xref: /openbmc/qemu/hw/ppc/spapr.c (revision dfeb8679)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 #include "sysemu/sysemu.h"
28 #include "sysemu/numa.h"
29 #include "hw/hw.h"
30 #include "hw/fw-path-provider.h"
31 #include "elf.h"
32 #include "net/net.h"
33 #include "sysemu/device_tree.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/cpus.h"
36 #include "sysemu/kvm.h"
37 #include "sysemu/device_tree.h"
38 #include "kvm_ppc.h"
39 #include "migration/migration.h"
40 #include "mmu-hash64.h"
41 #include "qom/cpu.h"
42 
43 #include "hw/boards.h"
44 #include "hw/ppc/ppc.h"
45 #include "hw/loader.h"
46 
47 #include "hw/ppc/spapr.h"
48 #include "hw/ppc/spapr_vio.h"
49 #include "hw/pci-host/spapr.h"
50 #include "hw/ppc/xics.h"
51 #include "hw/pci/msi.h"
52 
53 #include "hw/pci/pci.h"
54 #include "hw/scsi/scsi.h"
55 #include "hw/virtio/virtio-scsi.h"
56 
57 #include "exec/address-spaces.h"
58 #include "hw/usb.h"
59 #include "qemu/config-file.h"
60 #include "qemu/error-report.h"
61 #include "trace.h"
62 #include "hw/nmi.h"
63 
64 #include "hw/compat.h"
65 #include "qemu-common.h"
66 
67 #include <libfdt.h>
68 
69 /* SLOF memory layout:
70  *
71  * SLOF raw image loaded at 0, copies its romfs right below the flat
72  * device-tree, then position SLOF itself 31M below that
73  *
74  * So we set FW_OVERHEAD to 40MB which should account for all of that
75  * and more
76  *
77  * We load our kernel at 4M, leaving space for SLOF initial image
78  */
79 #define FDT_MAX_SIZE            0x100000
80 #define RTAS_MAX_SIZE           0x10000
81 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
82 #define FW_MAX_SIZE             0x400000
83 #define FW_FILE_NAME            "slof.bin"
84 #define FW_OVERHEAD             0x2800000
85 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
86 
87 #define MIN_RMA_SLOF            128UL
88 
89 #define TIMEBASE_FREQ           512000000ULL
90 
91 #define PHANDLE_XICP            0x00001111
92 
93 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
94 
95 static XICSState *try_create_xics(const char *type, int nr_servers,
96                                   int nr_irqs, Error **errp)
97 {
98     Error *err = NULL;
99     DeviceState *dev;
100 
101     dev = qdev_create(NULL, type);
102     qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
103     qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
104     object_property_set_bool(OBJECT(dev), true, "realized", &err);
105     if (err) {
106         error_propagate(errp, err);
107         object_unparent(OBJECT(dev));
108         return NULL;
109     }
110     return XICS_COMMON(dev);
111 }
112 
113 static XICSState *xics_system_init(MachineState *machine,
114                                    int nr_servers, int nr_irqs)
115 {
116     XICSState *icp = NULL;
117 
118     if (kvm_enabled()) {
119         Error *err = NULL;
120 
121         if (machine_kernel_irqchip_allowed(machine)) {
122             icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs, &err);
123         }
124         if (machine_kernel_irqchip_required(machine) && !icp) {
125             error_report("kernel_irqchip requested but unavailable: %s",
126                          error_get_pretty(err));
127         }
128     }
129 
130     if (!icp) {
131         icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs, &error_abort);
132     }
133 
134     return icp;
135 }
136 
137 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
138                                   int smt_threads)
139 {
140     int i, ret = 0;
141     uint32_t servers_prop[smt_threads];
142     uint32_t gservers_prop[smt_threads * 2];
143     int index = ppc_get_vcpu_dt_id(cpu);
144 
145     if (cpu->cpu_version) {
146         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
147         if (ret < 0) {
148             return ret;
149         }
150     }
151 
152     /* Build interrupt servers and gservers properties */
153     for (i = 0; i < smt_threads; i++) {
154         servers_prop[i] = cpu_to_be32(index + i);
155         /* Hack, direct the group queues back to cpu 0 */
156         gservers_prop[i*2] = cpu_to_be32(index + i);
157         gservers_prop[i*2 + 1] = 0;
158     }
159     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
160                       servers_prop, sizeof(servers_prop));
161     if (ret < 0) {
162         return ret;
163     }
164     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
165                       gservers_prop, sizeof(gservers_prop));
166 
167     return ret;
168 }
169 
170 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
171 {
172     int ret = 0;
173     PowerPCCPU *cpu = POWERPC_CPU(cs);
174     int index = ppc_get_vcpu_dt_id(cpu);
175     uint32_t associativity[] = {cpu_to_be32(0x5),
176                                 cpu_to_be32(0x0),
177                                 cpu_to_be32(0x0),
178                                 cpu_to_be32(0x0),
179                                 cpu_to_be32(cs->numa_node),
180                                 cpu_to_be32(index)};
181 
182     /* Advertise NUMA via ibm,associativity */
183     if (nb_numa_nodes > 1) {
184         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
185                           sizeof(associativity));
186     }
187 
188     return ret;
189 }
190 
191 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
192 {
193     int ret = 0, offset, cpus_offset;
194     CPUState *cs;
195     char cpu_model[32];
196     int smt = kvmppc_smt_threads();
197     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
198 
199     CPU_FOREACH(cs) {
200         PowerPCCPU *cpu = POWERPC_CPU(cs);
201         DeviceClass *dc = DEVICE_GET_CLASS(cs);
202         int index = ppc_get_vcpu_dt_id(cpu);
203 
204         if ((index % smt) != 0) {
205             continue;
206         }
207 
208         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
209 
210         cpus_offset = fdt_path_offset(fdt, "/cpus");
211         if (cpus_offset < 0) {
212             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
213                                           "cpus");
214             if (cpus_offset < 0) {
215                 return cpus_offset;
216             }
217         }
218         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
219         if (offset < 0) {
220             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
221             if (offset < 0) {
222                 return offset;
223             }
224         }
225 
226         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
227                           pft_size_prop, sizeof(pft_size_prop));
228         if (ret < 0) {
229             return ret;
230         }
231 
232         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
233         if (ret < 0) {
234             return ret;
235         }
236 
237         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
238                                      ppc_get_compat_smt_threads(cpu));
239         if (ret < 0) {
240             return ret;
241         }
242     }
243     return ret;
244 }
245 
246 
247 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
248                                      size_t maxsize)
249 {
250     size_t maxcells = maxsize / sizeof(uint32_t);
251     int i, j, count;
252     uint32_t *p = prop;
253 
254     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
255         struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
256 
257         if (!sps->page_shift) {
258             break;
259         }
260         for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
261             if (sps->enc[count].page_shift == 0) {
262                 break;
263             }
264         }
265         if ((p - prop) >= (maxcells - 3 - count * 2)) {
266             break;
267         }
268         *(p++) = cpu_to_be32(sps->page_shift);
269         *(p++) = cpu_to_be32(sps->slb_enc);
270         *(p++) = cpu_to_be32(count);
271         for (j = 0; j < count; j++) {
272             *(p++) = cpu_to_be32(sps->enc[j].page_shift);
273             *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
274         }
275     }
276 
277     return (p - prop) * sizeof(uint32_t);
278 }
279 
280 static hwaddr spapr_node0_size(void)
281 {
282     MachineState *machine = MACHINE(qdev_get_machine());
283 
284     if (nb_numa_nodes) {
285         int i;
286         for (i = 0; i < nb_numa_nodes; ++i) {
287             if (numa_info[i].node_mem) {
288                 return MIN(pow2floor(numa_info[i].node_mem),
289                            machine->ram_size);
290             }
291         }
292     }
293     return machine->ram_size;
294 }
295 
296 #define _FDT(exp) \
297     do { \
298         int ret = (exp);                                           \
299         if (ret < 0) {                                             \
300             fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
301                     #exp, fdt_strerror(ret));                      \
302             exit(1);                                               \
303         }                                                          \
304     } while (0)
305 
306 static void add_str(GString *s, const gchar *s1)
307 {
308     g_string_append_len(s, s1, strlen(s1) + 1);
309 }
310 
311 static void *spapr_create_fdt_skel(hwaddr initrd_base,
312                                    hwaddr initrd_size,
313                                    hwaddr kernel_size,
314                                    bool little_endian,
315                                    const char *kernel_cmdline,
316                                    uint32_t epow_irq)
317 {
318     void *fdt;
319     uint32_t start_prop = cpu_to_be32(initrd_base);
320     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
321     GString *hypertas = g_string_sized_new(256);
322     GString *qemu_hypertas = g_string_sized_new(256);
323     uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
324     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(max_cpus)};
325     unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
326     char *buf;
327 
328     add_str(hypertas, "hcall-pft");
329     add_str(hypertas, "hcall-term");
330     add_str(hypertas, "hcall-dabr");
331     add_str(hypertas, "hcall-interrupt");
332     add_str(hypertas, "hcall-tce");
333     add_str(hypertas, "hcall-vio");
334     add_str(hypertas, "hcall-splpar");
335     add_str(hypertas, "hcall-bulk");
336     add_str(hypertas, "hcall-set-mode");
337     add_str(qemu_hypertas, "hcall-memop1");
338 
339     fdt = g_malloc0(FDT_MAX_SIZE);
340     _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
341 
342     if (kernel_size) {
343         _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
344     }
345     if (initrd_size) {
346         _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
347     }
348     _FDT((fdt_finish_reservemap(fdt)));
349 
350     /* Root node */
351     _FDT((fdt_begin_node(fdt, "")));
352     _FDT((fdt_property_string(fdt, "device_type", "chrp")));
353     _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
354     _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
355 
356     /*
357      * Add info to guest to indentify which host is it being run on
358      * and what is the uuid of the guest
359      */
360     if (kvmppc_get_host_model(&buf)) {
361         _FDT((fdt_property_string(fdt, "host-model", buf)));
362         g_free(buf);
363     }
364     if (kvmppc_get_host_serial(&buf)) {
365         _FDT((fdt_property_string(fdt, "host-serial", buf)));
