xref: /openbmc/qemu/hw/ppc/spapr.c (revision 2bfe11c8)
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 void 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 
1128 static void ppc_spapr_reset(void)
1129 {
1130     MachineState *machine = MACHINE(qdev_get_machine());
1131     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1132     PowerPCCPU *first_ppc_cpu;
1133     uint32_t rtas_limit;
1134 
1135     /* Check for unknown sysbus devices */
1136     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1137 
1138     /* Allocate and/or reset the hash page table */
1139     spapr_reallocate_hpt(spapr,
1140                          spapr_hpt_shift_for_ramsize(machine->maxram_size),
1141                          &error_fatal);
1142 
1143     /* Update the RMA size if necessary */
1144     if (spapr->vrma_adjust) {
1145         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1146                                           spapr->htab_shift);
1147     }
1148 
1149     qemu_devices_reset();
1150 
1151     /*
1152      * We place the device tree and RTAS just below either the top of the RMA,
1153      * or just below 2GB, whichever is lowere, so that it can be
1154      * processed with 32-bit real mode code if necessary
1155      */
1156     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1157     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1158     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1159 
1160     /* Load the fdt */
1161     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
1162                        spapr->rtas_size);
1163 
1164     /* Copy RTAS over */
1165     cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
1166                               spapr->rtas_size);
1167 
1168     /* Set up the entry state */
1169     first_ppc_cpu = POWERPC_CPU(first_cpu);
1170     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
1171     first_ppc_cpu->env.gpr[5] = 0;
1172     first_cpu->halted = 0;
1173     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1174 
1175 }
1176 
1177 static void spapr_create_nvram(sPAPRMachineState *spapr)
1178 {
1179     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1180     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1181 
1182     if (dinfo) {
1183         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1184                             &error_fatal);
1185     }
1186 
1187     qdev_init_nofail(dev);
1188 
1189     spapr->nvram = (struct sPAPRNVRAM *)dev;
1190 }
1191 
1192 static void spapr_rtc_create(sPAPRMachineState *spapr)
1193 {
1194     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1195 
1196     qdev_init_nofail(dev);
1197     spapr->rtc = dev;
1198 
1199     object_property_add_alias(qdev_get_machine(), "rtc-time",
1200                               OBJECT(spapr->rtc), "date", NULL);
1201 }
1202 
1203 /* Returns whether we want to use VGA or not */
1204 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1205 {
1206     switch (vga_interface_type) {
1207     case VGA_NONE:
1208         return false;
1209     case VGA_DEVICE:
1210         return true;
1211     case VGA_STD:
1212     case VGA_VIRTIO:
1213         return pci_vga_init(pci_bus) != NULL;
1214     default:
1215         error_setg(errp,
1216                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1217         return false;
1218     }
1219 }
1220 
1221 static int spapr_post_load(void *opaque, int version_id)
1222 {
1223     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1224     int err = 0;
1225 
1226     /* In earlier versions, there was no separate qdev for the PAPR
1227      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1228      * So when migrating from those versions, poke the incoming offset
1229      * value into the RTC device */
1230     if (version_id < 3) {
1231         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1232     }
1233 
1234     return err;
1235 }
1236 
1237 static bool version_before_3(void *opaque, int version_id)
1238 {
1239     return version_id < 3;
1240 }
1241 
1242 static const VMStateDescription vmstate_spapr = {
1243     .name = "spapr",
1244     .version_id = 3,
1245     .minimum_version_id = 1,
1246     .post_load = spapr_post_load,
1247     .fields = (VMStateField[]) {
1248         /* used to be @next_irq */
1249         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1250 
1251         /* RTC offset */
1252         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1253 
1254         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1255         VMSTATE_END_OF_LIST()
1256     },
1257 };
1258 
1259 static int htab_save_setup(QEMUFile *f, void *opaque)
1260 {
1261     sPAPRMachineState *spapr = opaque;
1262 
1263     /* "Iteration" header */
1264     qemu_put_be32(f, spapr->htab_shift);
1265 
1266     if (spapr->htab) {
1267         spapr->htab_save_index = 0;
