xref: /openbmc/qemu/hw/ppc/spapr.c (revision 11ad93f68195f68cc94d988f2aa50b4d190ee52a)
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 "hw/hw.h"
29 #include "elf.h"
30 #include "net/net.h"
31 #include "sysemu/blockdev.h"
32 #include "sysemu/cpus.h"
33 #include "sysemu/kvm.h"
34 #include "kvm_ppc.h"
35 #include "mmu-hash64.h"
36 
37 #include "hw/boards.h"
38 #include "hw/ppc/ppc.h"
39 #include "hw/loader.h"
40 
41 #include "hw/ppc/spapr.h"
42 #include "hw/ppc/spapr_vio.h"
43 #include "hw/pci-host/spapr.h"
44 #include "hw/ppc/xics.h"
45 #include "hw/pci/msi.h"
46 
47 #include "hw/pci/pci.h"
48 
49 #include "exec/address-spaces.h"
50 #include "hw/usb.h"
51 #include "qemu/config-file.h"
52 
53 #include <libfdt.h>
54 
55 /* SLOF memory layout:
56  *
57  * SLOF raw image loaded at 0, copies its romfs right below the flat
58  * device-tree, then position SLOF itself 31M below that
59  *
60  * So we set FW_OVERHEAD to 40MB which should account for all of that
61  * and more
62  *
63  * We load our kernel at 4M, leaving space for SLOF initial image
64  */
65 #define FDT_MAX_SIZE            0x40000
66 #define RTAS_MAX_SIZE           0x10000
67 #define FW_MAX_SIZE             0x400000
68 #define FW_FILE_NAME            "slof.bin"
69 #define FW_OVERHEAD             0x2800000
70 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
71 
72 #define MIN_RMA_SLOF            128UL
73 
74 #define TIMEBASE_FREQ           512000000ULL
75 
76 #define MAX_CPUS                256
77 #define XICS_IRQS               1024
78 
79 #define PHANDLE_XICP            0x00001111
80 
81 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
82 
83 sPAPREnvironment *spapr;
84 
85 int spapr_allocate_irq(int hint, bool lsi)
86 {
87     int irq;
88 
89     if (hint) {
90         irq = hint;
91         if (hint >= spapr->next_irq) {
92             spapr->next_irq = hint + 1;
93         }
94         /* FIXME: we should probably check for collisions somehow */
95     } else {
96         irq = spapr->next_irq++;
97     }
98 
99     /* Configure irq type */
100     if (!xics_get_qirq(spapr->icp, irq)) {
101         return 0;
102     }
103 
104     xics_set_irq_type(spapr->icp, irq, lsi);
105 
106     return irq;
107 }
108 
109 /*
110  * Allocate block of consequtive IRQs, returns a number of the first.
111  * If msi==true, aligns the first IRQ number to num.
112  */
113 int spapr_allocate_irq_block(int num, bool lsi, bool msi)
114 {
115     int first = -1;
116     int i, hint = 0;
117 
118     /*
119      * MSIMesage::data is used for storing VIRQ so
120      * it has to be aligned to num to support multiple
121      * MSI vectors. MSI-X is not affected by this.
122      * The hint is used for the first IRQ, the rest should
123      * be allocated continously.
124      */
125     if (msi) {
126         assert((num == 1) || (num == 2) || (num == 4) ||
127                (num == 8) || (num == 16) || (num == 32));
128         hint = (spapr->next_irq + num - 1) & ~(num - 1);
129     }
130 
131     for (i = 0; i < num; ++i) {
132         int irq;
133 
134         irq = spapr_allocate_irq(hint, lsi);
135         if (!irq) {
136             return -1;
137         }
138 
139         if (0 == i) {
140             first = irq;
141             hint = 0;
142         }
143 
144         /* If the above doesn't create a consecutive block then that's
145          * an internal bug */
146         assert(irq == (first + i));
147     }
148 
149     return first;
150 }
151 
152 static XICSState *try_create_xics(const char *type, int nr_servers,
153                                   int nr_irqs)
154 {
155     DeviceState *dev;
156 
157     dev = qdev_create(NULL, type);
158     qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
159     qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
160     if (qdev_init(dev) < 0) {
161         return NULL;
162     }
163 
164     return XICS_COMMON(dev);
165 }
166 
167 static XICSState *xics_system_init(int nr_servers, int nr_irqs)
168 {
169     XICSState *icp = NULL;
170 
171     if (kvm_enabled()) {
172         QemuOpts *machine_opts = qemu_get_machine_opts();
173         bool irqchip_allowed = qemu_opt_get_bool(machine_opts,
174                                                 "kernel_irqchip", true);
175         bool irqchip_required = qemu_opt_get_bool(machine_opts,
176                                                   "kernel_irqchip", false);
177         if (irqchip_allowed) {
178             icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs);
179         }
180 
181         if (irqchip_required && !icp) {
182             perror("Failed to create in-kernel XICS\n");
183             abort();
184         }
185     }
186 
187     if (!icp) {
188         icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs);
189     }
190 
191     if (!icp) {
192         perror("Failed to create XICS\n");
193         abort();
194     }
195 
196     return icp;
197 }
198 
199 static int spapr_fixup_cpu_dt(void *fdt, sPAPREnvironment *spapr)
200 {
201     int ret = 0, offset;
202     CPUState *cpu;
203     char cpu_model[32];
204     int smt = kvmppc_smt_threads();
205     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
206 
207     assert(spapr->cpu_model);
208 
209     CPU_FOREACH(cpu) {
210         uint32_t associativity[] = {cpu_to_be32(0x5),
211                                     cpu_to_be32(0x0),
212                                     cpu_to_be32(0x0),
213                                     cpu_to_be32(0x0),
214                                     cpu_to_be32(cpu->numa_node),
215                                     cpu_to_be32(cpu->cpu_index)};
