xref: /openbmc/qemu/hw/ppc/spapr.c (revision 461a2753)
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 "hw/fw-path-provider.h"
30 #include "elf.h"
31 #include "net/net.h"
32 #include "sysemu/blockdev.h"
33 #include "sysemu/cpus.h"
34 #include "sysemu/kvm.h"
35 #include "kvm_ppc.h"
36 #include "mmu-hash64.h"
37 #include "qom/cpu.h"
38 
39 #include "hw/boards.h"
40 #include "hw/ppc/ppc.h"
41 #include "hw/loader.h"
42 
43 #include "hw/ppc/spapr.h"
44 #include "hw/ppc/spapr_vio.h"
45 #include "hw/pci-host/spapr.h"
46 #include "hw/ppc/xics.h"
47 #include "hw/pci/msi.h"
48 
49 #include "hw/pci/pci.h"
50 #include "hw/scsi/scsi.h"
51 #include "hw/virtio/virtio-scsi.h"
52 
53 #include "exec/address-spaces.h"
54 #include "hw/usb.h"
55 #include "qemu/config-file.h"
56 #include "qemu/error-report.h"
57 #include "trace.h"
58 #include "hw/nmi.h"
59 
60 #include <libfdt.h>
61 
62 /* SLOF memory layout:
63  *
64  * SLOF raw image loaded at 0, copies its romfs right below the flat
65  * device-tree, then position SLOF itself 31M below that
66  *
67  * So we set FW_OVERHEAD to 40MB which should account for all of that
68  * and more
69  *
70  * We load our kernel at 4M, leaving space for SLOF initial image
71  */
72 #define FDT_MAX_SIZE            0x40000
73 #define RTAS_MAX_SIZE           0x10000
74 #define FW_MAX_SIZE             0x400000
75 #define FW_FILE_NAME            "slof.bin"
76 #define FW_OVERHEAD             0x2800000
77 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
78 
79 #define MIN_RMA_SLOF            128UL
80 
81 #define TIMEBASE_FREQ           512000000ULL
82 
83 #define MAX_CPUS                256
84 
85 #define PHANDLE_XICP            0x00001111
86 
87 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
88 
89 typedef struct sPAPRMachineState sPAPRMachineState;
90 
91 #define TYPE_SPAPR_MACHINE      "spapr-machine"
92 #define SPAPR_MACHINE(obj) \
93     OBJECT_CHECK(sPAPRMachineState, (obj), TYPE_SPAPR_MACHINE)
94 
95 /**
96  * sPAPRMachineState:
97  */
98 struct sPAPRMachineState {
99     /*< private >*/
100     MachineState parent_obj;
101 
102     /*< public >*/
103     char *kvm_type;
104 };
105 
106 sPAPREnvironment *spapr;
107 
108 static XICSState *try_create_xics(const char *type, int nr_servers,
109                                   int nr_irqs)
110 {
111     DeviceState *dev;
112 
113     dev = qdev_create(NULL, type);
114     qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
115     qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
116     if (qdev_init(dev) < 0) {
117         return NULL;
118     }
119 
120     return XICS_COMMON(dev);
121 }
122 
123 static XICSState *xics_system_init(int nr_servers, int nr_irqs)
124 {
125     XICSState *icp = NULL;
126 
127     if (kvm_enabled()) {
128         QemuOpts *machine_opts = qemu_get_machine_opts();
129         bool irqchip_allowed = qemu_opt_get_bool(machine_opts,
130                                                 "kernel_irqchip", true);
131         bool irqchip_required = qemu_opt_get_bool(machine_opts,
132                                                   "kernel_irqchip", false);
133         if (irqchip_allowed) {
134             icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs);
135         }
136 
137         if (irqchip_required && !icp) {
138             perror("Failed to create in-kernel XICS\n");
139             abort();
140         }
141     }
142 
143     if (!icp) {
144         icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs);
145     }
146 
147     if (!icp) {
148         perror("Failed to create XICS\n");
149         abort();
150     }
151 
152     return icp;
153 }
154 
155 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
156                                   int smt_threads)
157 {
158     int i, ret = 0;
159     uint32_t servers_prop[smt_threads];
160     uint32_t gservers_prop[smt_threads * 2];
161     int index = ppc_get_vcpu_dt_id(cpu);
162 
163     if (cpu->cpu_version) {
164         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
165         if (ret < 0) {
166             return ret;
167         }
168     }
169 
170     /* Build interrupt servers and gservers properties */
171     for (i = 0; i < smt_threads; i++) {
172         servers_prop[i] = cpu_to_be32(index + i);
173         /* Hack, direct the group queues back to cpu 0 */
174         gservers_prop[i*2] = cpu_to_be32(index + i);
175         gservers_prop[i*2 + 1] = 0;
176     }
177     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
178                       servers_prop, sizeof(servers_prop));
179     if (ret < 0) {
180         return ret;
181     }
182     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
183                       gservers_prop, sizeof(gservers_prop));
184 
185     return ret;
186 }
187 
188 static int spapr_fixup_cpu_dt(void *fdt, sPAPREnvironment *spapr)
189 {
190     int ret = 0, offset, cpus_offset;
191     CPUState *cs;
192     char cpu_model[32];
193     int smt = kvmppc_smt_threads();
194     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
195 
196     CPU_FOREACH(cs) {
197         PowerPCCPU *cpu = POWERPC_CPU(cs);
198         DeviceClass *dc = DEVICE_GET_CLASS(cs);
199         int index = ppc_get_vcpu_dt_id(cpu);
200         uint32_t associativity[] = {cpu_to_be32(0x5),
201                                     cpu_to_be32(0x0),
202                                     cpu_to_be32(0x0),
203                                     cpu_to_be32(0x0),
204                                     cpu_to_be32(cs->numa_node),
205                                     cpu_to_be32(index)};
206 
207         if ((index % smt) != 0) {
208             continue;
209         }
210 
211         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
212 
213         cpus_offset = fdt_path_offset(fdt, "/cpus");
214         if (cpus_offset < 0) {
215             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
216                                           "cpus");
217             if (cpus_offset < 0) {
218                 return cpus_offset;
219             }
220         }
221         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
222         if (offset < 0) {
223             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
224             if (offset < 0) {
225                 return offset;
226             }
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         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
244                                      ppc_get_compat_smt_threads(cpu));
245         if (ret < 0) {
246             return ret;
247         }
248     }
249     return ret;
250 }
251 
252 
253 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
254                                      size_t maxsize)
255 {
256     size_t maxcells = maxsize / sizeof(uint32_t);
257     int i, j, count;
258     uint32_t *p = prop;
259 
260     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
261         struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
262 
263         if (!sps->page_shift) {
264             break;
265         }
266         for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
267             if (sps->enc[count].page_shift == 0) {
268                 break;
