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 #include <linux/kernel_stat.h>
  37
  38 #include <asm/pgtable.h>
  39 #include <asm/uaccess.h>
  40 #include <asm/system.h>
  41 #include <asm/io.h>
  42 #include <asm/processor.h>
  43 #include <asm/mmu.h>
  44 #include <asm/prom.h>
  45 #include <asm/machdep.h>
  46 #include <asm/time.h>
  47 #include <asm/syscalls.h>
  48 #ifdef CONFIG_PPC64
  49 #include <asm/firmware.h>
  50 #endif
  51 #include <linux/kprobes.h>
  52 #include <linux/kdebug.h>
  53
  54 extern unsigned long _get_SP(void);
  55
  56 #ifndef CONFIG_SMP
  57 struct task_struct *last_task_used_math = NULL;
  58 struct task_struct *last_task_used_altivec = NULL;
  59 struct task_struct *last_task_used_vsx = NULL;
  60 struct task_struct *last_task_used_spe = NULL;
  61 #endif
  62
  63 /*
  64  * Make sure the floating-point register state in the
  65  * the thread_struct is up to date for task tsk.
  66  */
  67 void flush_fp_to_thread(struct task_struct *tsk)
  68 {
  69         if (tsk->thread.regs) {
  70                 /*
  71                  * We need to disable preemption here because if we didn't,
  72                  * another process could get scheduled after the regs->msr
  73                  * test but before we have finished saving the FP registers
  74                  * to the thread_struct.  That process could take over the
  75                  * FPU, and then when we get scheduled again we would store
  76                  * bogus values for the remaining FP registers.
  77                  */
  78                 preempt_disable();
  79                 if (tsk->thread.regs->msr & MSR_FP) {
  80 #ifdef CONFIG_SMP
  81                         /*
  82                          * This should only ever be called for current or
  83                          * for a stopped child process.  Since we save away
  84                          * the FP register state on context switch on SMP,
  85                          * there is something wrong if a stopped child appears
  86                          * to still have its FP state in the CPU registers.
  87                          */
  88                         BUG_ON(tsk != current);
  89 #endif
  90                         giveup_fpu(tsk);
  91                 }
  92                 preempt_enable();
  93         }
  94 }
  95
  96 void enable_kernel_fp(void)
  97 {
  98         WARN_ON(preemptible());
  99
 100 #ifdef CONFIG_SMP
 101         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
 102                 giveup_fpu(current);
 103         else
 104                 giveup_fpu(NULL);       /* just enables FP for kernel */
 105 #else
 106         giveup_fpu(last_task_used_math);
 107 #endif /* CONFIG_SMP */
 108 }
 109 EXPORT_SYMBOL(enable_kernel_fp);
 110
 111 #ifdef CONFIG_ALTIVEC
 112 void enable_kernel_altivec(void)
 113 {
 114         WARN_ON(preemptible());
 115
 116 #ifdef CONFIG_SMP
 117         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 118                 giveup_altivec(current);
 119         else
 120                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
 121 #else
 122         giveup_altivec(last_task_used_altivec);
 123 #endif /* CONFIG_SMP */
 124 }
 125 EXPORT_SYMBOL(enable_kernel_altivec);
 126
 127 /*
 128  * Make sure the VMX/Altivec register state in the
 129  * the thread_struct is up to date for task tsk.
