xref: /openbmc/qemu/hw/ppc/spapr.c (revision 6d0ceb80)
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 "sysemu/device_tree.h"
41 #include "kvm_ppc.h"
42 #include "migration/migration.h"
43 #include "mmu-hash64.h"
44 #include "qom/cpu.h"
45 
46 #include "hw/boards.h"
47 #include "hw/ppc/ppc.h"
48 #include "hw/loader.h"
49 
50 #include "hw/ppc/fdt.h"
51 #include "hw/ppc/spapr.h"
52 #include "hw/ppc/spapr_vio.h"
53 #include "hw/pci-host/spapr.h"
54 #include "hw/ppc/xics.h"
55 #include "hw/pci/msi.h"
56 
57 #include "hw/pci/pci.h"
58 #include "hw/scsi/scsi.h"
59 #include "hw/virtio/virtio-scsi.h"
60 
61 #include "exec/address-spaces.h"
62 #include "hw/usb.h"
63 #include "qemu/config-file.h"
64 #include "qemu/error-report.h"
65 #include "trace.h"
66 #include "hw/nmi.h"
67 
68 #include "hw/compat.h"
69 #include "qemu/cutils.h"
70 #include "hw/ppc/spapr_cpu_core.h"
71 #include "qmp-commands.h"
72 
73 #include <libfdt.h>
74 
75 /* SLOF memory layout:
76  *
77  * SLOF raw image loaded at 0, copies its romfs right below the flat
78  * device-tree, then position SLOF itself 31M below that
79  *
80  * So we set FW_OVERHEAD to 40MB which should account for all of that
81  * and more
82  *
83  * We load our kernel at 4M, leaving space for SLOF initial image
84  */
85 #define FDT_MAX_SIZE            0x100000
86 #define RTAS_MAX_SIZE           0x10000
87 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
88 #define FW_MAX_SIZE             0x400000
89 #define FW_FILE_NAME            "slof.bin"
90 #define FW_OVERHEAD             0x2800000
91 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
92 
93 #define MIN_RMA_SLOF            128UL
94 
95 #define PHANDLE_XICP            0x00001111
96 
97 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
98 
99 static XICSState *try_create_xics(const char *type, int nr_servers,
100                                   int nr_irqs, Error **errp)
101 {
102     Error *err = NULL;
103     DeviceState *dev;
104 
105     dev = qdev_create(NULL, type);
106     qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
107     qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
108     object_property_set_bool(OBJECT(dev), true, "realized", &err);
109     if (err) {
110         error_propagate(errp, err);
111         object_unparent(OBJECT(dev));
112         return NULL;
113     }
114     return XICS_COMMON(dev);
115 }
116 
117 static XICSState *xics_system_init(MachineState *machine,
118                                    int nr_servers, int nr_irqs, Error **errp)
119 {
120     XICSState *xics = NULL;
121 
122     if (kvm_enabled()) {
123         Error *err = NULL;
124 
125         if (machine_kernel_irqchip_allowed(machine)) {
126             xics = try_create_xics(TYPE_XICS_SPAPR_KVM, nr_servers, nr_irqs,
127                                    &err);
128         }
129         if (machine_kernel_irqchip_required(machine) && !xics) {
130             error_reportf_err(err,
131                               "kernel_irqchip requested but unavailable: ");
132         } else {
133             error_free(err);
134         }
135     }
136 
137     if (!xics) {
138         xics = try_create_xics(TYPE_XICS_SPAPR, nr_servers, nr_irqs, errp);
139     }
140 
141     return xics;
142 }
143 
144 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
145                                   int smt_threads)
146 {
147     int i, ret = 0;
148     uint32_t servers_prop[smt_threads];
149     uint32_t gservers_prop[smt_threads * 2];
150     int index = ppc_get_vcpu_dt_id(cpu);
151 
152     if (cpu->cpu_version) {
153         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
154         if (ret < 0) {
155             return ret;
156         }
157     }
158 
159     /* Build interrupt servers and gservers properties */
160     for (i = 0; i < smt_threads; i++) {
161         servers_prop[i] = cpu_to_be32(index + i);
162         /* Hack, direct the group queues back to cpu 0 */
163         gservers_prop[i*2] = cpu_to_be32(index + i);
164         gservers_prop[i*2 + 1] = 0;
165     }
166     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
167                       servers_prop, sizeof(servers_prop));
168     if (ret < 0) {
169         return ret;
170     }
171     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
172                       gservers_prop, sizeof(gservers_prop));
173 
174     return ret;
175 }
176 
177 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
178 {
179     int ret = 0;
180     PowerPCCPU *cpu = POWERPC_CPU(cs);
181     int index = ppc_get_vcpu_dt_id(cpu);
182     uint32_t associativity[] = {cpu_to_be32(0x5),
183                                 cpu_to_be32(0x0),
184                                 cpu_to_be32(0x0),
185                                 cpu_to_be32(0x0),
186                                 cpu_to_be32(cs->numa_node),
187                                 cpu_to_be32(index)};
188 
189     /* Advertise NUMA via ibm,associativity */
190     if (nb_numa_nodes > 1) {
191         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
192                           sizeof(associativity));
193     }
194 
195     return ret;
196 }
197 
198 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
199 {
200     int ret = 0, offset, cpus_offset;
201     CPUState *cs;
202     char cpu_model[32];
203     int smt = kvmppc_smt_threads();
204     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
205 
206     CPU_FOREACH(cs) {
207         PowerPCCPU *cpu = POWERPC_CPU(cs);
208         DeviceClass *dc = DEVICE_GET_CLASS(cs);
209         int index = ppc_get_vcpu_dt_id(cpu);
210 
211         if ((index % smt) != 0) {
212             continue;
213         }
214 
215         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
216 
217         cpus_offset = fdt_path_offset(fdt, "/cpus");
218         if (cpus_offset < 0) {
219             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
220                                           "cpus");
221             if (cpus_offset < 0) {
222                 return cpus_offset;
223             }
224         }
225         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
226         if (offset < 0) {
227             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
228             if (offset < 0) {
229                 return offset;
230             }
231         }
232 
233         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
234                           pft_size_prop, sizeof(pft_size_prop));
235         if (ret < 0) {
236             return ret;
237         }
238 
239         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
240         if (ret < 0) {
241             return ret;
242         }
243 
244         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
245                                      ppc_get_compat_smt_threads(cpu));
246         if (ret < 0) {
247             return ret;
248         }
249     }
250     return ret;
251 }
252 
253 static hwaddr spapr_node0_size(void)
254 {
255     MachineState *machine = MACHINE(qdev_get_machine());
256 
257     if (nb_numa_nodes) {
258         int i;
259         for (i = 0; i < nb_numa_nodes; ++i) {
260             if (numa_info[i].node_mem) {
261                 return MIN(pow2floor(numa_info[i].node_mem),
262                            machine->ram_size);
263             }
264         }
265     }
266     return machine->ram_size;
267 }
268 
269 static void add_str(GString *s, const gchar *s1)
270 {
271     g_string_append_len(s, s1, strlen(s1) + 1);
272 }
273 
274 static void *spapr_create_fdt_skel(hwaddr initrd_base,
275                                    hwaddr initrd_size,
276                                    hwaddr kernel_size,
277                                    bool little_endian,
278                                    const char *kernel_cmdline,
279                                    uint32_t epow_irq)
280 {
281     void *fdt;
282     uint32_t start_prop = cpu_to_be32(initrd_base);
283     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
284     GString *hypertas = g_string_sized_new(256);
285     GString *qemu_hypertas = g_string_sized_new(256);
286     uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
287     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(max_cpus)};
288     unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
289     char *buf;
290 
291     add_str(hypertas, "hcall-pft");
292     add_str(hypertas, "hcall-term");
293     add_str(hypertas, "hcall-dabr");
294     add_str(hypertas, "hcall-interrupt");
295     add_str(hypertas, "hcall-tce");
296     add_str(hypertas, "hcall-vio");
297     add_str(hypertas, "hcall-splpar");
298     add_str(hypertas, "hcall-bulk");
299     add_str(hypertas, "hcall-set-mode");
300     add_str(hypertas, "hcall-sprg0");
301     add_str(hypertas, "hcall-copy");
302     add_str(hypertas, "hcall-debug");
303     add_str(qemu_hypertas, "hcall-memop1");
304 
305     fdt = g_malloc0(FDT_MAX_SIZE);
306     _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
307 
308     if (kernel_size) {
309         _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
310     }
311     if (initrd_size) {
312         _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
313     }
314     _FDT((fdt_finish_reservemap(fdt)));
315 
316     /* Root node */
317     _FDT((fdt_begin_node(fdt, "")));
318     _FDT((fdt_property_string(fdt, "device_type", "chrp")));
319     _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
320     _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
321 
322     /*
323      * Add info to guest to indentify which host is it being run on
324      * and what is the uuid of the guest
325      */
326     if (kvmppc_get_host_model(&buf)) {
327         _FDT((fdt_property_string(fdt, "host-model", buf)));
328         g_free(buf);
329     }
330     if (kvmppc_get_host_serial(&buf)) {
331         _FDT((fdt_property_string(fdt, "host-serial", buf)));
332         g_free(buf);
333     }
334 
335     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
336 
337     _FDT((fdt_property_string(fdt, "vm,uuid", buf)));
338     if (qemu_uuid_set) {
339         _FDT((fdt_property_string(fdt, "system-id", buf)));
340     }
341     g_free(buf);
342 
343     if (qemu_get_vm_name()) {
344         _FDT((fdt_property_string(fdt, "ibm,partition-name",
345                                   qemu_get_vm_name())));
346     }
347 
348     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
349     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
350 
351     /* /chosen */
352     _FDT((fdt_begin_node(fdt, "chosen")));
353 
354     /* Set Form1_affinity */
355     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
356 
357     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
358     _FDT((fdt_property(fdt, "linux,initrd-start",
359                        &start_prop, sizeof(start_prop))));
360     _FDT((fdt_property(fdt, "linux,initrd-end",
361                        &end_prop, sizeof(end_prop))));
362     if (kernel_size) {
363         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
364                               cpu_to_be64(kernel_size) };
365 
366         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
367         if (little_endian) {
368             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
369         }
370     }
371     if (boot_menu) {
372         _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
373     }
374     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
375     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
376     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
377 
378     _FDT((fdt_end_node(fdt)));
379 
380     /* RTAS */
381     _FDT((fdt_begin_node(fdt, "rtas")));
382 
383     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
384         add_str(hypertas, "hcall-multi-tce");
385     }
386     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
387                        hypertas->len)));
388     g_string_free(hypertas, TRUE);
389     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
390                        qemu_hypertas->len)));
391     g_string_free(qemu_hypertas, TRUE);
392 
393     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
394         refpoints, sizeof(refpoints))));
395 
396     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
397     _FDT((fdt_property_cell(fdt, "rtas-event-scan-rate",
398                             RTAS_EVENT_SCAN_RATE)));
399 
400     if (msi_nonbroken) {
401         _FDT((fdt_property(fdt, "ibm,change-msix-capable", NULL, 0)));
402     }
403 
404     /*
405      * According to PAPR, rtas ibm,os-term does not guarantee a return
406      * back to the guest cpu.
