git.oblomov.eu Git - linux-2.6/blob - arch/powerpc/kernel/process.c

   1 /*
   2  *  Derived from "arch/i386/kernel/process.c"
   3  *    Copyright (C) 1995  Linus Torvalds
   4  *
   5  *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
   6  *  Paul Mackerras (paulus@cs.anu.edu.au)
   7  *
   8  *  PowerPC version
   9  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  10  *
  11  *  This program is free software; you can redistribute it and/or
  12  *  modify it under the terms of the GNU General Public License
  13  *  as published by the Free Software Foundation; either version
  14  *  2 of the License, or (at your option) any later version.
  15  */
  16
  17 #include <linux/errno.h>
  18 #include <linux/sched.h>
  19 #include <linux/kernel.h>
  20 #include <linux/mm.h>
  21 #include <linux/smp.h>
  22 #include <linux/stddef.h>
  23 #include <linux/unistd.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/slab.h>
  26 #include <linux/user.h>
  27 #include <linux/elf.h>
  28 #include <linux/init.h>
  29 #include <linux/prctl.h>
  30 #include <linux/init_task.h>
  31 #include <linux/module.h>
  32 #include <linux/kallsyms.h>
  33 #include <linux/mqueue.h>
  34 #include <linux/hardirq.h>
  35 #include <linux/utsname.h>
  36
  37 #include <asm/pgtable.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/system.h>
  40 #include <asm/io.h>
  41 #include <asm/processor.h>
  42 #include <asm/mmu.h>
  43 #include <asm/prom.h>
  44 #include <asm/machdep.h>
  45 #include <asm/time.h>
  46 #include <asm/syscalls.h>
  47 #ifdef CONFIG_PPC64
  48 #include <asm/firmware.h>
  49 #endif
  50
  51 extern unsigned long _get_SP(void);
  52
  53 #ifndef CONFIG_SMP
  54 struct task_struct *last_task_used_math = NULL;
  55 struct task_struct *last_task_used_altivec = NULL;
  56 struct task_struct *last_task_used_vsx = NULL;
  57 struct task_struct *last_task_used_spe = NULL;
  58 #endif
  59
  60 /*
  61  * Make sure the floating-point register state in the
  62  * the thread_struct is up to date for task tsk.
  63  */
  64 void flush_fp_to_thread(struct task_struct *tsk)
  65 {
  66         if (tsk->thread.regs) {
  67                 /*
  68                  * We need to disable preemption here because if we didn't,
  69                  * another process could get scheduled after the regs->msr
  70                  * test but before we have finished saving the FP registers
  71                  * to the thread_struct.  That process could take over the
  72                  * FPU, and then when we get scheduled again we would store
  73                  * bogus values for the remaining FP registers.
  74                  */
  75                 preempt_disable();
  76                 if (tsk->thread.regs->msr & MSR_FP) {
  77 #ifdef CONFIG_SMP
  78                         /*
  79                          * This should only ever be called for current or
  80                          * for a stopped child process.  Since we save away
  81                          * the FP register state on context switch on SMP,
  82                          * there is something wrong if a stopped child appears
  83                          * to still have its FP state in the CPU registers.
  84                          */
  85                         BUG_ON(tsk != current);
  86 #endif
  87                         giveup_fpu(tsk);
  88                 }
  89                 preempt_enable();
  90         }
  91 }
  92
  93 void enable_kernel_fp(void)
  94 {
  95         WARN_ON(preemptible());
  96
  97 #ifdef CONFIG_SMP
  98         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
  99                 giveup_fpu(current);
 100         else
 101                 giveup_fpu(NULL);       /* just enables FP for kernel */
 102 #else
 103         giveup_fpu(last_task_used_math);
 104 #endif /* CONFIG_SMP */
 105 }
 106 EXPORT_SYMBOL(enable_kernel_fp);
 107
 108 #ifdef CONFIG_ALTIVEC
 109 void enable_kernel_altivec(void)
 110 {
 111         WARN_ON(preemptible());
 112
 113 #ifdef CONFIG_SMP
 114         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 115                 giveup_altivec(current);
 116         else
 117                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
 118 #else
 119         giveup_altivec(last_task_used_altivec);
 120 #endif /* CONFIG_SMP */
 121 }
 122 EXPORT_SYMBOL(enable_kernel_altivec);
 123
 124 /*
 125  * Make sure the VMX/Altivec register state in the
 126  * the thread_struct is up to date for task tsk.