216 
217         if ((cpu->cpu_index % smt) != 0) {
218             continue;
219         }
220 
221         snprintf(cpu_model, 32, "/cpus/%s@%x", spapr->cpu_model,
222                  cpu->cpu_index);
223 
224         offset = fdt_path_offset(fdt, cpu_model);
225         if (offset < 0) {
226             return offset;
227         }
228 
229         if (nb_numa_nodes > 1) {
230             ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
231                               sizeof(associativity));
232             if (ret < 0) {
233                 return ret;
234             }
235         }
236 
237         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
238                           pft_size_prop, sizeof(pft_size_prop));
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 #define _FDT(exp) \
281     do { \
282         int ret = (exp);                                           \
283         if (ret < 0) {                                             \
284             fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
285                     #exp, fdt_strerror(ret));                      \
286             exit(1);                                               \
287         }                                                          \
288     } while (0)
289 
290 
291 static void *spapr_create_fdt_skel(const char *cpu_model,
292                                    hwaddr initrd_base,
293                                    hwaddr initrd_size,
294                                    hwaddr kernel_size,
295                                    bool little_endian,
296                                    const char *boot_device,
297                                    const char *kernel_cmdline,
298                                    uint32_t epow_irq)
299 {
300     void *fdt;
301     CPUState *cs;
302     uint32_t start_prop = cpu_to_be32(initrd_base);
303     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
304     char hypertas_prop[] = "hcall-pft\0hcall-term\0hcall-dabr\0hcall-interrupt"
305         "\0hcall-tce\0hcall-vio\0hcall-splpar\0hcall-bulk\0hcall-set-mode";
306     char qemu_hypertas_prop[] = "hcall-memop1";
307     uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
308     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(smp_cpus)};
309     char *modelname;
310     int i, smt = kvmppc_smt_threads();
311     unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
312 
313     fdt = g_malloc0(FDT_MAX_SIZE);
314     _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
315 
316     if (kernel_size) {
317         _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
318     }
319     if (initrd_size) {
320         _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
321     }
322     _FDT((fdt_finish_reservemap(fdt)));
323 
324     /* Root node */
325     _FDT((fdt_begin_node(fdt, "")));
326     _FDT((fdt_property_string(fdt, "device_type", "chrp")));
327     _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
328     _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
329 
330     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
331     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
332 
333     /* /chosen */
334     _FDT((fdt_begin_node(fdt, "chosen")));
335 
336     /* Set Form1_affinity */
337     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
338 
339     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
340     _FDT((fdt_property(fdt, "linux,initrd-start",
341                        &start_prop, sizeof(start_prop))));
342     _FDT((fdt_property(fdt, "linux,initrd-end",
343                        &end_prop, sizeof(end_prop))));
344     if (kernel_size) {
345         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
346                               cpu_to_be64(kernel_size) };
347 
348         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
349         if (little_endian) {
350             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
351         }
352     }
353     if (boot_device) {
354         _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
355     }
356     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
357     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
358     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
359 
360     _FDT((fdt_end_node(fdt)));
361 
362     /* cpus */
363     _FDT((fdt_begin_node(fdt, "cpus")));
364 
365     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
366     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
367 
368     modelname = g_strdup(cpu_model);
369 
370     for (i = 0; i < strlen(modelname); i++) {
371         modelname[i] = toupper(modelname[i]);
372     }
373 
374     /* This is needed during FDT finalization */
375     spapr->cpu_model = g_strdup(modelname);
376 
377     CPU_FOREACH(cs) {
378         PowerPCCPU *cpu = POWERPC_CPU(cs);
379         CPUPPCState *env = &cpu->env;
380         PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
381         int index = cs->cpu_index;
382         uint32_t servers_prop[smp_threads];
383         uint32_t gservers_prop[smp_threads * 2];
384         char *nodename;
385         uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
386                            0xffffffff, 0xffffffff};
387         uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
388         uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
389         uint32_t page_sizes_prop[64];
390         size_t page_sizes_prop_size;
391 
392         if ((index % smt) != 0) {
393             continue;
394         }
395 
