1 /* 2 * QEMU PowerPC PowerNV machine model 3 * 4 * Copyright (c) 2016, IBM Corporation. 5 * 6 * This library is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU Lesser General Public 8 * License as published by the Free Software Foundation; either 9 * version 2 of the License, or (at your option) any later version. 10 * 11 * This library is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * Lesser General Public License for more details. 15 * 16 * You should have received a copy of the GNU Lesser General Public 17 * License along with this library; if not, see <http://www.gnu.org/licenses/>. 18 */ 19 20 #include "qemu/osdep.h" 21 #include "qapi/error.h" 22 #include "sysemu/sysemu.h" 23 #include "sysemu/numa.h" 24 #include "sysemu/cpus.h" 25 #include "hw/hw.h" 26 #include "target/ppc/cpu.h" 27 #include "qemu/log.h" 28 #include "hw/ppc/fdt.h" 29 #include "hw/ppc/ppc.h" 30 #include "hw/ppc/pnv.h" 31 #include "hw/ppc/pnv_core.h" 32 #include "hw/loader.h" 33 #include "exec/address-spaces.h" 34 #include "qemu/cutils.h" 35 #include "qapi/visitor.h" 36 #include "monitor/monitor.h" 37 #include "hw/intc/intc.h" 38 #include "hw/ipmi/ipmi.h" 39 40 #include "hw/ppc/xics.h" 41 #include "hw/ppc/pnv_xscom.h" 42 43 #include "hw/isa/isa.h" 44 #include "hw/char/serial.h" 45 #include "hw/timer/mc146818rtc.h" 46 47 #include <libfdt.h> 48 49 #define FDT_MAX_SIZE 0x00100000 50 51 #define FW_FILE_NAME "skiboot.lid" 52 #define FW_LOAD_ADDR 0x0 53 #define FW_MAX_SIZE 0x00400000 54 55 #define KERNEL_LOAD_ADDR 0x20000000 56 #define INITRD_LOAD_ADDR 0x40000000 57 58 /* 59 * On Power Systems E880 (POWER8), the max cpus (threads) should be : 60 * 4 * 4 sockets * 12 cores * 8 threads = 1536 61 * Let's make it 2^11 62 */ 63 #define MAX_CPUS 2048 64 65 /* 66 * Memory nodes are created by hostboot, one for each range of memory 67 * that has a different "affinity". In practice, it means one range 68 * per chip. 69 */ 70 static void powernv_populate_memory_node(void *fdt, int chip_id, hwaddr start, 71 hwaddr size) 72 { 73 char *mem_name; 74 uint64_t mem_reg_property[2]; 75 int off; 76 77 mem_reg_property[0] = cpu_to_be64(start); 78 mem_reg_property[1] = cpu_to_be64(size); 79 80 mem_name = g_strdup_printf("memory@%"HWADDR_PRIx, start); 81 off = fdt_add_subnode(fdt, 0, mem_name); 82 g_free(mem_name); 83 84 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory"))); 85 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property, 86 sizeof(mem_reg_property)))); 87 _FDT((fdt_setprop_cell(fdt, off, "ibm,chip-id", chip_id))); 88 } 89 90 static int get_cpus_node(void *fdt) 91 { 92 int cpus_offset = fdt_path_offset(fdt, "/cpus"); 93 94 if (cpus_offset < 0) { 95 cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"), 96 "cpus"); 97 if (cpus_offset) { 98 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1))); 99 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0))); 100 } 101 } 102 _FDT(cpus_offset); 103 return cpus_offset; 104 } 105 106 /* 107 * The PowerNV cores (and threads) need to use real HW ids and not an 108 * incremental index like it has been done on other platforms. This HW 109 * id is stored in the CPU PIR, it is used to create cpu nodes in the 110 * device tree, used in XSCOM to address cores and in interrupt 111 * servers. 112 */ 113 static void powernv_create_core_node(PnvChip *chip, PnvCore *pc, void *fdt) 114 { 115 CPUState *cs = CPU(DEVICE(pc->threads)); 116 DeviceClass *dc = DEVICE_GET_CLASS(cs); 117 PowerPCCPU *cpu = POWERPC_CPU(cs); 118 int smt_threads = CPU_CORE(pc)->nr_threads; 119 CPUPPCState *env = &cpu->env; 120 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs); 121 uint32_t servers_prop[smt_threads]; 122 int i; 123 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40), 124 0xffffffff, 0xffffffff}; 125 uint32_t tbfreq = PNV_TIMEBASE_FREQ; 126 uint32_t cpufreq = 1000000000; 127 uint32_t page_sizes_prop[64]; 128 size_t page_sizes_prop_size; 129 const uint8_t pa_features[] = { 24, 0, 130 0xf6, 0x3f, 0xc7, 0xc0, 0x80, 0xf0, 131 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 132 0x00, 0x00, 0x00, 0x00, 0x80, 0x00, 133 0x80, 0x00, 0x80, 0x00, 0x80, 0x00 }; 134 int offset; 135 char *nodename; 136 int cpus_offset = get_cpus_node(fdt); 137 138 nodename = g_strdup_printf("%s@%x", dc->fw_name, pc->pir); 139 offset = fdt_add_subnode(fdt, cpus_offset, nodename); 140 _FDT(offset); 141 g_free(nodename); 142 143 _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id", chip->chip_id))); 144 145 _FDT((fdt_setprop_cell(fdt, offset, "reg", pc->pir))); 146 _FDT((fdt_setprop_cell(fdt, offset, "ibm,pir", pc->pir))); 147 _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu"))); 148 149 _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR]))); 150 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size", 151 env->dcache_line_size))); 152 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size", 153 env->dcache_line_size))); 154 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size", 155 env->icache_line_size))); 156 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size", 157 env->icache_line_size))); 158 159 if (pcc->l1_dcache_size) { 160 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size", 161 pcc->l1_dcache_size))); 162 } else { 163 warn_report("Unknown L1 dcache size for cpu"); 164 } 165 if (pcc->l1_icache_size) { 166 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size", 167 pcc->l1_icache_size))); 168 } else { 169 warn_report("Unknown L1 icache size for cpu"); 170 } 171 172 _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq))); 173 _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq))); 174 _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr))); 175 _FDT((fdt_setprop_string(fdt, offset, "status", "okay"))); 176 _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0))); 177 178 if (env->spr_cb[SPR_PURR].oea_read) { 179 _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0))); 180 } 181 182 if (env->mmu_model & POWERPC_MMU_1TSEG) { 183 _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes", 184 segs, sizeof(segs)))); 185 } 186 187 /* Advertise VMX/VSX (vector extensions) if available 188 * 0 / no property == no vector extensions 189 * 1 == VMX / Altivec available 190 * 2 == VSX available */ 191 if (env->insns_flags & PPC_ALTIVEC) { 192 uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1; 193 194 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx))); 195 } 196 197 /* Advertise DFP (Decimal Floating Point) if available 198 * 0 / no property == no DFP 199 * 1 == DFP available */ 200 if (env->insns_flags2 & PPC2_DFP) { 201 _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1))); 202 } 203 204 page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop, 205 sizeof(page_sizes_prop)); 206 if (page_sizes_prop_size) { 207 _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes", 208 page_sizes_prop, page_sizes_prop_size))); 209 } 210 211 _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", 212 pa_features, sizeof(pa_features)))); 213 214 /* Build interrupt servers properties */ 215 for (i = 0; i < smt_threads; i++) { 216 servers_prop[i] = cpu_to_be32(pc->pir + i); 217 } 218 _FDT((fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s", 219 servers_prop, sizeof(servers_prop)))); 220 } 221 222 static void powernv_populate_icp(PnvChip *chip, void *fdt, uint32_t pir, 223 uint32_t nr_threads) 224 { 225 uint64_t addr = PNV_ICP_BASE(chip) | (pir << 12); 226 char *name; 227 const char compat[] = "IBM,power8-icp\0IBM,ppc-xicp"; 228 uint32_t irange[2], i, rsize; 229 uint64_t *reg; 230 int offset; 231 232 irange[0] = cpu_to_be32(pir); 233 irange[1] = cpu_to_be32(nr_threads); 234 235 rsize = sizeof(uint64_t) * 2 * nr_threads; 236 reg = g_malloc(rsize); 237 for (i = 0; i < nr_threads; i++) { 238 reg[i * 2] = cpu_to_be64(addr | ((pir + i) * 0x1000)); 239 reg[i * 2 + 1] = cpu_to_be64(0x1000); 240 } 241 242 name = g_strdup_printf("interrupt-controller@%"PRIX64, addr); 243 offset = fdt_add_subnode(fdt, 0, name); 244 _FDT(offset); 245 g_free(name); 246 247 _FDT((fdt_setprop(fdt, offset, "compatible", compat, sizeof(compat)))); 248 _FDT((fdt_setprop(fdt, offset, "reg", reg, rsize))); 249 _FDT((fdt_setprop_string(fdt, offset, "device_type", 250 "PowerPC-External-Interrupt-Presentation"))); 251 _FDT((fdt_setprop(fdt, offset, "interrupt-controller", NULL, 0))); 252 _FDT((fdt_setprop(fdt, offset, "ibm,interrupt-server-ranges", 253 irange, sizeof(irange)))); 254 _FDT((fdt_setprop_cell(fdt, offset, "#interrupt-cells", 1))); 255 _FDT((fdt_setprop_cell(fdt, offset, "#address-cells", 0))); 256 g_free(reg); 257 } 258 259 static int pnv_chip_lpc_offset(PnvChip *chip, void *fdt) 260 { 261 char *name; 262 int offset; 263 264 name = g_strdup_printf("/xscom@%" PRIx64 "/isa@%x", 265 (uint64_t) PNV_XSCOM_BASE(chip), PNV_XSCOM_LPC_BASE); 266 offset = fdt_path_offset(fdt, name); 267 g_free(name); 268 return offset; 269 } 270 271 static void powernv_populate_chip(PnvChip *chip, void *fdt) 272 { 273 PnvChipClass *pcc = PNV_CHIP_GET_CLASS(chip); 274 char *typename = pnv_core_typename(pcc->cpu_model); 275 size_t typesize = object_type_get_instance_size(typename); 276 int i; 277 278 pnv_xscom_populate(chip, fdt, 0); 279 280 /* The default LPC bus of a multichip system is on chip 0. It's 281 * recognized by the firmware (skiboot) using a "primary" 282 * property. 