 130  */
 131 void flush_altivec_to_thread(struct task_struct *tsk)
 132 {
 133         if (tsk->thread.regs) {
 134                 preempt_disable();
 135                 if (tsk->thread.regs->msr & MSR_VEC) {
 136 #ifdef CONFIG_SMP
 137                         BUG_ON(tsk != current);
 138 #endif
 139                         giveup_altivec(tsk);
 140                 }
 141                 preempt_enable();
 142         }
 143 }
 144 #endif /* CONFIG_ALTIVEC */
 145
 146 #ifdef CONFIG_VSX
 147 #if 0
 148 /* not currently used, but some crazy RAID module might want to later */
 149 void enable_kernel_vsx(void)
 150 {
 151         WARN_ON(preemptible());
 152
 153 #ifdef CONFIG_SMP
 154         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 155                 giveup_vsx(current);
 156         else
 157                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 158 #else
 159         giveup_vsx(last_task_used_vsx);
 160 #endif /* CONFIG_SMP */
 161 }
 162 EXPORT_SYMBOL(enable_kernel_vsx);
 163 #endif
 164
 165 void giveup_vsx(struct task_struct *tsk)
 166 {
 167         giveup_fpu(tsk);
 168         giveup_altivec(tsk);
 169         __giveup_vsx(tsk);
 170 }
 171
 172 void flush_vsx_to_thread(struct task_struct *tsk)
 173 {
 174         if (tsk->thread.regs) {
 175                 preempt_disable();
 176                 if (tsk->thread.regs->msr & MSR_VSX) {
 177 #ifdef CONFIG_SMP
 178                         BUG_ON(tsk != current);
 179 #endif
 180                         giveup_vsx(tsk);
 181                 }
 182                 preempt_enable();
 183         }
 184 }
 185 #endif /* CONFIG_VSX */
 186
 187 #ifdef CONFIG_SPE
 188
 189 void enable_kernel_spe(void)
 190 {
 191         WARN_ON(preemptible());
 192
 193 #ifdef CONFIG_SMP
 194         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 195                 giveup_spe(current);
 196         else
 197                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 198 #else
 199         giveup_spe(last_task_used_spe);
 200 #endif /* __SMP __ */
 201 }
 202 EXPORT_SYMBOL(enable_kernel_spe);
 203
 204 void flush_spe_to_thread(struct task_struct *tsk)
 205 {
 206         if (tsk->thread.regs) {
 207                 preempt_disable();
 208                 if (tsk->thread.regs->msr & MSR_SPE) {
 209 #ifdef CONFIG_SMP
 210                         BUG_ON(tsk != current);
 211 #endif
 212                         giveup_spe(tsk);
 213                 }
 214                 preempt_enable();
 215         }
 216 }
 217 #endif /* CONFIG_SPE */
 218
 219 #ifndef CONFIG_SMP
 220 /*
 221  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 222  * and the current task has some state, discard it.
 223  */
 224 void discard_lazy_cpu_state(void)
 225 {
 226         preempt_disable();
 227         if (last_task_used_math == current)
 228                 last_task_used_math = NULL;
 229 #ifdef CONFIG_ALTIVEC
 230         if (last_task_used_altivec == current)
 231                 last_task_used_altivec = NULL;
 232 #endif /* CONFIG_ALTIVEC */
 233 #ifdef CONFIG_VSX
 234         if (last_task_used_vsx == current)
 235                 last_task_used_vsx = NULL;
 236 #endif /* CONFIG_VSX */
 237 #ifdef CONFIG_SPE
 238         if (last_task_used_spe == current)
 239                 last_task_used_spe = NULL;
 240 #endif
 241         preempt_enable();
 242 }
 243 #endif /* CONFIG_SMP */
 244
 245 void do_dabr(struct pt_regs *regs, unsigned long address,
 246                     unsigned long error_code)
 247 {
 248         siginfo_t info;
 249
 250         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 251                         11, SIGSEGV) == NOTIFY_STOP)
 252                 return;
 253
 254         if (debugger_dabr_match(regs))
 255                 return;
 256
 257         /* Clear the DAC and struct entries.  One shot trigger */
 258 #if defined(CONFIG_BOOKE)
 259         mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~(DBSR_DAC1R | DBSR_DAC1W
 260                                                         | DBCR0_IDM));
 261 #endif
 262
 263         /* Clear the DABR */
 264         set_dabr(0);
 265
 266         /* Deliver the signal to userspace */
 267         info.si_signo = SIGTRAP;
 268         info.si_errno = 0;
 269         info.si_code = TRAP_HWBKPT;
 270         info.si_addr = (void __user *)address;
 271         force_sig_info(SIGTRAP, &info, current);
 272 }
 273
 274 static DEFINE_PER_CPU(unsigned long, current_dabr);
 275
 276 int set_dabr(unsigned long dabr)
 277 {
 278         __get_cpu_var(current_dabr) = dabr;
 279
 280         if (ppc_md.set_dabr)
 281                 return ppc_md.set_dabr(dabr);
 282
 283         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 284 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
 285         mtspr(SPRN_DABR, dabr);
 286 #endif
 287
 288 #if defined(CONFIG_BOOKE)
 289         mtspr(SPRN_DAC1, dabr);
 290 #endif
 291
 292         return 0;
 293 }
 294
 295 #ifdef CONFIG_PPC64
 296 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 297 #endif
 298
 299 struct task_struct *__switch_to(struct task_struct *prev,
 300         struct task_struct *new)
 301 {
 302         struct thread_struct *new_thread, *old_thread;
 303         unsigned long flags;
 304         struct task_struct *last;
 305
 306 #ifdef CONFIG_SMP
 307         /* avoid complexity of lazy save/restore of fpu
 308          * by just saving it every time we switch out if
 309          * this task used the fpu during the last quantum.