366         g_free(buf);
367     }
368 
369     buf = g_strdup_printf(UUID_FMT, qemu_uuid[0], qemu_uuid[1],
370                           qemu_uuid[2], qemu_uuid[3], qemu_uuid[4],
371                           qemu_uuid[5], qemu_uuid[6], qemu_uuid[7],
372                           qemu_uuid[8], qemu_uuid[9], qemu_uuid[10],
373                           qemu_uuid[11], qemu_uuid[12], qemu_uuid[13],
374                           qemu_uuid[14], qemu_uuid[15]);
375 
376     _FDT((fdt_property_string(fdt, "vm,uuid", buf)));
377     g_free(buf);
378 
379     if (qemu_get_vm_name()) {
380         _FDT((fdt_property_string(fdt, "ibm,partition-name",
381                                   qemu_get_vm_name())));
382     }
383 
384     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
385     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
386 
387     /* /chosen */
388     _FDT((fdt_begin_node(fdt, "chosen")));
389 
390     /* Set Form1_affinity */
391     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
392 
393     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
394     _FDT((fdt_property(fdt, "linux,initrd-start",
395                        &start_prop, sizeof(start_prop))));
396     _FDT((fdt_property(fdt, "linux,initrd-end",
397                        &end_prop, sizeof(end_prop))));
398     if (kernel_size) {
399         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
400                               cpu_to_be64(kernel_size) };
401 
402         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
403         if (little_endian) {
404             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
405         }
406     }
407     if (boot_menu) {
408         _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
409     }
410     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
411     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
412     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
413 
414     _FDT((fdt_end_node(fdt)));
415 
416     /* RTAS */
417     _FDT((fdt_begin_node(fdt, "rtas")));
418 
419     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
420         add_str(hypertas, "hcall-multi-tce");
421     }
422     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
423                        hypertas->len)));
424     g_string_free(hypertas, TRUE);
425     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
426                        qemu_hypertas->len)));
427     g_string_free(qemu_hypertas, TRUE);
428 
429     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
430         refpoints, sizeof(refpoints))));
431 
432     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
433     _FDT((fdt_property_cell(fdt, "rtas-event-scan-rate",
434                             RTAS_EVENT_SCAN_RATE)));
435 
436     if (msi_supported) {
437         _FDT((fdt_property(fdt, "ibm,change-msix-capable", NULL, 0)));
438     }
439 
440     /*
441      * According to PAPR, rtas ibm,os-term does not guarantee a return
442      * back to the guest cpu.
443      *
444      * While an additional ibm,extended-os-term property indicates that
445      * rtas call return will always occur. Set this property.
446      */
447     _FDT((fdt_property(fdt, "ibm,extended-os-term", NULL, 0)));
448 
449     _FDT((fdt_end_node(fdt)));
450 
451     /* interrupt controller */
452     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
453 
454     _FDT((fdt_property_string(fdt, "device_type",
455                               "PowerPC-External-Interrupt-Presentation")));
456     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
457     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
458     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
459                        interrupt_server_ranges_prop,
460                        sizeof(interrupt_server_ranges_prop))));
461     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
462     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
463     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
464 
465     _FDT((fdt_end_node(fdt)));
466 
467     /* vdevice */
468     _FDT((fdt_begin_node(fdt, "vdevice")));
469 
470     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
471     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
472     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
473     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
474     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
475     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
476 
477     _FDT((fdt_end_node(fdt)));
478 
479     /* event-sources */
480     spapr_events_fdt_skel(fdt, epow_irq);
481 
482     /* /hypervisor node */
483     if (kvm_enabled()) {
484         uint8_t hypercall[16];
485 
486         /* indicate KVM hypercall interface */
487         _FDT((fdt_begin_node(fdt, "hypervisor")));
488         _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
489         if (kvmppc_has_cap_fixup_hcalls()) {
490             /*
491              * Older KVM versions with older guest kernels were broken with the
492              * magic page, don't allow the guest to map it.
493              */
494             kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
495                                  sizeof(hypercall));
496             _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
497                               sizeof(hypercall))));
498         }
499         _FDT((fdt_end_node(fdt)));
500     }
501 
502     _FDT((fdt_end_node(fdt))); /* close root node */
503     _FDT((fdt_finish(fdt)));
504 
505     return fdt;
506 }
507 
508 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
509                                        hwaddr size)
510 {
511     uint32_t associativity[] = {
512         cpu_to_be32(0x4), /* length */
513         cpu_to_be32(0x0), cpu_to_be32(0x0),
514         cpu_to_be32(0x0), cpu_to_be32(nodeid)
515     };
516     char mem_name[32];
517     uint64_t mem_reg_property[2];
518     int off;
519 
520     mem_reg_property[0] = cpu_to_be64(start);
521     mem_reg_property[1] = cpu_to_be64(size);
522 
523     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
524     off = fdt_add_subnode(fdt, 0, mem_name);
525     _FDT(off);
526     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
527     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
528                       sizeof(mem_reg_property))));
529     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
530                       sizeof(associativity))));
531     return off;
532 }
533 
534 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
535 {
536     MachineState *machine = MACHINE(spapr);
537     hwaddr mem_start, node_size;
538     int i, nb_nodes = nb_numa_nodes;
539     NodeInfo *nodes = numa_info;
540     NodeInfo ramnode;
541 
542     /* No NUMA nodes, assume there is just one node with whole RAM */
543     if (!nb_numa_nodes) {
544         nb_nodes = 1;
545         ramnode.node_mem = machine->ram_size;
546         nodes = &ramnode;
547     }
548 
549     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
550         if (!nodes[i].node_mem) {
551             continue;
552         }
553         if (mem_start >= machine->ram_size) {
554             node_size = 0;
555         } else {
556             node_size = nodes[i].node_mem;
557             if (node_size > machine->ram_size - mem_start) {
558                 node_size = machine->ram_size - mem_start;
559             }
560         }
561         if (!mem_start) {
562             /* ppc_spapr_init() checks for rma_size <= node0_size already */
563             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
564             mem_start += spapr->rma_size;
565             node_size -= spapr->rma_size;
566         }
567         for ( ; node_size; ) {
568             hwaddr sizetmp = pow2floor(node_size);
569 
570             /* mem_start != 0 here */
571             if (ctzl(mem_start) < ctzl(sizetmp)) {
572                 sizetmp = 1ULL << ctzl(mem_start);
573             }
574 
575             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
576             node_size -= sizetmp;
577             mem_start += sizetmp;
578         }
579     }
580 
581     return 0;
582 }
583 
584 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
585                                   sPAPRMachineState *spapr)
586 {
587     PowerPCCPU *cpu = POWERPC_CPU(cs);
588     CPUPPCState *env = &cpu->env;
589     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
590     int index = ppc_get_vcpu_dt_id(cpu);
591     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
592                        0xffffffff, 0xffffffff};
593     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
594     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
595     uint32_t page_sizes_prop[64];
596     size_t page_sizes_prop_size;
597     uint32_t vcpus_per_socket = smp_threads * smp_cores;
598     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
599 
600     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
601     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
602 
603     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
604     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
605                            env->dcache_line_size)));
606     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
607                            env->dcache_line_size)));
608     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
609                            env->icache_line_size)));
610     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
611                            env->icache_line_size)));
612 
613     if (pcc->l1_dcache_size) {
614         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
615                                pcc->l1_dcache_size)));
616     } else {
617         fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
618     }
619     if (pcc->l1_icache_size) {
620         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
621                                pcc->l1_icache_size)));
622     } else {
623         fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
624     }
625 
626     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
627     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
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 = 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     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
666                            cs->cpu_index / vcpus_per_socket)));
667 
668     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
669                       pft_size_prop, sizeof(pft_size_prop))));
670 
671     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
672 
673     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
674                                 ppc_get_compat_smt_threads(cpu)));
675 }
676 
677 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
678 {
679     CPUState *cs;
680     int cpus_offset;
681     char *nodename;
682     int smt = kvmppc_smt_threads();
683 
684     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
685     _FDT(cpus_offset);
686     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
687     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
688 
689     /*
690      * We walk the CPUs in reverse order to ensure that CPU DT nodes
691      * created by fdt_add_subnode() end up in the right order in FDT
692      * for the guest kernel the enumerate the CPUs correctly.
