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/export.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/mman.h>
  21 #include <linux/personality.h>
  22 #include <linux/sys.h>
  23 #include <linux/user.h>
  24 #include <linux/init.h>
  25 #include <linux/completion.h>
  26 #include <linux/kallsyms.h>
  27 #include <linux/random.h>
  28
  29 #include <asm/asm.h>
  30 #include <asm/bootinfo.h>
  31 #include <asm/cpu.h>
  32 #include <asm/dsp.h>
  33 #include <asm/fpu.h>
  34 #include <asm/pgtable.h>
  35 #include <asm/system.h>
  36 #include <asm/mipsregs.h>
  37 #include <asm/processor.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/io.h>
  40 #include <asm/elf.h>
  41 #include <asm/isadep.h>
  42 #include <asm/inst.h>
  43 #include <asm/stacktrace.h>
  44
  45 /*
  46  * The idle thread. There's no useful work to be done, so just try to conserve
  47  * power and have a low exit latency (ie sit in a loop waiting for somebody to
  48  * say that they'd like to reschedule)
  49  */
  50 void __noreturn cpu_idle(void)
  51 {
  52         int cpu;
  53
  54         /* CPU is going idle. */
  55         cpu = smp_processor_id();
  56
  57         /* endless idle loop with no priority at all */
  58         while (1) {
  59                 tick_nohz_idle_enter();
  60                 rcu_idle_enter();
  61                 while (!need_resched() && cpu_online(cpu)) {
  62 #ifdef CONFIG_MIPS_MT_SMTC
  63                         extern void smtc_idle_loop_hook(void);
  64
  65                         smtc_idle_loop_hook();
  66 #endif
  67
  68                         if (cpu_wait) {
  69                                 /* Don't trace irqs off for idle */
  70                                 stop_critical_timings();
  71                                 (*cpu_wait)();
  72                                 start_critical_timings();
  73                         }
  74                 }
  75 #ifdef CONFIG_HOTPLUG_CPU
  76                 if (!cpu_online(cpu) && !cpu_isset(cpu, cpu_callin_map) &&
  77                     (system_state == SYSTEM_RUNNING ||
  78                      system_state == SYSTEM_BOOTING))
  79                         play_dead();
  80 #endif
  81                 rcu_idle_exit();
  82                 tick_nohz_idle_exit();
  83                 preempt_enable_no_resched();
  84                 schedule();
  85                 preempt_disable();
  86         }
  87 }
  88
  89 asmlinkage void ret_from_fork(void);
  90
  91 void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
  92 {
  93         unsigned long status;
  94
  95         /* New thread loses kernel privileges. */
  96         status = regs->cp0_status & ~(ST0_CU0|ST0_CU1|ST0_FR|KU_MASK);
  97 #ifdef CONFIG_64BIT
  98         status |= test_thread_flag(TIF_32BIT_REGS) ? 0 : ST0_FR;
  99 #endif
 100         status |= KU_USER;
 101         regs->cp0_status = status;
 102         clear_used_math();
 103         clear_fpu_owner();
 104         if (cpu_has_dsp)
 105                 __init_dsp();
 106         regs->cp0_epc = pc;
 107         regs->regs[29] = sp;
 108 }
 109
 110 void exit_thread(void)
 111 {
 112 }
 113
 114 void flush_thread(void)
 115 {
 116 }
 117
 118 int copy_thread(unsigned long clone_flags, unsigned long usp,
 119         unsigned long unused, struct task_struct *p, struct pt_regs *regs)
 120 {
 121         struct thread_info *ti = task_thread_info(p);
 122         struct pt_regs *childregs;
 123         unsigned long childksp;
 124         p->set_child_tid = p->clear_child_tid = NULL;
 125
 126         childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
 127
 128         preempt_disable();
 129
 130         if (is_fpu_owner())
 131                 save_fp(p);
 132
 133         if (cpu_has_dsp)
 134                 save_dsp(p);
 135
 136         preempt_enable();
 137
 138         /* set up new TSS. */
 139         childregs = (struct pt_regs *) childksp - 1;
 140         /*  Put the stack after the struct pt_regs.  */
 141         childksp = (unsigned long) childregs;
 142         *childregs = *regs;
 143         childregs->regs[7] = 0; /* Clear error flag */
 144
 145         childregs->regs[2] = 0; /* Child gets zero as return value */
 146
 147         if (childregs->cp0_status & ST0_CU0) {
 148                 childregs->regs[28] = (unsigned long) ti;
 149                 childregs->regs[29] = childksp;
 150                 ti->addr_limit = KERNEL_DS;
 151         } else {
 152                 childregs->regs[29] = usp;
 153                 ti->addr_limit = USER_DS;
 154         }
 155         p->thread.reg29 = (unsigned long) childregs;
 156         p->thread.reg31 = (unsigned long) ret_from_fork;
 157
 158         /*
 159          * New tasks lose permission to use the fpu. This accelerates context
 160          * switching for most programs since they don't use the fpu.