1268         spapr->htab_first_pass = true;
1269     } else {
1270         assert(kvm_enabled());
1271     }
1272 
1273 
1274     return 0;
1275 }
1276 
1277 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1278                                  int64_t max_ns)
1279 {
1280     bool has_timeout = max_ns != -1;
1281     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1282     int index = spapr->htab_save_index;
1283     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1284 
1285     assert(spapr->htab_first_pass);
1286 
1287     do {
1288         int chunkstart;
1289 
1290         /* Consume invalid HPTEs */
1291         while ((index < htabslots)
1292                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1293             index++;
1294             CLEAN_HPTE(HPTE(spapr->htab, index));
1295         }
1296 
1297         /* Consume valid HPTEs */
1298         chunkstart = index;
1299         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1300                && HPTE_VALID(HPTE(spapr->htab, index))) {
1301             index++;
1302             CLEAN_HPTE(HPTE(spapr->htab, index));
1303         }
1304 
1305         if (index > chunkstart) {
1306             int n_valid = index - chunkstart;
1307 
1308             qemu_put_be32(f, chunkstart);
1309             qemu_put_be16(f, n_valid);
1310             qemu_put_be16(f, 0);
1311             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1312                             HASH_PTE_SIZE_64 * n_valid);
1313 
1314             if (has_timeout &&
1315                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1316                 break;
1317             }
1318         }
1319     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1320 
1321     if (index >= htabslots) {
1322         assert(index == htabslots);
1323         index = 0;
1324         spapr->htab_first_pass = false;
1325     }
1326     spapr->htab_save_index = index;
1327 }
1328 
1329 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1330                                 int64_t max_ns)
1331 {
1332     bool final = max_ns < 0;
1333     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1334     int examined = 0, sent = 0;
1335     int index = spapr->htab_save_index;
1336     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1337 
1338     assert(!spapr->htab_first_pass);
1339 
1340     do {
1341         int chunkstart, invalidstart;
1342 
1343         /* Consume non-dirty HPTEs */
1344         while ((index < htabslots)
1345                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1346             index++;
1347             examined++;
1348         }
1349 
1350         chunkstart = index;
1351         /* Consume valid dirty HPTEs */
1352         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1353                && HPTE_DIRTY(HPTE(spapr->htab, index))
1354                && HPTE_VALID(HPTE(spapr->htab, index))) {
1355             CLEAN_HPTE(HPTE(spapr->htab, index));
1356             index++;
1357             examined++;
1358         }
1359 
1360         invalidstart = index;
1361         /* Consume invalid dirty HPTEs */
1362         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1363                && HPTE_DIRTY(HPTE(spapr->htab, index))
1364                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1365             CLEAN_HPTE(HPTE(spapr->htab, index));
1366             index++;
1367             examined++;
1368         }
1369 
1370         if (index > chunkstart) {
1371             int n_valid = invalidstart - chunkstart;
1372             int n_invalid = index - invalidstart;
1373 
1374             qemu_put_be32(f, chunkstart);
1375             qemu_put_be16(f, n_valid);
1376             qemu_put_be16(f, n_invalid);
1377             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1378                             HASH_PTE_SIZE_64 * n_valid);
1379             sent += index - chunkstart;
1380 
1381             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1382                 break;
1383             }
1384         }
1385 
1386         if (examined >= htabslots) {
1387             break;
1388         }
1389 
1390         if (index >= htabslots) {
1391             assert(index == htabslots);
1392             index = 0;
1393         }
1394     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1395 
1396     if (index >= htabslots) {
1397         assert(index == htabslots);
1398         index = 0;
1399     }
1400 
1401     spapr->htab_save_index = index;
1402 
1403     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1404 }
1405 
1406 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1407 #define MAX_KVM_BUF_SIZE    2048
1408 
1409 static int htab_save_iterate(QEMUFile *f, void *opaque)
1410 {
1411     sPAPRMachineState *spapr = opaque;
1412     int fd;
1413     int rc = 0;
1414 
1415     /* Iteration header */
1416     qemu_put_be32(f, 0);
1417 
