xref: /openbmc/qemu/hw/ppc/spapr.c (revision 416296a9)
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 = g_strdup_printf(UUID_FMT, qemu_uuid[0], qemu_uuid[1],
336                           qemu_uuid[2], qemu_uuid[3], qemu_uuid[4],
337                           qemu_uuid[5], qemu_uuid[6], qemu_uuid[7],
338                           qemu_uuid[8], qemu_uuid[9], qemu_uuid[10],
339                           qemu_uuid[11], qemu_uuid[12], qemu_uuid[13],
340                           qemu_uuid[14], qemu_uuid[15]);
341 
342     _FDT((fdt_property_string(fdt, "vm,uuid", buf)));
343     if (qemu_uuid_set) {
344         _FDT((fdt_property_string(fdt, "system-id", buf)));
345     }
346     g_free(buf);
347 
348     if (qemu_get_vm_name()) {
349         _FDT((fdt_property_string(fdt, "ibm,partition-name",
350                                   qemu_get_vm_name())));
351     }
352 
353     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
354     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
355 
356     /* /chosen */
357     _FDT((fdt_begin_node(fdt, "chosen")));
358 
359     /* Set Form1_affinity */
360     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
361 
362     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
363     _FDT((fdt_property(fdt, "linux,initrd-start",
364                        &start_prop, sizeof(start_prop))));
365     _FDT((fdt_property(fdt, "linux,initrd-end",
366                        &end_prop, sizeof(end_prop))));
367     if (kernel_size) {
368         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
369                               cpu_to_be64(kernel_size) };
370 
371         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
372         if (little_endian) {
373             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
374         }
375     }
376     if (boot_menu) {
377         _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
378     }
379     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
380     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
381     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
382 
383     _FDT((fdt_end_node(fdt)));
384 
385     /* RTAS */
386     _FDT((fdt_begin_node(fdt, "rtas")));
387 
388     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
389         add_str(hypertas, "hcall-multi-tce");
390     }
391     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
392                        hypertas->len)));
393     g_string_free(hypertas, TRUE);
394     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
395                        qemu_hypertas->len)));
396     g_string_free(qemu_hypertas, TRUE);
397 
398     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
399         refpoints, sizeof(refpoints))));
400 
401     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
402     _FDT((fdt_property_cell(fdt, "rtas-event-scan-rate",
403                             RTAS_EVENT_SCAN_RATE)));
404 
405     if (msi_nonbroken) {
406         _FDT((fdt_property(fdt, "ibm,change-msix-capable", NULL, 0)));
407     }
408 
409     /*
410      * According to PAPR, rtas ibm,os-term does not guarantee a return
411      * back to the guest cpu.
412      *
413      * While an additional ibm,extended-os-term property indicates that
414      * rtas call return will always occur. Set this property.
415      */
416     _FDT((fdt_property(fdt, "ibm,extended-os-term", NULL, 0)));
417 
418     _FDT((fdt_end_node(fdt)));
419 
420     /* interrupt controller */
421     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
422 
423     _FDT((fdt_property_string(fdt, "device_type",
424                               "PowerPC-External-Interrupt-Presentation")));
425     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
426     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
427     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
428                        interrupt_server_ranges_prop,
429                        sizeof(interrupt_server_ranges_prop))));
430     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
431     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
432     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
433 
434     _FDT((fdt_end_node(fdt)));
435 
436     /* vdevice */
437     _FDT((fdt_begin_node(fdt, "vdevice")));
438 
439     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
440     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
441     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
442     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
443     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
444     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
445 
446     _FDT((fdt_end_node(fdt)));
447 
448     /* event-sources */
449     spapr_events_fdt_skel(fdt, epow_irq);
450 
451     /* /hypervisor node */
452     if (kvm_enabled()) {
453         uint8_t hypercall[16];
454 
455         /* indicate KVM hypercall interface */
456         _FDT((fdt_begin_node(fdt, "hypervisor")));
457         _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
458         if (kvmppc_has_cap_fixup_hcalls()) {
459             /*
460              * Older KVM versions with older guest kernels were broken with the
461              * magic page, don't allow the guest to map it.
462              */
463             if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
464                                       sizeof(hypercall))) {
465                 _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
466                                    sizeof(hypercall))));
467             }
468         }
469         _FDT((fdt_end_node(fdt)));
470     }
471 
472     _FDT((fdt_end_node(fdt))); /* close root node */
473     _FDT((fdt_finish(fdt)));
474 
475     return fdt;
476 }
477 
478 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
479                                        hwaddr size)
480 {
481     uint32_t associativity[] = {
482         cpu_to_be32(0x4), /* length */
483         cpu_to_be32(0x0), cpu_to_be32(0x0),
484         cpu_to_be32(0x0), cpu_to_be32(nodeid)
485     };
486     char mem_name[32];
487     uint64_t mem_reg_property[2];
488     int off;
489 
490     mem_reg_property[0] = cpu_to_be64(start);
491     mem_reg_property[1] = cpu_to_be64(size);
492 
493     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
494     off = fdt_add_subnode(fdt, 0, mem_name);
495     _FDT(off);
496     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
497     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
498                       sizeof(mem_reg_property))));
499     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
500                       sizeof(associativity))));
501     return off;
502 }
503 
504 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
505 {
506     MachineState *machine = MACHINE(spapr);
507     hwaddr mem_start, node_size;
508     int i, nb_nodes = nb_numa_nodes;
509     NodeInfo *nodes = numa_info;
510     NodeInfo ramnode;
511 
512     /* No NUMA nodes, assume there is just one node with whole RAM */
513     if (!nb_numa_nodes) {
514         nb_nodes = 1;
515         ramnode.node_mem = machine->ram_size;
516         nodes = &ramnode;
517     }
518 
519     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
520         if (!nodes[i].node_mem) {
521             continue;
522         }
523         if (mem_start >= machine->ram_size) {
524             node_size = 0;
525         } else {
526             node_size = nodes[i].node_mem;
527             if (node_size > machine->ram_size - mem_start) {
528                 node_size = machine->ram_size - mem_start;
529             }
530         }
531         if (!mem_start) {
532             /* ppc_spapr_init() checks for rma_size <= node0_size already */
533             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
534             mem_start += spapr->rma_size;
535             node_size -= spapr->rma_size;
536         }
537         for ( ; node_size; ) {
538             hwaddr sizetmp = pow2floor(node_size);
539 
540             /* mem_start != 0 here */
541             if (ctzl(mem_start) < ctzl(sizetmp)) {
542                 sizetmp = 1ULL << ctzl(mem_start);
543             }
544 
545             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
546             node_size -= sizetmp;
547             mem_start += sizetmp;
548         }
549     }
550 
551     return 0;
552 }
553 
554 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
555                                   sPAPRMachineState *spapr)
556 {
557     PowerPCCPU *cpu = POWERPC_CPU(cs);
558     CPUPPCState *env = &cpu->env;
559     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
560     int index = ppc_get_vcpu_dt_id(cpu);
561     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
562                        0xffffffff, 0xffffffff};
563     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
564         : SPAPR_TIMEBASE_FREQ;
565     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
566     uint32_t page_sizes_prop[64];
567     size_t page_sizes_prop_size;
568     uint32_t vcpus_per_socket = smp_threads * smp_cores;
569     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
570     sPAPRDRConnector *drc;
571     sPAPRDRConnectorClass *drck;
572     int drc_index;
573 
574     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
575     if (drc) {
576         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
577         drc_index = drck->get_index(drc);
578         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
579     }
580 
581     /* Note: we keep CI large pages off for now because a 64K capable guest
582      * provisioned with large pages might otherwise try to map a qemu
583      * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
584      * even if that qemu runs on a 4k host.