 127  */
 128 void flush_altivec_to_thread(struct task_struct *tsk)
 129 {
 130         if (tsk->thread.regs) {
 131                 preempt_disable();
 132                 if (tsk->thread.regs->msr & MSR_VEC) {
 133 #ifdef CONFIG_SMP
 134                         BUG_ON(tsk != current);
 135 #endif
 136                         giveup_altivec(tsk);
 137                 }
 138                 preempt_enable();
 139         }
 140 }
 141 #endif /* CONFIG_ALTIVEC */
 142
 143 #ifdef CONFIG_VSX
 144 #if 0
 145 /* not currently used, but some crazy RAID module might want to later */
 146 void enable_kernel_vsx(void)
 147 {
 148         WARN_ON(preemptible());
 149
 150 #ifdef CONFIG_SMP
 151         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 152                 giveup_vsx(current);
 153         else
 154                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 155 #else
 156         giveup_vsx(last_task_used_vsx);
 157 #endif /* CONFIG_SMP */
 158 }
 159 EXPORT_SYMBOL(enable_kernel_vsx);
 160 #endif
 161
 162 void giveup_vsx(struct task_struct *tsk)
 163 {
 164         giveup_fpu(tsk);
 165         giveup_altivec(tsk);
 166         __giveup_vsx(tsk);
 167 }
 168
 169 void flush_vsx_to_thread(struct task_struct *tsk)
 170 {
 171         if (tsk->thread.regs) {
 172                 preempt_disable();
 173                 if (tsk->thread.regs->msr & MSR_VSX) {
 174 #ifdef CONFIG_SMP
 175                         BUG_ON(tsk != current);
 176 #endif
 177                         giveup_vsx(tsk);
 178                 }
 179                 preempt_enable();
 180         }
 181 }
 182 #endif /* CONFIG_VSX */
 183
 184 #ifdef CONFIG_SPE
 185
 186 void enable_kernel_spe(void)
 187 {
 188         WARN_ON(preemptible());
 189
 190 #ifdef CONFIG_SMP
 191         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 192                 giveup_spe(current);
 193         else
 194                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 195 #else
 196         giveup_spe(last_task_used_spe);
 197 #endif /* __SMP __ */
 198 }
 199 EXPORT_SYMBOL(enable_kernel_spe);
 200
 201 void flush_spe_to_thread(struct task_struct *tsk)
 202 {
 203         if (tsk->thread.regs) {
 204                 preempt_disable();
 205                 if (tsk->thread.regs->msr & MSR_SPE) {
 206 #ifdef CONFIG_SMP
 207                         BUG_ON(tsk != current);
 208 #endif
 209                         giveup_spe(tsk);
 210                 }
 211                 preempt_enable();
 212         }
 213 }
 214 #endif /* CONFIG_SPE */
 215
 216 #ifndef CONFIG_SMP
 217 /*
 218  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 219  * and the current task has some state, discard it.
 220  */
 221 void discard_lazy_cpu_state(void)
 222 {
 223         preempt_disable();
 224         if (last_task_used_math == current)
 225                 last_task_used_math = NULL;
 226 #ifdef CONFIG_ALTIVEC
 227         if (last_task_used_altivec == current)
 228                 last_task_used_altivec = NULL;
 229 #endif /* CONFIG_ALTIVEC */
 230 #ifdef CONFIG_VSX
 231         if (last_task_used_vsx == current)
 232                 last_task_used_vsx = NULL;
 233 #endif /* CONFIG_VSX */
 234 #ifdef CONFIG_SPE
 235         if (last_task_used_spe == current)
 236                 last_task_used_spe = NULL;
 237 #endif
 238         preempt_enable();
 239 }
 240 #endif /* CONFIG_SMP */
 241
 242 static DEFINE_PER_CPU(unsigned long, current_dabr);
 243
 244 int set_dabr(unsigned long dabr)
 245 {
 246         __get_cpu_var(current_dabr) = dabr;
 247
 248 #ifdef CONFIG_PPC_MERGE         /* XXX for now */
 249         if (ppc_md.set_dabr)
 250                 return ppc_md.set_dabr(dabr);
 251 #endif
 252
 253         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 254 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
 255         mtspr(SPRN_DABR, dabr);
 256 #endif
 257         return 0;
 258 }
 259
 260 #ifdef CONFIG_PPC64
 261 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 262 #endif
 263
 264 struct task_struct *__switch_to(struct task_struct *prev,
 265         struct task_struct *new)
 266 {
 267         struct thread_struct *new_thread, *old_thread;
 268         unsigned long flags;
 269         struct task_struct *last;
 270
 271 #ifdef CONFIG_SMP
 272         /* avoid complexity of lazy save/restore of fpu
 273          * by just saving it every time we switch out if
 274          * this task used the fpu during the last quantum.