693      */
694     CPU_FOREACH_REVERSE(cs) {
695         PowerPCCPU *cpu = POWERPC_CPU(cs);
696         int index = ppc_get_vcpu_dt_id(cpu);
697         DeviceClass *dc = DEVICE_GET_CLASS(cs);
698         int offset;
699 
700         if ((index % smt) != 0) {
701             continue;
702         }
703 
704         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
705         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
706         g_free(nodename);
707         _FDT(offset);
708         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
709     }
710 
711 }
712 
713 /*
714  * Adds ibm,dynamic-reconfiguration-memory node.
715  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
716  * of this device tree node.
717  */
718 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
719 {
720     MachineState *machine = MACHINE(spapr);
721     int ret, i, offset;
722     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
723     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
724     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
725     uint32_t *int_buf, *cur_index, buf_len;
726     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
727 
728     /*
729      * Allocate enough buffer size to fit in ibm,dynamic-memory
730      * or ibm,associativity-lookup-arrays
731      */
732     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
733               * sizeof(uint32_t);
734     cur_index = int_buf = g_malloc0(buf_len);
735 
736     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
737 
738     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
739                     sizeof(prop_lmb_size));
740     if (ret < 0) {
741         goto out;
742     }
743 
744     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
745     if (ret < 0) {
746         goto out;
747     }
748 
749     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
750     if (ret < 0) {
751         goto out;
752     }
753 
754     /* ibm,dynamic-memory */
755     int_buf[0] = cpu_to_be32(nr_lmbs);
756     cur_index++;
757     for (i = 0; i < nr_lmbs; i++) {
758         sPAPRDRConnector *drc;
759         sPAPRDRConnectorClass *drck;
760         uint64_t addr = i * lmb_size + spapr->hotplug_memory.base;;
761         uint32_t *dynamic_memory = cur_index;
762 
763         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
764                                        addr/lmb_size);
765         g_assert(drc);
766         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
767 
768         dynamic_memory[0] = cpu_to_be32(addr >> 32);
769         dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
770         dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
771         dynamic_memory[3] = cpu_to_be32(0); /* reserved */
772         dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
773         if (addr < machine->ram_size ||
774                     memory_region_present(get_system_memory(), addr)) {
775             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
776         } else {
777             dynamic_memory[5] = cpu_to_be32(0);
778         }
779 
780         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
781     }
782     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
783     if (ret < 0) {
784         goto out;
785     }
786 
787     /* ibm,associativity-lookup-arrays */
788     cur_index = int_buf;
789     int_buf[0] = cpu_to_be32(nr_nodes);
790     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
791     cur_index += 2;
792     for (i = 0; i < nr_nodes; i++) {
793         uint32_t associativity[] = {
794             cpu_to_be32(0x0),
795             cpu_to_be32(0x0),
796             cpu_to_be32(0x0),
797             cpu_to_be32(i)
798         };
799         memcpy(cur_index, associativity, sizeof(associativity));
800         cur_index += 4;
801     }
802     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
803             (cur_index - int_buf) * sizeof(uint32_t));
804 out:
805     g_free(int_buf);
806     return ret;
807 }
808 
809 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
810                                  target_ulong addr, target_ulong size,
811                                  bool cpu_update, bool memory_update)
812 {
813     void *fdt, *fdt_skel;
814     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
815     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
816 
817     size -= sizeof(hdr);
818 
819     /* Create sceleton */
820     fdt_skel = g_malloc0(size);
821     _FDT((fdt_create(fdt_skel, size)));
822     _FDT((fdt_begin_node(fdt_skel, "")));
823     _FDT((fdt_end_node(fdt_skel)));
824     _FDT((fdt_finish(fdt_skel)));
825     fdt = g_malloc0(size);
826     _FDT((fdt_open_into(fdt_skel, fdt, size)));
827     g_free(fdt_skel);
828 
829     /* Fixup cpu nodes */
830     if (cpu_update) {
831         _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
832     }
833 
834     /* Generate memory nodes or ibm,dynamic-reconfiguration-memory node */
835     if (memory_update && smc->dr_lmb_enabled) {
836         _FDT((spapr_populate_drconf_memory(spapr, fdt)));
837     }
838 
839     /* Pack resulting tree */
840     _FDT((fdt_pack(fdt)));
841 
842     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
843         trace_spapr_cas_failed(size);
844         return -1;
845     }
846 
847     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
848     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
849     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
850     g_free(fdt);
851 
852     return 0;
853 }
854 
855 static void spapr_finalize_fdt(sPAPRMachineState *spapr,
856                                hwaddr fdt_addr,
857                                hwaddr rtas_addr,
858                                hwaddr rtas_size)
859 {
860     MachineState *machine = MACHINE(qdev_get_machine());
861     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
862     const char *boot_device = machine->boot_order;
863     int ret, i;
864     size_t cb = 0;
865     char *bootlist;
866     void *fdt;
867     sPAPRPHBState *phb;
868 
869     fdt = g_malloc(FDT_MAX_SIZE);
870 
871     /* open out the base tree into a temp buffer for the final tweaks */
872     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
873 
874     ret = spapr_populate_memory(spapr, fdt);
875     if (ret < 0) {
876         fprintf(stderr, "couldn't setup memory nodes in fdt\n");
877         exit(1);
878     }
879 
880     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
881     if (ret < 0) {
882         fprintf(stderr, "couldn't setup vio devices in fdt\n");
883         exit(1);
884     }
885 
886     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
887         ret = spapr_rng_populate_dt(fdt);
888         if (ret < 0) {
889             fprintf(stderr, "could not set up rng device in the fdt\n");
890             exit(1);
891         }
892     }
893 
894     QLIST_FOREACH(phb, &spapr->phbs, list) {
895         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
896     }
897 
898     if (ret < 0) {
899         fprintf(stderr, "couldn't setup PCI devices in fdt\n");
900         exit(1);
901     }
902 
903     /* RTAS */
904     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
905     if (ret < 0) {
906         fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
907     }
908 
909     /* cpus */
910     spapr_populate_cpus_dt_node(fdt, spapr);
911 
912     bootlist = get_boot_devices_list(&cb, true);
913     if (cb && bootlist) {
914         int offset = fdt_path_offset(fdt, "/chosen");
915         if (offset < 0) {
916             exit(1);
917         }
918         for (i = 0; i < cb; i++) {
919             if (bootlist[i] == '\n') {
920                 bootlist[i] = ' ';
921             }
922 
923         }
924         ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
925     }
926 
927     if (boot_device && strlen(boot_device)) {
928         int offset = fdt_path_offset(fdt, "/chosen");
929 
930         if (offset < 0) {
931             exit(1);
932         }
933         fdt_setprop_string(fdt, offset, "qemu,boot-device", boot_device);
934     }
935 
936     if (!spapr->has_graphics) {
937         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
938     }
939 
940     if (smc->dr_lmb_enabled) {
941         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
942     }
943 
944     _FDT((fdt_pack(fdt)));
945 
946     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
947         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
948                      fdt_totalsize(fdt), FDT_MAX_SIZE);
949         exit(1);
950     }
951 
952     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
953     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
954 
955     g_free(bootlist);
956     g_free(fdt);
957 }
958 
959 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
960 {
961     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
962 }
963 
964 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
965 {
966     CPUPPCState *env = &cpu->env;
967 
968     if (msr_pr) {
969         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
970         env->gpr[3] = H_PRIVILEGE;
971     } else {
972         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
973     }
974 }
975 
976 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
977 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
978 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
979 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
980 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
981 
982 static void spapr_reset_htab(sPAPRMachineState *spapr)
983 {
984     long shift;
985     int index;
986 
987     /* allocate hash page table.  For now we always make this 16mb,
988      * later we should probably make it scale to the size of guest
989      * RAM */
990 
991     shift = kvmppc_reset_htab(spapr->htab_shift);
992 
993     if (shift > 0) {
994         /* Kernel handles htab, we don't need to allocate one */
995         spapr->htab_shift = shift;
996         kvmppc_kern_htab = true;
997 
998         /* Tell readers to update their file descriptor */
999         if (spapr->htab_fd >= 0) {
1000             spapr->htab_fd_stale = true;
1001         }
1002     } else {
1003         if (!spapr->htab) {
1004             /* Allocate an htab if we don't yet have one */
1005             spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
1006         }
1007 
1008         /* And clear it */
1009         memset(spapr->htab, 0, HTAB_SIZE(spapr));
1010 
1011         for (index = 0; index < HTAB_SIZE(spapr) / HASH_PTE_SIZE_64; index++) {
1012             DIRTY_HPTE(HPTE(spapr->htab, index));
1013         }
1014     }
1015 
1016     /* Update the RMA size if necessary */
1017     if (spapr->vrma_adjust) {
1018         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1019                                           spapr->htab_shift);
1020     }
1021 }
1022 
1023 static int find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1024 {
1025     bool matched = false;
1026 
1027     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1028         matched = true;
1029     }
1030 
1031     if (!matched) {
1032         error_report("Device %s is not supported by this machine yet.",
1033                      qdev_fw_name(DEVICE(sbdev)));
1034         exit(1);
1035     }
1036 
1037     return 0;
1038 }
1039 
1040 /*
1041  * A guest reset will cause spapr->htab_fd to become stale if being used.
