xref: /openbmc/qemu/hw/ppc/spapr.c (revision d6032e06)
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 continuously.
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     CPU_FOREACH(cpu) {
208         DeviceClass *dc = DEVICE_GET_CLASS(cpu);
209         uint32_t associativity[] = {cpu_to_be32(0x5),
210                                     cpu_to_be32(0x0),
211                                     cpu_to_be32(0x0),
212                                     cpu_to_be32(0x0),
213                                     cpu_to_be32(cpu->numa_node),
214                                     cpu_to_be32(cpu->cpu_index)};
215 
216         if ((cpu->cpu_index % smt) != 0) {
217             continue;
218         }
219 
220         snprintf(cpu_model, 32, "/cpus/%s@%x", dc->fw_name,
221                  cpu->cpu_index);
222 
223         offset = fdt_path_offset(fdt, cpu_model);
224         if (offset < 0) {
225             return offset;
226         }
227 
228         if (nb_numa_nodes > 1) {
229             ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
230                               sizeof(associativity));
231             if (ret < 0) {
232                 return ret;
233             }
234         }
235 
236         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
237                           pft_size_prop, sizeof(pft_size_prop));
238         if (ret < 0) {
239             return ret;
240         }
241     }
242     return ret;
243 }
244 
245 
246 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
247                                      size_t maxsize)
248 {
249     size_t maxcells = maxsize / sizeof(uint32_t);
250     int i, j, count;
251     uint32_t *p = prop;
252 
253     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
254         struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
255 
256         if (!sps->page_shift) {
257             break;
258         }
259         for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
260             if (sps->enc[count].page_shift == 0) {
261                 break;
262             }
263         }
264         if ((p - prop) >= (maxcells - 3 - count * 2)) {
265             break;
266         }
267         *(p++) = cpu_to_be32(sps->page_shift);
268         *(p++) = cpu_to_be32(sps->slb_enc);
269         *(p++) = cpu_to_be32(count);
270         for (j = 0; j < count; j++) {
271             *(p++) = cpu_to_be32(sps->enc[j].page_shift);
272             *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
273         }
274     }
275 
276     return (p - prop) * sizeof(uint32_t);
277 }
278 
279 #define _FDT(exp) \
280     do { \
281         int ret = (exp);                                           \
282         if (ret < 0) {                                             \
283             fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
284                     #exp, fdt_strerror(ret));                      \
285             exit(1);                                               \
286         }                                                          \
287     } while (0)
288 
289 
290 static void *spapr_create_fdt_skel(hwaddr initrd_base,
291                                    hwaddr initrd_size,
292                                    hwaddr kernel_size,
293                                    bool little_endian,
294                                    const char *boot_device,
295                                    const char *kernel_cmdline,
296                                    uint32_t epow_irq)
297 {
298     void *fdt;
299     CPUState *cs;
300     uint32_t start_prop = cpu_to_be32(initrd_base);
301     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
302     char hypertas_prop[] = "hcall-pft\0hcall-term\0hcall-dabr\0hcall-interrupt"
303         "\0hcall-tce\0hcall-vio\0hcall-splpar\0hcall-bulk\0hcall-set-mode";
304     char qemu_hypertas_prop[] = "hcall-memop1";
305     uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
306     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(smp_cpus)};
307     int i, smt = kvmppc_smt_threads();
308     unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
309 
310     fdt = g_malloc0(FDT_MAX_SIZE);
311     _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
312 
313     if (kernel_size) {
314         _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
315     }
316     if (initrd_size) {
317         _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
318     }
319     _FDT((fdt_finish_reservemap(fdt)));
320 
321     /* Root node */
322     _FDT((fdt_begin_node(fdt, "")));
323     _FDT((fdt_property_string(fdt, "device_type", "chrp")));
324     _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
325     _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
326 
327     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
328     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
329 
330     /* /chosen */
331     _FDT((fdt_begin_node(fdt, "chosen")));
332 
333     /* Set Form1_affinity */
334     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
335 
336     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
337     _FDT((fdt_property(fdt, "linux,initrd-start",
338                        &start_prop, sizeof(start_prop))));
339     _FDT((fdt_property(fdt, "linux,initrd-end",
340                        &end_prop, sizeof(end_prop))));
341     if (kernel_size) {
342         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
343                               cpu_to_be64(kernel_size) };
344 