396         nodename = g_strdup_printf("%s@%x", modelname, index);
397 
398         _FDT((fdt_begin_node(fdt, nodename)));
399 
400         g_free(nodename);
401 
402         _FDT((fdt_property_cell(fdt, "reg", index)));
403         _FDT((fdt_property_string(fdt, "device_type", "cpu")));
404 
405         _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
406         _FDT((fdt_property_cell(fdt, "d-cache-block-size",
407                                 env->dcache_line_size)));
408         _FDT((fdt_property_cell(fdt, "d-cache-line-size",
409                                 env->dcache_line_size)));
410         _FDT((fdt_property_cell(fdt, "i-cache-block-size",
411                                 env->icache_line_size)));
412         _FDT((fdt_property_cell(fdt, "i-cache-line-size",
413                                 env->icache_line_size)));
414 
415         if (pcc->l1_dcache_size) {
416             _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
417         } else {
418             fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
419         }
420         if (pcc->l1_icache_size) {
421             _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
422         } else {
423             fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
424         }
425 
426         _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
427         _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
428         _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
429         _FDT((fdt_property_string(fdt, "status", "okay")));
430         _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
431 
432         /* Build interrupt servers and gservers properties */
433         for (i = 0; i < smp_threads; i++) {
434             servers_prop[i] = cpu_to_be32(index + i);
435             /* Hack, direct the group queues back to cpu 0 */
436             gservers_prop[i*2] = cpu_to_be32(index + i);
437             gservers_prop[i*2 + 1] = 0;
438         }
439         _FDT((fdt_property(fdt, "ibm,ppc-interrupt-server#s",
440                            servers_prop, sizeof(servers_prop))));
441         _FDT((fdt_property(fdt, "ibm,ppc-interrupt-gserver#s",
442                            gservers_prop, sizeof(gservers_prop))));
443 
444         if (env->spr_cb[SPR_PURR].oea_read) {
445             _FDT((fdt_property(fdt, "ibm,purr", NULL, 0)));
446         }
447 
448         if (env->mmu_model & POWERPC_MMU_1TSEG) {
449             _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
450                                segs, sizeof(segs))));
451         }
452 
453         /* Advertise VMX/VSX (vector extensions) if available
454          *   0 / no property == no vector extensions
455          *   1               == VMX / Altivec available
456          *   2               == VSX available */
457         if (env->insns_flags & PPC_ALTIVEC) {
458             uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
459 
460             _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
461         }
462 
463         /* Advertise DFP (Decimal Floating Point) if available
464          *   0 / no property == no DFP
465          *   1               == DFP available */
466         if (env->insns_flags2 & PPC2_DFP) {
467             _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
468         }
469 
470         page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
471                                                       sizeof(page_sizes_prop));
472         if (page_sizes_prop_size) {
473             _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
474                                page_sizes_prop, page_sizes_prop_size)));
475         }
476 
477         _FDT((fdt_end_node(fdt)));
478     }
479 
480     g_free(modelname);
481 
482     _FDT((fdt_end_node(fdt)));
483 
484     /* RTAS */
485     _FDT((fdt_begin_node(fdt, "rtas")));
486 
487     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas_prop,
488                        sizeof(hypertas_prop))));
489     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas_prop,
490                        sizeof(qemu_hypertas_prop))));
491 
492     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
493         refpoints, sizeof(refpoints))));
494 
495     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
496 
497     _FDT((fdt_end_node(fdt)));
498 
499     /* interrupt controller */
500     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
501 
502     _FDT((fdt_property_string(fdt, "device_type",
503                               "PowerPC-External-Interrupt-Presentation")));
504     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
505     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
506     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
507                        interrupt_server_ranges_prop,
508                        sizeof(interrupt_server_ranges_prop))));
509     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
510     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
511     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
512 
513     _FDT((fdt_end_node(fdt)));
514 
515     /* vdevice */
516     _FDT((fdt_begin_node(fdt, "vdevice")));
517 
518     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
519     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
520     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
521     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
522     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
523     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
524 
