1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * SMP support for power macintosh. 4 * 5 * We support both the old "powersurge" SMP architecture 6 * and the current Core99 (G4 PowerMac) machines. 7 * 8 * Note that we don't support the very first rev. of 9 * Apple/DayStar 2 CPUs board, the one with the funky 10 * watchdog. Hopefully, none of these should be there except 11 * maybe internally to Apple. I should probably still add some 12 * code to detect this card though and disable SMP. --BenH. 13 * 14 * Support Macintosh G4 SMP by Troy Benjegerdes (hozer@drgw.net) 15 * and Ben Herrenschmidt <benh@kernel.crashing.org>. 16 * 17 * Support for DayStar quad CPU cards 18 * Copyright (C) XLR8, Inc. 1994-2000 19 */ 20 #include <linux/kernel.h> 21 #include <linux/sched.h> 22 #include <linux/sched/hotplug.h> 23 #include <linux/smp.h> 24 #include <linux/interrupt.h> 25 #include <linux/irqdomain.h> 26 #include <linux/kernel_stat.h> 27 #include <linux/delay.h> 28 #include <linux/init.h> 29 #include <linux/spinlock.h> 30 #include <linux/errno.h> 31 #include <linux/hardirq.h> 32 #include <linux/cpu.h> 33 #include <linux/compiler.h> 34 #include <linux/pgtable.h> 35 36 #include <asm/ptrace.h> 37 #include <linux/atomic.h> 38 #include <asm/code-patching.h> 39 #include <asm/irq.h> 40 #include <asm/page.h> 41 #include <asm/sections.h> 42 #include <asm/io.h> 43 #include <asm/smp.h> 44 #include <asm/machdep.h> 45 #include <asm/pmac_feature.h> 46 #include <asm/time.h> 47 #include <asm/mpic.h> 48 #include <asm/cacheflush.h> 49 #include <asm/keylargo.h> 50 #include <asm/pmac_low_i2c.h> 51 #include <asm/pmac_pfunc.h> 52 #include <asm/inst.h> 53 54 #include "pmac.h" 55 56 #undef DEBUG 57 58 #ifdef DEBUG 59 #define DBG(fmt...) udbg_printf(fmt) 60 #else 61 #define DBG(fmt...) 62 #endif 63 64 extern void __secondary_start_pmac_0(void); 65 66 static void (*pmac_tb_freeze)(int freeze); 67 static u64 timebase; 68 static int tb_req; 69 70 #ifdef CONFIG_PPC_PMAC32_PSURGE 71 72 /* 73 * Powersurge (old powermac SMP) support. 74 */ 75 76 /* Addresses for powersurge registers */ 77 #define HAMMERHEAD_BASE 0xf8000000 78 #define HHEAD_CONFIG 0x90 79 #define HHEAD_SEC_INTR 0xc0 80 81 /* register for interrupting the primary processor on the powersurge */ 82 /* N.B. this is actually the ethernet ROM! */ 83 #define PSURGE_PRI_INTR 0xf3019000 84 85 /* register for storing the start address for the secondary processor */ 86 /* N.B. this is the PCI config space address register for the 1st bridge */ 87 #define PSURGE_START 0xf2800000 88 89 /* Daystar/XLR8 4-CPU card */ 90 #define PSURGE_QUAD_REG_ADDR 0xf8800000 91 92 #define PSURGE_QUAD_IRQ_SET 0 93 #define PSURGE_QUAD_IRQ_CLR 1 94 #define PSURGE_QUAD_IRQ_PRIMARY 2 95 #define PSURGE_QUAD_CKSTOP_CTL 3 96 #define PSURGE_QUAD_PRIMARY_ARB 4 97 #define PSURGE_QUAD_BOARD_ID 6 98 #define PSURGE_QUAD_WHICH_CPU 7 99 #define PSURGE_QUAD_CKSTOP_RDBK 8 100 #define PSURGE_QUAD_RESET_CTL 11 101 102 #define PSURGE_QUAD_OUT(r, v) (out_8(quad_base + ((r) << 4) + 4, (v))) 103 #define PSURGE_QUAD_IN(r) (in_8(quad_base + ((r) << 4) + 4) & 0x0f) 104 #define PSURGE_QUAD_BIS(r, v) (PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) | (v))) 105 #define PSURGE_QUAD_BIC(r, v) (PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) & ~(v))) 106 107 /* virtual addresses for the above */ 108 static volatile u8 __iomem *hhead_base; 109 static volatile u8 __iomem *quad_base; 110 static volatile u32 __iomem *psurge_pri_intr; 111 static volatile u8 __iomem *psurge_sec_intr; 112 static volatile u32 __iomem *psurge_start; 113 114 /* values for psurge_type */ 115 #define PSURGE_NONE -1 116 #define PSURGE_DUAL 0 117 #define PSURGE_QUAD_OKEE 1 118 #define PSURGE_QUAD_COTTON 2 119 #define PSURGE_QUAD_ICEGRASS 3 120 121 /* what sort of powersurge board we have */ 122 static int psurge_type = PSURGE_NONE; 123 124 /* irq for secondary cpus to report */ 125 static struct irq_domain *psurge_host; 126 int psurge_secondary_virq; 127 128 /* 129 * Set and clear IPIs for powersurge. 