0x2800000
82 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
83 
84 #define MIN_RMA_SLOF            128UL
85 
86 #define TIMEBASE_FREQ           512000000ULL
87 
88 #define MAX_CPUS                255
89 
90 #define PHANDLE_XICP            0x00001111
91 
92 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
93 
94 static XICSState *try_create_xics(const char *type, int nr_servers,
95                                   int nr_irqs, Error **errp)
96 {
97     Error *err = NULL;
98     DeviceState *dev;
99 
100     dev = qdev_create(NULL, type);
101     qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
102     qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
103     object_property_set_bool(OBJECT(dev), true, "realized", &err);
104     if (err) {
105         error_propagate(errp, err);
106         object_unparent(OBJECT(dev));
107         return NULL;
108     }
109     return XICS_COMMON(dev);
110 }
111 
112 static XICSState *xics_system_init(MachineState *machine,
113                                    int nr_servers, int nr_irqs)
114 {
115     XICSState *icp = NULL;
116 
117     if (kvm_enabled()) {
118         Error *err = NULL;
119 
120         if (machine_kernel_irqchip_allowed(machine)) {
121             icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs, &err);
122         }
123         if (machine_kernel_irqchip_required(machine) && !icp) {
124             error_report("kernel_irqchip requested but unavailable: %s",
125                          error_get_pretty(err));
126         }
127     }
128 
129     if (!icp) {
130         icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs, &error_abort);
131     }
132 
133     return icp;
134 }
135 
136 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
137                                   int smt_threads)
138 {
139     int i, ret = 0;
140     uint32_t servers_prop[smt_threads];
141     uint32_t gservers_prop[smt_threads * 2];
142     int index = ppc_get_vcpu_dt_id(cpu);
143 
144     if (cpu->cpu_version) {
145         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
146         if (ret < 0) {
147             return ret;
148         }
149     }
150 
151     /* Build interrupt servers and gservers properties */
152     for (i = 0; i < smt_threads; i++) {
153         servers_prop[i] = cpu_to_be32(index + i);
154         /* Hack, direct the group queues back to cpu 0 */
155         gservers_prop[i*2] = cpu_to_be32(index + i);
156         gservers_prop[i*2 + 1] = 0;
157     }
158     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
159                       servers_prop, sizeof(servers_prop));
160     if (ret < 0) {
161         return ret;
162     }
163     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
164                       gservers_prop, sizeof(gservers_prop));
165 
166     return ret;
167 }
168 
169 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
170 {
171     int ret = 0;
172     PowerPCCPU *cpu = POWERPC_CPU(cs);
173     int index = ppc_get_vcpu_dt_id(cpu);
174     uint32_t associativity[] = {cpu_to_be32(0x5),
175                                 cpu_to_be32(0x0),
176                                 cpu_to_be32(0x0),
177                                 cpu_to_be32(0x0),
178                                 cpu_to_be32(cs->numa_node),
179                                 cpu_to_be32(index)};
180 
181     /* Advertise NUMA via ibm,associativity */
182     if (nb_numa_nodes > 1) {
183         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
184                           sizeof(associativity));
185     }
186 
187     return ret;
188 }
189 
190 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
191 {
192     int ret = 0, offset, cpus_offset;
193     CPUState *cs;
194     char cpu_model[32];
195     int smt = kvmppc_smt_threads();
196     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
197 
198     CPU_FOREACH(cs) {
199         PowerPCCPU *cpu = POWERPC_CPU(cs);
200         DeviceClass *dc = DEVICE_GET_CLASS(cs);
201         int index = ppc_get_vcpu_dt_id(cpu);
202 
203         if ((index % smt) != 0) {
204             continue;
205         }
206 
207         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
208 
209         cpus_offset = fdt_path_offset(fdt, "/cpus");
210         if (cpus_offset < 0) {
211             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
212                                           "cpus");
213             if (cpus_offset < 0) {
214                 return cpus_offset;
215             }
216         }
217         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
218         if (offset < 0) {
219             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
220             if (offset < 0) {
221                 return offset;
222             }
223         }
224 
225         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
226                           pft_size_prop, sizeof(pft_size_prop));
227         if (ret < 0) {
228             return ret;
229         }
230 
231         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
232         if (ret < 0) {
233             return ret;
234         }
235 
236         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
237                                      ppc_get_compat_smt_threads(cpu));
238         if (ret < 0) {
239             return ret;
240         }
241     }
242     return ret;
243 }
244 
245 
246 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
247                                      size_t maxsize)
248 {
249     size_t maxcells = maxsize / sizeof(uint32_t);
250     int i, j, count;
251     uint32_t *p = prop;
252 
253     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
254         struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
255 
256         if (!sps->page_shift) {
257             break;
258         }
259         for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
260             if (sps->enc[count].page_shift == 0) {
261                 break;
262             }
263         }
264         if ((p - prop) >= (maxcells - 3 - count * 2)) {
265             break;
266         }
267         *(p++) = cpu_to_be32(sps->page_shift);
268         *(p++) = cpu_to_be32(sps->slb_enc);
269         *(p++) = cpu_to_be32(count);
270         for (j = 0; j < count; j++) {
271             *(p++) = cpu_to_be32(sps->enc[j].page_shift);
272             *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
273         }
274     }
275 
276     return (p - prop) * sizeof(uint32_t);
277 }
278 
279 static hwaddr spapr_node0_size(void)
280 {
281     MachineState *machine = MACHINE(qdev_get_machine());
282 
283     if (nb_numa_nodes) {
284         int i;
285         for (i = 0; i < nb_numa_nodes; ++i) {
286             if (numa_info[i].node_mem) {
287                 return MIN(pow2floor(numa_info[i].node_mem),
288                            machine->ram_size);
289             }
290         }
291     }
292     return machine->ram_size;
293 }
294 
295 #define _FDT(exp) \
296     do { \
297         int ret = (exp);                                           \
298         if (ret < 0) {                                             \
299             fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
300                     #exp, fdt_strerror(ret));                      \
301             exit(1);                                               \
302         }                                                          \
303     } while (0)
304 
305 static void add_str(GString *s, const gchar *s1)
306 {
307     g_string_append_len(s, s1, strlen(s1) + 1);
308 }
309 
310 static void *spapr_create_fdt_skel(hwaddr initrd_base,
311                                    hwaddr initrd_size,
312                                    hwaddr kernel_size,
313                                    bool little_endian,
314                                    const char *kernel_cmdline,
315                                    uint32_t epow_irq)
316 {
317     void *fdt;
318     uint32_t start_prop = cpu_to_be32(initrd_base);
319     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
320     GString *hypertas = g_string_sized_new(256);
321     GString *qemu_hypertas = g_string_sized_new(256);
322     uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
323     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(max_cpus)};
324     unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
325     char *buf;
326 
327     add_str(hypertas, "hcall-pft");
328     add_str(hypertas, "hcall-term");
329     add_str(hypertas, "hcall-dabr");
330     add_str(hypertas, "hcall-interrupt");
331     add_str(hypertas, "hcall-tce");
332     add_str(hypertas, "hcall-vio");
333     add_str(hypertas, "hcall-splpar");
334     add_str(hypertas, "hcall-bulk");
335     add_str(hypertas, "hcall-set-mode");
336     add_str(qemu_hypertas, "hcall-memop1");
337 
338     fdt = g_malloc0(FDT_MAX_SIZE);
339     _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
340 
341     if (kernel_size) {
342         _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
343     }
344     if (initrd_size) {
345         _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
346     }
347     _FDT((fdt_finish_reservemap(fdt)));
348 
349     /* Root node */
350     _FDT((fdt_begin_node(fdt, "")));
351     _FDT((fdt_property_string(fdt, "device_type", "chrp")));
352     _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
353     _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
354 
355     /*
356      * Add info to guest to indentify which host is it being run on
357      * and what is the uuid of the guest
358      */
359     if (kvmppc_get_host_model(&buf)) {
360         _FDT((fdt_property_string(fdt, "host-model", buf)));
361         g_free(buf);
362     }
363     if (kvmppc_get_host_serial(&buf)) {
364         _FDT((fdt_property_string(fdt, "host-serial", buf)));
365         g_free(buf);
366     }
367 
