xref: /openbmc/qemu/hw/ppc/spapr.c (revision 3a3b8502e6f0c8d30865c5f36d2c3ae4114000b5)
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 
59 #include <libfdt.h>
60 
61 /* SLOF memory layout:
62  *
63  * SLOF raw image loaded at 0, copies its romfs right below the flat
64  * device-tree, then position SLOF itself 31M below that
65  *
66  * So we set FW_OVERHEAD to 40MB which should account for all of that
67  * and more
68  *
69  * We load our kernel at 4M, leaving space for SLOF initial image
70  */
71 #define FDT_MAX_SIZE            0x40000
72 #define RTAS_MAX_SIZE           0x10000
73 #define FW_MAX_SIZE             0x400000
74 #define FW_FILE_NAME            "slof.bin"
75 #define FW_OVERHEAD             0x2800000
76 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
77 
78 #define MIN_RMA_SLOF            128UL
79 
80 #define TIMEBASE_FREQ           512000000ULL
81 
82 #define MAX_CPUS                256
83 
84 #define PHANDLE_XICP            0x00001111
85 
86 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
87 
88 
89 typedef struct SPAPRMachine SPAPRMachine;
90 #define TYPE_SPAPR_MACHINE      "spapr-machine"
91 #define SPAPR_MACHINE(obj) \
92     OBJECT_CHECK(SPAPRMachine, (obj), TYPE_SPAPR_MACHINE)
93 
94 /**
95  * SPAPRMachine:
96  */
97 struct SPAPRMachine {
98     /*< private >*/
99     MachineState parent_obj;
100 
101     /*< public >*/
102     char *kvm_type;
103 };
104 
105 
106 sPAPREnvironment *spapr;
107 
108 int spapr_allocate_irq(int hint, bool lsi)
109 {
110     int irq;
111 
112     if (hint) {
113         irq = hint;
114         if (hint >= spapr->next_irq) {
115             spapr->next_irq = hint + 1;
116         }
117         /* FIXME: we should probably check for collisions somehow */
118     } else {
119         irq = spapr->next_irq++;
120     }
121 
122     /* Configure irq type */
123     if (!xics_get_qirq(spapr->icp, irq)) {
124         return 0;
125     }
126 
127     xics_set_irq_type(spapr->icp, irq, lsi);
128 
129     return irq;
130 }
131 
132 /*
133  * Allocate block of consequtive IRQs, returns a number of the first.
134  * If msi==true, aligns the first IRQ number to num.
135  */
136 int spapr_allocate_irq_block(int num, bool lsi, bool msi)
137 {
138     int first = -1;
139     int i, hint = 0;
140 
141     /*
142      * MSIMesage::data is used for storing VIRQ so
143      * it has to be aligned to num to support multiple
144      * MSI vectors. MSI-X is not affected by this.
145      * The hint is used for the first IRQ, the rest should
146      * be allocated continuously.
147      */
148     if (msi) {
149         assert((num == 1) || (num == 2) || (num == 4) ||
150                (num == 8) || (num == 16) || (num == 32));
151         hint = (spapr->next_irq + num - 1) & ~(num - 1);
152     }
153 
154     for (i = 0; i < num; ++i) {
155         int irq;
156 
157         irq = spapr_allocate_irq(hint, lsi);
158         if (!irq) {
159             return -1;
160         }
161 
162         if (0 == i) {
163             first = irq;
164             hint = 0;
165         }
166 
167         /* If the above doesn't create a consecutive block then that's
168          * an internal bug */
169         assert(irq == (first + i));
170     }
171 
172     return first;
173 }
174 
175 static XICSState *try_create_xics(const char *type, int nr_servers,
176                                   int nr_irqs)
177 {
178     DeviceState *dev;
179 
180     dev = qdev_create(NULL, type);
181     qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
182     qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
183     if (qdev_init(dev) < 0) {
184         return NULL;
185     }
186 
187     return XICS_COMMON(dev);
188 }
189 
190 static XICSState *xics_system_init(int nr_servers, int nr_irqs)
191 {
192     XICSState *icp = NULL;
193 
194     if (kvm_enabled()) {
195         QemuOpts *machine_opts = qemu_get_machine_opts();
196         bool irqchip_allowed = qemu_opt_get_bool(machine_opts,
197                                                 "kernel_irqchip", true);
198         bool irqchip_required = qemu_opt_get_bool(machine_opts,
199                                                   "kernel_irqchip", false);
200         if (irqchip_allowed) {
201             icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs);
202         }
203 
204         if (irqchip_required && !icp) {
205             perror("Failed to create in-kernel XICS\n");
206             abort();
207         }
208     }
209 
210     if (!icp) {
211         icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs);
212     }
213 
214     if (!icp) {
215         perror("Failed to create XICS\n");
216         abort();
217     }
218 
219     return icp;
220 }
221 
222 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
223                                   int smt_threads)
224 {
225     int i, ret = 0;
226     uint32_t servers_prop[smt_threads];
227     uint32_t gservers_prop[smt_threads * 2];
228     int index = ppc_get_vcpu_dt_id(cpu);
229 
230     if (cpu->cpu_version) {
231         ret = fdt_setprop(fdt, offset, "cpu-version",
232                           &cpu->cpu_version, sizeof(cpu->cpu_version));
233         if (ret < 0) {
234             return ret;
235         }
236     }
237 
238     /* Build interrupt servers and gservers properties */
239     for (i = 0; i < smt_threads; i++) {
240         servers_prop[i] = cpu_to_be32(index + i);
241         /* Hack, direct the group queues back to cpu 0 */
242         gservers_prop[i*2] = cpu_to_be32(index + i);
243         gservers_prop[i*2 + 1] = 0;
244     }
245     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
246                       servers_prop, sizeof(servers_prop));
247     if (ret < 0) {
248         return ret;
249     }
250     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
251                       gservers_prop, sizeof(gservers_prop));
252 
253     return ret;
254 }
255 
256 static int spapr_fixup_cpu_dt(void *fdt, sPAPREnvironment *spapr)
257 {
258     int ret = 0, offset, cpus_offset;
259     CPUState *cs;
260     char cpu_model[32];
261     int smt = kvmppc_smt_threads();
262     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
263 
264     CPU_FOREACH(cs) {
265         PowerPCCPU *cpu = POWERPC_CPU(cs);
266         DeviceClass *dc = DEVICE_GET_CLASS(cs);
267         int index = ppc_get_vcpu_dt_id(cpu);
268         uint32_t associativity[] = {cpu_to_be32(0x5),
269                                     cpu_to_be32(0x0),
270                                     cpu_to_be32(0x0),
271                                     cpu_to_be32(0x0),
272                                     cpu_to_be32(cs->numa_node),
273                                     cpu_to_be32(index)};
274 
275         if ((index % smt) != 0) {