269             }
270         }
271         if ((p - prop) >= (maxcells - 3 - count * 2)) {
272             break;
273         }
274         *(p++) = cpu_to_be32(sps->page_shift);
275         *(p++) = cpu_to_be32(sps->slb_enc);
276         *(p++) = cpu_to_be32(count);
277         for (j = 0; j < count; j++) {
278             *(p++) = cpu_to_be32(sps->enc[j].page_shift);
279             *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
280         }
281     }
282 
283     return (p - prop) * sizeof(uint32_t);
284 }
285 
286 #define _FDT(exp) \
287     do { \
288         int ret = (exp);                                           \
289         if (ret < 0) {                                             \
290             fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
291                     #exp, fdt_strerror(ret));                      \
292             exit(1);                                               \
293         }                                                          \
294     } while (0)
295 
296 static void add_str(GString *s, const gchar *s1)
297 {
298     g_string_append_len(s, s1, strlen(s1) + 1);
299 }
300 
301 static void *spapr_create_fdt_skel(hwaddr initrd_base,
302                                    hwaddr initrd_size,
303                                    hwaddr kernel_size,
304                                    bool little_endian,
305                                    const char *boot_device,
306                                    const char *kernel_cmdline,
307                                    uint32_t epow_irq)
308 {
309     void *fdt;
310     CPUState *cs;
311     uint32_t start_prop = cpu_to_be32(initrd_base);
312     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
313     GString *hypertas = g_string_sized_new(256);
314     GString *qemu_hypertas = g_string_sized_new(256);
315     uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
316     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(smp_cpus)};
317     int smt = kvmppc_smt_threads();
318     unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
319     QemuOpts *opts = qemu_opts_find(qemu_find_opts("smp-opts"), NULL);
320     unsigned sockets = opts ? qemu_opt_get_number(opts, "sockets", 0) : 0;
321     uint32_t cpus_per_socket = sockets ? (smp_cpus / sockets) : 1;
322 
323     add_str(hypertas, "hcall-pft");
324     add_str(hypertas, "hcall-term");
325     add_str(hypertas, "hcall-dabr");
326     add_str(hypertas, "hcall-interrupt");
327     add_str(hypertas, "hcall-tce");
328     add_str(hypertas, "hcall-vio");
329     add_str(hypertas, "hcall-splpar");
330     add_str(hypertas, "hcall-bulk");
331     add_str(hypertas, "hcall-set-mode");
332     add_str(qemu_hypertas, "hcall-memop1");
333 
334     fdt = g_malloc0(FDT_MAX_SIZE);
335     _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
336 
337     if (kernel_size) {
338         _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
339     }
340     if (initrd_size) {
341         _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
342     }
343     _FDT((fdt_finish_reservemap(fdt)));
344 
345     /* Root node */
346     _FDT((fdt_begin_node(fdt, "")));
347     _FDT((fdt_property_string(fdt, "device_type", "chrp")));
348     _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
349     _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
350 
351     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
352     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
353 
354     /* /chosen */
355     _FDT((fdt_begin_node(fdt, "chosen")));
356 
357     /* Set Form1_affinity */
358     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
359 
360     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
361     _FDT((fdt_property(fdt, "linux,initrd-start",
362                        &start_prop, sizeof(start_prop))));
363     _FDT((fdt_property(fdt, "linux,initrd-end",
364                        &end_prop, sizeof(end_prop))));
365     if (kernel_size) {
366         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
367                               cpu_to_be64(kernel_size) };
368 
369         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
370         if (little_endian) {
371             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
372         }
373     }
374     if (boot_device) {
375         _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
376     }
377     if (boot_menu) {
378         _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
379     }
380     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
381     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
382     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
383 
384     _FDT((fdt_end_node(fdt)));
385 
386     /* cpus */
387     _FDT((fdt_begin_node(fdt, "cpus")));
388 
389     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
390     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
391 
392     CPU_FOREACH(cs) {
393         PowerPCCPU *cpu = POWERPC_CPU(cs);
394         CPUPPCState *env = &cpu->env;
395         DeviceClass *dc = DEVICE_GET_CLASS(cs);
396         PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
397         int index = ppc_get_vcpu_dt_id(cpu);
398         char *nodename;
399         uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
400                            0xffffffff, 0xffffffff};
401         uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
402         uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
403         uint32_t page_sizes_prop[64];
404         size_t page_sizes_prop_size;
405 
406         if ((index % smt) != 0) {
407             continue;
408         }
409 
410         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
411 
412         _FDT((fdt_begin_node(fdt, nodename)));
413 
414         g_free(nodename);
415 
416         _FDT((fdt_property_cell(fdt, "reg", index)));
417         _FDT((fdt_property_string(fdt, "device_type", "cpu")));
418 
419         _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
420         _FDT((fdt_property_cell(fdt, "d-cache-block-size",
421                                 env->dcache_line_size)));
422         _FDT((fdt_property_cell(fdt, "d-cache-line-size",
423                                 env->dcache_line_size)));
424         _FDT((fdt_property_cell(fdt, "i-cache-block-size",
425                                 env->icache_line_size)));
426         _FDT((fdt_property_cell(fdt, "i-cache-line-size",
427                                 env->icache_line_size)));
428 
429         if (pcc->l1_dcache_size) {
430             _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
431         } else {
432             fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
433         }
434         if (pcc->l1_icache_size) {
435             _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
436         } else {
437             fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
438         }
439 
440         _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
441         _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
442         _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
443         _FDT((fdt_property_string(fdt, "status", "okay")));