 275          *
 276          * If it tries to use the fpu again, it'll trap and
 277          * reload its fp regs.  So we don't have to do a restore
 278          * every switch, just a save.
 279          *  -- Cort
 280          */
 281         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 282                 giveup_fpu(prev);
 283 #ifdef CONFIG_ALTIVEC
 284         /*
 285          * If the previous thread used altivec in the last quantum
 286          * (thus changing altivec regs) then save them.
 287          * We used to check the VRSAVE register but not all apps
 288          * set it, so we don't rely on it now (and in fact we need
 289          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 290          *
 291          * On SMP we always save/restore altivec regs just to avoid the
 292          * complexity of changing processors.
 293          *  -- Cort
 294          */
 295         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 296                 giveup_altivec(prev);
 297 #endif /* CONFIG_ALTIVEC */
 298 #ifdef CONFIG_VSX
 299         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 300                 /* VMX and FPU registers are already save here */
 301                 __giveup_vsx(prev);
 302 #endif /* CONFIG_VSX */
 303 #ifdef CONFIG_SPE
 304         /*
 305          * If the previous thread used spe in the last quantum
 306          * (thus changing spe regs) then save them.
 307          *
 308          * On SMP we always save/restore spe regs just to avoid the
 309          * complexity of changing processors.
 310          */
 311         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 312                 giveup_spe(prev);
 313 #endif /* CONFIG_SPE */
 314
 315 #else  /* CONFIG_SMP */
 316 #ifdef CONFIG_ALTIVEC
 317         /* Avoid the trap.  On smp this this never happens since
 318          * we don't set last_task_used_altivec -- Cort
 319          */
 320         if (new->thread.regs && last_task_used_altivec == new)
 321                 new->thread.regs->msr |= MSR_VEC;
 322 #endif /* CONFIG_ALTIVEC */
 323 #ifdef CONFIG_VSX
 324         if (new->thread.regs && last_task_used_vsx == new)
 325                 new->thread.regs->msr |= MSR_VSX;
 326 #endif /* CONFIG_VSX */
 327 #ifdef CONFIG_SPE
 328         /* Avoid the trap.  On smp this this never happens since
 329          * we don't set last_task_used_spe
 330          */
 331         if (new->thread.regs && last_task_used_spe == new)
 332                 new->thread.regs->msr |= MSR_SPE;
 333 #endif /* CONFIG_SPE */
 334
 335 #endif /* CONFIG_SMP */
 336
 337         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 338                 set_dabr(new->thread.dabr);
 339
 340         new_thread = &new->thread;
 341         old_thread = &current->thread;
 342
 343 #ifdef CONFIG_PPC64
 344         /*
 345          * Collect processor utilization data per process
 346          */
 347         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 348                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 349                 long unsigned start_tb, current_tb;
 350                 start_tb = old_thread->start_tb;
 351                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 352                 old_thread->accum_tb += (current_tb - start_tb);
 353                 new_thread->start_tb = current_tb;
 354         }
 355 #endif
 356
 357         local_irq_save(flags);
 358
 359         account_system_vtime(current);
 360         account_process_vtime(current);
 361         calculate_steal_time();
 362
 363         /*
 364          * We can't take a PMU exception inside _switch() since there is a
 365          * window where the kernel stack SLB and the kernel stack are out
 366          * of sync. Hard disable here.
 367          */
 368         hard_irq_disable();
 369         last = _switch(old_thread, new_thread);
 370
 371         local_irq_restore(flags);
 372
 373         return last;
 374 }
 375
 376 static int instructions_to_print = 16;
 377
 378 static void show_instructions(struct pt_regs *regs)
 379 {
 380         int i;
 381         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 382                         sizeof(int));
 383
 384         printk("Instruction dump:");
 385
 386         for (i = 0; i < instructions_to_print; i++) {
 387                 int instr;
 388
 389                 if (!(i % 8))
 390                         printk("\n");
 391
 392 #if !defined(CONFIG_BOOKE)
 393                 /* If executing with the IMMU off, adjust pc rather
 394                  * than print XXXXXXXX.