345         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
346         if (little_endian) {
347             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
348         }
349     }
350     if (boot_device) {
351         _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
352     }
353     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
354     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
355     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
356 
357     _FDT((fdt_end_node(fdt)));
358 
359     /* cpus */
360     _FDT((fdt_begin_node(fdt, "cpus")));
361 
362     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
363     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
364 
365     CPU_FOREACH(cs) {
366         PowerPCCPU *cpu = POWERPC_CPU(cs);
367         CPUPPCState *env = &cpu->env;
368         DeviceClass *dc = DEVICE_GET_CLASS(cs);
369         PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
370         int index = cs->cpu_index;
371         uint32_t servers_prop[smp_threads];
372         uint32_t gservers_prop[smp_threads * 2];
373         char *nodename;
374         uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
375                            0xffffffff, 0xffffffff};
376         uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
377         uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
378         uint32_t page_sizes_prop[64];
379         size_t page_sizes_prop_size;
380 
381         if ((index % smt) != 0) {
382             continue;
383         }
384 
385         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
386 
387         _FDT((fdt_begin_node(fdt, nodename)));
388 
389         g_free(nodename);
390 
391         _FDT((fdt_property_cell(fdt, "reg", index)));
392         _FDT((fdt_property_string(fdt, "device_type", "cpu")));
393 
394         _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
395         _FDT((fdt_property_cell(fdt, "d-cache-block-size",
396                                 env->dcache_line_size)));
397         _FDT((fdt_property_cell(fdt, "d-cache-line-size",
398                                 env->dcache_line_size)));
399         _FDT((fdt_property_cell(fdt, "i-cache-block-size",
400                                 env->icache_line_size)));
401         _FDT((fdt_property_cell(fdt, "i-cache-line-size",
402                                 env->icache_line_size)));
403 
404         if (pcc->l1_dcache_size) {
405             _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
406         } else {
407             fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
408         }
409         if (pcc->l1_icache_size) {
410             _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
411         } else {
412             fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
413         }
414 
415         _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
416         _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
417         _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
418         _FDT((fdt_property_string(fdt, "status", "okay")));
419         _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
420 
421         /* Build interrupt servers and gservers properties */
422         for (i = 0; i < smp_threads; i++) {
423             servers_prop[i] = cpu_to_be32(index + i);
424             /* Hack, direct the group queues back to cpu 0 */
425             gservers_prop[i*2] = cpu_to_be32(index + i);
426             gservers_prop[i*2 + 1] = 0;
427         }
428         _FDT((fdt_property(fdt, "ibm,ppc-interrupt-server#s",
429                            servers_prop, sizeof(servers_prop))));
430         _FDT((fdt_property(fdt, "ibm,ppc-interrupt-gserver#s",
431                            gservers_prop, sizeof(gservers_prop))));
432 
433         if (env->spr_cb[SPR_PURR].oea_read) {
434             _FDT((fdt_property(fdt, "ibm,purr", NULL, 0)));
435         }
436 
437         if (env->mmu_model & POWERPC_MMU_1TSEG) {
438             _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
439                                segs, sizeof(segs))));
440         }
441 
442         /* Advertise VMX/VSX (vector extensions) if available
443          *   0 / no property == no vector extensions
444          *   1               == VMX / Altivec available
445          *   2               == VSX available */
446         if (env->insns_flags & PPC_ALTIVEC) {
447             uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
448 
449             _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
450         }
451 
452         /* Advertise DFP (Decimal Floating Point) if available
453          *   0 / no property == no DFP
454          *   1               == DFP available */
455         if (env->insns_flags2 & PPC2_DFP) {
456             _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
457         }
458 
459         page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
460                                                       sizeof(page_sizes_prop));
461         if (page_sizes_prop_size) {
462             _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
463                                page_sizes_prop, page_sizes_prop_size)));
464         }
465 
466         _FDT((fdt_end_node(fdt)));
467     }
468 
469     _FDT((fdt_end_node(fdt)));
470 
471     /* RTAS */
472     _FDT((fdt_begin_node(fdt, "rtas")));
473 
474     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas_prop,
475                        sizeof(hypertas_prop))));
