arch/mips/kernel/process.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * Copyright (C) 1994 - 1999, 2000 by Ralf Baechle and others.
   7  * Copyright (C) 2005, 2006 by Ralf Baechle (ralf@linux-mips.org)
   8  * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
   9  * Copyright (C) 2004 Thiemo Seufer
  10  */
  11 #include <linux/errno.h>
  12 #include <linux/sched.h>
  13 #include <linux/tick.h>
  14 #include <linux/kernel.h>
  15 #include <linux/mm.h>
  16 #include <linux/stddef.h>
  17 #include <linux/unistd.h>
  18 #include <linux/ptrace.h>
  19 #include <linux/mman.h>
  20 #include <linux/personality.h>
  21 #include <linux/sys.h>
  22 #include <linux/user.h>
  23 #include <linux/init.h>
  24 #include <linux/completion.h>
  25 #include <linux/kallsyms.h>
  26 #include <linux/random.h>
  27
  28 #include <asm/asm.h>
  29 #include <asm/bootinfo.h>
  30 #include <asm/cpu.h>
  31 #include <asm/dsp.h>
  32 #include <asm/fpu.h>
  33 #include <asm/pgtable.h>
  34 #include <asm/system.h>
  35 #include <asm/mipsregs.h>
  36 #include <asm/processor.h>
  37 #include <asm/uaccess.h>
  38 #include <asm/io.h>
  39 #include <asm/elf.h>
  40 #include <asm/isadep.h>
  41 #include <asm/inst.h>
  42 #include <asm/stacktrace.h>
  43
  44 /*
  45  * The idle thread. There's no useful work to be done, so just try to conserve
  46  * power and have a low exit latency (ie sit in a loop waiting for somebody to
  47  * say that they'd like to reschedule)
  48  */
  49 void __noreturn cpu_idle(void)
  50 {
  51         int cpu;
  52
  53         /* CPU is going idle. */
  54         cpu = smp_processor_id();
  55
  56         /* endless idle loop with no priority at all */
  57         while (1) {
  58                 tick_nohz_stop_sched_tick(1);
  59                 while (!need_resched() && cpu_online(cpu)) {
  60 #ifdef CONFIG_MIPS_MT_SMTC
  61                         extern void smtc_idle_loop_hook(void);
  62
  63                         smtc_idle_loop_hook();
  64 #endif
  65
  66                         if (cpu_wait) {
  67                                 /* Don't trace irqs off for idle */
  68                                 stop_critical_timings();
  69                                 (*cpu_wait)();
  70                                 start_critical_timings();
  71                         }
  72                 }
  73 #ifdef CONFIG_HOTPLUG_CPU
  74                 if (!cpu_online(cpu) && !cpu_isset(cpu, cpu_callin_map) &&
  75                     (system_state == SYSTEM_RUNNING ||
  76                      system_state == SYSTEM_BOOTING))
  77                         play_dead();
  78 #endif
  79                 tick_nohz_restart_sched_tick();
  80                 preempt_enable_no_resched();
  81                 schedule();
  82                 preempt_disable();
  83         }
  84 }
  85
  86 asmlinkage void ret_from_fork(void);
  87
  88 void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
  89 {
  90         unsigned long status;
  91
  92         /* New thread loses kernel privileges. */
  93         status = regs->cp0_status & ~(ST0_CU0|ST0_CU1|ST0_FR|KU_MASK);
  94 #ifdef CONFIG_64BIT
  95         status |= test_thread_flag(TIF_32BIT_REGS) ? 0 : ST0_FR;
  96 #endif
  97         status |= KU_USER;
  98         regs->cp0_status = status;
  99         clear_used_math();
 100         clear_fpu_owner();
 101         if (cpu_has_dsp)
 102                 __init_dsp();
 103         regs->cp0_epc = pc;
 104         regs->regs[29] = sp;
 105 }
 106
 107 void exit_thread(void)
 108 {
 109 }
 110
 111 void flush_thread(void)
 112 {
 113 }
 114
 115 int copy_thread(unsigned long clone_flags, unsigned long usp,
 116         unsigned long unused, struct task_struct *p, struct pt_regs *regs)
 117 {
 118         struct thread_info *ti = task_thread_info(p);
 119         struct pt_regs *childregs;
 120         unsigned long childksp;