585      *
586      * We can later add this bit back when we are confident this is not
587      * an issue (!HV KVM or 64K host)
588      */
589     uint8_t pa_features_206[] = { 6, 0,
590         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
591     uint8_t pa_features_207[] = { 24, 0,
592         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
593         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
594         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
595         0x80, 0x00, 0x80, 0x00, 0x80, 0x00 };
596     uint8_t *pa_features;
597     size_t pa_size;
598 
599     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
600     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
601 
602     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
603     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
604                            env->dcache_line_size)));
605     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
606                            env->dcache_line_size)));
607     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
608                            env->icache_line_size)));
609     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
610                            env->icache_line_size)));
611 
612     if (pcc->l1_dcache_size) {
613         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
614                                pcc->l1_dcache_size)));
615     } else {
616         error_report("Warning: Unknown L1 dcache size for cpu");
617     }
618     if (pcc->l1_icache_size) {
619         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
620                                pcc->l1_icache_size)));
621     } else {
622         error_report("Warning: Unknown L1 icache size for cpu");
623     }
624 
625     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
626     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
627     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
628     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
629     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
630     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
631 
632     if (env->spr_cb[SPR_PURR].oea_read) {
633         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
634     }
635 
636     if (env->mmu_model & POWERPC_MMU_1TSEG) {
637         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
638                           segs, sizeof(segs))));
639     }
640 
641     /* Advertise VMX/VSX (vector extensions) if available
642      *   0 / no property == no vector extensions
643      *   1               == VMX / Altivec available
644      *   2               == VSX available */
645     if (env->insns_flags & PPC_ALTIVEC) {
646         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
647 
648         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
649     }
650 
651     /* Advertise DFP (Decimal Floating Point) if available
652      *   0 / no property == no DFP
653      *   1               == DFP available */
654     if (env->insns_flags2 & PPC2_DFP) {
655         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
656     }
657 
658     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
659                                                   sizeof(page_sizes_prop));
660     if (page_sizes_prop_size) {
661         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
662                           page_sizes_prop, page_sizes_prop_size)));
663     }
664 
665     /* Do the ibm,pa-features property, adjust it for ci-large-pages */
666     if (env->mmu_model == POWERPC_MMU_2_06) {
667         pa_features = pa_features_206;
668         pa_size = sizeof(pa_features_206);
669     } else /* env->mmu_model == POWERPC_MMU_2_07 */ {
670         pa_features = pa_features_207;
671         pa_size = sizeof(pa_features_207);
672     }
673     if (env->ci_large_pages) {
674         pa_features[3] |= 0x20;
675     }
676     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
677 
678     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
679                            cs->cpu_index / vcpus_per_socket)));
680 
681     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
682                       pft_size_prop, sizeof(pft_size_prop))));
683 
684     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
685 
686     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
687                                 ppc_get_compat_smt_threads(cpu)));
688 }
689 
690 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
691 {
692     CPUState *cs;
693     int cpus_offset;
694     char *nodename;
695     int smt = kvmppc_smt_threads();
696 
697     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
698     _FDT(cpus_offset);
699     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
700     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
701 
702     /*
703      * We walk the CPUs in reverse order to ensure that CPU DT nodes
704      * created by fdt_add_subnode() end up in the right order in FDT
705      * for the guest kernel the enumerate the CPUs correctly.
706      */
707     CPU_FOREACH_REVERSE(cs) {
708         PowerPCCPU *cpu = POWERPC_CPU(cs);
709         int index = ppc_get_vcpu_dt_id(cpu);
710         DeviceClass *dc = DEVICE_GET_CLASS(cs);
711         int offset;
712 
713         if ((index % smt) != 0) {
714             continue;
715         }
716 
717         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
718         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
719         g_free(nodename);
720         _FDT(offset);
721         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
722     }
723 
724 }
725 
726 /*
727  * Adds ibm,dynamic-reconfiguration-memory node.
728  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
729  * of this device tree node.
730  */
731 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
732 {
733     MachineState *machine = MACHINE(spapr);
734     int ret, i, offset;
735     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
736     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
737     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
738     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
739                        memory_region_size(&spapr->hotplug_memory.mr)) /
740                        lmb_size;
741     uint32_t *int_buf, *cur_index, buf_len;
742     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
743 
744     /*
745      * Don't create the node if there is no hotpluggable memory
746      */
747     if (machine->ram_size == machine->maxram_size) {
748         return 0;
749     }
750 
751     /*
752      * Allocate enough buffer size to fit in ibm,dynamic-memory
753      * or ibm,associativity-lookup-arrays
754      */
755     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
756               * sizeof(uint32_t);
757     cur_index = int_buf = g_malloc0(buf_len);
758 
759     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
760 
761     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
762                     sizeof(prop_lmb_size));
763     if (ret < 0) {
764         goto out;
765     }
766 
767     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
768     if (ret < 0) {
769         goto out;
770     }
771 
772     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
773     if (ret < 0) {
774         goto out;
775     }
776 
777     /* ibm,dynamic-memory */
778     int_buf[0] = cpu_to_be32(nr_lmbs);
779     cur_index++;
780     for (i = 0; i < nr_lmbs; i++) {
781         uint64_t addr = i * lmb_size;
782         uint32_t *dynamic_memory = cur_index;
783 
784         if (i >= hotplug_lmb_start) {
785             sPAPRDRConnector *drc;
786             sPAPRDRConnectorClass *drck;
787 
788             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
789             g_assert(drc);
790             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
791 
792             dynamic_memory[0] = cpu_to_be32(addr >> 32);
793             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
794             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
795             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
796             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
797             if (memory_region_present(get_system_memory(), addr)) {
798                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
799             } else {
800                 dynamic_memory[5] = cpu_to_be32(0);
801             }
802         } else {
803             /*
804              * LMB information for RMA, boot time RAM and gap b/n RAM and
805              * hotplug memory region -- all these are marked as reserved
806              * and as having no valid DRC.
807              */
808             dynamic_memory[0] = cpu_to_be32(addr >> 32);
809             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
810             dynamic_memory[2] = cpu_to_be32(0);
811             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
812             dynamic_memory[4] = cpu_to_be32(-1);
813             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
814                                             SPAPR_LMB_FLAGS_DRC_INVALID);
815         }
816 
817         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
818     }
819     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
820     if (ret < 0) {
821         goto out;
822     }
823 
824     /* ibm,associativity-lookup-arrays */
825     cur_index = int_buf;
826     int_buf[0] = cpu_to_be32(nr_nodes);
827     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
828     cur_index += 2;
829     for (i = 0; i < nr_nodes; i++) {
830         uint32_t associativity[] = {
831             cpu_to_be32(0x0),
832             cpu_to_be32(0x0),
833             cpu_to_be32(0x0),
834             cpu_to_be32(i)
835         };
836         memcpy(cur_index, associativity, sizeof(associativity));
837         cur_index += 4;
838     }
839     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
840             (cur_index - int_buf) * sizeof(uint32_t));
841 out:
842     g_free(int_buf);
843     return ret;
844 }
845 
846 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
847                                  target_ulong addr, target_ulong size,
848                                  bool cpu_update, bool memory_update)
849 {
850     void *fdt, *fdt_skel;
851     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
852     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
853 
854     size -= sizeof(hdr);
855 
856     /* Create sceleton */
857     fdt_skel = g_malloc0(size);
858     _FDT((fdt_create(fdt_skel, size)));
859     _FDT((fdt_begin_node(fdt_skel, "")));
860     _FDT((fdt_end_node(fdt_skel)));
861     _FDT((fdt_finish(fdt_skel)));
862     fdt = g_malloc0(size);
863     _FDT((fdt_open_into(fdt_skel, fdt, size)));
864     g_free(fdt_skel);
865 
866     /* Fixup cpu nodes */
867     if (cpu_update) {
868         _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
869     }
870 
871     /* Generate ibm,dynamic-reconfiguration-memory node if required */
872     if (memory_update && smc->dr_lmb_enabled) {
873         _FDT((spapr_populate_drconf_memory(spapr, fdt)));
874     }
875 
876     /* Pack resulting tree */
877     _FDT((fdt_pack(fdt)));
878 