444         _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
445 
446         if (env->spr_cb[SPR_PURR].oea_read) {
447             _FDT((fdt_property(fdt, "ibm,purr", NULL, 0)));
448         }
449 
450         if (env->mmu_model & POWERPC_MMU_1TSEG) {
451             _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
452                                segs, sizeof(segs))));
453         }
454 
455         /* Advertise VMX/VSX (vector extensions) if available
456          *   0 / no property == no vector extensions
457          *   1               == VMX / Altivec available
458          *   2               == VSX available */
459         if (env->insns_flags & PPC_ALTIVEC) {
460             uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
461 
462             _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
463         }
464 
465         /* Advertise DFP (Decimal Floating Point) if available
466          *   0 / no property == no DFP
467          *   1               == DFP available */
468         if (env->insns_flags2 & PPC2_DFP) {
469             _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
470         }
471 
472         page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
473                                                       sizeof(page_sizes_prop));
474         if (page_sizes_prop_size) {
475             _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
476                                page_sizes_prop, page_sizes_prop_size)));
477         }
478 
479         _FDT((fdt_property_cell(fdt, "ibm,chip-id",
480                                 cs->cpu_index / cpus_per_socket)));
481 
482         _FDT((fdt_end_node(fdt)));
483     }
484 
485     _FDT((fdt_end_node(fdt)));
486 
487     /* RTAS */
488     _FDT((fdt_begin_node(fdt, "rtas")));
489 
490     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
491         add_str(hypertas, "hcall-multi-tce");
492     }
493     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
494                        hypertas->len)));
495     g_string_free(hypertas, TRUE);
496     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
497                        qemu_hypertas->len)));
498     g_string_free(qemu_hypertas, TRUE);
499 
500     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
501         refpoints, sizeof(refpoints))));
502 
503     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
504 
505     _FDT((fdt_end_node(fdt)));
506 
507     /* interrupt controller */
508     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
509 
510     _FDT((fdt_property_string(fdt, "device_type",
511                               "PowerPC-External-Interrupt-Presentation")));
512     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
513     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
514     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
515                        interrupt_server_ranges_prop,
516                        sizeof(interrupt_server_ranges_prop))));
517     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
518     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
519     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
520 
521     _FDT((fdt_end_node(fdt)));
522 
523     /* vdevice */
524     _FDT((fdt_begin_node(fdt, "vdevice")));
525 
526     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
527     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
528     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
529     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
530     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
531     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
532 
533     _FDT((fdt_end_node(fdt)));
534 
535     /* event-sources */
536     spapr_events_fdt_skel(fdt, epow_irq);
537 
538     /* /hypervisor node */
539     if (kvm_enabled()) {
540         uint8_t hypercall[16];
541 
542         /* indicate KVM hypercall interface */
543         _FDT((fdt_begin_node(fdt, "hypervisor")));
544         _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
545         if (kvmppc_has_cap_fixup_hcalls()) {
546             /*
547              * Older KVM versions with older guest kernels were broken with the
548              * magic page, don't allow the guest to map it.
549              */
550             kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
551                                  sizeof(hypercall));
552             _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
553                               sizeof(hypercall))));
554         }
555         _FDT((fdt_end_node(fdt)));
556     }
557 
558     _FDT((fdt_end_node(fdt))); /* close root node */
559     _FDT((fdt_finish(fdt)));
560 
561     return fdt;
562 }
563 
564 int spapr_h_cas_compose_response(target_ulong addr, target_ulong size)
565 {
566     void *fdt, *fdt_skel;
567     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
568 
569     size -= sizeof(hdr);
570 
571     /* Create sceleton */
572     fdt_skel = g_malloc0(size);
573     _FDT((fdt_create(fdt_skel, size)));
574     _FDT((fdt_begin_node(fdt_skel, "")));
575     _FDT((fdt_end_node(fdt_skel)));
576     _FDT((fdt_finish(fdt_skel)));
577     fdt = g_malloc0(size);
578     _FDT((fdt_open_into(fdt_skel, fdt, size)));
579     g_free(fdt_skel);
580 
581     /* Fix skeleton up */
582     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
583 
584     /* Pack resulting tree */
585     _FDT((fdt_pack(fdt)));
586 
587     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
588         trace_spapr_cas_failed(size);
589         return -1;
590     }
591 
592     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
593     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
594     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
595     g_free(fdt);
596 
597     return 0;
598 }
599 
600 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
601 {
602     uint32_t associativity[] = {cpu_to_be32(0x4), cpu_to_be32(0x0),
603                                 cpu_to_be32(0x0), cpu_to_be32(0x0),
604                                 cpu_to_be32(0x0)};
605     char mem_name[32];
606     hwaddr node0_size, mem_start, node_size;
607     uint64_t mem_reg_property[2];
608     int i, off;
609 
610     /* memory node(s) */
611     if (nb_numa_nodes > 1 && numa_info[0].node_mem < ram_size) {
612         node0_size = numa_info[0].node_mem;
613     } else {
614         node0_size = ram_size;
615     }
616 
617     /* RMA */
618     mem_reg_property[0] = 0;
619     mem_reg_property[1] = cpu_to_be64(spapr->rma_size);
620     off = fdt_add_subnode(fdt, 0, "memory@0");
621     _FDT(off);
622     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
623     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
624                       sizeof(mem_reg_property))));
625     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
626                       sizeof(associativity))));
627 
628     /* RAM: Node 0 */
629     if (node0_size > spapr->rma_size) {
630         mem_reg_property[0] = cpu_to_be64(spapr->rma_size);
631         mem_reg_property[1] = cpu_to_be64(node0_size - spapr->rma_size);
632 