368     buf = g_strdup_printf(UUID_FMT, qemu_uuid[0], qemu_uuid[1],
369                           qemu_uuid[2], qemu_uuid[3], qemu_uuid[4],
370                           qemu_uuid[5], qemu_uuid[6], qemu_uuid[7],
371                           qemu_uuid[8], qemu_uuid[9], qemu_uuid[10],
372                           qemu_uuid[11], qemu_uuid[12], qemu_uuid[13],
373                           qemu_uuid[14], qemu_uuid[15]);
374 
375     _FDT((fdt_property_string(fdt, "vm,uuid", buf)));
376     g_free(buf);
377 
378     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
379     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
380 
381     /* /chosen */
382     _FDT((fdt_begin_node(fdt, "chosen")));
383 
384     /* Set Form1_affinity */
385     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
386 
387     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
388     _FDT((fdt_property(fdt, "linux,initrd-start",
389                        &start_prop, sizeof(start_prop))));
390     _FDT((fdt_property(fdt, "linux,initrd-end",
391                        &end_prop, sizeof(end_prop))));
392     if (kernel_size) {
393         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
394                               cpu_to_be64(kernel_size) };
395 
396         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
397         if (little_endian) {
398             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
399         }
400     }
401     if (boot_menu) {
402         _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
403     }
404     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
405     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
406     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
407 
408     _FDT((fdt_end_node(fdt)));
409 
410     /* RTAS */
411     _FDT((fdt_begin_node(fdt, "rtas")));
412 
413     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
414         add_str(hypertas, "hcall-multi-tce");
415     }
416     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
417                        hypertas->len)));
418     g_string_free(hypertas, TRUE);
419     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
420                        qemu_hypertas->len)));
421     g_string_free(qemu_hypertas, TRUE);
422 
423     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
424         refpoints, sizeof(refpoints))));
425 
426     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
427     _FDT((fdt_property_cell(fdt, "rtas-event-scan-rate",
428                             RTAS_EVENT_SCAN_RATE)));
429 
430     /*
431      * According to PAPR, rtas ibm,os-term does not guarantee a return
432      * back to the guest cpu.
433      *
434      * While an additional ibm,extended-os-term property indicates that
435      * rtas call return will always occur. Set this property.
436      */
437     _FDT((fdt_property(fdt, "ibm,extended-os-term", NULL, 0)));
438 
439     _FDT((fdt_end_node(fdt)));
440 
441     /* interrupt controller */
442     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
443 
444     _FDT((fdt_property_string(fdt, "device_type",
445                               "PowerPC-External-Interrupt-Presentation")));
446     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
447     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
448     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
449                        interrupt_server_ranges_prop,
450                        sizeof(interrupt_server_ranges_prop))));
451     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
452     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
453     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
454 
455     _FDT((fdt_end_node(fdt)));
456 
457     /* vdevice */
458     _FDT((fdt_begin_node(fdt, "vdevice")));
459 
460     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
461     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
462     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
463     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
464     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
465     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
466 
467     _FDT((fdt_end_node(fdt)));
468 
469     /* event-sources */
470     spapr_events_fdt_skel(fdt, epow_irq);
471 
472     /* /hypervisor node */
473     if (kvm_enabled()) {
474         uint8_t hypercall[16];
475 
476         /* indicate KVM hypercall interface */
477         _FDT((fdt_begin_node(fdt, "hypervisor")));
478         _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
479         if (kvmppc_has_cap_fixup_hcalls()) {
480             /*
481              * Older KVM versions with older guest kernels were broken with the
482              * magic page, don't allow the guest to map it.
483              */
484             kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
485                                  sizeof(hypercall));
486             _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
487                               sizeof(hypercall))));
488         }
489         _FDT((fdt_end_node(fdt)));
490     }
491 
492     _FDT((fdt_end_node(fdt))); /* close root node */
493     _FDT((fdt_finish(fdt)));
494 
495     return fdt;
496 }
497 
498 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
499                                  target_ulong addr, target_ulong size)
500 {
501     void *fdt, *fdt_skel;
502     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
503 
504     size -= sizeof(hdr);
505 
506     /* Create sceleton */
507     fdt_skel = g_malloc0(size);
508     _FDT((fdt_create(fdt_skel, size)));
509     _FDT((fdt_begin_node(fdt_skel, "")));
510     _FDT((fdt_end_node(fdt_skel)));
511     _FDT((fdt_finish(fdt_skel)));
512     fdt = g_malloc0(size);
513     _FDT((fdt_open_into(fdt_skel, fdt, size)));
514     g_free(fdt_skel);
515 
516     /* Fix skeleton up */
517     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
518 
519     /* Pack resulting tree */
520     _FDT((fdt_pack(fdt)));
521 
522     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
523         trace_spapr_cas_failed(size);
524         return -1;
525     }
526 
527     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
528     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
529     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
530     g_free(fdt);
531 
532     return 0;
533 }
534 
535 static void spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
536                                        hwaddr size)
537 {
538     uint32_t associativity[] = {
539         cpu_to_be32(0x4), /* length */
540         cpu_to_be32(0x0), cpu_to_be32(0x0),
541         cpu_to_be32(0x0), cpu_to_be32(nodeid)
542     };
543     char mem_name[32];
544     uint64_t mem_reg_property[2];
545     int off;
546 
547     mem_reg_property[0] = cpu_to_be64(start);
548     mem_reg_property[1] = cpu_to_be64(size);
549 
550     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
551     off = fdt_add_subnode(fdt, 0, mem_name);
552     _FDT(off);
553     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
554     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
555                       sizeof(mem_reg_property))));
556     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
557                       sizeof(associativity))));
558 }
559 
560 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
561 {
562     MachineState *machine = MACHINE(spapr);
563     hwaddr mem_start, node_size;
564     int i, nb_nodes = nb_numa_nodes;
565     NodeInfo *nodes = numa_info;
566     NodeInfo ramnode;
567 
568     /* No NUMA nodes, assume there is just one node with whole RAM */
569     if (!nb_numa_nodes) {
570         nb_nodes = 1;
571         ramnode.node_mem = machine->ram_size;
572         nodes = &ramnode;
573     }
574 
575     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
576         if (!nodes[i].node_mem) {
577             continue;
578         }
579         if (mem_start >= machine->ram_size) {
580             node_size = 0;
581         } else {
582             node_size = nodes[i].node_mem;
583             if (node_size > machine->ram_size - mem_start) {
584                 node_size = machine->ram_size - mem_start;
585             }
586         }
587         if (!mem_start) {
588             /* ppc_spapr_init() checks for rma_size <= node0_size already */
589             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
590             mem_start += spapr->rma_size;
591             node_size -= spapr->rma_size;
592         }
593         for ( ; node_size; ) {
594             hwaddr sizetmp = pow2floor(node_size);
595 
596             /* mem_start != 0 here */
597             if (ctzl(mem_start) < ctzl(sizetmp)) {
598                 sizetmp = 1ULL << ctzl(mem_start);
599             }
600 
601             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
602             node_size -= sizetmp;
603             mem_start += sizetmp;
604         }
605     }
606 
607     return 0;
608 }
609 
610 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
611                                   sPAPRMachineState *spapr)
612 {
613     PowerPCCPU *cpu = POWERPC_CPU(cs);
614     CPUPPCState *env = &cpu->env;
615     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
616     int index = ppc_get_vcpu_dt_id(cpu);
617     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
618                        0xffffffff, 0xffffffff};
619     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
620     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
621     uint32_t page_sizes_prop[64];
622     size_t page_sizes_prop_size;
623     QemuOpts *opts = qemu_opts_find(qemu_find_opts("smp-opts"), NULL);
624     unsigned sockets = opts ? qemu_opt_get_number(opts, "sockets", 0) : 0;
625     uint32_t cpus_per_socket = sockets ? (smp_cpus / sockets) : 1;
626     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