1418     if (!spapr->htab) {
1419         assert(kvm_enabled());
1420 
1421         fd = get_htab_fd(spapr);
1422         if (fd < 0) {
1423             return fd;
1424         }
1425 
1426         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1427         if (rc < 0) {
1428             return rc;
1429         }
1430     } else  if (spapr->htab_first_pass) {
1431         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1432     } else {
1433         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1434     }
1435 
1436     /* End marker */
1437     qemu_put_be32(f, 0);
1438     qemu_put_be16(f, 0);
1439     qemu_put_be16(f, 0);
1440 
1441     return rc;
1442 }
1443 
1444 static int htab_save_complete(QEMUFile *f, void *opaque)
1445 {
1446     sPAPRMachineState *spapr = opaque;
1447     int fd;
1448 
1449     /* Iteration header */
1450     qemu_put_be32(f, 0);
1451 
1452     if (!spapr->htab) {
1453         int rc;
1454 
1455         assert(kvm_enabled());
1456 
1457         fd = get_htab_fd(spapr);
1458         if (fd < 0) {
1459             return fd;
1460         }
1461 
1462         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1463         if (rc < 0) {
1464             return rc;
1465         }
1466     } else {
1467         if (spapr->htab_first_pass) {
1468             htab_save_first_pass(f, spapr, -1);
1469         }
1470         htab_save_later_pass(f, spapr, -1);
1471     }
1472 
1473     /* End marker */
1474     qemu_put_be32(f, 0);
1475     qemu_put_be16(f, 0);
1476     qemu_put_be16(f, 0);
1477 
1478     return 0;
1479 }
1480 
1481 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1482 {
1483     sPAPRMachineState *spapr = opaque;
1484     uint32_t section_hdr;
1485     int fd = -1;
1486 
1487     if (version_id < 1 || version_id > 1) {
1488         error_report("htab_load() bad version");
1489         return -EINVAL;
1490     }
1491 
1492     section_hdr = qemu_get_be32(f);
1493 
1494     if (section_hdr) {
1495         Error *local_err = NULL;
1496 
1497         /* First section gives the htab size */
1498         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1499         if (local_err) {
1500             error_report_err(local_err);
1501             return -EINVAL;
1502         }
1503         return 0;
1504     }
1505 
1506     if (!spapr->htab) {
1507         assert(kvm_enabled());
1508 
1509         fd = kvmppc_get_htab_fd(true);
1510         if (fd < 0) {
1511             error_report("Unable to open fd to restore KVM hash table: %s",
1512                          strerror(errno));
1513         }
1514     }
1515 
1516     while (true) {
1517         uint32_t index;
1518         uint16_t n_valid, n_invalid;
1519 
1520         index = qemu_get_be32(f);
1521         n_valid = qemu_get_be16(f);
1522         n_invalid = qemu_get_be16(f);
1523 
1524         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1525             /* End of Stream */
1526             break;
1527         }
1528 
1529         if ((index + n_valid + n_invalid) >
1530             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1531             /* Bad index in stream */
1532             error_report(
1533                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1534                 index, n_valid, n_invalid, spapr->htab_shift);
1535             return -EINVAL;
1536         }
1537 
1538         if (spapr->htab) {
1539             if (n_valid) {
1540                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1541                                 HASH_PTE_SIZE_64 * n_valid);
1542             }
1543             if (n_invalid) {
1544                 memset(HPTE(spapr->htab, index + n_valid), 0,
1545                        HASH_PTE_SIZE_64 * n_invalid);
1546             }
1547         } else {
1548             int rc;
1549 
1550             assert(fd >= 0);
1551 
1552             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1553             if (rc < 0) {
1554                 return rc;
1555             }
1556         }
1557     }
1558 
1559     if (!spapr->htab) {
1560         assert(fd >= 0);
1561         close(fd);
1562     }
1563 
1564     return 0;
1565 }
1566 
1567 static void htab_cleanup(void *opaque)
1568 {
1569     sPAPRMachineState *spapr = opaque;
1570 
1571     close_htab_fd(spapr);
1572 }
1573 
1574 static SaveVMHandlers savevm_htab_handlers = {
1575     .save_live_setup = htab_save_setup,
1576     .save_live_iterate = htab_save_iterate,
1577     .save_live_complete_precopy = htab_save_complete,
1578     .cleanup = htab_cleanup,
1579     .load_state = htab_load,
1580 };
1581 
1582 static void spapr_boot_set(void *opaque, const char *boot_device,
1583                            Error **errp)
1584 {
1585     MachineState *machine = MACHINE(qdev_get_machine());
1586     machine->boot_order = g_strdup(boot_device);
1587 }
1588 
1589 /*
1590  * Reset routine for LMB DR devices.