633         sprintf(mem_name, "memory@" TARGET_FMT_lx, spapr->rma_size);
634         off = fdt_add_subnode(fdt, 0, mem_name);
635         _FDT(off);
636         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
637         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
638                           sizeof(mem_reg_property))));
639         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
640                           sizeof(associativity))));
641     }
642 
643     /* RAM: Node 1 and beyond */
644     mem_start = node0_size;
645     for (i = 1; i < nb_numa_nodes; i++) {
646         mem_reg_property[0] = cpu_to_be64(mem_start);
647         if (mem_start >= ram_size) {
648             node_size = 0;
649         } else {
650             node_size = numa_info[i].node_mem;
651             if (node_size > ram_size - mem_start) {
652                 node_size = ram_size - mem_start;
653             }
654         }
655         mem_reg_property[1] = cpu_to_be64(node_size);
656         associativity[3] = associativity[4] = cpu_to_be32(i);
657         sprintf(mem_name, "memory@" TARGET_FMT_lx, mem_start);
658         off = fdt_add_subnode(fdt, 0, mem_name);
659         _FDT(off);
660         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
661         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
662                           sizeof(mem_reg_property))));
663         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
664                           sizeof(associativity))));
665         mem_start += node_size;
666     }
667 
668     return 0;
669 }
670 
671 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
672                                hwaddr fdt_addr,
673                                hwaddr rtas_addr,
674                                hwaddr rtas_size)
675 {
676     int ret, i;
677     size_t cb = 0;
678     char *bootlist;
679     void *fdt;
680     sPAPRPHBState *phb;
681 
682     fdt = g_malloc(FDT_MAX_SIZE);
683 
684     /* open out the base tree into a temp buffer for the final tweaks */
685     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
686 
687     ret = spapr_populate_memory(spapr, fdt);
688     if (ret < 0) {
689         fprintf(stderr, "couldn't setup memory nodes in fdt\n");
690         exit(1);
691     }
692 
693     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
694     if (ret < 0) {
695         fprintf(stderr, "couldn't setup vio devices in fdt\n");
696         exit(1);
697     }
698 
699     QLIST_FOREACH(phb, &spapr->phbs, list) {
700         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
701     }
702 
703     if (ret < 0) {
704         fprintf(stderr, "couldn't setup PCI devices in fdt\n");
705         exit(1);
706     }
707 
708     /* RTAS */
709     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
710     if (ret < 0) {
711         fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
712     }
713 
714     /* Advertise NUMA via ibm,associativity */
715     ret = spapr_fixup_cpu_dt(fdt, spapr);
716     if (ret < 0) {
717         fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
718     }
719 
720     bootlist = get_boot_devices_list(&cb, true);
721     if (cb && bootlist) {
722         int offset = fdt_path_offset(fdt, "/chosen");
723         if (offset < 0) {
724             exit(1);
725         }
726         for (i = 0; i < cb; i++) {
727             if (bootlist[i] == '\n') {
728                 bootlist[i] = ' ';
729             }
730 
731         }
732         ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
733     }
734 
735     if (!spapr->has_graphics) {
736         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
737     }
738 
739     _FDT((fdt_pack(fdt)));
740 
741     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
742         hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
743                  fdt_totalsize(fdt), FDT_MAX_SIZE);
744         exit(1);
745     }
746 
747     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
748 
749     g_free(fdt);
750 }
751 
752 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
753 {
754     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
755 }
756 
757 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
758 {
759     CPUPPCState *env = &cpu->env;
760 
761     if (msr_pr) {
762         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
763         env->gpr[3] = H_PRIVILEGE;
764     } else {
765         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
766     }
767 }
768 
769 static void spapr_reset_htab(sPAPREnvironment *spapr)
770 {
771     long shift;
772 
773     /* allocate hash page table.  For now we always make this 16mb,
774      * later we should probably make it scale to the size of guest
775      * RAM */
776 
777     shift = kvmppc_reset_htab(spapr->htab_shift);
778 
779     if (shift > 0) {
780         /* Kernel handles htab, we don't need to allocate one */
781         spapr->htab_shift = shift;
782         kvmppc_kern_htab = true;
783     } else {
784         if (!spapr->htab) {
785             /* Allocate an htab if we don't yet have one */
786             spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
787         }
788 
789         /* And clear it */
790         memset(spapr->htab, 0, HTAB_SIZE(spapr));
791     }
792 
793     /* Update the RMA size if necessary */
794     if (spapr->vrma_adjust) {
795         hwaddr node0_size = (nb_numa_nodes > 1) ?
796             numa_info[0].node_mem : ram_size;
797         spapr->rma_size = kvmppc_rma_size(node0_size, spapr->htab_shift);
798     }
799 }
800 
801 static void ppc_spapr_reset(void)
802 {
803     PowerPCCPU *first_ppc_cpu;
804 
805     /* Reset the hash table & recalc the RMA */
806     spapr_reset_htab(spapr);
807 
808     qemu_devices_reset();
809 
810     /* Load the fdt */
811     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
812                        spapr->rtas_size);
813 
814     /* Set up the entry state */
815     first_ppc_cpu = POWERPC_CPU(first_cpu);
816     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
817     first_ppc_cpu->env.gpr[5] = 0;
818     first_cpu->halted = 0;
819     first_ppc_cpu->env.nip = spapr->entry_point;
820 
821 }
822 
823 static void spapr_cpu_reset(void *opaque)
824 {
825     PowerPCCPU *cpu = opaque;
826     CPUState *cs = CPU(cpu);
827     CPUPPCState *env = &cpu->env;
828 
829     cpu_reset(cs);
830 
831     /* All CPUs start halted.  CPU0 is unhalted from the machine level
832      * reset code and the rest are explicitly started up by the guest
833      * using an RTAS call */
834     cs->halted = 1;
835 
836     env->spr[SPR_HIOR] = 0;
837 
838     env->external_htab = (uint8_t *)spapr->htab;
839     if (kvm_enabled() && !env->external_htab) {
840         /*
841          * HV KVM, set external_htab to 1 so our ppc_hash64_load_hpte*
842          * functions do the right thing.
843          */
844         env->external_htab = (void *)1;
845     }
846     env->htab_base = -1;
847     /*
848      * htab_mask is the mask used to normalize hash value to PTEG index.
849      * htab_shift is log2 of hash table size.
850      * We have 8 hpte per group, and each hpte is 16 bytes.
851      * ie have 128 bytes per hpte entry.