525     _FDT((fdt_end_node(fdt)));
526 
527     /* event-sources */
528     spapr_events_fdt_skel(fdt, epow_irq);
529 
530     _FDT((fdt_end_node(fdt))); /* close root node */
531     _FDT((fdt_finish(fdt)));
532 
533     return fdt;
534 }
535 
536 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
537 {
538     uint32_t associativity[] = {cpu_to_be32(0x4), cpu_to_be32(0x0),
539                                 cpu_to_be32(0x0), cpu_to_be32(0x0),
540                                 cpu_to_be32(0x0)};
541     char mem_name[32];
542     hwaddr node0_size, mem_start;
543     uint64_t mem_reg_property[2];
544     int i, off;
545 
546     /* memory node(s) */
547     node0_size = (nb_numa_nodes > 1) ? node_mem[0] : ram_size;
548     if (spapr->rma_size > node0_size) {
549         spapr->rma_size = node0_size;
550     }
551 
552     /* RMA */
553     mem_reg_property[0] = 0;
554     mem_reg_property[1] = cpu_to_be64(spapr->rma_size);
555     off = fdt_add_subnode(fdt, 0, "memory@0");
556     _FDT(off);
557     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
558     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
559                       sizeof(mem_reg_property))));
560     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
561                       sizeof(associativity))));
562 
563     /* RAM: Node 0 */
564     if (node0_size > spapr->rma_size) {
565         mem_reg_property[0] = cpu_to_be64(spapr->rma_size);
566         mem_reg_property[1] = cpu_to_be64(node0_size - spapr->rma_size);
567 
568         sprintf(mem_name, "memory@" TARGET_FMT_lx, spapr->rma_size);
569         off = fdt_add_subnode(fdt, 0, mem_name);
570         _FDT(off);
571         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
572         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
573                           sizeof(mem_reg_property))));
574         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
575                           sizeof(associativity))));
576     }
577 
578     /* RAM: Node 1 and beyond */
579     mem_start = node0_size;
580     for (i = 1; i < nb_numa_nodes; i++) {
581         mem_reg_property[0] = cpu_to_be64(mem_start);
582         mem_reg_property[1] = cpu_to_be64(node_mem[i]);
583         associativity[3] = associativity[4] = cpu_to_be32(i);
584         sprintf(mem_name, "memory@" TARGET_FMT_lx, mem_start);
585         off = fdt_add_subnode(fdt, 0, mem_name);
586         _FDT(off);
587         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
588         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
589                           sizeof(mem_reg_property))));
590         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
591                           sizeof(associativity))));
592         mem_start += node_mem[i];
593     }
594 
595     return 0;
596 }
597 
598 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
599                                hwaddr fdt_addr,
600                                hwaddr rtas_addr,
601                                hwaddr rtas_size)
602 {
603     int ret;
604     void *fdt;
605     sPAPRPHBState *phb;
606 
607     fdt = g_malloc(FDT_MAX_SIZE);
608 
609     /* open out the base tree into a temp buffer for the final tweaks */
610     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
611 
612     ret = spapr_populate_memory(spapr, fdt);
613     if (ret < 0) {
614         fprintf(stderr, "couldn't setup memory nodes in fdt\n");
615         exit(1);
616     }
617 
618     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
619     if (ret < 0) {
620         fprintf(stderr, "couldn't setup vio devices in fdt\n");
621         exit(1);
622     }
623 
624     QLIST_FOREACH(phb, &spapr->phbs, list) {
625         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
626     }
627 
628     if (ret < 0) {
629         fprintf(stderr, "couldn't setup PCI devices in fdt\n");
630         exit(1);
631     }
632 
633     /* RTAS */
634     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
635     if (ret < 0) {
636         fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
637     }
638 
639     /* Advertise NUMA via ibm,associativity */
640     ret = spapr_fixup_cpu_dt(fdt, spapr);
641     if (ret < 0) {
642         fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
643     }
644 
645     if (!spapr->has_graphics) {
646         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
647     }
648 
649     _FDT((fdt_pack(fdt)));
650 
651     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
652         hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
653                  fdt_totalsize(fdt), FDT_MAX_SIZE);
654         exit(1);
655     }
656 
657     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
658 
659     g_free(fdt);
660 }
661 
662 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
663 {
664     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
665 }
666 
667 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
668 {
669     CPUPPCState *env = &cpu->env;
670 
671     if (msr_pr) {
672         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
673         env->gpr[3] = H_PRIVILEGE;
674     } else {
675         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
676     }
677 }
678 
679 static void spapr_reset_htab(sPAPREnvironment *spapr)
680 {
681     long shift;
682 
683     /* allocate hash page table.  For now we always make this 16mb,
684      * later we should probably make it scale to the size of guest
685      * RAM */
686 
687     shift = kvmppc_reset_htab(spapr->htab_shift);
688 
689     if (shift > 0) {