 310          *
 311          * If it tries to use the fpu again, it'll trap and
 312          * reload its fp regs.  So we don't have to do a restore
 313          * every switch, just a save.
 314          *  -- Cort
 315          */
 316         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 317                 giveup_fpu(prev);
 318 #ifdef CONFIG_ALTIVEC
 319         /*
 320          * If the previous thread used altivec in the last quantum
 321          * (thus changing altivec regs) then save them.
 322          * We used to check the VRSAVE register but not all apps
 323          * set it, so we don't rely on it now (and in fact we need
 324          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 325          *
 326          * On SMP we always save/restore altivec regs just to avoid the
 327          * complexity of changing processors.
 328          *  -- Cort
 329          */
 330         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 331                 giveup_altivec(prev);
 332 #endif /* CONFIG_ALTIVEC */
 333 #ifdef CONFIG_VSX
 334         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 335                 /* VMX and FPU registers are already save here */
 336                 __giveup_vsx(prev);
 337 #endif /* CONFIG_VSX */
 338 #ifdef CONFIG_SPE
 339         /*
 340          * If the previous thread used spe in the last quantum
 341          * (thus changing spe regs) then save them.
 342          *
 343          * On SMP we always save/restore spe regs just to avoid the
 344          * complexity of changing processors.
 345          */
 346         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 347                 giveup_spe(prev);
 348 #endif /* CONFIG_SPE */
 349
 350 #else  /* CONFIG_SMP */
 351 #ifdef CONFIG_ALTIVEC
 352         /* Avoid the trap.  On smp this this never happens since
 353          * we don't set last_task_used_altivec -- Cort
 354          */
 355         if (new->thread.regs && last_task_used_altivec == new)
 356                 new->thread.regs->msr |= MSR_VEC;
 357 #endif /* CONFIG_ALTIVEC */
 358 #ifdef CONFIG_VSX
 359         if (new->thread.regs && last_task_used_vsx == new)
 360                 new->thread.regs->msr |= MSR_VSX;
 361 #endif /* CONFIG_VSX */
 362 #ifdef CONFIG_SPE
 363         /* Avoid the trap.  On smp this this never happens since
 364          * we don't set last_task_used_spe
 365          */
 366         if (new->thread.regs && last_task_used_spe == new)
 367                 new->thread.regs->msr |= MSR_SPE;
 368 #endif /* CONFIG_SPE */
 369
 370 #endif /* CONFIG_SMP */
 371
 372         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 373                 set_dabr(new->thread.dabr);
 374
 375 #if defined(CONFIG_BOOKE)
 376         /* If new thread DAC (HW breakpoint) is the same then leave it */
 377         if (new->thread.dabr)
 378                 set_dabr(new->thread.dabr);
 379 #endif
 380
 381         new_thread = &new->thread;
 382         old_thread = &current->thread;
 383
 384 #ifdef CONFIG_PPC64
 385         /*
 386          * Collect processor utilization data per process
 387          */
 388         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 389                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 390                 long unsigned start_tb, current_tb;
 391                 start_tb = old_thread->start_tb;
 392                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 393                 old_thread->accum_tb += (current_tb - start_tb);
 394                 new_thread->start_tb = current_tb;
 395         }
 396 #endif
 397
 398         local_irq_save(flags);
 399
 400         account_system_vtime(current);
 401         account_process_vtime(current);
 402         calculate_steal_time();
 403
 404         /*
 405          * We can't take a PMU exception inside _switch() since there is a
 406          * window where the kernel stack SLB and the kernel stack are out
 407          * of sync. Hard disable here.
 408          */
 409         hard_irq_disable();
 410         last = _switch(old_thread, new_thread);
 411
 412         local_irq_restore(flags);
 413
 414         return last;
 415 }
 416
 417 static int instructions_to_print = 16;
 418
 419 static void show_instructions(struct pt_regs *regs)
 420 {
 421         int i;
 422         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 423                         sizeof(int));
 424
 425         printk("Instruction dump:");
 426
 427         for (i = 0; i < instructions_to_print; i++) {
 428                 int instr;
 429
 430                 if (!(i % 8))
 431                         printk("\n");
 432
 433 #if !defined(CONFIG_BOOKE)
 434                 /* If executing with the IMMU off, adjust pc rather
 435                  * than print XXXXXXXX.