879     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
880         trace_spapr_cas_failed(size);
881         return -1;
882     }
883 
884     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
885     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
886     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
887     g_free(fdt);
888 
889     return 0;
890 }
891 
892 static void spapr_finalize_fdt(sPAPRMachineState *spapr,
893                                hwaddr fdt_addr,
894                                hwaddr rtas_addr,
895                                hwaddr rtas_size)
896 {
897     MachineState *machine = MACHINE(qdev_get_machine());
898     MachineClass *mc = MACHINE_GET_CLASS(machine);
899     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
900     const char *boot_device = machine->boot_order;
901     int ret, i;
902     size_t cb = 0;
903     char *bootlist;
904     void *fdt;
905     sPAPRPHBState *phb;
906 
907     fdt = g_malloc(FDT_MAX_SIZE);
908 
909     /* open out the base tree into a temp buffer for the final tweaks */
910     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
911 
912     ret = spapr_populate_memory(spapr, fdt);
913     if (ret < 0) {
914         error_report("couldn't setup memory nodes in fdt");
915         exit(1);
916     }
917 
918     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
919     if (ret < 0) {
920         error_report("couldn't setup vio devices in fdt");
921         exit(1);
922     }
923 
924     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
925         ret = spapr_rng_populate_dt(fdt);
926         if (ret < 0) {
927             error_report("could not set up rng device in the fdt");
928             exit(1);
929         }
930     }
931 
932     QLIST_FOREACH(phb, &spapr->phbs, list) {
933         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
934         if (ret < 0) {
935             error_report("couldn't setup PCI devices in fdt");
936             exit(1);
937         }
938     }
939 
940     /* RTAS */
941     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
942     if (ret < 0) {
943         error_report("Couldn't set up RTAS device tree properties");
944     }
945 
946     /* cpus */
947     spapr_populate_cpus_dt_node(fdt, spapr);
948 
949     bootlist = get_boot_devices_list(&cb, true);
950     if (cb && bootlist) {
951         int offset = fdt_path_offset(fdt, "/chosen");
952         if (offset < 0) {
953             exit(1);
954         }
955         for (i = 0; i < cb; i++) {
956             if (bootlist[i] == '\n') {
957                 bootlist[i] = ' ';
958             }
959 
960         }
961         ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
962     }
963 
964     if (boot_device && strlen(boot_device)) {
965         int offset = fdt_path_offset(fdt, "/chosen");
966 
967         if (offset < 0) {
968             exit(1);
969         }
970         fdt_setprop_string(fdt, offset, "qemu,boot-device", boot_device);
971     }
972 
973     if (!spapr->has_graphics) {
974         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
975     }
976 
977     if (smc->dr_lmb_enabled) {
978         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
979     }
980 
981     if (mc->query_hotpluggable_cpus) {
982         int offset = fdt_path_offset(fdt, "/cpus");
983         ret = spapr_drc_populate_dt(fdt, offset, NULL,
984                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
985         if (ret < 0) {
986             error_report("Couldn't set up CPU DR device tree properties");
987             exit(1);
988         }
989     }
990 
991     _FDT((fdt_pack(fdt)));
992 
993     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
994         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
995                      fdt_totalsize(fdt), FDT_MAX_SIZE);
996         exit(1);
997     }
998 
999     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1000     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1001 
1002     g_free(bootlist);
1003     g_free(fdt);
1004 }
1005 
1006 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1007 {
1008     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1009 }
1010 
1011 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
1012 {
1013     CPUPPCState *env = &cpu->env;
1014 
1015     if (msr_pr) {
1016         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1017         env->gpr[3] = H_PRIVILEGE;
1018     } else {
1019         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1020     }
1021 }
1022 
1023 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1024 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1025 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1026 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1027 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1028 
1029 /*
1030  * Get the fd to access the kernel htab, re-opening it if necessary
1031  */
1032 static int get_htab_fd(sPAPRMachineState *spapr)
1033 {
1034     if (spapr->htab_fd >= 0) {
1035         return spapr->htab_fd;
1036     }
1037 
1038     spapr->htab_fd = kvmppc_get_htab_fd(false);
1039     if (spapr->htab_fd < 0) {
1040         error_report("Unable to open fd for reading hash table from KVM: %s",
1041                      strerror(errno));
1042     }
1043 
1044     return spapr->htab_fd;
1045 }
1046 
1047 static void close_htab_fd(sPAPRMachineState *spapr)
1048 {
1049     if (spapr->htab_fd >= 0) {
1050         close(spapr->htab_fd);
1051     }
1052     spapr->htab_fd = -1;
1053 }
1054 
1055 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1056 {
1057     int shift;
1058 
1059     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1060      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1061      * that's much more than is needed for Linux guests */
1062     shift = ctz64(pow2ceil(ramsize)) - 7;
1063     shift = MAX(shift, 18); /* Minimum architected size */
1064     shift = MIN(shift, 46); /* Maximum architected size */
1065     return shift;
1066 }
1067 
1068 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1069                                  Error **errp)
1070 {
1071     long rc;
1072 
1073     /* Clean up any HPT info from a previous boot */
1074     g_free(spapr->htab);
1075     spapr->htab = NULL;
1076     spapr->htab_shift = 0;
1077     close_htab_fd(spapr);
1078 
1079     rc = kvmppc_reset_htab(shift);
1080     if (rc < 0) {
1081         /* kernel-side HPT needed, but couldn't allocate one */
1082         error_setg_errno(errp, errno,
1083                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1084                          shift);
1085         /* This is almost certainly fatal, but if the caller really
1086          * wants to carry on with shift == 0, it's welcome to try */
1087     } else if (rc > 0) {
1088         /* kernel-side HPT allocated */
1089         if (rc != shift) {
1090             error_setg(errp,
1091                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1092                        shift, rc);
1093         }
1094 
1095         spapr->htab_shift = shift;
1096         spapr->htab = NULL;
1097     } else {
1098         /* kernel-side HPT not needed, allocate in userspace instead */
1099         size_t size = 1ULL << shift;
1100         int i;
1101 
1102         spapr->htab = qemu_memalign(size, size);
1103         if (!spapr->htab) {
1104             error_setg_errno(errp, errno,
1105                              "Could not allocate HPT of order %d", shift);
1106             return;
1107         }
1108 
1109         memset(spapr->htab, 0, size);
1110         spapr->htab_shift = shift;
1111 
1112         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1113             DIRTY_HPTE(HPTE(spapr->htab, i));
1114         }
1115     }
1116 }
1117 
1118 static int find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1119 {
1120     bool matched = false;
1121 
1122     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1123         matched = true;
1124     }
1125 
1126     if (!matched) {
1127         error_report("Device %s is not supported by this machine yet.",
1128                      qdev_fw_name(DEVICE(sbdev)));
1129         exit(1);
1130     }
1131 
1132     return 0;
1133 }
1134 
1135 static void ppc_spapr_reset(void)
1136 {
1137     MachineState *machine = MACHINE(qdev_get_machine());
1138     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1139     PowerPCCPU *first_ppc_cpu;
1140     uint32_t rtas_limit;
1141 
1142     /* Check for unknown sysbus devices */
1143     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1144 
1145     /* Allocate and/or reset the hash page table */
1146     spapr_reallocate_hpt(spapr,
1147                          spapr_hpt_shift_for_ramsize(machine->maxram_size),
1148                          &error_fatal);
1149 
1150     /* Update the RMA size if necessary */
1151     if (spapr->vrma_adjust) {
1152         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1153                                           spapr->htab_shift);
1154     }
1155 
1156     qemu_devices_reset();
1157 
1158     /*
1159      * We place the device tree and RTAS just below either the top of the RMA,
1160      * or just below 2GB, whichever is lowere, so that it can be
1161      * processed with 32-bit real mode code if necessary
1162      */
1163     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1164     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1165     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1166 
1167     /* Load the fdt */
1168     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
1169                        spapr->rtas_size);
1170 
1171     /* Copy RTAS over */
1172     cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
1173                               spapr->rtas_size);
1174 
1175     /* Set up the entry state */
1176     first_ppc_cpu = POWERPC_CPU(first_cpu);
1177     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
1178     first_ppc_cpu->env.gpr[5] = 0;
1179     first_cpu->halted = 0;
1180     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1181 
1182 }
1183 
1184 static void spapr_create_nvram(sPAPRMachineState *spapr)
1185 {
1186     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1187     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1188 
1189     if (dinfo) {
1190         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1191                             &error_fatal);
1192     }
1193 
1194     qdev_init_nofail(dev);
1195 
1196     spapr->nvram = (struct sPAPRNVRAM *)dev;
1197 }
1198 
1199 static void spapr_rtc_create(sPAPRMachineState *spapr)
1200 {
1201     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1202 
1203     qdev_init_nofail(dev);
1204     spapr->rtc = dev;
1205 
1206     object_property_add_alias(qdev_get_machine(), "rtc-time",
1207                               OBJECT(spapr->rtc), "date", NULL);
1208 }
1209 
1210 /* Returns whether we want to use VGA or not */
1211 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1212 {
1213     switch (vga_interface_type) {
1214     case VGA_NONE:
1215         return false;
1216     case VGA_DEVICE:
1217         return true;
1218     case VGA_STD:
1219     case VGA_VIRTIO:
1220         return pci_vga_init(pci_bus) != NULL;
1221     default:
1222         error_setg(errp,
1223                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1224         return false;
