1 /* 2 * This program is free software; you can redistribute it and/or 3 * modify it under the terms of the GNU General Public License 4 * as published by the Free Software Foundation; either version 2 5 * of the License, or (at your option) any later version. 6 * 7 * This program is distributed in the hope that it will be useful, 8 * but WITHOUT ANY WARRANTY; without even the implied warranty of 9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 10 * GNU General Public License for more details. 11 * 12 * You should have received a copy of the GNU General Public License 13 * along with this program; if not, write to the Free Software 14 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. 15 * 16 * Copyright (C) 2000, 2001 Kanoj Sarcar 17 * Copyright (C) 2000, 2001 Ralf Baechle 18 * Copyright (C) 2000, 2001 Silicon Graphics, Inc. 19 * Copyright (C) 2000, 2001, 2003 Broadcom Corporation 20 */ 21 #include <linux/cache.h> 22 #include <linux/delay.h> 23 #include <linux/init.h> 24 #include <linux/interrupt.h> 25 #include <linux/smp.h> 26 #include <linux/spinlock.h> 27 #include <linux/threads.h> 28 #include <linux/module.h> 29 #include <linux/time.h> 30 #include <linux/timex.h> 31 #include <linux/sched.h> 32 #include <linux/cpumask.h> 33 #include <linux/cpu.h> 34 #include <linux/err.h> 35 #include <linux/ftrace.h> 36 37 #include <linux/atomic.h> 38 #include <asm/cpu.h> 39 #include <asm/processor.h> 40 #include <asm/idle.h> 41 #include <asm/r4k-timer.h> 42 #include <asm/mmu_context.h> 43 #include <asm/time.h> 44 #include <asm/setup.h> 45 46 volatile cpumask_t cpu_callin_map; /* Bitmask of started secondaries */ 47 48 int __cpu_number_map[NR_CPUS]; /* Map physical to logical */ 49 EXPORT_SYMBOL(__cpu_number_map); 50 51 int __cpu_logical_map[NR_CPUS]; /* Map logical to physical */ 52 EXPORT_SYMBOL(__cpu_logical_map); 53 54 /* Number of TCs (or siblings in Intel speak) per CPU core */ 55 int smp_num_siblings = 1; 56 EXPORT_SYMBOL(smp_num_siblings); 57 58 /* representing the TCs (or siblings in Intel speak) of each logical CPU */ 59 cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly; 60 EXPORT_SYMBOL(cpu_sibling_map); 61 62 /* representing cpus for which sibling maps can be computed */ 63 static cpumask_t cpu_sibling_setup_map; 64 65 cpumask_t cpu_coherent_mask; 66 67 static inline void set_cpu_sibling_map(int cpu) 68 { 69 int i; 70 71 cpu_set(cpu, cpu_sibling_setup_map); 72 73 if (smp_num_siblings > 1) { 74 for_each_cpu_mask(i, cpu_sibling_setup_map) { 75 if (cpu_data[cpu].core == cpu_data[i].core) { 76 cpu_set(i, cpu_sibling_map[cpu]); 77 cpu_set(cpu, cpu_sibling_map[i]); 78 } 79 } 80 } else 81 cpu_set(cpu, cpu_sibling_map[cpu]); 82 } 83 84 struct plat_smp_ops *mp_ops; 85 EXPORT_SYMBOL(mp_ops); 86 87 void register_smp_ops(struct plat_smp_ops *ops) 88 { 89 if (mp_ops) 90 printk(KERN_WARNING "Overriding previously set SMP ops\n"); 91 92 mp_ops = ops; 93 } 94 95 /* 96 * First C code run on the secondary CPUs after being started up by 97 * the master. 98 */ 99 asmlinkage void start_secondary(void) 100 { 101 unsigned int cpu; 102 103 cpu_probe(); 104 cpu_report(); 105 per_cpu_trap_init(false); 106 mips_clockevent_init(); 107 mp_ops->init_secondary(); 108 109 /* 110 * XXX parity protection should be folded in here when it's converted 111 * to an option instead of something based on .cputype 112 */ 113 114 calibrate_delay(); 115 preempt_disable(); 116 cpu = smp_processor_id(); 117 cpu_data[cpu].udelay_val = loops_per_jiffy; 118 119 cpu_set(cpu, cpu_coherent_mask); 120 notify_cpu_starting(cpu); 121 122 set_cpu_online(cpu, true); 123 124 set_cpu_sibling_map(cpu); 125 126 cpu_set(cpu, cpu_callin_map); 127 128 synchronise_count_slave(cpu); 129 130 /* 131 * irq will be enabled in ->smp_finish(), enabling it too early 132 * is dangerous. 