xref: /openbmc/qemu/hw/ppc/spapr.c (revision bfb27e60)
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 RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
75 #define FW_MAX_SIZE             0x400000
76 #define FW_FILE_NAME            "slof.bin"
77 #define FW_OVERHEAD             0x2800000
78 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
79 
80 #define MIN_RMA_SLOF            128UL
81 
82 #define TIMEBASE_FREQ           512000000ULL
83 
84 #define MAX_CPUS                255
85 
86 #define PHANDLE_XICP            0x00001111
87 
88 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
89 
90 typedef struct sPAPRMachineState sPAPRMachineState;
91 
92 #define TYPE_SPAPR_MACHINE      "spapr-machine"
93 #define SPAPR_MACHINE(obj) \
94     OBJECT_CHECK(sPAPRMachineState, (obj), TYPE_SPAPR_MACHINE)
95 
96 /**
97  * sPAPRMachineState:
98  */
99 struct sPAPRMachineState {
100     /*< private >*/
101     MachineState parent_obj;
102 
103     /*< public >*/
104     char *kvm_type;
105 };
106 
107 sPAPREnvironment *spapr;
108 
109 static XICSState *try_create_xics(const char *type, int nr_servers,
110                                   int nr_irqs)
111 {
112     DeviceState *dev;
113 
114     dev = qdev_create(NULL, type);
115     qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
116     qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
117     if (qdev_init(dev) < 0) {
118         return NULL;
119     }
120 
121     return XICS_COMMON(dev);
122 }
123 
124 static XICSState *xics_system_init(int nr_servers, int nr_irqs)
125 {
126     XICSState *icp = NULL;
127 
128     if (kvm_enabled()) {
129         QemuOpts *machine_opts = qemu_get_machine_opts();
130         bool irqchip_allowed = qemu_opt_get_bool(machine_opts,
131                                                 "kernel_irqchip", true);
132         bool irqchip_required = qemu_opt_get_bool(machine_opts,
133                                                   "kernel_irqchip", false);
134         if (irqchip_allowed) {
135             icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs);
136         }
137 
138         if (irqchip_required && !icp) {
139             perror("Failed to create in-kernel XICS\n");
140             abort();
141         }
142     }
143 
144     if (!icp) {
145         icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs);
146     }
147 
148     if (!icp) {
149         perror("Failed to create XICS\n");
150         abort();
151     }
152 
153     return icp;
154 }
155 
156 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
157                                   int smt_threads)
158 {
159     int i, ret = 0;
160     uint32_t servers_prop[smt_threads];
161     uint32_t gservers_prop[smt_threads * 2];
162     int index = ppc_get_vcpu_dt_id(cpu);
163 
164     if (cpu->cpu_version) {
165         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
166         if (ret < 0) {
167             return ret;
168         }
169     }
170 
171     /* Build interrupt servers and gservers properties */
172     for (i = 0; i < smt_threads; i++) {
173         servers_prop[i] = cpu_to_be32(index + i);
174         /* Hack, direct the group queues back to cpu 0 */
175         gservers_prop[i*2] = cpu_to_be32(index + i);
176         gservers_prop[i*2 + 1] = 0;
177     }
178     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
179                       servers_prop, sizeof(servers_prop));
180     if (ret < 0) {
181         return ret;
182     }
183     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
184                       gservers_prop, sizeof(gservers_prop));
185 
186     return ret;
187 }
188 
189 static int spapr_fixup_cpu_dt(void *fdt, sPAPREnvironment *spapr)
190 {
191     int ret = 0, offset, cpus_offset;
192     CPUState *cs;
193     char cpu_model[32];
194     int smt = kvmppc_smt_threads();
195     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
196 
197     CPU_FOREACH(cs) {
198         PowerPCCPU *cpu = POWERPC_CPU(cs);
199         DeviceClass *dc = DEVICE_GET_CLASS(cs);
200         int index = ppc_get_vcpu_dt_id(cpu);
201         uint32_t associativity[] = {cpu_to_be32(0x5),
202                                     cpu_to_be32(0x0),
203                                     cpu_to_be32(0x0),
204                                     cpu_to_be32(0x0),
205                                     cpu_to_be32(cs->numa_node),
206                                     cpu_to_be32(index)};
207 
208         if ((index % smt) != 0) {
209             continue;
210         }
211 
212         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
213 
214         cpus_offset = fdt_path_offset(fdt, "/cpus");
215         if (cpus_offset < 0) {
216             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
217                                           "cpus");
218             if (cpus_offset < 0) {
219                 return cpus_offset;
220             }
221         }
222         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
223         if (offset < 0) {
224             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
225             if (offset < 0) {
226                 return offset;
227             }
228         }
229 
230         if (nb_numa_nodes > 1) {
231             ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
232                               sizeof(associativity));
233             if (ret < 0) {
234                 return ret;
235             }
236         }
237 
238         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
239                           pft_size_prop, sizeof(pft_size_prop));
240         if (ret < 0) {
241             return ret;
242         }
243 
244         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
245                                      ppc_get_compat_smt_threads(cpu));
246         if (ret < 0) {
247             return ret;
248         }
249     }
250     return ret;
251 }
252 
253 
254 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
255                                      size_t maxsize)
256 {
257     size_t maxcells = maxsize / sizeof(uint32_t);
258     int i, j, count;
259     uint32_t *p = prop;
260 
261     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
262         struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
263 
264         if (!sps->page_shift) {
265             break;
266         }
267         for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
268             if (sps->enc[count].page_shift == 0) {
269                 break;
270             }
271         }
272         if ((p - prop) >= (maxcells - 3 - count * 2)) {
273             break;
274         }
275         *(p++) = cpu_to_be32(sps->page_shift);
276         *(p++) = cpu_to_be32(sps->slb_enc);
277         *(p++) = cpu_to_be32(count);
278         for (j = 0; j < count; j++) {
279             *(p++) = cpu_to_be32(sps->enc[j].page_shift);
280             *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
281         }
282     }
283 
284     return (p - prop) * sizeof(uint32_t);
285 }
286 
287 static hwaddr spapr_node0_size(void)
288 {
289     if (nb_numa_nodes) {
290         int i;
291         for (i = 0; i < nb_numa_nodes; ++i) {
292             if (numa_info[i].node_mem) {
293                 return MIN(pow2floor(numa_info[i].node_mem), ram_size);
294             }
295         }
296     }
297     return ram_size;
298 }
299 
300 #define _FDT(exp) \
301     do { \