1591  *
1592  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1593  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1594  * when it walks all its children devices. LMB devices reset occurs
1595  * as part of spapr_ppc_reset().
1596  */
1597 static void spapr_drc_reset(void *opaque)
1598 {
1599     sPAPRDRConnector *drc = opaque;
1600     DeviceState *d = DEVICE(drc);
1601 
1602     if (d) {
1603         device_reset(d);
1604     }
1605 }
1606 
1607 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1608 {
1609     MachineState *machine = MACHINE(spapr);
1610     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1611     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1612     int i;
1613 
1614     for (i = 0; i < nr_lmbs; i++) {
1615         sPAPRDRConnector *drc;
1616         uint64_t addr;
1617 
1618         addr = i * lmb_size + spapr->hotplug_memory.base;
1619         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1620                                      addr/lmb_size);
1621         qemu_register_reset(spapr_drc_reset, drc);
1622     }
1623 }
1624 
1625 /*
1626  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1627  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1628  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1629  */
1630 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1631 {
1632     int i;
1633 
1634     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1635         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1636                    " is not aligned to %llu MiB",
1637                    machine->ram_size,
1638                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1639         return;
1640     }
1641 
1642     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1643         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1644                    " is not aligned to %llu MiB",
1645                    machine->ram_size,
1646                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1647         return;
1648     }
1649 
1650     for (i = 0; i < nb_numa_nodes; i++) {
1651         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1652             error_setg(errp,
1653                        "Node %d memory size 0x%" PRIx64
1654                        " is not aligned to %llu MiB",
1655                        i, numa_info[i].node_mem,
1656                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1657             return;
1658         }
1659     }
1660 }
1661 
1662 /* pSeries LPAR / sPAPR hardware init */
1663 static void ppc_spapr_init(MachineState *machine)
1664 {
1665     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1666     MachineClass *mc = MACHINE_GET_CLASS(machine);
1667     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1668     const char *kernel_filename = machine->kernel_filename;
1669     const char *kernel_cmdline = machine->kernel_cmdline;
1670     const char *initrd_filename = machine->initrd_filename;
1671     PCIHostState *phb;
1672     int i;
1673     MemoryRegion *sysmem = get_system_memory();
1674     MemoryRegion *ram = g_new(MemoryRegion, 1);
1675     MemoryRegion *rma_region;
1676     void *rma = NULL;
1677     hwaddr rma_alloc_size;
1678     hwaddr node0_size = spapr_node0_size();
1679     uint32_t initrd_base = 0;
1680     long kernel_size = 0, initrd_size = 0;
1681     long load_limit, fw_size;
1682     bool kernel_le = false;
1683     char *filename;
1684     int smt = kvmppc_smt_threads();
1685     int spapr_cores = smp_cpus / smp_threads;
1686     int spapr_max_cores = max_cpus / smp_threads;
1687 
1688     if (mc->query_hotpluggable_cpus) {
1689         if (smp_cpus % smp_threads) {
1690             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1691                          smp_cpus, smp_threads);
1692             exit(1);
1693         }
1694         if (max_cpus % smp_threads) {
1695             error_report("max_cpus (%u) must be multiple of threads (%u)",
1696                          max_cpus, smp_threads);
1697             exit(1);
1698         }
1699     }
1700 
1701     msi_nonbroken = true;
1702 
1703     QLIST_INIT(&spapr->phbs);
1704 
1705     cpu_ppc_hypercall = emulate_spapr_hypercall;
1706 
1707     /* Allocate RMA if necessary */
1708     rma_alloc_size = kvmppc_alloc_rma(&rma);
1709 
1710     if (rma_alloc_size == -1) {
1711         error_report("Unable to create RMA");
1712         exit(1);
1713     }
1714 
1715     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1716         spapr->rma_size = rma_alloc_size;
1717     } else {
1718         spapr->rma_size = node0_size;
1719 
1720         /* With KVM, we don't actually know whether KVM supports an
1721          * unbounded RMA (PR KVM) or is limited by the hash table size
1722          * (HV KVM using VRMA), so we always assume the latter
1723          *
1724          * In that case, we also limit the initial allocations for RTAS
1725          * etc... to 256M since we have no way to know what the VRMA size
1726          * is going to be as it depends on the size of the hash table
1727          * isn't determined yet.