476     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas_prop,
477                        sizeof(qemu_hypertas_prop))));
478 
479     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
480         refpoints, sizeof(refpoints))));
481 
482     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
483 
484     _FDT((fdt_end_node(fdt)));
485 
486     /* interrupt controller */
487     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
488 
489     _FDT((fdt_property_string(fdt, "device_type",
490                               "PowerPC-External-Interrupt-Presentation")));
491     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
492     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
493     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
494                        interrupt_server_ranges_prop,
495                        sizeof(interrupt_server_ranges_prop))));
496     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
497     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
498     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
499 
500     _FDT((fdt_end_node(fdt)));
501 
502     /* vdevice */
503     _FDT((fdt_begin_node(fdt, "vdevice")));
504 
505     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
506     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
507     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
508     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
509     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
510     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
511 
512     _FDT((fdt_end_node(fdt)));
513 
514     /* event-sources */
515     spapr_events_fdt_skel(fdt, epow_irq);
516 
517     _FDT((fdt_end_node(fdt))); /* close root node */
518     _FDT((fdt_finish(fdt)));
519 
520     return fdt;
521 }
522 
523 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
524 {
525     uint32_t associativity[] = {cpu_to_be32(0x4), cpu_to_be32(0x0),
526                                 cpu_to_be32(0x0), cpu_to_be32(0x0),
527                                 cpu_to_be32(0x0)};
528     char mem_name[32];
529     hwaddr node0_size, mem_start, node_size;
530     uint64_t mem_reg_property[2];
531     int i, off;
532 
533     /* memory node(s) */
534     if (nb_numa_nodes > 1 && node_mem[0] < ram_size) {
535         node0_size = node_mem[0];
536     } else {
537         node0_size = ram_size;
538     }
539 
540     /* RMA */
541     mem_reg_property[0] = 0;
542     mem_reg_property[1] = cpu_to_be64(spapr->rma_size);
543     off = fdt_add_subnode(fdt, 0, "memory@0");
544     _FDT(off);
545     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
546     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
547                       sizeof(mem_reg_property))));
548     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
549                       sizeof(associativity))));
550 
551     /* RAM: Node 0 */
552     if (node0_size > spapr->rma_size) {
553         mem_reg_property[0] = cpu_to_be64(spapr->rma_size);
554         mem_reg_property[1] = cpu_to_be64(node0_size - spapr->rma_size);
555 
556         sprintf(mem_name, "memory@" TARGET_FMT_lx, spapr->rma_size);
557         off = fdt_add_subnode(fdt, 0, mem_name);
558         _FDT(off);
559         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
560         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
561                           sizeof(mem_reg_property))));
562         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
563                           sizeof(associativity))));
564     }
565 
566     /* RAM: Node 1 and beyond */
567     mem_start = node0_size;
568     for (i = 1; i < nb_numa_nodes; i++) {
569         mem_reg_property[0] = cpu_to_be64(mem_start);
570         if (mem_start >= ram_size) {
571             node_size = 0;
572         } else {
573             node_size = node_mem[i];
574             if (node_size > ram_size - mem_start) {
575                 node_size = ram_size - mem_start;
576             }
577         }
578         mem_reg_property[1] = cpu_to_be64(node_size);
579         associativity[3] = associativity[4] = cpu_to_be32(i);
580         sprintf(mem_name, "memory@" TARGET_FMT_lx, mem_start);
581         off = fdt_add_subnode(fdt, 0, mem_name);
582         _FDT(off);
583         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
584         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
585                           sizeof(mem_reg_property))));
586         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
587                           sizeof(associativity))));
588         mem_start += node_size;
589     }
590 
591     return 0;
592 }
593 
594 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
595                                hwaddr fdt_addr,
596                                hwaddr rtas_addr,
597                                hwaddr rtas_size)
598 {
599     int ret;
600     void *fdt;
601     sPAPRPHBState *phb;
602 
603     fdt = g_malloc(FDT_MAX_SIZE);
604 
605     /* open out the base tree into a temp buffer for the final tweaks */
606     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
607 
608     ret = spapr_populate_memory(spapr, fdt);
609     if (ret < 0) {
610         fprintf(stderr, "couldn't setup memory nodes in fdt\n");
611         exit(1);
612     }
613 
614     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
615     if (ret < 0) {
616         fprintf(stderr, "couldn't setup vio devices in fdt\n");
617         exit(1);
618     }
619 