130 */ 131 static inline void psurge_set_ipi(int cpu) 132 { 133 if (psurge_type == PSURGE_NONE) 134 return; 135 if (cpu == 0) 136 in_be32(psurge_pri_intr); 137 else if (psurge_type == PSURGE_DUAL) 138 out_8(psurge_sec_intr, 0); 139 else 140 PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu); 141 } 142 143 static inline void psurge_clr_ipi(int cpu) 144 { 145 if (cpu > 0) { 146 switch(psurge_type) { 147 case PSURGE_DUAL: 148 out_8(psurge_sec_intr, ~0); 149 break; 150 case PSURGE_NONE: 151 break; 152 default: 153 PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu); 154 } 155 } 156 } 157 158 /* 159 * On powersurge (old SMP powermac architecture) we don't have 160 * separate IPIs for separate messages like openpic does. Instead 161 * use the generic demux helpers 162 * -- paulus. 163 */ 164 static irqreturn_t psurge_ipi_intr(int irq, void *d) 165 { 166 psurge_clr_ipi(smp_processor_id()); 167 smp_ipi_demux(); 168 169 return IRQ_HANDLED; 170 } 171 172 static void smp_psurge_cause_ipi(int cpu) 173 { 174 psurge_set_ipi(cpu); 175 } 176 177 static int psurge_host_map(struct irq_domain *h, unsigned int virq, 178 irq_hw_number_t hw) 179 { 180 irq_set_chip_and_handler(virq, &dummy_irq_chip, handle_percpu_irq); 181 182 return 0; 183 } 184 185 static const struct irq_domain_ops psurge_host_ops = { 186 .map = psurge_host_map, 187 }; 188 189 static int __init psurge_secondary_ipi_init(void) 190 { 191 int rc = -ENOMEM; 192 193 psurge_host = irq_domain_add_nomap(NULL, ~0, &psurge_host_ops, NULL); 194 195 if (psurge_host) 196 psurge_secondary_virq = irq_create_direct_mapping(psurge_host); 197 198 if (psurge_secondary_virq) 199 rc = request_irq(psurge_secondary_virq, psurge_ipi_intr, 200 IRQF_PERCPU | IRQF_NO_THREAD, "IPI", NULL); 201 202 if (rc) 203 pr_err("Failed to setup secondary cpu IPI\n"); 204 205 return rc; 206 } 207 208 /* 209 * Determine a quad card presence. We read the board ID register, we 210 * force the data bus to change to something else, and we read it again. 211 * It it's stable, then the register probably exist (ugh !) 212 */ 213 static int __init psurge_quad_probe(void) 214 { 215 int type; 216 unsigned int i; 217 218 type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID); 219 if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS 220 || type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID)) 221 return PSURGE_DUAL; 222 223 /* looks OK, try a slightly more rigorous test */ 224 /* bogus is not necessarily cacheline-aligned, 225 though I don't suppose that really matters. -- paulus */ 226 for (i = 0; i < 100; i++) { 227 volatile u32 bogus[8]; 228 bogus[(0+i)%8] = 0x00000000; 229 bogus[(1+i)%8] = 0x55555555; 230 bogus[(2+i)%8] = 0xFFFFFFFF; 231 bogus[(3+i)%8] = 0xAAAAAAAA; 232 bogus[(4+i)%8] = 0x33333333; 233 bogus[(5+i)%8] = 0xCCCCCCCC; 234 bogus[(6+i)%8] = 0xCCCCCCCC; 235 bogus[(7+i)%8] = 0x33333333; 236 wmb(); 237 asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory"); 238 mb(); 239 