302         int ret = (exp);                                           \
303         if (ret < 0) {                                             \
304             fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
305                     #exp, fdt_strerror(ret));                      \
306             exit(1);                                               \
307         }                                                          \
308     } while (0)
309 
310 static void add_str(GString *s, const gchar *s1)
311 {
312     g_string_append_len(s, s1, strlen(s1) + 1);
313 }
314 
315 static void *spapr_create_fdt_skel(hwaddr initrd_base,
316                                    hwaddr initrd_size,
317                                    hwaddr kernel_size,
318                                    bool little_endian,
319                                    const char *boot_device,
320                                    const char *kernel_cmdline,
321                                    uint32_t epow_irq)
322 {
323     void *fdt;
324     CPUState *cs;
325     uint32_t start_prop = cpu_to_be32(initrd_base);
326     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
327     GString *hypertas = g_string_sized_new(256);
328     GString *qemu_hypertas = g_string_sized_new(256);
329     uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
330     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(smp_cpus)};
331     int smt = kvmppc_smt_threads();
332     unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
333     QemuOpts *opts = qemu_opts_find(qemu_find_opts("smp-opts"), NULL);
334     unsigned sockets = opts ? qemu_opt_get_number(opts, "sockets", 0) : 0;
335     uint32_t cpus_per_socket = sockets ? (smp_cpus / sockets) : 1;
336     char *buf;
337 
338     add_str(hypertas, "hcall-pft");
339     add_str(hypertas, "hcall-term");
340     add_str(hypertas, "hcall-dabr");
341     add_str(hypertas, "hcall-interrupt");
342     add_str(hypertas, "hcall-tce");
343     add_str(hypertas, "hcall-vio");
344     add_str(hypertas, "hcall-splpar");
345     add_str(hypertas, "hcall-bulk");
346     add_str(hypertas, "hcall-set-mode");
347     add_str(qemu_hypertas, "hcall-memop1");
348 
349     fdt = g_malloc0(FDT_MAX_SIZE);
350     _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
351 
352     if (kernel_size) {
353         _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
354     }
355     if (initrd_size) {
356         _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
357     }
358     _FDT((fdt_finish_reservemap(fdt)));
359 
360     /* Root node */
361     _FDT((fdt_begin_node(fdt, "")));
362     _FDT((fdt_property_string(fdt, "device_type", "chrp")));
363     _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
364     _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
365 
366     /*
367      * Add info to guest to indentify which host is it being run on
368      * and what is the uuid of the guest
369      */
370     if (kvmppc_get_host_model(&buf)) {
371         _FDT((fdt_property_string(fdt, "host-model", buf)));
372         g_free(buf);
373     }
374     if (kvmppc_get_host_serial(&buf)) {
375         _FDT((fdt_property_string(fdt, "host-serial", buf)));
376         g_free(buf);
377     }
378 
379     buf = g_strdup_printf(UUID_FMT, qemu_uuid[0], qemu_uuid[1],
380                           qemu_uuid[2], qemu_uuid[3], qemu_uuid[4],
381                           qemu_uuid[5], qemu_uuid[6], qemu_uuid[7],
382                           qemu_uuid[8], qemu_uuid[9], qemu_uuid[10],
383                           qemu_uuid[11], qemu_uuid[12], qemu_uuid[13],
384                           qemu_uuid[14], qemu_uuid[15]);
385 
386     _FDT((fdt_property_string(fdt, "vm,uuid", buf)));
387     g_free(buf);
388 
389     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
390     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
391 
392     /* /chosen */
393     _FDT((fdt_begin_node(fdt, "chosen")));
394 
395     /* Set Form1_affinity */
396     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
397 
398     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
399     _FDT((fdt_property(fdt, "linux,initrd-start",
400                        &start_prop, sizeof(start_prop))));
401     _FDT((fdt_property(fdt, "linux,initrd-end",
402                        &end_prop, sizeof(end_prop))));
403     if (kernel_size) {
404         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
405                               cpu_to_be64(kernel_size) };
406 
407         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
408         if (little_endian) {
409             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
410         }
411     }
412     if (boot_device) {
413         _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
414     }
415     if (boot_menu) {
416         _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
417     }
418     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
419     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
420     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
421 
422     _FDT((fdt_end_node(fdt)));
423 
424     /* cpus */
425     _FDT((fdt_begin_node(fdt, "cpus")));
426 
427     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
428     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
429 
430     CPU_FOREACH(cs) {
431         PowerPCCPU *cpu = POWERPC_CPU(cs);
432         CPUPPCState *env = &cpu->env;
433         DeviceClass *dc = DEVICE_GET_CLASS(cs);
434         PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
435         int index = ppc_get_vcpu_dt_id(cpu);
436         char *nodename;
437         uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
438                            0xffffffff, 0xffffffff};
439         uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
440         uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
441         uint32_t page_sizes_prop[64];
442         size_t page_sizes_prop_size;
443 
444         if ((index % smt) != 0) {
445             continue;
446         }
447 
448         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
449 
450         _FDT((fdt_begin_node(fdt, nodename)));
451 
452         g_free(nodename);
453 
454         _FDT((fdt_property_cell(fdt, "reg", index)));
455         _FDT((fdt_property_string(fdt, "device_type", "cpu")));
456 
457         _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
458         _FDT((fdt_property_cell(fdt, "d-cache-block-size",
459                                 env->dcache_line_size)));
460         _FDT((fdt_property_cell(fdt, "d-cache-line-size",
461                                 env->dcache_line_size)));
462         _FDT((fdt_property_cell(fdt, "i-cache-block-size",
463                                 env->icache_line_size)));
464         _FDT((fdt_property_cell(fdt, "i-cache-line-size",
465                                 env->icache_line_size)));
466 
467         if (pcc->l1_dcache_size) {
468             _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
469         } else {
470             fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
471         }
472         if (pcc->l1_icache_size) {
473             _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
474         } else {
475             fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
476         }
477 
478         _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