276             continue;
277         }
278 
279         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
280 
281         cpus_offset = fdt_path_offset(fdt, "/cpus");
282         if (cpus_offset < 0) {
283             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
284                                           "cpus");
285             if (cpus_offset < 0) {
286                 return cpus_offset;
287             }
288         }
289         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
290         if (offset < 0) {
291             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
292             if (offset < 0) {
293                 return offset;
294             }
295         }
296 
297         if (nb_numa_nodes > 1) {
298             ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
299                               sizeof(associativity));
300             if (ret < 0) {
301                 return ret;
302             }
303         }
304 
305         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
306                           pft_size_prop, sizeof(pft_size_prop));
307         if (ret < 0) {
308             return ret;
309         }
310 
311         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
312                                      ppc_get_compat_smt_threads(cpu));
313         if (ret < 0) {
314             return ret;
315         }
316     }
317     return ret;
318 }
319 
320 
321 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
322                                      size_t maxsize)
323 {
324     size_t maxcells = maxsize / sizeof(uint32_t);
325     int i, j, count;
326     uint32_t *p = prop;
327 
328     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
329         struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
330 
331         if (!sps->page_shift) {
332             break;
333         }
334         for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
335             if (sps->enc[count].page_shift == 0) {
336                 break;
337             }
338         }
339         if ((p - prop) >= (maxcells - 3 - count * 2)) {
340             break;
341         }
342         *(p++) = cpu_to_be32(sps->page_shift);
343         *(p++) = cpu_to_be32(sps->slb_enc);
344         *(p++) = cpu_to_be32(count);
345         for (j = 0; j < count; j++) {
346             *(p++) = cpu_to_be32(sps->enc[j].page_shift);
347             *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
348         }
349     }
350 
351     return (p - prop) * sizeof(uint32_t);
352 }
353 
354 #define _FDT(exp) \
355     do { \
356         int ret = (exp);                                           \
357         if (ret < 0) {                                             \
358             fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
359                     #exp, fdt_strerror(ret));                      \
360             exit(1);                                               \
361         }                                                          \
362     } while (0)
363 
364 static void add_str(GString *s, const gchar *s1)
365 {
366     g_string_append_len(s, s1, strlen(s1) + 1);
367 }
368 
369 static void *spapr_create_fdt_skel(hwaddr initrd_base,
370                                    hwaddr initrd_size,
371                                    hwaddr kernel_size,
372                                    bool little_endian,
373                                    const char *boot_device,
374                                    const char *kernel_cmdline,
375                                    uint32_t epow_irq)
376 {
377     void *fdt;
378     CPUState *cs;
379     uint32_t start_prop = cpu_to_be32(initrd_base);
380     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
381     GString *hypertas = g_string_sized_new(256);
382     GString *qemu_hypertas = g_string_sized_new(256);
383     uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
384     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(smp_cpus)};
385     int smt = kvmppc_smt_threads();
386     unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
387     QemuOpts *opts = qemu_opts_find(qemu_find_opts("smp-opts"), NULL);
388     unsigned sockets = opts ? qemu_opt_get_number(opts, "sockets", 0) : 0;
389     uint32_t cpus_per_socket = sockets ? (smp_cpus / sockets) : 1;
390 
391     add_str(hypertas, "hcall-pft");
392     add_str(hypertas, "hcall-term");
393     add_str(hypertas, "hcall-dabr");
394     add_str(hypertas, "hcall-interrupt");
395     add_str(hypertas, "hcall-tce");
396     add_str(hypertas, "hcall-vio");
397     add_str(hypertas, "hcall-splpar");
398     add_str(hypertas, "hcall-bulk");
399     add_str(hypertas, "hcall-set-mode");
400     add_str(qemu_hypertas, "hcall-memop1");
401 
402     fdt = g_malloc0(FDT_MAX_SIZE);
403     _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
404 
405     if (kernel_size) {
406         _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
407     }
408     if (initrd_size) {
409         _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
410     }
411     _FDT((fdt_finish_reservemap(fdt)));
412 
413     /* Root node */
414     _FDT((fdt_begin_node(fdt, "")));
415     _FDT((fdt_property_string(fdt, "device_type", "chrp")));
416     _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
417     _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
418 
419     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
420     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
421 
422     /* /chosen */
423     _FDT((fdt_begin_node(fdt, "chosen")));
424 
425     /* Set Form1_affinity */
426     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
427 
428     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
429     _FDT((fdt_property(fdt, "linux,initrd-start",
430                        &start_prop, sizeof(start_prop))));
431     _FDT((fdt_property(fdt, "linux,initrd-end",
432                        &end_prop, sizeof(end_prop))));
433     if (kernel_size) {
434         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
435                               cpu_to_be64(kernel_size) };
436 
437         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
438         if (little_endian) {
439             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
440         }
441     }
442     if (boot_device) {
443         _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
444     }
445     if (boot_menu) {
446         _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
447     }
448     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
449     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
450     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
451 
452     _FDT((fdt_end_node(fdt)));
453 