627 
628     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
629     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
630 
631     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
632     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
633                            env->dcache_line_size)));
634     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
635                            env->dcache_line_size)));
636     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
637                            env->icache_line_size)));
638     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
639                            env->icache_line_size)));
640 
641     if (pcc->l1_dcache_size) {
642         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
643                                pcc->l1_dcache_size)));
644     } else {
645         fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
646     }
647     if (pcc->l1_icache_size) {
648         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
649                                pcc->l1_icache_size)));
650     } else {
651         fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
652     }
653 
654     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
655     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
656     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
657     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
658     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
659 
660     if (env->spr_cb[SPR_PURR].oea_read) {
661         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
662     }
663 
664     if (env->mmu_model & POWERPC_MMU_1TSEG) {
665         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
666                           segs, sizeof(segs))));
667     }
668 
669     /* Advertise VMX/VSX (vector extensions) if available
670      *   0 / no property == no vector extensions
671      *   1               == VMX / Altivec available
672      *   2               == VSX available */
673     if (env->insns_flags & PPC_ALTIVEC) {
674         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
675 
676         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
677     }
678 
679     /* Advertise DFP (Decimal Floating Point) if available
680      *   0 / no property == no DFP
681      *   1               == DFP available */
682     if (env->insns_flags2 & PPC2_DFP) {
683         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
684     }
685 
686     page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
687                                                   sizeof(page_sizes_prop));
688     if (page_sizes_prop_size) {
689         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
690                           page_sizes_prop, page_sizes_prop_size)));
691     }
692 
693     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
694                            cs->cpu_index / cpus_per_socket)));
695 
696     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
697                       pft_size_prop, sizeof(pft_size_prop))));
698 
699     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
700 
701     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
702                                 ppc_get_compat_smt_threads(cpu)));
703 }
704 
705 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
706 {
707     CPUState *cs;
708     int cpus_offset;
709     char *nodename;
710     int smt = kvmppc_smt_threads();
711 
712     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
713     _FDT(cpus_offset);
714     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
715     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
716 
717     /*
718      * We walk the CPUs in reverse order to ensure that CPU DT nodes
719      * created by fdt_add_subnode() end up in the right order in FDT
720      * for the guest kernel the enumerate the CPUs correctly.
721      */
722     CPU_FOREACH_REVERSE(cs) {
723         PowerPCCPU *cpu = POWERPC_CPU(cs);
724         int index = ppc_get_vcpu_dt_id(cpu);
725         DeviceClass *dc = DEVICE_GET_CLASS(cs);
726         int offset;
727 
728         if ((index % smt) != 0) {
729             continue;
730         }
731 
732         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
733         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
734         g_free(nodename);
735         _FDT(offset);
736         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
737     }
738 
739 }
740 
741 static void spapr_finalize_fdt(sPAPRMachineState *spapr,
742                                hwaddr fdt_addr,
743                                hwaddr rtas_addr,
744                                hwaddr rtas_size)
745 {
746     MachineState *machine = MACHINE(qdev_get_machine());
747     const char *boot_device = machine->boot_order;
748     int ret, i;
749     size_t cb = 0;
750     char *bootlist;
751     void *fdt;
752     sPAPRPHBState *phb;
753 
754     fdt = g_malloc(FDT_MAX_SIZE);
755 
756     /* open out the base tree into a temp buffer for the final tweaks */
757     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
758 
759     ret = spapr_populate_memory(spapr, fdt);
760     if (ret < 0) {
761         fprintf(stderr, "couldn't setup memory nodes in fdt\n");
762         exit(1);
763     }
764 
765     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
766     if (ret < 0) {
767         fprintf(stderr, "couldn't setup vio devices in fdt\n");
768         exit(1);
769     }
770 
771     QLIST_FOREACH(phb, &spapr->phbs, list) {
772         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
773     }
774 
775     if (ret < 0) {
776         fprintf(stderr, "couldn't setup PCI devices in fdt\n");
777         exit(1);
778     }
779 
780     /* RTAS */
781     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
782     if (ret < 0) {
783         fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
784     }
785 
786     /* cpus */
787     spapr_populate_cpus_dt_node(fdt, spapr);
788 
789     bootlist = get_boot_devices_list(&cb, true);
790     if (cb && bootlist) {
791         int offset = fdt_path_offset(fdt, "/chosen");
792         if (offset < 0) {
793             exit(1);
794         }
795         for (i = 0; i < cb; i++) {
796             if (bootlist[i] == '\n') {
797                 bootlist[i] = ' ';
798             }
799 
800         }
801         ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
802     }
803 
804     if (boot_device && strlen(boot_device)) {
805         int offset = fdt_path_offset(fdt, "/chosen");
806 
807         if (offset < 0) {
808             exit(1);
809         }
810         fdt_setprop_string(fdt, offset, "qemu,boot-device", boot_device);
811     }
812 
813     if (!spapr->has_graphics) {
814         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
815     }
816 
817     _FDT((fdt_pack(fdt)));
818 
819     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
820         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
821                      fdt_totalsize(fdt), FDT_MAX_SIZE);
822         exit(1);
823     }
824 
825     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
826 
827     g_free(bootlist);
828     g_free(fdt);
829 }
830 
831 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
832 {
833     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
834 }
835 
836 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
837 {
838     CPUPPCState *env = &cpu->env;
839 
840     if (msr_pr) {
841         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
842         env->gpr[3] = H_PRIVILEGE;
843     } else {
844         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
845     }
846 }
847 
848 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
849 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
850 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
851 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
852 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
853 
854 static void spapr_reset_htab(sPAPRMachineState *spapr)
855 {
856     long shift;
857     int index;
858 
859     /* allocate hash page table.  For now we always make this 16mb,
860      * later we should probably make it scale to the size of guest
861      * RAM */
862 
863     shift = kvmppc_reset_htab(spapr->htab_shift);
864 
865     if (shift > 0) {
866         /* Kernel handles htab, we don't need to allocate one */
867         spapr->htab_shift = shift;
868         kvmppc_kern_htab = true;
869 
870         /* Tell readers to update their file descriptor */
871         if (spapr->htab_fd >= 0) {
872             spapr->htab_fd_stale = true;
873         }
874     } else {
875         if (!spapr->htab) {
876             /* Allocate an htab if we don't yet have one */
877             spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
878         }
879 
880         /* And clear it */
881         memset(spapr->htab, 0, HTAB_SIZE(spapr));
882 
883         for (index = 0; index < HTAB_SIZE(spapr) / HASH_PTE_SIZE_64; index++) {
884             DIRTY_HPTE(HPTE(spapr->htab, index));
885         }
886     }
887 
888     /* Update the RMA size if necessary */
889     if (spapr->vrma_adjust) {
890         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
891                                           spapr->htab_shift);
892     }
893 }
894 
895 static int find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
896 {
897     bool matched = false;
898 
899     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
900         matched = true;
901     }
902 
903     if (!matched) {
904         error_report("Device %s is not supported by this machine yet.",
905                      qdev_fw_name(DEVICE(sbdev)));
906         exit(1);
907     }
908 
909     return 0;
910 }
911 
912 /*
913  * A guest reset will cause spapr->htab_fd to become stale if being used.
