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_spe = NULL;
  57 #endif
  58
  59 /*
  60  * Make sure the floating-point register state in the
  61  * the thread_struct is up to date for task tsk.
  62  */
  63 void flush_fp_to_thread(struct task_struct *tsk)
  64 {
  65         if (tsk->thread.regs) {
  66                 /*
  67                  * We need to disable preemption here because if we didn't,
  68                  * another process could get scheduled after the regs->msr
  69                  * test but before we have finished saving the FP registers
  70                  * to the thread_struct.  That process could take over the
  71                  * FPU, and then when we get scheduled again we would store
  72                  * bogus values for the remaining FP registers.
  73                  */
  74                 preempt_disable();
  75                 if (tsk->thread.regs->msr & MSR_FP) {
  76 #ifdef CONFIG_SMP
  77                         /*
  78                          * This should only ever be called for current or
  79                          * for a stopped child process.  Since we save away
  80                          * the FP register state on context switch on SMP,
  81                          * there is something wrong if a stopped child appears
  82                          * to still have its FP state in the CPU registers.
  83                          */
  84                         BUG_ON(tsk != current);
  85 #endif
  86                         giveup_fpu(tsk);
  87                 }
  88                 preempt_enable();
  89         }
  90 }
  91
  92 void enable_kernel_fp(void)
  93 {
  94         WARN_ON(preemptible());
  95
  96 #ifdef CONFIG_SMP
  97         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
  98                 giveup_fpu(current);
  99         else
 100                 giveup_fpu(NULL);       /* just enables FP for kernel */
 101 #else
 102         giveup_fpu(last_task_used_math);
 103 #endif /* CONFIG_SMP */
 104 }
 105 EXPORT_SYMBOL(enable_kernel_fp);
 106
 107 int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs)
 108 {
 109         if (!tsk->thread.regs)
 110                 return 0;
 111         flush_fp_to_thread(current);
 112
 113         memcpy(fpregs, &tsk->thread.fpr[0], sizeof(*fpregs));
 114
 115         return 1;
 116 }
 117
 118 #ifdef CONFIG_ALTIVEC
 119 void enable_kernel_altivec(void)
 120 {
 121         WARN_ON(preemptible());
 122
 123 #ifdef CONFIG_SMP
 124         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 125                 giveup_altivec(current);
 126         else
 127                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
 128 #else
 129         giveup_altivec(last_task_used_altivec);
 130 #endif /* CONFIG_SMP */
 131 }
 132 EXPORT_SYMBOL(enable_kernel_altivec);
 133
 134 /*
 135  * Make sure the VMX/Altivec register state in the
 136  * the thread_struct is up to date for task tsk.
 137  */
 138 void flush_altivec_to_thread(struct task_struct *tsk)
 139 {
 140         if (tsk->thread.regs) {
 141                 preempt_disable();
 142                 if (tsk->thread.regs->msr & MSR_VEC) {
 143 #ifdef CONFIG_SMP
 144                         BUG_ON(tsk != current);
 145 #endif
 146                         giveup_altivec(tsk);
 147                 }
 148                 preempt_enable();
 149         }
 150 }
 151
 152 int dump_task_altivec(struct task_struct *tsk, elf_vrregset_t *vrregs)
 153 {
 154         /* ELF_NVRREG includes the VSCR and VRSAVE which we need to save
 155          * separately, see below */
 156         const int nregs = ELF_NVRREG - 2;
 157         elf_vrreg_t *reg;
 158         u32 *dest;
 159
 160         if (tsk == current)
 161                 flush_altivec_to_thread(tsk);
 162
 163         reg = (elf_vrreg_t *)vrregs;
 164
 165         /* copy the 32 vr registers */
 166         memcpy(reg, &tsk->thread.vr[0], nregs * sizeof(*reg));
 167         reg += nregs;
 168
 169         /* copy the vscr */
 170         memcpy(reg, &tsk->thread.vscr, sizeof(*reg));
 171         reg++;
 172
 173         /* vrsave is stored in the high 32bit slot of the final 128bits */
 174         memset(reg, 0, sizeof(*reg));
 175         dest = (u32 *)reg;
 176         *dest = tsk->thread.vrsave;
 177
 178         return 1;
 179 }
 180 #endif /* CONFIG_ALTIVEC */
 181
 182 #ifdef CONFIG_SPE
 183
 184 void enable_kernel_spe(void)
 185 {
 186         WARN_ON(preemptible());
 187
 188 #ifdef CONFIG_SMP
 189         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 190                 giveup_spe(current);
 191         else
 192                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 193 #else
 194         giveup_spe(last_task_used_spe);
 195 #endif /* __SMP __ */
 196 }
 197 EXPORT_SYMBOL(enable_kernel_spe);
 198
 199 void flush_spe_to_thread(struct task_struct *tsk)
 200 {
 201         if (tsk->thread.regs) {
 202                 preempt_disable();
 203                 if (tsk->thread.regs->msr & MSR_SPE) {
 204 #ifdef CONFIG_SMP
 205                         BUG_ON(tsk != current);
 206 #endif
 207                         giveup_spe(tsk);
 208                 }
 209                 preempt_enable();
 210         }
 211 }
 212
 213 int dump_spe(struct pt_regs *regs, elf_vrregset_t *evrregs)
 214 {
 215         flush_spe_to_thread(current);
 216         /* We copy u32 evr[32] + u64 acc + u32 spefscr -> 35 */
 217         memcpy(evrregs, &current->thread.evr[0], sizeof(u32) * 35);
 218         return 1;
 219 }
 220 #endif /* CONFIG_SPE */
 221
 222 #ifndef CONFIG_SMP
 223 /*
 224  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 225  * and the current task has some state, discard it.