133 */ 134 WARN_ON_ONCE(!irqs_disabled()); 135 mp_ops->smp_finish(); 136 137 cpu_startup_entry(CPUHP_ONLINE); 138 } 139 140 /* 141 * Call into both interrupt handlers, as we share the IPI for them 142 */ 143 void __irq_entry smp_call_function_interrupt(void) 144 { 145 irq_enter(); 146 generic_smp_call_function_interrupt(); 147 irq_exit(); 148 } 149 150 static void stop_this_cpu(void *dummy) 151 { 152 /* 153 * Remove this CPU: 154 */ 155 set_cpu_online(smp_processor_id(), false); 156 for (;;) { 157 if (cpu_wait) 158 (*cpu_wait)(); /* Wait if available. */ 159 } 160 } 161 162 void smp_send_stop(void) 163 { 164 smp_call_function(stop_this_cpu, NULL, 0); 165 } 166 167 void __init smp_cpus_done(unsigned int max_cpus) 168 { 169 } 170 171 /* called from main before smp_init() */ 172 void __init smp_prepare_cpus(unsigned int max_cpus) 173 { 174 init_new_context(current, &init_mm); 175 current_thread_info()->cpu = 0; 176 mp_ops->prepare_cpus(max_cpus); 177 set_cpu_sibling_map(0); 178 #ifndef CONFIG_HOTPLUG_CPU 179 init_cpu_present(cpu_possible_mask); 180 #endif 181 cpumask_copy(&cpu_coherent_mask, cpu_possible_mask); 182 } 183 184 /* preload SMP state for boot cpu */ 185 void smp_prepare_boot_cpu(void) 186 { 187 set_cpu_possible(0, true); 188 set_cpu_online(0, true); 189 cpu_set(0, cpu_callin_map); 190 } 191 192 int __cpu_up(unsigned int cpu, struct task_struct *tidle) 193 { 194 mp_ops->boot_secondary(cpu, tidle); 195 196 /* 197 * Trust is futile. We should really have timeouts ... 198 */ 199 while (!cpu_isset(cpu, cpu_callin_map)) 200 udelay(100); 201 202 synchronise_count_master(cpu); 203 return 0; 204 } 205 206 /* Not really SMP stuff ... */ 207 int setup_profiling_timer(unsigned int multiplier) 208 { 209 return 0; 210 } 211 212 static void flush_tlb_all_ipi(void *info) 213 { 214 local_flush_tlb_all(); 215 } 216 217 void flush_tlb_all(void) 218 { 219 on_each_cpu(flush_tlb_all_ipi, NULL, 1); 220 } 221 222 static void flush_tlb_mm_ipi(void *mm) 223 { 224 local_flush_tlb_mm((struct mm_struct *)mm); 225 } 226 227 /* 228 * Special Variant of smp_call_function for use by TLB functions: 229 * 230 * o No return value 231 * o collapses to normal function call on UP kernels 232 * o collapses to normal function call on systems with a single shared 233 * primary cache. 234 */ 235 static inline void smp_on_other_tlbs(void (*func) (void *info), void *info) 236 { 237 smp_call_function(func, info, 1); 238 } 239 240 static inline void smp_on_each_tlb(void (*func) (void *info), void *info) 241 { 242 preempt_disable(); 243 244 smp_on_other_tlbs(func, info); 245 func(info); 246 247 preempt_enable(); 248 } 249 250 /* 251 * The following tlb flush calls are invoked when old translations are 252 * being torn down, or pte attributes are changing. For single threaded 253 * address spaces, a new context is obtained on the current cpu, and tlb 254 * context on other cpus are invalidated to force a new context allocation 255 * at switch_mm time, should the mm ever be used on other cpus. For 256 * multithreaded address spaces, intercpu interrupts have to be sent. 257 * Another case where intercpu interrupts are required is when the target 258 * mm might be active on another cpu (eg debuggers doing the flushes on 259 * behalf of debugees, kswapd stealing pages from another process etc). 260 * Kanoj 07/00. 