 436                  */
 437                 if (!(regs->msr & MSR_IR))
 438                         pc = (unsigned long)phys_to_virt(pc);
 439 #endif
 440
 441                 /* We use __get_user here *only* to avoid an OOPS on a
 442                  * bad address because the pc *should* only be a
 443                  * kernel address.
 444                  */
 445                 if (!__kernel_text_address(pc) ||
 446                      __get_user(instr, (unsigned int __user *)pc)) {
 447                         printk("XXXXXXXX ");
 448                 } else {
 449                         if (regs->nip == pc)
 450                                 printk("<%08x> ", instr);
 451                         else
 452                                 printk("%08x ", instr);
 453                 }
 454
 455                 pc += sizeof(int);
 456         }
 457
 458         printk("\n");
 459 }
 460
 461 static struct regbit {
 462         unsigned long bit;
 463         const char *name;
 464 } msr_bits[] = {
 465         {MSR_EE,        "EE"},
 466         {MSR_PR,        "PR"},
 467         {MSR_FP,        "FP"},
 468         {MSR_VEC,       "VEC"},
 469         {MSR_VSX,       "VSX"},
 470         {MSR_ME,        "ME"},
 471         {MSR_CE,        "CE"},
 472         {MSR_DE,        "DE"},
 473         {MSR_IR,        "IR"},
 474         {MSR_DR,        "DR"},
 475         {0,             NULL}
 476 };
 477
 478 static void printbits(unsigned long val, struct regbit *bits)
 479 {
 480         const char *sep = "";
 481
 482         printk("<");
 483         for (; bits->bit; ++bits)
 484                 if (val & bits->bit) {
 485                         printk("%s%s", sep, bits->name);
 486                         sep = ",";
 487                 }
 488         printk(">");
 489 }
 490
 491 #ifdef CONFIG_PPC64
 492 #define REG             "%016lx"
 493 #define REGS_PER_LINE   4
 494 #define LAST_VOLATILE   13
 495 #else
 496 #define REG             "%08lx"
 497 #define REGS_PER_LINE   8
 498 #define LAST_VOLATILE   12
 499 #endif
 500
 501 void show_regs(struct pt_regs * regs)
 502 {
 503         int i, trap;
 504
 505         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 506                regs->nip, regs->link, regs->ctr);
 507         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 508                regs, regs->trap, print_tainted(), init_utsname()->release);
 509         printk("MSR: "REG" ", regs->msr);
 510         printbits(regs->msr, msr_bits);
 511         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 512         trap = TRAP(regs);
 513         if (trap == 0x300 || trap == 0x600)
 514 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 515                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 516 #else
 517                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 518 #endif
 519         printk("TASK = %p[%d] '%s' THREAD: %p",
 520                current, task_pid_nr(current), current->comm, task_thread_info(current));
 521
 522 #ifdef CONFIG_SMP
 523         printk(" CPU: %d", raw_smp_processor_id());
 524 #endif /* CONFIG_SMP */
 525
 526         for (i = 0;  i < 32;  i++) {
 527                 if ((i % REGS_PER_LINE) == 0)
 528                         printk("\n" KERN_INFO "GPR%02d: ", i);
 529                 printk(REG " ", regs->gpr[i]);
 530                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 531                         break;
 532         }
 533         printk("\n");
 534 #ifdef CONFIG_KALLSYMS
 535         /*
 536          * Lookup NIP late so we have the best change of getting the
 537          * above info out without failing
 538          */
 539         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 540         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 541 #endif
 542         show_stack(current, (unsigned long *) regs->gpr[1]);
 543         if (!user_mode(regs))
 544                 show_instructions(regs);
 545 }
 546
 547 void exit_thread(void)
 548 {
 549         discard_lazy_cpu_state();
 550 }
 551
 552 void flush_thread(void)
 553 {
 554 #ifdef CONFIG_PPC64
 555         struct thread_info *t = current_thread_info();
 556
 557         if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
 558                 clear_ti_thread_flag(t, TIF_ABI_PENDING);
 559                 if (test_ti_thread_flag(t, TIF_32BIT))
 560                         clear_ti_thread_flag(t, TIF_32BIT);
 561                 else
 562                         set_ti_thread_flag(t, TIF_32BIT);
 563         }
 564 #endif
 565
 566         discard_lazy_cpu_state();
 567
 568         if (current->thread.dabr) {
 569                 current->thread.dabr = 0;
 570                 set_dabr(0);
 571
 572 #if defined(CONFIG_BOOKE)
 573                 current->thread.dbcr0 &= ~(DBSR_DAC1R | DBSR_DAC1W);
 574 #endif
 575         }
 576 }
 577
 578 void
 579 release_thread(struct task_struct *t)
 580 {
 581 }
 582
 583 /*
 584  * This gets called before we allocate a new thread and copy
 585  * the current task into it.