479         _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
480         _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
481         _FDT((fdt_property_string(fdt, "status", "okay")));
482         _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
483 
484         if (env->spr_cb[SPR_PURR].oea_read) {
485             _FDT((fdt_property(fdt, "ibm,purr", NULL, 0)));
486         }
487 
488         if (env->mmu_model & POWERPC_MMU_1TSEG) {
489             _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
490                                segs, sizeof(segs))));
491         }
492 
493         /* Advertise VMX/VSX (vector extensions) if available
494          *   0 / no property == no vector extensions
495          *   1               == VMX / Altivec available
496          *   2               == VSX available */
497         if (env->insns_flags & PPC_ALTIVEC) {
498             uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
499 
500             _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
501         }
502 
503         /* Advertise DFP (Decimal Floating Point) if available
504          *   0 / no property == no DFP
505          *   1               == DFP available */
506         if (env->insns_flags2 & PPC2_DFP) {
507             _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
508         }
509 
510         page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
511                                                       sizeof(page_sizes_prop));
512         if (page_sizes_prop_size) {
513             _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
514                                page_sizes_prop, page_sizes_prop_size)));
515         }
516 
517         _FDT((fdt_property_cell(fdt, "ibm,chip-id",
518                                 cs->cpu_index / cpus_per_socket)));
519 
520         _FDT((fdt_end_node(fdt)));
521     }
522 
523     _FDT((fdt_end_node(fdt)));
524 
525     /* RTAS */
526     _FDT((fdt_begin_node(fdt, "rtas")));
527 
528     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
529         add_str(hypertas, "hcall-multi-tce");
530     }
531     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
532                        hypertas->len)));
533     g_string_free(hypertas, TRUE);
534     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
535                        qemu_hypertas->len)));
536     g_string_free(qemu_hypertas, TRUE);
537 
538     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
539         refpoints, sizeof(refpoints))));
540 
541     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
542 
543     /*
544      * According to PAPR, rtas ibm,os-term, does not gaurantee a return
545      * back to the guest cpu.
546      *
547      * While an additional ibm,extended-os-term property indicates that
548      * rtas call return will always occur. Set this property.
549      */
550     _FDT((fdt_property(fdt, "ibm,extended-os-term", NULL, 0)));
551 
552     _FDT((fdt_end_node(fdt)));
553 
554     /* interrupt controller */
555     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
556 
557     _FDT((fdt_property_string(fdt, "device_type",
558                               "PowerPC-External-Interrupt-Presentation")));
559     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
560     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
561     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
562                        interrupt_server_ranges_prop,
563                        sizeof(interrupt_server_ranges_prop))));
564     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
565     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
566     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
567 
568     _FDT((fdt_end_node(fdt)));
569 
570     /* vdevice */
571     _FDT((fdt_begin_node(fdt, "vdevice")));
572 
573     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
574     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
575     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
576     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
577     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
578     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
579 
580     _FDT((fdt_end_node(fdt)));
581 
582     /* event-sources */
583     spapr_events_fdt_skel(fdt, epow_irq);
584 
585     /* /hypervisor node */
586     if (kvm_enabled()) {
587         uint8_t hypercall[16];
588 
589         /* indicate KVM hypercall interface */
590         _FDT((fdt_begin_node(fdt, "hypervisor")));
591         _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
592         if (kvmppc_has_cap_fixup_hcalls()) {
593             /*
594              * Older KVM versions with older guest kernels were broken with the
595              * magic page, don't allow the guest to map it.
596              */
597             kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
598                                  sizeof(hypercall));
599             _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
600                               sizeof(hypercall))));
601         }
602         _FDT((fdt_end_node(fdt)));
603     }
604 
605     _FDT((fdt_end_node(fdt))); /* close root node */
606     _FDT((fdt_finish(fdt)));
607 
608     return fdt;
609 }
610 
611 int spapr_h_cas_compose_response(target_ulong addr, target_ulong size)
612 {
613     void *fdt, *fdt_skel;
614     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
615 
616     size -= sizeof(hdr);
617 
618     /* Create sceleton */
619     fdt_skel = g_malloc0(size);
620     _FDT((fdt_create(fdt_skel, size)));
621     _FDT((fdt_begin_node(fdt_skel, "")));
622     _FDT((fdt_end_node(fdt_skel)));
623     _FDT((fdt_finish(fdt_skel)));
624     fdt = g_malloc0(size);
625     _FDT((fdt_open_into(fdt_skel, fdt, size)));
626     g_free(fdt_skel);
627 
628     /* Fix skeleton up */
629     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
630 
631     /* Pack resulting tree */
632     _FDT((fdt_pack(fdt)));
633 
634     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
635         trace_spapr_cas_failed(size);
636         return -1;
637     }
638 
639     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
640     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
641     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
642     g_free(fdt);
643 
644     return 0;
645 }
646 
647 static void spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
648                                        hwaddr size)
649 {
650     uint32_t associativity[] = {
651         cpu_to_be32(0x4), /* length */
652         cpu_to_be32(0x0), cpu_to_be32(0x0),
653         cpu_to_be32(0x0), cpu_to_be32(nodeid)
654     };
655     char mem_name[32];
656     uint64_t mem_reg_property[2];
657     int off;
658 
659     mem_reg_property[0] = cpu_to_be64(start);
660     mem_reg_property[1] = cpu_to_be64(size);
661 
662     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
663     off = fdt_add_subnode(fdt, 0, mem_name);
664     _FDT(off);
665     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
666     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
667                       sizeof(mem_reg_property))));
668     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
669                       sizeof(associativity))));
670 }
671 
672 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