1042  * Reopen the file descriptor to make sure the whole HTAB is properly read.
1043  */
1044 static int spapr_check_htab_fd(sPAPRMachineState *spapr)
1045 {
1046     int rc = 0;
1047 
1048     if (spapr->htab_fd_stale) {
1049         close(spapr->htab_fd);
1050         spapr->htab_fd = kvmppc_get_htab_fd(false);
1051         if (spapr->htab_fd < 0) {
1052             error_report("Unable to open fd for reading hash table from KVM: "
1053                          "%s", strerror(errno));
1054             rc = -1;
1055         }
1056         spapr->htab_fd_stale = false;
1057     }
1058 
1059     return rc;
1060 }
1061 
1062 static void ppc_spapr_reset(void)
1063 {
1064     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1065     PowerPCCPU *first_ppc_cpu;
1066     uint32_t rtas_limit;
1067 
1068     /* Check for unknown sysbus devices */
1069     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1070 
1071     /* Reset the hash table & recalc the RMA */
1072     spapr_reset_htab(spapr);
1073 
1074     qemu_devices_reset();
1075 
1076     /*
1077      * We place the device tree and RTAS just below either the top of the RMA,
1078      * or just below 2GB, whichever is lowere, so that it can be
1079      * processed with 32-bit real mode code if necessary
1080      */
1081     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1082     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1083     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1084 
1085     /* Load the fdt */
1086     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
1087                        spapr->rtas_size);
1088 
1089     /* Copy RTAS over */
1090     cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
1091                               spapr->rtas_size);
1092 
1093     /* Set up the entry state */
1094     first_ppc_cpu = POWERPC_CPU(first_cpu);
1095     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
1096     first_ppc_cpu->env.gpr[5] = 0;
1097     first_cpu->halted = 0;
1098     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1099 
1100 }
1101 
1102 static void spapr_cpu_reset(void *opaque)
1103 {
1104     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
1105     PowerPCCPU *cpu = opaque;
1106     CPUState *cs = CPU(cpu);
1107     CPUPPCState *env = &cpu->env;
1108 
1109     cpu_reset(cs);
1110 
1111     /* All CPUs start halted.  CPU0 is unhalted from the machine level
1112      * reset code and the rest are explicitly started up by the guest
1113      * using an RTAS call */
1114     cs->halted = 1;
1115 
1116     env->spr[SPR_HIOR] = 0;
1117 
1118     env->external_htab = (uint8_t *)spapr->htab;
1119     if (kvm_enabled() && !env->external_htab) {
1120         /*
1121          * HV KVM, set external_htab to 1 so our ppc_hash64_load_hpte*
1122          * functions do the right thing.
1123          */
1124         env->external_htab = (void *)1;
1125     }
1126     env->htab_base = -1;
1127     /*
1128      * htab_mask is the mask used to normalize hash value to PTEG index.
1129      * htab_shift is log2 of hash table size.
1130      * We have 8 hpte per group, and each hpte is 16 bytes.
1131      * ie have 128 bytes per hpte entry.
1132      */
1133     env->htab_mask = (1ULL << (spapr->htab_shift - 7)) - 1;
1134     env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
1135         (spapr->htab_shift - 18);
1136 }
1137 
1138 static void spapr_create_nvram(sPAPRMachineState *spapr)
1139 {
1140     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1141     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1142 
1143     if (dinfo) {
1144         qdev_prop_set_drive_nofail(dev, "drive", blk_by_legacy_dinfo(dinfo));
1145     }
1146 
1147     qdev_init_nofail(dev);
1148 
1149     spapr->nvram = (struct sPAPRNVRAM *)dev;
1150 }
1151 
1152 static void spapr_rtc_create(sPAPRMachineState *spapr)
1153 {
1154     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1155 
1156     qdev_init_nofail(dev);
1157     spapr->rtc = dev;
1158 
1159     object_property_add_alias(qdev_get_machine(), "rtc-time",
1160                               OBJECT(spapr->rtc), "date", NULL);
1161 }
1162 
1163 /* Returns whether we want to use VGA or not */
1164 static int spapr_vga_init(PCIBus *pci_bus)
1165 {
1166     switch (vga_interface_type) {
1167     case VGA_NONE:
1168         return false;
1169     case VGA_DEVICE:
1170         return true;
1171     case VGA_STD:
1172         return pci_vga_init(pci_bus) != NULL;
1173     default:
1174         fprintf(stderr, "This vga model is not supported,"
1175                 "currently it only supports -vga std\n");
1176         exit(0);
1177     }
1178 }
1179 
1180 static int spapr_post_load(void *opaque, int version_id)
1181 {
1182     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1183     int err = 0;
1184 
1185     /* In earlier versions, there was no separate qdev for the PAPR
1186      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1187      * So when migrating from those versions, poke the incoming offset
1188      * value into the RTC device */
1189     if (version_id < 3) {
1190         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1191     }
1192 
1193     return err;
1194 }
1195 
1196 static bool version_before_3(void *opaque, int version_id)
1197 {
1198     return version_id < 3;
1199 }
1200 
1201 static const VMStateDescription vmstate_spapr = {
1202     .name = "spapr",
1203     .version_id = 3,
1204     .minimum_version_id = 1,
1205     .post_load = spapr_post_load,
1206     .fields = (VMStateField[]) {
1207         /* used to be @next_irq */
1208         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1209 
1210         /* RTC offset */
1211         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1212 
1213         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1214         VMSTATE_END_OF_LIST()
1215     },
1216 };
1217 
1218 static int htab_save_setup(QEMUFile *f, void *opaque)
1219 {
1220     sPAPRMachineState *spapr = opaque;
1221 
1222     /* "Iteration" header */
1223     qemu_put_be32(f, spapr->htab_shift);
1224 
1225     if (spapr->htab) {
1226         spapr->htab_save_index = 0;
1227         spapr->htab_first_pass = true;
1228     } else {
1229         assert(kvm_enabled());
1230 
1231         spapr->htab_fd = kvmppc_get_htab_fd(false);
1232         spapr->htab_fd_stale = false;
1233         if (spapr->htab_fd < 0) {
1234             fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
1235                     strerror(errno));
1236             return -1;
1237         }
1238     }
1239 
1240 
1241     return 0;
1242 }
1243 
1244 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1245                                  int64_t max_ns)
1246 {
1247     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1248     int index = spapr->htab_save_index;
1249     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1250 
1251     assert(spapr->htab_first_pass);
1252 
1253     do {
1254         int chunkstart;
1255 
1256         /* Consume invalid HPTEs */
1257         while ((index < htabslots)
1258                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1259             index++;
1260             CLEAN_HPTE(HPTE(spapr->htab, index));
1261         }
1262 
1263         /* Consume valid HPTEs */
1264         chunkstart = index;
1265         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1266                && HPTE_VALID(HPTE(spapr->htab, index))) {
1267             index++;
1268             CLEAN_HPTE(HPTE(spapr->htab, index));
1269         }
1270 
1271         if (index > chunkstart) {
1272             int n_valid = index - chunkstart;
1273 
1274             qemu_put_be32(f, chunkstart);
1275             qemu_put_be16(f, n_valid);
1276             qemu_put_be16(f, 0);
1277             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1278                             HASH_PTE_SIZE_64 * n_valid);
1279 
1280             if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1281                 break;
1282             }
1283         }
1284     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1285 
1286     if (index >= htabslots) {
1287         assert(index == htabslots);
1288         index = 0;
1289         spapr->htab_first_pass = false;
1290     }
1291     spapr->htab_save_index = index;
1292 }
1293 
1294 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1295                                 int64_t max_ns)