1225     }
1226 }
1227 
1228 static int spapr_post_load(void *opaque, int version_id)
1229 {
1230     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1231     int err = 0;
1232 
1233     /* In earlier versions, there was no separate qdev for the PAPR
1234      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1235      * So when migrating from those versions, poke the incoming offset
1236      * value into the RTC device */
1237     if (version_id < 3) {
1238         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1239     }
1240 
1241     return err;
1242 }
1243 
1244 static bool version_before_3(void *opaque, int version_id)
1245 {
1246     return version_id < 3;
1247 }
1248 
1249 static const VMStateDescription vmstate_spapr = {
1250     .name = "spapr",
1251     .version_id = 3,
1252     .minimum_version_id = 1,
1253     .post_load = spapr_post_load,
1254     .fields = (VMStateField[]) {
1255         /* used to be @next_irq */
1256         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1257 
1258         /* RTC offset */
1259         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1260 
1261         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1262         VMSTATE_END_OF_LIST()
1263     },
1264 };
1265 
1266 static int htab_save_setup(QEMUFile *f, void *opaque)
1267 {
1268     sPAPRMachineState *spapr = opaque;
1269 
1270     /* "Iteration" header */
1271     qemu_put_be32(f, spapr->htab_shift);
1272 
1273     if (spapr->htab) {
1274         spapr->htab_save_index = 0;
1275         spapr->htab_first_pass = true;
1276     } else {
1277         assert(kvm_enabled());
1278     }
1279 
1280 
1281     return 0;
1282 }
1283 
1284 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1285                                  int64_t max_ns)
1286 {
1287     bool has_timeout = max_ns != -1;
1288     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1289     int index = spapr->htab_save_index;
1290     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1291 
1292     assert(spapr->htab_first_pass);
1293 
1294     do {
1295         int chunkstart;
1296 
1297         /* Consume invalid HPTEs */
1298         while ((index < htabslots)
1299                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1300             index++;
1301             CLEAN_HPTE(HPTE(spapr->htab, index));
1302         }
1303 
1304         /* Consume valid HPTEs */
1305         chunkstart = index;
1306         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1307                && HPTE_VALID(HPTE(spapr->htab, index))) {
1308             index++;
1309             CLEAN_HPTE(HPTE(spapr->htab, index));
1310         }
1311 
1312         if (index > chunkstart) {
1313             int n_valid = index - chunkstart;
1314 
1315             qemu_put_be32(f, chunkstart);
1316             qemu_put_be16(f, n_valid);
1317             qemu_put_be16(f, 0);
1318             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1319                             HASH_PTE_SIZE_64 * n_valid);
1320 
1321             if (has_timeout &&
1322                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1323                 break;
1324             }
1325         }
1326     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1327 
1328     if (index >= htabslots) {
1329         assert(index == htabslots);
1330         index = 0;
1331         spapr->htab_first_pass = false;
1332     }
1333     spapr->htab_save_index = index;
1334 }
1335 
1336 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1337                                 int64_t max_ns)
1338 {
1339     bool final = max_ns < 0;
1340     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1341     int examined = 0, sent = 0;
1342     int index = spapr->htab_save_index;
1343     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1344 
1345     assert(!spapr->htab_first_pass);
1346 
1347     do {
1348         int chunkstart, invalidstart;
1349 
1350         /* Consume non-dirty HPTEs */
1351         while ((index < htabslots)
1352                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1353             index++;
1354             examined++;
1355         }
1356 
1357         chunkstart = index;
1358         /* Consume valid dirty HPTEs */
1359         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1360                && HPTE_DIRTY(HPTE(spapr->htab, index))
1361                && HPTE_VALID(HPTE(spapr->htab, index))) {
1362             CLEAN_HPTE(HPTE(spapr->htab, index));
1363             index++;
1364             examined++;
1365         }
1366 
1367         invalidstart = index;
1368         /* Consume invalid dirty HPTEs */
1369         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1370                && HPTE_DIRTY(HPTE(spapr->htab, index))
1371                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1372             CLEAN_HPTE(HPTE(spapr->htab, index));
1373             index++;
1374             examined++;
1375         }
1376 
1377         if (index > chunkstart) {
1378             int n_valid = invalidstart - chunkstart;
1379             int n_invalid = index - invalidstart;
1380 
1381             qemu_put_be32(f, chunkstart);
1382             qemu_put_be16(f, n_valid);
1383             qemu_put_be16(f, n_invalid);
1384             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1385                             HASH_PTE_SIZE_64 * n_valid);
1386             sent += index - chunkstart;
1387 
1388             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1389                 break;
1390             }
1391         }
1392 
1393         if (examined >= htabslots) {
1394             break;
1395         }
1396 
1397         if (index >= htabslots) {
1398             assert(index == htabslots);
1399             index = 0;
1400         }
1401     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1402 
1403     if (index >= htabslots) {
1404         assert(index == htabslots);
1405         index = 0;
1406     }
1407 
1408     spapr->htab_save_index = index;
1409 
1410     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1411 }
1412 
1413 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1414 #define MAX_KVM_BUF_SIZE    2048
1415 
1416 static int htab_save_iterate(QEMUFile *f, void *opaque)
1417 {
1418     sPAPRMachineState *spapr = opaque;
1419     int fd;
1420     int rc = 0;
1421 
1422     /* Iteration header */
1423     qemu_put_be32(f, 0);
1424 
1425     if (!spapr->htab) {
1426         assert(kvm_enabled());
1427 
1428         fd = get_htab_fd(spapr);
1429         if (fd < 0) {
1430             return fd;
1431         }
1432 
1433         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1434         if (rc < 0) {
1435             return rc;
1436         }
1437     } else  if (spapr->htab_first_pass) {
1438         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1439     } else {
1440         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1441     }
1442 
1443     /* End marker */
1444     qemu_put_be32(f, 0);
1445     qemu_put_be16(f, 0);
1446     qemu_put_be16(f, 0);
1447 
1448     return rc;
1449 }
1450 
1451 static int htab_save_complete(QEMUFile *f, void *opaque)
1452 {
1453     sPAPRMachineState *spapr = opaque;
1454     int fd;
1455 
1456     /* Iteration header */
1457     qemu_put_be32(f, 0);
1458 
1459     if (!spapr->htab) {
1460         int rc;
1461 
1462         assert(kvm_enabled());
1463 
1464         fd = get_htab_fd(spapr);
1465         if (fd < 0) {
1466             return fd;
1467         }
1468 
1469         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1470         if (rc < 0) {
1471             return rc;
1472         }
1473     } else {
1474         if (spapr->htab_first_pass) {
1475             htab_save_first_pass(f, spapr, -1);
1476         }
1477         htab_save_later_pass(f, spapr, -1);
1478     }
1479 
1480     /* End marker */
1481     qemu_put_be32(f, 0);
1482     qemu_put_be16(f, 0);
1483     qemu_put_be16(f, 0);
1484 
1485     return 0;
1486 }
1487 
1488 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1489 {
1490     sPAPRMachineState *spapr = opaque;
1491     uint32_t section_hdr;
1492     int fd = -1;
1493 
1494     if (version_id < 1 || version_id > 1) {
1495         error_report("htab_load() bad version");
1496         return -EINVAL;
1497     }
1498 
1499     section_hdr = qemu_get_be32(f);
1500 
1501     if (section_hdr) {
1502         Error *local_err = NULL;
1503 
1504         /* First section gives the htab size */
1505         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1506         if (local_err) {
1507             error_report_err(local_err);
1508             return -EINVAL;
1509         }
1510         return 0;
1511     }
1512 
1513     if (!spapr->htab) {
1514         assert(kvm_enabled());
1515 
1516         fd = kvmppc_get_htab_fd(true);
1517         if (fd < 0) {
1518             error_report("Unable to open fd to restore KVM hash table: %s",
1519                          strerror(errno));
1520         }
1521     }
1522 
1523     while (true) {
1524         uint32_t index;
1525         uint16_t n_valid, n_invalid;
1526 
1527         index = qemu_get_be32(f);
1528         n_valid = qemu_get_be16(f);
1529         n_invalid = qemu_get_be16(f);
1530 
1531         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1532             /* End of Stream */
1533             break;
1534         }
1535 
1536         if ((index + n_valid + n_invalid) >
1537             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1538             /* Bad index in stream */
1539             error_report(
1540                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1541                 index, n_valid, n_invalid, spapr->htab_shift);
1542             return -EINVAL;
1543         }
1544 
1545         if (spapr->htab) {
1546             if (n_valid) {
1547                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1548                                 HASH_PTE_SIZE_64 * n_valid);
1549             }
1550             if (n_invalid) {
1551                 memset(HPTE(spapr->htab, index + n_valid), 0,
1552                        HASH_PTE_SIZE_64 * n_invalid);
1553             }
1554         } else {
1555             int rc;
1556 
1557             assert(fd >= 0);
1558 
1559             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1560             if (rc < 0) {
1561                 return rc;
1562             }
1563         }
1564     }
1565 
1566     if (!spapr->htab) {
1567         assert(fd >= 0);
1568         close(fd);
1569     }
1570 
1571     return 0;
1572 }
1573 
1574 static void htab_cleanup(void *opaque)
1575 {
1576     sPAPRMachineState *spapr = opaque;
1577 
1578     close_htab_fd(spapr);
1579 }
1580 
1581 static SaveVMHandlers savevm_htab_handlers = {
1582     .save_live_setup = htab_save_setup,
1583     .save_live_iterate = htab_save_iterate,
1584     .save_live_complete_precopy = htab_save_complete,
1585     .cleanup = htab_cleanup,
1586     .load_state = htab_load,
1587 };
1588 
1589 static void spapr_boot_set(void *opaque, const char *boot_device,
1590                            Error **errp)
1591 {
1592     MachineState *machine = MACHINE(qdev_get_machine());
1593     machine->boot_order = g_strdup(boot_device);
1594 }
1595 
1596 /*
1597  * Reset routine for LMB DR devices.
