xref: /openbmc/qemu/hw/ppc/spapr.c (revision 8c46f7ec85a4dd9663489b2fa2b425cd7b3653e1)
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     if (kvm_enabled()) {
367         _FDT((fdt_property_string(fdt, "hypervisor", "kvm")));
368     }
369 
370     /*
371      * Add info to guest to indentify which host is it being run on
372      * and what is the uuid of the guest
373      */
374     if (kvmppc_get_host_model(&buf)) {
375         _FDT((fdt_property_string(fdt, "host-model", buf)));
376         g_free(buf);
377     }
378     if (kvmppc_get_host_serial(&buf)) {
379         _FDT((fdt_property_string(fdt, "host-serial", buf)));
380         g_free(buf);
381     }
382 
383     buf = g_strdup_printf(UUID_FMT, qemu_uuid[0], qemu_uuid[1],
384                           qemu_uuid[2], qemu_uuid[3], qemu_uuid[4],
385                           qemu_uuid[5], qemu_uuid[6], qemu_uuid[7],
386                           qemu_uuid[8], qemu_uuid[9], qemu_uuid[10],
387                           qemu_uuid[11], qemu_uuid[12], qemu_uuid[13],
388                           qemu_uuid[14], qemu_uuid[15]);
389 
390     _FDT((fdt_property_string(fdt, "vm,uuid", buf)));
391     g_free(buf);
392 
393     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
394     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
395 
396     /* /chosen */
397     _FDT((fdt_begin_node(fdt, "chosen")));
398 
399     /* Set Form1_affinity */
400     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
401 
402     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
403     _FDT((fdt_property(fdt, "linux,initrd-start",
404                        &start_prop, sizeof(start_prop))));
405     _FDT((fdt_property(fdt, "linux,initrd-end",
406                        &end_prop, sizeof(end_prop))));
407     if (kernel_size) {
408         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
409                               cpu_to_be64(kernel_size) };
410 
411         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
412         if (little_endian) {
413             _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
414         }
415     }
416     if (boot_device) {
417         _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
418     }
419     if (boot_menu) {
420         _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
421     }
422     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
423     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
424     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
425 
426     _FDT((fdt_end_node(fdt)));
427 
428     /* cpus */
429     _FDT((fdt_begin_node(fdt, "cpus")));
430 
431     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
432     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
433 
434     CPU_FOREACH(cs) {
435         PowerPCCPU *cpu = POWERPC_CPU(cs);
436         CPUPPCState *env = &cpu->env;
437         DeviceClass *dc = DEVICE_GET_CLASS(cs);
438         PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
439         int index = ppc_get_vcpu_dt_id(cpu);
440         char *nodename;
441         uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
442                            0xffffffff, 0xffffffff};
443         uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
444         uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
445         uint32_t page_sizes_prop[64];
446         size_t page_sizes_prop_size;
447 
448         if ((index % smt) != 0) {
449             continue;
450         }
451 
452         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
453 
454         _FDT((fdt_begin_node(fdt, nodename)));
455 
456         g_free(nodename);
457 
458         _FDT((fdt_property_cell(fdt, "reg", index)));
459         _FDT((fdt_property_string(fdt, "device_type", "cpu")));
460 
461         _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
462         _FDT((fdt_property_cell(fdt, "d-cache-block-size",
463                                 env->dcache_line_size)));
464         _FDT((fdt_property_cell(fdt, "d-cache-line-size",
465                                 env->dcache_line_size)));
466         _FDT((fdt_property_cell(fdt, "i-cache-block-size",
467                                 env->icache_line_size)));
468         _FDT((fdt_property_cell(fdt, "i-cache-line-size",
469                                 env->icache_line_size)));
470 
471         if (pcc->l1_dcache_size) {
472             _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
473         } else {
474             fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
475         }
476         if (pcc->l1_icache_size) {
477             _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
478         } else {
479             fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
480         }
481 
482         _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