if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID)) 240 return PSURGE_DUAL; 241 } 242 return type; 243 } 244 245 static void __init psurge_quad_init(void) 246 { 247 int procbits; 248 249 if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351); 250 procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU); 251 if (psurge_type == PSURGE_QUAD_ICEGRASS) 252 PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits); 253 else 254 PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits); 255 mdelay(33); 256 out_8(psurge_sec_intr, ~0); 257 PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits); 258 PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits); 259 if (psurge_type != PSURGE_QUAD_ICEGRASS) 260 PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits); 261 PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits); 262 mdelay(33); 263 PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits); 264 mdelay(33); 265 PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits); 266 mdelay(33); 267 } 268 269 static void __init smp_psurge_probe(void) 270 { 271 int i, ncpus; 272 struct device_node *dn; 273 274 /* 275 * The powersurge cpu board can be used in the generation 276 * of powermacs that have a socket for an upgradeable cpu card, 277 * including the 7500, 8500, 9500, 9600. 278 * The device tree doesn't tell you if you have 2 cpus because 279 * OF doesn't know anything about the 2nd processor. 280 * Instead we look for magic bits in magic registers, 281 * in the hammerhead memory controller in the case of the 282 * dual-cpu powersurge board. -- paulus. 283 */ 284 dn = of_find_node_by_name(NULL, "hammerhead"); 285 if (dn == NULL) 286 return; 287 of_node_put(dn); 288 289 hhead_base = ioremap(HAMMERHEAD_BASE, 0x800); 290 quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024); 291 psurge_sec_intr = hhead_base + HHEAD_SEC_INTR; 292 293 psurge_type = psurge_quad_probe(); 294 if (psurge_type != PSURGE_DUAL) { 295 psurge_quad_init(); 296 /* All released cards using this HW design have 4 CPUs */ 297 ncpus = 4; 298 /* No sure how timebase sync works on those, let's use SW */ 299 smp_ops->give_timebase = smp_generic_give_timebase; 300 smp_ops->take_timebase = smp_generic_take_timebase; 301 } else { 302 iounmap(quad_base); 303 if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) { 304 /* not a dual-cpu card */ 305 iounmap(hhead_base); 306 psurge_type = PSURGE_NONE; 307 return; 308 } 309 ncpus = 2; 310 } 311 312 if (psurge_secondary_ipi_init()) 313 return; 314 315 psurge_start = ioremap(PSURGE_START, 4); 316 psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4); 317 318 /* This is necessary because OF doesn't know about the 319 * secondary cpu(s), and thus there aren't nodes in the 320 * device tree for them, and smp_setup_cpu_maps hasn't 321 * set their bits in cpu_present_mask. 322 */ 323 if (ncpus > NR_CPUS) 324 ncpus = NR_CPUS; 325 for (i = 1; i < ncpus ; ++i) 326 set_cpu_present(i, true); 327 328 if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352); 329 } 330 331 static int __init smp_psurge_kick_cpu(int nr) 332 { 333 unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8; 334 unsigned long a, flags; 335 int i, j; 336 337 /* Defining this here is evil ... but I prefer hiding that 338 * crap to avoid giving people ideas that they can do the 339 * same. 340 */ 341 extern volatile unsigned int cpu_callin_map[NR_CPUS]; 342 343 /* may need to flush here if secondary bats aren't setup */ 344 for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32) 345 asm volatile("dcbf 0,%0" : : "r" (a) : "memory"); 346 asm volatile("sync"); 347 348 if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353); 349 350 /* This is going to freeze the timeebase, we disable interrupts */ 351 local_irq_save(flags); 352 353 out_be32(psurge_start, start); 354 mb(); 355 356 psurge_set_ipi(nr); 357 358 /* 359 * We can't use udelay here because the timebase is now frozen. 