1598  *
1599  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1600  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1601  * when it walks all its children devices. LMB devices reset occurs
1602  * as part of spapr_ppc_reset().
1603  */
1604 static void spapr_drc_reset(void *opaque)
1605 {
1606     sPAPRDRConnector *drc = opaque;
1607     DeviceState *d = DEVICE(drc);
1608 
1609     if (d) {
1610         device_reset(d);
1611     }
1612 }
1613 
1614 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1615 {
1616     MachineState *machine = MACHINE(spapr);
1617     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1618     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1619     int i;
1620 
1621     for (i = 0; i < nr_lmbs; i++) {
1622         sPAPRDRConnector *drc;
1623         uint64_t addr;
1624 
1625         addr = i * lmb_size + spapr->hotplug_memory.base;
1626         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1627                                      addr/lmb_size);
1628         qemu_register_reset(spapr_drc_reset, drc);
1629     }
1630 }
1631 
1632 /*
1633  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1634  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1635  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1636  */
1637 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1638 {
1639     int i;
1640 
1641     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1642         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1643                    " is not aligned to %llu MiB",
1644                    machine->ram_size,
1645                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1646         return;
1647     }
1648 
1649     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1650         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1651                    " is not aligned to %llu MiB",
1652                    machine->ram_size,
1653                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1654         return;
1655     }
1656 
1657     for (i = 0; i < nb_numa_nodes; i++) {
1658         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1659             error_setg(errp,
1660                        "Node %d memory size 0x%" PRIx64
1661                        " is not aligned to %llu MiB",
1662                        i, numa_info[i].node_mem,
1663                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1664             return;
1665         }
1666     }
1667 }
1668 
1669 /* pSeries LPAR / sPAPR hardware init */
1670 static void ppc_spapr_init(MachineState *machine)
1671 {
1672     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1673     MachineClass *mc = MACHINE_GET_CLASS(machine);
1674     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1675     const char *kernel_filename = machine->kernel_filename;
1676     const char *kernel_cmdline = machine->kernel_cmdline;
1677     const char *initrd_filename = machine->initrd_filename;
1678     PCIHostState *phb;
1679     int i;
1680     MemoryRegion *sysmem = get_system_memory();
1681     MemoryRegion *ram = g_new(MemoryRegion, 1);
1682     MemoryRegion *rma_region;
1683     void *rma = NULL;
1684     hwaddr rma_alloc_size;
1685     hwaddr node0_size = spapr_node0_size();
1686     uint32_t initrd_base = 0;
1687     long kernel_size = 0, initrd_size = 0;
1688     long load_limit, fw_size;
1689     bool kernel_le = false;
1690     char *filename;
1691     int smt = kvmppc_smt_threads();
1692     int spapr_cores = smp_cpus / smp_threads;
1693     int spapr_max_cores = max_cpus / smp_threads;
1694 
1695     if (mc->query_hotpluggable_cpus) {
1696         if (smp_cpus % smp_threads) {
1697             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1698                          smp_cpus, smp_threads);
1699             exit(1);
1700         }
1701         if (max_cpus % smp_threads) {
1702             error_report("max_cpus (%u) must be multiple of threads (%u)",
1703                          max_cpus, smp_threads);
1704             exit(1);
1705         }
1706     }
1707 
1708     msi_nonbroken = true;
1709 
1710     QLIST_INIT(&spapr->phbs);
1711 
1712     cpu_ppc_hypercall = emulate_spapr_hypercall;
1713 
1714     /* Allocate RMA if necessary */
1715     rma_alloc_size = kvmppc_alloc_rma(&rma);
1716 
1717     if (rma_alloc_size == -1) {
1718         error_report("Unable to create RMA");
1719         exit(1);
1720     }
1721 
1722     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1723         spapr->rma_size = rma_alloc_size;
1724     } else {
1725         spapr->rma_size = node0_size;
1726 
1727         /* With KVM, we don't actually know whether KVM supports an
1728          * unbounded RMA (PR KVM) or is limited by the hash table size
1729          * (HV KVM using VRMA), so we always assume the latter
1730          *
1731          * In that case, we also limit the initial allocations for RTAS
1732          * etc... to 256M since we have no way to know what the VRMA size
1733          * is going to be as it depends on the size of the hash table
1734          * isn't determined yet.
1735          */
1736         if (kvm_enabled()) {
1737             spapr->vrma_adjust = 1;
1738             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1739         }
1740 
1741         /* Actually we don't support unbounded RMA anymore since we
1742          * added proper emulation of HV mode. The max we can get is
1743          * 16G which also happens to be what we configure for PAPR
1744          * mode so make sure we don't do anything bigger than that
1745          */
1746         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1747     }
1748 
1749     if (spapr->rma_size > node0_size) {
1750         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1751                      spapr->rma_size);
1752         exit(1);
1753     }
1754 
1755     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1756     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1757 
1758     /* Set up Interrupt Controller before we create the VCPUs */
1759     spapr->xics = xics_system_init(machine,
1760                                    DIV_ROUND_UP(max_cpus * smt, smp_threads),
1761                                    XICS_IRQS_SPAPR, &error_fatal);
1762 
1763     if (smc->dr_lmb_enabled) {
1764         spapr_validate_node_memory(machine, &error_fatal);
1765     }
1766 
1767     /* init CPUs */
1768     if (machine->cpu_model == NULL) {
1769         machine->cpu_model = kvm_enabled() ? "host" : "POWER7";
1770     }
1771 
1772     ppc_cpu_parse_features(machine->cpu_model);
1773 
1774     if (mc->query_hotpluggable_cpus) {
1775         char *type = spapr_get_cpu_core_type(machine->cpu_model);
1776 
1777         if (type == NULL) {
1778             error_report("Unable to find sPAPR CPU Core definition");
1779             exit(1);
1780         }
1781 
1782         spapr->cores = g_new0(Object *, spapr_max_cores);
1783         for (i = 0; i < spapr_max_cores; i++) {
1784             int core_id = i * smp_threads;
1785             sPAPRDRConnector *drc =
1786                 spapr_dr_connector_new(OBJECT(spapr),
1787                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
1788                                        (core_id / smp_threads) * smt);
1789 
1790             qemu_register_reset(spapr_drc_reset, drc);
1791 
1792             if (i < spapr_cores) {
1793                 Object *core  = object_new(type);
1794                 object_property_set_int(core, smp_threads, "nr-threads",
1795                                         &error_fatal);
1796                 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1797                                         &error_fatal);
1798                 object_property_set_bool(core, true, "realized", &error_fatal);
1799             }
1800         }
1801         g_free(type);
1802     } else {
1803         for (i = 0; i < smp_cpus; i++) {
1804             PowerPCCPU *cpu = cpu_ppc_init(machine->cpu_model);
1805             if (cpu == NULL) {
1806                 error_report("Unable to find PowerPC CPU definition");
1807                 exit(1);
1808             }
1809             spapr_cpu_init(spapr, cpu, &error_fatal);
1810        }
1811     }
1812 
1813     if (kvm_enabled()) {
1814         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1815         kvmppc_enable_logical_ci_hcalls();
1816         kvmppc_enable_set_mode_hcall();
1817     }
1818 
1819     /* allocate RAM */
1820     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1821                                          machine->ram_size);
1822     memory_region_add_subregion(sysmem, 0, ram);
1823 
1824     if (rma_alloc_size && rma) {
1825         rma_region = g_new(MemoryRegion, 1);
1826         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1827                                    rma_alloc_size, rma);
1828         vmstate_register_ram_global(rma_region);
1829         memory_region_add_subregion(sysmem, 0, rma_region);
1830     }
1831 
1832     /* initialize hotplug memory address space */
1833     if (machine->ram_size < machine->maxram_size) {
1834         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
1835         /*
1836          * Limit the number of hotpluggable memory slots to half the number
1837          * slots that KVM supports, leaving the other half for PCI and other
1838          * devices. However ensure that number of slots doesn't drop below 32.