 395                  */
 396                 if (!(regs->msr & MSR_IR))
 397                         pc = (unsigned long)phys_to_virt(pc);
 398 #endif
 399
 400                 /* We use __get_user here *only* to avoid an OOPS on a
 401                  * bad address because the pc *should* only be a
 402                  * kernel address.
 403                  */
 404                 if (!__kernel_text_address(pc) ||
 405                      __get_user(instr, (unsigned int __user *)pc)) {
 406                         printk("XXXXXXXX ");
 407                 } else {
 408                         if (regs->nip == pc)
 409                                 printk("<%08x> ", instr);
 410                         else
 411                                 printk("%08x ", instr);
 412                 }
 413
 414                 pc += sizeof(int);
 415         }
 416
 417         printk("\n");
 418 }
 419
 420 static struct regbit {
 421         unsigned long bit;
 422         const char *name;
 423 } msr_bits[] = {
 424         {MSR_EE,        "EE"},
 425         {MSR_PR,        "PR"},
 426         {MSR_FP,        "FP"},
 427         {MSR_VEC,       "VEC"},
 428         {MSR_VSX,       "VSX"},
 429         {MSR_ME,        "ME"},
 430         {MSR_IR,        "IR"},
 431         {MSR_DR,        "DR"},
 432         {0,             NULL}
 433 };
 434
 435 static void printbits(unsigned long val, struct regbit *bits)
 436 {
 437         const char *sep = "";
 438
 439         printk("<");
 440         for (; bits->bit; ++bits)
 441                 if (val & bits->bit) {
 442                         printk("%s%s", sep, bits->name);
 443                         sep = ",";
 444                 }
 445         printk(">");
 446 }
 447
 448 #ifdef CONFIG_PPC64
 449 #define REG             "%016lx"
 450 #define REGS_PER_LINE   4
 451 #define LAST_VOLATILE   13
 452 #else
 453 #define REG             "%08lx"
 454 #define REGS_PER_LINE   8
 455 #define LAST_VOLATILE   12
 456 #endif
 457
 458 void show_regs(struct pt_regs * regs)
 459 {
 460         int i, trap;
 461
 462         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 463                regs->nip, regs->link, regs->ctr);
 464         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 465                regs, regs->trap, print_tainted(), init_utsname()->release);
 466         printk("MSR: "REG" ", regs->msr);
 467         printbits(regs->msr, msr_bits);
 468         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 469         trap = TRAP(regs);
 470         if (trap == 0x300 || trap == 0x600)
 471 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 472                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 473 #else
 474                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 475 #endif
 476         printk("TASK = %p[%d] '%s' THREAD: %p",
 477                current, task_pid_nr(current), current->comm, task_thread_info(current));
 478
 479 #ifdef CONFIG_SMP
 480         printk(" CPU: %d", raw_smp_processor_id());
 481 #endif /* CONFIG_SMP */
 482
 483         for (i = 0;  i < 32;  i++) {
 484                 if ((i % REGS_PER_LINE) == 0)
 485                         printk("\n" KERN_INFO "GPR%02d: ", i);
 486                 printk(REG " ", regs->gpr[i]);
 487                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 488                         break;
 489         }
 490         printk("\n");
 491 #ifdef CONFIG_KALLSYMS
 492         /*
 493          * Lookup NIP late so we have the best change of getting the
 494          * above info out without failing
 495          */
 496         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 497         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 498 #endif
 499         show_stack(current, (unsigned long *) regs->gpr[1]);
 500         if (!user_mode(regs))
 501                 show_instructions(regs);
 502 }
 503
 504 void exit_thread(void)
 505 {
 506         discard_lazy_cpu_state();
 507 }
 508
 509 void flush_thread(void)
 510 {
 511 #ifdef CONFIG_PPC64
 512         struct thread_info *t = current_thread_info();
 513
 514         if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
 515                 clear_ti_thread_flag(t, TIF_ABI_PENDING);
 516                 if (test_ti_thread_flag(t, TIF_32BIT))
 517                         clear_ti_thread_flag(t, TIF_32BIT);
 518                 else
 519                         set_ti_thread_flag(t, TIF_32BIT);
 520         }
 521 #endif
 522
 523         discard_lazy_cpu_state();
 524
 525         if (current->thread.dabr) {
 526                 current->thread.dabr = 0;
 527                 set_dabr(0);
 528         }
 529 }
 530
 531 void
 532 release_thread(struct task_struct *t)
 533 {
 534 }
 535
 536 /*
 537  * This gets called before we allocate a new thread and copy
 538  * the current task into it.