852      */
853     env->htab_mask = (1ULL << ((spapr)->htab_shift - 7)) - 1;
854     env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
855         (spapr->htab_shift - 18);
856 }
857 
858 static void spapr_create_nvram(sPAPREnvironment *spapr)
859 {
860     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
861     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
862 
863     if (dinfo) {
864         qdev_prop_set_drive_nofail(dev, "drive", dinfo->bdrv);
865     }
866 
867     qdev_init_nofail(dev);
868 
869     spapr->nvram = (struct sPAPRNVRAM *)dev;
870 }
871 
872 /* Returns whether we want to use VGA or not */
873 static int spapr_vga_init(PCIBus *pci_bus)
874 {
875     switch (vga_interface_type) {
876     case VGA_NONE:
877         return false;
878     case VGA_DEVICE:
879         return true;
880     case VGA_STD:
881         return pci_vga_init(pci_bus) != NULL;
882     default:
883         fprintf(stderr, "This vga model is not supported,"
884                 "currently it only supports -vga std\n");
885         exit(0);
886     }
887 }
888 
889 static const VMStateDescription vmstate_spapr = {
890     .name = "spapr",
891     .version_id = 2,
892     .minimum_version_id = 1,
893     .fields = (VMStateField[]) {
894         VMSTATE_UNUSED(4), /* used to be @next_irq */
895 
896         /* RTC offset */
897         VMSTATE_UINT64(rtc_offset, sPAPREnvironment),
898         VMSTATE_PPC_TIMEBASE_V(tb, sPAPREnvironment, 2),
899         VMSTATE_END_OF_LIST()
900     },
901 };
902 
903 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
904 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
905 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
906 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
907 
908 static int htab_save_setup(QEMUFile *f, void *opaque)
909 {
910     sPAPREnvironment *spapr = opaque;
911 
912     /* "Iteration" header */
913     qemu_put_be32(f, spapr->htab_shift);
914 
915     if (spapr->htab) {
916         spapr->htab_save_index = 0;
917         spapr->htab_first_pass = true;
918     } else {
919         assert(kvm_enabled());
920 
921         spapr->htab_fd = kvmppc_get_htab_fd(false);
922         if (spapr->htab_fd < 0) {
923             fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
924                     strerror(errno));
925             return -1;
926         }
927     }
928 
929 
930     return 0;
931 }
932 
933 static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
934                                  int64_t max_ns)
935 {
936     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
937     int index = spapr->htab_save_index;
938     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
939 
940     assert(spapr->htab_first_pass);
941 
942     do {
943         int chunkstart;
944 
945         /* Consume invalid HPTEs */
946         while ((index < htabslots)
947                && !HPTE_VALID(HPTE(spapr->htab, index))) {
948             index++;
949             CLEAN_HPTE(HPTE(spapr->htab, index));
950         }
951 
952         /* Consume valid HPTEs */
953         chunkstart = index;
954         while ((index < htabslots)
955                && HPTE_VALID(HPTE(spapr->htab, index))) {
956             index++;
957             CLEAN_HPTE(HPTE(spapr->htab, index));
958         }
959 
960         if (index > chunkstart) {
961             int n_valid = index - chunkstart;
962 
963             qemu_put_be32(f, chunkstart);
964             qemu_put_be16(f, n_valid);
965             qemu_put_be16(f, 0);
966             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
967                             HASH_PTE_SIZE_64 * n_valid);
968 
969             if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
970                 break;
971             }
972         }
973     } while ((index < htabslots) && !qemu_file_rate_limit(f));
974 
975     if (index >= htabslots) {
976         assert(index == htabslots);
977         index = 0;
978         spapr->htab_first_pass = false;
979     }
980     spapr->htab_save_index = index;
981 }
982 
983 static int htab_save_later_pass(QEMUFile *f, sPAPREnvironment *spapr,
984                                 int64_t max_ns)
985 {
986     bool final = max_ns < 0;
987     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
988     int examined = 0, sent = 0;
989     int index = spapr->htab_save_index;
990     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
991 
992     assert(!spapr->htab_first_pass);
993 
994     do {
995         int chunkstart, invalidstart;
996 
997         /* Consume non-dirty HPTEs */
998         while ((index < htabslots)
999                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1000             index++;
1001             examined++;
1002         }
1003 
1004         chunkstart = index;
1005         /* Consume valid dirty HPTEs */
1006         while ((index < htabslots)
1007                && HPTE_DIRTY(HPTE(spapr->htab, index))
1008                && HPTE_VALID(HPTE(spapr->htab, index))) {
1009             CLEAN_HPTE(HPTE(spapr->htab, index));
1010             index++;
1011             examined++;
1012         }
1013 
1014         invalidstart = index;
1015         /* Consume invalid dirty HPTEs */
1016         while ((index < htabslots)
1017                && HPTE_DIRTY(HPTE(spapr->htab, index))
1018                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1019             CLEAN_HPTE(HPTE(spapr->htab, index));
1020             index++;
1021             examined++;
1022         }
1023 
1024         if (index > chunkstart) {
1025             int n_valid = invalidstart - chunkstart;
1026             int n_invalid = index - invalidstart;
1027 
1028             qemu_put_be32(f, chunkstart);
1029             qemu_put_be16(f, n_valid);
1030             qemu_put_be16(f, n_invalid);
1031             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1032                             HASH_PTE_SIZE_64 * n_valid);
1033             sent += index - chunkstart;
1034 
1035             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1036                 break;
1037             }
1038         }
1039 
1040         if (examined >= htabslots) {
1041             break;
1042         }
1043 
1044         if (index >= htabslots) {
1045             assert(index == htabslots);
1046             index = 0;
1047         }
1048     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1049 
1050     if (index >= htabslots) {
1051         assert(index == htabslots);
1052         index = 0;
1053     }
1054 
1055     spapr->htab_save_index = index;
1056 