914  * Reopen the file descriptor to make sure the whole HTAB is properly read.
915  */
916 static int spapr_check_htab_fd(sPAPRMachineState *spapr)
917 {
918     int rc = 0;
919 
920     if (spapr->htab_fd_stale) {
921         close(spapr->htab_fd);
922         spapr->htab_fd = kvmppc_get_htab_fd(false);
923         if (spapr->htab_fd < 0) {
924             error_report("Unable to open fd for reading hash table from KVM: "
925                          "%s", strerror(errno));
926             rc = -1;
927         }
928         spapr->htab_fd_stale = false;
929     }
930 
931     return rc;
932 }
933 
934 static void ppc_spapr_reset(void)
935 {
936     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
937     PowerPCCPU *first_ppc_cpu;
938     uint32_t rtas_limit;
939 
940     /* Check for unknown sysbus devices */
941     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
942 
943     /* Reset the hash table & recalc the RMA */
944     spapr_reset_htab(spapr);
945 
946     qemu_devices_reset();
947 
948     /*
949      * We place the device tree and RTAS just below either the top of the RMA,
950      * or just below 2GB, whichever is lowere, so that it can be
951      * processed with 32-bit real mode code if necessary
952      */
953     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
954     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
955     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
956 
957     /* Load the fdt */
958     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
959                        spapr->rtas_size);
960 
961     /* Copy RTAS over */
962     cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
963                               spapr->rtas_size);
964 
965     /* Set up the entry state */
966     first_ppc_cpu = POWERPC_CPU(first_cpu);
967     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
968     first_ppc_cpu->env.gpr[5] = 0;
969     first_cpu->halted = 0;
970     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
971 
972 }
973 
974 static void spapr_cpu_reset(void *opaque)
975 {
976     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
977     PowerPCCPU *cpu = opaque;
978     CPUState *cs = CPU(cpu);
979     CPUPPCState *env = &cpu->env;
980 
981     cpu_reset(cs);
982 
983     /* All CPUs start halted.  CPU0 is unhalted from the machine level
984      * reset code and the rest are explicitly started up by the guest
985      * using an RTAS call */
986     cs->halted = 1;
987 
988     env->spr[SPR_HIOR] = 0;
989 
990     env->external_htab = (uint8_t *)spapr->htab;
991     if (kvm_enabled() && !env->external_htab) {
992         /*
993          * HV KVM, set external_htab to 1 so our ppc_hash64_load_hpte*
994          * functions do the right thing.
995          */
996         env->external_htab = (void *)1;
997     }
998     env->htab_base = -1;
999     /*
1000      * htab_mask is the mask used to normalize hash value to PTEG index.
1001      * htab_shift is log2 of hash table size.
1002      * We have 8 hpte per group, and each hpte is 16 bytes.
1003      * ie have 128 bytes per hpte entry.
1004      */
1005     env->htab_mask = (1ULL << (spapr->htab_shift - 7)) - 1;
1006     env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
1007         (spapr->htab_shift - 18);
1008 }
1009 
1010 static void spapr_create_nvram(sPAPRMachineState *spapr)
1011 {
1012     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1013     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1014 
1015     if (dinfo) {
1016         qdev_prop_set_drive_nofail(dev, "drive", blk_by_legacy_dinfo(dinfo));
1017     }
1018 
1019     qdev_init_nofail(dev);
1020 
1021     spapr->nvram = (struct sPAPRNVRAM *)dev;
1022 }
1023 
1024 static void spapr_rtc_create(sPAPRMachineState *spapr)
1025 {
1026     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1027 
1028     qdev_init_nofail(dev);
1029     spapr->rtc = dev;
1030 
1031     object_property_add_alias(qdev_get_machine(), "rtc-time",
1032                               OBJECT(spapr->rtc), "date", NULL);
1033 }
1034 
1035 /* Returns whether we want to use VGA or not */
1036 static int spapr_vga_init(PCIBus *pci_bus)
1037 {
1038     switch (vga_interface_type) {
1039     case VGA_NONE:
1040         return false;
1041     case VGA_DEVICE:
1042         return true;
1043     case VGA_STD:
1044         return pci_vga_init(pci_bus) != NULL;
1045     default:
1046         fprintf(stderr, "This vga model is not supported,"
1047                 "currently it only supports -vga std\n");
1048         exit(0);
1049     }
1050 }
1051 
1052 static int spapr_post_load(void *opaque, int version_id)
1053 {
1054     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1055     int err = 0;
1056 
1057     /* In earlier versions, there was no separate qdev for the PAPR
1058      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1059      * So when migrating from those versions, poke the incoming offset
1060      * value into the RTC device */
1061     if (version_id < 3) {
1062         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1063     }
1064 
1065     return err;
1066 }
1067 
1068 static bool version_before_3(void *opaque, int version_id)
1069 {
1070     return version_id < 3;
1071 }
1072 
1073 static const VMStateDescription vmstate_spapr = {
1074     .name = "spapr",
1075     .version_id = 3,
1076     .minimum_version_id = 1,
1077     .post_load = spapr_post_load,
1078     .fields = (VMStateField[]) {
1079         /* used to be @next_irq */
1080         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1081 
1082         /* RTC offset */
1083         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1084 
1085         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1086         VMSTATE_END_OF_LIST()
1087     },
1088 };
1089 
1090 static int htab_save_setup(QEMUFile *f, void *opaque)
1091 {
1092     sPAPRMachineState *spapr = opaque;
1093 
1094     /* "Iteration" header */
1095     qemu_put_be32(f, spapr->htab_shift);
1096 
1097     if (spapr->htab) {
1098         spapr->htab_save_index = 0;
1099         spapr->htab_first_pass = true;
1100     } else {
1101         assert(kvm_enabled());
1102 
1103         spapr->htab_fd = kvmppc_get_htab_fd(false);
1104         spapr->htab_fd_stale = false;
1105         if (spapr->htab_fd < 0) {
1106             fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
1107                     strerror(errno));
1108             return -1;
1109         }
1110     }
1111 
1112 
1113     return 0;
1114 }
1115 
1116 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1117                                  int64_t max_ns)
1118 {
1119     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1120     int index = spapr->htab_save_index;
1121     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1122 
1123     assert(spapr->htab_first_pass);
1124 
1125     do {
1126         int chunkstart;
1127 
1128         /* Consume invalid HPTEs */
1129         while ((index < htabslots)
1130                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1131             index++;
1132             CLEAN_HPTE(HPTE(spapr->htab, index));
1133         }
1134 
1135         /* Consume valid HPTEs */
1136         chunkstart = index;