 539  */
 540 void prepare_to_copy(struct task_struct *tsk)
 541 {
 542         flush_fp_to_thread(current);
 543         flush_altivec_to_thread(current);
 544         flush_vsx_to_thread(current);
 545         flush_spe_to_thread(current);
 546 }
 547
 548 /*
 549  * Copy a thread..
 550  */
 551 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 552                 unsigned long unused, struct task_struct *p,
 553                 struct pt_regs *regs)
 554 {
 555         struct pt_regs *childregs, *kregs;
 556         extern void ret_from_fork(void);
 557         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 558
 559         CHECK_FULL_REGS(regs);
 560         /* Copy registers */
 561         sp -= sizeof(struct pt_regs);
 562         childregs = (struct pt_regs *) sp;
 563         *childregs = *regs;
 564         if ((childregs->msr & MSR_PR) == 0) {
 565                 /* for kernel thread, set `current' and stackptr in new task */
 566                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 567 #ifdef CONFIG_PPC32
 568                 childregs->gpr[2] = (unsigned long) p;
 569 #else
 570                 clear_tsk_thread_flag(p, TIF_32BIT);
 571 #endif
 572                 p->thread.regs = NULL;  /* no user register state */
 573         } else {
 574                 childregs->gpr[1] = usp;
 575                 p->thread.regs = childregs;
 576                 if (clone_flags & CLONE_SETTLS) {
 577 #ifdef CONFIG_PPC64
 578                         if (!test_thread_flag(TIF_32BIT))
 579                                 childregs->gpr[13] = childregs->gpr[6];
 580                         else
 581 #endif
 582                                 childregs->gpr[2] = childregs->gpr[6];
 583                 }
 584         }
 585         childregs->gpr[3] = 0;  /* Result from fork() */
 586         sp -= STACK_FRAME_OVERHEAD;
 587
 588         /*
 589          * The way this works is that at some point in the future
 590          * some task will call _switch to switch to the new task.
 591          * That will pop off the stack frame created below and start
 592          * the new task running at ret_from_fork.  The new task will
 593          * do some house keeping and then return from the fork or clone
 594          * system call, using the stack frame created above.
 595          */
 596         sp -= sizeof(struct pt_regs);
 597         kregs = (struct pt_regs *) sp;
 598         sp -= STACK_FRAME_OVERHEAD;
 599         p->thread.ksp = sp;
 600         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 601                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 602
 603 #ifdef CONFIG_PPC64
 604         if (cpu_has_feature(CPU_FTR_SLB)) {
 605                 unsigned long sp_vsid;
 606                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 607
 608                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
 609                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 610                                 << SLB_VSID_SHIFT_1T;
 611                 else
 612                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 613                                 << SLB_VSID_SHIFT;
 614                 sp_vsid |= SLB_VSID_KERNEL | llp;
 615                 p->thread.ksp_vsid = sp_vsid;
 616         }
 617
 618         /*
 619          * The PPC64 ABI makes use of a TOC to contain function
 620          * pointers.  The function (ret_from_except) is actually a pointer
 621          * to the TOC entry.  The first entry is a pointer to the actual
 622          * function.
 623          */
 624         kregs->nip = *((unsigned long *)ret_from_fork);
 625 #else
 626         kregs->nip = (unsigned long)ret_from_fork;
 627 #endif
 628
 629         return 0;
 630 }
 631
 632 /*
 633  * Set up a thread for executing a new program
 634  */
 635 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 636 {
 637 #ifdef CONFIG_PPC64
 638         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 639 #endif
 640
 641         set_fs(USER_DS);
 642
 643         /*
 644          * If we exec out of a kernel thread then thread.regs will not be
 645          * set.  Do it now.
 646          */
 647         if (!current->thread.regs) {
 648                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 649                 current->thread.regs = regs - 1;
 650         }
 651
 652         memset(regs->gpr, 0, sizeof(regs->gpr));
 653         regs->ctr = 0;
 654         regs->link = 0;
 655         regs->xer = 0;
 656         regs->ccr = 0;
 657         regs->gpr[1] = sp;
 658
 659         /*
 660          * We have just cleared all the nonvolatile GPRs, so make
 661          * FULL_REGS(regs) return true.  This is necessary to allow
 662          * ptrace to examine the thread immediately after exec.