620     QLIST_FOREACH(phb, &spapr->phbs, list) {
621         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
622     }
623 
624     if (ret < 0) {
625         fprintf(stderr, "couldn't setup PCI devices in fdt\n");
626         exit(1);
627     }
628 
629     /* RTAS */
630     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
631     if (ret < 0) {
632         fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
633     }
634 
635     /* Advertise NUMA via ibm,associativity */
636     ret = spapr_fixup_cpu_dt(fdt, spapr);
637     if (ret < 0) {
638         fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
639     }
640 
641     if (!spapr->has_graphics) {
642         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
643     }
644 
645     _FDT((fdt_pack(fdt)));
646 
647     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
648         hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
649                  fdt_totalsize(fdt), FDT_MAX_SIZE);
650         exit(1);
651     }
652 
653     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
654 
655     g_free(fdt);
656 }
657 
658 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
659 {
660     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
661 }
662 
663 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
664 {
665     CPUPPCState *env = &cpu->env;
666 
667     if (msr_pr) {
668         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
669         env->gpr[3] = H_PRIVILEGE;
670     } else {
671         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
672     }
673 }
674 
675 static void spapr_reset_htab(sPAPREnvironment *spapr)
676 {
677     long shift;
678 
679     /* allocate hash page table.  For now we always make this 16mb,
680      * later we should probably make it scale to the size of guest
681      * RAM */
682 
683     shift = kvmppc_reset_htab(spapr->htab_shift);
684 
685     if (shift > 0) {
686         /* Kernel handles htab, we don't need to allocate one */
687         spapr->htab_shift = shift;
688     } else {
689         if (!spapr->htab) {
690             /* Allocate an htab if we don't yet have one */
691             spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
692         }
693 
694         /* And clear it */
695         memset(spapr->htab, 0, HTAB_SIZE(spapr));
696     }
697 
698     /* Update the RMA size if necessary */
699     if (spapr->vrma_adjust) {
700         hwaddr node0_size = (nb_numa_nodes > 1) ? node_mem[0] : ram_size;
701         spapr->rma_size = kvmppc_rma_size(node0_size, spapr->htab_shift);
702     }
703 }
704 
705 static void ppc_spapr_reset(void)
706 {
707     PowerPCCPU *first_ppc_cpu;
708 
709     /* Reset the hash table & recalc the RMA */
710     spapr_reset_htab(spapr);
711 
712     qemu_devices_reset();
713 
714     /* Load the fdt */
715     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
716                        spapr->rtas_size);
717 
718     /* Set up the entry state */
719     first_ppc_cpu = POWERPC_CPU(first_cpu);
720     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
721     first_ppc_cpu->env.gpr[5] = 0;
722     first_cpu->halted = 0;
723     first_ppc_cpu->env.nip = spapr->entry_point;
724 
725 }
726 
727 static void spapr_cpu_reset(void *opaque)
728 {
729     PowerPCCPU *cpu = opaque;
730     CPUState *cs = CPU(cpu);
731     CPUPPCState *env = &cpu->env;
732 
733     cpu_reset(cs);
734 
735     /* All CPUs start halted.  CPU0 is unhalted from the machine level
736      * reset code and the rest are explicitly started up by the guest
737      * using an RTAS call */
738     cs->halted = 1;
739 
740     env->spr[SPR_HIOR] = 0;
741 
742     env->external_htab = (uint8_t *)spapr->htab;
743     env->htab_base = -1;
744     env->htab_mask = HTAB_SIZE(spapr) - 1;
745     env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
746         (spapr->htab_shift - 18);
747 }
748 
749 static void spapr_create_nvram(sPAPREnvironment *spapr)
750 {
751     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
752     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
753 
754     if (dinfo) {
755         qdev_prop_set_drive_nofail(dev, "drive", dinfo->bdrv);
756     }
757 
758     qdev_init_nofail(dev);
759 
760     spapr->nvram = (struct sPAPRNVRAM *)dev;
761 }
762 
763 /* Returns whether we want to use VGA or not */
764 static int spapr_vga_init(PCIBus *pci_bus)
765 {
766     switch (vga_interface_type) {
767     case VGA_NONE:
768     case VGA_STD:
769         return pci_vga_init(pci_bus) != NULL;
770     default:
771         fprintf(stderr, "This vga model is not supported,"
772                 "currently it only supports -vga std\n");
773         exit(0);
774         break;
775     }
776 }
777 
778 static const VMStateDescription vmstate_spapr = {
779     .name = "spapr",
780     .version_id = 1,
781     .minimum_version_id = 1,
782     .minimum_version_id_old = 1,
783     .fields      = (VMStateField []) {
784         VMSTATE_UINT32(next_irq, sPAPREnvironment),
785 
786         /* RTC offset */
787         VMSTATE_UINT64(rtc_offset, sPAPREnvironment),
788 
789         VMSTATE_END_OF_LIST()
790     },
791 };
792 
793 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
794 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
795 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
796 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
797 
798 static int htab_save_setup(QEMUFile *f, void *opaque)
799 {
800     sPAPREnvironment *spapr = opaque;
801 
802     /* "Iteration" header */