1137         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1138                && HPTE_VALID(HPTE(spapr->htab, index))) {
1139             index++;
1140             CLEAN_HPTE(HPTE(spapr->htab, index));
1141         }
1142 
1143         if (index > chunkstart) {
1144             int n_valid = index - chunkstart;
1145 
1146             qemu_put_be32(f, chunkstart);
1147             qemu_put_be16(f, n_valid);
1148             qemu_put_be16(f, 0);
1149             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1150                             HASH_PTE_SIZE_64 * n_valid);
1151 
1152             if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1153                 break;
1154             }
1155         }
1156     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1157 
1158     if (index >= htabslots) {
1159         assert(index == htabslots);
1160         index = 0;
1161         spapr->htab_first_pass = false;
1162     }
1163     spapr->htab_save_index = index;
1164 }
1165 
1166 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1167                                 int64_t max_ns)
1168 {
1169     bool final = max_ns < 0;
1170     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1171     int examined = 0, sent = 0;
1172     int index = spapr->htab_save_index;
1173     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1174 
1175     assert(!spapr->htab_first_pass);
1176 
1177     do {
1178         int chunkstart, invalidstart;
1179 
1180         /* Consume non-dirty HPTEs */
1181         while ((index < htabslots)
1182                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1183             index++;
1184             examined++;
1185         }
1186 
1187         chunkstart = index;
1188         /* Consume valid dirty HPTEs */
1189         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1190                && HPTE_DIRTY(HPTE(spapr->htab, index))
1191                && HPTE_VALID(HPTE(spapr->htab, index))) {
1192             CLEAN_HPTE(HPTE(spapr->htab, index));
1193             index++;
1194             examined++;
1195         }
1196 
1197         invalidstart = index;
1198         /* Consume invalid dirty HPTEs */
1199         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1200                && HPTE_DIRTY(HPTE(spapr->htab, index))
1201                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1202             CLEAN_HPTE(HPTE(spapr->htab, index));
1203             index++;
1204             examined++;
1205         }
1206 
1207         if (index > chunkstart) {
1208             int n_valid = invalidstart - chunkstart;
1209             int n_invalid = index - invalidstart;
1210 
1211             qemu_put_be32(f, chunkstart);
1212             qemu_put_be16(f, n_valid);
1213             qemu_put_be16(f, n_invalid);
1214             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1215                             HASH_PTE_SIZE_64 * n_valid);
1216             sent += index - chunkstart;
1217 
1218             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1219                 break;
1220             }
1221         }
1222 
1223         if (examined >= htabslots) {
1224             break;
1225         }
1226 
1227         if (index >= htabslots) {
1228             assert(index == htabslots);
1229             index = 0;
1230         }
1231     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1232 
1233     if (index >= htabslots) {
1234         assert(index == htabslots);
1235         index = 0;
1236     }
1237 
1238     spapr->htab_save_index = index;
1239 
1240     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1241 }
1242 
1243 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1244 #define MAX_KVM_BUF_SIZE    2048
1245 
1246 static int htab_save_iterate(QEMUFile *f, void *opaque)
1247 {
1248     sPAPRMachineState *spapr = opaque;
1249     int rc = 0;
1250 
1251     /* Iteration header */
1252     qemu_put_be32(f, 0);
1253 
1254     if (!spapr->htab) {
1255         assert(kvm_enabled());
1256 
1257         rc = spapr_check_htab_fd(spapr);
1258         if (rc < 0) {
1259             return rc;
1260         }
1261 
1262         rc = kvmppc_save_htab(f, spapr->htab_fd,
1263                               MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1264         if (rc < 0) {
1265             return rc;
1266         }
1267     } else  if (spapr->htab_first_pass) {
1268         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1269     } else {
1270         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1271     }
1272 
1273     /* End marker */
1274     qemu_put_be32(f, 0);
1275     qemu_put_be16(f, 0);
1276     qemu_put_be16(f, 0);
1277 
1278     return rc;
1279 }
1280 
1281 static int htab_save_complete(QEMUFile *f, void *opaque)
1282 {
1283     sPAPRMachineState *spapr = opaque;
1284 
1285     /* Iteration header */
1286     qemu_put_be32(f, 0);
1287 
1288     if (!spapr->htab) {
1289         int rc;
1290 
1291         assert(kvm_enabled());
1292 
1293         rc = spapr_check_htab_fd(spapr);
1294         if (rc < 0) {
1295             return rc;
1296         }
1297 
1298         rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
1299         if (rc < 0) {
1300             return rc;
1301         }
1302         close(spapr->htab_fd);
1303         spapr->htab_fd = -1;
1304     } else {
1305         htab_save_later_pass(f, spapr, -1);
1306     }
1307 
1308     /* End marker */
1309     qemu_put_be32(f, 0);
1310     qemu_put_be16(f, 0);
1311     qemu_put_be16(f, 0);
1312 
1313     return 0;
1314 }
1315 
1316 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1317 {
1318     sPAPRMachineState *spapr = opaque;
1319     uint32_t section_hdr;
1320     int fd = -1;
1321 
1322     if (version_id < 1 || version_id > 1) {
1323         fprintf(stderr, "htab_load() bad version\n");
1324         return -EINVAL;
1325     }
1326 
1327     section_hdr = qemu_get_be32(f);
1328 
1329     if (section_hdr) {
1330         /* First section, just the hash shift */
1331         if (spapr->htab_shift != section_hdr) {
1332             return -EINVAL;
1333         }
1334         return 0;
1335     }
1336 
1337     if (!spapr->htab) {
1338         assert(kvm_enabled());
1339 
1340         fd = kvmppc_get_htab_fd(true);
1341         if (fd < 0) {
1342             fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1343                     strerror(errno));
1344         }
1345     }
1346 
1347     while (true) {
1348         uint32_t index;
1349         uint16_t n_valid, n_invalid;
1350 
1351         index = qemu_get_be32(f);
1352         n_valid = qemu_get_be16(f);
1353         n_invalid = qemu_get_be16(f);
1354 
1355         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1356             /* End of Stream */
1357             break;
1358         }
1359 
1360         if ((index + n_valid + n_invalid) >
1361             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1362             /* Bad index in stream */
1363             fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1364                     "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1365                     spapr->htab_shift);
1366             return -EINVAL;
1367         }
1368 
1369         if (spapr->htab) {
1370             if (n_valid) {