673 {
674     hwaddr mem_start, node_size;
675     int i, nb_nodes = nb_numa_nodes;
676     NodeInfo *nodes = numa_info;
677     NodeInfo ramnode;
678 
679     /* No NUMA nodes, assume there is just one node with whole RAM */
680     if (!nb_numa_nodes) {
681         nb_nodes = 1;
682         ramnode.node_mem = ram_size;
683         nodes = &ramnode;
684     }
685 
686     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
687         if (!nodes[i].node_mem) {
688             continue;
689         }
690         if (mem_start >= ram_size) {
691             node_size = 0;
692         } else {
693             node_size = nodes[i].node_mem;
694             if (node_size > ram_size - mem_start) {
695                 node_size = ram_size - mem_start;
696             }
697         }
698         if (!mem_start) {
699             /* ppc_spapr_init() checks for rma_size <= node0_size already */
700             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
701             mem_start += spapr->rma_size;
702             node_size -= spapr->rma_size;
703         }
704         for ( ; node_size; ) {
705             hwaddr sizetmp = pow2floor(node_size);
706 
707             /* mem_start != 0 here */
708             if (ctzl(mem_start) < ctzl(sizetmp)) {
709                 sizetmp = 1ULL << ctzl(mem_start);
710             }
711 
712             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
713             node_size -= sizetmp;
714             mem_start += sizetmp;
715         }
716     }
717 
718     return 0;
719 }
720 
721 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
722                                hwaddr fdt_addr,
723                                hwaddr rtas_addr,
724                                hwaddr rtas_size)
725 {
726     int ret, i;
727     size_t cb = 0;
728     char *bootlist;
729     void *fdt;
730     sPAPRPHBState *phb;
731 
732     fdt = g_malloc(FDT_MAX_SIZE);
733 
734     /* open out the base tree into a temp buffer for the final tweaks */
735     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
736 
737     ret = spapr_populate_memory(spapr, fdt);
738     if (ret < 0) {
739         fprintf(stderr, "couldn't setup memory nodes in fdt\n");
740         exit(1);
741     }
742 
743     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
744     if (ret < 0) {
745         fprintf(stderr, "couldn't setup vio devices in fdt\n");
746         exit(1);
747     }
748 
749     QLIST_FOREACH(phb, &spapr->phbs, list) {
750         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
751     }
752 
753     if (ret < 0) {
754         fprintf(stderr, "couldn't setup PCI devices in fdt\n");
755         exit(1);
756     }
757 
758     /* RTAS */
759     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
760     if (ret < 0) {
761         fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
762     }
763 
764     /* Advertise NUMA via ibm,associativity */
765     ret = spapr_fixup_cpu_dt(fdt, spapr);
766     if (ret < 0) {
767         fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
768     }
769 
770     bootlist = get_boot_devices_list(&cb, true);
771     if (cb && bootlist) {
772         int offset = fdt_path_offset(fdt, "/chosen");
773         if (offset < 0) {
774             exit(1);
775         }
776         for (i = 0; i < cb; i++) {
777             if (bootlist[i] == '\n') {
778                 bootlist[i] = ' ';
779             }
780 
781         }
782         ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
783     }
784 
785     if (!spapr->has_graphics) {
786         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
787     }
788 
789     _FDT((fdt_pack(fdt)));
790 
791     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
792         hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
793                  fdt_totalsize(fdt), FDT_MAX_SIZE);
794         exit(1);
795     }
796 
797     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
798 
799     g_free(bootlist);
800     g_free(fdt);
801 }
802 
803 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
804 {
805     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
806 }
807 
808 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
809 {
810     CPUPPCState *env = &cpu->env;
811 
812     if (msr_pr) {
813         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
814         env->gpr[3] = H_PRIVILEGE;
815     } else {
816         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
817     }
818 }
819 
820 static void spapr_reset_htab(sPAPREnvironment *spapr)
821 {
822     long shift;
823 
824     /* allocate hash page table.  For now we always make this 16mb,
825      * later we should probably make it scale to the size of guest
826      * RAM */
827 
828     shift = kvmppc_reset_htab(spapr->htab_shift);
829 
830     if (shift > 0) {
831         /* Kernel handles htab, we don't need to allocate one */
832         spapr->htab_shift = shift;
833         kvmppc_kern_htab = true;
834     } else {
835         if (!spapr->htab) {
836             /* Allocate an htab if we don't yet have one */
837             spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
838         }
839 
840         /* And clear it */
841         memset(spapr->htab, 0, HTAB_SIZE(spapr));
842     }
843 
844     /* Update the RMA size if necessary */
845     if (spapr->vrma_adjust) {
846         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
847                                           spapr->htab_shift);
848     }
849 }
850 
851 static void ppc_spapr_reset(void)
852 {
853     PowerPCCPU *first_ppc_cpu;
854     uint32_t rtas_limit;
855 
856     /* Reset the hash table & recalc the RMA */
857     spapr_reset_htab(spapr);
858 
859     qemu_devices_reset();
860 
861     /*
862      * We place the device tree and RTAS just below either the top of the RMA,
863      * or just below 2GB, whichever is lowere, so that it can be
864      * processed with 32-bit real mode code if necessary
865      */
866     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
867     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
868     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
869 
870     /* Load the fdt */
871     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
872                        spapr->rtas_size);
873 
874     /* Copy RTAS over */
875     cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
876                               spapr->rtas_size);
877 
878     /* Set up the entry state */
879     first_ppc_cpu = POWERPC_CPU(first_cpu);
880     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
881     first_ppc_cpu->env.gpr[5] = 0;
882     first_cpu->halted = 0;
883     first_ppc_cpu->env.nip = spapr->entry_point;
884 
885 }
886 
887 static void spapr_cpu_reset(void *opaque)
888 {
889     PowerPCCPU *cpu = opaque;
890     CPUState *cs = CPU(cpu);
891     CPUPPCState *env = &cpu->env;
892 
893     cpu_reset(cs);
894 
895     /* All CPUs start halted.  CPU0 is unhalted from the machine level
896      * reset code and the rest are explicitly started up by the guest
897      * using an RTAS call */
898     cs->halted = 1;
899 
900     env->spr[SPR_HIOR] = 0;
901 
902     env->external_htab = (uint8_t *)spapr->htab;
903     if (kvm_enabled() && !env->external_htab) {
904         /*
905          * HV KVM, set external_htab to 1 so our ppc_hash64_load_hpte*
906          * functions do the right thing.