283 */ 284 if (chip->chip_id == 0x0) { 285 int lpc_offset = pnv_chip_lpc_offset(chip, fdt); 286 287 _FDT((fdt_setprop(fdt, lpc_offset, "primary", NULL, 0))); 288 } 289 290 for (i = 0; i < chip->nr_cores; i++) { 291 PnvCore *pnv_core = PNV_CORE(chip->cores + i * typesize); 292 293 powernv_create_core_node(chip, pnv_core, fdt); 294 295 /* Interrupt Control Presenters (ICP). One per core. */ 296 powernv_populate_icp(chip, fdt, pnv_core->pir, 297 CPU_CORE(pnv_core)->nr_threads); 298 } 299 300 if (chip->ram_size) { 301 powernv_populate_memory_node(fdt, chip->chip_id, chip->ram_start, 302 chip->ram_size); 303 } 304 g_free(typename); 305 } 306 307 static void powernv_populate_rtc(ISADevice *d, void *fdt, int lpc_off) 308 { 309 uint32_t io_base = d->ioport_id; 310 uint32_t io_regs[] = { 311 cpu_to_be32(1), 312 cpu_to_be32(io_base), 313 cpu_to_be32(2) 314 }; 315 char *name; 316 int node; 317 318 name = g_strdup_printf("%s@i%x", qdev_fw_name(DEVICE(d)), io_base); 319 node = fdt_add_subnode(fdt, lpc_off, name); 320 _FDT(node); 321 g_free(name); 322 323 _FDT((fdt_setprop(fdt, node, "reg", io_regs, sizeof(io_regs)))); 324 _FDT((fdt_setprop_string(fdt, node, "compatible", "pnpPNP,b00"))); 325 } 326 327 static void powernv_populate_serial(ISADevice *d, void *fdt, int lpc_off) 328 { 329 const char compatible[] = "ns16550\0pnpPNP,501"; 330 uint32_t io_base = d->ioport_id; 331 uint32_t io_regs[] = { 332 cpu_to_be32(1), 333 cpu_to_be32(io_base), 334 cpu_to_be32(8) 335 }; 336 char *name; 337 int node; 338 339 name = g_strdup_printf("%s@i%x", qdev_fw_name(DEVICE(d)), io_base); 340 node = fdt_add_subnode(fdt, lpc_off, name); 341 _FDT(node); 342 g_free(name); 343 344 _FDT((fdt_setprop(fdt, node, "reg", io_regs, sizeof(io_regs)))); 345 _FDT((fdt_setprop(fdt, node, "compatible", compatible, 346 sizeof(compatible)))); 347 348 _FDT((fdt_setprop_cell(fdt, node, "clock-frequency", 1843200))); 349 _FDT((fdt_setprop_cell(fdt, node, "current-speed", 115200))); 350 _FDT((fdt_setprop_cell(fdt, node, "interrupts", d->isairq[0]))); 351 _FDT((fdt_setprop_cell(fdt, node, "interrupt-parent", 352 fdt_get_phandle(fdt, lpc_off)))); 353 354 /* This is needed by Linux */ 355 _FDT((fdt_setprop_string(fdt, node, "device_type", "serial"))); 356 } 357 358 static void powernv_populate_ipmi_bt(ISADevice *d, void *fdt, int lpc_off) 359 { 360 const char compatible[] = "bt\0ipmi-bt"; 361 uint32_t io_base; 362 uint32_t io_regs[] = { 363 cpu_to_be32(1), 364 0, /* 'io_base' retrieved from the 'ioport' property of 'isa-ipmi-bt' */ 365 cpu_to_be32(3) 366 }; 367 uint32_t irq; 368 char *name; 369 int node; 370 371 io_base = object_property_get_int(OBJECT(d), "ioport", &error_fatal); 372 io_regs[1] = cpu_to_be32(io_base); 373 374 irq = object_property_get_int(OBJECT(d), "irq", &error_fatal); 375 376 name = g_strdup_printf("%s@i%x", qdev_fw_name(DEVICE(d)), io_base); 377 node = fdt_add_subnode(fdt, lpc_off, name); 378 _FDT(node); 379 g_free(name); 380 381 _FDT((fdt_setprop(fdt, node, "reg", io_regs, sizeof(io_regs)))); 382 _FDT((fdt_setprop(fdt, node, "compatible", compatible, 383 sizeof(compatible)))); 384 385 /* Mark it as reserved to avoid Linux trying to claim it */ 386 _FDT((fdt_setprop_string(fdt, node, "status", "reserved"))); 387 _FDT((fdt_setprop_cell(fdt, node, "interrupts", irq))); 388 _FDT((fdt_setprop_cell(fdt, node, "interrupt-parent", 389 fdt_get_phandle(fdt, lpc_off)))); 390 } 391 392 typedef struct ForeachPopulateArgs { 393 void *fdt; 394 int offset; 395 } ForeachPopulateArgs; 396 397 static int powernv_populate_isa_device(DeviceState *dev, void *opaque) 398 { 399 ForeachPopulateArgs *args = opaque; 400 ISADevice *d = ISA_DEVICE(dev); 401 402 if (object_dynamic_cast(OBJECT(dev), TYPE_MC146818_RTC)) { 403 powernv_populate_rtc(d, args->fdt, args->offset); 404 } else if (object_dynamic_cast(OBJECT(dev), TYPE_ISA_SERIAL)) { 405 powernv_populate_serial(d, args->fdt, args->offset); 406 } else if (object_dynamic_cast(OBJECT(dev), "isa-ipmi-bt")) { 407 powernv_populate_ipmi_bt(d, args->fdt, args->offset); 408 } else { 409 error_report("unknown isa device %s@i%x", qdev_fw_name(dev), 410 d->ioport_id); 411 } 412 413 return 0; 414 } 415 416 static void powernv_populate_isa(ISABus *bus, void *fdt, int lpc_offset) 417 { 418 ForeachPopulateArgs args = { 419 .fdt = fdt, 420 .offset = lpc_offset, 