1296 {
1297     bool final = max_ns < 0;
1298     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1299     int examined = 0, sent = 0;
1300     int index = spapr->htab_save_index;
1301     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1302 
1303     assert(!spapr->htab_first_pass);
1304 
1305     do {
1306         int chunkstart, invalidstart;
1307 
1308         /* Consume non-dirty HPTEs */
1309         while ((index < htabslots)
1310                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1311             index++;
1312             examined++;
1313         }
1314 
1315         chunkstart = index;
1316         /* Consume valid dirty HPTEs */
1317         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1318                && HPTE_DIRTY(HPTE(spapr->htab, index))
1319                && HPTE_VALID(HPTE(spapr->htab, index))) {
1320             CLEAN_HPTE(HPTE(spapr->htab, index));
1321             index++;
1322             examined++;
1323         }
1324 
1325         invalidstart = index;
1326         /* Consume invalid dirty HPTEs */
1327         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1328                && HPTE_DIRTY(HPTE(spapr->htab, index))
1329                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1330             CLEAN_HPTE(HPTE(spapr->htab, index));
1331             index++;
1332             examined++;
1333         }
1334 
1335         if (index > chunkstart) {
1336             int n_valid = invalidstart - chunkstart;
1337             int n_invalid = index - invalidstart;
1338 
1339             qemu_put_be32(f, chunkstart);
1340             qemu_put_be16(f, n_valid);
1341             qemu_put_be16(f, n_invalid);
1342             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1343                             HASH_PTE_SIZE_64 * n_valid);
1344             sent += index - chunkstart;
1345 
1346             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1347                 break;
1348             }
1349         }
1350 
1351         if (examined >= htabslots) {
1352             break;
1353         }
1354 
1355         if (index >= htabslots) {
1356             assert(index == htabslots);
1357             index = 0;
1358         }
1359     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1360 
1361     if (index >= htabslots) {
1362         assert(index == htabslots);
1363         index = 0;
1364     }
1365 
1366     spapr->htab_save_index = index;
1367 
1368     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1369 }
1370 
1371 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1372 #define MAX_KVM_BUF_SIZE    2048
1373 
1374 static int htab_save_iterate(QEMUFile *f, void *opaque)
1375 {
1376     sPAPRMachineState *spapr = opaque;
1377     int rc = 0;
1378 
1379     /* Iteration header */
1380     qemu_put_be32(f, 0);
1381 
1382     if (!spapr->htab) {
1383         assert(kvm_enabled());
1384 
1385         rc = spapr_check_htab_fd(spapr);
1386         if (rc < 0) {
1387             return rc;
1388         }
1389 
1390         rc = kvmppc_save_htab(f, spapr->htab_fd,
1391                               MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1392         if (rc < 0) {
1393             return rc;
1394         }
1395     } else  if (spapr->htab_first_pass) {
1396         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1397     } else {
1398         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1399     }
1400 
1401     /* End marker */
1402     qemu_put_be32(f, 0);
1403     qemu_put_be16(f, 0);
1404     qemu_put_be16(f, 0);
1405 
1406     return rc;
1407 }
1408 
1409 static int htab_save_complete(QEMUFile *f, void *opaque)
1410 {
1411     sPAPRMachineState *spapr = opaque;
1412 
1413     /* Iteration header */
1414     qemu_put_be32(f, 0);
1415 
1416     if (!spapr->htab) {
1417         int rc;
1418 
1419         assert(kvm_enabled());
1420 
1421         rc = spapr_check_htab_fd(spapr);
1422         if (rc < 0) {
1423             return rc;
1424         }
1425 
1426         rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
1427         if (rc < 0) {
1428             return rc;
1429         }
1430         close(spapr->htab_fd);
1431         spapr->htab_fd = -1;
1432     } else {
1433         htab_save_later_pass(f, spapr, -1);
1434     }
1435 
1436     /* End marker */
1437     qemu_put_be32(f, 0);
1438     qemu_put_be16(f, 0);
1439     qemu_put_be16(f, 0);
1440 
1441     return 0;
1442 }
1443 
1444 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1445 {
1446     sPAPRMachineState *spapr = opaque;
1447     uint32_t section_hdr;
1448     int fd = -1;
1449 
1450     if (version_id < 1 || version_id > 1) {
1451         fprintf(stderr, "htab_load() bad version\n");
1452         return -EINVAL;
1453     }
1454 
1455     section_hdr = qemu_get_be32(f);
1456 
1457     if (section_hdr) {
1458         /* First section, just the hash shift */
1459         if (spapr->htab_shift != section_hdr) {
1460             error_report("htab_shift mismatch: source %d target %d",
1461                          section_hdr, spapr->htab_shift);
1462             return -EINVAL;
1463         }
1464         return 0;
1465     }
1466 
1467     if (!spapr->htab) {
1468         assert(kvm_enabled());
1469 
1470         fd = kvmppc_get_htab_fd(true);
1471         if (fd < 0) {
1472             fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1473                     strerror(errno));
1474         }
1475     }
1476 
1477     while (true) {
1478         uint32_t index;
1479         uint16_t n_valid, n_invalid;
1480 
1481         index = qemu_get_be32(f);
1482         n_valid = qemu_get_be16(f);
1483         n_invalid = qemu_get_be16(f);
1484 
1485         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1486             /* End of Stream */
1487             break;
1488         }
1489 
1490         if ((index + n_valid + n_invalid) >
1491             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1492             /* Bad index in stream */
1493             fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1494                     "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1495                     spapr->htab_shift);
1496             return -EINVAL;
1497         }
1498 
1499         if (spapr->htab) {
1500             if (n_valid) {
1501                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1502                                 HASH_PTE_SIZE_64 * n_valid);
1503             }
1504             if (n_invalid) {
1505                 memset(HPTE(spapr->htab, index + n_valid), 0,
1506                        HASH_PTE_SIZE_64 * n_invalid);
1507             }
1508         } else {
1509             int rc;
1510 
1511             assert(fd >= 0);
1512 
1513             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1514             if (rc < 0) {
1515                 return rc;
1516             }
1517         }
1518     }
1519 
1520     if (!spapr->htab) {
1521         assert(fd >= 0);
1522         close(fd);
1523     }
1524 
1525     return 0;
1526 }
1527 
1528 static SaveVMHandlers savevm_htab_handlers = {
1529     .save_live_setup = htab_save_setup,
1530     .save_live_iterate = htab_save_iterate,
1531     .save_live_complete = htab_save_complete,
1532     .load_state = htab_load,
1533 };
1534 
1535 static void spapr_boot_set(void *opaque, const char *boot_device,
1536                            Error **errp)
1537 {
1538     MachineState *machine = MACHINE(qdev_get_machine());
1539     machine->boot_order = g_strdup(boot_device);
1540 }
1541 
1542 static void spapr_cpu_init(sPAPRMachineState *spapr, PowerPCCPU *cpu)
1543 {
1544     CPUPPCState *env = &cpu->env;
1545 
1546     /* Set time-base frequency to 512 MHz */
1547     cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1548 
1549     /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1550      * MSR[IP] should never be set.
1551      */
1552     env->msr_mask &= ~(1 << 6);
1553 
1554     /* Tell KVM that we're in PAPR mode */
1555     if (kvm_enabled()) {
1556         kvmppc_set_papr(cpu);
1557     }
1558 
1559     if (cpu->max_compat) {
1560         if (ppc_set_compat(cpu, cpu->max_compat) < 0) {
1561             exit(1);
1562         }
1563     }
1564 
1565     xics_cpu_setup(spapr->icp, cpu);
1566 
1567     qemu_register_reset(spapr_cpu_reset, cpu);
1568 }
1569 
1570 /*
1571  * Reset routine for LMB DR devices.