690         /* Kernel handles htab, we don't need to allocate one */
691         spapr->htab_shift = shift;
692     } else {
693         if (!spapr->htab) {
694             /* Allocate an htab if we don't yet have one */
695             spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
696         }
697 
698         /* And clear it */
699         memset(spapr->htab, 0, HTAB_SIZE(spapr));
700     }
701 
702     /* Update the RMA size if necessary */
703     if (spapr->vrma_adjust) {
704         spapr->rma_size = kvmppc_rma_size(ram_size, spapr->htab_shift);
705     }
706 }
707 
708 static void ppc_spapr_reset(void)
709 {
710     PowerPCCPU *first_ppc_cpu;
711 
712     /* Reset the hash table & recalc the RMA */
713     spapr_reset_htab(spapr);
714 
715     qemu_devices_reset();
716 
717     /* Load the fdt */
718     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
719                        spapr->rtas_size);
720 
721     /* Set up the entry state */
722     first_ppc_cpu = POWERPC_CPU(first_cpu);
723     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
724     first_ppc_cpu->env.gpr[5] = 0;
725     first_cpu->halted = 0;
726     first_ppc_cpu->env.nip = spapr->entry_point;
727 
728 }
729 
730 static void spapr_cpu_reset(void *opaque)
731 {
732     PowerPCCPU *cpu = opaque;
733     CPUState *cs = CPU(cpu);
734     CPUPPCState *env = &cpu->env;
735 
736     cpu_reset(cs);
737 
738     /* All CPUs start halted.  CPU0 is unhalted from the machine level
739      * reset code and the rest are explicitly started up by the guest
740      * using an RTAS call */
741     cs->halted = 1;
742 
743     env->spr[SPR_HIOR] = 0;
744 
745     env->external_htab = (uint8_t *)spapr->htab;
746     env->htab_base = -1;
747     env->htab_mask = HTAB_SIZE(spapr) - 1;
748     env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
749         (spapr->htab_shift - 18);
750 }
751 
752 static void spapr_create_nvram(sPAPREnvironment *spapr)
753 {
754     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
755     const char *drivename = qemu_opt_get(qemu_get_machine_opts(), "nvram");
756 
757     if (drivename) {
758         BlockDriverState *bs;
759 
760         bs = bdrv_find(drivename);
761         if (!bs) {
762             fprintf(stderr, "No such block device \"%s\" for nvram\n",
763                     drivename);
764             exit(1);
765         }
766         qdev_prop_set_drive_nofail(dev, "drive", bs);
767     }
768 
769     qdev_init_nofail(dev);
770 
771     spapr->nvram = (struct sPAPRNVRAM *)dev;
772 }
773 
774 /* Returns whether we want to use VGA or not */
775 static int spapr_vga_init(PCIBus *pci_bus)
776 {
777     switch (vga_interface_type) {
778     case VGA_NONE:
779     case VGA_STD:
780         return pci_vga_init(pci_bus) != NULL;
781     default:
782         fprintf(stderr, "This vga model is not supported,"
783                 "currently it only supports -vga std\n");
784         exit(0);
785         break;
786     }
787 }
788 
789 static const VMStateDescription vmstate_spapr = {
790     .name = "spapr",
791     .version_id = 1,
792     .minimum_version_id = 1,
793     .minimum_version_id_old = 1,
794     .fields      = (VMStateField []) {
795         VMSTATE_UINT32(next_irq, sPAPREnvironment),
796 
797         /* RTC offset */
798         VMSTATE_UINT64(rtc_offset, sPAPREnvironment),
799 
800         VMSTATE_END_OF_LIST()
801     },
802 };
803 
804 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
805 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
806 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
807 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
808 
809 static int htab_save_setup(QEMUFile *f, void *opaque)
810 {
811     sPAPREnvironment *spapr = opaque;
812 
813     /* "Iteration" header */
814     qemu_put_be32(f, spapr->htab_shift);
815 
816     if (spapr->htab) {
817         spapr->htab_save_index = 0;
818         spapr->htab_first_pass = true;
819     } else {
820         assert(kvm_enabled());
821 
822         spapr->htab_fd = kvmppc_get_htab_fd(false);
823         if (spapr->htab_fd < 0) {
824             fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
825                     strerror(errno));
826             return -1;
827         }
828     }
829 
830 
831     return 0;
832 }
833 
834 static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
835                                  int64_t max_ns)
836 {
837     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
838     int index = spapr->htab_save_index;
839     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
840 
841     assert(spapr->htab_first_pass);
842 
843     do {
844         int chunkstart;
845 
846         /* Consume invalid HPTEs */
847         while ((index < htabslots)
848                && !HPTE_VALID(HPTE(spapr->htab, index))) {
849             index++;
850             CLEAN_HPTE(HPTE(spapr->htab, index));
851         }
852 
853         /* Consume valid HPTEs */
854         chunkstart = index;
855         while ((index < htabslots)
856                && HPTE_VALID(HPTE(spapr->htab, index))) {
857             index++;
858             CLEAN_HPTE(HPTE(spapr->htab, index));
859         }
860 
861         if (index > chunkstart) {
862             int n_valid = index - chunkstart;
863 
864             qemu_put_be32(f, chunkstart);
865             qemu_put_be16(f, n_valid);
866             qemu_put_be16(f, 0);
867             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
868                             HASH_PTE_SIZE_64 * n_valid);
869 
870             if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