421 }; 422 423 /* ISA devices are not necessarily parented to the ISA bus so we 424 * can not use object_child_foreach() */ 425 qbus_walk_children(BUS(bus), powernv_populate_isa_device, 426 NULL, NULL, NULL, &args); 427 } 428 429 static void *powernv_create_fdt(MachineState *machine) 430 { 431 const char plat_compat[] = "qemu,powernv\0ibm,powernv"; 432 PnvMachineState *pnv = POWERNV_MACHINE(machine); 433 void *fdt; 434 char *buf; 435 int off; 436 int i; 437 int lpc_offset; 438 439 fdt = g_malloc0(FDT_MAX_SIZE); 440 _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE))); 441 442 /* Root node */ 443 _FDT((fdt_setprop_cell(fdt, 0, "#address-cells", 0x2))); 444 _FDT((fdt_setprop_cell(fdt, 0, "#size-cells", 0x2))); 445 _FDT((fdt_setprop_string(fdt, 0, "model", 446 "IBM PowerNV (emulated by qemu)"))); 447 _FDT((fdt_setprop(fdt, 0, "compatible", plat_compat, 448 sizeof(plat_compat)))); 449 450 buf = qemu_uuid_unparse_strdup(&qemu_uuid); 451 _FDT((fdt_setprop_string(fdt, 0, "vm,uuid", buf))); 452 if (qemu_uuid_set) { 453 _FDT((fdt_property_string(fdt, "system-id", buf))); 454 } 455 g_free(buf); 456 457 off = fdt_add_subnode(fdt, 0, "chosen"); 458 if (machine->kernel_cmdline) { 459 _FDT((fdt_setprop_string(fdt, off, "bootargs", 460 machine->kernel_cmdline))); 461 } 462 463 if (pnv->initrd_size) { 464 uint32_t start_prop = cpu_to_be32(pnv->initrd_base); 465 uint32_t end_prop = cpu_to_be32(pnv->initrd_base + pnv->initrd_size); 466 467 _FDT((fdt_setprop(fdt, off, "linux,initrd-start", 468 &start_prop, sizeof(start_prop)))); 469 _FDT((fdt_setprop(fdt, off, "linux,initrd-end", 470 &end_prop, sizeof(end_prop)))); 471 } 472 473 /* Populate device tree for each chip */ 474 for (i = 0; i < pnv->num_chips; i++) { 475 powernv_populate_chip(pnv->chips[i], fdt); 476 } 477 478 /* Populate ISA devices on chip 0 */ 479 lpc_offset = pnv_chip_lpc_offset(pnv->chips[0], fdt); 480 powernv_populate_isa(pnv->isa_bus, fdt, lpc_offset); 481 482 if (pnv->bmc) { 483 pnv_bmc_populate_sensors(pnv->bmc, fdt); 484 } 485 486 return fdt; 487 } 488 489 static void pnv_powerdown_notify(Notifier *n, void *opaque) 490 { 491 PnvMachineState *pnv = POWERNV_MACHINE(qdev_get_machine()); 492 493 if (pnv->bmc) { 494 pnv_bmc_powerdown(pnv->bmc); 495 } 496 } 497 498 static void ppc_powernv_reset(void) 499 { 500 MachineState *machine = MACHINE(qdev_get_machine()); 501 PnvMachineState *pnv = POWERNV_MACHINE(machine); 502 void *fdt; 503 Object *obj; 504 505 qemu_devices_reset(); 506 507 /* OpenPOWER systems have a BMC, which can be defined on the 508 * command line with: 509 * 510 * -device ipmi-bmc-sim,id=bmc0 511 * 512 * This is the internal simulator but it could also be an external 513 * BMC. 514 */ 515 obj = object_resolve_path_type("", "ipmi-bmc-sim", NULL); 516 if (obj) { 517 pnv->bmc = IPMI_BMC(obj); 518 } 519 520 fdt = powernv_create_fdt(machine); 521 522 /* Pack resulting tree */ 523 _FDT((fdt_pack(fdt))); 524 525 cpu_physical_memory_write(PNV_FDT_ADDR, fdt, fdt_totalsize(fdt)); 526 } 527 528 static ISABus *pnv_isa_create(PnvChip *chip) 529 { 530 PnvLpcController *lpc = &chip->lpc; 531 ISABus *isa_bus; 532 qemu_irq *irqs; 533 PnvChipClass *pcc = PNV_CHIP_GET_CLASS(chip); 534 535 /* let isa_bus_new() create its own bridge on SysBus otherwise 536 * devices speficied on the command line won't find the bus and 537 * will fail to create. 538 */ 539 isa_bus = isa_bus_new(NULL, &lpc->isa_mem, &lpc->isa_io, 540 &error_fatal); 541 542 irqs = pnv_lpc_isa_irq_create(lpc, pcc->chip_type, ISA_NUM_IRQS); 543 544 isa_bus_irqs(isa_bus, irqs); 545 return isa_bus; 546 } 547 548 static void ppc_powernv_init(MachineState *machine) 549 { 550 PnvMachineState *pnv = POWERNV_MACHINE(machine); 551 MemoryRegion *ram; 552 char *fw_filename; 553 long fw_size; 554 int i; 555 char *chip_typename; 556 557 /* allocate RAM */ 558 if (machine->ram_size < (1 * G_BYTE)) { 559 warn_report("skiboot may not work with < 1GB of RAM"); 560 } 561 562 ram = g_new(MemoryRegion, 1); 563 memory_region_allocate_system_memory(ram, NULL, "ppc_powernv.ram", 564 machine->ram_size); 565 memory_region_add_subregion(get_system_memory(), 0, ram); 566 567 /* load skiboot firmware */ 568 if (bios_name == NULL) { 569 bios_name = FW_FILE_NAME; 570 } 571 572 fw_filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name); 573 if (!fw_filename) { 574 error_report("Could not find OPAL firmware '%s'", bios_name); 575 exit(1); 576 } 577 578 fw_size = load_image_targphys(fw_filename, FW_LOAD_ADDR, FW_MAX_SIZE); 579 if (fw_size < 0) { 580 error_report("Could not load OPAL firmware '%s'", fw_filename); 581 exit(1); 582 } 583 g_free(fw_filename); 584 585 /* load kernel */ 586 if (machine->kernel_filename) { 587 long kernel_size; 588 589 kernel_size = load_image_targphys(machine->kernel_filename, 590 KERNEL_LOAD_ADDR, 0x2000000); 591 if (kernel_size < 0) { 592 error_report("Could not load kernel '%s'", 593 machine->kernel_filename); 594 exit(1); 595 } 596 } 597 598 /* load initrd */ 599 if (machine->initrd_filename) { 600 pnv->initrd_base = INITRD_LOAD_ADDR; 601 pnv->initrd_size = load_image_targphys(machine->initrd_filename, 602 pnv->initrd_base, 0x10000000); /* 128MB max */ 603 if (pnv->initrd_size < 0) { 604 error_report("Could not load initial ram disk '%s'", 605 machine->initrd_filename); 606 exit(1); 607 } 608 } 609 610 /* We need some cpu model to instantiate the PnvChip class */ 611 if (machine->cpu_model == NULL) { 612 machine->cpu_model = "POWER8"; 613 } 614 615 /* Create the processor chips */ 616 chip_typename = g_strdup_printf(TYPE_PNV_CHIP "-%s", machine->cpu_model); 617 if (!object_class_by_name(chip_typename)) { 618 error_report("invalid CPU model '%s' for %s machine", 619 machine->cpu_model, MACHINE_GET_CLASS(machine)->name); 620 exit(1); 621 } 622 623 pnv->chips = g_new0(PnvChip *, pnv->num_chips); 624 for (i = 0; i < pnv->num_chips; i++) { 625 char chip_name[32]; 626 Object *chip = object_new(chip_typename); 627 628 pnv->chips[i] = PNV_CHIP(chip); 629 630 /* TODO: put all the memory in one node on chip 0 until we find a 631 * way to specify different ranges for each chip 632 */ 633 if (i == 0) { 634 object_property_set_int(chip, machine->ram_size, "ram-size", 635 &error_fatal); 636 } 637 638 snprintf(chip_name, sizeof(chip_name), "chip[%d]", PNV_CHIP_HWID(i)); 639 object_property_add_child(OBJECT(pnv), chip_name, chip, &error_fatal); 640 object_property_set_int(chip, PNV_CHIP_HWID(i), "chip-id", 641 &error_fatal); 642 object_property_set_int(chip, smp_cores, "nr-cores", &error_fatal); 643 object_property_set_bool(chip, true, "realized", &error_fatal); 644 } 645 g_free(chip_typename); 646 647 /* Instantiate ISA bus on chip 0 */ 648 pnv->isa_bus = pnv_isa_create(pnv->chips[0]); 649 650 /* Create serial port */ 651 serial_hds_isa_init(pnv->isa_bus, 0, MAX_SERIAL_PORTS); 652 653 /* Create an RTC ISA device too */ 654 rtc_init(pnv->isa_bus, 2000, NULL); 655 656 /* OpenPOWER systems use a IPMI SEL Event message to notify the 657 * host to powerdown */ 658 pnv->powerdown_notifier.notify = pnv_powerdown_notify; 659 qemu_register_powerdown_notifier(&pnv->powerdown_notifier); 660 } 661 662 /* 663 * 0:21 Reserved - Read as zeros 664 * 22:24 Chip ID 665 * 25:28 Core number 666 * 29:31 Thread ID 667 */ 668 static uint32_t pnv_chip_core_pir_p8(PnvChip *chip, uint32_t core_id) 669 { 670 return (chip->chip_id << 7) | (core_id << 3); 671 } 672 673 /* 674 * 0:48 Reserved - Read as zeroes 675 * 49:52 Node ID 676 * 53:55 Chip ID 677 * 56 Reserved - Read as zero 678 * 57:61 Core number 679 * 62:63 Thread ID 680 * 681 * We only care about the lower bits. uint32_t is fine for the moment. 682 */ 683 static uint32_t pnv_chip_core_pir_p9(PnvChip *chip, uint32_t core_id) 684 { 685 return (chip->chip_id << 8) | (core_id << 2); 686 } 687 688 /* Allowed core identifiers on a POWER8 Processor Chip : 689 * 690 * <EX0 reserved> 691 * EX1 - Venice only 692 * EX2 - Venice only 693 * EX3 - Venice only 694 * EX4 695 * EX5 696 * EX6 697 * <EX7,8 reserved> <reserved> 698 * EX9 - Venice only 699 * EX10 - Venice only 700 * EX11 - Venice only 701 * EX12 702 * EX13 703 * EX14 704 * <EX15 reserved> 705 */ 706 #define POWER8E_CORE_MASK (0x7070ull) 707 #define POWER8_CORE_MASK (0x7e7eull) 708 709 /* 710 * POWER9 has 24 cores, ids starting at 0x20 711 */ 712 #define POWER9_CORE_MASK (0xffffff00000000ull) 713 714 static void pnv_chip_power8e_class_init(ObjectClass *klass, void *data) 715 { 716 DeviceClass *dc = DEVICE_CLASS(klass); 717 PnvChipClass *k = PNV_CHIP_CLASS(klass); 718 719 k->cpu_model = "POWER8E"; 720 k->chip_type = PNV_CHIP_POWER8E; 721 k->chip_cfam_id = 0x221ef04980000000ull; /* P8 Murano DD2.1 */ 722 k->cores_mask = POWER8E_CORE_MASK; 723 k->core_pir = pnv_chip_core_pir_p8; 724 k->xscom_base = 0x003fc0000000000ull; 725 k->xscom_core_base = 0x10000000ull; 726 dc->desc = "PowerNV Chip POWER8E"; 727 } 728 729 static const TypeInfo pnv_chip_power8e_info = { 730 .name = TYPE_PNV_CHIP_POWER8E, 731 .parent = TYPE_PNV_CHIP, 732 .instance_size = sizeof(PnvChip), 733 .class_init = pnv_chip_power8e_class_init, 734 }; 735 736 static void pnv_chip_power8_class_init(ObjectClass *klass, void *data) 737 { 738 DeviceClass *dc = DEVICE_CLASS(klass); 739 PnvChipClass *k = PNV_CHIP_CLASS(klass); 740 741 k->cpu_model = "POWER8"; 742 k->chip_type = PNV_CHIP_POWER8; 743 k->chip_cfam_id = 0x220ea04980000000ull; /* P8 Venice DD2.0 */ 744 k->cores_mask = POWER8_CORE_MASK; 745 k->core_pir = pnv_chip_core_pir_p8; 746 k->xscom_base = 0x003fc0000000000ull; 747 k->xscom_core_base = 0x10000000ull; 748 dc->desc = "PowerNV Chip POWER8"; 749 } 750 751 static const TypeInfo pnv_chip_power8_info = { 752 .name = TYPE_PNV_CHIP_POWER8, 753 .parent = TYPE_PNV_CHIP, 754 .instance_size = sizeof(PnvChip), 755 .class_init = pnv_chip_power8_class_init, 756 }; 757 758 static void pnv_chip_power8nvl_class_init(ObjectClass *klass, void *data) 759 { 760 DeviceClass *dc = DEVICE_CLASS(klass); 761 PnvChipClass *k = PNV_CHIP_CLASS(klass); 762 763 k->cpu_model = "POWER8NVL"; 764 k->chip_type = PNV_CHIP_POWER8NVL; 765 k->chip_cfam_id = 0x120d304980000000ull; /* P8 Naples DD1.0 */ 766 k->cores_mask = POWER8_CORE_MASK; 767 k->core_pir = pnv_chip_core_pir_p8; 768 k->xscom_base = 0x003fc0000000000ull; 769 k->xscom_core_base = 0x10000000ull; 770 dc->desc = "PowerNV Chip POWER8NVL"; 771 } 772 773 static const TypeInfo pnv_chip_power8nvl_info = { 774 .name = TYPE_PNV_CHIP_POWER8NVL, 775 .parent = TYPE_PNV_CHIP, 776 .instance_size = sizeof(PnvChip), 777 .class_init = pnv_chip_power8nvl_class_init, 778 }; 779 780 static void pnv_chip_power9_class_init(ObjectClass *klass, void *data) 781 { 782 DeviceClass *dc = DEVICE_CLASS(klass); 783 PnvChipClass *k = PNV_CHIP_CLASS(klass); 784 785 k->cpu_model = "POWER9"; 786 k->chip_type = PNV_CHIP_POWER9; 787 k->chip_cfam_id = 0x100d104980000000ull; /* P9 Nimbus DD1.0 */ 788 k->cores_mask = POWER9_CORE_MASK; 789 k->core_pir = pnv_chip_core_pir_p9; 790 k->xscom_base = 0x00603fc00000000ull; 791 k->xscom_core_base = 0x0ull; 792 dc->desc = "PowerNV Chip POWER9"; 793 } 794 795 static const TypeInfo pnv_chip_power9_info = { 796 .name = TYPE_PNV_CHIP_POWER9, 797 .parent = TYPE_PNV_CHIP, 798 .instance_size = sizeof(PnvChip), 799 .class_init = pnv_chip_power9_class_init, 800 }; 801 802 static void pnv_chip_core_sanitize(PnvChip *chip, Error **errp) 803 { 804 PnvChipClass *pcc = PNV_CHIP_GET_CLASS(chip); 805 int cores_max; 806 807 /* 808 * No custom mask for this chip, let's use the default one from * 809 * the chip class 810 */ 811 if (!chip->cores_mask) { 812 chip->cores_mask = pcc->cores_mask; 813 } 814 815 /* filter alien core ids ! some are reserved */ 816 if ((chip->cores_mask & pcc->cores_mask) != chip->cores_mask) { 817 error_setg(errp, "warning: invalid core mask for chip Ox%"PRIx64" !", 818 chip->cores_mask); 819 return; 820 } 821 chip->cores_mask &= pcc->cores_mask; 822 823 /* now that we have a sane layout, let check the number of cores */ 824 cores_max = ctpop64(chip->cores_mask); 825 if (chip->nr_cores > cores_max) { 826 error_setg(errp, "warning: too many cores for chip ! Limit is %d", 827 cores_max); 828 return; 829 } 830 } 831 832 static void pnv_chip_init(Object *obj) 833 { 834 PnvChip *chip = PNV_CHIP(obj); 835 PnvChipClass *pcc = PNV_CHIP_GET_CLASS(chip); 836 837 chip->xscom_base = pcc->xscom_base; 838 839 object_initialize(&chip->lpc, sizeof(chip->lpc), TYPE_PNV_LPC); 840 object_property_add_child(obj, "lpc", OBJECT(&chip->lpc), NULL); 841 842 object_initialize(&chip->psi, sizeof(chip->psi), TYPE_PNV_PSI); 843 object_property_add_child(obj, "psi", OBJECT(&chip->psi), NULL); 844 object_property_add_const_link(OBJECT(&chip->psi), "xics", 845 OBJECT(qdev_get_machine()), &error_abort); 846 847 object_initialize(&chip->occ, sizeof(chip->occ), TYPE_PNV_OCC); 848 object_property_add_child(obj, "occ", OBJECT(&chip->occ), NULL); 849 object_property_add_const_link(OBJECT(&chip->occ), "psi", 850 OBJECT(&chip->psi), &error_abort); 851 852 /* The LPC controller needs PSI to generate interrupts */ 853 object_property_add_const_link(OBJECT(&chip->lpc), "psi", 854 OBJECT(&chip->psi), &error_abort); 855 } 856 857 static void pnv_chip_icp_realize(PnvChip *chip, Error **errp) 858 { 859 PnvChipClass *pcc = PNV_CHIP_GET_CLASS(chip); 860 char *typename = pnv_core_typename(pcc->cpu_model); 861 size_t typesize = object_type_get_instance_size(typename); 862 int i, j; 863 char *name; 864 XICSFabric *xi = XICS_FABRIC(qdev_get_machine()); 865 866 name = g_strdup_printf("icp-%x", chip->chip_id); 867 memory_region_init(&chip->icp_mmio, OBJECT(chip), name, PNV_ICP_SIZE); 868 sysbus_init_mmio(SYS_BUS_DEVICE(chip), &chip->icp_mmio); 869 g_free(name); 870 871 