1371                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1372                                 HASH_PTE_SIZE_64 * n_valid);
1373             }
1374             if (n_invalid) {
1375                 memset(HPTE(spapr->htab, index + n_valid), 0,
1376                        HASH_PTE_SIZE_64 * n_invalid);
1377             }
1378         } else {
1379             int rc;
1380 
1381             assert(fd >= 0);
1382 
1383             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1384             if (rc < 0) {
1385                 return rc;
1386             }
1387         }
1388     }
1389 
1390     if (!spapr->htab) {
1391         assert(fd >= 0);
1392         close(fd);
1393     }
1394 
1395     return 0;
1396 }
1397 
1398 static SaveVMHandlers savevm_htab_handlers = {
1399     .save_live_setup = htab_save_setup,
1400     .save_live_iterate = htab_save_iterate,
1401     .save_live_complete = htab_save_complete,
1402     .load_state = htab_load,
1403 };
1404 
1405 static void spapr_boot_set(void *opaque, const char *boot_device,
1406                            Error **errp)
1407 {
1408     MachineState *machine = MACHINE(qdev_get_machine());
1409     machine->boot_order = g_strdup(boot_device);
1410 }
1411 
1412 static void spapr_cpu_init(sPAPRMachineState *spapr, PowerPCCPU *cpu)
1413 {
1414     CPUPPCState *env = &cpu->env;
1415 
1416     /* Set time-base frequency to 512 MHz */
1417     cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1418 
1419     /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1420      * MSR[IP] should never be set.
1421      */
1422     env->msr_mask &= ~(1 << 6);
1423 
1424     /* Tell KVM that we're in PAPR mode */
1425     if (kvm_enabled()) {
1426         kvmppc_set_papr(cpu);
1427     }
1428 
1429     if (cpu->max_compat) {
1430         if (ppc_set_compat(cpu, cpu->max_compat) < 0) {
1431             exit(1);
1432         }
1433     }
1434 
1435     xics_cpu_setup(spapr->icp, cpu);
1436 
1437     qemu_register_reset(spapr_cpu_reset, cpu);
1438 }
1439 
1440 /* pSeries LPAR / sPAPR hardware init */
1441 static void ppc_spapr_init(MachineState *machine)
1442 {
1443     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1444     const char *kernel_filename = machine->kernel_filename;
1445     const char *kernel_cmdline = machine->kernel_cmdline;
1446     const char *initrd_filename = machine->initrd_filename;
1447     PowerPCCPU *cpu;
1448     PCIHostState *phb;
1449     int i;
1450     MemoryRegion *sysmem = get_system_memory();
1451     MemoryRegion *ram = g_new(MemoryRegion, 1);
1452     MemoryRegion *rma_region;
1453     void *rma = NULL;
1454     hwaddr rma_alloc_size;
1455     hwaddr node0_size = spapr_node0_size();
1456     uint32_t initrd_base = 0;
1457     long kernel_size = 0, initrd_size = 0;
1458     long load_limit, fw_size;
1459     bool kernel_le = false;
1460     char *filename;
1461 
1462     msi_supported = true;
1463 
1464     QLIST_INIT(&spapr->phbs);
1465 
1466     cpu_ppc_hypercall = emulate_spapr_hypercall;
1467 
1468     /* Allocate RMA if necessary */
1469     rma_alloc_size = kvmppc_alloc_rma(&rma);
1470 
1471     if (rma_alloc_size == -1) {
1472         error_report("Unable to create RMA");
1473         exit(1);
1474     }
1475 
1476     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1477         spapr->rma_size = rma_alloc_size;
1478     } else {
1479         spapr->rma_size = node0_size;
1480 
1481         /* With KVM, we don't actually know whether KVM supports an
1482          * unbounded RMA (PR KVM) or is limited by the hash table size
1483          * (HV KVM using VRMA), so we always assume the latter
1484          *
1485          * In that case, we also limit the initial allocations for RTAS
1486          * etc... to 256M since we have no way to know what the VRMA size
1487          * is going to be as it depends on the size of the hash table
1488          * isn't determined yet.
1489          */
1490         if (kvm_enabled()) {
1491             spapr->vrma_adjust = 1;
1492             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1493         }
1494     }
1495 
1496     if (spapr->rma_size > node0_size) {
1497         fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1498                 spapr->rma_size);
1499         exit(1);
1500     }
1501 
1502     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1503     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1504 
1505     /* We aim for a hash table of size 1/128 the size of RAM.  The
1506      * normal rule of thumb is 1/64 the size of RAM, but that's much
1507      * more than needed for the Linux guests we support. */
1508     spapr->htab_shift = 18; /* Minimum architected size */
1509     while (spapr->htab_shift <= 46) {
1510         if ((1ULL << (spapr->htab_shift + 7)) >= machine->ram_size) {
1511             break;
1512         }
1513         spapr->htab_shift++;
1514     }
1515 
1516     /* Set up Interrupt Controller before we create the VCPUs */
1517     spapr->icp = xics_system_init(machine,
1518                                   DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(),
1519                                                smp_threads),
1520                                   XICS_IRQS);
1521 
1522     /* init CPUs */
1523     if (machine->cpu_model == NULL) {
1524         machine->cpu_model = kvm_enabled() ? "host" : "POWER7";
1525     }
1526     for (i = 0; i < smp_cpus; i++) {
1527         cpu = cpu_ppc_init(machine->cpu_model);
1528         if (cpu == NULL) {
1529             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1530             exit(1);
1531         }
1532         spapr_cpu_init(spapr, cpu);
1533     }
1534 
1535     if (kvm_enabled()) {
1536         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1537         kvmppc_enable_logical_ci_hcalls();
1538     }
1539 
1540     /* allocate RAM */
1541     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1542                                          machine->ram_size);
1543     memory_region_add_subregion(sysmem, 0, ram);
1544 
1545     if (rma_alloc_size && rma) {
1546         rma_region = g_new(MemoryRegion, 1);
1547         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1548                                    rma_alloc_size, rma);
1549         vmstate_register_ram_global(rma_region);
1550         memory_region_add_subregion(sysmem, 0, rma_region);
1551     }
1552 
1553     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1554     if (!filename) {
1555         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1556         exit(1);
1557     }
1558     spapr->rtas_size = get_image_size(filename);
1559     spapr->rtas_blob = g_malloc(spapr->rtas_size);
1560     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1561         error_report("Could not load LPAR rtas '%s'", filename);
1562         exit(1);
1563     }
1564     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1565         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1566                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1567         exit(1);
1568     }
1569     g_free(filename);
1570 
1571     /* Set up EPOW events infrastructure */
1572     spapr_events_init(spapr);
1573 
1574     /* Set up the RTC RTAS interfaces */
1575     spapr_rtc_create(spapr);
1576 