 586  */
 587 void prepare_to_copy(struct task_struct *tsk)
 588 {
 589         flush_fp_to_thread(current);
 590         flush_altivec_to_thread(current);
 591         flush_vsx_to_thread(current);
 592         flush_spe_to_thread(current);
 593 }
 594
 595 /*
 596  * Copy a thread..
 597  */
 598 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 599                 unsigned long unused, struct task_struct *p,
 600                 struct pt_regs *regs)
 601 {
 602         struct pt_regs *childregs, *kregs;
 603         extern void ret_from_fork(void);
 604         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 605
 606         CHECK_FULL_REGS(regs);
 607         /* Copy registers */
 608         sp -= sizeof(struct pt_regs);
 609         childregs = (struct pt_regs *) sp;
 610         *childregs = *regs;
 611         if ((childregs->msr & MSR_PR) == 0) {
 612                 /* for kernel thread, set `current' and stackptr in new task */
 613                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 614 #ifdef CONFIG_PPC32
 615                 childregs->gpr[2] = (unsigned long) p;
 616 #else
 617                 clear_tsk_thread_flag(p, TIF_32BIT);
 618 #endif
 619                 p->thread.regs = NULL;  /* no user register state */
 620         } else {
 621                 childregs->gpr[1] = usp;
 622                 p->thread.regs = childregs;
 623                 if (clone_flags & CLONE_SETTLS) {
 624 #ifdef CONFIG_PPC64
 625                         if (!test_thread_flag(TIF_32BIT))
 626                                 childregs->gpr[13] = childregs->gpr[6];
 627                         else
 628 #endif
 629                                 childregs->gpr[2] = childregs->gpr[6];
 630                 }
 631         }
 632         childregs->gpr[3] = 0;  /* Result from fork() */
 633         sp -= STACK_FRAME_OVERHEAD;
 634
 635         /*
 636          * The way this works is that at some point in the future
 637          * some task will call _switch to switch to the new task.
 638          * That will pop off the stack frame created below and start
 639          * the new task running at ret_from_fork.  The new task will
 640          * do some house keeping and then return from the fork or clone
 641          * system call, using the stack frame created above.
 642          */
 643         sp -= sizeof(struct pt_regs);
 644         kregs = (struct pt_regs *) sp;
 645         sp -= STACK_FRAME_OVERHEAD;
 646         p->thread.ksp = sp;
 647         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 648                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 649
 650 #ifdef CONFIG_PPC64
 651         if (cpu_has_feature(CPU_FTR_SLB)) {
 652                 unsigned long sp_vsid;
 653                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 654
 655                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
 656                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 657                                 << SLB_VSID_SHIFT_1T;
 658                 else
 659                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 660                                 << SLB_VSID_SHIFT;
 661                 sp_vsid |= SLB_VSID_KERNEL | llp;
 662                 p->thread.ksp_vsid = sp_vsid;
 663         }
 664
 665         /*
 666          * The PPC64 ABI makes use of a TOC to contain function
 667          * pointers.  The function (ret_from_except) is actually a pointer
 668          * to the TOC entry.  The first entry is a pointer to the actual
 669          * function.
 670          */
 671         kregs->nip = *((unsigned long *)ret_from_fork);
 672 #else
 673         kregs->nip = (unsigned long)ret_from_fork;
 674 #endif
 675
 676         return 0;
 677 }
 678
 679 /*
 680  * Set up a thread for executing a new program
 681  */
 682 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 683 {
 684 #ifdef CONFIG_PPC64
 685         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 686 #endif
 687
 688         set_fs(USER_DS);
 689
 690         /*
 691          * If we exec out of a kernel thread then thread.regs will not be
 692          * set.  Do it now.