907          */
908         env->external_htab = (void *)1;
909     }
910     env->htab_base = -1;
911     /*
912      * htab_mask is the mask used to normalize hash value to PTEG index.
913      * htab_shift is log2 of hash table size.
914      * We have 8 hpte per group, and each hpte is 16 bytes.
915      * ie have 128 bytes per hpte entry.
916      */
917     env->htab_mask = (1ULL << ((spapr)->htab_shift - 7)) - 1;
918     env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
919         (spapr->htab_shift - 18);
920 }
921 
922 static void spapr_create_nvram(sPAPREnvironment *spapr)
923 {
924     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
925     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
926 
927     if (dinfo) {
928         qdev_prop_set_drive_nofail(dev, "drive", dinfo->bdrv);
929     }
930 
931     qdev_init_nofail(dev);
932 
933     spapr->nvram = (struct sPAPRNVRAM *)dev;
934 }
935 
936 /* Returns whether we want to use VGA or not */
937 static int spapr_vga_init(PCIBus *pci_bus)
938 {
939     switch (vga_interface_type) {
940     case VGA_NONE:
941         return false;
942     case VGA_DEVICE:
943         return true;
944     case VGA_STD:
945         return pci_vga_init(pci_bus) != NULL;
946     default:
947         fprintf(stderr, "This vga model is not supported,"
948                 "currently it only supports -vga std\n");
949         exit(0);
950     }
951 }
952 
953 static const VMStateDescription vmstate_spapr = {
954     .name = "spapr",
955     .version_id = 2,
956     .minimum_version_id = 1,
957     .fields = (VMStateField[]) {
958         VMSTATE_UNUSED(4), /* used to be @next_irq */
959 
960         /* RTC offset */
961         VMSTATE_UINT64(rtc_offset, sPAPREnvironment),
962         VMSTATE_PPC_TIMEBASE_V(tb, sPAPREnvironment, 2),
963         VMSTATE_END_OF_LIST()
964     },
965 };
966 
967 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
968 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
969 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
970 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
971 
972 static int htab_save_setup(QEMUFile *f, void *opaque)
973 {
974     sPAPREnvironment *spapr = opaque;
975 
976     /* "Iteration" header */
977     qemu_put_be32(f, spapr->htab_shift);
978 
979     if (spapr->htab) {
980         spapr->htab_save_index = 0;
981         spapr->htab_first_pass = true;
982     } else {
983         assert(kvm_enabled());
984 
985         spapr->htab_fd = kvmppc_get_htab_fd(false);
986         if (spapr->htab_fd < 0) {
987             fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
988                     strerror(errno));
989             return -1;
990         }
991     }
992 
993 
994     return 0;
995 }
996 
997 static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
998                                  int64_t max_ns)
999 {
1000     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1001     int index = spapr->htab_save_index;
1002     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1003 
1004     assert(spapr->htab_first_pass);
1005 
1006     do {
1007         int chunkstart;
1008 
1009         /* Consume invalid HPTEs */
1010         while ((index < htabslots)
1011                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1012             index++;
1013             CLEAN_HPTE(HPTE(spapr->htab, index));
1014         }
1015 
1016         /* Consume valid HPTEs */
1017         chunkstart = index;
1018         while ((index < htabslots)
1019                && HPTE_VALID(HPTE(spapr->htab, index))) {
1020             index++;
1021             CLEAN_HPTE(HPTE(spapr->htab, index));
1022         }
1023 
1024         if (index > chunkstart) {
1025             int n_valid = index - chunkstart;
1026 
1027             qemu_put_be32(f, chunkstart);
1028             qemu_put_be16(f, n_valid);
1029             qemu_put_be16(f, 0);
1030             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1031                             HASH_PTE_SIZE_64 * n_valid);
1032 
1033             if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1034                 break;
1035             }
1036         }
1037     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1038 
1039     if (index >= htabslots) {
1040         assert(index == htabslots);
1041         index = 0;
1042         spapr->htab_first_pass = false;
1043     }
1044     spapr->htab_save_index = index;
1045 }
1046 
1047 static int htab_save_later_pass(QEMUFile *f, sPAPREnvironment *spapr,
1048                                 int64_t max_ns)
1049 {
1050     bool final = max_ns < 0;
1051     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1052     int examined = 0, sent = 0;
1053     int index = spapr->htab_save_index;
1054     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1055 
1056     assert(!spapr->htab_first_pass);
1057 
1058     do {
1059         int chunkstart, invalidstart;
1060 
1061         /* Consume non-dirty HPTEs */
1062         while ((index < htabslots)
1063                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1064             index++;
1065             examined++;
1066         }
1067 
1068         chunkstart = index;
1069         /* Consume valid dirty HPTEs */
1070         while ((index < htabslots)
1071                && HPTE_DIRTY(HPTE(spapr->htab, index))
1072                && HPTE_VALID(HPTE(spapr->htab, index))) {
1073             CLEAN_HPTE(HPTE(spapr->htab, index));
1074             index++;
1075             examined++;
1076         }
1077 
1078         invalidstart = index;
1079         /* Consume invalid dirty HPTEs */
1080         while ((index < htabslots)
1081                && HPTE_DIRTY(HPTE(spapr->htab, index))
1082                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1083             CLEAN_HPTE(HPTE(spapr->htab, index));
1084             index++;
1085             examined++;
1086         }
1087 
1088         if (index > chunkstart) {
1089             int n_valid = invalidstart - chunkstart;
1090             int n_invalid = index - invalidstart;
1091 
1092             qemu_put_be32(f, chunkstart);
1093             qemu_put_be16(f, n_valid);
1094             qemu_put_be16(f, n_invalid);
1095             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1096                             HASH_PTE_SIZE_64 * n_valid);
1097             sent += index - chunkstart;
1098 
1099             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1100                 break;
1101             }
1102         }
1103 
1104         if (examined >= htabslots) {
1105             break;
1106         }
1107 
1108         if (index >= htabslots) {
1109             assert(index == htabslots);
1110             index = 0;
1111         }