360 */ 361 for (i = 0; i < 2000; ++i) 362 asm volatile("nop" : : : "memory"); 363 psurge_clr_ipi(nr); 364 365 /* 366 * Also, because the timebase is frozen, we must not return to the 367 * caller which will try to do udelay's etc... Instead, we wait -here- 368 * for the CPU to callin. 369 */ 370 for (i = 0; i < 100000 && !cpu_callin_map[nr]; ++i) { 371 for (j = 1; j < 10000; j++) 372 asm volatile("nop" : : : "memory"); 373 asm volatile("sync" : : : "memory"); 374 } 375 if (!cpu_callin_map[nr]) 376 goto stuck; 377 378 /* And we do the TB sync here too for standard dual CPU cards */ 379 if (psurge_type == PSURGE_DUAL) { 380 while(!tb_req) 381 barrier(); 382 tb_req = 0; 383 mb(); 384 timebase = get_tb(); 385 mb(); 386 while (timebase) 387 barrier(); 388 mb(); 389 } 390 stuck: 391 /* now interrupt the secondary, restarting both TBs */ 392 if (psurge_type == PSURGE_DUAL) 393 psurge_set_ipi(1); 394 395 if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354); 396 397 return 0; 398 } 399 400 static void __init smp_psurge_setup_cpu(int cpu_nr) 401 { 402 unsigned long flags = IRQF_PERCPU | IRQF_NO_THREAD; 403 int irq; 404 405 if (cpu_nr != 0 || !psurge_start) 406 return; 407 408 /* reset the entry point so if we get another intr we won't 409 * try to startup again */ 410 out_be32(psurge_start, 0x100); 411 irq = irq_create_mapping(NULL, 30); 412 if (request_irq(irq, psurge_ipi_intr, flags, "primary IPI", NULL)) 413 printk(KERN_ERR "Couldn't get primary IPI interrupt"); 414 } 415 416 void __init smp_psurge_take_timebase(void) 417 { 418 if (psurge_type != PSURGE_DUAL) 419 return; 420 421 tb_req = 1; 422 mb(); 423 while (!timebase) 424 barrier(); 425 mb(); 426 set_tb(timebase >> 32, timebase & 0xffffffff); 427 timebase = 0; 428 mb(); 429 set_dec(tb_ticks_per_jiffy/2); 430 } 431 432 void __init smp_psurge_give_timebase(void) 433 { 434 /* Nothing to do here */ 435 } 436 437 /* PowerSurge-style Macs */ 438 struct smp_ops_t psurge_smp_ops = { 439 .message_pass = NULL, /* Use smp_muxed_ipi_message_pass */ 440 .cause_ipi = smp_psurge_cause_ipi, 441 .cause_nmi_ipi = NULL, 442 .probe = smp_psurge_probe, 443 .kick_cpu = smp_psurge_kick_cpu, 444 .setup_cpu = smp_psurge_setup_cpu, 445 .give_timebase = smp_psurge_give_timebase, 446 .take_timebase = smp_psurge_take_timebase, 447 }; 448 #endif /* CONFIG_PPC_PMAC32_PSURGE */ 449 450 /* 451 * Core 99 and later support 452 */ 453 454 455 static void smp_core99_give_timebase(void) 456 { 457 unsigned long flags; 458 459 local_irq_save(flags); 460 461 while(!tb_req) 462 barrier(); 463 tb_req = 0; 464 (*pmac_tb_freeze)(1); 465 mb(); 466 timebase = get_tb(); 467 mb(); 468 while (timebase) 469 barrier(); 470 mb(); 471 (*pmac_tb_freeze)(0); 472 mb(); 473 474 local_irq_restore(flags); 475 } 476 477 478 static void smp_core99_take_timebase(void) 479 { 480 unsigned long flags; 481 482 local_irq_save(flags); 483 484 tb_req = 1; 485 mb(); 486 while (!timebase) 487 barrier(); 488 mb(); 489 set_tb(timebase >> 32, timebase & 0xffffffff); 490 timebase = 0; 491 mb(); 492 493 local_irq_restore(flags); 494 } 495 496 #ifdef CONFIG_PPC64 497 /* 498 * G5s enable/disable the timebase via an i2c-connected clock chip. 499 */ 500 static struct pmac_i2c_bus *pmac_tb_clock_chip_host; 501 static u8 pmac_tb_pulsar_addr; 502 503 static void smp_core99_cypress_tb_freeze(int freeze) 504 { 505 u8 data; 506 int rc; 507 508 /* Strangely, the device-tree says address is 0xd2, but darwin 509 * accesses 0xd0 ... 