1728          */
1729         if (kvm_enabled()) {
1730             spapr->vrma_adjust = 1;
1731             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1732         }
1733 
1734         /* Actually we don't support unbounded RMA anymore since we
1735          * added proper emulation of HV mode. The max we can get is
1736          * 16G which also happens to be what we configure for PAPR
1737          * mode so make sure we don't do anything bigger than that
1738          */
1739         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1740     }
1741 
1742     if (spapr->rma_size > node0_size) {
1743         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1744                      spapr->rma_size);
1745         exit(1);
1746     }
1747 
1748     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1749     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1750 
1751     /* Set up Interrupt Controller before we create the VCPUs */
1752     spapr->xics = xics_system_init(machine,
1753                                    DIV_ROUND_UP(max_cpus * smt, smp_threads),
1754                                    XICS_IRQS_SPAPR, &error_fatal);
1755 
1756     if (smc->dr_lmb_enabled) {
1757         spapr_validate_node_memory(machine, &error_fatal);
1758     }
1759 
1760     /* init CPUs */
1761     if (machine->cpu_model == NULL) {
1762         machine->cpu_model = kvm_enabled() ? "host" : "POWER7";
1763     }
1764 
1765     ppc_cpu_parse_features(machine->cpu_model);
1766 
1767     if (mc->query_hotpluggable_cpus) {
1768         char *type = spapr_get_cpu_core_type(machine->cpu_model);
1769 
1770         if (type == NULL) {
1771             error_report("Unable to find sPAPR CPU Core definition");
1772             exit(1);
1773         }
1774 
1775         spapr->cores = g_new0(Object *, spapr_max_cores);
1776         for (i = 0; i < spapr_max_cores; i++) {
1777             int core_id = i * smp_threads;
1778             sPAPRDRConnector *drc =
1779                 spapr_dr_connector_new(OBJECT(spapr),
1780                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
1781                                        (core_id / smp_threads) * smt);
1782 
1783             qemu_register_reset(spapr_drc_reset, drc);
1784 
1785             if (i < spapr_cores) {
1786                 Object *core  = object_new(type);
1787                 object_property_set_int(core, smp_threads, "nr-threads",
1788                                         &error_fatal);
1789                 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1790                                         &error_fatal);
1791                 object_property_set_bool(core, true, "realized", &error_fatal);
1792             }
1793         }
1794         g_free(type);
1795     } else {
1796         for (i = 0; i < smp_cpus; i++) {
1797             PowerPCCPU *cpu = cpu_ppc_init(machine->cpu_model);
1798             if (cpu == NULL) {
1799                 error_report("Unable to find PowerPC CPU definition");
1800                 exit(1);
1801             }
1802             spapr_cpu_init(spapr, cpu, &error_fatal);
1803        }
1804     }
1805 
1806     if (kvm_enabled()) {
1807         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1808         kvmppc_enable_logical_ci_hcalls();
1809         kvmppc_enable_set_mode_hcall();
1810 
1811         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
1812         kvmppc_enable_clear_ref_mod_hcalls();
1813     }
1814 
1815     /* allocate RAM */
1816     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1817                                          machine->ram_size);
1818     memory_region_add_subregion(sysmem, 0, ram);
1819 
1820     if (rma_alloc_size && rma) {
1821         rma_region = g_new(MemoryRegion, 1);
1822         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1823                                    rma_alloc_size, rma);
1824         vmstate_register_ram_global(rma_region);
1825         memory_region_add_subregion(sysmem, 0, rma_region);
1826     }
1827 
1828     /* initialize hotplug memory address space */
1829     if (machine->ram_size < machine->maxram_size) {
1830         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
1831         /*
1832          * Limit the number of hotpluggable memory slots to half the number
1833          * slots that KVM supports, leaving the other half for PCI and other
1834          * devices. However ensure that number of slots doesn't drop below 32.