803     qemu_put_be32(f, spapr->htab_shift);
804 
805     if (spapr->htab) {
806         spapr->htab_save_index = 0;
807         spapr->htab_first_pass = true;
808     } else {
809         assert(kvm_enabled());
810 
811         spapr->htab_fd = kvmppc_get_htab_fd(false);
812         if (spapr->htab_fd < 0) {
813             fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
814                     strerror(errno));
815             return -1;
816         }
817     }
818 
819 
820     return 0;
821 }
822 
823 static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
824                                  int64_t max_ns)
825 {
826     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
827     int index = spapr->htab_save_index;
828     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
829 
830     assert(spapr->htab_first_pass);
831 
832     do {
833         int chunkstart;
834 
835         /* Consume invalid HPTEs */
836         while ((index < htabslots)
837                && !HPTE_VALID(HPTE(spapr->htab, index))) {
838             index++;
839             CLEAN_HPTE(HPTE(spapr->htab, index));
840         }
841 
842         /* Consume valid HPTEs */
843         chunkstart = index;
844         while ((index < htabslots)
845                && HPTE_VALID(HPTE(spapr->htab, index))) {
846             index++;
847             CLEAN_HPTE(HPTE(spapr->htab, index));
848         }
849 
850         if (index > chunkstart) {
851             int n_valid = index - chunkstart;
852 
853             qemu_put_be32(f, chunkstart);
854             qemu_put_be16(f, n_valid);
855             qemu_put_be16(f, 0);
856             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
857                             HASH_PTE_SIZE_64 * n_valid);
858 
859             if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
860                 break;
861             }
862         }
863     } while ((index < htabslots) && !qemu_file_rate_limit(f));
864 
865     if (index >= htabslots) {
866         assert(index == htabslots);
867         index = 0;
868         spapr->htab_first_pass = false;
869     }
870     spapr->htab_save_index = index;
871 }
872 
873 static int htab_save_later_pass(QEMUFile *f, sPAPREnvironment *spapr,
874                                 int64_t max_ns)
875 {
876     bool final = max_ns < 0;
877     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
878     int examined = 0, sent = 0;
879     int index = spapr->htab_save_index;
880     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
881 
882     assert(!spapr->htab_first_pass);
883 
884     do {
885         int chunkstart, invalidstart;
886 
887         /* Consume non-dirty HPTEs */
888         while ((index < htabslots)
889                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
890             index++;
891             examined++;
892         }
893 
894         chunkstart = index;
895         /* Consume valid dirty HPTEs */
896         while ((index < htabslots)
897                && HPTE_DIRTY(HPTE(spapr->htab, index))
898                && HPTE_VALID(HPTE(spapr->htab, index))) {
899             CLEAN_HPTE(HPTE(spapr->htab, index));
900             index++;
901             examined++;
902         }
903 
904         invalidstart = index;
905         /* Consume invalid dirty HPTEs */
906         while ((index < htabslots)
907                && HPTE_DIRTY(HPTE(spapr->htab, index))
908                && !HPTE_VALID(HPTE(spapr->htab, index))) {
909             CLEAN_HPTE(HPTE(spapr->htab, index));
910             index++;
911             examined++;
912         }
913 
914         if (index > chunkstart) {
915             int n_valid = invalidstart - chunkstart;
916             int n_invalid = index - invalidstart;
917 
918             qemu_put_be32(f, chunkstart);
919             qemu_put_be16(f, n_valid);
920             qemu_put_be16(f, n_invalid);
921             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
922                             HASH_PTE_SIZE_64 * n_valid);
923             sent += index - chunkstart;
924 
925             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
926                 break;
927             }
928         }
929 
930         if (examined >= htabslots) {
931             break;
932         }
933 
934         if (index >= htabslots) {
935             assert(index == htabslots);
936             index = 0;
937         }
938     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
939 
940     if (index >= htabslots) {
941         assert(index == htabslots);
942         index = 0;
943     }
944 
945     spapr->htab_save_index = index;
946 
947     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
948 }
949 
950 #define MAX_ITERATION_NS    5000000 /* 5 ms */
951 #define MAX_KVM_BUF_SIZE    2048
952 
953 static int htab_save_iterate(QEMUFile *f, void *opaque)
954 {
955     sPAPREnvironment *spapr = opaque;
956     int rc = 0;
957 
958     /* Iteration header */
959     qemu_put_be32(f, 0);
960 
961     if (!spapr->htab) {
962         assert(kvm_enabled());
963 
964         rc = kvmppc_save_htab(f, spapr->htab_fd,
965                               MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
966         if (rc < 0) {
967             return rc;
968         }
969     } else  if (spapr->htab_first_pass) {
970         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
971     } else {
972         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
973     }
974 
975     /* End marker */
976     qemu_put_be32(f, 0);
977     qemu_put_be16(f, 0);