454     /* cpus */
455     _FDT((fdt_begin_node(fdt, "cpus")));
456 
457     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
458     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
459 
460     CPU_FOREACH(cs) {
461         PowerPCCPU *cpu = POWERPC_CPU(cs);
462         CPUPPCState *env = &cpu->env;
463         DeviceClass *dc = DEVICE_GET_CLASS(cs);
464         PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
465         int index = ppc_get_vcpu_dt_id(cpu);
466         char *nodename;
467         uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
468                            0xffffffff, 0xffffffff};
469         uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
470         uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
471         uint32_t page_sizes_prop[64];
472         size_t page_sizes_prop_size;
473 
474         if ((index % smt) != 0) {
475             continue;
476         }
477 
478         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
479 
480         _FDT((fdt_begin_node(fdt, nodename)));
481 
482         g_free(nodename);
483 
484         _FDT((fdt_property_cell(fdt, "reg", index)));
485         _FDT((fdt_property_string(fdt, "device_type", "cpu")));
486 
487         _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
488         _FDT((fdt_property_cell(fdt, "d-cache-block-size",
489                                 env->dcache_line_size)));
490         _FDT((fdt_property_cell(fdt, "d-cache-line-size",
491                                 env->dcache_line_size)));
492         _FDT((fdt_property_cell(fdt, "i-cache-block-size",
493                                 env->icache_line_size)));
494         _FDT((fdt_property_cell(fdt, "i-cache-line-size",
495                                 env->icache_line_size)));
496 
497         if (pcc->l1_dcache_size) {
498             _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
499         } else {
500             fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
501         }
502         if (pcc->l1_icache_size) {
503             _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
504         } else {
505             fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
506         }
507 
508         _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
509         _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
510         _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
511         _FDT((fdt_property_string(fdt, "status", "okay")));
512         _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
513 
514         if (env->spr_cb[SPR_PURR].oea_read) {
515             _FDT((fdt_property(fdt, "ibm,purr", NULL, 0)));
516         }
517 
518         if (env->mmu_model & POWERPC_MMU_1TSEG) {
519             _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
520                                segs, sizeof(segs))));
521         }
522 
523         /* Advertise VMX/VSX (vector extensions) if available
524          *   0 / no property == no vector extensions
525          *   1               == VMX / Altivec available
526          *   2               == VSX available */
527         if (env->insns_flags & PPC_ALTIVEC) {
528             uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
529 
530             _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
531         }
532 
533         /* Advertise DFP (Decimal Floating Point) if available
534          *   0 / no property == no DFP
535          *   1               == DFP available */
536         if (env->insns_flags2 & PPC2_DFP) {
537             _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
538         }
539 
540         page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
541                                                       sizeof(page_sizes_prop));
542         if (page_sizes_prop_size) {
543             _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
544                                page_sizes_prop, page_sizes_prop_size)));
545         }
546 
547         _FDT((fdt_property_cell(fdt, "ibm,chip-id",
548                                 cs->cpu_index / cpus_per_socket)));
549 
550         _FDT((fdt_end_node(fdt)));
551     }
552 
553     _FDT((fdt_end_node(fdt)));
554 
555     /* RTAS */
556     _FDT((fdt_begin_node(fdt, "rtas")));
557 
558     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
559         add_str(hypertas, "hcall-multi-tce");
560     }
561     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
562                        hypertas->len)));
563     g_string_free(hypertas, TRUE);
564     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
565                        qemu_hypertas->len)));
566     g_string_free(qemu_hypertas, TRUE);
567 
568     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
569         refpoints, sizeof(refpoints))));
570 
571     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
572 
573     _FDT((fdt_end_node(fdt)));
574 
575     /* interrupt controller */
576     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
577 
578     _FDT((fdt_property_string(fdt, "device_type",
579                               "PowerPC-External-Interrupt-Presentation")));
580     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
581     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
582     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
583                        interrupt_server_ranges_prop,
584                        sizeof(interrupt_server_ranges_prop))));
585     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
586     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
587     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
588 
589     _FDT((fdt_end_node(fdt)));
590 
591     /* vdevice */
592     _FDT((fdt_begin_node(fdt, "vdevice")));
593 
594     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
595     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
596     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
597     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
598     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
599     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
600 
601     _FDT((fdt_end_node(fdt)));
602 
603     /* event-sources */
604     spapr_events_fdt_skel(fdt, epow_irq);
605 
606     /* /hypervisor node */
607     if (kvm_enabled()) {
608         uint8_t hypercall[16];
609 
610         /* indicate KVM hypercall interface */
611         _FDT((fdt_begin_node(fdt, "hypervisor")));
612         _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
613         if (kvmppc_has_cap_fixup_hcalls()) {
614             /*
615              * Older KVM versions with older guest kernels were broken with the
616              * magic page, don't allow the guest to map it.