1112     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1113 
1114     if (index >= htabslots) {
1115         assert(index == htabslots);
1116         index = 0;
1117     }
1118 
1119     spapr->htab_save_index = index;
1120 
1121     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1122 }
1123 
1124 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1125 #define MAX_KVM_BUF_SIZE    2048
1126 
1127 static int htab_save_iterate(QEMUFile *f, void *opaque)
1128 {
1129     sPAPREnvironment *spapr = opaque;
1130     int rc = 0;
1131 
1132     /* Iteration header */
1133     qemu_put_be32(f, 0);
1134 
1135     if (!spapr->htab) {
1136         assert(kvm_enabled());
1137 
1138         rc = kvmppc_save_htab(f, spapr->htab_fd,
1139                               MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1140         if (rc < 0) {
1141             return rc;
1142         }
1143     } else  if (spapr->htab_first_pass) {
1144         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1145     } else {
1146         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1147     }
1148 
1149     /* End marker */
1150     qemu_put_be32(f, 0);
1151     qemu_put_be16(f, 0);
1152     qemu_put_be16(f, 0);
1153 
1154     return rc;
1155 }
1156 
1157 static int htab_save_complete(QEMUFile *f, void *opaque)
1158 {
1159     sPAPREnvironment *spapr = opaque;
1160 
1161     /* Iteration header */
1162     qemu_put_be32(f, 0);
1163 
1164     if (!spapr->htab) {
1165         int rc;
1166 
1167         assert(kvm_enabled());
1168 
1169         rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
1170         if (rc < 0) {
1171             return rc;
1172         }
1173         close(spapr->htab_fd);
1174         spapr->htab_fd = -1;
1175     } else {
1176         htab_save_later_pass(f, spapr, -1);
1177     }
1178 
1179     /* End marker */
1180     qemu_put_be32(f, 0);
1181     qemu_put_be16(f, 0);
1182     qemu_put_be16(f, 0);
1183 
1184     return 0;
1185 }
1186 
1187 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1188 {
1189     sPAPREnvironment *spapr = opaque;
1190     uint32_t section_hdr;
1191     int fd = -1;
1192 
1193     if (version_id < 1 || version_id > 1) {
1194         fprintf(stderr, "htab_load() bad version\n");
1195         return -EINVAL;
1196     }
1197 
1198     section_hdr = qemu_get_be32(f);
1199 
1200     if (section_hdr) {
1201         /* First section, just the hash shift */
1202         if (spapr->htab_shift != section_hdr) {
1203             return -EINVAL;
1204         }
1205         return 0;
1206     }
1207 
1208     if (!spapr->htab) {
1209         assert(kvm_enabled());
1210 
1211         fd = kvmppc_get_htab_fd(true);
1212         if (fd < 0) {
1213             fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1214                     strerror(errno));
1215         }
1216     }
1217 
1218     while (true) {
1219         uint32_t index;
1220         uint16_t n_valid, n_invalid;
1221 
1222         index = qemu_get_be32(f);
1223         n_valid = qemu_get_be16(f);
1224         n_invalid = qemu_get_be16(f);
1225 
1226         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1227             /* End of Stream */
1228             break;
1229         }
1230 
1231         if ((index + n_valid + n_invalid) >
1232             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1233             /* Bad index in stream */
1234             fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1235                     "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1236                     spapr->htab_shift);
1237             return -EINVAL;
1238         }
1239 
1240         if (spapr->htab) {
1241             if (n_valid) {
1242                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1243                                 HASH_PTE_SIZE_64 * n_valid);
1244             }
1245             if (n_invalid) {
1246                 memset(HPTE(spapr->htab, index + n_valid), 0,
1247                        HASH_PTE_SIZE_64 * n_invalid);
1248             }
1249         } else {
1250             int rc;
1251 
1252             assert(fd >= 0);
1253 
1254             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1255             if (rc < 0) {
1256                 return rc;
1257             }
1258         }
1259     }
1260 
1261     if (!spapr->htab) {
1262         assert(fd >= 0);
1263         close(fd);
1264     }
1265 
1266     return 0;
1267 }
1268 
1269 static SaveVMHandlers savevm_htab_handlers = {
1270     .save_live_setup = htab_save_setup,
1271     .save_live_iterate = htab_save_iterate,
1272     .save_live_complete = htab_save_complete,
1273     .load_state = htab_load,
1274 };
1275 
1276 /* pSeries LPAR / sPAPR hardware init */
1277 static void ppc_spapr_init(MachineState *machine)
1278 {
1279     ram_addr_t ram_size = machine->ram_size;
1280     const char *cpu_model = machine->cpu_model;
1281     const char *kernel_filename = machine->kernel_filename;
1282     const char *kernel_cmdline = machine->kernel_cmdline;
1283     const char *initrd_filename = machine->initrd_filename;
1284     const char *boot_device = machine->boot_order;
1285     PowerPCCPU *cpu;
1286     CPUPPCState *env;
1287     PCIHostState *phb;
1288     int i;
1289     MemoryRegion *sysmem = get_system_memory();
1290     MemoryRegion *ram = g_new(MemoryRegion, 1);
1291     MemoryRegion *rma_region;
1292     void *rma = NULL;
1293     hwaddr rma_alloc_size;
1294     hwaddr node0_size = spapr_node0_size();
1295     uint32_t initrd_base = 0;
1296     long kernel_size = 0, initrd_size = 0;
1297     long load_limit, fw_size;
1298     bool kernel_le = false;
1299     char *filename;
1300 
1301     msi_supported = true;
1302 
1303     spapr = g_malloc0(sizeof(*spapr));
1304     QLIST_INIT(&spapr->phbs);
1305 
1306     cpu_ppc_hypercall = emulate_spapr_hypercall;
1307 
1308     /* Allocate RMA if necessary */
1309     rma_alloc_size = kvmppc_alloc_rma(&rma);
1310 
1311     if (rma_alloc_size == -1) {
1312         hw_error("qemu: Unable to create RMA\n");
1313         exit(1);
1314     }
1315 
1316     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1317         spapr->rma_size = rma_alloc_size;
1318     } else {
1319         spapr->rma_size = node0_size;
1320 
1321         /* With KVM, we don't actually know whether KVM supports an
1322          * unbounded RMA (PR KVM) or is limited by the hash table size
1323          * (HV KVM using VRMA), so we always assume the latter
1324          *
1325          * In that case, we also limit the initial allocations for RTAS
1326          * etc... to 256M since we have no way to know what the VRMA size
1327          * is going to be as it depends on the size of the hash table
1328          * isn't determined yet.