 663          */
 664         regs->trap &= ~1UL;
 665
 666 #ifdef CONFIG_PPC32
 667         regs->mq = 0;
 668         regs->nip = start;
 669         regs->msr = MSR_USER;
 670 #else
 671         if (!test_thread_flag(TIF_32BIT)) {
 672                 unsigned long entry, toc;
 673
 674                 /* start is a relocated pointer to the function descriptor for
 675                  * the elf _start routine.  The first entry in the function
 676                  * descriptor is the entry address of _start and the second
 677                  * entry is the TOC value we need to use.
 678                  */
 679                 __get_user(entry, (unsigned long __user *)start);
 680                 __get_user(toc, (unsigned long __user *)start+1);
 681
 682                 /* Check whether the e_entry function descriptor entries
 683                  * need to be relocated before we can use them.
 684                  */
 685                 if (load_addr != 0) {
 686                         entry += load_addr;
 687                         toc   += load_addr;
 688                 }
 689                 regs->nip = entry;
 690                 regs->gpr[2] = toc;
 691                 regs->msr = MSR_USER64;
 692         } else {
 693                 regs->nip = start;
 694                 regs->gpr[2] = 0;
 695                 regs->msr = MSR_USER32;
 696         }
 697 #endif
 698
 699         discard_lazy_cpu_state();
 700 #ifdef CONFIG_VSX
 701         current->thread.used_vsr = 0;
 702 #endif
 703         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 704         current->thread.fpscr.val = 0;
 705 #ifdef CONFIG_ALTIVEC
 706         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 707         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 708         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 709         current->thread.vrsave = 0;
 710         current->thread.used_vr = 0;
 711 #endif /* CONFIG_ALTIVEC */
 712 #ifdef CONFIG_SPE
 713         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 714         current->thread.acc = 0;
 715         current->thread.spefscr = 0;
 716         current->thread.used_spe = 0;
 717 #endif /* CONFIG_SPE */
 718 }
 719
 720 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 721                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 722
 723 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 724 {
 725         struct pt_regs *regs = tsk->thread.regs;
 726
 727         /* This is a bit hairy.  If we are an SPE enabled  processor
 728          * (have embedded fp) we store the IEEE exception enable flags in
 729          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 730          * mode (asyn, precise, disabled) for 'Classic' FP. */
 731         if (val & PR_FP_EXC_SW_ENABLE) {
 732 #ifdef CONFIG_SPE
 733                 if (cpu_has_feature(CPU_FTR_SPE)) {
 734                         tsk->thread.fpexc_mode = val &
 735                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 736                         return 0;
 737                 } else {
 738                         return -EINVAL;
 739                 }
 740 #else
 741                 return -EINVAL;
 742 #endif
 743         }
 744
 745         /* on a CONFIG_SPE this does not hurt us.  The bits that
 746          * __pack_fe01 use do not overlap with bits used for
 747          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 748          * on CONFIG_SPE implementations are reserved so writing to
 749          * them does not change anything */
 750         if (val > PR_FP_EXC_PRECISE)
 751                 return -EINVAL;
 752         tsk->thread.fpexc_mode = __pack_fe01(val);
 753         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 754                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 755                         | tsk->thread.fpexc_mode;
 756         return 0;
 757 }
 758
 759 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 760 {
 761         unsigned int val;
 762
 763         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 764 #ifdef CONFIG_SPE
 765                 if (cpu_has_feature(CPU_FTR_SPE))
 766                         val = tsk->thread.fpexc_mode;
 767                 else
 768                         return -EINVAL;
 769 #else
 770                 return -EINVAL;
 771 #endif
 772         else