978     qemu_put_be16(f, 0);
979 
980     return rc;
981 }
982 
983 static int htab_save_complete(QEMUFile *f, void *opaque)
984 {
985     sPAPREnvironment *spapr = opaque;
986 
987     /* Iteration header */
988     qemu_put_be32(f, 0);
989 
990     if (!spapr->htab) {
991         int rc;
992 
993         assert(kvm_enabled());
994 
995         rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
996         if (rc < 0) {
997             return rc;
998         }
999         close(spapr->htab_fd);
1000         spapr->htab_fd = -1;
1001     } else {
1002         htab_save_later_pass(f, spapr, -1);
1003     }
1004 
1005     /* End marker */
1006     qemu_put_be32(f, 0);
1007     qemu_put_be16(f, 0);
1008     qemu_put_be16(f, 0);
1009 
1010     return 0;
1011 }
1012 
1013 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1014 {
1015     sPAPREnvironment *spapr = opaque;
1016     uint32_t section_hdr;
1017     int fd = -1;
1018 
1019     if (version_id < 1 || version_id > 1) {
1020         fprintf(stderr, "htab_load() bad version\n");
1021         return -EINVAL;
1022     }
1023 
1024     section_hdr = qemu_get_be32(f);
1025 
1026     if (section_hdr) {
1027         /* First section, just the hash shift */
1028         if (spapr->htab_shift != section_hdr) {
1029             return -EINVAL;
1030         }
1031         return 0;
1032     }
1033 
1034     if (!spapr->htab) {
1035         assert(kvm_enabled());
1036 
1037         fd = kvmppc_get_htab_fd(true);
1038         if (fd < 0) {
1039             fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1040                     strerror(errno));
1041         }
1042     }
1043 
1044     while (true) {
1045         uint32_t index;
1046         uint16_t n_valid, n_invalid;
1047 
1048         index = qemu_get_be32(f);
1049         n_valid = qemu_get_be16(f);
1050         n_invalid = qemu_get_be16(f);
1051 
1052         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1053             /* End of Stream */
1054             break;
1055         }
1056 
1057         if ((index + n_valid + n_invalid) >
1058             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1059             /* Bad index in stream */
1060             fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1061                     "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1062                     spapr->htab_shift);
1063             return -EINVAL;
1064         }
1065 
1066         if (spapr->htab) {
1067             if (n_valid) {
1068                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1069                                 HASH_PTE_SIZE_64 * n_valid);
1070             }
1071             if (n_invalid) {
1072                 memset(HPTE(spapr->htab, index + n_valid), 0,
1073                        HASH_PTE_SIZE_64 * n_invalid);
1074             }
1075         } else {
1076             int rc;
1077 
1078             assert(fd >= 0);
1079 
1080             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1081             if (rc < 0) {
1082                 return rc;
1083             }
1084         }
1085     }
1086 
1087     if (!spapr->htab) {
1088         assert(fd >= 0);
1089         close(fd);
1090     }
1091 
1092     return 0;
1093 }
1094 
1095 static SaveVMHandlers savevm_htab_handlers = {
1096     .save_live_setup = htab_save_setup,
1097     .save_live_iterate = htab_save_iterate,
1098     .save_live_complete = htab_save_complete,
1099     .load_state = htab_load,
1100 };
1101 
1102 /* pSeries LPAR / sPAPR hardware init */
1103 static void ppc_spapr_init(QEMUMachineInitArgs *args)
1104 {
1105     ram_addr_t ram_size = args->ram_size;
1106     const char *cpu_model = args->cpu_model;
1107     const char *kernel_filename = args->kernel_filename;
1108     const char *kernel_cmdline = args->kernel_cmdline;
1109     const char *initrd_filename = args->initrd_filename;
1110     const char *boot_device = args->boot_order;
1111     PowerPCCPU *cpu;
1112     CPUPPCState *env;
1113     PCIHostState *phb;
1114     int i;
1115     MemoryRegion *sysmem = get_system_memory();
1116     MemoryRegion *ram = g_new(MemoryRegion, 1);
1117     hwaddr rma_alloc_size;
1118     hwaddr node0_size = (nb_numa_nodes > 1) ? node_mem[0] : ram_size;
1119     uint32_t initrd_base = 0;
1120     long kernel_size = 0, initrd_size = 0;
1121     long load_limit, rtas_limit, fw_size;
1122     bool kernel_le = false;
1123     char *filename;
1124 
1125     msi_supported = true;
1126 
1127     spapr = g_malloc0(sizeof(*spapr));
1128     QLIST_INIT(&spapr->phbs);
1129 
1130     cpu_ppc_hypercall = emulate_spapr_hypercall;
1131 
1132     /* Allocate RMA if necessary */
1133     rma_alloc_size = kvmppc_alloc_rma("ppc_spapr.rma", sysmem);
1134 
1135     if (rma_alloc_size == -1) {
1136         hw_error("qemu: Unable to create RMA\n");
1137         exit(1);
1138     }
1139 
1140     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1141         spapr->rma_size = rma_alloc_size;
1142     } else {
1143         spapr->rma_size = node0_size;
1144 
1145         /* With KVM, we don't actually know whether KVM supports an
1146          * unbounded RMA (PR KVM) or is limited by the hash table size
1147          * (HV KVM using VRMA), so we always assume the latter
1148          *
1149          * In that case, we also limit the initial allocations for RTAS
1150          * etc... to 256M since we have no way to know what the VRMA size
1151          * is going to be as it depends on the size of the hash table
1152          * isn't determined yet.