483         _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
484         _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
485         _FDT((fdt_property_string(fdt, "status", "okay")));
486         _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
487 
488         if (env->spr_cb[SPR_PURR].oea_read) {
489             _FDT((fdt_property(fdt, "ibm,purr", NULL, 0)));
490         }
491 
492         if (env->mmu_model & POWERPC_MMU_1TSEG) {
493             _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
494                                segs, sizeof(segs))));
495         }
496 
497         /* Advertise VMX/VSX (vector extensions) if available
498          *   0 / no property == no vector extensions
499          *   1               == VMX / Altivec available
500          *   2               == VSX available */
501         if (env->insns_flags & PPC_ALTIVEC) {
502             uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
503 
504             _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
505         }
506 
507         /* Advertise DFP (Decimal Floating Point) if available
508          *   0 / no property == no DFP
509          *   1               == DFP available */
510         if (env->insns_flags2 & PPC2_DFP) {
511             _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
512         }
513 
514         page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
515                                                       sizeof(page_sizes_prop));
516         if (page_sizes_prop_size) {
517             _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
518                                page_sizes_prop, page_sizes_prop_size)));
519         }
520 
521         _FDT((fdt_property_cell(fdt, "ibm,chip-id",
522                                 cs->cpu_index / cpus_per_socket)));
523 
524         _FDT((fdt_end_node(fdt)));
525     }
526 
527     _FDT((fdt_end_node(fdt)));
528 
529     /* RTAS */
530     _FDT((fdt_begin_node(fdt, "rtas")));
531 
532     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
533         add_str(hypertas, "hcall-multi-tce");
534     }
535     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
536                        hypertas->len)));
537     g_string_free(hypertas, TRUE);
538     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
539                        qemu_hypertas->len)));
540     g_string_free(qemu_hypertas, TRUE);
541 
542     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
543         refpoints, sizeof(refpoints))));
544 
545     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
546 
547     /*
548      * According to PAPR, rtas ibm,os-term, does not gaurantee a return
549      * back to the guest cpu.
550      *
551      * While an additional ibm,extended-os-term property indicates that
552      * rtas call return will always occur. Set this property.
553      */
554     _FDT((fdt_property(fdt, "ibm,extended-os-term", NULL, 0)));
555 
556     _FDT((fdt_end_node(fdt)));
557 
558     /* interrupt controller */
559     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
560 
561     _FDT((fdt_property_string(fdt, "device_type",
562                               "PowerPC-External-Interrupt-Presentation")));
563     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
564     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
565     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
566                        interrupt_server_ranges_prop,
567                        sizeof(interrupt_server_ranges_prop))));
568     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
569     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
570     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
571 
572     _FDT((fdt_end_node(fdt)));
573 
574     /* vdevice */
575     _FDT((fdt_begin_node(fdt, "vdevice")));
576 
577     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
578     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
579     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
580     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
581     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
582     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
583 
584     _FDT((fdt_end_node(fdt)));
585 
586     /* event-sources */
587     spapr_events_fdt_skel(fdt, epow_irq);
588 
589     /* /hypervisor node */
590     if (kvm_enabled()) {
591         uint8_t hypercall[16];
592 
593         /* indicate KVM hypercall interface */
594         _FDT((fdt_begin_node(fdt, "hypervisor")));
595         _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
596         if (kvmppc_has_cap_fixup_hcalls()) {
597             /*
598              * Older KVM versions with older guest kernels were broken with the
599              * magic page, don't allow the guest to map it.