871                 break;
872             }
873         }
874     } while ((index < htabslots) && !qemu_file_rate_limit(f));
875 
876     if (index >= htabslots) {
877         assert(index == htabslots);
878         index = 0;
879         spapr->htab_first_pass = false;
880     }
881     spapr->htab_save_index = index;
882 }
883 
884 static int htab_save_later_pass(QEMUFile *f, sPAPREnvironment *spapr,
885                                 int64_t max_ns)
886 {
887     bool final = max_ns < 0;
888     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
889     int examined = 0, sent = 0;
890     int index = spapr->htab_save_index;
891     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
892 
893     assert(!spapr->htab_first_pass);
894 
895     do {
896         int chunkstart, invalidstart;
897 
898         /* Consume non-dirty HPTEs */
899         while ((index < htabslots)
900                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
901             index++;
902             examined++;
903         }
904 
905         chunkstart = index;
906         /* Consume valid dirty HPTEs */
907         while ((index < htabslots)
908                && HPTE_DIRTY(HPTE(spapr->htab, index))
909                && HPTE_VALID(HPTE(spapr->htab, index))) {
910             CLEAN_HPTE(HPTE(spapr->htab, index));
911             index++;
912             examined++;
913         }
914 
915         invalidstart = index;
916         /* Consume invalid dirty HPTEs */
917         while ((index < htabslots)
918                && HPTE_DIRTY(HPTE(spapr->htab, index))
919                && !HPTE_VALID(HPTE(spapr->htab, index))) {
920             CLEAN_HPTE(HPTE(spapr->htab, index));
921             index++;
922             examined++;
923         }
924 
925         if (index > chunkstart) {
926             int n_valid = invalidstart - chunkstart;
927             int n_invalid = index - invalidstart;
928 
929             qemu_put_be32(f, chunkstart);
930             qemu_put_be16(f, n_valid);
931             qemu_put_be16(f, n_invalid);
932             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
933                             HASH_PTE_SIZE_64 * n_valid);
934             sent += index - chunkstart;
935 
936             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
937                 break;
938             }
939         }
940 
941         if (examined >= htabslots) {
942             break;
943         }
944 
945         if (index >= htabslots) {
946             assert(index == htabslots);
947             index = 0;
948         }
949     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
950 
951     if (index >= htabslots) {
952         assert(index == htabslots);
953         index = 0;
954     }
955 
956     spapr->htab_save_index = index;
957 
958     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
959 }
960 
961 #define MAX_ITERATION_NS    5000000 /* 5 ms */
962 #define MAX_KVM_BUF_SIZE    2048
963 
964 static int htab_save_iterate(QEMUFile *f, void *opaque)
965 {
966     sPAPREnvironment *spapr = opaque;
967     int rc = 0;
968 
969     /* Iteration header */
970     qemu_put_be32(f, 0);
971 
972     if (!spapr->htab) {
973         assert(kvm_enabled());
974 
975         rc = kvmppc_save_htab(f, spapr->htab_fd,
976                               MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
977         if (rc < 0) {
978             return rc;
979         }
980     } else  if (spapr->htab_first_pass) {
981         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
982     } else {
983         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
984     }
985 
986     /* End marker */
987     qemu_put_be32(f, 0);
988     qemu_put_be16(f, 0);
989     qemu_put_be16(f, 0);
990 
991     return rc;
992 }
993 
994 static int htab_save_complete(QEMUFile *f, void *opaque)
995 {
996     sPAPREnvironment *spapr = opaque;
997 
998     /* Iteration header */
999     qemu_put_be32(f, 0);
1000 
1001     if (!spapr->htab) {
1002         int rc;
1003 
1004         assert(kvm_enabled());
1005 
1006         rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
1007         if (rc < 0) {
1008             return rc;
1009         }
1010         close(spapr->htab_fd);
1011         spapr->htab_fd = -1;
1012     } else {
1013         htab_save_later_pass(f, spapr, -1);
1014     }
1015 
1016     /* End marker */
1017     qemu_put_be32(f, 0);
1018     qemu_put_be16(f, 0);
1019     qemu_put_be16(f, 0);
1020 
1021     return 0;
1022 }
1023 
1024 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1025 {
1026     sPAPREnvironment *spapr = opaque;
1027     uint32_t section_hdr;
1028     int fd = -1;
1029 
1030     if (version_id < 1 || version_id > 1) {
1031         fprintf(stderr, "htab_load() bad version\n");
1032         return -EINVAL;
1033     }
1034 
1035     section_hdr = qemu_get_be32(f);
1036 
1037     if (section_hdr) {
1038         /* First section, just the hash shift */
1039         if (spapr->htab_shift != section_hdr) {
1040             return -EINVAL;
1041         }
1042         return 0;
1043     }
1044 
1045     if (!spapr->htab) {
1046         assert(kvm_enabled());
1047 
1048         fd = kvmppc_get_htab_fd(true);
1049         if (fd < 0) {
1050             fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1051                     strerror(errno));
1052         }
1053     }
1054 
1055     while (true) {
1056         uint32_t index;
1057         uint16_t n_valid, n_invalid;
1058 
1059         index = qemu_get_be32(f);
1060         n_valid = qemu_get_be16(f);
1061         n_invalid = qemu_get_be16(f);