1839          */
1840         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
1841                            SPAPR_MAX_RAM_SLOTS;
1842 
1843         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
1844             max_memslots = SPAPR_MAX_RAM_SLOTS;
1845         }
1846         if (machine->ram_slots > max_memslots) {
1847             error_report("Specified number of memory slots %"
1848                          PRIu64" exceeds max supported %d",
1849                          machine->ram_slots, max_memslots);
1850             exit(1);
1851         }
1852 
1853         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
1854                                               SPAPR_HOTPLUG_MEM_ALIGN);
1855         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
1856                            "hotplug-memory", hotplug_mem_size);
1857         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
1858                                     &spapr->hotplug_memory.mr);
1859     }
1860 
1861     if (smc->dr_lmb_enabled) {
1862         spapr_create_lmb_dr_connectors(spapr);
1863     }
1864 
1865     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1866     if (!filename) {
1867         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1868         exit(1);
1869     }
1870     spapr->rtas_size = get_image_size(filename);
1871     if (spapr->rtas_size < 0) {
1872         error_report("Could not get size of LPAR rtas '%s'", filename);
1873         exit(1);
1874     }
1875     spapr->rtas_blob = g_malloc(spapr->rtas_size);
1876     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1877         error_report("Could not load LPAR rtas '%s'", filename);
1878         exit(1);
1879     }
1880     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1881         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1882                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1883         exit(1);
1884     }
1885     g_free(filename);
1886 
1887     /* Set up EPOW events infrastructure */
1888     spapr_events_init(spapr);
1889 
1890     /* Set up the RTC RTAS interfaces */
1891     spapr_rtc_create(spapr);
1892 
1893     /* Set up VIO bus */
1894     spapr->vio_bus = spapr_vio_bus_init();
1895 
1896     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1897         if (serial_hds[i]) {
1898             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1899         }
1900     }
1901 
1902     /* We always have at least the nvram device on VIO */
1903     spapr_create_nvram(spapr);
1904 
1905     /* Set up PCI */
1906     spapr_pci_rtas_init();
1907 
1908     phb = spapr_create_phb(spapr, 0);
1909 
1910     for (i = 0; i < nb_nics; i++) {
1911         NICInfo *nd = &nd_table[i];
1912 
1913         if (!nd->model) {
1914             nd->model = g_strdup("ibmveth");
1915         }
1916 
1917         if (strcmp(nd->model, "ibmveth") == 0) {
1918             spapr_vlan_create(spapr->vio_bus, nd);
1919         } else {
1920             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1921         }
1922     }
1923 
1924     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1925         spapr_vscsi_create(spapr->vio_bus);
1926     }
1927 
1928     /* Graphics */
1929     if (spapr_vga_init(phb->bus, &error_fatal)) {
1930         spapr->has_graphics = true;
1931         machine->usb |= defaults_enabled() && !machine->usb_disabled;
1932     }
1933 
1934     if (machine->usb) {
1935         if (smc->use_ohci_by_default) {
1936             pci_create_simple(phb->bus, -1, "pci-ohci");
1937         } else {
1938             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
1939         }
1940 
1941         if (spapr->has_graphics) {
1942             USBBus *usb_bus = usb_bus_find(-1);
1943 
1944             usb_create_simple(usb_bus, "usb-kbd");
1945             usb_create_simple(usb_bus, "usb-mouse");
1946         }
1947     }
1948 
1949     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1950         error_report(
1951             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
1952             MIN_RMA_SLOF);
1953         exit(1);
1954     }
1955 
1956     if (kernel_filename) {
1957         uint64_t lowaddr = 0;
1958 
1959         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1960                                NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE,
1961                                0, 0);
1962         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1963             kernel_size = load_elf(kernel_filename,
1964                                    translate_kernel_address, NULL,
1965                                    NULL, &lowaddr, NULL, 0, PPC_ELF_MACHINE,
1966                                    0, 0);
1967             kernel_le = kernel_size > 0;
1968         }
1969         if (kernel_size < 0) {
1970             error_report("error loading %s: %s",
1971                          kernel_filename, load_elf_strerror(kernel_size));
1972             exit(1);
1973         }
1974 
1975         /* load initrd */
1976         if (initrd_filename) {
1977             /* Try to locate the initrd in the gap between the kernel
1978              * and the firmware. Add a bit of space just in case
1979              */
1980             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1981             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1982                                               load_limit - initrd_base);
1983             if (initrd_size < 0) {
1984                 error_report("could not load initial ram disk '%s'",
1985                              initrd_filename);
1986                 exit(1);
1987             }
1988         } else {
1989             initrd_base = 0;
1990             initrd_size = 0;
1991         }
1992     }
1993 
1994     if (bios_name == NULL) {
1995         bios_name = FW_FILE_NAME;
1996     }
1997     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1998     if (!filename) {
1999         error_report("Could not find LPAR firmware '%s'", bios_name);
2000         exit(1);
2001     }
2002     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2003     if (fw_size <= 0) {
2004         error_report("Could not load LPAR firmware '%s'", filename);
2005         exit(1);
2006     }
2007     g_free(filename);
2008 
2009     /* FIXME: Should register things through the MachineState's qdev
2010      * interface, this is a legacy from the sPAPREnvironment structure
2011      * which predated MachineState but had a similar function */
2012     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2013     register_savevm_live(NULL, "spapr/htab", -1, 1,
2014                          &savevm_htab_handlers, spapr);
2015 
2016     /* Prepare the device tree */
2017     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
2018                                             kernel_size, kernel_le,
2019                                             kernel_cmdline,
2020                                             spapr->check_exception_irq);
2021     assert(spapr->fdt_skel != NULL);
2022 
2023     /* used by RTAS */
2024     QTAILQ_INIT(&spapr->ccs_list);
2025     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2026 
2027     qemu_register_boot_set(spapr_boot_set, spapr);
2028 }
2029 
2030 static int spapr_kvm_type(const char *vm_type)
2031 {
2032     if (!vm_type) {
2033         return 0;
2034     }
2035 
2036     if (!strcmp(vm_type, "HV")) {
2037         return 1;
2038     }
2039 
2040     if (!strcmp(vm_type, "PR")) {
2041         return 2;
2042     }
2043 
2044     error_report("Unknown kvm-type specified '%s'", vm_type);
2045     exit(1);
2046 }
2047 
2048 /*
2049  * Implementation of an interface to adjust firmware path
2050  * for the bootindex property handling.