 121         p->set_child_tid = p->clear_child_tid = NULL;
 122
 123         childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
 124
 125         preempt_disable();
 126
 127         if (is_fpu_owner())
 128                 save_fp(p);
 129
 130         if (cpu_has_dsp)
 131                 save_dsp(p);
 132
 133         preempt_enable();
 134
 135         /* set up new TSS. */
 136         childregs = (struct pt_regs *) childksp - 1;
 137         /*  Put the stack after the struct pt_regs.  */
 138         childksp = (unsigned long) childregs;
 139         *childregs = *regs;
 140         childregs->regs[7] = 0; /* Clear error flag */
 141
 142         childregs->regs[2] = 0; /* Child gets zero as return value */
 143
 144         if (childregs->cp0_status & ST0_CU0) {
 145                 childregs->regs[28] = (unsigned long) ti;
 146                 childregs->regs[29] = childksp;
 147                 ti->addr_limit = KERNEL_DS;
 148         } else {
 149                 childregs->regs[29] = usp;
 150                 ti->addr_limit = USER_DS;
 151         }
 152         p->thread.reg29 = (unsigned long) childregs;
 153         p->thread.reg31 = (unsigned long) ret_from_fork;
 154
 155         /*
 156          * New tasks lose permission to use the fpu. This accelerates context
 157          * switching for most programs since they don't use the fpu.
 158          */
 159         p->thread.cp0_status = read_c0_status() & ~(ST0_CU2|ST0_CU1);
 160         childregs->cp0_status &= ~(ST0_CU2|ST0_CU1);
 161
 162 #ifdef CONFIG_MIPS_MT_SMTC
 163         /*
 164          * SMTC restores TCStatus after Status, and the CU bits
 165          * are aliased there.
 166          */
 167         childregs->cp0_tcstatus &= ~(ST0_CU2|ST0_CU1);
 168 #endif
 169         clear_tsk_thread_flag(p, TIF_USEDFPU);
 170
 171 #ifdef CONFIG_MIPS_MT_FPAFF
 172         clear_tsk_thread_flag(p, TIF_FPUBOUND);
 173 #endif /* CONFIG_MIPS_MT_FPAFF */
 174
 175         if (clone_flags & CLONE_SETTLS)
 176                 ti->tp_value = regs->regs[7];
 177
 178         return 0;
 179 }
 180
 181 /* Fill in the fpu structure for a core dump.. */
 182 int dump_fpu(struct pt_regs *regs, elf_fpregset_t *r)
 183 {
 184         memcpy(r, &current->thread.fpu, sizeof(current->thread.fpu));
 185
 186         return 1;
 187 }
 188
 189 void elf_dump_regs(elf_greg_t *gp, struct pt_regs *regs)
 190 {
 191         int i;
 192
 193         for (i = 0; i < EF_R0; i++)
 194                 gp[i] = 0;
 195         gp[EF_R0] = 0;
 196         for (i = 1; i <= 31; i++)
 197                 gp[EF_R0 + i] = regs->regs[i];
 198         gp[EF_R26] = 0;
 199         gp[EF_R27] = 0;
 200         gp[EF_LO] = regs->lo;
 201         gp[EF_HI] = regs->hi;
 202         gp[EF_CP0_EPC] = regs->cp0_epc;
 203         gp[EF_CP0_BADVADDR] = regs->cp0_badvaddr;
 204         gp[EF_CP0_STATUS] = regs->cp0_status;
 205         gp[EF_CP0_CAUSE] = regs->cp0_cause;
 206 #ifdef EF_UNUSED0
 207         gp[EF_UNUSED0] = 0;
 208 #endif
 209 }
 210
 211 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 212 {
 213         elf_dump_regs(*regs, task_pt_regs(tsk));
 214         return 1;
 215 }
 216
 217 int dump_task_fpu(struct task_struct *t, elf_fpregset_t *fpr)
 218 {
 219         memcpy(fpr, &t->thread.fpu, sizeof(current->thread.fpu));
 220
 221         return 1;
 222 }
 223
 224 /*
 225  * Create a kernel thread
 226  */
 227 static void __noreturn kernel_thread_helper(void *arg, int (*fn)(void *))
 228 {
 229         do_exit(fn(arg));
 230 }
 231
 232 long kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
 233 {
 234         struct pt_regs regs;
 235
 236         memset(&regs, 0, sizeof(regs));
 237
 238         regs.regs[4] = (unsigned long) arg;
 239         regs.regs[5] = (unsigned long) fn;
 240         regs.cp0_epc = (unsigned long) kernel_thread_helper;
 241         regs.cp0_status = read_c0_status();
 242 #if defined(CONFIG_CPU_R3000) || defined(CONFIG_CPU_TX39XX)