407      *
408      * While an additional ibm,extended-os-term property indicates that
409      * rtas call return will always occur. Set this property.
410      */
411     _FDT((fdt_property(fdt, "ibm,extended-os-term", NULL, 0)));
412 
413     _FDT((fdt_end_node(fdt)));
414 
415     /* interrupt controller */
416     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
417 
418     _FDT((fdt_property_string(fdt, "device_type",
419                               "PowerPC-External-Interrupt-Presentation")));
420     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
421     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
422     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
423                        interrupt_server_ranges_prop,
424                        sizeof(interrupt_server_ranges_prop))));
425     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
426     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
427     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
428 
429     _FDT((fdt_end_node(fdt)));
430 
431     /* vdevice */
432     _FDT((fdt_begin_node(fdt, "vdevice")));
433 
434     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
435     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
436     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
437     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
438     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
439     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
440 
441     _FDT((fdt_end_node(fdt)));
442 
443     /* event-sources */
444     spapr_events_fdt_skel(fdt, epow_irq);
445 
446     /* /hypervisor node */
447     if (kvm_enabled()) {
448         uint8_t hypercall[16];
449 
450         /* indicate KVM hypercall interface */
451         _FDT((fdt_begin_node(fdt, "hypervisor")));
452         _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
453         if (kvmppc_has_cap_fixup_hcalls()) {
454             /*
455              * Older KVM versions with older guest kernels were broken with the
456              * magic page, don't allow the guest to map it.
457              */
458             if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
459                                       sizeof(hypercall))) {
460                 _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
461                                    sizeof(hypercall))));
462             }
463         }
464         _FDT((fdt_end_node(fdt)));
465     }
466 
467     _FDT((fdt_end_node(fdt))); /* close root node */
468     _FDT((fdt_finish(fdt)));
469 
470     return fdt;
471 }
472 
473 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
474                                        hwaddr size)
475 {
476     uint32_t associativity[] = {
477         cpu_to_be32(0x4), /* length */
478         cpu_to_be32(0x0), cpu_to_be32(0x0),
479         cpu_to_be32(0x0), cpu_to_be32(nodeid)
480     };
481     char mem_name[32];
482     uint64_t mem_reg_property[2];
483     int off;
484 
485     mem_reg_property[0] = cpu_to_be64(start);
486     mem_reg_property[1] = cpu_to_be64(size);
487 
488     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
489     off = fdt_add_subnode(fdt, 0, mem_name);
490     _FDT(off);
491     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
492     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
493                       sizeof(mem_reg_property))));
494     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
495                       sizeof(associativity))));
496     return off;
497 }
498 
499 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
500 {
501     MachineState *machine = MACHINE(spapr);
502     hwaddr mem_start, node_size;
503     int i, nb_nodes = nb_numa_nodes;
504     NodeInfo *nodes = numa_info;
505     NodeInfo ramnode;
506 
507     /* No NUMA nodes, assume there is just one node with whole RAM */
508     if (!nb_numa_nodes) {
509         nb_nodes = 1;
510         ramnode.node_mem = machine->ram_size;
511         nodes = &ramnode;
512     }
513 
514     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
515         if (!nodes[i].node_mem) {
516             continue;
517         }
518         if (mem_start >= machine->ram_size) {
519             node_size = 0;
520         } else {
521             node_size = nodes[i].node_mem;
522             if (node_size > machine->ram_size - mem_start) {
523                 node_size = machine->ram_size - mem_start;
524             }
525         }
526         if (!mem_start) {
527             /* ppc_spapr_init() checks for rma_size <= node0_size already */
528             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
529             mem_start += spapr->rma_size;
530             node_size -= spapr->rma_size;
531         }
532         for ( ; node_size; ) {
533             hwaddr sizetmp = pow2floor(node_size);
534 
535             /* mem_start != 0 here */
536             if (ctzl(mem_start) < ctzl(sizetmp)) {
537                 sizetmp = 1ULL << ctzl(mem_start);
538             }
539 
540             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
541             node_size -= sizetmp;
542             mem_start += sizetmp;
543         }
544     }
545 
546     return 0;
547 }
548 
549 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
550                                   sPAPRMachineState *spapr)
551 {
552     PowerPCCPU *cpu = POWERPC_CPU(cs);
553     CPUPPCState *env = &cpu->env;
554     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
555     int index = ppc_get_vcpu_dt_id(cpu);
556     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
557                        0xffffffff, 0xffffffff};
558     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
559         : SPAPR_TIMEBASE_FREQ;
560     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
561     uint32_t page_sizes_prop[64];
562     size_t page_sizes_prop_size;
563     uint32_t vcpus_per_socket = smp_threads * smp_cores;
564     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
565     sPAPRDRConnector *drc;
566     sPAPRDRConnectorClass *drck;
567     int drc_index;
568 
569     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
570     if (drc) {
571         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
572         drc_index = drck->get_index(drc);
573         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
574     }
575 
576     /* Note: we keep CI large pages off for now because a 64K capable guest
577      * provisioned with large pages might otherwise try to map a qemu
578      * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
579      * even if that qemu runs on a 4k host.
580      *
581      * We can later add this bit back when we are confident this is not
582      * an issue (!HV KVM or 64K host)
583      */
584     uint8_t pa_features_206[] = { 6, 0,
585         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
586     uint8_t pa_features_207[] = { 24, 0,
587         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
588         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
589         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
590         0x80, 0x00, 0x80, 0x00, 0x80, 0x00 };
591     uint8_t *pa_features;
592     size_t pa_size;
593 
594     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
595     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
596 
597     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
598     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
599                            env->dcache_line_size)));
600     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
601                            env->dcache_line_size)));
602     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
603                            env->icache_line_size)));
604     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
605                            env->icache_line_size)));
606 
607     if (pcc->l1_dcache_size) {
608         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
609                                pcc->l1_dcache_size)));
610     } else {
611         error_report("Warning: Unknown L1 dcache size for cpu");
612     }
613     if (pcc->l1_icache_size) {
614         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
615                                pcc->l1_icache_size)));
616     } else {
617         error_report("Warning: Unknown L1 icache size for cpu");
618     }
619 
620     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
621     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
622     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
623     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
624     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
625     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
626 
627     if (env->spr_cb[SPR_PURR].oea_read) {
628         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
629     }
630 
631     if (env->mmu_model & POWERPC_MMU_1TSEG) {
632         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
633                           segs, sizeof(segs))));
634     }
635 
636     /* Advertise VMX/VSX (vector extensions) if available
637      *   0 / no property == no vector extensions
638      *   1               == VMX / Altivec available
639      *   2               == VSX available */
640     if (env->insns_flags & PPC_ALTIVEC) {
641         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
642 
643         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
644     }
645 
646     /* Advertise DFP (Decimal Floating Point) if available
647      *   0 / no property == no DFP
648      *   1               == DFP available */
649     if (env->insns_flags2 & PPC2_DFP) {
650         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
651     }
652 
653     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
654                                                   sizeof(page_sizes_prop));
655     if (page_sizes_prop_size) {
656         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
657                           page_sizes_prop, page_sizes_prop_size)));
658     }
659 
660     /* Do the ibm,pa-features property, adjust it for ci-large-pages */
661     if (env->mmu_model == POWERPC_MMU_2_06) {
662         pa_features = pa_features_206;
663         pa_size = sizeof(pa_features_206);
664     } else /* env->mmu_model == POWERPC_MMU_2_07 */ {
665         pa_features = pa_features_207;
666         pa_size = sizeof(pa_features_207);
667     }
668     if (env->ci_large_pages) {
669         pa_features[3] |= 0x20;
670     }
671     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
672 
673     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
674                            cs->cpu_index / vcpus_per_socket)));
675 
676     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
677                       pft_size_prop, sizeof(pft_size_prop))));
678 
679     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
680 
681     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
682                                 ppc_get_compat_smt_threads(cpu)));
683 }
684 
685 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
686 {
687     CPUState *cs;
688     int cpus_offset;
689     char *nodename;
690     int smt = kvmppc_smt_threads();
691 
692     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
693     _FDT(cpus_offset);
694     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
695     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
696 
697     /*
698      * We walk the CPUs in reverse order to ensure that CPU DT nodes
699      * created by fdt_add_subnode() end up in the right order in FDT
700      * for the guest kernel the enumerate the CPUs correctly.