617              */
618             kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
619                                  sizeof(hypercall));
620             _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
621                               sizeof(hypercall))));
622         }
623         _FDT((fdt_end_node(fdt)));
624     }
625 
626     _FDT((fdt_end_node(fdt))); /* close root node */
627     _FDT((fdt_finish(fdt)));
628 
629     return fdt;
630 }
631 
632 int spapr_h_cas_compose_response(target_ulong addr, target_ulong size)
633 {
634     void *fdt, *fdt_skel;
635     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
636 
637     size -= sizeof(hdr);
638 
639     /* Create sceleton */
640     fdt_skel = g_malloc0(size);
641     _FDT((fdt_create(fdt_skel, size)));
642     _FDT((fdt_begin_node(fdt_skel, "")));
643     _FDT((fdt_end_node(fdt_skel)));
644     _FDT((fdt_finish(fdt_skel)));
645     fdt = g_malloc0(size);
646     _FDT((fdt_open_into(fdt_skel, fdt, size)));
647     g_free(fdt_skel);
648 
649     /* Fix skeleton up */
650     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
651 
652     /* Pack resulting tree */
653     _FDT((fdt_pack(fdt)));
654 
655     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
656         trace_spapr_cas_failed(size);
657         return -1;
658     }
659 
660     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
661     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
662     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
663     g_free(fdt);
664 
665     return 0;
666 }
667 
668 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
669 {
670     uint32_t associativity[] = {cpu_to_be32(0x4), cpu_to_be32(0x0),
671                                 cpu_to_be32(0x0), cpu_to_be32(0x0),
672                                 cpu_to_be32(0x0)};
673     char mem_name[32];
674     hwaddr node0_size, mem_start, node_size;
675     uint64_t mem_reg_property[2];
676     int i, off;
677 
678     /* memory node(s) */
679     if (nb_numa_nodes > 1 && numa_info[0].node_mem < ram_size) {
680         node0_size = numa_info[0].node_mem;
681     } else {
682         node0_size = ram_size;
683     }
684 
685     /* RMA */
686     mem_reg_property[0] = 0;
687     mem_reg_property[1] = cpu_to_be64(spapr->rma_size);
688     off = fdt_add_subnode(fdt, 0, "memory@0");
689     _FDT(off);
690     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
691     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
692                       sizeof(mem_reg_property))));
693     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
694                       sizeof(associativity))));
695 
696     /* RAM: Node 0 */
697     if (node0_size > spapr->rma_size) {
698         mem_reg_property[0] = cpu_to_be64(spapr->rma_size);
699         mem_reg_property[1] = cpu_to_be64(node0_size - spapr->rma_size);
700 
701         sprintf(mem_name, "memory@" TARGET_FMT_lx, spapr->rma_size);
702         off = fdt_add_subnode(fdt, 0, mem_name);
703         _FDT(off);
704         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
705         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
706                           sizeof(mem_reg_property))));
707         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
708                           sizeof(associativity))));
709     }
710 
711     /* RAM: Node 1 and beyond */
712     mem_start = node0_size;
713     for (i = 1; i < nb_numa_nodes; i++) {
714         mem_reg_property[0] = cpu_to_be64(mem_start);
715         if (mem_start >= ram_size) {
716             node_size = 0;
717         } else {
718             node_size = numa_info[i].node_mem;
719             if (node_size > ram_size - mem_start) {
720                 node_size = ram_size - mem_start;
721             }
722         }
723         mem_reg_property[1] = cpu_to_be64(node_size);
724         associativity[3] = associativity[4] = cpu_to_be32(i);
725         sprintf(mem_name, "memory@" TARGET_FMT_lx, mem_start);
726         off = fdt_add_subnode(fdt, 0, mem_name);
727         _FDT(off);
728         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
729         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
730                           sizeof(mem_reg_property))));
731         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
732                           sizeof(associativity))));
733         mem_start += node_size;
734     }
735 
736     return 0;
737 }
738 
739 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
740                                hwaddr fdt_addr,
741                                hwaddr rtas_addr,
742                                hwaddr rtas_size)
743 {
744     int ret, i;
745     size_t cb = 0;
746     char *bootlist;
747     void *fdt;
748     sPAPRPHBState *phb;
749 
750     fdt = g_malloc(FDT_MAX_SIZE);
751 
752     /* open out the base tree into a temp buffer for the final tweaks */
753     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
754 
755     ret = spapr_populate_memory(spapr, fdt);
756     if (ret < 0) {
757         fprintf(stderr, "couldn't setup memory nodes in fdt\n");
758         exit(1);
759     }
760 
761     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
762     if (ret < 0) {
763         fprintf(stderr, "couldn't setup vio devices in fdt\n");
764         exit(1);
765     }
766 
767     QLIST_FOREACH(phb, &spapr->phbs, list) {
768         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
769     }
770 
771     if (ret < 0) {
772         fprintf(stderr, "couldn't setup PCI devices in fdt\n");
773         exit(1);
774     }
775 
776     /* RTAS */
777     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
778     if (ret < 0) {
779         fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
780     }
781 
782     /* Advertise NUMA via ibm,associativity */
783     ret = spapr_fixup_cpu_dt(fdt, spapr);
784     if (ret < 0) {
785         fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
786     }
787 
788     bootlist = get_boot_devices_list(&cb, true);
789     if (cb && bootlist) {
790         int offset = fdt_path_offset(fdt, "/chosen");
791         if (offset < 0) {
792             exit(1);
793         }
794         for (i = 0; i < cb; i++) {
795             if (bootlist[i] == '\n') {
796                 bootlist[i] = ' ';
797             }
798 
799         }
800         ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
801     }
802 
803     if (!spapr->has_graphics) {
804         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
805     }
806 
807     _FDT((fdt_pack(fdt)));
808 
809     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
810         hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
811                  fdt_totalsize(fdt), FDT_MAX_SIZE);
812         exit(1);
813     }
814 
815     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
816 
817     g_free(fdt);
818 }
819 
820 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
821 {
822     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
823 }
824 
825 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
826 {
827     CPUPPCState *env = &cpu->env;
828 
829     if (msr_pr) {
830         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
831         env->gpr[3] = H_PRIVILEGE;
832     } else {
833         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
834     }
835 }
836 
837 static void spapr_reset_htab(sPAPREnvironment *spapr)
838 {
839     long shift;
840 
841     /* allocate hash page table.  For now we always make this 16mb,
842      * later we should probably make it scale to the size of guest
843      * RAM */
844 
845     shift = kvmppc_reset_htab(spapr->htab_shift);
846 
847     if (shift > 0) {
848         /* Kernel handles htab, we don't need to allocate one */
849         spapr->htab_shift = shift;
850         kvmppc_kern_htab = true;
851     } else {
852         if (!spapr->htab) {
853             /* Allocate an htab if we don't yet have one */
854             spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
855         }
856 
857         /* And clear it */
858         memset(spapr->htab, 0, HTAB_SIZE(spapr));
859     }
860 
861     /* Update the RMA size if necessary */
862     if (spapr->vrma_adjust) {
863         hwaddr node0_size = (nb_numa_nodes > 1) ?