 693          */
 694         if (!current->thread.regs) {
 695                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 696                 current->thread.regs = regs - 1;
 697         }
 698
 699         memset(regs->gpr, 0, sizeof(regs->gpr));
 700         regs->ctr = 0;
 701         regs->link = 0;
 702         regs->xer = 0;
 703         regs->ccr = 0;
 704         regs->gpr[1] = sp;
 705
 706         /*
 707          * We have just cleared all the nonvolatile GPRs, so make
 708          * FULL_REGS(regs) return true.  This is necessary to allow
 709          * ptrace to examine the thread immediately after exec.
 710          */
 711         regs->trap &= ~1UL;
 712
 713 #ifdef CONFIG_PPC32
 714         regs->mq = 0;
 715         regs->nip = start;
 716         regs->msr = MSR_USER;
 717 #else
 718         if (!test_thread_flag(TIF_32BIT)) {
 719                 unsigned long entry, toc;
 720
 721                 /* start is a relocated pointer to the function descriptor for
 722                  * the elf _start routine.  The first entry in the function
 723                  * descriptor is the entry address of _start and the second
 724                  * entry is the TOC value we need to use.
 725                  */
 726                 __get_user(entry, (unsigned long __user *)start);
 727                 __get_user(toc, (unsigned long __user *)start+1);
 728
 729                 /* Check whether the e_entry function descriptor entries
 730                  * need to be relocated before we can use them.
 731                  */
 732                 if (load_addr != 0) {
 733                         entry += load_addr;
 734                         toc   += load_addr;
 735                 }
 736                 regs->nip = entry;
 737                 regs->gpr[2] = toc;
 738                 regs->msr = MSR_USER64;
 739         } else {
 740                 regs->nip = start;
 741                 regs->gpr[2] = 0;
 742                 regs->msr = MSR_USER32;
 743         }
 744 #endif
 745
 746         discard_lazy_cpu_state();
 747 #ifdef CONFIG_VSX
 748         current->thread.used_vsr = 0;
 749 #endif
 750         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 751         current->thread.fpscr.val = 0;
 752 #ifdef CONFIG_ALTIVEC
 753         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 754         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 755         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 756         current->thread.vrsave = 0;
 757         current->thread.used_vr = 0;
 758 #endif /* CONFIG_ALTIVEC */
 759 #ifdef CONFIG_SPE
 760         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 761         current->thread.acc = 0;
 762         current->thread.spefscr = 0;
 763         current->thread.used_spe = 0;
 764 #endif /* CONFIG_SPE */
 765 }
 766
 767 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 768                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 769
 770 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 771 {
 772         struct pt_regs *regs = tsk->thread.regs;
 773
 774         /* This is a bit hairy.  If we are an SPE enabled  processor
 775          * (have embedded fp) we store the IEEE exception enable flags in
 776          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 777          * mode (asyn, precise, disabled) for 'Classic' FP. */
 778         if (val & PR_FP_EXC_SW_ENABLE) {
 779 #ifdef CONFIG_SPE
 780                 if (cpu_has_feature(CPU_FTR_SPE)) {
 781                         tsk->thread.fpexc_mode = val &
 782                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 783                         return 0;
 784                 } else {
 785                         return -EINVAL;
 786                 }
 787 #else
 788                 return -EINVAL;
 789 #endif
 790         }
 791
 792         /* on a CONFIG_SPE this does not hurt us.  The bits that
 793          * __pack_fe01 use do not overlap with bits used for
 794          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 795          * on CONFIG_SPE implementations are reserved so writing to
 796          * them does not change anything */
 797         if (val > PR_FP_EXC_PRECISE)
 798                 return -EINVAL;
 799         tsk->thread.fpexc_mode = __pack_fe01(val);
 800         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 801                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 802                         | tsk->thread.fpexc_mode;
 803         return 0;
 804 }
 805
 806 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 807 {
 808         unsigned int val;
 809