 243         regs.cp0_status = (regs.cp0_status & ~(ST0_KUP | ST0_IEP | ST0_IEC)) |
 244                           ((regs.cp0_status & (ST0_KUC | ST0_IEC)) << 2);
 245 #else
 246         regs.cp0_status |= ST0_EXL;
 247 #endif
 248
 249         /* Ok, create the new process.. */
 250         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 251 }
 252
 253 /*
 254  *
 255  */
 256 struct mips_frame_info {
 257         void            *func;
 258         unsigned long   func_size;
 259         int             frame_size;
 260         int             pc_offset;
 261 };
 262
 263 static inline int is_ra_save_ins(union mips_instruction *ip)
 264 {
 265         /* sw / sd $ra, offset($sp) */
 266         return (ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
 267                 ip->i_format.rs == 29 &&
 268                 ip->i_format.rt == 31;
 269 }
 270
 271 static inline int is_jal_jalr_jr_ins(union mips_instruction *ip)
 272 {
 273         if (ip->j_format.opcode == jal_op)
 274                 return 1;
 275         if (ip->r_format.opcode != spec_op)
 276                 return 0;
 277         return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
 278 }
 279
 280 static inline int is_sp_move_ins(union mips_instruction *ip)
 281 {
 282         /* addiu/daddiu sp,sp,-imm */
 283         if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
 284                 return 0;
 285         if (ip->i_format.opcode == addiu_op || ip->i_format.opcode == daddiu_op)
 286                 return 1;
 287         return 0;
 288 }
 289
 290 static int get_frame_info(struct mips_frame_info *info)
 291 {
 292         union mips_instruction *ip = info->func;
 293         unsigned max_insns = info->func_size / sizeof(union mips_instruction);
 294         unsigned i;
 295
 296         info->pc_offset = -1;
 297         info->frame_size = 0;
 298
 299         if (!ip)
 300                 goto err;
 301
 302         if (max_insns == 0)
 303                 max_insns = 128U;       /* unknown function size */
 304         max_insns = min(128U, max_insns);
 305
 306         for (i = 0; i < max_insns; i++, ip++) {
 307
 308                 if (is_jal_jalr_jr_ins(ip))
 309                         break;
 310                 if (!info->frame_size) {
 311                         if (is_sp_move_ins(ip))
 312                                 info->frame_size = - ip->i_format.simmediate;
 313                         continue;
 314                 }
 315                 if (info->pc_offset == -1 && is_ra_save_ins(ip)) {
 316                         info->pc_offset =
 317                                 ip->i_format.simmediate / sizeof(long);
 318                         break;
 319                 }
 320         }
 321         if (info->frame_size && info->pc_offset >= 0) /* nested */
 322                 return 0;
 323         if (info->pc_offset < 0) /* leaf */
 324                 return 1;
 325         /* prologue seems boggus... */
 326 err:
 327         return -1;
 328 }
 329
 330 static struct mips_frame_info schedule_mfi __read_mostly;
 331
 332 static int __init frame_info_init(void)
 333 {
 334         unsigned long size = 0;
 335 #ifdef CONFIG_KALLSYMS
 336         unsigned long ofs;
 337
 338         kallsyms_lookup_size_offset((unsigned long)schedule, &size, &ofs);
 339 #endif
 340         schedule_mfi.func = schedule;
 341         schedule_mfi.func_size = size;
 342
 343         get_frame_info(&schedule_mfi);
 344
 345         /*
 346          * Without schedule() frame info, result given by
 347          * thread_saved_pc() and get_wchan() are not reliable.
 348          */
 349         if (schedule_mfi.pc_offset < 0)
 350                 printk("Can't analyze schedule() prologue at %p\n", schedule);
 351
 352         return 0;
 353 }
 354
 355 arch_initcall(frame_info_init);
 356
 357 /*
 358  * Return saved PC of a blocked thread.