1057     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1058 }
1059 
1060 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1061 #define MAX_KVM_BUF_SIZE    2048
1062 
1063 static int htab_save_iterate(QEMUFile *f, void *opaque)
1064 {
1065     sPAPREnvironment *spapr = opaque;
1066     int rc = 0;
1067 
1068     /* Iteration header */
1069     qemu_put_be32(f, 0);
1070 
1071     if (!spapr->htab) {
1072         assert(kvm_enabled());
1073 
1074         rc = kvmppc_save_htab(f, spapr->htab_fd,
1075                               MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1076         if (rc < 0) {
1077             return rc;
1078         }
1079     } else  if (spapr->htab_first_pass) {
1080         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1081     } else {
1082         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1083     }
1084 
1085     /* End marker */
1086     qemu_put_be32(f, 0);
1087     qemu_put_be16(f, 0);
1088     qemu_put_be16(f, 0);
1089 
1090     return rc;
1091 }
1092 
1093 static int htab_save_complete(QEMUFile *f, void *opaque)
1094 {
1095     sPAPREnvironment *spapr = opaque;
1096 
1097     /* Iteration header */
1098     qemu_put_be32(f, 0);
1099 
1100     if (!spapr->htab) {
1101         int rc;
1102 
1103         assert(kvm_enabled());
1104 
1105         rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
1106         if (rc < 0) {
1107             return rc;
1108         }
1109         close(spapr->htab_fd);
1110         spapr->htab_fd = -1;
1111     } else {
1112         htab_save_later_pass(f, spapr, -1);
1113     }
1114 
1115     /* End marker */
1116     qemu_put_be32(f, 0);
1117     qemu_put_be16(f, 0);
1118     qemu_put_be16(f, 0);
1119 
1120     return 0;
1121 }
1122 
1123 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1124 {
1125     sPAPREnvironment *spapr = opaque;
1126     uint32_t section_hdr;
1127     int fd = -1;
1128 
1129     if (version_id < 1 || version_id > 1) {
1130         fprintf(stderr, "htab_load() bad version\n");
1131         return -EINVAL;
1132     }
1133 
1134     section_hdr = qemu_get_be32(f);
1135 
1136     if (section_hdr) {
1137         /* First section, just the hash shift */
1138         if (spapr->htab_shift != section_hdr) {
1139             return -EINVAL;
1140         }
1141         return 0;
1142     }
1143 
1144     if (!spapr->htab) {
1145         assert(kvm_enabled());
1146 
1147         fd = kvmppc_get_htab_fd(true);
1148         if (fd < 0) {
1149             fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1150                     strerror(errno));
1151         }
1152     }
1153 
1154     while (true) {
1155         uint32_t index;
1156         uint16_t n_valid, n_invalid;
1157 
1158         index = qemu_get_be32(f);
1159         n_valid = qemu_get_be16(f);
1160         n_invalid = qemu_get_be16(f);
1161 
1162         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1163             /* End of Stream */
1164             break;
1165         }
1166 
1167         if ((index + n_valid + n_invalid) >
1168             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1169             /* Bad index in stream */
1170             fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1171                     "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1172                     spapr->htab_shift);
1173             return -EINVAL;
1174         }
1175 
1176         if (spapr->htab) {
1177             if (n_valid) {
1178                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1179                                 HASH_PTE_SIZE_64 * n_valid);
1180             }
1181             if (n_invalid) {
1182                 memset(HPTE(spapr->htab, index + n_valid), 0,
1183                        HASH_PTE_SIZE_64 * n_invalid);
1184             }
1185         } else {
1186             int rc;
1187 
1188             assert(fd >= 0);
1189 
1190             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1191             if (rc < 0) {
1192                 return rc;
1193             }
1194         }
1195     }
1196 
1197     if (!spapr->htab) {
1198         assert(fd >= 0);
1199         close(fd);
1200     }
1201 
1202     return 0;
1203 }
1204 
1205 static SaveVMHandlers savevm_htab_handlers = {
1206     .save_live_setup = htab_save_setup,
1207     .save_live_iterate = htab_save_iterate,
1208     .save_live_complete = htab_save_complete,
1209     .load_state = htab_load,
1210 };
1211 
1212 /* pSeries LPAR / sPAPR hardware init */
1213 static void ppc_spapr_init(MachineState *machine)
1214 {
1215     ram_addr_t ram_size = machine->ram_size;
1216     const char *cpu_model = machine->cpu_model;
1217     const char *kernel_filename = machine->kernel_filename;
1218     const char *kernel_cmdline = machine->kernel_cmdline;
1219     const char *initrd_filename = machine->initrd_filename;
1220     const char *boot_device = machine->boot_order;
1221     PowerPCCPU *cpu;
1222     CPUPPCState *env;
1223     PCIHostState *phb;
1224     int i;
1225     MemoryRegion *sysmem = get_system_memory();
1226     MemoryRegion *ram = g_new(MemoryRegion, 1);
1227     MemoryRegion *rma_region;
1228     void *rma = NULL;
1229     hwaddr rma_alloc_size;
1230     hwaddr node0_size = (nb_numa_nodes > 1) ? numa_info[0].node_mem : ram_size;
1231     uint32_t initrd_base = 0;
1232     long kernel_size = 0, initrd_size = 0;
1233     long load_limit, rtas_limit, fw_size;
1234     bool kernel_le = false;
1235     char *filename;
1236 
1237     msi_supported = true;
1238 
1239     spapr = g_malloc0(sizeof(*spapr));
1240     QLIST_INIT(&spapr->phbs);
1241 
1242     cpu_ppc_hypercall = emulate_spapr_hypercall;
1243 
1244     /* Allocate RMA if necessary */
1245     rma_alloc_size = kvmppc_alloc_rma(&rma);
1246 
1247     if (rma_alloc_size == -1) {
1248         hw_error("qemu: Unable to create RMA\n");
1249         exit(1);
1250     }
1251 
1252     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1253         spapr->rma_size = rma_alloc_size;
1254     } else {
1255         spapr->rma_size = node0_size;
1256 
1257         /* With KVM, we don't actually know whether KVM supports an
1258          * unbounded RMA (PR KVM) or is limited by the hash table size
1259          * (HV KVM using VRMA), so we always assume the latter
1260          *
1261          * In that case, we also limit the initial allocations for RTAS
1262          * etc... to 256M since we have no way to know what the VRMA size
1263          * is going to be as it depends on the size of the hash table
1264          * isn't determined yet.