 161          */
 162         p->thread.cp0_status = read_c0_status() & ~(ST0_CU2|ST0_CU1);
 163         childregs->cp0_status &= ~(ST0_CU2|ST0_CU1);
 164
 165 #ifdef CONFIG_MIPS_MT_SMTC
 166         /*
 167          * SMTC restores TCStatus after Status, and the CU bits
 168          * are aliased there.
 169          */
 170         childregs->cp0_tcstatus &= ~(ST0_CU2|ST0_CU1);
 171 #endif
 172         clear_tsk_thread_flag(p, TIF_USEDFPU);
 173
 174 #ifdef CONFIG_MIPS_MT_FPAFF
 175         clear_tsk_thread_flag(p, TIF_FPUBOUND);
 176 #endif /* CONFIG_MIPS_MT_FPAFF */
 177
 178         if (clone_flags & CLONE_SETTLS)
 179                 ti->tp_value = regs->regs[7];
 180
 181         return 0;
 182 }
 183
 184 /* Fill in the fpu structure for a core dump.. */
 185 int dump_fpu(struct pt_regs *regs, elf_fpregset_t *r)
 186 {
 187         memcpy(r, &current->thread.fpu, sizeof(current->thread.fpu));
 188
 189         return 1;
 190 }
 191
 192 void elf_dump_regs(elf_greg_t *gp, struct pt_regs *regs)
 193 {
 194         int i;
 195
 196         for (i = 0; i < EF_R0; i++)
 197                 gp[i] = 0;
 198         gp[EF_R0] = 0;
 199         for (i = 1; i <= 31; i++)
 200                 gp[EF_R0 + i] = regs->regs[i];
 201         gp[EF_R26] = 0;
 202         gp[EF_R27] = 0;
 203         gp[EF_LO] = regs->lo;
 204         gp[EF_HI] = regs->hi;
 205         gp[EF_CP0_EPC] = regs->cp0_epc;
 206         gp[EF_CP0_BADVADDR] = regs->cp0_badvaddr;
 207         gp[EF_CP0_STATUS] = regs->cp0_status;
 208         gp[EF_CP0_CAUSE] = regs->cp0_cause;
 209 #ifdef EF_UNUSED0
 210         gp[EF_UNUSED0] = 0;
 211 #endif
 212 }
 213
 214 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 215 {
 216         elf_dump_regs(*regs, task_pt_regs(tsk));
 217         return 1;
 218 }
 219
 220 int dump_task_fpu(struct task_struct *t, elf_fpregset_t *fpr)
 221 {
 222         memcpy(fpr, &t->thread.fpu, sizeof(current->thread.fpu));
 223
 224         return 1;
 225 }
 226
 227 /*
 228  * Create a kernel thread
 229  */
 230 static void __noreturn kernel_thread_helper(void *arg, int (*fn)(void *))
 231 {
 232         do_exit(fn(arg));
 233 }
 234
 235 long kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
 236 {
 237         struct pt_regs regs;
 238
 239         memset(&regs, 0, sizeof(regs));
 240
 241         regs.regs[4] = (unsigned long) arg;
 242         regs.regs[5] = (unsigned long) fn;
 243         regs.cp0_epc = (unsigned long) kernel_thread_helper;
 244         regs.cp0_status = read_c0_status();
 245 #if defined(CONFIG_CPU_R3000) || defined(CONFIG_CPU_TX39XX)
 246         regs.cp0_status = (regs.cp0_status & ~(ST0_KUP | ST0_IEP | ST0_IEC)) |
 247                           ((regs.cp0_status & (ST0_KUC | ST0_IEC)) << 2);
 248 #else
 249         regs.cp0_status |= ST0_EXL;
 250 #endif
 251
 252         /* Ok, create the new process.. */
 253         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 254 }
 255
 256 /*
 257  *
 258  */
 259 struct mips_frame_info {
 260         void            *func;
 261         unsigned long   func_size;
 262         int             frame_size;
 263         int             pc_offset;
 264 };
 265
 266 static inline int is_ra_save_ins(union mips_instruction *ip)
 267 {