701      */
702     CPU_FOREACH_REVERSE(cs) {
703         PowerPCCPU *cpu = POWERPC_CPU(cs);
704         int index = ppc_get_vcpu_dt_id(cpu);
705         DeviceClass *dc = DEVICE_GET_CLASS(cs);
706         int offset;
707 
708         if ((index % smt) != 0) {
709             continue;
710         }
711 
712         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
713         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
714         g_free(nodename);
715         _FDT(offset);
716         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
717     }
718 
719 }
720 
721 /*
722  * Adds ibm,dynamic-reconfiguration-memory node.
723  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
724  * of this device tree node.
725  */
726 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
727 {
728     MachineState *machine = MACHINE(spapr);
729     int ret, i, offset;
730     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
731     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
732     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
733     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
734                        memory_region_size(&spapr->hotplug_memory.mr)) /
735                        lmb_size;
736     uint32_t *int_buf, *cur_index, buf_len;
737     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
738 
739     /*
740      * Don't create the node if there is no hotpluggable memory
741      */
742     if (machine->ram_size == machine->maxram_size) {
743         return 0;
744     }
745 
746     /*
747      * Allocate enough buffer size to fit in ibm,dynamic-memory
748      * or ibm,associativity-lookup-arrays
749      */
750     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
751               * sizeof(uint32_t);
752     cur_index = int_buf = g_malloc0(buf_len);
753 
754     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
755 
756     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
757                     sizeof(prop_lmb_size));
758     if (ret < 0) {
759         goto out;
760     }
761 
762     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
763     if (ret < 0) {
764         goto out;
765     }
766 
767     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
768     if (ret < 0) {
769         goto out;
770     }
771 
772     /* ibm,dynamic-memory */
773     int_buf[0] = cpu_to_be32(nr_lmbs);
774     cur_index++;
775     for (i = 0; i < nr_lmbs; i++) {
776         uint64_t addr = i * lmb_size;
777         uint32_t *dynamic_memory = cur_index;
778 
779         if (i >= hotplug_lmb_start) {
780             sPAPRDRConnector *drc;
781             sPAPRDRConnectorClass *drck;
782 
783             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
784             g_assert(drc);
785             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
786 
787             dynamic_memory[0] = cpu_to_be32(addr >> 32);
788             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
789             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
790             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
791             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
792             if (memory_region_present(get_system_memory(), addr)) {
793                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
794             } else {
795                 dynamic_memory[5] = cpu_to_be32(0);
796             }
797         } else {
798             /*
799              * LMB information for RMA, boot time RAM and gap b/n RAM and
800              * hotplug memory region -- all these are marked as reserved
801              * and as having no valid DRC.
802              */
803             dynamic_memory[0] = cpu_to_be32(addr >> 32);
804             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
805             dynamic_memory[2] = cpu_to_be32(0);
806             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
807             dynamic_memory[4] = cpu_to_be32(-1);
808             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
809                                             SPAPR_LMB_FLAGS_DRC_INVALID);
810         }
811 
812         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
813     }
814     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
815     if (ret < 0) {
816         goto out;
817     }
818 
819     /* ibm,associativity-lookup-arrays */
820     cur_index = int_buf;
821     int_buf[0] = cpu_to_be32(nr_nodes);
822     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
823     cur_index += 2;
824     for (i = 0; i < nr_nodes; i++) {
825         uint32_t associativity[] = {
826             cpu_to_be32(0x0),
827             cpu_to_be32(0x0),
828             cpu_to_be32(0x0),
829             cpu_to_be32(i)
830         };
831         memcpy(cur_index, associativity, sizeof(associativity));
832         cur_index += 4;
833     }
834     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
835             (cur_index - int_buf) * sizeof(uint32_t));
836 out:
837     g_free(int_buf);
838     return ret;
839 }
840 
841 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
842                                  target_ulong addr, target_ulong size,
843                                  bool cpu_update, bool memory_update)
844 {
845     void *fdt, *fdt_skel;
846     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
847     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
848 
849     size -= sizeof(hdr);
850 
851     /* Create sceleton */
852     fdt_skel = g_malloc0(size);
853     _FDT((fdt_create(fdt_skel, size)));
854     _FDT((fdt_begin_node(fdt_skel, "")));
855     _FDT((fdt_end_node(fdt_skel)));
856     _FDT((fdt_finish(fdt_skel)));
857     fdt = g_malloc0(size);
858     _FDT((fdt_open_into(fdt_skel, fdt, size)));
859     g_free(fdt_skel);
860 
861     /* Fixup cpu nodes */
862     if (cpu_update) {
863         _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
864     }
865 
866     /* Generate ibm,dynamic-reconfiguration-memory node if required */
867     if (memory_update && smc->dr_lmb_enabled) {
868         _FDT((spapr_populate_drconf_memory(spapr, fdt)));
869     }
870 
871     /* Pack resulting tree */
872     _FDT((fdt_pack(fdt)));
873 
874     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
875         trace_spapr_cas_failed(size);
876         return -1;
877     }
878 
879     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
880     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
881     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
882     g_free(fdt);
883 
884     return 0;
885 }
886 
887 static void spapr_finalize_fdt(sPAPRMachineState *spapr,
888                                hwaddr fdt_addr,
889                                hwaddr rtas_addr,
890                                hwaddr rtas_size)
891 {
892     MachineState *machine = MACHINE(qdev_get_machine());
893     MachineClass *mc = MACHINE_GET_CLASS(machine);
894     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
895     const char *boot_device = machine->boot_order;
896     int ret, i;
897     size_t cb = 0;
898     char *bootlist;
899     void *fdt;
900     sPAPRPHBState *phb;
901 
902     fdt = g_malloc(FDT_MAX_SIZE);
903 
904     /* open out the base tree into a temp buffer for the final tweaks */
905     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
906 
907     ret = spapr_populate_memory(spapr, fdt);
908     if (ret < 0) {
909         error_report("couldn't setup memory nodes in fdt");
910         exit(1);
911     }
912 
913     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
914     if (ret < 0) {
915         error_report("couldn't setup vio devices in fdt");
916         exit(1);
917     }
918 
919     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
920         ret = spapr_rng_populate_dt(fdt);
921         if (ret < 0) {
922             error_report("could not set up rng device in the fdt");
923             exit(1);
924         }
925     }
926 
927     QLIST_FOREACH(phb, &spapr->phbs, list) {
928         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
929         if (ret < 0) {
930             error_report("couldn't setup PCI devices in fdt");
931             exit(1);
932         }
933     }
934 
935     /* RTAS */
936     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
937     if (ret < 0) {
938         error_report("Couldn't set up RTAS device tree properties");
939     }
940 
941     /* cpus */
942     spapr_populate_cpus_dt_node(fdt, spapr);
943 
944     bootlist = get_boot_devices_list(&cb, true);
945     if (cb && bootlist) {
946         int offset = fdt_path_offset(fdt, "/chosen");
947         if (offset < 0) {
948             exit(1);
949         }
950         for (i = 0; i < cb; i++) {
951             if (bootlist[i] == '\n') {
952                 bootlist[i] = ' ';
953             }
954 
955         }
956         ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
957     }
958 
959     if (boot_device && strlen(boot_device)) {
960         int offset = fdt_path_offset(fdt, "/chosen");
961 
962         if (offset < 0) {
963             exit(1);
964         }
965         fdt_setprop_string(fdt, offset, "qemu,boot-device", boot_device);
966     }
967 
968     if (!spapr->has_graphics) {
969         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
970     }
971 
972     if (smc->dr_lmb_enabled) {
973         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
974     }
975 
976     if (mc->query_hotpluggable_cpus) {
977         int offset = fdt_path_offset(fdt, "/cpus");
978         ret = spapr_drc_populate_dt(fdt, offset, NULL,
979                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
980         if (ret < 0) {
981             error_report("Couldn't set up CPU DR device tree properties");
982             exit(1);
983         }
984     }
985 
986     _FDT((fdt_pack(fdt)));
987 
988     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
989         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
990                      fdt_totalsize(fdt), FDT_MAX_SIZE);
991         exit(1);
992     }
993 
994     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
995     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
996 
997     g_free(bootlist);
998     g_free(fdt);
999 }
1000 
1001 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1002 {
1003     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1004 }
1005 
1006 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
1007 {
1008     CPUPPCState *env = &cpu->env;
1009 
1010     if (msr_pr) {
1011         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1012         env->gpr[3] = H_PRIVILEGE;
1013     } else {
1014         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1015     }
1016 }
1017 
1018 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1019 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1020 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1021 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1022 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1023 
1024 /*
1025  * Get the fd to access the kernel htab, re-opening it if necessary
1026  */
1027 static int get_htab_fd(sPAPRMachineState *spapr)
1028 {
1029     if (spapr->htab_fd >= 0) {
1030         return spapr->htab_fd;
1031     }
1032 
1033     spapr->htab_fd = kvmppc_get_htab_fd(false);