1577     /* Set up VIO bus */
1578     spapr->vio_bus = spapr_vio_bus_init();
1579 
1580     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1581         if (serial_hds[i]) {
1582             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1583         }
1584     }
1585 
1586     /* We always have at least the nvram device on VIO */
1587     spapr_create_nvram(spapr);
1588 
1589     /* Set up PCI */
1590     spapr_pci_rtas_init();
1591 
1592     phb = spapr_create_phb(spapr, 0);
1593 
1594     for (i = 0; i < nb_nics; i++) {
1595         NICInfo *nd = &nd_table[i];
1596 
1597         if (!nd->model) {
1598             nd->model = g_strdup("ibmveth");
1599         }
1600 
1601         if (strcmp(nd->model, "ibmveth") == 0) {
1602             spapr_vlan_create(spapr->vio_bus, nd);
1603         } else {
1604             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1605         }
1606     }
1607 
1608     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1609         spapr_vscsi_create(spapr->vio_bus);
1610     }
1611 
1612     /* Graphics */
1613     if (spapr_vga_init(phb->bus)) {
1614         spapr->has_graphics = true;
1615         machine->usb |= defaults_enabled() && !machine->usb_disabled;
1616     }
1617 
1618     if (machine->usb) {
1619         pci_create_simple(phb->bus, -1, "pci-ohci");
1620 
1621         if (spapr->has_graphics) {
1622             USBBus *usb_bus = usb_bus_find(-1);
1623 
1624             usb_create_simple(usb_bus, "usb-kbd");
1625             usb_create_simple(usb_bus, "usb-mouse");
1626         }
1627     }
1628 
1629     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1630         fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1631                 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1632         exit(1);
1633     }
1634 
1635     if (kernel_filename) {
1636         uint64_t lowaddr = 0;
1637 
1638         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1639                                NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1640         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1641             kernel_size = load_elf(kernel_filename,
1642                                    translate_kernel_address, NULL,
1643                                    NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1644             kernel_le = kernel_size > 0;
1645         }
1646         if (kernel_size < 0) {
1647             fprintf(stderr, "qemu: error loading %s: %s\n",
1648                     kernel_filename, load_elf_strerror(kernel_size));
1649             exit(1);
1650         }
1651 
1652         /* load initrd */
1653         if (initrd_filename) {
1654             /* Try to locate the initrd in the gap between the kernel
1655              * and the firmware. Add a bit of space just in case
1656              */
1657             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1658             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1659                                               load_limit - initrd_base);
1660             if (initrd_size < 0) {
1661                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1662                         initrd_filename);
1663                 exit(1);
1664             }
1665         } else {
1666             initrd_base = 0;
1667             initrd_size = 0;
1668         }
1669     }
1670 
1671     if (bios_name == NULL) {
1672         bios_name = FW_FILE_NAME;
1673     }
1674     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1675     if (!filename) {
1676         error_report("Could not find LPAR firmware '%s'", bios_name);
1677         exit(1);
1678     }
1679     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1680     if (fw_size <= 0) {
1681         error_report("Could not load LPAR firmware '%s'", filename);
1682         exit(1);
1683     }
1684     g_free(filename);
1685 
1686     /* FIXME: Should register things through the MachineState's qdev
1687      * interface, this is a legacy from the sPAPREnvironment structure
1688      * which predated MachineState but had a similar function */
1689     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1690     register_savevm_live(NULL, "spapr/htab", -1, 1,
1691                          &savevm_htab_handlers, spapr);
1692 
1693     /* Prepare the device tree */
1694     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1695                                             kernel_size, kernel_le,
1696                                             kernel_cmdline,
1697                                             spapr->check_exception_irq);
1698     assert(spapr->fdt_skel != NULL);
1699 
1700     /* used by RTAS */
1701     QTAILQ_INIT(&spapr->ccs_list);
1702     qemu_register_reset(spapr_ccs_reset_hook, spapr);
1703 
1704     qemu_register_boot_set(spapr_boot_set, spapr);
1705 }
1706 
1707 static int spapr_kvm_type(const char *vm_type)
1708 {
1709     if (!vm_type) {
1710         return 0;
1711     }
1712 
1713     if (!strcmp(vm_type, "HV")) {
1714         return 1;
1715     }
1716 
1717     if (!strcmp(vm_type, "PR")) {
1718         return 2;
1719     }
1720 
1721     error_report("Unknown kvm-type specified '%s'", vm_type);
1722     exit(1);
1723 }
1724 
1725 /*
1726  * Implementation of an interface to adjust firmware path
1727  * for the bootindex property handling.
1728  */
1729 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
1730                                    DeviceState *dev)
1731 {
1732 #define CAST(type, obj, name) \
1733     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
1734     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
1735     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
1736 
1737     if (d) {
1738         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
1739         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
1740         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
1741 
1742         if (spapr) {
1743             /*
1744              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
1745              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
1746              * in the top 16 bits of the 64-bit LUN
1747              */
1748             unsigned id = 0x8000 | (d->id << 8) | d->lun;
1749             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1750                                    (uint64_t)id << 48);
1751         } else if (virtio) {
1752             /*
1753              * We use SRP luns of the form 01000000 | (target << 8) | lun
1754              * in the top 32 bits of the 64-bit LUN
1755              * Note: the quote above is from SLOF and it is wrong,
1756              * the actual binding is:
1757              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
1758              */
1759             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
1760             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1761                                    (uint64_t)id << 32);
1762         } else if (usb) {
1763             /*
1764              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
1765              * in the top 32 bits of the 64-bit LUN
1766              */
1767             unsigned usb_port = atoi(usb->port->path);
1768             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