 773                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 774         return put_user(val, (unsigned int __user *) adr);
 775 }
 776
 777 int set_endian(struct task_struct *tsk, unsigned int val)
 778 {
 779         struct pt_regs *regs = tsk->thread.regs;
 780
 781         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 782             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 783                 return -EINVAL;
 784
 785         if (regs == NULL)
 786                 return -EINVAL;
 787
 788         if (val == PR_ENDIAN_BIG)
 789                 regs->msr &= ~MSR_LE;
 790         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 791                 regs->msr |= MSR_LE;
 792         else
 793                 return -EINVAL;
 794
 795         return 0;
 796 }
 797
 798 int get_endian(struct task_struct *tsk, unsigned long adr)
 799 {
 800         struct pt_regs *regs = tsk->thread.regs;
 801         unsigned int val;
 802
 803         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 804             !cpu_has_feature(CPU_FTR_REAL_LE))
 805                 return -EINVAL;
 806
 807         if (regs == NULL)
 808                 return -EINVAL;
 809
 810         if (regs->msr & MSR_LE) {
 811                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 812                         val = PR_ENDIAN_LITTLE;
 813                 else
 814                         val = PR_ENDIAN_PPC_LITTLE;
 815         } else
 816                 val = PR_ENDIAN_BIG;
 817
 818         return put_user(val, (unsigned int __user *)adr);
 819 }
 820
 821 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 822 {
 823         tsk->thread.align_ctl = val;
 824         return 0;
 825 }
 826
 827 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 828 {
 829         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 830 }
 831
 832 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 833
 834 int sys_clone(unsigned long clone_flags, unsigned long usp,
 835               int __user *parent_tidp, void __user *child_threadptr,
 836               int __user *child_tidp, int p6,
 837               struct pt_regs *regs)
 838 {
 839         CHECK_FULL_REGS(regs);
 840         if (usp == 0)
 841                 usp = regs->gpr[1];     /* stack pointer for child */
 842 #ifdef CONFIG_PPC64
 843         if (test_thread_flag(TIF_32BIT)) {
 844                 parent_tidp = TRUNC_PTR(parent_tidp);
 845                 child_tidp = TRUNC_PTR(child_tidp);
 846         }
 847 #endif
 848         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 849 }
 850
 851 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 852              unsigned long p4, unsigned long p5, unsigned long p6,
 853              struct pt_regs *regs)
 854 {
 855         CHECK_FULL_REGS(regs);
 856         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 857 }
 858
 859 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
 860               unsigned long p4, unsigned long p5, unsigned long p6,
 861               struct pt_regs *regs)
 862 {
 863         CHECK_FULL_REGS(regs);
 864         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 865                         regs, 0, NULL, NULL);
 866 }
 867
 868 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 869                unsigned long a3, unsigned long a4, unsigned long a5,
 870                struct pt_regs *regs)
 871 {
 872         int error;
 873         char *filename;
 874
 875         filename = getname((char __user *) a0);
 876         error = PTR_ERR(filename);
 877         if (IS_ERR(filename))
 878                 goto out;
 879         flush_fp_to_thread(current);
 880         flush_altivec_to_thread(current);
 881         flush_spe_to_thread(current);
 882         error = do_execve(filename, (char __user * __user *) a1,
 883                           (char __user * __user *) a2, regs);
 884         putname(filename);
 885 out:
 886         return error;
 887 }
 888
 889 #ifdef CONFIG_IRQSTACKS
 890 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
 891                                   unsigned long nbytes)
 892 {
 893         unsigned long stack_page;
 894         unsigned long cpu = task_cpu(p);
 895
 896         /*
 897          * Avoid crashing if the stack has overflowed and corrupted
 898          * task_cpu(p), which is in the thread_info struct.