1034     if (spapr->htab_fd < 0) {
1035         error_report("Unable to open fd for reading hash table from KVM: %s",
1036                      strerror(errno));
1037     }
1038 
1039     return spapr->htab_fd;
1040 }
1041 
1042 static void close_htab_fd(sPAPRMachineState *spapr)
1043 {
1044     if (spapr->htab_fd >= 0) {
1045         close(spapr->htab_fd);
1046     }
1047     spapr->htab_fd = -1;
1048 }
1049 
1050 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1051 {
1052     int shift;
1053 
1054     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1055      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1056      * that's much more than is needed for Linux guests */
1057     shift = ctz64(pow2ceil(ramsize)) - 7;
1058     shift = MAX(shift, 18); /* Minimum architected size */
1059     shift = MIN(shift, 46); /* Maximum architected size */
1060     return shift;
1061 }
1062 
1063 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1064                                  Error **errp)
1065 {
1066     long rc;
1067 
1068     /* Clean up any HPT info from a previous boot */
1069     g_free(spapr->htab);
1070     spapr->htab = NULL;
1071     spapr->htab_shift = 0;
1072     close_htab_fd(spapr);
1073 
1074     rc = kvmppc_reset_htab(shift);
1075     if (rc < 0) {
1076         /* kernel-side HPT needed, but couldn't allocate one */
1077         error_setg_errno(errp, errno,
1078                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1079                          shift);
1080         /* This is almost certainly fatal, but if the caller really
1081          * wants to carry on with shift == 0, it's welcome to try */
1082     } else if (rc > 0) {
1083         /* kernel-side HPT allocated */
1084         if (rc != shift) {
1085             error_setg(errp,
1086                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1087                        shift, rc);
1088         }
1089 
1090         spapr->htab_shift = shift;
1091         spapr->htab = NULL;
1092     } else {
1093         /* kernel-side HPT not needed, allocate in userspace instead */
1094         size_t size = 1ULL << shift;
1095         int i;
1096 
1097         spapr->htab = qemu_memalign(size, size);
1098         if (!spapr->htab) {
1099             error_setg_errno(errp, errno,
1100                              "Could not allocate HPT of order %d", shift);
1101             return;
1102         }
1103 
1104         memset(spapr->htab, 0, size);
1105         spapr->htab_shift = shift;
1106 
1107         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1108             DIRTY_HPTE(HPTE(spapr->htab, i));
1109         }
1110     }
1111 }
1112 
1113 static int find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1114 {
1115     bool matched = false;
1116 
1117     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1118         matched = true;
1119     }
1120 
1121     if (!matched) {
1122         error_report("Device %s is not supported by this machine yet.",
1123                      qdev_fw_name(DEVICE(sbdev)));
1124         exit(1);
1125     }
1126 
1127     return 0;
1128 }
1129 
1130 static void ppc_spapr_reset(void)
1131 {
1132     MachineState *machine = MACHINE(qdev_get_machine());
1133     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1134     PowerPCCPU *first_ppc_cpu;
1135     uint32_t rtas_limit;
1136 
1137     /* Check for unknown sysbus devices */
1138     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1139 
1140     /* Allocate and/or reset the hash page table */
1141     spapr_reallocate_hpt(spapr,
1142                          spapr_hpt_shift_for_ramsize(machine->maxram_size),
1143                          &error_fatal);
1144 
1145     /* Update the RMA size if necessary */
1146     if (spapr->vrma_adjust) {
1147         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1148                                           spapr->htab_shift);
1149     }
1150 
1151     qemu_devices_reset();
1152 
1153     /*
1154      * We place the device tree and RTAS just below either the top of the RMA,
1155      * or just below 2GB, whichever is lowere, so that it can be
1156      * processed with 32-bit real mode code if necessary
1157      */
1158     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1159     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1160     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1161 
1162     /* Load the fdt */
1163     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
1164                        spapr->rtas_size);
1165 
1166     /* Copy RTAS over */
1167     cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
1168                               spapr->rtas_size);
1169 
1170     /* Set up the entry state */
1171     first_ppc_cpu = POWERPC_CPU(first_cpu);
1172     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
1173     first_ppc_cpu->env.gpr[5] = 0;
1174     first_cpu->halted = 0;
1175     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1176 
1177 }
1178 
1179 static void spapr_create_nvram(sPAPRMachineState *spapr)
1180 {
1181     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1182     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1183 
1184     if (dinfo) {
1185         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1186                             &error_fatal);
1187     }
1188 
1189     qdev_init_nofail(dev);
1190 
1191     spapr->nvram = (struct sPAPRNVRAM *)dev;
1192 }
1193 
1194 static void spapr_rtc_create(sPAPRMachineState *spapr)
1195 {
1196     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1197 
1198     qdev_init_nofail(dev);
1199     spapr->rtc = dev;
1200 
1201     object_property_add_alias(qdev_get_machine(), "rtc-time",
1202                               OBJECT(spapr->rtc), "date", NULL);
1203 }
1204 
1205 /* Returns whether we want to use VGA or not */
1206 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1207 {
1208     switch (vga_interface_type) {
1209     case VGA_NONE:
1210         return false;
1211     case VGA_DEVICE:
1212         return true;
1213     case VGA_STD:
1214     case VGA_VIRTIO:
1215         return pci_vga_init(pci_bus) != NULL;
1216     default:
1217         error_setg(errp,
1218                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1219         return false;
1220     }
1221 }
1222 
1223 static int spapr_post_load(void *opaque, int version_id)
1224 {
1225     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1226     int err = 0;
1227 
1228     /* In earlier versions, there was no separate qdev for the PAPR
1229      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1230      * So when migrating from those versions, poke the incoming offset
1231      * value into the RTC device */
1232     if (version_id < 3) {
1233         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1234     }
1235 
1236     return err;
1237 }
1238 
1239 static bool version_before_3(void *opaque, int version_id)
1240 {
1241     return version_id < 3;
1242 }
1243 
1244 static const VMStateDescription vmstate_spapr = {
1245     .name = "spapr",
1246     .version_id = 3,
1247     .minimum_version_id = 1,
1248     .post_load = spapr_post_load,
1249     .fields = (VMStateField[]) {
1250         /* used to be @next_irq */
1251         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1252 
1253         /* RTC offset */
1254         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1255 
1256         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1257         VMSTATE_END_OF_LIST()
1258     },
1259 };
1260 
1261 static int htab_save_setup(QEMUFile *f, void *opaque)
1262 {
1263     sPAPRMachineState *spapr = opaque;
1264 
1265     /* "Iteration" header */
1266     qemu_put_be32(f, spapr->htab_shift);
1267 
1268     if (spapr->htab) {
1269         spapr->htab_save_index = 0;
1270         spapr->htab_first_pass = true;
1271     } else {
1272         assert(kvm_enabled());
1273     }
1274 
1275 
1276     return 0;
1277 }
1278 
1279 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1280                                  int64_t max_ns)
1281 {
1282     bool has_timeout = max_ns != -1;
1283     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1284     int index = spapr->htab_save_index;
1285     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1286 
1287     assert(spapr->htab_first_pass);
1288 
1289     do {
1290         int chunkstart;
1291 
1292         /* Consume invalid HPTEs */
1293         while ((index < htabslots)
1294                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1295             index++;
1296             CLEAN_HPTE(HPTE(spapr->htab, index));
1297         }
1298 
1299         /* Consume valid HPTEs */
1300         chunkstart = index;
1301         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1302                && HPTE_VALID(HPTE(spapr->htab, index))) {
1303             index++;
1304             CLEAN_HPTE(HPTE(spapr->htab, index));
1305         }
1306 
1307         if (index > chunkstart) {
1308             int n_valid = index - chunkstart;
1309 
1310             qemu_put_be32(f, chunkstart);
1311             qemu_put_be16(f, n_valid);
1312             qemu_put_be16(f, 0);
1313             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1314                             HASH_PTE_SIZE_64 * n_valid);
1315 
1316             if (has_timeout &&
1317                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1318                 break;
1319             }
1320         }
1321     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1322 
1323     if (index >= htabslots) {
1324         assert(index == htabslots);
1325         index = 0;
1326         spapr->htab_first_pass = false;
1327     }
1328     spapr->htab_save_index = index;
1329 }
1330 
1331 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1332                                 int64_t max_ns)
1333 {
1334     bool final = max_ns < 0;
1335     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1336     int examined = 0, sent = 0;
1337     int index = spapr->htab_save_index;
1338     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1339 
1340     assert(!spapr->htab_first_pass);
1341 
1342     do {
1343         int chunkstart, invalidstart;
1344 
1345         /* Consume non-dirty HPTEs */
1346         while ((index < htabslots)
1347                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1348             index++;
1349             examined++;
1350         }
1351 
1352         chunkstart = index;
1353         /* Consume valid dirty HPTEs */
1354         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1355                && HPTE_DIRTY(HPTE(spapr->htab, index))
1356                && HPTE_VALID(HPTE(spapr->htab, index))) {
1357             CLEAN_HPTE(HPTE(spapr->htab, index));
1358             index++;
1359             examined++;
1360         }
1361 
1362         invalidstart = index;
1363         /* Consume invalid dirty HPTEs */
1364         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1365                && HPTE_DIRTY(HPTE(spapr->htab, index))
1366                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1367             CLEAN_HPTE(HPTE(spapr->htab, index));
1368             index++;
1369             examined++;
1370         }
1371 
1372         if (index > chunkstart) {
1373             int n_valid = invalidstart - chunkstart;