510 */ 511 pmac_i2c_setmode(pmac_tb_clock_chip_host, 512 pmac_i2c_mode_combined); 513 rc = pmac_i2c_xfer(pmac_tb_clock_chip_host, 514 0xd0 | pmac_i2c_read, 515 1, 0x81, &data, 1); 516 if (rc != 0) 517 goto bail; 518 519 data = (data & 0xf3) | (freeze ? 0x00 : 0x0c); 520 521 pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub); 522 rc = pmac_i2c_xfer(pmac_tb_clock_chip_host, 523 0xd0 | pmac_i2c_write, 524 1, 0x81, &data, 1); 525 526 bail: 527 if (rc != 0) { 528 printk("Cypress Timebase %s rc: %d\n", 529 freeze ? "freeze" : "unfreeze", rc); 530 panic("Timebase freeze failed !\n"); 531 } 532 } 533 534 535 static void smp_core99_pulsar_tb_freeze(int freeze) 536 { 537 u8 data; 538 int rc; 539 540 pmac_i2c_setmode(pmac_tb_clock_chip_host, 541 pmac_i2c_mode_combined); 542 rc = pmac_i2c_xfer(pmac_tb_clock_chip_host, 543 pmac_tb_pulsar_addr | pmac_i2c_read, 544 1, 0x2e, &data, 1); 545 if (rc != 0) 546 goto bail; 547 548 data = (data & 0x88) | (freeze ? 0x11 : 0x22); 549 550 pmac_i2c_setmode(pmac_tb_clock_chip_host, pmac_i2c_mode_stdsub); 551 rc = pmac_i2c_xfer(pmac_tb_clock_chip_host, 552 pmac_tb_pulsar_addr | pmac_i2c_write, 553 1, 0x2e, &data, 1); 554 bail: 555 if (rc != 0) { 556 printk(KERN_ERR "Pulsar Timebase %s rc: %d\n", 557 freeze ? "freeze" : "unfreeze", rc); 558 panic("Timebase freeze failed !\n"); 559 } 560 } 561 562 static void __init smp_core99_setup_i2c_hwsync(int ncpus) 563 { 564 struct device_node *cc = NULL; 565 struct device_node *p; 566 const char *name = NULL; 567 const u32 *reg; 568 int ok; 569 570 /* Look for the clock chip */ 571 for_each_node_by_name(cc, "i2c-hwclock") { 572 p = of_get_parent(cc); 573 ok = p && of_device_is_compatible(p, "uni-n-i2c"); 574 of_node_put(p); 575 if (!ok) 576 continue; 577 578 pmac_tb_clock_chip_host = pmac_i2c_find_bus(cc); 579 if (pmac_tb_clock_chip_host == NULL) 580 continue; 581 reg = of_get_property(cc, "reg", NULL); 582 if (reg == NULL) 583 continue; 584 switch (*reg) { 585 case 0xd2: 586 if (of_device_is_compatible(cc,"pulsar-legacy-slewing")) { 587 pmac_tb_freeze = smp_core99_pulsar_tb_freeze; 588 pmac_tb_pulsar_addr = 0xd2; 589 name = "Pulsar"; 590 } else if (of_device_is_compatible(cc, "cy28508")) { 591 pmac_tb_freeze = smp_core99_cypress_tb_freeze; 592 name = "Cypress"; 593 } 594 break; 595 case 0xd4: 596 pmac_tb_freeze = smp_core99_pulsar_tb_freeze; 597 pmac_tb_pulsar_addr = 0xd4; 598 name = "Pulsar"; 599 break; 600 } 601 if (pmac_tb_freeze != NULL) 602 break; 603 } 604 if (pmac_tb_freeze != NULL) { 605 /* Open i2c bus for synchronous access */ 606 if (pmac_i2c_open(pmac_tb_clock_chip_host, 1)) { 607 printk(KERN_ERR "Failed top open i2c bus for clock" 608 " sync, fallback to software sync !\n"); 609 goto no_i2c_sync; 610 } 611 printk(KERN_INFO "Processor timebase sync using %s i2c clock\n", 612 name); 613 return; 614 } 615 no_i2c_sync: 616 pmac_tb_freeze = NULL; 617 pmac_tb_clock_chip_host = NULL; 618 } 619 620 621 622 /* 623 * Newer G5s uses a platform function 624 */ 625 626 static void smp_core99_pfunc_tb_freeze(int freeze) 627 { 628 struct device_node *cpus; 629 struct pmf_args args; 630 631 cpus = of_find_node_by_path("/cpus"); 632 BUG_ON(cpus == NULL); 633 args.count = 1; 634 args.u[0].v = !freeze; 635 pmf_call_function(cpus, "cpu-timebase", &args); 636 of_node_put(cpus); 637 } 638 639 #else /* CONFIG_PPC64 */ 640 641 /* 642 * SMP G4 use a GPIO to enable/disable the timebase. 