1265          */
1266         if (kvm_enabled()) {
1267             spapr->vrma_adjust = 1;
1268             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1269         }
1270     }
1271 
1272     if (spapr->rma_size > node0_size) {
1273         fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1274                 spapr->rma_size);
1275         exit(1);
1276     }
1277 
1278     /* We place the device tree and RTAS just below either the top of the RMA,
1279      * or just below 2GB, whichever is lowere, so that it can be
1280      * processed with 32-bit real mode code if necessary */
1281     rtas_limit = MIN(spapr->rma_size, 0x80000000);
1282     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1283     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1284     load_limit = spapr->fdt_addr - FW_OVERHEAD;
1285 
1286     /* We aim for a hash table of size 1/128 the size of RAM.  The
1287      * normal rule of thumb is 1/64 the size of RAM, but that's much
1288      * more than needed for the Linux guests we support. */
1289     spapr->htab_shift = 18; /* Minimum architected size */
1290     while (spapr->htab_shift <= 46) {
1291         if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
1292             break;
1293         }
1294         spapr->htab_shift++;
1295     }
1296 
1297     /* Set up Interrupt Controller before we create the VCPUs */
1298     spapr->icp = xics_system_init(smp_cpus * kvmppc_smt_threads() / smp_threads,
1299                                   XICS_IRQS);
1300 
1301     /* init CPUs */
1302     if (cpu_model == NULL) {
1303         cpu_model = kvm_enabled() ? "host" : "POWER7";
1304     }
1305     for (i = 0; i < smp_cpus; i++) {
1306         cpu = cpu_ppc_init(cpu_model);
1307         if (cpu == NULL) {
1308             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1309             exit(1);
1310         }
1311         env = &cpu->env;
1312 
1313         /* Set time-base frequency to 512 MHz */
1314         cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1315 
1316         /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1317          * MSR[IP] should never be set.
1318          */
1319         env->msr_mask &= ~(1 << 6);
1320 
1321         /* Tell KVM that we're in PAPR mode */
1322         if (kvm_enabled()) {
1323             kvmppc_set_papr(cpu);
1324         }
1325 
1326         if (cpu->max_compat) {
1327             if (ppc_set_compat(cpu, cpu->max_compat) < 0) {
1328                 exit(1);
1329             }
1330         }
1331 
1332         xics_cpu_setup(spapr->icp, cpu);
1333 
1334         qemu_register_reset(spapr_cpu_reset, cpu);
1335     }
1336 
1337     /* allocate RAM */
1338     spapr->ram_limit = ram_size;
1339     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1340                                          spapr->ram_limit);
1341     memory_region_add_subregion(sysmem, 0, ram);
1342 
1343     if (rma_alloc_size && rma) {
1344         rma_region = g_new(MemoryRegion, 1);
1345         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1346                                    rma_alloc_size, rma);
1347         vmstate_register_ram_global(rma_region);
1348         memory_region_add_subregion(sysmem, 0, rma_region);
1349     }
1350 
1351     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1352     spapr->rtas_size = load_image_targphys(filename, spapr->rtas_addr,
1353                                            rtas_limit - spapr->rtas_addr);
1354     if (spapr->rtas_size < 0) {
1355         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1356         exit(1);
1357     }
1358     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1359         hw_error("RTAS too big ! 0x%lx bytes (max is 0x%x)\n",
1360                  spapr->rtas_size, RTAS_MAX_SIZE);
1361         exit(1);
1362     }
1363     g_free(filename);
1364 
1365     /* Set up EPOW events infrastructure */
1366     spapr_events_init(spapr);
1367 
1368     /* Set up VIO bus */
1369     spapr->vio_bus = spapr_vio_bus_init();
1370 
1371     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1372         if (serial_hds[i]) {
1373             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1374         }
1375     }
1376 
1377     /* We always have at least the nvram device on VIO */
1378     spapr_create_nvram(spapr);
1379 
1380     /* Set up PCI */
1381     spapr_pci_msi_init(spapr, SPAPR_PCI_MSI_WINDOW);
1382     spapr_pci_rtas_init();
1383 
1384     phb = spapr_create_phb(spapr, 0);
1385 
1386     for (i = 0; i < nb_nics; i++) {
1387         NICInfo *nd = &nd_table[i];
1388 
1389         if (!nd->model) {
1390             nd->model = g_strdup("ibmveth");
1391         }
1392 
1393         if (strcmp(nd->model, "ibmveth") == 0) {
1394             spapr_vlan_create(spapr->vio_bus, nd);
1395         } else {
1396             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1397         }
1398     }
1399 
1400     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1401         spapr_vscsi_create(spapr->vio_bus);
1402     }
1403 
1404     /* Graphics */
1405     if (spapr_vga_init(phb->bus)) {
1406         spapr->has_graphics = true;
1407     }
1408 
1409     if (usb_enabled(spapr->has_graphics)) {
1410         pci_create_simple(phb->bus, -1, "pci-ohci");
1411         if (spapr->has_graphics) {
1412             usbdevice_create("keyboard");
1413             usbdevice_create("mouse");
1414         }
1415     }
1416 
1417     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1418         fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1419                 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1420         exit(1);
1421     }
1422 
1423     if (kernel_filename) {
1424         uint64_t lowaddr = 0;
1425 
1426         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1427                                NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1428         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1429             kernel_size = load_elf(kernel_filename,
1430                                    translate_kernel_address, NULL,
1431                                    NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1432             kernel_le = kernel_size > 0;
1433         }
1434         if (kernel_size < 0) {
1435             fprintf(stderr, "qemu: error loading %s: %s\n",
1436                     kernel_filename, load_elf_strerror(kernel_size));
1437             exit(1);
1438         }
1439 
1440         /* load initrd */
1441         if (initrd_filename) {
1442             /* Try to locate the initrd in the gap between the kernel
1443              * and the firmware. Add a bit of space just in case
1444              */
1445             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1446             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1447                                               load_limit - initrd_base);
1448             if (initrd_size < 0) {
1449                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1450                         initrd_filename);
1451                 exit(1);
1452             }
1453         } else {
1454             initrd_base = 0;
1455             initrd_size = 0;
1456         }
1457     }
1458 
1459     if (bios_name == NULL) {
1460         bios_name = FW_FILE_NAME;
1461     }
1462     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1463     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1464     if (fw_size < 0) {
1465         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1466         exit(1);
1467     }
1468     g_free(filename);
1469 
1470     spapr->entry_point = 0x100;
1471 
1472     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1473     register_savevm_live(NULL, "spapr/htab", -1, 1,
1474                          &savevm_htab_handlers, spapr);
1475 
1476     /* Prepare the device tree */
1477     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1478                                             kernel_size, kernel_le,
1479                                             boot_device, kernel_cmdline,
1480                                             spapr->epow_irq);
1481     assert(spapr->fdt_skel != NULL);
1482 }
1483 
1484 static int spapr_kvm_type(const char *vm_type)
1485 {
1486     if (!vm_type) {
1487         return 0;
1488     }
1489 
1490     if (!strcmp(vm_type, "HV")) {
1491         return 1;
1492     }
1493 
1494     if (!strcmp(vm_type, "PR")) {
1495         return 2;
1496     }
1497 
1498     error_report("Unknown kvm-type specified '%s'", vm_type);
1499     exit(1);
1500 }
1501 
1502 /*
1503  * Implementation of an interface to adjust firmware patch
1504  * for the bootindex property handling.