600              */
601             kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
602                                  sizeof(hypercall));
603             _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
604                               sizeof(hypercall))));
605         }
606         _FDT((fdt_end_node(fdt)));
607     }
608 
609     _FDT((fdt_end_node(fdt))); /* close root node */
610     _FDT((fdt_finish(fdt)));
611 
612     return fdt;
613 }
614 
615 int spapr_h_cas_compose_response(target_ulong addr, target_ulong size)
616 {
617     void *fdt, *fdt_skel;
618     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
619 
620     size -= sizeof(hdr);
621 
622     /* Create sceleton */
623     fdt_skel = g_malloc0(size);
624     _FDT((fdt_create(fdt_skel, size)));
625     _FDT((fdt_begin_node(fdt_skel, "")));
626     _FDT((fdt_end_node(fdt_skel)));
627     _FDT((fdt_finish(fdt_skel)));
628     fdt = g_malloc0(size);
629     _FDT((fdt_open_into(fdt_skel, fdt, size)));
630     g_free(fdt_skel);
631 
632     /* Fix skeleton up */
633     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
634 
635     /* Pack resulting tree */
636     _FDT((fdt_pack(fdt)));
637 
638     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
639         trace_spapr_cas_failed(size);
640         return -1;
641     }
642 
643     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
644     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
645     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
646     g_free(fdt);
647 
648     return 0;
649 }
650 
651 static void spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
652                                        hwaddr size)
653 {
654     uint32_t associativity[] = {
655         cpu_to_be32(0x4), /* length */
656         cpu_to_be32(0x0), cpu_to_be32(0x0),
657         cpu_to_be32(0x0), cpu_to_be32(nodeid)
658     };
659     char mem_name[32];
660     uint64_t mem_reg_property[2];
661     int off;
662 
663     mem_reg_property[0] = cpu_to_be64(start);
664     mem_reg_property[1] = cpu_to_be64(size);
665 
666     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
667     off = fdt_add_subnode(fdt, 0, mem_name);
668     _FDT(off);
669     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
670     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
671                       sizeof(mem_reg_property))));
672     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
673                       sizeof(associativity))));
674 }
675 
676 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
677 {
678     hwaddr mem_start, node_size;
679     int i, nb_nodes = nb_numa_nodes;
680     NodeInfo *nodes = numa_info;
681     NodeInfo ramnode;
682 
683     /* No NUMA nodes, assume there is just one node with whole RAM */
684     if (!nb_numa_nodes) {
685         nb_nodes = 1;
686         ramnode.node_mem = ram_size;
687         nodes = &ramnode;
688     }
689 
690     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
691         if (!nodes[i].node_mem) {
692             continue;
693         }
694         if (mem_start >= ram_size) {
695             node_size = 0;
696         } else {
697             node_size = nodes[i].node_mem;
698             if (node_size > ram_size - mem_start) {
699                 node_size = ram_size - mem_start;
700             }
701         }
702         if (!mem_start) {
703             /* ppc_spapr_init() checks for rma_size <= node0_size already */
704             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
705             mem_start += spapr->rma_size;
706             node_size -= spapr->rma_size;
707         }
708         for ( ; node_size; ) {
709             hwaddr sizetmp = pow2floor(node_size);
710 
711             /* mem_start != 0 here */
712             if (ctzl(mem_start) < ctzl(sizetmp)) {
713                 sizetmp = 1ULL << ctzl(mem_start);
714             }
715 
716             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
717             node_size -= sizetmp;
718             mem_start += sizetmp;
719         }
720     }
721 
722     return 0;
723 }
724 
725 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
726                                hwaddr fdt_addr,
727                                hwaddr rtas_addr,
728                                hwaddr rtas_size)
729 {
730     int ret, i;
731     size_t cb = 0;
732     char *bootlist;
733     void *fdt;
734     sPAPRPHBState *phb;
735 
736     fdt = g_malloc(FDT_MAX_SIZE);
737 
738     /* open out the base tree into a temp buffer for the final tweaks */
739     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
740 
741     ret = spapr_populate_memory(spapr, fdt);
742     if (ret < 0) {
743         fprintf(stderr, "couldn't setup memory nodes in fdt\n");
744         exit(1);
745     }
746 
747     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
748     if (ret < 0) {
749         fprintf(stderr, "couldn't setup vio devices in fdt\n");
750         exit(1);
751     }
752 
753     QLIST_FOREACH(phb, &spapr->phbs, list) {
754         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
755     }
756 
757     if (ret < 0) {
758         fprintf(stderr, "couldn't setup PCI devices in fdt\n");
759         exit(1);
760     }
761 