643 */ 644 645 static unsigned int core99_tb_gpio; /* Timebase freeze GPIO */ 646 647 static void smp_core99_gpio_tb_freeze(int freeze) 648 { 649 if (freeze) 650 pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4); 651 else 652 pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0); 653 pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0); 654 } 655 656 657 #endif /* !CONFIG_PPC64 */ 658 659 static void core99_init_caches(int cpu) 660 { 661 #ifndef CONFIG_PPC64 662 /* L2 and L3 cache settings to pass from CPU0 to CPU1 on G4 cpus */ 663 static long int core99_l2_cache; 664 static long int core99_l3_cache; 665 666 if (!cpu_has_feature(CPU_FTR_L2CR)) 667 return; 668 669 if (cpu == 0) { 670 core99_l2_cache = _get_L2CR(); 671 printk("CPU0: L2CR is %lx\n", core99_l2_cache); 672 } else { 673 printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR()); 674 _set_L2CR(0); 675 _set_L2CR(core99_l2_cache); 676 printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache); 677 } 678 679 if (!cpu_has_feature(CPU_FTR_L3CR)) 680 return; 681 682 if (cpu == 0){ 683 core99_l3_cache = _get_L3CR(); 684 printk("CPU0: L3CR is %lx\n", core99_l3_cache); 685 } else { 686 printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR()); 687 _set_L3CR(0); 688 _set_L3CR(core99_l3_cache); 689 printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache); 690 } 691 #endif /* !CONFIG_PPC64 */ 692 } 693 694 static void __init smp_core99_setup(int ncpus) 695 { 696 #ifdef CONFIG_PPC64 697 698 /* i2c based HW sync on some G5s */ 699 if (of_machine_is_compatible("PowerMac7,2") || 700 of_machine_is_compatible("PowerMac7,3") || 701 of_machine_is_compatible("RackMac3,1")) 702 smp_core99_setup_i2c_hwsync(ncpus); 703 704 /* pfunc based HW sync on recent G5s */ 705 if (pmac_tb_freeze == NULL) { 706 struct device_node *cpus = 707 of_find_node_by_path("/cpus"); 708 if (cpus && 709 of_get_property(cpus, "platform-cpu-timebase", NULL)) { 710 pmac_tb_freeze = smp_core99_pfunc_tb_freeze; 711 printk(KERN_INFO "Processor timebase sync using" 712 " platform function\n"); 713 } 714 } 715 716 #else /* CONFIG_PPC64 */ 717 718 /* GPIO based HW sync on ppc32 Core99 */ 719 if (pmac_tb_freeze == NULL && !of_machine_is_compatible("MacRISC4")) { 720 struct device_node *cpu; 721 const u32 *tbprop = NULL; 722 723 core99_tb_gpio = KL_GPIO_TB_ENABLE; /* default value */ 724 cpu = of_find_node_by_type(NULL, "cpu"); 725 if (cpu != NULL) { 726 tbprop = of_get_property(cpu, "timebase-enable", NULL); 727 if (tbprop) 728 core99_tb_gpio = *tbprop; 729 of_node_put(cpu); 730 } 731 pmac_tb_freeze = smp_core99_gpio_tb_freeze; 732 printk(KERN_INFO "Processor timebase sync using" 733 " GPIO 0x%02x\n", core99_tb_gpio); 734 } 735 736 #endif /* CONFIG_PPC64 */ 737 738 /* No timebase sync, fallback to software */ 739 if (pmac_tb_freeze == NULL) { 740 smp_ops->give_timebase = smp_generic_give_timebase; 741 smp_ops->take_timebase = smp_generic_take_timebase; 742 printk(KERN_INFO "Processor timebase sync using software\n"); 743 } 744 745 #ifndef CONFIG_PPC64 746 { 747 int i; 748 749 /* XXX should get this from reg properties */ 750 for (i = 1; i < ncpus; ++i) 751 set_hard_smp_processor_id(i, i); 752 } 753 #endif 754 755 /* 32 bits SMP can't NAP */ 