 268         /* sw / sd $ra, offset($sp) */
 269         return (ip->i_format.opcode == sw_op || ip->i_format.opcode == sd_op) &&
 270                 ip->i_format.rs == 29 &&
 271                 ip->i_format.rt == 31;
 272 }
 273
 274 static inline int is_jal_jalr_jr_ins(union mips_instruction *ip)
 275 {
 276         if (ip->j_format.opcode == jal_op)
 277                 return 1;
 278         if (ip->r_format.opcode != spec_op)
 279                 return 0;
 280         return ip->r_format.func == jalr_op || ip->r_format.func == jr_op;
 281 }
 282
 283 static inline int is_sp_move_ins(union mips_instruction *ip)
 284 {
 285         /* addiu/daddiu sp,sp,-imm */
 286         if (ip->i_format.rs != 29 || ip->i_format.rt != 29)
 287                 return 0;
 288         if (ip->i_format.opcode == addiu_op || ip->i_format.opcode == daddiu_op)
 289                 return 1;
 290         return 0;
 291 }
 292
 293 static int get_frame_info(struct mips_frame_info *info)
 294 {
 295         union mips_instruction *ip = info->func;
 296         unsigned max_insns = info->func_size / sizeof(union mips_instruction);
 297         unsigned i;
 298
 299         info->pc_offset = -1;
 300         info->frame_size = 0;
 301
 302         if (!ip)
 303                 goto err;
 304
 305         if (max_insns == 0)
 306                 max_insns = 128U;       /* unknown function size */
 307         max_insns = min(128U, max_insns);
 308
 309         for (i = 0; i < max_insns; i++, ip++) {
 310
 311                 if (is_jal_jalr_jr_ins(ip))
 312                         break;
 313                 if (!info->frame_size) {
 314                         if (is_sp_move_ins(ip))
 315                                 info->frame_size = - ip->i_format.simmediate;
 316                         continue;
 317                 }
 318                 if (info->pc_offset == -1 && is_ra_save_ins(ip)) {
 319                         info->pc_offset =
 320                                 ip->i_format.simmediate / sizeof(long);
 321                         break;
 322                 }
 323         }
 324         if (info->frame_size && info->pc_offset >= 0) /* nested */
 325                 return 0;
 326         if (info->pc_offset < 0) /* leaf */
 327                 return 1;
 328         /* prologue seems boggus... */
 329 err:
 330         return -1;
 331 }
 332
 333 static struct mips_frame_info schedule_mfi __read_mostly;
 334
 335 static int __init frame_info_init(void)
 336 {
 337         unsigned long size = 0;
 338 #ifdef CONFIG_KALLSYMS
 339         unsigned long ofs;
 340
 341         kallsyms_lookup_size_offset((unsigned long)schedule, &size, &ofs);
 342 #endif
 343         schedule_mfi.func = schedule;
 344         schedule_mfi.func_size = size;
 345
 346         get_frame_info(&schedule_mfi);
 347
 348         /*
 349          * Without schedule() frame info, result given by
 350          * thread_saved_pc() and get_wchan() are not reliable.
 351          */
 352         if (schedule_mfi.pc_offset < 0)
 353                 printk("Can't analyze schedule() prologue at %p\n", schedule);
 354
 355         return 0;
 356 }
 357
 358 arch_initcall(frame_info_init);
 359
 360 /*
 361  * Return saved PC of a blocked thread.