sysbus_mmio_map(SYS_BUS_DEVICE(chip), 1, PNV_ICP_BASE(chip)); 872 873 /* Map the ICP registers for each thread */ 874 for (i = 0; i < chip->nr_cores; i++) { 875 PnvCore *pnv_core = PNV_CORE(chip->cores + i * typesize); 876 int core_hwid = CPU_CORE(pnv_core)->core_id; 877 878 for (j = 0; j < CPU_CORE(pnv_core)->nr_threads; j++) { 879 uint32_t pir = pcc->core_pir(chip, core_hwid) + j; 880 PnvICPState *icp = PNV_ICP(xics_icp_get(xi, pir)); 881 882 memory_region_add_subregion(&chip->icp_mmio, pir << 12, &icp->mmio); 883 } 884 } 885 886 g_free(typename); 887 } 888 889 static void pnv_chip_realize(DeviceState *dev, Error **errp) 890 { 891 PnvChip *chip = PNV_CHIP(dev); 892 Error *error = NULL; 893 PnvChipClass *pcc = PNV_CHIP_GET_CLASS(chip); 894 char *typename = pnv_core_typename(pcc->cpu_model); 895 size_t typesize = object_type_get_instance_size(typename); 896 int i, core_hwid; 897 898 if (!object_class_by_name(typename)) { 899 error_setg(errp, "Unable to find PowerNV CPU Core '%s'", typename); 900 return; 901 } 902 903 /* XSCOM bridge */ 904 pnv_xscom_realize(chip, &error); 905 if (error) { 906 error_propagate(errp, error); 907 return; 908 } 909 sysbus_mmio_map(SYS_BUS_DEVICE(chip), 0, PNV_XSCOM_BASE(chip)); 910 911 /* Cores */ 912 pnv_chip_core_sanitize(chip, &error); 913 if (error) { 914 error_propagate(errp, error); 915 return; 916 } 917 918 chip->cores = g_malloc0(typesize * chip->nr_cores); 919 920 for (i = 0, core_hwid = 0; (core_hwid < sizeof(chip->cores_mask) * 8) 921 && (i < chip->nr_cores); core_hwid++) { 922 char core_name[32]; 923 void *pnv_core = chip->cores + i * typesize; 924 925 if (!(chip->cores_mask & (1ull << core_hwid))) { 926 continue; 927 } 928 929 object_initialize(pnv_core, typesize, typename); 930 snprintf(core_name, sizeof(core_name), "core[%d]", core_hwid); 931 object_property_add_child(OBJECT(chip), core_name, OBJECT(pnv_core), 932 &error_fatal); 933 object_property_set_int(OBJECT(pnv_core), smp_threads, "nr-threads", 934 &error_fatal); 935 object_property_set_int(OBJECT(pnv_core), core_hwid, 936 CPU_CORE_PROP_CORE_ID, &error_fatal); 937 object_property_set_int(OBJECT(pnv_core), 938 pcc->core_pir(chip, core_hwid), 939 "pir", &error_fatal); 940 object_property_add_const_link(OBJECT(pnv_core), "xics", 941 qdev_get_machine(), &error_fatal); 942 object_property_set_bool(OBJECT(pnv_core), true, "realized", 943 &error_fatal); 944 object_unref(OBJECT(pnv_core)); 945 946 /* Each core has an XSCOM MMIO region */ 947 pnv_xscom_add_subregion(chip, 948 PNV_XSCOM_EX_CORE_BASE(pcc->xscom_core_base, 949 core_hwid), 950 &PNV_CORE(pnv_core)->xscom_regs); 951 i++; 952 } 953 g_free(typename); 954 955 /* Create LPC controller */ 956 object_property_set_bool(OBJECT(&chip->lpc), true, "realized", 957 &error_fatal); 958 pnv_xscom_add_subregion(chip, PNV_XSCOM_LPC_BASE, &chip->lpc.xscom_regs); 959 960 /* Interrupt Management Area. This is the memory region holding 961 * all the Interrupt Control Presenter (ICP) registers */ 962 pnv_chip_icp_realize(chip, &error); 963 if (error) { 964 error_propagate(errp, error); 965 return; 966 } 967 968 /* Processor Service Interface (PSI) Host Bridge */ 969 object_property_set_int(OBJECT(&chip->psi), PNV_PSIHB_BASE(chip), 970 "bar", &error_fatal); 971 object_property_set_bool(OBJECT(&chip->psi), true, "realized", &error); 972 if (error) { 973 error_propagate(errp, error); 974 return; 975 } 976 pnv_xscom_add_subregion(chip, PNV_XSCOM_PSIHB_BASE, &chip->psi.xscom_regs); 977 978 /* Create the simplified OCC model */ 979 object_property_set_bool(OBJECT(&chip->occ), true, "realized", &error); 980 if (error) { 981 error_propagate(errp, error); 982 return; 983 } 984 pnv_xscom_add_subregion(chip, PNV_XSCOM_OCC_BASE, &chip->occ.xscom_regs); 985 } 986 987 static Property pnv_chip_properties[] = { 988 DEFINE_PROP_UINT32("chip-id", PnvChip, chip_id, 0), 989 DEFINE_PROP_UINT64("ram-start", PnvChip, ram_start, 0), 990 DEFINE_PROP_UINT64("ram-size", PnvChip, ram_size, 0), 991 DEFINE_PROP_UINT32("nr-cores", PnvChip, nr_cores, 1), 992 DEFINE_PROP_UINT64("cores-mask", PnvChip, cores_mask, 0x0), 993 DEFINE_PROP_END_OF_LIST(), 994 }; 995 996 static void pnv_chip_class_init(ObjectClass *klass, void *data) 997 { 998 DeviceClass *dc = DEVICE_CLASS(klass); 999 1000 set_bit(DEVICE_CATEGORY_CPU, dc->categories); 1001 dc->realize = pnv_chip_realize; 1002 dc->props = pnv_chip_properties; 1003 dc->desc = "PowerNV Chip"; 1004 } 1005 1006 static const TypeInfo pnv_chip_info = { 1007 .name = TYPE_PNV_CHIP, 1008 .parent = TYPE_SYS_BUS_DEVICE, 1009 .class_init = pnv_chip_class_init, 