864             numa_info[0].node_mem : ram_size;
865         spapr->rma_size = kvmppc_rma_size(node0_size, spapr->htab_shift);
866     }
867 }
868 
869 static void ppc_spapr_reset(void)
870 {
871     PowerPCCPU *first_ppc_cpu;
872 
873     /* Reset the hash table & recalc the RMA */
874     spapr_reset_htab(spapr);
875 
876     qemu_devices_reset();
877 
878     /* Load the fdt */
879     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
880                        spapr->rtas_size);
881 
882     /* Set up the entry state */
883     first_ppc_cpu = POWERPC_CPU(first_cpu);
884     first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
885     first_ppc_cpu->env.gpr[5] = 0;
886     first_cpu->halted = 0;
887     first_ppc_cpu->env.nip = spapr->entry_point;
888 
889 }
890 
891 static void spapr_cpu_reset(void *opaque)
892 {
893     PowerPCCPU *cpu = opaque;
894     CPUState *cs = CPU(cpu);
895     CPUPPCState *env = &cpu->env;
896 
897     cpu_reset(cs);
898 
899     /* All CPUs start halted.  CPU0 is unhalted from the machine level
900      * reset code and the rest are explicitly started up by the guest
901      * using an RTAS call */
902     cs->halted = 1;
903 
904     env->spr[SPR_HIOR] = 0;
905 
906     env->external_htab = (uint8_t *)spapr->htab;
907     if (kvm_enabled() && !env->external_htab) {
908         /*
909          * HV KVM, set external_htab to 1 so our ppc_hash64_load_hpte*
910          * functions do the right thing.
911          */
912         env->external_htab = (void *)1;
913     }
914     env->htab_base = -1;
915     /*
916      * htab_mask is the mask used to normalize hash value to PTEG index.
917      * htab_shift is log2 of hash table size.
918      * We have 8 hpte per group, and each hpte is 16 bytes.
919      * ie have 128 bytes per hpte entry.
920      */
921     env->htab_mask = (1ULL << ((spapr)->htab_shift - 7)) - 1;
922     env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
923         (spapr->htab_shift - 18);
924 }
925 
926 static void spapr_create_nvram(sPAPREnvironment *spapr)
927 {
928     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
929     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
930 
931     if (dinfo) {
932         qdev_prop_set_drive_nofail(dev, "drive", dinfo->bdrv);
933     }
934 
935     qdev_init_nofail(dev);
936 
937     spapr->nvram = (struct sPAPRNVRAM *)dev;
938 }
939 
940 /* Returns whether we want to use VGA or not */
941 static int spapr_vga_init(PCIBus *pci_bus)
942 {
943     switch (vga_interface_type) {
944     case VGA_NONE:
945         return false;
946     case VGA_DEVICE:
947         return true;
948     case VGA_STD:
949         return pci_vga_init(pci_bus) != NULL;
950     default:
951         fprintf(stderr, "This vga model is not supported,"
952                 "currently it only supports -vga std\n");
953         exit(0);
954     }
955 }
956 
957 static const VMStateDescription vmstate_spapr = {
958     .name = "spapr",
959     .version_id = 2,
960     .minimum_version_id = 1,
961     .fields = (VMStateField[]) {
962         VMSTATE_UINT32(next_irq, sPAPREnvironment),
963 
964         /* RTC offset */
965         VMSTATE_UINT64(rtc_offset, sPAPREnvironment),
966         VMSTATE_PPC_TIMEBASE_V(tb, sPAPREnvironment, 2),
967         VMSTATE_END_OF_LIST()
968     },
969 };
970 
971 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
972 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
973 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
974 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
975 
976 static int htab_save_setup(QEMUFile *f, void *opaque)
977 {
978     sPAPREnvironment *spapr = opaque;
979 
980     /* "Iteration" header */
981     qemu_put_be32(f, spapr->htab_shift);
982 
983     if (spapr->htab) {
984         spapr->htab_save_index = 0;
985         spapr->htab_first_pass = true;
986     } else {
987         assert(kvm_enabled());
988 
989         spapr->htab_fd = kvmppc_get_htab_fd(false);
990         if (spapr->htab_fd < 0) {
991             fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
992                     strerror(errno));
993             return -1;
994         }
995     }
996 
997 
998     return 0;
999 }
1000 
1001 static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
1002                                  int64_t max_ns)
1003 {
1004     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1005     int index = spapr->htab_save_index;
1006     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1007 
1008     assert(spapr->htab_first_pass);
1009 
1010     do {
1011         int chunkstart;
1012 
1013         /* Consume invalid HPTEs */
1014         while ((index < htabslots)
1015                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1016             index++;
1017             CLEAN_HPTE(HPTE(spapr->htab, index));
1018         }
1019 
1020         /* Consume valid HPTEs */
1021         chunkstart = index;
1022         while ((index < htabslots)
1023                && HPTE_VALID(HPTE(spapr->htab, index))) {
1024             index++;
1025             CLEAN_HPTE(HPTE(spapr->htab, index));
1026         }
1027 
1028         if (index > chunkstart) {
1029             int n_valid = index - chunkstart;
1030 
1031             qemu_put_be32(f, chunkstart);
1032             qemu_put_be16(f, n_valid);
1033             qemu_put_be16(f, 0);
1034             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1035                             HASH_PTE_SIZE_64 * n_valid);
1036 
1037             if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1038                 break;
1039             }
1040         }
1041     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1042 
1043     if (index >= htabslots) {
1044         assert(index == htabslots);
1045         index = 0;
1046         spapr->htab_first_pass = false;
1047     }
1048     spapr->htab_save_index = index;
1049 }
1050 
1051 static int htab_save_later_pass(QEMUFile *f, sPAPREnvironment *spapr,
1052                                 int64_t max_ns)
1053 {
1054     bool final = max_ns < 0;
1055     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1056     int examined = 0, sent = 0;
1057     int index = spapr->htab_save_index;
1058     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1059 
1060     assert(!spapr->htab_first_pass);
1061 
1062     do {
1063         int chunkstart, invalidstart;
1064 
1065         /* Consume non-dirty HPTEs */
1066         while ((index < htabslots)
1067                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1068             index++;
1069             examined++;
1070         }
1071 
1072         chunkstart = index;
1073         /* Consume valid dirty HPTEs */
1074         while ((index < htabslots)
1075                && HPTE_DIRTY(HPTE(spapr->htab, index))
1076                && HPTE_VALID(HPTE(spapr->htab, index))) {
1077             CLEAN_HPTE(HPTE(spapr->htab, index));
1078             index++;
1079             examined++;
1080         }
1081 
1082         invalidstart = index;
1083         /* Consume invalid dirty HPTEs */
1084         while ((index < htabslots)
1085                && HPTE_DIRTY(HPTE(spapr->htab, index))
1086                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1087             CLEAN_HPTE(HPTE(spapr->htab, index));
1088             index++;
1089             examined++;
1090         }
1091 
1092         if (index > chunkstart) {
1093             int n_valid = invalidstart - chunkstart;