2051  */
2052 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2053                                    DeviceState *dev)
2054 {
2055 #define CAST(type, obj, name) \
2056     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2057     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2058     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2059 
2060     if (d) {
2061         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2062         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2063         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2064 
2065         if (spapr) {
2066             /*
2067              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2068              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2069              * in the top 16 bits of the 64-bit LUN
2070              */
2071             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2072             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2073                                    (uint64_t)id << 48);
2074         } else if (virtio) {
2075             /*
2076              * We use SRP luns of the form 01000000 | (target << 8) | lun
2077              * in the top 32 bits of the 64-bit LUN
2078              * Note: the quote above is from SLOF and it is wrong,
2079              * the actual binding is:
2080              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2081              */
2082             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2083             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2084                                    (uint64_t)id << 32);
2085         } else if (usb) {
2086             /*
2087              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2088              * in the top 32 bits of the 64-bit LUN
2089              */
2090             unsigned usb_port = atoi(usb->port->path);
2091             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2092             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2093                                    (uint64_t)id << 32);
2094         }
2095     }
2096 
2097     if (phb) {
2098         /* Replace "pci" with "pci@800000020000000" */
2099         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2100     }
2101 
2102     return NULL;
2103 }
2104 
2105 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2106 {
2107     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2108 
2109     return g_strdup(spapr->kvm_type);
2110 }
2111 
2112 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2113 {
2114     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2115 
2116     g_free(spapr->kvm_type);
2117     spapr->kvm_type = g_strdup(value);
2118 }
2119 
2120 static void spapr_machine_initfn(Object *obj)
2121 {
2122     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2123 
2124     spapr->htab_fd = -1;
2125     object_property_add_str(obj, "kvm-type",
2126                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2127     object_property_set_description(obj, "kvm-type",
2128                                     "Specifies the KVM virtualization mode (HV, PR)",
2129                                     NULL);
2130 }
2131 
2132 static void spapr_machine_finalizefn(Object *obj)
2133 {
2134     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2135 
2136     g_free(spapr->kvm_type);
2137 }
2138 
2139 static void ppc_cpu_do_nmi_on_cpu(void *arg)
2140 {
2141     CPUState *cs = arg;
2142 
2143     cpu_synchronize_state(cs);
2144     ppc_cpu_do_system_reset(cs);
2145 }
2146 
2147 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2148 {
2149     CPUState *cs;
2150 
2151     CPU_FOREACH(cs) {
2152         async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, cs);
2153     }
2154 }
2155 
2156 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr, uint64_t size,
2157                            uint32_t node, Error **errp)
2158 {
2159     sPAPRDRConnector *drc;
2160     sPAPRDRConnectorClass *drck;
2161     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2162     int i, fdt_offset, fdt_size;
2163     void *fdt;
2164 
2165     for (i = 0; i < nr_lmbs; i++) {
2166         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2167                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2168         g_assert(drc);
2169 
2170         fdt = create_device_tree(&fdt_size);
2171         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2172                                                 SPAPR_MEMORY_BLOCK_SIZE);
2173 
2174         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2175         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2176         addr += SPAPR_MEMORY_BLOCK_SIZE;
2177     }
2178     /* send hotplug notification to the
2179      * guest only in case of hotplugged memory
2180      */
2181     if (dev->hotplugged) {
2182        spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB, nr_lmbs);
2183     }
2184 }
2185 
2186 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2187                               uint32_t node, Error **errp)
2188 {
2189     Error *local_err = NULL;
2190     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2191     PCDIMMDevice *dimm = PC_DIMM(dev);
2192     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2193     MemoryRegion *mr = ddc->get_memory_region(dimm);
2194     uint64_t align = memory_region_get_alignment(mr);
2195     uint64_t size = memory_region_size(mr);
2196     uint64_t addr;
2197 
2198     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2199         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2200                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2201         goto out;
2202     }
2203 
2204     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2205     if (local_err) {
2206         goto out;
2207     }
2208 
2209     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2210     if (local_err) {
2211         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2212         goto out;
2213     }
2214 
2215     spapr_add_lmbs(dev, addr, size, node, &error_abort);
2216 
2217 out:
2218     error_propagate(errp, local_err);
2219 }
2220 
2221 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2222                                     sPAPRMachineState *spapr)
2223 {
2224     PowerPCCPU *cpu = POWERPC_CPU(cs);
2225     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2226     int id = ppc_get_vcpu_dt_id(cpu);
2227     void *fdt;
2228     int offset, fdt_size;
2229     char *nodename;
2230 
2231     fdt = create_device_tree(&fdt_size);
2232     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2233     offset = fdt_add_subnode(fdt, 0, nodename);
2234 
2235     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2236     g_free(nodename);
2237 
2238     *fdt_offset = offset;
2239     return fdt;
2240 }
2241 
2242 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2243                                       DeviceState *dev, Error **errp)
2244 {
2245     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2246 
2247     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2248         int node;
2249 
2250         if (!smc->dr_lmb_enabled) {
2251             error_setg(errp, "Memory hotplug not supported for this machine");
2252             return;
2253         }
2254         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2255         if (*errp) {
2256             return;
2257         }
2258         if (node < 0 || node >= MAX_NODES) {
2259             error_setg(errp, "Invaild node %d", node);
2260             return;
2261         }
2262 
2263         /*
2264          * Currently PowerPC kernel doesn't allow hot-adding memory to
2265          * memory-less node, but instead will silently add the memory
2266          * to the first node that has some memory. This causes two
2267          * unexpected behaviours for the user.
2268          *
2269          * - Memory gets hotplugged to a different node than what the user
2270          *   specified.
2271          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2272          *   to memory-less node, a reboot will set things accordingly
2273          *   and the previously hotplugged memory now ends in the right node.
2274          *   This appears as if some memory moved from one node to another.
2275          *
2276          * So until kernel starts supporting memory hotplug to memory-less
2277          * nodes, just prevent such attempts upfront in QEMU.
2278          */
2279         if (nb_numa_nodes && !numa_info[node].node_mem) {
2280             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2281                        node);
2282             return;
2283         }
2284 
2285         spapr_memory_plug(hotplug_dev, dev, node, errp);
2286     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2287         spapr_core_plug(hotplug_dev, dev, errp);
2288     }
2289 }
2290 
2291 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2292                                       DeviceState *dev, Error **errp)
2293 {
2294     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2295 
2296     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2297         error_setg(errp, "Memory hot unplug not supported by sPAPR");
2298     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2299         if (!mc->query_hotpluggable_cpus) {
2300             error_setg(errp, "CPU hot unplug not supported on this machine");
2301             return;
2302         }
2303         spapr_core_unplug(hotplug_dev, dev, errp);
2304     }
2305 }
2306 
2307 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2308                                           DeviceState *dev, Error **errp)
2309 {
2310     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2311         spapr_core_pre_plug(hotplug_dev, dev, errp);
2312     }
2313 }
2314 
2315 static HotplugHandler *spapr_get_hotpug_handler(MachineState *machine,
2316                                              DeviceState *dev)
2317 {
2318     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2319         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2320         return HOTPLUG_HANDLER(machine);
2321     }