1329          */
1330         if (kvm_enabled()) {
1331             spapr->vrma_adjust = 1;
1332             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1333         }
1334     }
1335 
1336     if (spapr->rma_size > node0_size) {
1337         fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1338                 spapr->rma_size);
1339         exit(1);
1340     }
1341 
1342     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1343     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1344 
1345     /* We aim for a hash table of size 1/128 the size of RAM.  The
1346      * normal rule of thumb is 1/64 the size of RAM, but that's much
1347      * more than needed for the Linux guests we support. */
1348     spapr->htab_shift = 18; /* Minimum architected size */
1349     while (spapr->htab_shift <= 46) {
1350         if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
1351             break;
1352         }
1353         spapr->htab_shift++;
1354     }
1355 
1356     /* Set up Interrupt Controller before we create the VCPUs */
1357     spapr->icp = xics_system_init(smp_cpus * kvmppc_smt_threads() / smp_threads,
1358                                   XICS_IRQS);
1359 
1360     /* init CPUs */
1361     if (cpu_model == NULL) {
1362         cpu_model = kvm_enabled() ? "host" : "POWER7";
1363     }
1364     for (i = 0; i < smp_cpus; i++) {
1365         cpu = cpu_ppc_init(cpu_model);
1366         if (cpu == NULL) {
1367             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1368             exit(1);
1369         }
1370         env = &cpu->env;
1371 
1372         /* Set time-base frequency to 512 MHz */
1373         cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1374 
1375         /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1376          * MSR[IP] should never be set.
1377          */
1378         env->msr_mask &= ~(1 << 6);
1379 
1380         /* Tell KVM that we're in PAPR mode */
1381         if (kvm_enabled()) {
1382             kvmppc_set_papr(cpu);
1383         }
1384 
1385         if (cpu->max_compat) {
1386             if (ppc_set_compat(cpu, cpu->max_compat) < 0) {
1387                 exit(1);
1388             }
1389         }
1390 
1391         xics_cpu_setup(spapr->icp, cpu);
1392 
1393         qemu_register_reset(spapr_cpu_reset, cpu);
1394     }
1395 
1396     /* allocate RAM */
1397     spapr->ram_limit = ram_size;
1398     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1399                                          spapr->ram_limit);
1400     memory_region_add_subregion(sysmem, 0, ram);
1401 
1402     if (rma_alloc_size && rma) {
1403         rma_region = g_new(MemoryRegion, 1);
1404         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1405                                    rma_alloc_size, rma);
1406         vmstate_register_ram_global(rma_region);
1407         memory_region_add_subregion(sysmem, 0, rma_region);
1408     }
1409 
1410     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1411     spapr->rtas_size = get_image_size(filename);
1412     spapr->rtas_blob = g_malloc(spapr->rtas_size);
1413     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1414         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1415         exit(1);
1416     }
1417     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1418         hw_error("RTAS too big ! 0x%zx bytes (max is 0x%x)\n",
1419                  spapr->rtas_size, RTAS_MAX_SIZE);
1420         exit(1);
1421     }
1422     g_free(filename);
1423 
1424     /* Set up EPOW events infrastructure */
1425     spapr_events_init(spapr);
1426 
1427     /* Set up VIO bus */
1428     spapr->vio_bus = spapr_vio_bus_init();
1429 
1430     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1431         if (serial_hds[i]) {
1432             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1433         }
1434     }
1435 
1436     /* We always have at least the nvram device on VIO */
1437     spapr_create_nvram(spapr);
1438 
1439     /* Set up PCI */
1440     spapr_pci_rtas_init();
1441 
1442     phb = spapr_create_phb(spapr, 0);
1443 
1444     for (i = 0; i < nb_nics; i++) {
1445         NICInfo *nd = &nd_table[i];
1446 
1447         if (!nd->model) {
1448             nd->model = g_strdup("ibmveth");
1449         }
1450 
1451         if (strcmp(nd->model, "ibmveth") == 0) {
1452             spapr_vlan_create(spapr->vio_bus, nd);
1453         } else {
1454             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1455         }
1456     }
1457 
1458     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1459         spapr_vscsi_create(spapr->vio_bus);
1460     }
1461 
1462     /* Graphics */
1463     if (spapr_vga_init(phb->bus)) {
1464         spapr->has_graphics = true;
1465     }
1466 
1467     if (usb_enabled(spapr->has_graphics)) {
1468         pci_create_simple(phb->bus, -1, "pci-ohci");
1469         if (spapr->has_graphics) {
1470             usbdevice_create("keyboard");
1471             usbdevice_create("mouse");
1472         }
1473     }
1474 
1475     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1476         fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1477                 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1478         exit(1);
1479     }
1480 
1481     if (kernel_filename) {
1482         uint64_t lowaddr = 0;
1483 
1484         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1485                                NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1486         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1487             kernel_size = load_elf(kernel_filename,
1488                                    translate_kernel_address, NULL,
1489                                    NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1490             kernel_le = kernel_size > 0;
1491         }
1492         if (kernel_size < 0) {
1493             fprintf(stderr, "qemu: error loading %s: %s\n",
1494                     kernel_filename, load_elf_strerror(kernel_size));
1495             exit(1);
1496         }
1497 
1498         /* load initrd */
1499         if (initrd_filename) {
1500             /* Try to locate the initrd in the gap between the kernel
1501              * and the firmware. Add a bit of space just in case
1502              */
1503             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1504             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1505                                               load_limit - initrd_base);
1506             if (initrd_size < 0) {
1507                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1508                         initrd_filename);
1509                 exit(1);
1510             }
1511         } else {
1512             initrd_base = 0;
1513             initrd_size = 0;
1514         }
1515     }
1516 
1517     if (bios_name == NULL) {
1518         bios_name = FW_FILE_NAME;
1519     }
1520     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1521     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1522     if (fw_size < 0) {
1523         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1524         exit(1);
1525     }
1526     g_free(filename);
1527 
1528     spapr->entry_point = 0x100;
1529 
1530     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1531     register_savevm_live(NULL, "spapr/htab", -1, 1,
1532                          &savevm_htab_handlers, spapr);
1533 
1534     /* Prepare the device tree */
1535     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1536                                             kernel_size, kernel_le,
1537                                             boot_device, kernel_cmdline,
1538                                             spapr->epow_irq);
1539     assert(spapr->fdt_skel != NULL);
1540 }
1541 
1542 static int spapr_kvm_type(const char *vm_type)
1543 {
1544     if (!vm_type) {
1545         return 0;
1546     }
1547 
1548     if (!strcmp(vm_type, "HV")) {
1549         return 1;
1550     }
1551 
1552     if (!strcmp(vm_type, "PR")) {
1553         return 2;
1554     }
1555 
1556     error_report("Unknown kvm-type specified '%s'", vm_type);
1557     exit(1);
1558 }
1559 
1560 /*
1561  * Implementation of an interface to adjust firmware patch
1562  * for the bootindex property handling.