1094             int n_invalid = index - invalidstart;
1095 
1096             qemu_put_be32(f, chunkstart);
1097             qemu_put_be16(f, n_valid);
1098             qemu_put_be16(f, n_invalid);
1099             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1100                             HASH_PTE_SIZE_64 * n_valid);
1101             sent += index - chunkstart;
1102 
1103             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1104                 break;
1105             }
1106         }
1107 
1108         if (examined >= htabslots) {
1109             break;
1110         }
1111 
1112         if (index >= htabslots) {
1113             assert(index == htabslots);
1114             index = 0;
1115         }
1116     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1117 
1118     if (index >= htabslots) {
1119         assert(index == htabslots);
1120         index = 0;
1121     }
1122 
1123     spapr->htab_save_index = index;
1124 
1125     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1126 }
1127 
1128 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1129 #define MAX_KVM_BUF_SIZE    2048
1130 
1131 static int htab_save_iterate(QEMUFile *f, void *opaque)
1132 {
1133     sPAPREnvironment *spapr = opaque;
1134     int rc = 0;
1135 
1136     /* Iteration header */
1137     qemu_put_be32(f, 0);
1138 
1139     if (!spapr->htab) {
1140         assert(kvm_enabled());
1141 
1142         rc = kvmppc_save_htab(f, spapr->htab_fd,
1143                               MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1144         if (rc < 0) {
1145             return rc;
1146         }
1147     } else  if (spapr->htab_first_pass) {
1148         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1149     } else {
1150         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1151     }
1152 
1153     /* End marker */
1154     qemu_put_be32(f, 0);
1155     qemu_put_be16(f, 0);
1156     qemu_put_be16(f, 0);
1157 
1158     return rc;
1159 }
1160 
1161 static int htab_save_complete(QEMUFile *f, void *opaque)
1162 {
1163     sPAPREnvironment *spapr = opaque;
1164 
1165     /* Iteration header */
1166     qemu_put_be32(f, 0);
1167 
1168     if (!spapr->htab) {
1169         int rc;
1170 
1171         assert(kvm_enabled());
1172 
1173         rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
1174         if (rc < 0) {
1175             return rc;
1176         }
1177         close(spapr->htab_fd);
1178         spapr->htab_fd = -1;
1179     } else {
1180         htab_save_later_pass(f, spapr, -1);
1181     }
1182 
1183     /* End marker */
1184     qemu_put_be32(f, 0);
1185     qemu_put_be16(f, 0);
1186     qemu_put_be16(f, 0);
1187 
1188     return 0;
1189 }
1190 
1191 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1192 {
1193     sPAPREnvironment *spapr = opaque;
1194     uint32_t section_hdr;
1195     int fd = -1;
1196 
1197     if (version_id < 1 || version_id > 1) {
1198         fprintf(stderr, "htab_load() bad version\n");
1199         return -EINVAL;
1200     }
1201 
1202     section_hdr = qemu_get_be32(f);
1203 
1204     if (section_hdr) {
1205         /* First section, just the hash shift */
1206         if (spapr->htab_shift != section_hdr) {
1207             return -EINVAL;
1208         }
1209         return 0;
1210     }
1211 
1212     if (!spapr->htab) {
1213         assert(kvm_enabled());
1214 
1215         fd = kvmppc_get_htab_fd(true);
1216         if (fd < 0) {
1217             fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1218                     strerror(errno));
1219         }
1220     }
1221 
1222     while (true) {
1223         uint32_t index;
1224         uint16_t n_valid, n_invalid;
1225 
1226         index = qemu_get_be32(f);
1227         n_valid = qemu_get_be16(f);
1228         n_invalid = qemu_get_be16(f);
1229 
1230         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1231             /* End of Stream */
1232             break;
1233         }
1234 
1235         if ((index + n_valid + n_invalid) >
1236             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1237             /* Bad index in stream */
1238             fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1239                     "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1240                     spapr->htab_shift);
1241             return -EINVAL;
1242         }
1243 
1244         if (spapr->htab) {
1245             if (n_valid) {
1246                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1247                                 HASH_PTE_SIZE_64 * n_valid);
1248             }
1249             if (n_invalid) {
1250                 memset(HPTE(spapr->htab, index + n_valid), 0,
1251                        HASH_PTE_SIZE_64 * n_invalid);
1252             }
1253         } else {
1254             int rc;
1255 
1256             assert(fd >= 0);
1257 
1258             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1259             if (rc < 0) {
1260                 return rc;
1261             }
1262         }
1263     }
1264 
1265     if (!spapr->htab) {
1266         assert(fd >= 0);
1267         close(fd);
1268     }
1269 
1270     return 0;
1271 }
1272 
1273 static SaveVMHandlers savevm_htab_handlers = {
1274     .save_live_setup = htab_save_setup,
1275     .save_live_iterate = htab_save_iterate,
1276     .save_live_complete = htab_save_complete,
1277     .load_state = htab_load,
1278 };
1279 
1280 /* pSeries LPAR / sPAPR hardware init */
1281 static void ppc_spapr_init(MachineState *machine)
1282 {
1283     ram_addr_t ram_size = machine->ram_size;
1284     const char *cpu_model = machine->cpu_model;
1285     const char *kernel_filename = machine->kernel_filename;
1286     const char *kernel_cmdline = machine->kernel_cmdline;
1287     const char *initrd_filename = machine->initrd_filename;
1288     const char *boot_device = machine->boot_order;
1289     PowerPCCPU *cpu;
1290     CPUPPCState *env;
1291     PCIHostState *phb;
1292     int i;
1293     MemoryRegion *sysmem = get_system_memory();
1294     MemoryRegion *ram = g_new(MemoryRegion, 1);
1295     hwaddr rma_alloc_size;
1296     hwaddr node0_size = (nb_numa_nodes > 1) ? numa_info[0].node_mem : ram_size;
1297     uint32_t initrd_base = 0;
1298     long kernel_size = 0, initrd_size = 0;
1299     long load_limit, rtas_limit, fw_size;
1300     bool kernel_le = false;
1301     char *filename;
1302 
1303     msi_supported = true;
1304 
1305     spapr = g_malloc0(sizeof(*spapr));
1306     QLIST_INIT(&spapr->phbs);
1307 
1308     cpu_ppc_hypercall = emulate_spapr_hypercall;
1309 
1310     /* Allocate RMA if necessary */
1311     rma_alloc_size = kvmppc_alloc_rma("ppc_spapr.rma", sysmem);
1312 
1313     if (rma_alloc_size == -1) {
1314         hw_error("qemu: Unable to create RMA\n");
1315         exit(1);
1316     }
1317 
1318     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1319         spapr->rma_size = rma_alloc_size;
1320     } else {
1321         spapr->rma_size = node0_size;
1322 
1323         /* With KVM, we don't actually know whether KVM supports an
1324          * unbounded RMA (PR KVM) or is limited by the hash table size
1325          * (HV KVM using VRMA), so we always assume the latter
1326          *
1327          * In that case, we also limit the initial allocations for RTAS
1328          * etc... to 256M since we have no way to know what the VRMA size
1329          * is going to be as it depends on the size of the hash table
1330          * isn't determined yet.