1572  *
1573  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1574  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1575  * when it walks all its children devices. LMB devices reset occurs
1576  * as part of spapr_ppc_reset().
1577  */
1578 static void spapr_drc_reset(void *opaque)
1579 {
1580     sPAPRDRConnector *drc = opaque;
1581     DeviceState *d = DEVICE(drc);
1582 
1583     if (d) {
1584         device_reset(d);
1585     }
1586 }
1587 
1588 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1589 {
1590     MachineState *machine = MACHINE(spapr);
1591     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1592     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1593     int i;
1594 
1595     for (i = 0; i < nr_lmbs; i++) {
1596         sPAPRDRConnector *drc;
1597         uint64_t addr;
1598 
1599         addr = i * lmb_size + spapr->hotplug_memory.base;
1600         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1601                                      addr/lmb_size);
1602         qemu_register_reset(spapr_drc_reset, drc);
1603     }
1604 }
1605 
1606 /*
1607  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1608  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1609  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1610  */
1611 static void spapr_validate_node_memory(MachineState *machine)
1612 {
1613     int i;
1614 
1615     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE ||
1616         machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1617         error_report("Can't support memory configuration where RAM size "
1618                      "0x" RAM_ADDR_FMT " or maxmem size "
1619                      "0x" RAM_ADDR_FMT " isn't aligned to %llu MB",
1620                      machine->ram_size, machine->maxram_size,
1621                      SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
1622         exit(EXIT_FAILURE);
1623     }
1624 
1625     for (i = 0; i < nb_numa_nodes; i++) {
1626         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1627             error_report("Can't support memory configuration where memory size"
1628                          " %" PRIx64 " of node %d isn't aligned to %llu MB",
1629                          numa_info[i].node_mem, i,
1630                          SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
1631             exit(EXIT_FAILURE);
1632         }
1633     }
1634 }
1635 
1636 /* pSeries LPAR / sPAPR hardware init */
1637 static void ppc_spapr_init(MachineState *machine)
1638 {
1639     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1640     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1641     const char *kernel_filename = machine->kernel_filename;
1642     const char *kernel_cmdline = machine->kernel_cmdline;
1643     const char *initrd_filename = machine->initrd_filename;
1644     PowerPCCPU *cpu;
1645     PCIHostState *phb;
1646     int i;
1647     MemoryRegion *sysmem = get_system_memory();
1648     MemoryRegion *ram = g_new(MemoryRegion, 1);
1649     MemoryRegion *rma_region;
1650     void *rma = NULL;
1651     hwaddr rma_alloc_size;
1652     hwaddr node0_size = spapr_node0_size();
1653     uint32_t initrd_base = 0;
1654     long kernel_size = 0, initrd_size = 0;
1655     long load_limit, fw_size;
1656     bool kernel_le = false;
1657     char *filename;
1658 
1659     msi_supported = true;
1660 
1661     QLIST_INIT(&spapr->phbs);
1662 
1663     cpu_ppc_hypercall = emulate_spapr_hypercall;
1664 
1665     /* Allocate RMA if necessary */
1666     rma_alloc_size = kvmppc_alloc_rma(&rma);
1667 
1668     if (rma_alloc_size == -1) {
1669         error_report("Unable to create RMA");
1670         exit(1);
1671     }
1672 
1673     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1674         spapr->rma_size = rma_alloc_size;
1675     } else {
1676         spapr->rma_size = node0_size;
1677 
1678         /* With KVM, we don't actually know whether KVM supports an
1679          * unbounded RMA (PR KVM) or is limited by the hash table size
1680          * (HV KVM using VRMA), so we always assume the latter
1681          *
1682          * In that case, we also limit the initial allocations for RTAS
1683          * etc... to 256M since we have no way to know what the VRMA size
1684          * is going to be as it depends on the size of the hash table
1685          * isn't determined yet.
1686          */
1687         if (kvm_enabled()) {
1688             spapr->vrma_adjust = 1;
1689             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1690         }
1691     }
1692 
1693     if (spapr->rma_size > node0_size) {
1694         fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1695                 spapr->rma_size);
1696         exit(1);
1697     }
1698 
1699     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1700     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1701 
1702     /* We aim for a hash table of size 1/128 the size of RAM.  The
1703      * normal rule of thumb is 1/64 the size of RAM, but that's much
1704      * more than needed for the Linux guests we support. */
1705     spapr->htab_shift = 18; /* Minimum architected size */
1706     while (spapr->htab_shift <= 46) {
1707         if ((1ULL << (spapr->htab_shift + 7)) >= machine->maxram_size) {
1708             break;
1709         }
1710         spapr->htab_shift++;
1711     }
1712 
1713     /* Set up Interrupt Controller before we create the VCPUs */
1714     spapr->icp = xics_system_init(machine,
1715                                   DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(),
1716                                                smp_threads),
1717                                   XICS_IRQS);
1718 
1719     if (smc->dr_lmb_enabled) {
1720         spapr_validate_node_memory(machine);
1721     }
1722 
1723     /* init CPUs */
1724     if (machine->cpu_model == NULL) {
1725         machine->cpu_model = kvm_enabled() ? "host" : "POWER7";
1726     }
1727     for (i = 0; i < smp_cpus; i++) {
1728         cpu = cpu_ppc_init(machine->cpu_model);
1729         if (cpu == NULL) {
1730             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1731             exit(1);
1732         }
1733         spapr_cpu_init(spapr, cpu);
1734     }
1735 
1736     if (kvm_enabled()) {
1737         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1738         kvmppc_enable_logical_ci_hcalls();
1739         kvmppc_enable_set_mode_hcall();
1740     }
1741 
1742     /* allocate RAM */
1743     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1744                                          machine->ram_size);
1745     memory_region_add_subregion(sysmem, 0, ram);
1746 
1747     if (rma_alloc_size && rma) {
1748         rma_region = g_new(MemoryRegion, 1);
1749         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1750                                    rma_alloc_size, rma);
1751         vmstate_register_ram_global(rma_region);
1752         memory_region_add_subregion(sysmem, 0, rma_region);
1753     }
1754 
1755     /* initialize hotplug memory address space */
1756     if (machine->ram_size < machine->maxram_size) {
1757         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
1758 
1759         if (machine->ram_slots > SPAPR_MAX_RAM_SLOTS) {
1760             error_report("Specified number of memory slots %"PRIu64" exceeds max supported %d\n",
1761                          machine->ram_slots, SPAPR_MAX_RAM_SLOTS);
1762             exit(EXIT_FAILURE);
1763         }
1764 
1765         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
1766                                               SPAPR_HOTPLUG_MEM_ALIGN);
1767         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
1768                            "hotplug-memory", hotplug_mem_size);
1769         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
1770                                     &spapr->hotplug_memory.mr);
1771     }
1772 
1773     if (smc->dr_lmb_enabled) {
1774         spapr_create_lmb_dr_connectors(spapr);
1775     }
1776 
1777     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1778     if (!filename) {
1779         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1780         exit(1);
1781     }
1782     spapr->rtas_size = get_image_size(filename);
1783     spapr->rtas_blob = g_malloc(spapr->rtas_size);
1784     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1785         error_report("Could not load LPAR rtas '%s'", filename);
1786         exit(1);
1787     }
1788     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1789         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1790                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1791         exit(1);
1792     }
1793     g_free(filename);
1794 
1795     /* Set up EPOW events infrastructure */
1796     spapr_events_init(spapr);
1797 
1798     /* Set up the RTC RTAS interfaces */
1799     spapr_rtc_create(spapr);
1800 
1801     /* Set up VIO bus */
1802     spapr->vio_bus = spapr_vio_bus_init();
1803 
1804     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1805         if (serial_hds[i]) {
1806             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1807         }
1808     }
1809 
1810     /* We always have at least the nvram device on VIO */
1811     spapr_create_nvram(spapr);
1812 
1813     /* Set up PCI */
1814     spapr_pci_rtas_init();
1815 
1816     phb = spapr_create_phb(spapr, 0);
1817 
1818     for (i = 0; i < nb_nics; i++) {
1819         NICInfo *nd = &nd_table[i];
1820 
1821         if (!nd->model) {
1822             nd->model = g_strdup("ibmveth");
1823         }
1824 
1825         if (strcmp(nd->model, "ibmveth") == 0) {
1826             spapr_vlan_create(spapr->vio_bus, nd);
1827         } else {
1828             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1829         }
1830     }
1831 
1832     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1833         spapr_vscsi_create(spapr->vio_bus);
1834     }
1835 
1836     /* Graphics */
1837     if (spapr_vga_init(phb->bus)) {
1838         spapr->has_graphics = true;
1839         machine->usb |= defaults_enabled() && !machine->usb_disabled;
1840     }
1841 
1842     if (machine->usb) {
1843         pci_create_simple(phb->bus, -1, "pci-ohci");
1844 
1845         if (spapr->has_graphics) {
1846             USBBus *usb_bus = usb_bus_find(-1);
1847 
1848             usb_create_simple(usb_bus, "usb-kbd");
1849             usb_create_simple(usb_bus, "usb-mouse");
1850         }
1851     }
1852 
1853     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1854         fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1855                 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1856         exit(1);
1857     }
1858 
1859     if (kernel_filename) {
1860         uint64_t lowaddr = 0;
1861 