1563  */
1564 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
1565                                    DeviceState *dev)
1566 {
1567 #define CAST(type, obj, name) \
1568     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
1569     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
1570     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
1571 
1572     if (d) {
1573         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
1574         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
1575         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
1576 
1577         if (spapr) {
1578             /*
1579              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
1580              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
1581              * in the top 16 bits of the 64-bit LUN
1582              */
1583             unsigned id = 0x8000 | (d->id << 8) | d->lun;
1584             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1585                                    (uint64_t)id << 48);
1586         } else if (virtio) {
1587             /*
1588              * We use SRP luns of the form 01000000 | (target << 8) | lun
1589              * in the top 32 bits of the 64-bit LUN
1590              * Note: the quote above is from SLOF and it is wrong,
1591              * the actual binding is:
1592              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
1593              */
1594             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
1595             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1596                                    (uint64_t)id << 32);
1597         } else if (usb) {
1598             /*
1599              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
1600              * in the top 32 bits of the 64-bit LUN
1601              */
1602             unsigned usb_port = atoi(usb->port->path);
1603             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
1604             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1605                                    (uint64_t)id << 32);
1606         }
1607     }
1608 
1609     if (phb) {
1610         /* Replace "pci" with "pci@800000020000000" */
1611         return g_strdup_printf("pci@%"PRIX64, phb->buid);
1612     }
1613 
1614     return NULL;
1615 }
1616 
1617 static char *spapr_get_kvm_type(Object *obj, Error **errp)
1618 {
1619     sPAPRMachineState *sm = SPAPR_MACHINE(obj);
1620 
1621     return g_strdup(sm->kvm_type);
1622 }
1623 
1624 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
1625 {
1626     sPAPRMachineState *sm = SPAPR_MACHINE(obj);
1627 
1628     g_free(sm->kvm_type);
1629     sm->kvm_type = g_strdup(value);
1630 }
1631 
1632 static void spapr_machine_initfn(Object *obj)
1633 {
1634     object_property_add_str(obj, "kvm-type",
1635                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
1636 }
1637 
1638 static void ppc_cpu_do_nmi_on_cpu(void *arg)
1639 {
1640     CPUState *cs = arg;
1641 
1642     cpu_synchronize_state(cs);
1643     ppc_cpu_do_system_reset(cs);
1644 }
1645 
1646 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
1647 {
1648     CPUState *cs;
1649 
1650     CPU_FOREACH(cs) {
1651         async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, cs);
1652     }
1653 }
1654 
1655 static void spapr_machine_class_init(ObjectClass *oc, void *data)
1656 {
1657     MachineClass *mc = MACHINE_CLASS(oc);
1658     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
1659     NMIClass *nc = NMI_CLASS(oc);
1660 
1661     mc->name = "pseries";
1662     mc->desc = "pSeries Logical Partition (PAPR compliant)";
1663     mc->is_default = 1;
1664     mc->init = ppc_spapr_init;
1665     mc->reset = ppc_spapr_reset;
1666     mc->block_default_type = IF_SCSI;
1667     mc->max_cpus = MAX_CPUS;
1668     mc->no_parallel = 1;
1669     mc->default_boot_order = NULL;
1670     mc->kvm_type = spapr_kvm_type;
1671 
1672     fwc->get_dev_path = spapr_get_fw_dev_path;
1673     nc->nmi_monitor_handler = spapr_nmi;
1674 }
1675 
1676 static const TypeInfo spapr_machine_info = {
1677     .name          = TYPE_SPAPR_MACHINE,
1678     .parent        = TYPE_MACHINE,
1679     .instance_size = sizeof(sPAPRMachineState),
1680     .instance_init = spapr_machine_initfn,
1681     .class_init    = spapr_machine_class_init,
1682     .interfaces = (InterfaceInfo[]) {
1683         { TYPE_FW_PATH_PROVIDER },
1684         { TYPE_NMI },
1685         { }
1686     },
1687 };
1688 
1689 static void spapr_machine_2_1_class_init(ObjectClass *oc, void *data)
1690 {
1691     MachineClass *mc = MACHINE_CLASS(oc);
1692 
1693     mc->name = "pseries-2.1";
1694     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.1";
1695     mc->is_default = 0;
1696 }
1697 
1698 static const TypeInfo spapr_machine_2_1_info = {
1699     .name          = TYPE_SPAPR_MACHINE "2.1",
1700     .parent        = TYPE_SPAPR_MACHINE,
1701     .class_init    = spapr_machine_2_1_class_init,
1702 };
1703 
1704 static void spapr_machine_register_types(void)
1705 {
1706     type_register_static(&spapr_machine_info);
1707     type_register_static(&spapr_machine_2_1_info);
1708 }
1709 
1710 type_init(spapr_machine_register_types)
1711