1374             int n_invalid = index - invalidstart;
1375 
1376             qemu_put_be32(f, chunkstart);
1377             qemu_put_be16(f, n_valid);
1378             qemu_put_be16(f, n_invalid);
1379             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1380                             HASH_PTE_SIZE_64 * n_valid);
1381             sent += index - chunkstart;
1382 
1383             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1384                 break;
1385             }
1386         }
1387 
1388         if (examined >= htabslots) {
1389             break;
1390         }
1391 
1392         if (index >= htabslots) {
1393             assert(index == htabslots);
1394             index = 0;
1395         }
1396     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1397 
1398     if (index >= htabslots) {
1399         assert(index == htabslots);
1400         index = 0;
1401     }
1402 
1403     spapr->htab_save_index = index;
1404 
1405     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1406 }
1407 
1408 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1409 #define MAX_KVM_BUF_SIZE    2048
1410 
1411 static int htab_save_iterate(QEMUFile *f, void *opaque)
1412 {
1413     sPAPRMachineState *spapr = opaque;
1414     int fd;
1415     int rc = 0;
1416 
1417     /* Iteration header */
1418     qemu_put_be32(f, 0);
1419 
1420     if (!spapr->htab) {
1421         assert(kvm_enabled());
1422 
1423         fd = get_htab_fd(spapr);
1424         if (fd < 0) {
1425             return fd;
1426         }
1427 
1428         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1429         if (rc < 0) {
1430             return rc;
1431         }
1432     } else  if (spapr->htab_first_pass) {
1433         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1434     } else {
1435         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1436     }
1437 
1438     /* End marker */
1439     qemu_put_be32(f, 0);
1440     qemu_put_be16(f, 0);
1441     qemu_put_be16(f, 0);
1442 
1443     return rc;
1444 }
1445 
1446 static int htab_save_complete(QEMUFile *f, void *opaque)
1447 {
1448     sPAPRMachineState *spapr = opaque;
1449     int fd;
1450 
1451     /* Iteration header */
1452     qemu_put_be32(f, 0);
1453 
1454     if (!spapr->htab) {
1455         int rc;
1456 
1457         assert(kvm_enabled());
1458 
1459         fd = get_htab_fd(spapr);
1460         if (fd < 0) {
1461             return fd;
1462         }
1463 
1464         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1465         if (rc < 0) {
1466             return rc;
1467         }
1468     } else {
1469         if (spapr->htab_first_pass) {
1470             htab_save_first_pass(f, spapr, -1);
1471         }
1472         htab_save_later_pass(f, spapr, -1);
1473     }
1474 
1475     /* End marker */
1476     qemu_put_be32(f, 0);
1477     qemu_put_be16(f, 0);
1478     qemu_put_be16(f, 0);
1479 
1480     return 0;
1481 }
1482 
1483 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1484 {
1485     sPAPRMachineState *spapr = opaque;
1486     uint32_t section_hdr;
1487     int fd = -1;
1488 
1489     if (version_id < 1 || version_id > 1) {
1490         error_report("htab_load() bad version");
1491         return -EINVAL;
1492     }
1493 
1494     section_hdr = qemu_get_be32(f);
1495 
1496     if (section_hdr) {
1497         Error *local_err = NULL;
1498 
1499         /* First section gives the htab size */
1500         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1501         if (local_err) {
1502             error_report_err(local_err);
1503             return -EINVAL;
1504         }
1505         return 0;
1506     }
1507 
1508     if (!spapr->htab) {
1509         assert(kvm_enabled());
1510 
1511         fd = kvmppc_get_htab_fd(true);
1512         if (fd < 0) {
1513             error_report("Unable to open fd to restore KVM hash table: %s",
1514                          strerror(errno));
1515         }
1516     }
1517 
1518     while (true) {
1519         uint32_t index;
1520         uint16_t n_valid, n_invalid;
1521 
1522         index = qemu_get_be32(f);
1523         n_valid = qemu_get_be16(f);
1524         n_invalid = qemu_get_be16(f);
1525 
1526         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1527             /* End of Stream */
1528             break;
1529         }
1530 
1531         if ((index + n_valid + n_invalid) >
1532             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1533             /* Bad index in stream */
1534             error_report(
1535                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1536                 index, n_valid, n_invalid, spapr->htab_shift);
1537             return -EINVAL;
1538         }
1539 
1540         if (spapr->htab) {
1541             if (n_valid) {
1542                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1543                                 HASH_PTE_SIZE_64 * n_valid);
1544             }
1545             if (n_invalid) {
1546                 memset(HPTE(spapr->htab, index + n_valid), 0,
1547                        HASH_PTE_SIZE_64 * n_invalid);
1548             }
1549         } else {
1550             int rc;
1551 
1552             assert(fd >= 0);
1553 
1554             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1555             if (rc < 0) {
1556                 return rc;
1557             }
1558         }
1559     }
1560 
1561     if (!spapr->htab) {
1562         assert(fd >= 0);
1563         close(fd);
1564     }
1565 
1566     return 0;
1567 }
1568 
1569 static void htab_cleanup(void *opaque)
1570 {
1571     sPAPRMachineState *spapr = opaque;
1572 
1573     close_htab_fd(spapr);
1574 }
1575 
1576 static SaveVMHandlers savevm_htab_handlers = {
1577     .save_live_setup = htab_save_setup,
1578     .save_live_iterate = htab_save_iterate,
1579     .save_live_complete_precopy = htab_save_complete,
1580     .cleanup = htab_cleanup,
1581     .load_state = htab_load,
1582 };
1583 
1584 static void spapr_boot_set(void *opaque, const char *boot_device,
1585                            Error **errp)
1586 {
1587     MachineState *machine = MACHINE(qdev_get_machine());
1588     machine->boot_order = g_strdup(boot_device);
1589 }
1590 
1591 /*
1592  * Reset routine for LMB DR devices.
1593  *
1594  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1595  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1596  * when it walks all its children devices. LMB devices reset occurs
1597  * as part of spapr_ppc_reset().
1598  */
1599 static void spapr_drc_reset(void *opaque)
1600 {
1601     sPAPRDRConnector *drc = opaque;
1602     DeviceState *d = DEVICE(drc);
1603 
1604     if (d) {
1605         device_reset(d);
1606     }
1607 }
1608 
1609 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1610 {
1611     MachineState *machine = MACHINE(spapr);
1612     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1613     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1614     int i;
1615 
1616     for (i = 0; i < nr_lmbs; i++) {
1617         sPAPRDRConnector *drc;
1618         uint64_t addr;
1619 
1620         addr = i * lmb_size + spapr->hotplug_memory.base;
1621         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1622                                      addr/lmb_size);
1623         qemu_register_reset(spapr_drc_reset, drc);
1624     }
1625 }
1626 
1627 /*
1628  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1629  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1630  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1631  */
1632 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1633 {
1634     int i;
1635 
1636     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1637         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1638                    " is not aligned to %llu MiB",
1639                    machine->ram_size,
1640                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1641         return;
1642     }
1643 
1644     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1645         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1646                    " is not aligned to %llu MiB",
1647                    machine->ram_size,
1648                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1649         return;
1650     }
1651 
1652     for (i = 0; i < nb_numa_nodes; i++) {
1653         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1654             error_setg(errp,
1655                        "Node %d memory size 0x%" PRIx64
1656                        " is not aligned to %llu MiB",
1657                        i, numa_info[i].node_mem,
1658                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1659             return;
1660         }
1661     }
1662 }
1663 
1664 /* pSeries LPAR / sPAPR hardware init */
1665 static void ppc_spapr_init(MachineState *machine)
1666 {
1667     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1668     MachineClass *mc = MACHINE_GET_CLASS(machine);
1669     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1670     const char *kernel_filename = machine->kernel_filename;
1671     const char *kernel_cmdline = machine->kernel_cmdline;
1672     const char *initrd_filename = machine->initrd_filename;
1673     PCIHostState *phb;
1674     int i;
1675     MemoryRegion *sysmem = get_system_memory();
1676     MemoryRegion *ram = g_new(MemoryRegion, 1);
1677     MemoryRegion *rma_region;
1678     void *rma = NULL;
1679     hwaddr rma_alloc_size;
1680     hwaddr node0_size = spapr_node0_size();
1681     uint32_t initrd_base = 0;
1682     long kernel_size = 0, initrd_size = 0;
1683     long load_limit, fw_size;
1684     bool kernel_le = false;
1685     char *filename;
1686     int smt = kvmppc_smt_threads();
1687     int spapr_cores = smp_cpus / smp_threads;
1688     int spapr_max_cores = max_cpus / smp_threads;
1689 
1690     if (mc->query_hotpluggable_cpus) {
1691         if (smp_cpus % smp_threads) {
1692             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1693                          smp_cpus, smp_threads);
1694             exit(1);
1695         }
1696         if (max_cpus % smp_threads) {
1697             error_report("max_cpus (%u) must be multiple of threads (%u)",
1698                          max_cpus, smp_threads);
1699             exit(1);
1700         }
1701     }
1702 
1703     msi_nonbroken = true;
1704 
1705     QLIST_INIT(&spapr->phbs);
1706 
1707     cpu_ppc_hypercall = emulate_spapr_hypercall;
1708 
1709     /* Allocate RMA if necessary */
1710     rma_alloc_size = kvmppc_alloc_rma(&rma);
1711 
1712     if (rma_alloc_size == -1) {
1713         error_report("Unable to create RMA");
1714         exit(1);
1715     }
1716 
1717     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1718         spapr->rma_size = rma_alloc_size;
1719     } else {
1720         spapr->rma_size = node0_size;
1721 
1722         /* With KVM, we don't actually know whether KVM supports an
1723          * unbounded RMA (PR KVM) or is limited by the hash table size
1724          * (HV KVM using VRMA), so we always assume the latter
1725          *
1726          * In that case, we also limit the initial allocations for RTAS
1727          * etc... to 256M since we have no way to know what the VRMA size
1728          * is going to be as it depends on the size of the hash table
1729          * isn't determined yet.