1862         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1863                                NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE, 0);
1864         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1865             kernel_size = load_elf(kernel_filename,
1866                                    translate_kernel_address, NULL,
1867                                    NULL, &lowaddr, NULL, 0, PPC_ELF_MACHINE, 0);
1868             kernel_le = kernel_size > 0;
1869         }
1870         if (kernel_size < 0) {
1871             fprintf(stderr, "qemu: error loading %s: %s\n",
1872                     kernel_filename, load_elf_strerror(kernel_size));
1873             exit(1);
1874         }
1875 
1876         /* load initrd */
1877         if (initrd_filename) {
1878             /* Try to locate the initrd in the gap between the kernel
1879              * and the firmware. Add a bit of space just in case
1880              */
1881             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1882             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1883                                               load_limit - initrd_base);
1884             if (initrd_size < 0) {
1885                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1886                         initrd_filename);
1887                 exit(1);
1888             }
1889         } else {
1890             initrd_base = 0;
1891             initrd_size = 0;
1892         }
1893     }
1894 
1895     if (bios_name == NULL) {
1896         bios_name = FW_FILE_NAME;
1897     }
1898     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1899     if (!filename) {
1900         error_report("Could not find LPAR firmware '%s'", bios_name);
1901         exit(1);
1902     }
1903     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1904     if (fw_size <= 0) {
1905         error_report("Could not load LPAR firmware '%s'", filename);
1906         exit(1);
1907     }
1908     g_free(filename);
1909 
1910     /* FIXME: Should register things through the MachineState's qdev
1911      * interface, this is a legacy from the sPAPREnvironment structure
1912      * which predated MachineState but had a similar function */
1913     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1914     register_savevm_live(NULL, "spapr/htab", -1, 1,
1915                          &savevm_htab_handlers, spapr);
1916 
1917     /* Prepare the device tree */
1918     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1919                                             kernel_size, kernel_le,
1920                                             kernel_cmdline,
1921                                             spapr->check_exception_irq);
1922     assert(spapr->fdt_skel != NULL);
1923 
1924     /* used by RTAS */
1925     QTAILQ_INIT(&spapr->ccs_list);
1926     qemu_register_reset(spapr_ccs_reset_hook, spapr);
1927 
1928     qemu_register_boot_set(spapr_boot_set, spapr);
1929 }
1930 
1931 static int spapr_kvm_type(const char *vm_type)
1932 {
1933     if (!vm_type) {
1934         return 0;
1935     }
1936 
1937     if (!strcmp(vm_type, "HV")) {
1938         return 1;
1939     }
1940 
1941     if (!strcmp(vm_type, "PR")) {
1942         return 2;
1943     }
1944 
1945     error_report("Unknown kvm-type specified '%s'", vm_type);
1946     exit(1);
1947 }
1948 
1949 /*
1950  * Implementation of an interface to adjust firmware path
1951  * for the bootindex property handling.
1952  */
1953 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
1954                                    DeviceState *dev)
1955 {
1956 #define CAST(type, obj, name) \
1957     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
1958     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
1959     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
1960 
1961     if (d) {
1962         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
1963         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
1964         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
1965 
1966         if (spapr) {
1967             /*
1968              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
1969              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
1970              * in the top 16 bits of the 64-bit LUN
1971              */
1972             unsigned id = 0x8000 | (d->id << 8) | d->lun;
1973             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1974                                    (uint64_t)id << 48);
1975         } else if (virtio) {
1976             /*
1977              * We use SRP luns of the form 01000000 | (target << 8) | lun
1978              * in the top 32 bits of the 64-bit LUN
1979              * Note: the quote above is from SLOF and it is wrong,
1980              * the actual binding is:
1981              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
1982              */
1983             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
1984             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1985                                    (uint64_t)id << 32);
1986         } else if (usb) {
1987             /*
1988              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
1989              * in the top 32 bits of the 64-bit LUN
1990              */
1991             unsigned usb_port = atoi(usb->port->path);
1992             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
1993             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1994                                    (uint64_t)id << 32);
1995         }
1996     }
1997 
1998     if (phb) {
1999         /* Replace "pci" with "pci@800000020000000" */
2000         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2001     }
2002 
2003     return NULL;
2004 }
2005 
2006 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2007 {
2008     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2009 
2010     return g_strdup(spapr->kvm_type);
2011 }
2012 
2013 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2014 {
2015     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2016 
2017     g_free(spapr->kvm_type);
2018     spapr->kvm_type = g_strdup(value);
2019 }
2020 
2021 static void spapr_machine_initfn(Object *obj)
2022 {
2023     object_property_add_str(obj, "kvm-type",
2024                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2025     object_property_set_description(obj, "kvm-type",
2026                                     "Specifies the KVM virtualization mode (HV, PR)",
2027                                     NULL);
2028 }
2029 
2030 static void ppc_cpu_do_nmi_on_cpu(void *arg)
2031 {
2032     CPUState *cs = arg;
2033 
2034     cpu_synchronize_state(cs);
2035     ppc_cpu_do_system_reset(cs);
2036 }
2037 
2038 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2039 {
2040     CPUState *cs;
2041 
2042     CPU_FOREACH(cs) {
2043         async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, cs);
2044     }
2045 }
2046 
2047 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr, uint64_t size,
2048                            uint32_t node, Error **errp)
2049 {
2050     sPAPRDRConnector *drc;
2051     sPAPRDRConnectorClass *drck;
2052     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2053     int i, fdt_offset, fdt_size;
2054     void *fdt;
2055 
2056     /*
2057      * Check for DRC connectors and send hotplug notification to the
2058      * guest only in case of hotplugged memory. This allows cold plugged
2059      * memory to be specified at boot time.
2060      */
2061     if (!dev->hotplugged) {
2062         return;
2063     }
2064 
2065     for (i = 0; i < nr_lmbs; i++) {
2066         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2067                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2068         g_assert(drc);
2069 
2070         fdt = create_device_tree(&fdt_size);
2071         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2072                                                 SPAPR_MEMORY_BLOCK_SIZE);
2073 
2074         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2075         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2076         addr += SPAPR_MEMORY_BLOCK_SIZE;
2077     }
2078     spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB, nr_lmbs);
2079 }
2080 
2081 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2082                               uint32_t node, Error **errp)
2083 {
2084     Error *local_err = NULL;
2085     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2086     PCDIMMDevice *dimm = PC_DIMM(dev);
2087     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2088     MemoryRegion *mr = ddc->get_memory_region(dimm);
2089     uint64_t align = memory_region_get_alignment(mr);
2090     uint64_t size = memory_region_size(mr);
2091     uint64_t addr;
2092 
2093     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2094         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2095                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2096         goto out;
2097     }
2098 
2099     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, false, &local_err);
2100     if (local_err) {
2101         goto out;
2102     }
2103 
2104     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2105     if (local_err) {
2106         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2107         goto out;
2108     }
2109 
2110     spapr_add_lmbs(dev, addr, size, node, &error_abort);
2111 
2112 out:
2113     error_propagate(errp, local_err);
2114 }
2115 
2116 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2117                                       DeviceState *dev, Error **errp)
2118 {
2119     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2120 
2121     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2122         int node;
2123 
2124         if (!smc->dr_lmb_enabled) {
2125             error_setg(errp, "Memory hotplug not supported for this machine");
2126             return;
2127         }
2128         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2129         if (*errp) {
2130             return;
2131         }
2132 
2133         /*
2134          * Currently PowerPC kernel doesn't allow hot-adding memory to
2135          * memory-less node, but instead will silently add the memory
2136          * to the first node that has some memory. This causes two
2137          * unexpected behaviours for the user.
2138          *
2139          * - Memory gets hotplugged to a different node than what the user
2140          *   specified.
2141          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2142          *   to memory-less node, a reboot will set things accordingly
2143          *   and the previously hotplugged memory now ends in the right node.
2144          *   This appears as if some memory moved from one node to another.