 810         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 811 #ifdef CONFIG_SPE
 812                 if (cpu_has_feature(CPU_FTR_SPE))
 813                         val = tsk->thread.fpexc_mode;
 814                 else
 815                         return -EINVAL;
 816 #else
 817                 return -EINVAL;
 818 #endif
 819         else
 820                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 821         return put_user(val, (unsigned int __user *) adr);
 822 }
 823
 824 int set_endian(struct task_struct *tsk, unsigned int val)
 825 {
 826         struct pt_regs *regs = tsk->thread.regs;
 827
 828         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 829             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 830                 return -EINVAL;
 831
 832         if (regs == NULL)
 833                 return -EINVAL;
 834
 835         if (val == PR_ENDIAN_BIG)
 836                 regs->msr &= ~MSR_LE;
 837         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 838                 regs->msr |= MSR_LE;
 839         else
 840                 return -EINVAL;
 841
 842         return 0;
 843 }
 844
 845 int get_endian(struct task_struct *tsk, unsigned long adr)
 846 {
 847         struct pt_regs *regs = tsk->thread.regs;
 848         unsigned int val;
 849
 850         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 851             !cpu_has_feature(CPU_FTR_REAL_LE))
 852                 return -EINVAL;
 853
 854         if (regs == NULL)
 855                 return -EINVAL;
 856
 857         if (regs->msr & MSR_LE) {
 858                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 859                         val = PR_ENDIAN_LITTLE;
 860                 else
 861                         val = PR_ENDIAN_PPC_LITTLE;
 862         } else
 863                 val = PR_ENDIAN_BIG;
 864
 865         return put_user(val, (unsigned int __user *)adr);
 866 }
 867
 868 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 869 {
 870         tsk->thread.align_ctl = val;
 871         return 0;
 872 }
 873
 874 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 875 {
 876         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 877 }
 878
 879 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 880
 881 int sys_clone(unsigned long clone_flags, unsigned long usp,
 882               int __user *parent_tidp, void __user *child_threadptr,
 883               int __user *child_tidp, int p6,
 884               struct pt_regs *regs)
 885 {
 886         CHECK_FULL_REGS(regs);
 887         if (usp == 0)
 888                 usp = regs->gpr[1];     /* stack pointer for child */
 889 #ifdef CONFIG_PPC64
 890         if (test_thread_flag(TIF_32BIT)) {
 891                 parent_tidp = TRUNC_PTR(parent_tidp);
 892                 child_tidp = TRUNC_PTR(child_tidp);
 893         }
 894 #endif
 895         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 896 }
 897
 898 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 899              unsigned long p4, unsigned long p5, unsigned long p6,
 900              struct pt_regs *regs)
 901 {
 902         CHECK_FULL_REGS(regs);
 903         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 904 }
 905
 906 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
 907               unsigned long p4, unsigned long p5, unsigned long p6,
 908               struct pt_regs *regs)
 909 {
 910         CHECK_FULL_REGS(regs);
 911         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 912                         regs, 0, NULL, NULL);
 913 }
 914
 915 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 916                unsigned long a3, unsigned long a4, unsigned long a5,
 917                struct pt_regs *regs)
 918 {
 919         int error;
 920         char *filename;
 921
 922         filename = getname((char __user *) a0);
 923         error = PTR_ERR(filename);
 924         if (IS_ERR(filename))
 925                 goto out;
 926         flush_fp_to_thread(current);
 927         flush_altivec_to_thread(current);
 928         flush_spe_to_thread(current);
 929         error = do_execve(filename, (char __user * __user *) a1,
 930                           (char __user * __user *) a2, regs);
 931         putname(filename);
 932 out:
 933         return error;
 934 }
 935
 936 #ifdef CONFIG_IRQSTACKS
 937 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
 938                                   unsigned long nbytes)
 939 {
 940         unsigned long stack_page;
 941         unsigned long cpu = task_cpu(p);
 942
 943         /*
 944          * Avoid crashing if the stack has overflowed and corrupted
 945          * task_cpu(p), which is in the thread_info struct.