756 if (!of_machine_is_compatible("MacRISC4")) 757 powersave_nap = 0; 758 } 759 760 static void __init smp_core99_probe(void) 761 { 762 struct device_node *cpus; 763 int ncpus = 0; 764 765 if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345); 766 767 /* Count CPUs in the device-tree */ 768 for_each_node_by_type(cpus, "cpu") 769 ++ncpus; 770 771 printk(KERN_INFO "PowerMac SMP probe found %d cpus\n", ncpus); 772 773 /* Nothing more to do if less than 2 of them */ 774 if (ncpus <= 1) 775 return; 776 777 /* We need to perform some early initialisations before we can start 778 * setting up SMP as we are running before initcalls 779 */ 780 pmac_pfunc_base_install(); 781 pmac_i2c_init(); 782 783 /* Setup various bits like timebase sync method, ability to nap, ... */ 784 smp_core99_setup(ncpus); 785 786 /* Install IPIs */ 787 mpic_request_ipis(); 788 789 /* Collect l2cr and l3cr values from CPU 0 */ 790 core99_init_caches(0); 791 } 792 793 static int smp_core99_kick_cpu(int nr) 794 { 795 unsigned int save_vector; 796 unsigned long target, flags; 797 unsigned int *vector = (unsigned int *)(PAGE_OFFSET+0x100); 798 799 if (nr < 0 || nr > 3) 800 return -ENOENT; 801 802 if (ppc_md.progress) 803 ppc_md.progress("smp_core99_kick_cpu", 0x346); 804 805 local_irq_save(flags); 806 807 /* Save reset vector */ 808 save_vector = *vector; 809 810 /* Setup fake reset vector that does 811 * b __secondary_start_pmac_0 + nr*8 812 */ 813 target = (unsigned long) __secondary_start_pmac_0 + nr * 8; 814 patch_branch(vector, target, BRANCH_SET_LINK); 815 816 /* Put some life in our friend */ 817 pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0); 818 819 /* FIXME: We wait a bit for the CPU to take the exception, I should 820 * instead wait for the entry code to set something for me. Well, 821 * ideally, all that crap will be done in prom.c and the CPU left 822 * in a RAM-based wait loop like CHRP. 823 */ 824 mdelay(1); 825 826 /* Restore our exception vector */ 827 patch_instruction(vector, ppc_inst(save_vector)); 828 829 local_irq_restore(flags); 830 if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347); 831 832 return 0; 833 } 834 835 static void smp_core99_setup_cpu(int cpu_nr) 836 { 837 /* Setup L2/L3 */ 838 if (cpu_nr != 0) 839 core99_init_caches(cpu_nr); 840 841 /* Setup openpic */ 842 mpic_setup_this_cpu(); 843 } 844 845 #ifdef CONFIG_PPC64 846 #ifdef CONFIG_HOTPLUG_CPU 847 static unsigned int smp_core99_host_open; 848 849 static int smp_core99_cpu_prepare(unsigned int cpu) 850 { 851 int rc; 852 853 /* Open i2c bus if it was used for tb sync */ 854 if (pmac_tb_clock_chip_host && !smp_core99_host_open) { 855 rc = pmac_i2c_open(pmac_tb_clock_chip_host, 1); 856 if (rc) { 857 pr_err("Failed to open i2c bus for time sync\n"); 858 return notifier_from_errno(rc); 859 } 860 smp_core99_host_open = 1; 861 } 862 return 0; 863 } 864 865 static int smp_core99_cpu_online(unsigned int cpu) 866 { 867 /* Close i2c bus if it was used for tb sync */ 868 if (pmac_tb_clock_chip_host && smp_core99_host_open) { 869 pmac_i2c_close(pmac_tb_clock_chip_host); 870 smp_core99_host_open = 0; 871 } 872 return 0; 873 } 874 #endif /* CONFIG_HOTPLUG_CPU */ 875 876 static void __init smp_core99_bringup_done(void) 877 { 878 /* Close i2c bus if it was used for tb sync */ 879 if (pmac_tb_clock_chip_host) 880 pmac_i2c_close(pmac_tb_clock_chip_host); 881 882 /* If we didn't start the second CPU, we must take 883 * it off the bus. 884 */ 885 if (of_machine_is_compatible("MacRISC4") && 886 