762     /* RTAS */
763     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
764     if (ret < 0) {
765         fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
766     }
767 
768     /* Advertise NUMA via ibm,associativity */
769     ret = spapr_fixup_cpu_dt(fdt, spapr);
770     if (ret < 0) {
771         fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
772     }
773 
774     bootlist = get_boot_devices_list(&cb, true);
775     if (cb && bootlist) {
776         int offset = fdt_path_offset(fdt, "/chosen");
777         if (offset < 0) {
778             exit(1);
779         }
780         for (i = 0; i < cb; i++) {
781             if (bootlist[i] == '\n') {
782                 bootlist[i] = ' ';
783             }
784 
785         }
786         ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
787     }
788 
789     if (!spapr->has_graphics) {
790         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
791     }
792 
793     _FDT((fdt_pack(fdt)));
794 
795     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
796         hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
797                  fdt_totalsize(fdt), FDT_MAX_SIZE);
798         exit(1);
799     }
800 
801     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
802 
803     g_free(bootlist);
804     g_free(fdt);
805 }
806 
807 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
808 {
809     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
810 }
811 
812 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
813 {
814     CPUPPCState *env = &cpu->env;
815 
816     if (msr_pr) {
817         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
818         env->gpr[3] = H_PRIVILEGE;
819     } else {
820         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
821     }
822 }
823 
824 static void spapr_reset_htab(sPAPREnvironment *spapr)
825 {
826     long shift;
827 
828     /* allocate hash page table.  For now we always make this 16mb,
829      * later we should probably make it scale to the size of guest
830      * RAM */
831 
832     shift = kvmppc_reset_htab(spapr->htab_shift);
833 
834     if (shift > 0) {
835         /* Kernel handles htab, we don't need to allocate one */
836         spapr->htab_shift = shift;
837         kvmppc_kern_htab = true;
838     } else {
839         if (!spapr->htab) {
840             /* Allocate an htab if we don't yet have one */
841             spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
842         }
843 
844         /* And clear it */
845         memset(spapr->htab, 0, HTAB_SIZE(spapr));
846     }
847 
848     /* Update the RMA size if necessary */
849     if (spapr->vrma_adjust) {
850         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
851                                           spapr->htab_shift);
852     }
853 }
854 
855 static void ppc_spapr_reset(void)
856 {
857     PowerPCCPU *first_ppc_cpu;
858     uint32_t rtas_limit;
859 
860     /* Reset the hash table & recalc the RMA */
861     spapr_reset_htab(spapr);
862 
863     qemu_devices_reset();
864 
865     /*
866      * We place the device tree and RTAS just below either the top of the RMA,
867      * or just below 2GB, whichever is lowere, so that it can be
868      * processed with 32-bit real mode code if necessary
869      */
870     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
871     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
872     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
873 
874     /* Load the fdt */
875     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
876                        spapr->rtas_size);
877 
878     /* Copy RTAS over */
879     cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
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_UNUSED(4), /* used to be @next_irq */
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     MemoryRegion *rma_region;
1296     void *rma = NULL;
1297     hwaddr rma_alloc_size;
1298     hwaddr node0_size = spapr_node0_size();
1299     uint32_t initrd_base = 0;
1300     long kernel_size = 0, initrd_size = 0;
1301     long load_limit, fw_size;
1302     bool kernel_le = false;
1303     char *filename;
1304 
1305     msi_supported = true;
1306 
1307     spapr = g_malloc0(sizeof(*spapr));
1308     QLIST_INIT(&spapr->phbs);
1309 
1310     cpu_ppc_hypercall = emulate_spapr_hypercall;
1311 
1312     /* Allocate RMA if necessary */
1313     rma_alloc_size = kvmppc_alloc_rma(&rma);
1314 
1315     if (rma_alloc_size == -1) {
1316         hw_error("qemu: Unable to create RMA\n");
1317         exit(1);
1318     }
1319 
1320     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1321         spapr->rma_size = rma_alloc_size;
1322     } else {
1323         spapr->rma_size = node0_size;
1324 
1325         /* With KVM, we don't actually know whether KVM supports an
1326          * unbounded RMA (PR KVM) or is limited by the hash table size
1327          * (HV KVM using VRMA), so we always assume the latter
1328          *
1329          * In that case, we also limit the initial allocations for RTAS
1330          * etc... to 256M since we have no way to know what the VRMA size
1331          * is going to be as it depends on the size of the hash table
1332          * isn't determined yet.