1062 
1063         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1064             /* End of Stream */
1065             break;
1066         }
1067 
1068         if ((index + n_valid + n_invalid) >
1069             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1070             /* Bad index in stream */
1071             fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1072                     "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1073                     spapr->htab_shift);
1074             return -EINVAL;
1075         }
1076 
1077         if (spapr->htab) {
1078             if (n_valid) {
1079                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1080                                 HASH_PTE_SIZE_64 * n_valid);
1081             }
1082             if (n_invalid) {
1083                 memset(HPTE(spapr->htab, index + n_valid), 0,
1084                        HASH_PTE_SIZE_64 * n_invalid);
1085             }
1086         } else {
1087             int rc;
1088 
1089             assert(fd >= 0);
1090 
1091             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1092             if (rc < 0) {
1093                 return rc;
1094             }
1095         }
1096     }
1097 
1098     if (!spapr->htab) {
1099         assert(fd >= 0);
1100         close(fd);
1101     }
1102 
1103     return 0;
1104 }
1105 
1106 static SaveVMHandlers savevm_htab_handlers = {
1107     .save_live_setup = htab_save_setup,
1108     .save_live_iterate = htab_save_iterate,
1109     .save_live_complete = htab_save_complete,
1110     .load_state = htab_load,
1111 };
1112 
1113 /* pSeries LPAR / sPAPR hardware init */
1114 static void ppc_spapr_init(QEMUMachineInitArgs *args)
1115 {
1116     ram_addr_t ram_size = args->ram_size;
1117     const char *cpu_model = args->cpu_model;
1118     const char *kernel_filename = args->kernel_filename;
1119     const char *kernel_cmdline = args->kernel_cmdline;
1120     const char *initrd_filename = args->initrd_filename;
1121     const char *boot_device = args->boot_order;
1122     PowerPCCPU *cpu;
1123     CPUPPCState *env;
1124     PCIHostState *phb;
1125     int i;
1126     MemoryRegion *sysmem = get_system_memory();
1127     MemoryRegion *ram = g_new(MemoryRegion, 1);
1128     hwaddr rma_alloc_size;
1129     uint32_t initrd_base = 0;
1130     long kernel_size = 0, initrd_size = 0;
1131     long load_limit, rtas_limit, fw_size;
1132     bool kernel_le = false;
1133     char *filename;
1134 
1135     msi_supported = true;
1136 
1137     spapr = g_malloc0(sizeof(*spapr));
1138     QLIST_INIT(&spapr->phbs);
1139 
1140     cpu_ppc_hypercall = emulate_spapr_hypercall;
1141 
1142     /* Allocate RMA if necessary */
1143     rma_alloc_size = kvmppc_alloc_rma("ppc_spapr.rma", sysmem);
1144 
1145     if (rma_alloc_size == -1) {
1146         hw_error("qemu: Unable to create RMA\n");
1147         exit(1);
1148     }
1149 
1150     if (rma_alloc_size && (rma_alloc_size < ram_size)) {
1151         spapr->rma_size = rma_alloc_size;
1152     } else {
1153         spapr->rma_size = ram_size;
1154 
1155         /* With KVM, we don't actually know whether KVM supports an
1156          * unbounded RMA (PR KVM) or is limited by the hash table size
1157          * (HV KVM using VRMA), so we always assume the latter
1158          *
1159          * In that case, we also limit the initial allocations for RTAS
1160          * etc... to 256M since we have no way to know what the VRMA size
1161          * is going to be as it depends on the size of the hash table
1162          * isn't determined yet.
1163          */
1164         if (kvm_enabled()) {
1165             spapr->vrma_adjust = 1;
1166             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1167         }
1168     }
1169 
1170     /* We place the device tree and RTAS just below either the top of the RMA,
1171      * or just below 2GB, whichever is lowere, so that it can be
1172      * processed with 32-bit real mode code if necessary */
1173     rtas_limit = MIN(spapr->rma_size, 0x80000000);
1174     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1175     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1176     load_limit = spapr->fdt_addr - FW_OVERHEAD;
1177 
1178     /* We aim for a hash table of size 1/128 the size of RAM.  The
1179      * normal rule of thumb is 1/64 the size of RAM, but that's much
1180      * more than needed for the Linux guests we support. */
1181     spapr->htab_shift = 18; /* Minimum architected size */
1182     while (spapr->htab_shift <= 46) {
1183         if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
1184             break;
1185         }
1186         spapr->htab_shift++;
1187     }
1188 
1189     /* Set up Interrupt Controller before we create the VCPUs */
1190     spapr->icp = xics_system_init(smp_cpus * kvmppc_smt_threads() / smp_threads,
1191                                   XICS_IRQS);
1192     spapr->next_irq = XICS_IRQ_BASE;
1193 
1194     /* init CPUs */
1195     if (cpu_model == NULL) {
1196         cpu_model = kvm_enabled() ? "host" : "POWER7";
1197     }
1198     for (i = 0; i < smp_cpus; i++) {
1199         cpu = cpu_ppc_init(cpu_model);
1200         if (cpu == NULL) {
1201             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1202             exit(1);
1203         }
1204         env = &cpu->env;
1205 
1206         /* Set time-base frequency to 512 MHz */
1207         cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1208 
1209         /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1210          * MSR[IP] should never be set.