2145          *
2146          * So until kernel starts supporting memory hotplug to memory-less
2147          * nodes, just prevent such attempts upfront in QEMU.
2148          */
2149         if (nb_numa_nodes && !numa_info[node].node_mem) {
2150             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2151                        node);
2152             return;
2153         }
2154 
2155         spapr_memory_plug(hotplug_dev, dev, node, errp);
2156     }
2157 }
2158 
2159 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2160                                       DeviceState *dev, Error **errp)
2161 {
2162     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2163         error_setg(errp, "Memory hot unplug not supported by sPAPR");
2164     }
2165 }
2166 
2167 static HotplugHandler *spapr_get_hotpug_handler(MachineState *machine,
2168                                              DeviceState *dev)
2169 {
2170     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2171         return HOTPLUG_HANDLER(machine);
2172     }
2173     return NULL;
2174 }
2175 
2176 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2177 {
2178     /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2179      * socket means much for the paravirtualized PAPR platform) */
2180     return cpu_index / smp_threads / smp_cores;
2181 }
2182 
2183 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2184 {
2185     MachineClass *mc = MACHINE_CLASS(oc);
2186     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2187     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2188     NMIClass *nc = NMI_CLASS(oc);
2189     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2190 
2191     mc->init = ppc_spapr_init;
2192     mc->reset = ppc_spapr_reset;
2193     mc->block_default_type = IF_SCSI;
2194     mc->max_cpus = MAX_CPUMASK_BITS;
2195     mc->no_parallel = 1;
2196     mc->default_boot_order = "";
2197     mc->default_ram_size = 512 * M_BYTE;
2198     mc->kvm_type = spapr_kvm_type;
2199     mc->has_dynamic_sysbus = true;
2200     mc->pci_allow_0_address = true;
2201     mc->get_hotplug_handler = spapr_get_hotpug_handler;
2202     hc->plug = spapr_machine_device_plug;
2203     hc->unplug = spapr_machine_device_unplug;
2204     mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
2205 
2206     smc->dr_lmb_enabled = false;
2207     fwc->get_dev_path = spapr_get_fw_dev_path;
2208     nc->nmi_monitor_handler = spapr_nmi;
2209 }
2210 
2211 static const TypeInfo spapr_machine_info = {
2212     .name          = TYPE_SPAPR_MACHINE,
2213     .parent        = TYPE_MACHINE,
2214     .abstract      = true,
2215     .instance_size = sizeof(sPAPRMachineState),
2216     .instance_init = spapr_machine_initfn,
2217     .class_size    = sizeof(sPAPRMachineClass),
2218     .class_init    = spapr_machine_class_init,
2219     .interfaces = (InterfaceInfo[]) {
2220         { TYPE_FW_PATH_PROVIDER },
2221         { TYPE_NMI },
2222         { TYPE_HOTPLUG_HANDLER },
2223         { }
2224     },
2225 };
2226 
2227 #define SPAPR_COMPAT_2_3 \
2228         HW_COMPAT_2_3 \
2229         {\
2230             .driver   = "spapr-pci-host-bridge",\
2231             .property = "dynamic-reconfiguration",\
2232             .value    = "off",\
2233         },
2234 
2235 #define SPAPR_COMPAT_2_2 \
2236         SPAPR_COMPAT_2_3 \
2237         HW_COMPAT_2_2 \
2238         {\
2239             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2240             .property = "mem_win_size",\
2241             .value    = "0x20000000",\
2242         },
2243 
2244 #define SPAPR_COMPAT_2_1 \
2245         SPAPR_COMPAT_2_2 \
2246         HW_COMPAT_2_1
2247 
2248 static void spapr_compat_2_3(Object *obj)
2249 {
2250     savevm_skip_section_footers();
2251     global_state_set_optional();
2252 }
2253 
2254 static void spapr_compat_2_2(Object *obj)
2255 {
2256     spapr_compat_2_3(obj);
2257 }
2258 
2259 static void spapr_compat_2_1(Object *obj)
2260 {
2261     spapr_compat_2_2(obj);
2262 }
2263 
2264 static void spapr_machine_2_3_instance_init(Object *obj)
2265 {
2266     spapr_compat_2_3(obj);
2267     spapr_machine_initfn(obj);
2268 }
2269 
2270 static void spapr_machine_2_2_instance_init(Object *obj)
2271 {
2272     spapr_compat_2_2(obj);
2273     spapr_machine_initfn(obj);
2274 }
2275 
2276 static void spapr_machine_2_1_instance_init(Object *obj)
2277 {
2278     spapr_compat_2_1(obj);
2279     spapr_machine_initfn(obj);
2280 }
2281 
2282 static void spapr_machine_2_1_class_init(ObjectClass *oc, void *data)
2283 {
2284     MachineClass *mc = MACHINE_CLASS(oc);
2285     static GlobalProperty compat_props[] = {
2286         SPAPR_COMPAT_2_1
2287         { /* end of list */ }
2288     };
2289 
2290     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.1";
2291     mc->compat_props = compat_props;
2292 }
2293 
2294 static const TypeInfo spapr_machine_2_1_info = {
2295     .name          = MACHINE_TYPE_NAME("pseries-2.1"),
2296     .parent        = TYPE_SPAPR_MACHINE,
2297     .class_init    = spapr_machine_2_1_class_init,
2298     .instance_init = spapr_machine_2_1_instance_init,
2299 };
2300 
2301 static void spapr_machine_2_2_class_init(ObjectClass *oc, void *data)
2302 {
2303     static GlobalProperty compat_props[] = {
2304         SPAPR_COMPAT_2_2
2305         { /* end of list */ }
2306     };
2307     MachineClass *mc = MACHINE_CLASS(oc);
2308 
2309     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.2";
2310     mc->compat_props = compat_props;
2311 }
2312 
2313 static const TypeInfo spapr_machine_2_2_info = {
2314     .name          = MACHINE_TYPE_NAME("pseries-2.2"),
2315     .parent        = TYPE_SPAPR_MACHINE,
2316     .class_init    = spapr_machine_2_2_class_init,
2317     .instance_init = spapr_machine_2_2_instance_init,
2318 };
2319 
2320 static void spapr_machine_2_3_class_init(ObjectClass *oc, void *data)
2321 {
2322     static GlobalProperty compat_props[] = {
2323         SPAPR_COMPAT_2_3
2324         { /* end of list */ }
2325     };
2326     MachineClass *mc = MACHINE_CLASS(oc);
2327 
2328     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.3";
2329     mc->compat_props = compat_props;
2330 }
2331 
2332 static const TypeInfo spapr_machine_2_3_info = {
2333     .name          = MACHINE_TYPE_NAME("pseries-2.3"),
2334     .parent        = TYPE_SPAPR_MACHINE,
2335     .class_init    = spapr_machine_2_3_class_init,
2336     .instance_init = spapr_machine_2_3_instance_init,
2337 };
2338 
2339 static void spapr_machine_2_4_class_init(ObjectClass *oc, void *data)
2340 {
2341     MachineClass *mc = MACHINE_CLASS(oc);
2342 
2343     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.4";
2344     mc->alias = "pseries";
2345     mc->is_default = 0;
2346 }
2347 
2348 static const TypeInfo spapr_machine_2_4_info = {
2349     .name          = MACHINE_TYPE_NAME("pseries-2.4"),
2350     .parent        = TYPE_SPAPR_MACHINE,
2351     .class_init    = spapr_machine_2_4_class_init,
2352 };
2353 
2354 static void spapr_machine_2_5_class_init(ObjectClass *oc, void *data)
2355 {
2356     MachineClass *mc = MACHINE_CLASS(oc);
2357     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2358 
2359     mc->name = "pseries-2.5";
2360     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.5";
2361     mc->alias = "pseries";
2362     mc->is_default = 1;
2363     smc->dr_lmb_enabled = true;
2364 }
2365 
2366 static const TypeInfo spapr_machine_2_5_info = {
2367     .name          = MACHINE_TYPE_NAME("pseries-2.5"),
2368     .parent        = TYPE_SPAPR_MACHINE,
2369     .class_init    = spapr_machine_2_5_class_init,
2370 };
2371 
2372 static void spapr_machine_register_types(void)
2373 {
2374     type_register_static(&spapr_machine_info);
2375     type_register_static(&spapr_machine_2_1_info);
2376     type_register_static(&spapr_machine_2_2_info);
2377     type_register_static(&spapr_machine_2_3_info);
2378     type_register_static(&spapr_machine_2_4_info);
2379     type_register_static(&spapr_machine_2_5_info);
2380 }
2381 
2382 type_init(spapr_machine_register_types)
2383