1333          */
1334         if (kvm_enabled()) {
1335             spapr->vrma_adjust = 1;
1336             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1337         }
1338     }
1339 
1340     if (spapr->rma_size > node0_size) {
1341         fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1342                 spapr->rma_size);
1343         exit(1);
1344     }
1345 
1346     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1347     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1348 
1349     /* We aim for a hash table of size 1/128 the size of RAM.  The
1350      * normal rule of thumb is 1/64 the size of RAM, but that's much
1351      * more than needed for the Linux guests we support. */
1352     spapr->htab_shift = 18; /* Minimum architected size */
1353     while (spapr->htab_shift <= 46) {
1354         if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
1355             break;
1356         }
1357         spapr->htab_shift++;
1358     }
1359 
1360     /* Set up Interrupt Controller before we create the VCPUs */
1361     spapr->icp = xics_system_init(smp_cpus * kvmppc_smt_threads() / smp_threads,
1362                                   XICS_IRQS);
1363 
1364     /* init CPUs */
1365     if (cpu_model == NULL) {
1366         cpu_model = kvm_enabled() ? "host" : "POWER7";
1367     }
1368     for (i = 0; i < smp_cpus; i++) {
1369         cpu = cpu_ppc_init(cpu_model);
1370         if (cpu == NULL) {
1371             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1372             exit(1);
1373         }
1374         env = &cpu->env;
1375 
1376         /* Set time-base frequency to 512 MHz */
1377         cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1378 
1379         /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1380          * MSR[IP] should never be set.
1381          */
1382         env->msr_mask &= ~(1 << 6);
1383 
1384         /* Tell KVM that we're in PAPR mode */
1385         if (kvm_enabled()) {
1386             kvmppc_set_papr(cpu);
1387         }
1388 
1389         if (cpu->max_compat) {
1390             if (ppc_set_compat(cpu, cpu->max_compat) < 0) {
1391                 exit(1);
1392             }
1393         }
1394 
1395         xics_cpu_setup(spapr->icp, cpu);
1396 
1397         qemu_register_reset(spapr_cpu_reset, cpu);
1398     }
1399 
1400     /* allocate RAM */
1401     spapr->ram_limit = ram_size;
1402     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1403                                          spapr->ram_limit);
1404     memory_region_add_subregion(sysmem, 0, ram);
1405 
1406     if (rma_alloc_size && rma) {
1407         rma_region = g_new(MemoryRegion, 1);
1408         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1409                                    rma_alloc_size, rma);
1410         vmstate_register_ram_global(rma_region);
1411         memory_region_add_subregion(sysmem, 0, rma_region);
1412     }
1413 
1414     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1415     spapr->rtas_size = get_image_size(filename);
1416     spapr->rtas_blob = g_malloc(spapr->rtas_size);
1417     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1418         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1419         exit(1);
1420     }
1421     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1422         hw_error("RTAS too big ! 0x%zx bytes (max is 0x%x)\n",
1423                  spapr->rtas_size, RTAS_MAX_SIZE);
1424         exit(1);
1425     }
1426     g_free(filename);
1427 
1428     /* Set up EPOW events infrastructure */
1429     spapr_events_init(spapr);
1430 
1431     /* Set up VIO bus */
1432     spapr->vio_bus = spapr_vio_bus_init();
1433 
1434     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1435         if (serial_hds[i]) {
1436             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1437         }
1438     }
1439 
1440     /* We always have at least the nvram device on VIO */
1441     spapr_create_nvram(spapr);
1442 
1443     /* Set up PCI */
1444     spapr_pci_rtas_init();
1445 
1446     phb = spapr_create_phb(spapr, 0);
1447 
1448     for (i = 0; i < nb_nics; i++) {
1449         NICInfo *nd = &nd_table[i];