 899          */
 900         if (cpu < NR_CPUS && cpu_possible(cpu)) {
 901                 stack_page = (unsigned long) hardirq_ctx[cpu];
 902                 if (sp >= stack_page + sizeof(struct thread_struct)
 903                     && sp <= stack_page + THREAD_SIZE - nbytes)
 904                         return 1;
 905
 906                 stack_page = (unsigned long) softirq_ctx[cpu];
 907                 if (sp >= stack_page + sizeof(struct thread_struct)
 908                     && sp <= stack_page + THREAD_SIZE - nbytes)
 909                         return 1;
 910         }
 911         return 0;
 912 }
 913
 914 #else
 915 #define valid_irq_stack(sp, p, nb)      0
 916 #endif /* CONFIG_IRQSTACKS */
 917
 918 int validate_sp(unsigned long sp, struct task_struct *p,
 919                        unsigned long nbytes)
 920 {
 921         unsigned long stack_page = (unsigned long)task_stack_page(p);
 922
 923         if (sp >= stack_page + sizeof(struct thread_struct)
 924             && sp <= stack_page + THREAD_SIZE - nbytes)
 925                 return 1;
 926
 927         return valid_irq_stack(sp, p, nbytes);
 928 }
 929
 930 EXPORT_SYMBOL(validate_sp);
 931
 932 unsigned long get_wchan(struct task_struct *p)
 933 {
 934         unsigned long ip, sp;
 935         int count = 0;
 936
 937         if (!p || p == current || p->state == TASK_RUNNING)
 938                 return 0;
 939
 940         sp = p->thread.ksp;
 941         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 942                 return 0;
 943
 944         do {
 945                 sp = *(unsigned long *)sp;
 946                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 947                         return 0;
 948                 if (count > 0) {
 949                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
 950                         if (!in_sched_functions(ip))
 951                                 return ip;
 952                 }
 953         } while (count++ < 16);
 954         return 0;
 955 }
 956
 957 static int kstack_depth_to_print = 64;
 958
 959 void show_stack(struct task_struct *tsk, unsigned long *stack)
 960 {
 961         unsigned long sp, ip, lr, newsp;
 962         int count = 0;
 963         int firstframe = 1;
 964
 965         sp = (unsigned long) stack;
 966         if (tsk == NULL)
 967                 tsk = current;
 968         if (sp == 0) {
 969                 if (tsk == current)
 970                         asm("mr %0,1" : "=r" (sp));
 971                 else
 972                         sp = tsk->thread.ksp;
 973         }
 974
 975         lr = 0;
 976         printk("Call Trace:\n");
 977         do {
 978                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
 979                         return;
 980
 981                 stack = (unsigned long *) sp;
 982                 newsp = stack[0];
 983                 ip = stack[STACK_FRAME_LR_SAVE];
 984                 if (!firstframe || ip != lr) {
 985                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
 986                         if (firstframe)
 987                                 printk(" (unreliable)");
 988                         printk("\n");
 989                 }
 990                 firstframe = 0;
 991
 992                 /*
 993                  * See if this is an exception frame.
 994                  * We look for the "regshere" marker in the current frame.
 995                  */
 996                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
 997                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
 998                         struct pt_regs *regs = (struct pt_regs *)
 999                                 (sp + STACK_FRAME_OVERHEAD);
1000                         lr = regs->link;
1001                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1002                                regs->trap, (void *)regs->nip, (void *)lr);
1003                         firstframe = 1;
1004                 }
1005
1006                 sp = newsp;
1007         } while (count++ < kstack_depth_to_print);
1008 }
1009
1010 void dump_stack(void)
1011 {
1012         show_stack(current, NULL);
1013 }
1014 EXPORT_SYMBOL(dump_stack);
1015
1016 #ifdef CONFIG_PPC64
1017 void ppc64_runlatch_on(void)
1018 {
1019         unsigned long ctrl;
1020
1021         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1022                 HMT_medium();
1023
1024                 ctrl = mfspr(SPRN_CTRLF);
1025                 ctrl |= CTRL_RUNLATCH;
1026                 mtspr(SPRN_CTRLT, ctrl);
1027
1028                 set_thread_flag(TIF_RUNLATCH);
1029         }
1030 }
1031
1032 void ppc64_runlatch_off(void)
1033 {
1034         unsigned long ctrl;
1035
1036         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1037                 HMT_medium();
1038
1039                 clear_thread_flag(TIF_RUNLATCH);
1040
1041                 ctrl = mfspr(SPRN_CTRLF);
1042                 ctrl &= ~CTRL_RUNLATCH;
1043                 mtspr(SPRN_CTRLT, ctrl);
1044         }
1045 }
1046 #endif
1047
1048 #if THREAD_SHIFT < PAGE_SHIFT
1049
1050 static struct kmem_cache *thread_info_cache;
1051
1052 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1053 {
1054         struct thread_info *ti;
1055
1056         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1057         if (unlikely(ti == NULL))
1058                 return NULL;
1059 #ifdef CONFIG_DEBUG_STACK_USAGE
1060         memset(ti, 0, THREAD_SIZE);
1061 #endif
1062         return ti;
1063 }
1064
1065 void free_thread_info(struct thread_info *ti)
1066 {
1067         kmem_cache_free(thread_info_cache, ti);
1068 }
1069
1070 void thread_info_cache_init(void)
1071 {
1072         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1073                                               THREAD_SIZE, 0, NULL);
1074         BUG_ON(thread_info_cache == NULL);
1075 }
1076
1077 #endif /* THREAD_SHIFT < PAGE_SHIFT */