1835          */
1836         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
1837                            SPAPR_MAX_RAM_SLOTS;
1838 
1839         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
1840             max_memslots = SPAPR_MAX_RAM_SLOTS;
1841         }
1842         if (machine->ram_slots > max_memslots) {
1843             error_report("Specified number of memory slots %"
1844                          PRIu64" exceeds max supported %d",
1845                          machine->ram_slots, max_memslots);
1846             exit(1);
1847         }
1848 
1849         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
1850                                               SPAPR_HOTPLUG_MEM_ALIGN);
1851         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
1852                            "hotplug-memory", hotplug_mem_size);
1853         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
1854                                     &spapr->hotplug_memory.mr);
1855     }
1856 
1857     if (smc->dr_lmb_enabled) {
1858         spapr_create_lmb_dr_connectors(spapr);
1859     }
1860 
1861     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1862     if (!filename) {
1863         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1864         exit(1);
1865     }
1866     spapr->rtas_size = get_image_size(filename);
1867     if (spapr->rtas_size < 0) {
1868         error_report("Could not get size of LPAR rtas '%s'", filename);
1869         exit(1);
1870     }
1871     spapr->rtas_blob = g_malloc(spapr->rtas_size);
1872     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1873         error_report("Could not load LPAR rtas '%s'", filename);
1874         exit(1);
1875     }
1876     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1877         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1878                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1879         exit(1);
1880     }
1881     g_free(filename);
1882 
1883     /* Set up EPOW events infrastructure */
1884     spapr_events_init(spapr);
1885 
1886     /* Set up the RTC RTAS interfaces */
1887     spapr_rtc_create(spapr);
1888 
1889     /* Set up VIO bus */
1890     spapr->vio_bus = spapr_vio_bus_init();
1891 
1892     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1893         if (serial_hds[i]) {
1894             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1895         }
1896     }
1897 
1898     /* We always have at least the nvram device on VIO */
1899     spapr_create_nvram(spapr);
1900 
1901     /* Set up PCI */
1902     spapr_pci_rtas_init();
1903 
1904     phb = spapr_create_phb(spapr, 0);
1905 
1906     for (i = 0; i < nb_nics; i++) {
1907         NICInfo *nd = &nd_table[i];
1908 
1909         if (!nd->model) {
1910             nd->model = g_strdup("ibmveth");
1911         }
1912 
1913         if (strcmp(nd->model, "ibmveth") == 0) {
1914             spapr_vlan_create(spapr->vio_bus, nd);
1915         } else {
1916             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1917         }
1918     }
1919 
1920     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1921         spapr_vscsi_create(spapr->vio_bus);
1922     }
1923 
1924     /* Graphics */
1925     if (spapr_vga_init(phb->bus, &error_fatal)) {
1926         spapr->has_graphics = true;
1927         machine->usb |= defaults_enabled() && !machine->usb_disabled;
1928     }
1929 
1930     if (machine->usb) {
1931         if (smc->use_ohci_by_default) {
1932             pci_create_simple(phb->bus, -1, "pci-ohci");
1933         } else {
1934             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
1935         }
1936 
1937         if (spapr->has_graphics) {
1938             USBBus *usb_bus = usb_bus_find(-1);
1939 
1940             usb_create_simple(usb_bus, "usb-kbd");
1941             usb_create_simple(usb_bus, "usb-mouse");
1942         }
1943     }
1944 
1945     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1946         error_report(
1947             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
1948             MIN_RMA_SLOF);
1949         exit(1);
1950     }
1951 
1952     if (kernel_filename) {
1953         uint64_t lowaddr = 0;
1954 
1955         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1956                                NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE,
1957                                0, 0);
1958         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1959             kernel_size = load_elf(kernel_filename,
1960                                    translate_kernel_address, NULL,
1961                                    NULL, &lowaddr, NULL, 0, PPC_ELF_MACHINE,
1962                                    0, 0);
1963             kernel_le = kernel_size > 0;
1964         }
1965         if (kernel_size < 0) {
1966             error_report("error loading %s: %s",
1967                          kernel_filename, load_elf_strerror(kernel_size));
1968             exit(1);
1969         }
1970 
1971         /* load initrd */
1972         if (initrd_filename) {
1973             /* Try to locate the initrd in the gap between the kernel
1974              * and the firmware. Add a bit of space just in case
1975              */
1976             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1977             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1978                                               load_limit - initrd_base);
1979             if (initrd_size < 0) {
1980                 error_report("could not load initial ram disk '%s'",
1981                              initrd_filename);
1982                 exit(1);
1983             }