1505  */
1506 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
1507                                    DeviceState *dev)
1508 {
1509 #define CAST(type, obj, name) \
1510     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
1511     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
1512     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
1513 
1514     if (d) {
1515         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
1516         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
1517         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
1518 
1519         if (spapr) {
1520             /*
1521              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
1522              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
1523              * in the top 16 bits of the 64-bit LUN
1524              */
1525             unsigned id = 0x8000 | (d->id << 8) | d->lun;
1526             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1527                                    (uint64_t)id << 48);
1528         } else if (virtio) {
1529             /*
1530              * We use SRP luns of the form 01000000 | (target << 8) | lun
1531              * in the top 32 bits of the 64-bit LUN
1532              * Note: the quote above is from SLOF and it is wrong,
1533              * the actual binding is:
1534              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
1535              */
1536             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
1537             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1538                                    (uint64_t)id << 32);
1539         } else if (usb) {
1540             /*
1541              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
1542              * in the top 32 bits of the 64-bit LUN
1543              */
1544             unsigned usb_port = atoi(usb->port->path);
1545             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
1546             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1547                                    (uint64_t)id << 32);
1548         }
1549     }
1550 
1551     if (phb) {
1552         /* Replace "pci" with "pci@800000020000000" */
1553         return g_strdup_printf("pci@%"PRIX64, phb->buid);
1554     }
1555 
1556     return NULL;
1557 }
1558 
1559 static char *spapr_get_kvm_type(Object *obj, Error **errp)
1560 {
1561     sPAPRMachineState *sm = SPAPR_MACHINE(obj);
1562 
1563     return g_strdup(sm->kvm_type);
1564 }
1565 
1566 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
1567 {
1568     sPAPRMachineState *sm = SPAPR_MACHINE(obj);
1569 
1570     g_free(sm->kvm_type);
1571     sm->kvm_type = g_strdup(value);
1572 }
1573 
1574 static void spapr_machine_initfn(Object *obj)
1575 {
1576     object_property_add_str(obj, "kvm-type",
1577                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
1578 }
1579 
1580 static void ppc_cpu_do_nmi_on_cpu(void *arg)
1581 {
1582     CPUState *cs = arg;
1583 
1584     cpu_synchronize_state(cs);
1585     ppc_cpu_do_system_reset(cs);
1586 }
1587 
1588 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
1589 {
1590     CPUState *cs;
1591 
1592     CPU_FOREACH(cs) {
1593         async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, cs);
1594     }
1595 }
1596 
1597 static void spapr_machine_class_init(ObjectClass *oc, void *data)
1598 {
1599     MachineClass *mc = MACHINE_CLASS(oc);
1600     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
1601     NMIClass *nc = NMI_CLASS(oc);
1602 
1603     mc->name = "pseries";
1604     mc->desc = "pSeries Logical Partition (PAPR compliant)";
1605     mc->is_default = 1;
1606     mc->init = ppc_spapr_init;
1607     mc->reset = ppc_spapr_reset;
1608     mc->block_default_type = IF_SCSI;
1609     mc->max_cpus = MAX_CPUS;
1610     mc->no_parallel = 1;
1611     mc->default_boot_order = NULL;
1612     mc->kvm_type = spapr_kvm_type;
1613 
1614     fwc->get_dev_path = spapr_get_fw_dev_path;
1615     nc->nmi_monitor_handler = spapr_nmi;
1616 }
1617 
1618 static const TypeInfo spapr_machine_info = {
1619     .name          = TYPE_SPAPR_MACHINE,
1620     .parent        = TYPE_MACHINE,
1621     .instance_size = sizeof(sPAPRMachineState),
1622     .instance_init = spapr_machine_initfn,
1623     .class_init    = spapr_machine_class_init,
1624     .interfaces = (InterfaceInfo[]) {
1625         { TYPE_FW_PATH_PROVIDER },
1626         { TYPE_NMI },
1627         { }
1628     },
1629 };
1630 
1631 static void spapr_machine_2_1_class_init(ObjectClass *oc, void *data)
1632 {
1633     MachineClass *mc = MACHINE_CLASS(oc);
1634 
1635     mc->name = "pseries-2.1";
1636     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.1";
1637     mc->is_default = 0;
1638 }
1639 
1640 static const TypeInfo spapr_machine_2_1_info = {
1641     .name          = TYPE_SPAPR_MACHINE "2.1",
1642     .parent        = TYPE_SPAPR_MACHINE,
1643     .class_init    = spapr_machine_2_1_class_init,
1644 };
1645 
1646 static void spapr_machine_register_types(void)
1647 {
1648     type_register_static(&spapr_machine_info);
1649     type_register_static(&spapr_machine_2_1_info);
1650 }
1651 
1652 type_init(spapr_machine_register_types)
1653