1450 
1451         if (!nd->model) {
1452             nd->model = g_strdup("ibmveth");
1453         }
1454 
1455         if (strcmp(nd->model, "ibmveth") == 0) {
1456             spapr_vlan_create(spapr->vio_bus, nd);
1457         } else {
1458             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1459         }
1460     }
1461 
1462     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1463         spapr_vscsi_create(spapr->vio_bus);
1464     }
1465 
1466     /* Graphics */
1467     if (spapr_vga_init(phb->bus)) {
1468         spapr->has_graphics = true;
1469     }
1470 
1471     if (usb_enabled(spapr->has_graphics)) {
1472         pci_create_simple(phb->bus, -1, "pci-ohci");
1473         if (spapr->has_graphics) {
1474             usbdevice_create("keyboard");
1475             usbdevice_create("mouse");
1476         }
1477     }
1478 
1479     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1480         fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1481                 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1482         exit(1);
1483     }
1484 
1485     if (kernel_filename) {
1486         uint64_t lowaddr = 0;
1487 
1488         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1489                                NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1490         if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1491             kernel_size = load_elf(kernel_filename,
1492                                    translate_kernel_address, NULL,
1493                                    NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1494             kernel_le = kernel_size > 0;
1495         }
1496         if (kernel_size < 0) {
1497             fprintf(stderr, "qemu: error loading %s: %s\n",
1498                     kernel_filename, load_elf_strerror(kernel_size));
1499             exit(1);
1500         }
1501 
1502         /* load initrd */
1503         if (initrd_filename) {
1504             /* Try to locate the initrd in the gap between the kernel
1505              * and the firmware. Add a bit of space just in case
1506              */
1507             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1508             initrd_size = load_image_targphys(initrd_filename, initrd_base,
1509                                               load_limit - initrd_base);
1510             if (initrd_size < 0) {
1511                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1512                         initrd_filename);
1513                 exit(1);
1514             }
1515         } else {
1516             initrd_base = 0;
1517             initrd_size = 0;
1518         }
1519     }
1520 
1521     if (bios_name == NULL) {
1522         bios_name = FW_FILE_NAME;
1523     }
1524     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1525     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1526     if (fw_size < 0) {
1527         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1528         exit(1);
1529     }
1530     g_free(filename);
1531 
1532     spapr->entry_point = 0x100;
1533 
1534     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1535     register_savevm_live(NULL, "spapr/htab", -1, 1,
1536                          &savevm_htab_handlers, spapr);
1537 
1538     /* Prepare the device tree */
1539     spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1540                                             kernel_size, kernel_le,
1541                                             boot_device, kernel_cmdline,
1542                                             spapr->epow_irq);
1543     assert(spapr->fdt_skel != NULL);
1544 }
1545 
1546 static int spapr_kvm_type(const char *vm_type)
1547 {
1548     if (!vm_type) {
1549         return 0;
1550     }
1551 
1552     if (!strcmp(vm_type, "HV")) {
1553         return 1;
1554     }
1555 
1556     if (!strcmp(vm_type, "PR")) {
1557         return 2;
1558     }
1559 
1560     error_report("Unknown kvm-type specified '%s'", vm_type);
1561     exit(1);
1562 }
1563 
1564 /*
1565  * Implementation of an interface to adjust firmware patch
1566  * for the bootindex property handling.