1153          */
1154         if (kvm_enabled()) {
1155             spapr->vrma_adjust = 1;
1156             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1157         }
1158     }
1159 
1160     if (spapr->rma_size > node0_size) {
1161         fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1162                 spapr->rma_size);
1163         exit(1);
1164     }
1165 
1166     /* We place the device tree and RTAS just below either the top of the RMA,
1167      * or just below 2GB, whichever is lowere, so that it can be
1168      * processed with 32-bit real mode code if necessary */
1169     rtas_limit = MIN(spapr->rma_size, 0x80000000);
1170     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1171     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1172     load_limit = spapr->fdt_addr - FW_OVERHEAD;
1173 
1174     /* We aim for a hash table of size 1/128 the size of RAM.  The
1175      * normal rule of thumb is 1/64 the size of RAM, but that's much
1176      * more than needed for the Linux guests we support. */
1177     spapr->htab_shift = 18; /* Minimum architected size */
1178     while (spapr->htab_shift <= 46) {
1179         if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
1180             break;
1181         }
1182         spapr->htab_shift++;
1183     }
1184 
1185     /* Set up Interrupt Controller before we create the VCPUs */
1186     spapr->icp = xics_system_init(smp_cpus * kvmppc_smt_threads() / smp_threads,
1187                                   XICS_IRQS);
1188     spapr->next_irq = XICS_IRQ_BASE;
1189 
1190     /* init CPUs */
1191     if (cpu_model == NULL) {
1192         cpu_model = kvm_enabled() ? "host" : "POWER7";
1193     }
1194     for (i = 0; i < smp_cpus; i++) {
1195         cpu = cpu_ppc_init(cpu_model);
1196         if (cpu == NULL) {
1197             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1198             exit(1);
1199         }
1200         env = &cpu->env;
1201 
1202         /* Set time-base frequency to 512 MHz */
1203         cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1204 
1205         /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1206          * MSR[IP] should never be set.
