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