261 */ 262 263 void flush_tlb_mm(struct mm_struct *mm) 264 { 265 preempt_disable(); 266 267 if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) { 268 smp_on_other_tlbs(flush_tlb_mm_ipi, mm); 269 } else { 270 unsigned int cpu; 271 272 for_each_online_cpu(cpu) { 273 if (cpu != smp_processor_id() && cpu_context(cpu, mm)) 274 cpu_context(cpu, mm) = 0; 275 } 276 } 277 local_flush_tlb_mm(mm); 278 279 preempt_enable(); 280 } 281 282 struct flush_tlb_data { 283 struct vm_area_struct *vma; 284 unsigned long addr1; 285 unsigned long addr2; 286 }; 287 288 static void flush_tlb_range_ipi(void *info) 289 { 290 struct flush_tlb_data *fd = info; 291 292 local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2); 293 } 294 295 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned long end) 296 { 297 struct mm_struct *mm = vma->vm_mm; 298 299 preempt_disable(); 300 if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) { 301 struct flush_tlb_data fd = { 302 .vma = vma, 303 .addr1 = start, 304 .addr2 = end, 305 }; 306 307 smp_on_other_tlbs(flush_tlb_range_ipi, &fd); 308 } else { 309 unsigned int cpu; 310 311 for_each_online_cpu(cpu) { 312 if (cpu != smp_processor_id() && cpu_context(cpu, mm)) 313 cpu_context(cpu, mm) = 0; 314 } 315 } 316 local_flush_tlb_range(vma, start, end); 317 preempt_enable(); 318 } 319 320 static void flush_tlb_kernel_range_ipi(void *info) 321 { 322 struct flush_tlb_data *fd = info; 323 324 local_flush_tlb_kernel_range(fd->addr1, fd->addr2); 325 } 326 327 void flush_tlb_kernel_range(unsigned long start, unsigned long end) 328 { 329 struct flush_tlb_data fd = { 330 .addr1 = start, 331 .addr2 = end, 332 }; 333 334 on_each_cpu(flush_tlb_kernel_range_ipi, &fd, 1); 335 } 336 337 static void flush_tlb_page_ipi(void *info) 338 { 339 struct flush_tlb_data *fd = info; 340 341 local_flush_tlb_page(fd->vma, fd->addr1); 342 } 343 344 void flush_tlb_page(struct vm_area_struct *vma, unsigned long page) 345 { 346 preempt_disable(); 347 if ((atomic_read(&vma->vm_mm->mm_users) != 1) || (current->mm != vma->vm_mm)) { 348 struct flush_tlb_data fd = { 349 .vma = vma, 350 .addr1 = page, 351 }; 352 353 smp_on_other_tlbs(flush_tlb_page_ipi, &fd); 354 } else { 355 unsigned int cpu; 356 357 for_each_online_cpu(cpu) { 358 if (cpu != smp_processor_id() && cpu_context(cpu, vma->vm_mm)) 359 cpu_context(cpu, vma->vm_mm) = 0; 360 } 361 } 362 local_flush_tlb_page(vma, page); 363 preempt_enable(); 364 } 365 366 static void flush_tlb_one_ipi(void *info) 367 { 368 unsigned long vaddr = (unsigned long) info; 369 370 local_flush_tlb_one(vaddr); 371 } 372 373 void flush_tlb_one(unsigned long vaddr) 374 { 375 smp_on_each_tlb(flush_tlb_one_ipi, (void *) vaddr); 376 } 377 378 EXPORT_SYMBOL(flush_tlb_page); 379 EXPORT_SYMBOL(flush_tlb_one); 380 381 #if defined(CONFIG_KEXEC) 382 void (*dump_ipi_function_ptr)(void *) = NULL; 383 void dump_send_ipi(void (*dump_ipi_callback)(void *)) 384 { 385 int i; 386 int cpu = smp_processor_id(); 387 388 dump_ipi_function_ptr = dump_ipi_callback; 389 smp_mb(); 390 for_each_online_cpu(i) 391 if (i != cpu) 392 mp_ops->send_ipi_single(i, SMP_DUMP); 393 394 } 395 EXPORT_SYMBOL(dump_send_ipi); 396 #endif 397 398 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST 399 400 static DEFINE_PER_CPU(atomic_t, tick_broadcast_count); 401 static DEFINE_PER_CPU(struct call_single_data, tick_broadcast_csd); 402 403 void tick_broadcast(const struct cpumask *mask) 404 { 405 atomic_t *count; 406 struct call_single_data *csd; 407 int cpu; 408 409 for_each_cpu(cpu, mask) { 410 count = &per_cpu(tick_broadcast_count, cpu); 411 csd = &per_cpu(tick_broadcast_csd, cpu); 412 413 if (atomic_inc_return(count) == 1) 414 smp_call_function_single_async(cpu, csd); 415 } 416 } 417 418 static void tick_broadcast_callee(void *info) 419 { 420 int cpu = smp_processor_id(); 421 tick_receive_broadcast(); 422 atomic_set(&per_cpu(tick_broadcast_count, cpu), 0); 423 } 424 425 static int __init tick_broadcast_init(void) 426 { 427 struct call_single_data *csd; 428 int cpu; 429 430 for (cpu = 0; cpu < NR_CPUS; cpu++) { 431 csd = &per_cpu(tick_broadcast_csd, cpu); 432 csd->func = tick_broadcast_callee; 433 } 434 435 return 0; 436 } 437 early_initcall(tick_broadcast_init); 438 439 #endif /* CONFIG_GENERIC_CLOCKEVENTS_BROADCAST */ 440