1769             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1770                                    (uint64_t)id << 32);
1771         }
1772     }
1773 
1774     if (phb) {
1775         /* Replace "pci" with "pci@800000020000000" */
1776         return g_strdup_printf("pci@%"PRIX64, phb->buid);
1777     }
1778 
1779     return NULL;
1780 }
1781 
1782 static char *spapr_get_kvm_type(Object *obj, Error **errp)
1783 {
1784     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
1785 
1786     return g_strdup(spapr->kvm_type);
1787 }
1788 
1789 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
1790 {
1791     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
1792 
1793     g_free(spapr->kvm_type);
1794     spapr->kvm_type = g_strdup(value);
1795 }
1796 
1797 static void spapr_machine_initfn(Object *obj)
1798 {
1799     object_property_add_str(obj, "kvm-type",
1800                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
1801     object_property_set_description(obj, "kvm-type",
1802                                     "Specifies the KVM virtualization mode (HV, PR)",
1803                                     NULL);
1804 }
1805 
1806 static void ppc_cpu_do_nmi_on_cpu(void *arg)
1807 {
1808     CPUState *cs = arg;
1809 
1810     cpu_synchronize_state(cs);
1811     ppc_cpu_do_system_reset(cs);
1812 }
1813 
1814 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
1815 {
1816     CPUState *cs;
1817 
1818     CPU_FOREACH(cs) {
1819         async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, cs);
1820     }
1821 }
1822 
1823 static void spapr_machine_class_init(ObjectClass *oc, void *data)
1824 {
1825     MachineClass *mc = MACHINE_CLASS(oc);
1826     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
1827     NMIClass *nc = NMI_CLASS(oc);
1828 
1829     mc->init = ppc_spapr_init;
1830     mc->reset = ppc_spapr_reset;
1831     mc->block_default_type = IF_SCSI;
1832     mc->max_cpus = MAX_CPUS;
1833     mc->no_parallel = 1;
1834     mc->default_boot_order = "";
1835     mc->default_ram_size = 512 * M_BYTE;
1836     mc->kvm_type = spapr_kvm_type;
1837     mc->has_dynamic_sysbus = true;
1838     mc->pci_allow_0_address = true;
1839 
1840     fwc->get_dev_path = spapr_get_fw_dev_path;
1841     nc->nmi_monitor_handler = spapr_nmi;
1842 }
1843 
1844 static const TypeInfo spapr_machine_info = {
1845     .name          = TYPE_SPAPR_MACHINE,
1846     .parent        = TYPE_MACHINE,
1847     .abstract      = true,
1848     .instance_size = sizeof(sPAPRMachineState),
1849     .instance_init = spapr_machine_initfn,
1850     .class_size    = sizeof(sPAPRMachineClass),
1851     .class_init    = spapr_machine_class_init,
1852     .interfaces = (InterfaceInfo[]) {
1853         { TYPE_FW_PATH_PROVIDER },
1854         { TYPE_NMI },
1855         { }
1856     },
1857 };
1858 
1859 #define SPAPR_COMPAT_2_3 \
1860         HW_COMPAT_2_3 \
1861         {\
1862             .driver   = "spapr-pci-host-bridge",\
1863             .property = "dynamic-reconfiguration",\
1864             .value    = "off",\
1865         },
1866 
1867 #define SPAPR_COMPAT_2_2 \
1868         SPAPR_COMPAT_2_3 \
1869         HW_COMPAT_2_2 \
1870         {\
1871             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
1872             .property = "mem_win_size",\
1873             .value    = "0x20000000",\
1874         },
1875 
1876 #define SPAPR_COMPAT_2_1 \
1877         SPAPR_COMPAT_2_2 \
1878         HW_COMPAT_2_1
1879 
1880 static void spapr_compat_2_3(Object *obj)
1881 {
1882     savevm_skip_section_footers();
1883     global_state_set_optional();
1884 }
1885 
1886 static void spapr_compat_2_2(Object *obj)
1887 {
1888     spapr_compat_2_3(obj);
1889 }
1890 
1891 static void spapr_compat_2_1(Object *obj)
1892 {
1893     spapr_compat_2_2(obj);
1894 }
1895 
1896 static void spapr_machine_2_3_instance_init(Object *obj)
1897 {
1898     spapr_compat_2_3(obj);
1899     spapr_machine_initfn(obj);
1900 }
1901 
1902 static void spapr_machine_2_2_instance_init(Object *obj)
1903 {
1904     spapr_compat_2_2(obj);
1905     spapr_machine_initfn(obj);
1906 }
1907 
1908 static void spapr_machine_2_1_instance_init(Object *obj)
1909 {
1910     spapr_compat_2_1(obj);
1911     spapr_machine_initfn(obj);
1912 }
1913 
1914 static void spapr_machine_2_1_class_init(ObjectClass *oc, void *data)
1915 {
1916     MachineClass *mc = MACHINE_CLASS(oc);
1917     static GlobalProperty compat_props[] = {
1918         SPAPR_COMPAT_2_1
1919         { /* end of list */ }
1920     };
1921 
1922     mc->name = "pseries-2.1";
1923     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.1";
1924     mc->compat_props = compat_props;
1925 }
1926 
1927 static const TypeInfo spapr_machine_2_1_info = {
1928     .name          = TYPE_SPAPR_MACHINE "2.1",
1929     .parent        = TYPE_SPAPR_MACHINE,
1930     .class_init    = spapr_machine_2_1_class_init,
1931     .instance_init = spapr_machine_2_1_instance_init,
1932 };
1933 
1934 static void spapr_machine_2_2_class_init(ObjectClass *oc, void *data)
1935 {
1936     static GlobalProperty compat_props[] = {
1937         SPAPR_COMPAT_2_2
1938         { /* end of list */ }
1939     };
1940     MachineClass *mc = MACHINE_CLASS(oc);
1941 
1942     mc->name = "pseries-2.2";
1943     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.2";
1944     mc->compat_props = compat_props;
1945 }
1946 
1947 static const TypeInfo spapr_machine_2_2_info = {
1948     .name          = TYPE_SPAPR_MACHINE "2.2",
1949     .parent        = TYPE_SPAPR_MACHINE,
1950     .class_init    = spapr_machine_2_2_class_init,
1951     .instance_init = spapr_machine_2_2_instance_init,
1952 };
1953 
1954 static void spapr_machine_2_3_class_init(ObjectClass *oc, void *data)
1955 {
1956     static GlobalProperty compat_props[] = {
1957         SPAPR_COMPAT_2_3
1958         { /* end of list */ }
1959     };
1960     MachineClass *mc = MACHINE_CLASS(oc);
1961 
1962     mc->name = "pseries-2.3";
1963     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.3";
1964     mc->compat_props = compat_props;
1965 }
1966 
1967 static const TypeInfo spapr_machine_2_3_info = {
1968     .name          = TYPE_SPAPR_MACHINE "2.3",
1969     .parent        = TYPE_SPAPR_MACHINE,
1970     .class_init    = spapr_machine_2_3_class_init,
1971     .instance_init = spapr_machine_2_3_instance_init,
1972 };
1973 
1974 static void spapr_machine_2_4_class_init(ObjectClass *oc, void *data)
1975 {
1976     MachineClass *mc = MACHINE_CLASS(oc);
1977 
1978     mc->name = "pseries-2.4";
1979     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.4";
1980     mc->alias = "pseries";
1981     mc->is_default = 1;
1982 }
1983 
1984 static const TypeInfo spapr_machine_2_4_info = {
1985     .name          = TYPE_SPAPR_MACHINE "2.4",
1986     .parent        = TYPE_SPAPR_MACHINE,
1987     .class_init    = spapr_machine_2_4_class_init,
1988 };
1989 
1990 static void spapr_machine_register_types(void)
1991 {
1992     type_register_static(&spapr_machine_info);
1993     type_register_static(&spapr_machine_2_1_info);
1994     type_register_static(&spapr_machine_2_2_info);
1995     type_register_static(&spapr_machine_2_3_info);
1996     type_register_static(&spapr_machine_2_4_info);
1997 }
1998 
1999 type_init(spapr_machine_register_types)
2000