1211          */
1212         env->msr_mask &= ~(1 << 6);
1213 
1214         /* Tell KVM that we're in PAPR mode */
1215         if (kvm_enabled()) {
1216             kvmppc_set_papr(cpu);
1217         }
1218 
1219         xics_cpu_setup(spapr->icp, cpu);
1220 
1221         qemu_register_reset(spapr_cpu_reset, cpu);
1222     }
1223 
1224     /* allocate RAM */
1225     spapr->ram_limit = ram_size;
1226     if (spapr->ram_limit > rma_alloc_size) {
1227         ram_addr_t nonrma_base = rma_alloc_size;
1228         ram_addr_t nonrma_size = spapr->ram_limit - rma_alloc_size;
1229 
1230         memory_region_init_ram(ram, NULL, "ppc_spapr.ram", nonrma_size);
1231         vmstate_register_ram_global(ram);
1232         memory_region_add_subregion(sysmem, nonrma_base, ram);
1233     }
1234 
1235     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1236     spapr->rtas_size = load_image_targphys(filename, spapr->rtas_addr,
1237                                            rtas_limit - spapr->rtas_addr);
1238     if (spapr->rtas_size < 0) {
1239         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1240         exit(1);
1241     }
1242     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1243         hw_error("RTAS too big ! 0x%lx bytes (max is 0x%x)\n",
1244                  spapr->rtas_size, RTAS_MAX_SIZE);
1245         exit(1);
1246     }
1247     g_free(filename);
1248 
1249     /* Set up EPOW events infrastructure */
1250     spapr_events_init(spapr);
1251 
1252     /* Set up VIO bus */
1253     spapr->vio_bus = spapr_vio_bus_init();
1254 
1255     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1256         if (serial_hds[i]) {
1257             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1258         }
1259     }
1260 
1261     /* We always have at least the nvram device on VIO */
1262     spapr_create_nvram(spapr);
1263 
1264     /* Set up PCI */
1265     spapr_pci_msi_init(spapr, SPAPR_PCI_MSI_WINDOW);
1266     spapr_pci_rtas_init();
1267 
1268     phb = spapr_create_phb(spapr, 0);
1269 
1270     for (i = 0; i < nb_nics; i++) {
1271         NICInfo *nd = &nd_table[i];
1272 
1273         if (!nd->model) {
1274             nd->model = g_strdup("ibmveth");
1275         }
1276 
1277         if (strcmp(nd->model, "ibmveth") == 0) {
1278             spapr_vlan_create(spapr->vio_bus, nd);
1279         } else {
1280             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1281         }
1282     }
1283 
1284     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1285         spapr_vscsi_create(spapr->vio_bus);
1286     }
1287 
1288     /* Graphics */
1289     if (spapr_vga_init(phb->bus)) {
1290         spapr->has_graphics = true;
1291     }
1292 
1293     if (usb_enabled(spapr->has_graphics)) {
1294         pci_create_simple(phb->bus, -1, "pci-ohci");
1295         if (spapr->has_graphics) {
1296             usbdevice_create("keyboard");
1297             usbdevice_create("mouse");
1298         }
1299     }
1300 
1301     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1302         fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1303                 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1304         exit(1);
1305     }
1306 
1307     if (kernel_filename) {
1308         uint64_t lowaddr = 0;
1309 
1310         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1311                                NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1312         if (kernel_size < 0) {
1313             kernel_size = load_elf(kernel_filename,
1314                                    translate_kernel_address, NULL,
1315                                    NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1316             kernel_le = kernel_size > 0;
1317         }
1318         if (kernel_size < 0) {
1319             kernel_size = load_image_targphys(kernel_filename,
1320                                               KERNEL_LOAD_ADDR,
1321                                               load_limit - KERNEL_LOAD_ADDR);
1322         }
1323         if (kernel_size < 0) {
1324             fprintf(stderr, "qemu: could not load kernel '%s'\n",
1325                     kernel_filename);
1326             exit(1);
1327         }
1328 
1329         /* load initrd */
1330         if (initrd_filename) {
1331             /* Try to locate the initrd in the gap between the kernel
1332              * and the firmware. Add a bit of space just in case
1333              */
1334             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1335             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1336                                               load_limit - initrd_base);
1337             if (initrd_size < 0) {
1338                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1339                         initrd_filename);
1340                 exit(1);
1341             }
1342         } else {
1343             initrd_base = 0;
1344             initrd_size = 0;
1345         }
1346     }
1347 
1348     if (bios_name == NULL) {
1349         bios_name = FW_FILE_NAME;
1350     }
1351     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1352     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1353     if (fw_size < 0) {
1354         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1355         exit(1);
1356     }
1357     g_free(filename);
1358 
1359     spapr->entry_point = 0x100;
1360 
1361     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1362     register_savevm_live(NULL, "spapr/htab", -1, 1,
1363                          &savevm_htab_handlers, spapr);
1364 
1365     /* Prepare the device tree */
1366     spapr->fdt_skel = spapr_create_fdt_skel(cpu_model,
1367                                             initrd_base, initrd_size,
1368                                             kernel_size, kernel_le,
1369                                             boot_device, kernel_cmdline,
1370                                             spapr->epow_irq);
1371     assert(spapr->fdt_skel != NULL);
1372 }
1373 
1374 static QEMUMachine spapr_machine = {
1375     .name = "pseries",
1376     .desc = "pSeries Logical Partition (PAPR compliant)",
1377     .is_default = 1,
1378     .init = ppc_spapr_init,
1379     .reset = ppc_spapr_reset,
1380     .block_default_type = IF_SCSI,
1381     .max_cpus = MAX_CPUS,
1382     .no_parallel = 1,
1383     .default_boot_order = NULL,
1384 };
1385 
1386 static void spapr_machine_init(void)
1387 {
1388     qemu_register_machine(&spapr_machine);
1389 }
1390 
1391 machine_init(spapr_machine_init);
1392