1567  */
1568 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
1569                                    DeviceState *dev)
1570 {
1571 #define CAST(type, obj, name) \
1572     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
1573     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
1574     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
1575 
1576     if (d) {
1577         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
1578         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
1579         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
1580 
1581         if (spapr) {
1582             /*
1583              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
1584              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
1585              * in the top 16 bits of the 64-bit LUN
1586              */
1587             unsigned id = 0x8000 | (d->id << 8) | d->lun;
1588             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1589                                    (uint64_t)id << 48);
1590         } else if (virtio) {
1591             /*
1592              * We use SRP luns of the form 01000000 | (target << 8) | lun
1593              * in the top 32 bits of the 64-bit LUN
1594              * Note: the quote above is from SLOF and it is wrong,
1595              * the actual binding is:
1596              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
1597              */
1598             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
1599             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1600                                    (uint64_t)id << 32);
1601         } else if (usb) {
1602             /*
1603              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
1604              * in the top 32 bits of the 64-bit LUN
1605              */
1606             unsigned usb_port = atoi(usb->port->path);
1607             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
1608             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1609                                    (uint64_t)id << 32);
1610         }
1611     }
1612 
1613     if (phb) {
1614         /* Replace "pci" with "pci@800000020000000" */
1615         return g_strdup_printf("pci@%"PRIX64, phb->buid);
1616     }
1617 
1618     return NULL;
1619 }
1620 
1621 static char *spapr_get_kvm_type(Object *obj, Error **errp)
1622 {
1623     sPAPRMachineState *sm = SPAPR_MACHINE(obj);
1624 
1625     return g_strdup(sm->kvm_type);
1626 }
1627 
1628 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
1629 {
1630     sPAPRMachineState *sm = SPAPR_MACHINE(obj);
1631 
1632     g_free(sm->kvm_type);
1633     sm->kvm_type = g_strdup(value);
1634 }
1635 
1636 static void spapr_machine_initfn(Object *obj)
1637 {
1638     object_property_add_str(obj, "kvm-type",
1639                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
1640 }
1641 
1642 static void ppc_cpu_do_nmi_on_cpu(void *arg)
1643 {
1644     CPUState *cs = arg;
1645 
1646     cpu_synchronize_state(cs);
1647     ppc_cpu_do_system_reset(cs);
1648 }
1649 
1650 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
1651 {
1652     CPUState *cs;
1653 
1654     CPU_FOREACH(cs) {
1655         async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, cs);
1656     }
1657 }
1658 
1659 static void spapr_machine_class_init(ObjectClass *oc, void *data)
1660 {
1661     MachineClass *mc = MACHINE_CLASS(oc);
1662     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
1663     NMIClass *nc = NMI_CLASS(oc);
1664 
1665     mc->name = "pseries";
1666     mc->desc = "pSeries Logical Partition (PAPR compliant)";
1667     mc->is_default = 1;
1668     mc->init = ppc_spapr_init;
1669     mc->reset = ppc_spapr_reset;
1670     mc->block_default_type = IF_SCSI;
1671     mc->max_cpus = MAX_CPUS;
1672     mc->no_parallel = 1;
1673     mc->default_boot_order = NULL;
1674     mc->kvm_type = spapr_kvm_type;
1675 
1676     fwc->get_dev_path = spapr_get_fw_dev_path;
1677     nc->nmi_monitor_handler = spapr_nmi;
1678 }
1679 
1680 static const TypeInfo spapr_machine_info = {
1681     .name          = TYPE_SPAPR_MACHINE,
1682     .parent        = TYPE_MACHINE,
1683     .instance_size = sizeof(sPAPRMachineState),
1684     .instance_init = spapr_machine_initfn,
1685     .class_init    = spapr_machine_class_init,
1686     .interfaces = (InterfaceInfo[]) {
1687         { TYPE_FW_PATH_PROVIDER },
1688         { TYPE_NMI },
1689         { }
1690     },
1691 };
1692 
1693 static void spapr_machine_2_1_class_init(ObjectClass *oc, void *data)
1694 {
1695     MachineClass *mc = MACHINE_CLASS(oc);
1696 
1697     mc->name = "pseries-2.1";
1698     mc->desc = "pSeries Logical Partition (PAPR compliant) v2.1";
1699     mc->is_default = 0;
1700 }
1701 
1702 static const TypeInfo spapr_machine_2_1_info = {
1703     .name          = TYPE_SPAPR_MACHINE "2.1",
1704     .parent        = TYPE_SPAPR_MACHINE,
1705     .class_init    = spapr_machine_2_1_class_init,
1706 };
1707 
1708 static void spapr_machine_register_types(void)
1709 {
1710     type_register_static(&spapr_machine_info);
1711     type_register_static(&spapr_machine_2_1_info);
1712 }
1713 
1714 type_init(spapr_machine_register_types)
1715