1730          */
1731         if (kvm_enabled()) {
1732             spapr->vrma_adjust = 1;
1733             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1734         }
1735 
1736         /* Actually we don't support unbounded RMA anymore since we
1737          * added proper emulation of HV mode. The max we can get is
1738          * 16G which also happens to be what we configure for PAPR
1739          * mode so make sure we don't do anything bigger than that
1740          */
1741         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1742     }
1743 
1744     if (spapr->rma_size > node0_size) {
1745         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1746                      spapr->rma_size);
1747         exit(1);
1748     }
1749 
1750     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1751     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1752 
1753     /* Set up Interrupt Controller before we create the VCPUs */
1754     spapr->xics = xics_system_init(machine,
1755                                    DIV_ROUND_UP(max_cpus * smt, smp_threads),
1756                                    XICS_IRQS_SPAPR, &error_fatal);
1757 
1758     if (smc->dr_lmb_enabled) {
1759         spapr_validate_node_memory(machine, &error_fatal);
1760     }
1761 
1762     /* init CPUs */
1763     if (machine->cpu_model == NULL) {
1764         machine->cpu_model = kvm_enabled() ? "host" : "POWER7";
1765     }
1766 
1767     ppc_cpu_parse_features(machine->cpu_model);
1768 
1769     if (mc->query_hotpluggable_cpus) {
1770         char *type = spapr_get_cpu_core_type(machine->cpu_model);
1771 
1772         if (type == NULL) {
1773             error_report("Unable to find sPAPR CPU Core definition");
1774             exit(1);
1775         }
1776 
1777         spapr->cores = g_new0(Object *, spapr_max_cores);
1778         for (i = 0; i < spapr_max_cores; i++) {
1779             int core_id = i * smp_threads;
1780             sPAPRDRConnector *drc =
1781                 spapr_dr_connector_new(OBJECT(spapr),
1782                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
1783                                        (core_id / smp_threads) * smt);
1784 
1785             qemu_register_reset(spapr_drc_reset, drc);
1786 
1787             if (i < spapr_cores) {
1788                 Object *core  = object_new(type);
1789                 object_property_set_int(core, smp_threads, "nr-threads",
1790                                         &error_fatal);
1791                 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1792                                         &error_fatal);
1793                 object_property_set_bool(core, true, "realized", &error_fatal);
1794             }
1795         }
1796         g_free(type);
1797     } else {
1798         for (i = 0; i < smp_cpus; i++) {
1799             PowerPCCPU *cpu = cpu_ppc_init(machine->cpu_model);
1800             if (cpu == NULL) {
1801                 error_report("Unable to find PowerPC CPU definition");
1802                 exit(1);
1803             }
1804             spapr_cpu_init(spapr, cpu, &error_fatal);
1805        }
1806     }
1807 
1808     if (kvm_enabled()) {
1809         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1810         kvmppc_enable_logical_ci_hcalls();
1811         kvmppc_enable_set_mode_hcall();
1812 
1813         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
1814         kvmppc_enable_clear_ref_mod_hcalls();
1815     }
1816 
1817     /* allocate RAM */
1818     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1819                                          machine->ram_size);
1820     memory_region_add_subregion(sysmem, 0, ram);
1821 
1822     if (rma_alloc_size && rma) {
1823         rma_region = g_new(MemoryRegion, 1);
1824         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1825                                    rma_alloc_size, rma);
1826         vmstate_register_ram_global(rma_region);
1827         memory_region_add_subregion(sysmem, 0, rma_region);
1828     }
1829 
1830     /* initialize hotplug memory address space */
1831     if (machine->ram_size < machine->maxram_size) {
1832         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
1833         /*
1834          * Limit the number of hotpluggable memory slots to half the number
1835          * slots that KVM supports, leaving the other half for PCI and other
1836          * devices. However ensure that number of slots doesn't drop below 32.
1837          */
1838         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
1839                            SPAPR_MAX_RAM_SLOTS;
1840 
1841         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
1842             max_memslots = SPAPR_MAX_RAM_SLOTS;
1843         }
1844         if (machine->ram_slots > max_memslots) {
1845             error_report("Specified number of memory slots %"
1846                          PRIu64" exceeds max supported %d",
1847                          machine->ram_slots, max_memslots);
1848             exit(1);
1849         }
1850 
1851         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
1852                                               SPAPR_HOTPLUG_MEM_ALIGN);
1853         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
1854                            "hotplug-memory", hotplug_mem_size);
1855         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
1856                                     &spapr->hotplug_memory.mr);
1857     }
1858 
1859     if (smc->dr_lmb_enabled) {
1860         spapr_create_lmb_dr_connectors(spapr);
1861     }
1862 
1863     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1864     if (!filename) {
1865         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1866         exit(1);
1867     }
1868     spapr->rtas_size = get_image_size(filename);
1869     if (spapr->rtas_size < 0) {
1870         error_report("Could not get size of LPAR rtas '%s'", filename);
1871         exit(1);
1872     }
1873     spapr->rtas_blob = g_malloc(spapr->rtas_size);
1874     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1875         error_report("Could not load LPAR rtas '%s'", filename);
1876         exit(1);
1877     }
1878     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1879         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1880                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1881         exit(1);
1882     }
1883     g_free(filename);
1884 
1885     /* Set up EPOW events infrastructure */
1886     spapr_events_init(spapr);
1887 
1888     /* Set up the RTC RTAS interfaces */
1889     spapr_rtc_create(spapr);
1890 
1891     /* Set up VIO bus */
1892     spapr->vio_bus = spapr_vio_bus_init();
1893 
1894     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1895         if (serial_hds[i]) {
1896             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1897         }
1898     }
1899 
1900     /* We always have at least the nvram device on VIO */
1901     spapr_create_nvram(spapr);
1902 
1903     /* Set up PCI */
1904     spapr_pci_rtas_init();
1905 
1906     phb = spapr_create_phb(spapr, 0);
1907 
1908     for (i = 0; i < nb_nics; i++) {
1909         NICInfo *nd = &nd_table[i];
1910 
1911         if (!nd->model) {
1912             nd->model = g_strdup("ibmveth");
1913         }
1914 
1915         if (strcmp(nd->model, "ibmveth") == 0) {
1916             spapr_vlan_create(spapr->vio_bus, nd);
1917         } else {
1918             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1919         }
1920     }
1921 
1922     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1923         spapr_vscsi_create(spapr->vio_bus);
1924     }
1925 
1926     /* Graphics */
1927     if (spapr_vga_init(phb->bus, &error_fatal)) {
1928         spapr->has_graphics = true;
1929         machine->usb |= defaults_enabled() && !machine->usb_disabled;
1930     }
1931 
1932     if (machine->usb) {
1933         if (smc->use_ohci_by_default) {
1934             pci_create_simple(phb->bus, -1, "pci-ohci");
1935         } else {
1936             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
1937         }
1938 
1939         if (spapr->has_graphics) {
1940             USBBus *usb_bus = usb_bus_find(-1);
1941 
1942             usb_create_simple(usb_bus, "usb-kbd");
1943             usb_create_simple(usb_bus, "usb-mouse");
1944         }
1945     }
1946 
1947     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1948         error_report(
1949             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
1950             MIN_RMA_SLOF);
1951         exit(1);
1952     }
1953 
1954     if (kernel_filename) {
1955         uint64_t lowaddr = 0;
1956 
1957         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1958                                NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE,
1959                                0, 0);
1960         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1961             kernel_size = load_elf(kernel_filename,
1962                                    translate_kernel_address, NULL,
1963                                    NULL, &lowaddr, NULL, 0, PPC_ELF_MACHINE,
1964                                    0, 0);
1965             kernel_le = kernel_size > 0;
1966         }
1967         if (kernel_size < 0) {
1968             error_report("error loading %s: %s",
1969                          kernel_filename, load_elf_strerror(kernel_size));
1970             exit(1);
1971         }
1972 
1973         /* load initrd */
1974         if (initrd_filename) {
1975             /* Try to locate the initrd in the gap between the kernel
1976              * and the firmware. Add a bit of space just in case
1977              */
1978             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1979             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1980                                               load_limit - initrd_base);
1981             if (initrd_size < 0) {
1982                 error_report("could not load initial ram disk '%s'",
1983                              initrd_filename);
1984                 exit(1);
1985             }
1986         } else {
1987             initrd_base = 0;
1988             initrd_size = 0;
1989         }
1990     }
1991 
1992     if (bios_name == NULL) {
1993         bios_name = FW_FILE_NAME;
1994     }
1995     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1996     if (!filename) {
1997         error_report("Could not find LPAR firmware '%s'", bios_name);
1998         exit(1);
1999     }
2000     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2001     if (fw_size <= 0) {
2002         error_report("Could not load LPAR firmware '%s'", filename);
2003         exit(1);
2004     }
2005     g_free(filename);
2006 
2007     /* FIXME: Should register things through the MachineState's qdev
2008      * interface, this is a legacy from the sPAPREnvironment structure
2009      * which predated MachineState but had a similar function */
2010     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2011     register_savevm_live(NULL, "spapr/htab", -1, 1,
2012                          &savevm_htab_handlers, spapr);
2013 
2014     /* Prepare the device tree */
2015     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
2016                                             kernel_size, kernel_le,
2017                                             kernel_cmdline,
2018                                             spapr->check_exception_irq);
2019     assert(spapr->fdt_skel != NULL);
2020 
2021     /* used by RTAS */
2022     QTAILQ_INIT(&spapr->ccs_list);
2023     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2024 
2025     qemu_register_boot_set(spapr_boot_set, spapr);
2026 }
2027 
2028 static int spapr_kvm_type(const char *vm_type)
2029 {
2030     if (!vm_type) {
2031         return 0;
2032     }
2033 
2034     if (!strcmp(vm_type, "HV")) {
2035         return 1;
2036     }
2037 
2038     if (!strcmp(vm_type, "PR")) {
2039         return 2;
2040     }
2041 
2042     error_report("Unknown kvm-type specified '%s'", vm_type);
2043     exit(1);
2044 }
2045 
2046 /*
2047  * Implementation of an interface to adjust firmware path
2048  * for the bootindex property handling.