num_online_cpus() < 2) { 887 set_cpu_present(1, false); 888 g5_phy_disable_cpu1(); 889 } 890 #ifdef CONFIG_HOTPLUG_CPU 891 cpuhp_setup_state_nocalls(CPUHP_POWERPC_PMAC_PREPARE, 892 "powerpc/pmac:prepare", smp_core99_cpu_prepare, 893 NULL); 894 cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "powerpc/pmac:online", 895 smp_core99_cpu_online, NULL); 896 #endif 897 898 if (ppc_md.progress) 899 ppc_md.progress("smp_core99_bringup_done", 0x349); 900 } 901 #endif /* CONFIG_PPC64 */ 902 903 #ifdef CONFIG_HOTPLUG_CPU 904 905 static int smp_core99_cpu_disable(void) 906 { 907 int rc = generic_cpu_disable(); 908 if (rc) 909 return rc; 910 911 mpic_cpu_set_priority(0xf); 912 913 cleanup_cpu_mmu_context(); 914 915 return 0; 916 } 917 918 #ifdef CONFIG_PPC32 919 920 static void pmac_cpu_offline_self(void) 921 { 922 int cpu = smp_processor_id(); 923 924 local_irq_disable(); 925 idle_task_exit(); 926 pr_debug("CPU%d offline\n", cpu); 927 generic_set_cpu_dead(cpu); 928 smp_wmb(); 929 mb(); 930 low_cpu_offline_self(); 931 } 932 933 #else /* CONFIG_PPC32 */ 934 935 static void pmac_cpu_offline_self(void) 936 { 937 int cpu = smp_processor_id(); 938 939 local_irq_disable(); 940 idle_task_exit(); 941 942 /* 943 * turn off as much as possible, we'll be 944 * kicked out as this will only be invoked 945 * on core99 platforms for now ... 946 */ 947 948 printk(KERN_INFO "CPU#%d offline\n", cpu); 949 generic_set_cpu_dead(cpu); 950 smp_wmb(); 951 952 /* 953 * Re-enable interrupts. The NAP code needs to enable them 954 * anyways, do it now so we deal with the case where one already 955 * happened while soft-disabled. 956 * We shouldn't get any external interrupts, only decrementer, and the 957 * decrementer handler is safe for use on offline CPUs 958 */ 959 local_irq_enable(); 960 961 while (1) { 962 /* let's not take timer interrupts too often ... */ 963 set_dec(0x7fffffff); 964 965 /* Enter NAP mode */ 966 power4_idle(); 967 } 968 } 969 970 #endif /* else CONFIG_PPC32 */ 971 #endif /* CONFIG_HOTPLUG_CPU */ 972 973 /* Core99 Macs (dual G4s and G5s) */ 974 static struct smp_ops_t core99_smp_ops = { 975 .message_pass = smp_mpic_message_pass, 976 .probe = smp_core99_probe, 977 #ifdef CONFIG_PPC64 978 .bringup_done = smp_core99_bringup_done, 979 #endif 980 .kick_cpu = smp_core99_kick_cpu, 981 .setup_cpu = smp_core99_setup_cpu, 982 .give_timebase = smp_core99_give_timebase, 983 .take_timebase = smp_core99_take_timebase, 984 #if defined(CONFIG_HOTPLUG_CPU) 985 .cpu_disable = smp_core99_cpu_disable, 986 .cpu_die = generic_cpu_die, 987 #endif 988 }; 989 990 void __init pmac_setup_smp(void) 991 { 992 struct device_node *np; 993 994 /* Check for Core99 */ 995 np = of_find_node_by_name(NULL, "uni-n"); 996 if (!np) 997 np = of_find_node_by_name(NULL, "u3"); 998 if (!np) 999 np = of_find_node_by_name(NULL, "u4"); 1000 if (np) { 1001 of_node_put(np); 1002 smp_ops = &core99_smp_ops; 1003 } 1004 #ifdef CONFIG_PPC_PMAC32_PSURGE 1005 else { 1006 /* We have to set bits in cpu_possible_mask here since the 1007 * secondary CPU(s) aren't in the device tree. Various 1008 * things won't be initialized for CPUs not in the possible 1009 * map, so we really need to fix it up here. 1010 */ 1011 int cpu; 1012 1013 for (cpu = 1; cpu < 4 && cpu < NR_CPUS; ++cpu) 1014 set_cpu_possible(cpu, true); 1015 smp_ops = &psurge_smp_ops; 1016 } 1017 #endif /* CONFIG_PPC_PMAC32_PSURGE */ 1018 1019 #ifdef CONFIG_HOTPLUG_CPU 1020 smp_ops->cpu_offline_self = pmac_cpu_offline_self; 1021 #endif 1022 } 1023 1024 1025