 946          */
 947         if (cpu < NR_CPUS && cpu_possible(cpu)) {
 948                 stack_page = (unsigned long) hardirq_ctx[cpu];
 949                 if (sp >= stack_page + sizeof(struct thread_struct)
 950                     && sp <= stack_page + THREAD_SIZE - nbytes)
 951                         return 1;
 952
 953                 stack_page = (unsigned long) softirq_ctx[cpu];
 954                 if (sp >= stack_page + sizeof(struct thread_struct)
 955                     && sp <= stack_page + THREAD_SIZE - nbytes)
 956                         return 1;
 957         }
 958         return 0;
 959 }
 960
 961 #else
 962 #define valid_irq_stack(sp, p, nb)      0
 963 #endif /* CONFIG_IRQSTACKS */
 964
 965 int validate_sp(unsigned long sp, struct task_struct *p,
 966                        unsigned long nbytes)
 967 {
 968         unsigned long stack_page = (unsigned long)task_stack_page(p);
 969
 970         if (sp >= stack_page + sizeof(struct thread_struct)
 971             && sp <= stack_page + THREAD_SIZE - nbytes)
 972                 return 1;
 973
 974         return valid_irq_stack(sp, p, nbytes);
 975 }
 976
 977 EXPORT_SYMBOL(validate_sp);
 978
 979 unsigned long get_wchan(struct task_struct *p)
 980 {
 981         unsigned long ip, sp;
 982         int count = 0;
 983
 984         if (!p || p == current || p->state == TASK_RUNNING)
 985                 return 0;
 986
 987         sp = p->thread.ksp;
 988         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 989                 return 0;
 990
 991         do {
 992                 sp = *(unsigned long *)sp;
 993                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 994                         return 0;
 995                 if (count > 0) {
 996                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
 997                         if (!in_sched_functions(ip))
 998                                 return ip;
 999                 }
1000         } while (count++ < 16);
1001         return 0;
1002 }
1003
1004 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1005
1006 void show_stack(struct task_struct *tsk, unsigned long *stack)
1007 {
1008         unsigned long sp, ip, lr, newsp;
1009         int count = 0;
1010         int firstframe = 1;
1011
1012         sp = (unsigned long) stack;
1013         if (tsk == NULL)
1014                 tsk = current;
1015         if (sp == 0) {
1016                 if (tsk == current)
1017                         asm("mr %0,1" : "=r" (sp));
1018                 else
1019                         sp = tsk->thread.ksp;
1020         }
1021
1022         lr = 0;
1023         printk("Call Trace:\n");
1024         do {
1025                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1026                         return;
1027
1028                 stack = (unsigned long *) sp;
1029                 newsp = stack[0];
1030                 ip = stack[STACK_FRAME_LR_SAVE];
1031                 if (!firstframe || ip != lr) {
1032                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1033                         if (firstframe)
1034                                 printk(" (unreliable)");
1035                         printk("\n");
1036                 }
1037                 firstframe = 0;
1038
1039                 /*
1040                  * See if this is an exception frame.
1041                  * We look for the "regshere" marker in the current frame.
1042                  */
1043                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1044                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1045                         struct pt_regs *regs = (struct pt_regs *)
1046                                 (sp + STACK_FRAME_OVERHEAD);
1047                         lr = regs->link;
1048                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1049                                regs->trap, (void *)regs->nip, (void *)lr);
1050                         firstframe = 1;
1051                 }
1052
1053                 sp = newsp;
1054         } while (count++ < kstack_depth_to_print);
1055 }
1056
1057 void dump_stack(void)
1058 {
1059         show_stack(current, NULL);
1060 }
1061 EXPORT_SYMBOL(dump_stack);
1062
1063 #ifdef CONFIG_PPC64
1064 void ppc64_runlatch_on(void)
1065 {
1066         unsigned long ctrl;
1067
1068         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1069                 HMT_medium();
1070
1071                 ctrl = mfspr(SPRN_CTRLF);
1072                 ctrl |= CTRL_RUNLATCH;
1073                 mtspr(SPRN_CTRLT, ctrl);
1074
1075                 set_thread_flag(TIF_RUNLATCH);
1076         }
1077 }
1078
1079 void ppc64_runlatch_off(void)
1080 {
1081         unsigned long ctrl;
1082
1083         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1084                 HMT_medium();
1085
1086                 clear_thread_flag(TIF_RUNLATCH);
1087
1088                 ctrl = mfspr(SPRN_CTRLF);
1089                 ctrl &= ~CTRL_RUNLATCH;
1090                 mtspr(SPRN_CTRLT, ctrl);
1091         }
1092 }
1093 #endif
1094
1095 #if THREAD_SHIFT < PAGE_SHIFT
1096
1097 static struct kmem_cache *thread_info_cache;
1098
1099 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1100 {
1101         struct thread_info *ti;
1102
1103         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1104         if (unlikely(ti == NULL))
1105                 return NULL;
1106 #ifdef CONFIG_DEBUG_STACK_USAGE
1107         memset(ti, 0, THREAD_SIZE);
1108 #endif
1109         return ti;
1110 }
1111
1112 void free_thread_info(struct thread_info *ti)
1113 {
1114         kmem_cache_free(thread_info_cache, ti);
1115 }
1116
1117 void thread_info_cache_init(void)
1118 {
1119         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1120                                               THREAD_SIZE, 0, NULL);
1121         BUG_ON(thread_info_cache == NULL);
1122 }
1123
1124 #endif /* THREAD_SHIFT < PAGE_SHIFT */