2049  */
2050 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2051                                    DeviceState *dev)
2052 {
2053 #define CAST(type, obj, name) \
2054     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2055     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2056     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2057 
2058     if (d) {
2059         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2060         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2061         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2062 
2063         if (spapr) {
2064             /*
2065              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2066              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2067              * in the top 16 bits of the 64-bit LUN
2068              */
2069             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2070             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2071                                    (uint64_t)id << 48);
2072         } else if (virtio) {
2073             /*
2074              * We use SRP luns of the form 01000000 | (target << 8) | lun
2075              * in the top 32 bits of the 64-bit LUN
2076              * Note: the quote above is from SLOF and it is wrong,
2077              * the actual binding is:
2078              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2079              */
2080             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2081             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2082                                    (uint64_t)id << 32);
2083         } else if (usb) {
2084             /*
2085              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2086              * in the top 32 bits of the 64-bit LUN
2087              */
2088             unsigned usb_port = atoi(usb->port->path);
2089             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2090             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2091                                    (uint64_t)id << 32);
2092         }
2093     }
2094 
2095     if (phb) {
2096         /* Replace "pci" with "pci@800000020000000" */
2097         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2098     }
2099 
2100     return NULL;
2101 }
2102 
2103 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2104 {
2105     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2106 
2107     return g_strdup(spapr->kvm_type);
2108 }
2109 
2110 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2111 {
2112     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2113 
2114     g_free(spapr->kvm_type);
2115     spapr->kvm_type = g_strdup(value);
2116 }
2117 
2118 static void spapr_machine_initfn(Object *obj)
2119 {
2120     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2121 
2122     spapr->htab_fd = -1;
2123     object_property_add_str(obj, "kvm-type",
2124                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2125     object_property_set_description(obj, "kvm-type",
2126                                     "Specifies the KVM virtualization mode (HV, PR)",
2127                                     NULL);
2128 }
2129 
2130 static void spapr_machine_finalizefn(Object *obj)
2131 {
2132     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2133 
2134     g_free(spapr->kvm_type);
2135 }
2136 
2137 static void ppc_cpu_do_nmi_on_cpu(CPUState *cs, void *arg)
2138 {
2139     cpu_synchronize_state(cs);
2140     ppc_cpu_do_system_reset(cs);
2141 }
2142 
2143 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2144 {
2145     CPUState *cs;
2146 
2147     CPU_FOREACH(cs) {
2148         async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, NULL);
2149     }
2150 }
2151 
2152 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr, uint64_t size,
2153                            uint32_t node, Error **errp)
2154 {
2155     sPAPRDRConnector *drc;
2156     sPAPRDRConnectorClass *drck;
2157     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2158     int i, fdt_offset, fdt_size;
2159     void *fdt;
2160 
2161     for (i = 0; i < nr_lmbs; i++) {
2162         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2163                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2164         g_assert(drc);
2165 
2166         fdt = create_device_tree(&fdt_size);
2167         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2168                                                 SPAPR_MEMORY_BLOCK_SIZE);
2169 
2170         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2171         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2172         addr += SPAPR_MEMORY_BLOCK_SIZE;
2173     }
2174     /* send hotplug notification to the
2175      * guest only in case of hotplugged memory
2176      */
2177     if (dev->hotplugged) {
2178        spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB, nr_lmbs);
2179     }
2180 }
2181 
2182 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2183                               uint32_t node, Error **errp)
2184 {
2185     Error *local_err = NULL;
2186     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2187     PCDIMMDevice *dimm = PC_DIMM(dev);
2188     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2189     MemoryRegion *mr = ddc->get_memory_region(dimm);
2190     uint64_t align = memory_region_get_alignment(mr);
2191     uint64_t size = memory_region_size(mr);
2192     uint64_t addr;
2193 
2194     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2195         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2196                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2197         goto out;
2198     }
2199 
2200     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2201     if (local_err) {
2202         goto out;
2203     }
2204 
2205     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2206     if (local_err) {
2207         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2208         goto out;
2209     }
2210 
2211     spapr_add_lmbs(dev, addr, size, node, &error_abort);
2212 
2213 out:
2214     error_propagate(errp, local_err);
2215 }
2216 
2217 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2218                                     sPAPRMachineState *spapr)
2219 {
2220     PowerPCCPU *cpu = POWERPC_CPU(cs);
2221     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2222     int id = ppc_get_vcpu_dt_id(cpu);
2223     void *fdt;
2224     int offset, fdt_size;
2225     char *nodename;
2226 
2227     fdt = create_device_tree(&fdt_size);
2228     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2229     offset = fdt_add_subnode(fdt, 0, nodename);
2230 
2231     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2232     g_free(nodename);
2233 
2234     *fdt_offset = offset;
2235     return fdt;
2236 }
2237 
2238 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2239                                       DeviceState *dev, Error **errp)
2240 {
2241     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2242 
2243     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2244         int node;
2245 
2246         if (!smc->dr_lmb_enabled) {
2247             error_setg(errp, "Memory hotplug not supported for this machine");
2248             return;
2249         }
2250         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2251         if (*errp) {
2252             return;
2253         }
2254         if (node < 0 || node >= MAX_NODES) {
2255             error_setg(errp, "Invaild node %d", node);
2256             return;
2257         }
2258 
2259         /*
2260          * Currently PowerPC kernel doesn't allow hot-adding memory to
2261          * memory-less node, but instead will silently add the memory
2262          * to the first node that has some memory. This causes two
2263          * unexpected behaviours for the user.
2264          *
2265          * - Memory gets hotplugged to a different node than what the user
2266          *   specified.
2267          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2268          *   to memory-less node, a reboot will set things accordingly
2269          *   and the previously hotplugged memory now ends in the right node.
2270          *   This appears as if some memory moved from one node to another.
2271          *
2272          * So until kernel starts supporting memory hotplug to memory-less
2273          * nodes, just prevent such attempts upfront in QEMU.