1207          */
1208         env->msr_mask &= ~(1 << 6);
1209 
1210         /* Tell KVM that we're in PAPR mode */
1211         if (kvm_enabled()) {
1212             kvmppc_set_papr(cpu);
1213         }
1214 
1215         xics_cpu_setup(spapr->icp, cpu);
1216 
1217         qemu_register_reset(spapr_cpu_reset, cpu);
1218     }
1219 
1220     /* allocate RAM */
1221     spapr->ram_limit = ram_size;
1222     if (spapr->ram_limit > rma_alloc_size) {
1223         ram_addr_t nonrma_base = rma_alloc_size;
1224         ram_addr_t nonrma_size = spapr->ram_limit - rma_alloc_size;
1225 
1226         memory_region_init_ram(ram, NULL, "ppc_spapr.ram", nonrma_size);
1227         vmstate_register_ram_global(ram);
1228         memory_region_add_subregion(sysmem, nonrma_base, ram);
1229     }
1230 
1231     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1232     spapr->rtas_size = load_image_targphys(filename, spapr->rtas_addr,
1233                                            rtas_limit - spapr->rtas_addr);
1234     if (spapr->rtas_size < 0) {
1235         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1236         exit(1);
1237     }
1238     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1239         hw_error("RTAS too big ! 0x%lx bytes (max is 0x%x)\n",
1240                  spapr->rtas_size, RTAS_MAX_SIZE);
1241         exit(1);
1242     }
1243     g_free(filename);
1244 
1245     /* Set up EPOW events infrastructure */
1246     spapr_events_init(spapr);
1247 
1248     /* Set up VIO bus */
1249     spapr->vio_bus = spapr_vio_bus_init();
1250 
1251     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1252         if (serial_hds[i]) {
1253             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1254         }
1255     }
1256 
1257     /* We always have at least the nvram device on VIO */
1258     spapr_create_nvram(spapr);
1259 
1260     /* Set up PCI */
1261     spapr_pci_msi_init(spapr, SPAPR_PCI_MSI_WINDOW);
1262     spapr_pci_rtas_init();
1263 
1264     phb = spapr_create_phb(spapr, 0);
1265 
1266     for (i = 0; i < nb_nics; i++) {
1267         NICInfo *nd = &nd_table[i];
1268 
1269         if (!nd->model) {
1270             nd->model = g_strdup("ibmveth");
1271         }
1272 
1273         if (strcmp(nd->model, "ibmveth") == 0) {
1274             spapr_vlan_create(spapr->vio_bus, nd);
1275         } else {
1276             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1277         }
1278     }
1279 
1280     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1281         spapr_vscsi_create(spapr->vio_bus);
1282     }
1283 
1284     /* Graphics */
1285     if (spapr_vga_init(phb->bus)) {
1286         spapr->has_graphics = true;
1287     }
1288 
1289     if (usb_enabled(spapr->has_graphics)) {
1290         pci_create_simple(phb->bus, -1, "pci-ohci");
1291         if (spapr->has_graphics) {
1292             usbdevice_create("keyboard");
1293             usbdevice_create("mouse");
1294         }
1295     }
1296 
1297     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1298         fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1299                 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1300         exit(1);
1301     }
1302 
1303     if (kernel_filename) {
1304         uint64_t lowaddr = 0;
1305 
1306         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1307                                NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1308         if (kernel_size < 0) {
1309             kernel_size = load_elf(kernel_filename,
1310                                    translate_kernel_address, NULL,
1311                                    NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1312             kernel_le = kernel_size > 0;
1313         }
1314         if (kernel_size < 0) {
1315             kernel_size = load_image_targphys(kernel_filename,
1316                                               KERNEL_LOAD_ADDR,
1317                                               load_limit - KERNEL_LOAD_ADDR);
1318         }
1319         if (kernel_size < 0) {
1320             fprintf(stderr, "qemu: could not load kernel '%s'\n",
1321                     kernel_filename);
1322             exit(1);
1323         }
1324 
1325         /* load initrd */
1326         if (initrd_filename) {
1327             /* Try to locate the initrd in the gap between the kernel
1328              * and the firmware. Add a bit of space just in case
1329              */
1330             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1331             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1332                                               load_limit - initrd_base);
1333             if (initrd_size < 0) {
1334                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1335                         initrd_filename);
1336                 exit(1);
1337             }
1338         } else {
1339             initrd_base = 0;
1340             initrd_size = 0;
1341         }
1342     }
1343 
1344     if (bios_name == NULL) {
1345         bios_name = FW_FILE_NAME;
1346     }
1347     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1348     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1349     if (fw_size < 0) {
1350         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1351         exit(1);
1352     }
1353     g_free(filename);
1354 
1355     spapr->entry_point = 0x100;
1356 
1357     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1358     register_savevm_live(NULL, "spapr/htab", -1, 1,
1359                          &savevm_htab_handlers, spapr);
1360 
1361     /* Prepare the device tree */
1362     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1363                                             kernel_size, kernel_le,
1364                                             boot_device, kernel_cmdline,
1365                                             spapr->epow_irq);
1366     assert(spapr->fdt_skel != NULL);
1367 }
1368 
1369 static QEMUMachine spapr_machine = {
1370     .name = "pseries",
1371     .desc = "pSeries Logical Partition (PAPR compliant)",
1372     .is_default = 1,
1373     .init = ppc_spapr_init,
1374     .reset = ppc_spapr_reset,
1375     .block_default_type = IF_SCSI,
1376     .max_cpus = MAX_CPUS,
1377     .no_parallel = 1,
1378     .default_boot_order = NULL,
1379 };
1380 
1381 static void spapr_machine_init(void)
1382 {
1383     qemu_register_machine(&spapr_machine);
1384 }
1385 
1386 machine_init(spapr_machine_init);
1387