 362  */
 363 unsigned long thread_saved_pc(struct task_struct *tsk)
 364 {
 365         struct thread_struct *t = &tsk->thread;
 366
 367         /* New born processes are a special case */
 368         if (t->reg31 == (unsigned long) ret_from_fork)
 369                 return t->reg31;
 370         if (schedule_mfi.pc_offset < 0)
 371                 return 0;
 372         return ((unsigned long *)t->reg29)[schedule_mfi.pc_offset];
 373 }
 374
 375
 376 #ifdef CONFIG_KALLSYMS
 377 /* generic stack unwinding function */
 378 unsigned long notrace unwind_stack_by_address(unsigned long stack_page,
 379                                               unsigned long *sp,
 380                                               unsigned long pc,
 381                                               unsigned long *ra)
 382 {
 383         struct mips_frame_info info;
 384         unsigned long size, ofs;
 385         int leaf;
 386         extern void ret_from_irq(void);
 387         extern void ret_from_exception(void);
 388
 389         if (!stack_page)
 390                 return 0;
 391
 392         /*
 393          * If we reached the bottom of interrupt context,
 394          * return saved pc in pt_regs.
 395          */
 396         if (pc == (unsigned long)ret_from_irq ||
 397             pc == (unsigned long)ret_from_exception) {
 398                 struct pt_regs *regs;
 399                 if (*sp >= stack_page &&
 400                     *sp + sizeof(*regs) <= stack_page + THREAD_SIZE - 32) {
 401                         regs = (struct pt_regs *)*sp;
 402                         pc = regs->cp0_epc;
 403                         if (__kernel_text_address(pc)) {
 404                                 *sp = regs->regs[29];
 405                                 *ra = regs->regs[31];
 406                                 return pc;
 407                         }
 408                 }
 409                 return 0;
 410         }
 411         if (!kallsyms_lookup_size_offset(pc, &size, &ofs))
 412                 return 0;
 413         /*
 414          * Return ra if an exception occurred at the first instruction
 415          */
 416         if (unlikely(ofs == 0)) {
 417                 pc = *ra;
 418                 *ra = 0;
 419                 return pc;
 420         }
 421
 422         info.func = (void *)(pc - ofs);
 423         info.func_size = ofs;   /* analyze from start to ofs */
 424         leaf = get_frame_info(&info);
 425         if (leaf < 0)
 426                 return 0;
 427
 428         if (*sp < stack_page ||
 429             *sp + info.frame_size > stack_page + THREAD_SIZE - 32)
 430                 return 0;
 431
 432         if (leaf)
 433                 /*
 434                  * For some extreme cases, get_frame_info() can
 435                  * consider wrongly a nested function as a leaf
 436                  * one. In that cases avoid to return always the
 437                  * same value.
 438                  */
 439                 pc = pc != *ra ? *ra : 0;
 440         else
 441                 pc = ((unsigned long *)(*sp))[info.pc_offset];
 442
 443         *sp += info.frame_size;
 444         *ra = 0;
 445         return __kernel_text_address(pc) ? pc : 0;
 446 }
 447 EXPORT_SYMBOL(unwind_stack_by_address);
 448
 449 /* used by show_backtrace() */
 450 unsigned long unwind_stack(struct task_struct *task, unsigned long *sp,
 451                            unsigned long pc, unsigned long *ra)
 452 {
 453         unsigned long stack_page = (unsigned long)task_stack_page(task);
 454         return unwind_stack_by_address(stack_page, sp, pc, ra);
 455 }
 456 #endif
 457
 458 /*
 459  * get_wchan - a maintenance nightmare^W^Wpain in the ass ...
 460  */
 461 unsigned long get_wchan(struct task_struct *task)
 462 {
 463         unsigned long pc = 0;
 464 #ifdef CONFIG_KALLSYMS
 465         unsigned long sp;
 466         unsigned long ra = 0;
 467 #endif
 468
 469         if (!task || task == current || task->state == TASK_RUNNING)
 470                 goto out;
 471         if (!task_stack_page(task))
 472                 goto out;
 473
 474         pc = thread_saved_pc(task);
 475
 476 #ifdef CONFIG_KALLSYMS
 477         sp = task->thread.reg29 + schedule_mfi.frame_size;
 478
 479         while (in_sched_functions(pc))
 480                 pc = unwind_stack(task, &sp, pc, &ra);
 481 #endif
 482
 483 out:
 484         return pc;
 485 }
 486
 487 /*
 488  * Don't forget that the stack pointer must be aligned on a 8 bytes
 489  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
 490  */
 491 unsigned long arch_align_stack(unsigned long sp)
 492 {
 493         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 494                 sp -= get_random_int() & ~PAGE_MASK;
 495
 496         return sp & ALMASK;
 497 }