1331          */
1332         if (kvm_enabled()) {
1333             spapr->vrma_adjust = 1;
1334             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1335         }
1336     }
1337 
1338     if (spapr->rma_size > node0_size) {
1339         fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1340                 spapr->rma_size);
1341         exit(1);
1342     }
1343 
1344     /* We place the device tree and RTAS just below either the top of the RMA,
1345      * or just below 2GB, whichever is lowere, so that it can be
1346      * processed with 32-bit real mode code if necessary */
1347     rtas_limit = MIN(spapr->rma_size, 0x80000000);
1348     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1349     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1350     load_limit = spapr->fdt_addr - FW_OVERHEAD;
1351 
1352     /* We aim for a hash table of size 1/128 the size of RAM.  The
1353      * normal rule of thumb is 1/64 the size of RAM, but that's much
1354      * more than needed for the Linux guests we support. */
1355     spapr->htab_shift = 18; /* Minimum architected size */
1356     while (spapr->htab_shift <= 46) {
1357         if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
1358             break;
1359         }
1360         spapr->htab_shift++;
1361     }
1362 
1363     /* Set up Interrupt Controller before we create the VCPUs */
1364     spapr->icp = xics_system_init(smp_cpus * kvmppc_smt_threads() / smp_threads,
1365                                   XICS_IRQS);
1366     spapr->next_irq = XICS_IRQ_BASE;
1367 
1368     /* init CPUs */
1369     if (cpu_model == NULL) {
1370         cpu_model = kvm_enabled() ? "host" : "POWER7";
1371     }
1372     for (i = 0; i < smp_cpus; i++) {
1373         cpu = cpu_ppc_init(cpu_model);
1374         if (cpu == NULL) {
1375             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1376             exit(1);
1377         }
1378         env = &cpu->env;
1379 
1380         /* Set time-base frequency to 512 MHz */
1381         cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1382 
1383         /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1384          * MSR[IP] should never be set.
1385          */
1386         env->msr_mask &= ~(1 << 6);
1387 
1388         /* Tell KVM that we're in PAPR mode */
1389         if (kvm_enabled()) {
1390             kvmppc_set_papr(cpu);
1391         }
1392 
1393         if (cpu->max_compat) {
1394             if (ppc_set_compat(cpu, cpu->max_compat) < 0) {
1395                 exit(1);
1396             }
1397         }
1398 
1399         xics_cpu_setup(spapr->icp, cpu);
1400 
1401         qemu_register_reset(spapr_cpu_reset, cpu);
1402     }
1403 
1404     /* allocate RAM */
1405     spapr->ram_limit = ram_size;
1406     if (spapr->ram_limit > rma_alloc_size) {
1407         ram_addr_t nonrma_base = rma_alloc_size;
1408         ram_addr_t nonrma_size = spapr->ram_limit - rma_alloc_size;
1409 
1410         memory_region_init_ram(ram, NULL, "ppc_spapr.ram", nonrma_size);
1411         vmstate_register_ram_global(ram);
1412         memory_region_add_subregion(sysmem, nonrma_base, ram);
1413     }
1414 
1415     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1416     spapr->rtas_size = load_image_targphys(filename, spapr->rtas_addr,
1417                                            rtas_limit - spapr->rtas_addr);
1418     if (spapr->rtas_size < 0) {
1419         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1420         exit(1);
1421     }
1422     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1423         hw_error("RTAS too big ! 0x%lx bytes (max is 0x%x)\n",
1424                  spapr->rtas_size, RTAS_MAX_SIZE);
1425         exit(1);
1426     }
1427     g_free(filename);
1428 
1429     /* Set up EPOW events infrastructure */
1430     spapr_events_init(spapr);
1431 
1432     /* Set up VIO bus */
1433     spapr->vio_bus = spapr_vio_bus_init();
1434 
1435     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1436         if (serial_hds[i]) {
1437             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1438         }
1439     }
1440 
1441     /* We always have at least the nvram device on VIO */
1442     spapr_create_nvram(spapr);
1443 
1444     /* Set up PCI */
1445     spapr_pci_msi_init(spapr, SPAPR_PCI_MSI_WINDOW);
1446     spapr_pci_rtas_init();
1447 
1448     phb = spapr_create_phb(spapr, 0);
1449 
1450     for (i = 0; i < nb_nics; i++) {
1451         NICInfo *nd = &nd_table[i];
1452 
1453         if (!nd->model) {
1454             nd->model = g_strdup("ibmveth");
1455         }
1456 
1457         if (strcmp(nd->model, "ibmveth") == 0) {
1458             spapr_vlan_create(spapr->vio_bus, nd);
1459         } else {
1460             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1461         }
1462     }
1463 
1464     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1465         spapr_vscsi_create(spapr->vio_bus);
1466     }
1467 
1468     /* Graphics */
1469     if (spapr_vga_init(phb->bus)) {
1470         spapr->has_graphics = true;
1471     }
1472 
1473     if (usb_enabled(spapr->has_graphics)) {
1474         pci_create_simple(phb->bus, -1, "pci-ohci");
1475         if (spapr->has_graphics) {
1476             usbdevice_create("keyboard");
1477             usbdevice_create("mouse");
1478         }
1479     }
1480 
1481     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1482         fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1483                 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1484         exit(1);
1485     }
1486 
1487     if (kernel_filename) {
1488         uint64_t lowaddr = 0;
1489 
1490         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1491                                NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1492         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1493             kernel_size = load_elf(kernel_filename,
1494                                    translate_kernel_address, NULL,
1495                                    NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1496             kernel_le = kernel_size > 0;
1497         }
1498         if (kernel_size < 0) {
1499             fprintf(stderr, "qemu: error loading %s: %s\n",
1500                     kernel_filename, load_elf_strerror(kernel_size));
1501             exit(1);
1502         }