 359  */
 360 unsigned long thread_saved_pc(struct task_struct *tsk)
 361 {
 362         struct thread_struct *t = &tsk->thread;
 363
 364         /* New born processes are a special case */
 365         if (t->reg31 == (unsigned long) ret_from_fork)
 366                 return t->reg31;
 367         if (schedule_mfi.pc_offset < 0)
 368                 return 0;
 369         return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
 370 }
 371
 372
 373 #ifdef CONFIG_KALLSYMS
 374 /* generic stack unwinding function */
 375 unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
 376                                               unsigned long *sp,
 377                                               unsigned long pc,
 378                                               unsigned long *ra)
 379 {
 380         struct mips_frame_info info;
 381         unsigned long size, ofs;
 382         int leaf;
 383         extern void ret_from_irq(void);
 384         extern void ret_from_exception(void);
 385
 386         if (!stack_page)
 387                 return 0;
 388
 389         /*
 390          * If we reached the bottom of interrupt context,
 391          * return saved pc in pt_regs.
 392          */
 393         if (pc == (unsigned long)ret_from_irq ||
 394             pc == (unsigned long)ret_from_exception) {
 395                 struct pt_regs *regs;
 396                 if (*sp >= stack_page &&
 397                     *sp + sizeof(*regs) <= stack_page + THREAD_SIZE - 32) {
 398                         regs = (struct pt_regs *)*sp;
 399                         pc = regs->cp0_epc;
 400                         if (__kernel_text_address(pc)) {
 401                                 *sp = regs->regs[29];
 402                                 *ra = regs->regs[31];
 403                                 return pc;
 404                         }
 405                 }
 406                 return 0;
 407         }
 408         if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
 409                 return 0;
 410         /*
 411          * Return ra if an exception occurred at the first instruction
 412          */
 413         if (unlikely(ofs == 0)) {
 414                 pc = *ra;
 415                 *ra = 0;
 416                 return pc;
 417         }
 418
 419         info.func = (void *)(pc - ofs);
 420         info.func_size = ofs;   /* analyze from start to ofs */
 421         leaf = get_frame_info(&info);
 422         if (leaf < 0)
 423                 return 0;
 424
 425         if (*sp < stack_page ||
 426             *sp + info.frame_size > stack_page + THREAD_SIZE - 32)
 427                 return 0;
 428
 429         if (leaf)
 430                 /*
 431                  * For some extreme cases, get_frame_info() can
 432                  * consider wrongly a nested function as a leaf
 433                  * one. In that cases avoid to return always the
 434                  * same value.
 435                  */
 436                 pc = pc != *ra ? *ra : 0;
 437         else
 438                 pc = ((unsigned long *)(*sp))[info.pc_offset];
 439
 440         *sp += info.frame_size;
 441         *ra = 0;
 442         return __kernel_text_address(pc) ? pc : 0;
 443 }
 444 EXPORT_SYMBOL(unwind_stack_by_address);
 445
 446 /* used by show_backtrace() */
 447 unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
 448                            unsigned long pc, unsigned long *ra)
 449 {
 450         unsigned long stack_page = (unsigned long)task_stack_page(task);
 451         return unwind_stack_by_address(stack_page, sp, pc, ra);
 452 }
 453 #endif
 454
 455 /*
 456  * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
 457  */
 458 unsigned long get_wchan(struct task_struct *task)
 459 {
 460         unsigned long pc = 0;
 461 #ifdef CONFIG_KALLSYMS
 462         unsigned long sp;
 463         unsigned long ra = 0;
 464 #endif
 465
 466         if (!task || task == current || task->state == TASK_RUNNING)
 467                 goto out;
 468         if (!task_stack_page(task))
 469                 goto out;
 470
 471         pc = thread_saved_pc(task);
 472
 473 #ifdef CONFIG_KALLSYMS
 474         sp = task->thread.reg29 + schedule_mfi.frame_size;
 475
 476         while (in_sched_functions(pc))
 477                 pc = unwind_stack(task, &sp, pc, &ra);
 478 #endif
 479
 480 out:
 481         return pc;
 482 }
 483
 484 /*
 485  * Don't forget that the stack pointer must be aligned on a 8 bytes
 486  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
 487  */
 488 unsigned long arch_align_stack(unsigned long sp)
 489 {
 490         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 491                 sp -= get_random_int() & ~PAGE_MASK;
 492
 493         return sp & ALMASK;
 494 }