1984         } else {
1985             initrd_base = 0;
1986             initrd_size = 0;
1987         }
1988     }
1989 
1990     if (bios_name == NULL) {
1991         bios_name = FW_FILE_NAME;
1992     }
1993     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1994     if (!filename) {
1995         error_report("Could not find LPAR firmware '%s'", bios_name);
1996         exit(1);
1997     }
1998     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1999     if (fw_size <= 0) {
2000         error_report("Could not load LPAR firmware '%s'", filename);
2001         exit(1);
2002     }
2003     g_free(filename);
2004 
2005     /* FIXME: Should register things through the MachineState's qdev
2006      * interface, this is a legacy from the sPAPREnvironment structure
2007      * which predated MachineState but had a similar function */
2008     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2009     register_savevm_live(NULL, "spapr/htab", -1, 1,
2010                          &savevm_htab_handlers, spapr);
2011 
2012     /* Prepare the device tree */
2013     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
2014                                             kernel_size, kernel_le,
2015                                             kernel_cmdline,
2016                                             spapr->check_exception_irq);
2017     assert(spapr->fdt_skel != NULL);
2018 
2019     /* used by RTAS */
2020     QTAILQ_INIT(&spapr->ccs_list);
2021     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2022 
2023     qemu_register_boot_set(spapr_boot_set, spapr);
2024 }
2025 
2026 static int spapr_kvm_type(const char *vm_type)
2027 {
2028     if (!vm_type) {
2029         return 0;
2030     }
2031 
2032     if (!strcmp(vm_type, "HV")) {
2033         return 1;
2034     }
2035 
2036     if (!strcmp(vm_type, "PR")) {
2037         return 2;
2038     }
2039 
2040     error_report("Unknown kvm-type specified '%s'", vm_type);
2041     exit(1);
2042 }
2043 
2044 /*
2045  * Implementation of an interface to adjust firmware path
2046  * for the bootindex property handling.
2047  */
2048 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2049                                    DeviceState *dev)
2050 {
2051 #define CAST(type, obj, name) \
2052     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2053     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2054     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2055 
2056     if (d) {
2057         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2058         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2059         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2060 
2061         if (spapr) {
2062             /*
2063              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2064              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2065              * in the top 16 bits of the 64-bit LUN
2066              */
2067             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2068             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2069                                    (uint64_t)id << 48);
2070         } else if (virtio) {
2071             /*
2072              * We use SRP luns of the form 01000000 | (target << 8) | lun
2073              * in the top 32 bits of the 64-bit LUN
2074              * Note: the quote above is from SLOF and it is wrong,
2075              * the actual binding is:
2076              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2077              */
2078             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2079             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2080                                    (uint64_t)id << 32);
2081         } else if (usb) {
2082             /*
2083              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2084              * in the top 32 bits of the 64-bit LUN
2085              */
2086             unsigned usb_port = atoi(usb->port->path);
2087             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2088             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2089                                    (uint64_t)id << 32);
2090         }
2091     }
2092 
2093     if (phb) {
2094         /* Replace "pci" with "pci@800000020000000" */
2095         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2096     }
2097 
2098     return NULL;
2099 }
2100 
2101 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2102 {
2103     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2104 
2105     return g_strdup(spapr->kvm_type);
2106 }
2107 
2108 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2109 {
2110     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2111 
2112     g_free(spapr->kvm_type);
2113     spapr->kvm_type = g_strdup(value);
2114 }
2115 
2116 static void spapr_machine_initfn(Object *obj)
2117 {
2118     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2119 
2120     spapr->htab_fd = -1;
2121     object_property_add_str(obj, "kvm-type",
2122                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2123     object_property_set_description(obj, "kvm-type",
2124                                     "Specifies the KVM virtualization mode (HV, PR)",
2125                                     NULL);
2126 }
2127 
2128 static void spapr_machine_finalizefn(Object *obj)
2129 {
2130     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2131 
2132     g_free(spapr->kvm_type);
2133 }
2134 
2135 static void ppc_cpu_do_nmi_on_cpu(void *arg)
2136 {
2137     CPUState *cs = 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, cs);
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