1503 
1504         /* load initrd */
1505         if (initrd_filename) {
1506             /* Try to locate the initrd in the gap between the kernel
1507              * and the firmware. Add a bit of space just in case
1508              */
1509             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1510             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1511                                               load_limit - initrd_base);
1512             if (initrd_size < 0) {
1513                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1514                         initrd_filename);
1515                 exit(1);
1516             }
1517         } else {
1518             initrd_base = 0;
1519             initrd_size = 0;
1520         }
1521     }
1522 
1523     if (bios_name == NULL) {
1524         bios_name = FW_FILE_NAME;
1525     }
1526     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1527     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1528     if (fw_size < 0) {
1529         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1530         exit(1);
1531     }
1532     g_free(filename);
1533 
1534     spapr->entry_point = 0x100;
1535 
1536     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1537     register_savevm_live(NULL, "spapr/htab", -1, 1,
1538                          &savevm_htab_handlers, spapr);
1539 
1540     /* Prepare the device tree */
1541     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1542                                             kernel_size, kernel_le,
1543                                             boot_device, kernel_cmdline,
1544                                             spapr->epow_irq);
1545     assert(spapr->fdt_skel != NULL);
1546 }
1547 
1548 static int spapr_kvm_type(const char *vm_type)
1549 {
1550     if (!vm_type) {
1551         return 0;
1552     }
1553 
1554     if (!strcmp(vm_type, "HV")) {
1555         return 1;
1556     }
1557 
1558     if (!strcmp(vm_type, "PR")) {
1559         return 2;
1560     }
1561 
1562     error_report("Unknown kvm-type specified '%s'", vm_type);
1563     exit(1);
1564 }
1565 
1566 /*
1567  * Implementation of an interface to adjust firmware patch
1568  * for the bootindex property handling.
1569  */
1570 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
1571                                    DeviceState *dev)
1572 {
1573 #define CAST(type, obj, name) \
1574     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
1575     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
1576     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
1577 
1578     if (d) {
1579         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
1580         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
1581         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
1582 
1583         if (spapr) {
1584             /*
1585              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
1586              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
1587              * in the top 16 bits of the 64-bit LUN
1588              */
1589             unsigned id = 0x8000 | (d->id << 8) | d->lun;
1590             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1591                                    (uint64_t)id << 48);
1592         } else if (virtio) {
1593             /*
1594              * We use SRP luns of the form 01000000 | (target << 8) | lun
1595              * in the top 32 bits of the 64-bit LUN
1596              * Note: the quote above is from SLOF and it is wrong,
1597              * the actual binding is:
1598              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
1599              */
1600             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
1601             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1602                                    (uint64_t)id << 32);
1603         } else if (usb) {
1604             /*
1605              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
1606              * in the top 32 bits of the 64-bit LUN
1607              */
1608             unsigned usb_port = atoi(usb->port->path);
1609             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
1610             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1611                                    (uint64_t)id << 32);
1612         }
1613     }
1614 
1615     if (phb) {
1616         /* Replace "pci" with "pci@800000020000000" */
1617         return g_strdup_printf("pci@%"PRIX64, phb->buid);
1618     }
1619 
1620     return NULL;
1621 }
1622 
1623 static char *spapr_get_kvm_type(Object *obj, Error **errp)
1624 {
1625     SPAPRMachine *sm = SPAPR_MACHINE(obj);
1626 
1627     return g_strdup(sm->kvm_type);
1628 }
1629 
1630 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
1631 {
1632     SPAPRMachine *sm = SPAPR_MACHINE(obj);
1633 
1634     g_free(sm->kvm_type);
1635     sm->kvm_type = g_strdup(value);
1636 }
1637 
1638 static void spapr_machine_initfn(Object *obj)
1639 {
1640     object_property_add_str(obj, "kvm-type",
1641                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
1642 }
1643 
1644 static void spapr_machine_class_init(ObjectClass *oc, void *data)
1645 {
1646     MachineClass *mc = MACHINE_CLASS(oc);
1647     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
1648 
1649     mc->name = "pseries";
1650     mc->desc = "pSeries Logical Partition (PAPR compliant)";
1651     mc->is_default = 1;
1652     mc->init = ppc_spapr_init;
1653     mc->reset = ppc_spapr_reset;
1654     mc->block_default_type = IF_SCSI;
1655     mc->max_cpus = MAX_CPUS;
1656     mc->no_parallel = 1;
1657     mc->default_boot_order = NULL;
1658     mc->kvm_type = spapr_kvm_type;
1659 
1660     fwc->get_dev_path = spapr_get_fw_dev_path;
1661 }
1662 
1663 static const TypeInfo spapr_machine_info = {
1664     .name          = TYPE_SPAPR_MACHINE,
1665     .parent        = TYPE_MACHINE,
1666     .instance_size = sizeof(SPAPRMachine),
1667     .instance_init = spapr_machine_initfn,
1668     .class_init    = spapr_machine_class_init,
1669     .interfaces = (InterfaceInfo[]) {
1670         { TYPE_FW_PATH_PROVIDER },
1671         { }
1672     },
1673 };
1674 
1675 static void spapr_machine_register_types(void)
1676 {
1677     type_register_static(&spapr_machine_info);
1678 }
1679 
1680 type_init(spapr_machine_register_types)
1681