2274          */
2275         if (nb_numa_nodes && !numa_info[node].node_mem) {
2276             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2277                        node);
2278             return;
2279         }
2280 
2281         spapr_memory_plug(hotplug_dev, dev, node, errp);
2282     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2283         spapr_core_plug(hotplug_dev, dev, errp);
2284     }
2285 }
2286 
2287 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2288                                       DeviceState *dev, Error **errp)
2289 {
2290     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2291 
2292     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2293         error_setg(errp, "Memory hot unplug not supported by sPAPR");
2294     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2295         if (!mc->query_hotpluggable_cpus) {
2296             error_setg(errp, "CPU hot unplug not supported on this machine");
2297             return;
2298         }
2299         spapr_core_unplug(hotplug_dev, dev, errp);
2300     }
2301 }
2302 
2303 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2304                                           DeviceState *dev, Error **errp)
2305 {
2306     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2307         spapr_core_pre_plug(hotplug_dev, dev, errp);
2308     }
2309 }
2310 
2311 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2312                                                  DeviceState *dev)
2313 {
2314     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2315         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2316         return HOTPLUG_HANDLER(machine);
2317     }
2318     return NULL;
2319 }
2320 
2321 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2322 {
2323     /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2324      * socket means much for the paravirtualized PAPR platform) */
2325     return cpu_index / smp_threads / smp_cores;
2326 }
2327 
2328 static HotpluggableCPUList *spapr_query_hotpluggable_cpus(MachineState *machine)
2329 {
2330     int i;
2331     HotpluggableCPUList *head = NULL;
2332     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2333     int spapr_max_cores = max_cpus / smp_threads;
2334 
2335     for (i = 0; i < spapr_max_cores; i++) {
2336         HotpluggableCPUList *list_item = g_new0(typeof(*list_item), 1);
2337         HotpluggableCPU *cpu_item = g_new0(typeof(*cpu_item), 1);
2338         CpuInstanceProperties *cpu_props = g_new0(typeof(*cpu_props), 1);
2339 
2340         cpu_item->type = spapr_get_cpu_core_type(machine->cpu_model);
2341         cpu_item->vcpus_count = smp_threads;
2342         cpu_props->has_core_id = true;
2343         cpu_props->core_id = i * smp_threads;
2344         /* TODO: add 'has_node/node' here to describe
2345            to which node core belongs */
2346 
2347         cpu_item->props = cpu_props;
2348         if (spapr->cores[i]) {
2349             cpu_item->has_qom_path = true;
2350             cpu_item->qom_path = object_get_canonical_path(spapr->cores[i]);
2351         }
2352         list_item->value = cpu_item;
2353         list_item->next = head;
2354         head = list_item;
2355     }
2356     return head;
2357 }
2358 
2359 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2360 {
2361     MachineClass *mc = MACHINE_CLASS(oc);
2362     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2363     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2364     NMIClass *nc = NMI_CLASS(oc);
2365     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2366 
2367     mc->desc = "pSeries Logical Partition (PAPR compliant)";
2368 
2369     /*
2370      * We set up the default / latest behaviour here.  The class_init
2371      * functions for the specific versioned machine types can override
2372      * these details for backwards compatibility
2373      */
2374     mc->init = ppc_spapr_init;
2375     mc->reset = ppc_spapr_reset;
2376     mc->block_default_type = IF_SCSI;
2377     mc->max_cpus = MAX_CPUMASK_BITS;
2378     mc->no_parallel = 1;
2379     mc->default_boot_order = "";
2380     mc->default_ram_size = 512 * M_BYTE;
2381     mc->kvm_type = spapr_kvm_type;
2382     mc->has_dynamic_sysbus = true;
2383     mc->pci_allow_0_address = true;
2384     mc->get_hotplug_handler = spapr_get_hotplug_handler;
2385     hc->pre_plug = spapr_machine_device_pre_plug;
2386     hc->plug = spapr_machine_device_plug;
2387     hc->unplug = spapr_machine_device_unplug;
2388     mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
2389 
2390     smc->dr_lmb_enabled = true;
2391     mc->query_hotpluggable_cpus = spapr_query_hotpluggable_cpus;
2392     fwc->get_dev_path = spapr_get_fw_dev_path;
2393     nc->nmi_monitor_handler = spapr_nmi;
2394 }
2395 
2396 static const TypeInfo spapr_machine_info = {
2397     .name          = TYPE_SPAPR_MACHINE,
2398     .parent        = TYPE_MACHINE,
2399     .abstract      = true,
2400     .instance_size = sizeof(sPAPRMachineState),
2401     .instance_init = spapr_machine_initfn,
2402     .instance_finalize = spapr_machine_finalizefn,
2403     .class_size    = sizeof(sPAPRMachineClass),
2404     .class_init    = spapr_machine_class_init,
2405     .interfaces = (InterfaceInfo[]) {
2406         { TYPE_FW_PATH_PROVIDER },
2407         { TYPE_NMI },
2408         { TYPE_HOTPLUG_HANDLER },
2409         { }
2410     },
2411 };
2412 
2413 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
2414     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
2415                                                     void *data)      \
2416     {                                                                \
2417         MachineClass *mc = MACHINE_CLASS(oc);                        \
2418         spapr_machine_##suffix##_class_options(mc);                  \
2419         if (latest) {                                                \
2420             mc->alias = "pseries";                                   \
2421             mc->is_default = 1;                                      \
2422         }                                                            \
2423     }                                                                \
2424     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
2425     {                                                                \
2426         MachineState *machine = MACHINE(obj);                        \
2427         spapr_machine_##suffix##_instance_options(machine);          \
2428     }                                                                \
2429     static const TypeInfo spapr_machine_##suffix##_info = {          \
2430         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
2431         .parent = TYPE_SPAPR_MACHINE,                                \
2432         .class_init = spapr_machine_##suffix##_class_init,           \
2433         .instance_init = spapr_machine_##suffix##_instance_init,     \
2434     };                                                               \
2435     static void spapr_machine_register_##suffix(void)                \
2436     {                                                                \
2437         type_register(&spapr_machine_##suffix##_info);               \
2438     }                                                                \
2439     type_init(spapr_machine_register_##suffix)
2440 
2441 /*
2442  * pseries-2.7
2443  */
2444 static void spapr_machine_2_7_instance_options(MachineState *machine)
2445 {
2446 }
2447 
2448 static void spapr_machine_2_7_class_options(MachineClass *mc)
2449 {
2450     /* Defaults for the latest behaviour inherited from the base class */
2451 }
2452 
2453 DEFINE_SPAPR_MACHINE(2_7, "2.7", true);
2454 
2455 /*
2456  * pseries-2.6
2457  */
2458 #define SPAPR_COMPAT_2_6 \
2459     HW_COMPAT_2_6 \
2460     { \
2461         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2462         .property = "ddw",\
2463         .value    = stringify(off),\
2464     },
2465 
2466 static void spapr_machine_2_6_instance_options(MachineState *machine)
2467 {
2468 }
2469 
2470 static void spapr_machine_2_6_class_options(MachineClass *mc)
2471 {
2472     spapr_machine_2_7_class_options(mc);
2473     mc->query_hotpluggable_cpus = NULL;
2474     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
2475 }
2476 
2477 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
2478 
2479 /*
2480  * pseries-2.5
2481  */
2482 #define SPAPR_COMPAT_2_5 \
2483     HW_COMPAT_2_5 \
2484     { \
2485         .driver   = "spapr-vlan", \
2486         .property = "use-rx-buffer-pools", \
2487         .value    = "off", \
2488     },
2489 
2490 static void spapr_machine_2_5_instance_options(MachineState *machine)
2491 {
2492 }
2493 
2494 static void spapr_machine_2_5_class_options(MachineClass *mc)
2495 {
2496     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2497 
2498     spapr_machine_2_6_class_options(mc);
2499     smc->use_ohci_by_default = true;
2500     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
2501 }
2502 
2503 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
2504 
2505 /*
2506  * pseries-2.4
2507  */
2508 #define SPAPR_COMPAT_2_4 \
2509         HW_COMPAT_2_4
2510 
2511 static void spapr_machine_2_4_instance_options(MachineState *machine)
2512 {
2513     spapr_machine_2_5_instance_options(machine);
2514 }
2515 
2516 static void spapr_machine_2_4_class_options(MachineClass *mc)
2517 {
2518     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2519 
2520     spapr_machine_2_5_class_options(mc);
2521     smc->dr_lmb_enabled = false;
2522     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
2523 }
2524 
2525 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
2526 
2527 /*
2528  * pseries-2.3
2529  */
2530 #define SPAPR_COMPAT_2_3 \
2531         HW_COMPAT_2_3 \
2532         {\
2533             .driver   = "spapr-pci-host-bridge",\
2534             .property = "dynamic-reconfiguration",\
2535             .value    = "off",\
2536         },
2537 
2538 static void spapr_machine_2_3_instance_options(MachineState *machine)
2539 {
2540     spapr_machine_2_4_instance_options(machine);
2541     savevm_skip_section_footers();
2542     global_state_set_optional();
2543     savevm_skip_configuration();
2544 }
2545 
2546 static void spapr_machine_2_3_class_options(MachineClass *mc)
2547 {
2548     spapr_machine_2_4_class_options(mc);
2549     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
2550 }
2551 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
2552 
2553 /*
2554  * pseries-2.2
2555  */
2556 
2557 #define SPAPR_COMPAT_2_2 \
2558         HW_COMPAT_2_2 \
2559         {\
2560             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2561             .property = "mem_win_size",\
2562             .value    = "0x20000000",\
2563         },
2564 
2565 static void spapr_machine_2_2_instance_options(MachineState *machine)
2566 {
2567     spapr_machine_2_3_instance_options(machine);
2568     machine->suppress_vmdesc = true;
2569 }
2570 
2571 static void spapr_machine_2_2_class_options(MachineClass *mc)
2572 {
2573     spapr_machine_2_3_class_options(mc);
2574     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
2575 }
2576 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
2577 
2578 /*
2579  * pseries-2.1
2580  */
2581 #define SPAPR_COMPAT_2_1 \
2582         HW_COMPAT_2_1
2583 
2584 static void spapr_machine_2_1_instance_options(MachineState *machine)
2585 {
2586     spapr_machine_2_2_instance_options(machine);
2587 }
2588 
2589 static void spapr_machine_2_1_class_options(MachineClass *mc)
2590 {
2591     spapr_machine_2_2_class_options(mc);
2592     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
2593 }
2594 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
2595 
2596 static void spapr_machine_register_types(void)
2597 {
2598     type_register_static(&spapr_machine_info);
2599 }
2600 
2601 type_init(spapr_machine_register_types)
2602