2322     return NULL;
2323 }
2324 
2325 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2326 {
2327     /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2328      * socket means much for the paravirtualized PAPR platform) */
2329     return cpu_index / smp_threads / smp_cores;
2330 }
2331 
2332 static HotpluggableCPUList *spapr_query_hotpluggable_cpus(MachineState *machine)
2333 {
2334     int i;
2335     HotpluggableCPUList *head = NULL;
2336     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2337     int spapr_max_cores = max_cpus / smp_threads;
2338 
2339     for (i = 0; i < spapr_max_cores; i++) {
2340         HotpluggableCPUList *list_item = g_new0(typeof(*list_item), 1);
2341         HotpluggableCPU *cpu_item = g_new0(typeof(*cpu_item), 1);
2342         CpuInstanceProperties *cpu_props = g_new0(typeof(*cpu_props), 1);
2343 
2344         cpu_item->type = spapr_get_cpu_core_type(machine->cpu_model);
2345         cpu_item->vcpus_count = smp_threads;
2346         cpu_props->has_core_id = true;
2347         cpu_props->core_id = i * smp_threads;
2348         /* TODO: add 'has_node/node' here to describe
2349            to which node core belongs */
2350 
2351         cpu_item->props = cpu_props;
2352         if (spapr->cores[i]) {
2353             cpu_item->has_qom_path = true;
2354             cpu_item->qom_path = object_get_canonical_path(spapr->cores[i]);
2355         }
2356         list_item->value = cpu_item;
2357         list_item->next = head;
2358         head = list_item;
2359     }
2360     return head;
2361 }
2362 
2363 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2364 {
2365     MachineClass *mc = MACHINE_CLASS(oc);
2366     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2367     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2368     NMIClass *nc = NMI_CLASS(oc);
2369     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2370 
2371     mc->desc = "pSeries Logical Partition (PAPR compliant)";
2372 
2373     /*
2374      * We set up the default / latest behaviour here.  The class_init
2375      * functions for the specific versioned machine types can override
2376      * these details for backwards compatibility
2377      */
2378     mc->init = ppc_spapr_init;
2379     mc->reset = ppc_spapr_reset;
2380     mc->block_default_type = IF_SCSI;
2381     mc->max_cpus = MAX_CPUMASK_BITS;
2382     mc->no_parallel = 1;
2383     mc->default_boot_order = "";
2384     mc->default_ram_size = 512 * M_BYTE;
2385     mc->kvm_type = spapr_kvm_type;
2386     mc->has_dynamic_sysbus = true;
2387     mc->pci_allow_0_address = true;
2388     mc->get_hotplug_handler = spapr_get_hotpug_handler;
2389     hc->pre_plug = spapr_machine_device_pre_plug;
2390     hc->plug = spapr_machine_device_plug;
2391     hc->unplug = spapr_machine_device_unplug;
2392     mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
2393 
2394     smc->dr_lmb_enabled = true;
2395     mc->query_hotpluggable_cpus = spapr_query_hotpluggable_cpus;
2396     fwc->get_dev_path = spapr_get_fw_dev_path;
2397     nc->nmi_monitor_handler = spapr_nmi;
2398 }
2399 
2400 static const TypeInfo spapr_machine_info = {
2401     .name          = TYPE_SPAPR_MACHINE,
2402     .parent        = TYPE_MACHINE,
2403     .abstract      = true,
2404     .instance_size = sizeof(sPAPRMachineState),
2405     .instance_init = spapr_machine_initfn,
2406     .instance_finalize = spapr_machine_finalizefn,
2407     .class_size    = sizeof(sPAPRMachineClass),
2408     .class_init    = spapr_machine_class_init,
2409     .interfaces = (InterfaceInfo[]) {
2410         { TYPE_FW_PATH_PROVIDER },
2411         { TYPE_NMI },
2412         { TYPE_HOTPLUG_HANDLER },
2413         { }
2414     },
2415 };
2416 
2417 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
2418     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
2419                                                     void *data)      \
2420     {                                                                \
2421         MachineClass *mc = MACHINE_CLASS(oc);                        \
2422         spapr_machine_##suffix##_class_options(mc);                  \
2423         if (latest) {                                                \
2424             mc->alias = "pseries";                                   \
2425             mc->is_default = 1;                                      \
2426         }                                                            \
2427     }                                                                \
2428     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
2429     {                                                                \
2430         MachineState *machine = MACHINE(obj);                        \
2431         spapr_machine_##suffix##_instance_options(machine);          \
2432     }                                                                \
2433     static const TypeInfo spapr_machine_##suffix##_info = {          \
2434         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
2435         .parent = TYPE_SPAPR_MACHINE,                                \
2436         .class_init = spapr_machine_##suffix##_class_init,           \
2437         .instance_init = spapr_machine_##suffix##_instance_init,     \
2438     };                                                               \
2439     static void spapr_machine_register_##suffix(void)                \
2440     {                                                                \
2441         type_register(&spapr_machine_##suffix##_info);               \
2442     }                                                                \
2443     type_init(spapr_machine_register_##suffix)
2444 
2445 /*
2446  * pseries-2.7
2447  */
2448 static void spapr_machine_2_7_instance_options(MachineState *machine)
2449 {
2450 }
2451 
2452 static void spapr_machine_2_7_class_options(MachineClass *mc)
2453 {
2454     /* Defaults for the latest behaviour inherited from the base class */
2455 }
2456 
2457 DEFINE_SPAPR_MACHINE(2_7, "2.7", true);
2458 
2459 /*
2460  * pseries-2.6
2461  */
2462 #define SPAPR_COMPAT_2_6 \
2463     HW_COMPAT_2_6 \
2464     { \
2465         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2466         .property = "ddw",\
2467         .value    = stringify(off),\
2468     },
2469 
2470 static void spapr_machine_2_6_instance_options(MachineState *machine)
2471 {
2472 }
2473 
2474 static void spapr_machine_2_6_class_options(MachineClass *mc)
2475 {
2476     spapr_machine_2_7_class_options(mc);
2477     mc->query_hotpluggable_cpus = NULL;
2478     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
2479 }
2480 
2481 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
2482 
2483 /*
2484  * pseries-2.5
2485  */
2486 #define SPAPR_COMPAT_2_5 \
2487     HW_COMPAT_2_5 \
2488     { \
2489         .driver   = "spapr-vlan", \
2490         .property = "use-rx-buffer-pools", \
2491         .value    = "off", \
2492     },
2493 
2494 static void spapr_machine_2_5_instance_options(MachineState *machine)
2495 {
2496 }
2497 
2498 static void spapr_machine_2_5_class_options(MachineClass *mc)
2499 {
2500     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2501 
2502     spapr_machine_2_6_class_options(mc);
2503     smc->use_ohci_by_default = true;
2504     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
2505 }
2506 
2507 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
2508 
2509 /*
2510  * pseries-2.4
2511  */
2512 #define SPAPR_COMPAT_2_4 \
2513         HW_COMPAT_2_4
2514 
2515 static void spapr_machine_2_4_instance_options(MachineState *machine)
2516 {
2517     spapr_machine_2_5_instance_options(machine);
2518 }
2519 
2520 static void spapr_machine_2_4_class_options(MachineClass *mc)
2521 {
2522     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2523 
2524     spapr_machine_2_5_class_options(mc);
2525     smc->dr_lmb_enabled = false;
2526     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
2527 }
2528 
2529 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
2530 
2531 /*
2532  * pseries-2.3
2533  */
2534 #define SPAPR_COMPAT_2_3 \
2535         HW_COMPAT_2_3 \
2536         {\
2537             .driver   = "spapr-pci-host-bridge",\
2538             .property = "dynamic-reconfiguration",\
2539             .value    = "off",\
2540         },
2541 
2542 static void spapr_machine_2_3_instance_options(MachineState *machine)
2543 {
2544     spapr_machine_2_4_instance_options(machine);
2545     savevm_skip_section_footers();
2546     global_state_set_optional();
2547     savevm_skip_configuration();
2548 }
2549 
2550 static void spapr_machine_2_3_class_options(MachineClass *mc)
2551 {
2552     spapr_machine_2_4_class_options(mc);
2553     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
2554 }
2555 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
2556 
2557 /*
2558  * pseries-2.2
2559  */
2560 
2561 #define SPAPR_COMPAT_2_2 \
2562         HW_COMPAT_2_2 \
2563         {\
2564             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2565             .property = "mem_win_size",\
2566             .value    = "0x20000000",\
2567         },
2568 
2569 static void spapr_machine_2_2_instance_options(MachineState *machine)
2570 {
2571     spapr_machine_2_3_instance_options(machine);
2572     machine->suppress_vmdesc = true;
2573 }
2574 
2575 static void spapr_machine_2_2_class_options(MachineClass *mc)
2576 {
2577     spapr_machine_2_3_class_options(mc);
2578     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
2579 }
2580 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
2581 
2582 /*
2583  * pseries-2.1
2584  */
2585 #define SPAPR_COMPAT_2_1 \
2586         HW_COMPAT_2_1
2587 
2588 static void spapr_machine_2_1_instance_options(MachineState *machine)
2589 {
2590     spapr_machine_2_2_instance_options(machine);
2591 }
2592 
2593 static void spapr_machine_2_1_class_options(MachineClass *mc)
2594 {
2595     spapr_machine_2_2_class_options(mc);
2596     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
2597 }
2598 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
2599 
2600 static void spapr_machine_register_types(void)
2601 {
2602     type_register_static(&spapr_machine_info);
2603 }
2604 
2605 type_init(spapr_machine_register_types)
2606