1010 .instance_init = pnv_chip_init, 1011 .class_size = sizeof(PnvChipClass), 1012 .abstract = true, 1013 }; 1014 1015 static ICSState *pnv_ics_get(XICSFabric *xi, int irq) 1016 { 1017 PnvMachineState *pnv = POWERNV_MACHINE(xi); 1018 int i; 1019 1020 for (i = 0; i < pnv->num_chips; i++) { 1021 if (ics_valid_irq(&pnv->chips[i]->psi.ics, irq)) { 1022 return &pnv->chips[i]->psi.ics; 1023 } 1024 } 1025 return NULL; 1026 } 1027 1028 static void pnv_ics_resend(XICSFabric *xi) 1029 { 1030 PnvMachineState *pnv = POWERNV_MACHINE(xi); 1031 int i; 1032 1033 for (i = 0; i < pnv->num_chips; i++) { 1034 ics_resend(&pnv->chips[i]->psi.ics); 1035 } 1036 } 1037 1038 static PowerPCCPU *ppc_get_vcpu_by_pir(int pir) 1039 { 1040 CPUState *cs; 1041 1042 CPU_FOREACH(cs) { 1043 PowerPCCPU *cpu = POWERPC_CPU(cs); 1044 CPUPPCState *env = &cpu->env; 1045 1046 if (env->spr_cb[SPR_PIR].default_value == pir) { 1047 return cpu; 1048 } 1049 } 1050 1051 return NULL; 1052 } 1053 1054 static ICPState *pnv_icp_get(XICSFabric *xi, int pir) 1055 { 1056 PowerPCCPU *cpu = ppc_get_vcpu_by_pir(pir); 1057 1058 return cpu ? ICP(cpu->intc) : NULL; 1059 } 1060 1061 static void pnv_pic_print_info(InterruptStatsProvider *obj, 1062 Monitor *mon) 1063 { 1064 PnvMachineState *pnv = POWERNV_MACHINE(obj); 1065 int i; 1066 CPUState *cs; 1067 1068 CPU_FOREACH(cs) { 1069 PowerPCCPU *cpu = POWERPC_CPU(cs); 1070 1071 icp_pic_print_info(ICP(cpu->intc), mon); 1072 } 1073 1074 for (i = 0; i < pnv->num_chips; i++) { 1075 ics_pic_print_info(&pnv->chips[i]->psi.ics, mon); 1076 } 1077 } 1078 1079 static void pnv_get_num_chips(Object *obj, Visitor *v, const char *name, 1080 void *opaque, Error **errp) 1081 { 1082 visit_type_uint32(v, name, &POWERNV_MACHINE(obj)->num_chips, errp); 1083 } 1084 1085 static void pnv_set_num_chips(Object *obj, Visitor *v, const char *name, 1086 void *opaque, Error **errp) 1087 { 1088 PnvMachineState *pnv = POWERNV_MACHINE(obj); 1089 uint32_t num_chips; 1090 Error *local_err = NULL; 1091 1092 visit_type_uint32(v, name, &num_chips, &local_err); 1093 if (local_err) { 1094 error_propagate(errp, local_err); 1095 return; 1096 } 1097 1098 /* 1099 * TODO: should we decide on how many chips we can create based 1100 * on #cores and Venice vs. Murano vs. Naples chip type etc..., 1101 */ 1102 if (!is_power_of_2(num_chips) || num_chips > 4) { 1103 error_setg(errp, "invalid number of chips: '%d'", num_chips); 1104 return; 1105 } 1106 1107 pnv->num_chips = num_chips; 1108 } 1109 1110 static void powernv_machine_initfn(Object *obj) 1111 { 1112 PnvMachineState *pnv = POWERNV_MACHINE(obj); 1113 pnv->num_chips = 1; 1114 } 1115 1116 static void powernv_machine_class_props_init(ObjectClass *oc) 1117 { 1118 object_class_property_add(oc, "num-chips", "uint32", 1119 pnv_get_num_chips, pnv_set_num_chips, 1120 NULL, NULL, NULL); 1121 object_class_property_set_description(oc, "num-chips", 1122 "Specifies the number of processor chips", 1123 NULL); 1124 } 1125 1126 static void powernv_machine_class_init(ObjectClass *oc, void *data) 1127 { 1128 MachineClass *mc = MACHINE_CLASS(oc); 1129 XICSFabricClass *xic = XICS_FABRIC_CLASS(oc); 1130 InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc); 1131 1132 mc->desc = "IBM PowerNV (Non-Virtualized)"; 1133 mc->init = ppc_powernv_init; 1134 mc->reset = ppc_powernv_reset; 1135 mc->max_cpus = MAX_CPUS; 1136 mc->block_default_type = IF_IDE; /* Pnv provides a AHCI device for 1137 * storage */ 1138 mc->no_parallel = 1; 1139 mc->default_boot_order = NULL; 1140 mc->default_ram_size = 1 * G_BYTE; 1141 xic->icp_get = pnv_icp_get; 1142 xic->ics_get = pnv_ics_get; 1143 xic->ics_resend = pnv_ics_resend; 1144 ispc->print_info = pnv_pic_print_info; 1145 1146 powernv_machine_class_props_init(oc); 1147 } 1148 1149 static const TypeInfo powernv_machine_info = { 1150 .name = TYPE_POWERNV_MACHINE, 1151 .parent = TYPE_MACHINE, 1152 .instance_size = sizeof(PnvMachineState), 1153 .instance_init = powernv_machine_initfn, 1154 .class_init = powernv_machine_class_init, 1155 .interfaces = (InterfaceInfo[]) { 1156 { TYPE_XICS_FABRIC }, 1157 { TYPE_INTERRUPT_STATS_PROVIDER }, 1158 { }, 1159 }, 1160 }; 1161 1162 static void powernv_machine_register_types(void) 1163 { 1164 type_register_static(&powernv_machine_info); 1165 type_register_static(&pnv_chip_info); 1166 type_register_static(&pnv_chip_power8e_info); 1167 type_register_static(&pnv_chip_power8_info); 1168 type_register_static(&pnv_chip_power8nvl_info); 1169 type_register_static(&pnv_chip_power9_info); 1170 } 1171 1172 type_init(powernv_machine_register_types) 1173