 226  */
 227 void discard_lazy_cpu_state(void)
 228 {
 229         preempt_disable();
 230         if (last_task_used_math == current)
 231                 last_task_used_math = NULL;
 232 #ifdef CONFIG_ALTIVEC
 233         if (last_task_used_altivec == current)
 234                 last_task_used_altivec = NULL;
 235 #endif /* CONFIG_ALTIVEC */
 236 #ifdef CONFIG_SPE
 237         if (last_task_used_spe == current)
 238                 last_task_used_spe = NULL;
 239 #endif
 240         preempt_enable();
 241 }
 242 #endif /* CONFIG_SMP */
 243
 244 static DEFINE_PER_CPU(unsigned long, current_dabr);
 245
 246 int set_dabr(unsigned long dabr)
 247 {
 248         __get_cpu_var(current_dabr) = dabr;
 249
 250 #ifdef CONFIG_PPC_MERGE         /* XXX for now */
 251         if (ppc_md.set_dabr)
 252                 return ppc_md.set_dabr(dabr);
 253 #endif
 254
 255         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 256 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
 257         mtspr(SPRN_DABR, dabr);
 258 #endif
 259         return 0;
 260 }
 261
 262 #ifdef CONFIG_PPC64
 263 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 264 #endif
 265
 266 struct task_struct *__switch_to(struct task_struct *prev,
 267         struct task_struct *new)
 268 {
 269         struct thread_struct *new_thread, *old_thread;
 270         unsigned long flags;
 271         struct task_struct *last;
 272
 273 #ifdef CONFIG_SMP
 274         /* avoid complexity of lazy save/restore of fpu
 275          * by just saving it every time we switch out if
 276          * this task used the fpu during the last quantum.
 277          *
 278          * If it tries to use the fpu again, it'll trap and
 279          * reload its fp regs.  So we don't have to do a restore
 280          * every switch, just a save.
 281          *  -- Cort
 282          */
 283         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 284                 giveup_fpu(prev);
 285 #ifdef CONFIG_ALTIVEC
 286         /*
 287          * If the previous thread used altivec in the last quantum
 288          * (thus changing altivec regs) then save them.
 289          * We used to check the VRSAVE register but not all apps
 290          * set it, so we don't rely on it now (and in fact we need
 291          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 292          *
 293          * On SMP we always save/restore altivec regs just to avoid the
 294          * complexity of changing processors.
 295          *  -- Cort
 296          */
 297         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 298                 giveup_altivec(prev);
 299 #endif /* CONFIG_ALTIVEC */
 300 #ifdef CONFIG_SPE
 301         /*
 302          * If the previous thread used spe in the last quantum
 303          * (thus changing spe regs) then save them.
 304          *
 305          * On SMP we always save/restore spe regs just to avoid the
 306          * complexity of changing processors.
 307          */
 308         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 309                 giveup_spe(prev);
 310 #endif /* CONFIG_SPE */
 311
 312 #else  /* CONFIG_SMP */
 313 #ifdef CONFIG_ALTIVEC
 314         /* Avoid the trap.  On smp this this never happens since
 315          * we don't set last_task_used_altivec -- Cort
 316          */
 317         if (new->thread.regs && last_task_used_altivec == new)
 318                 new->thread.regs->msr |= MSR_VEC;
 319 #endif /* CONFIG_ALTIVEC */
 320 #ifdef CONFIG_SPE
 321         /* Avoid the trap.  On smp this this never happens since
 322          * we don't set last_task_used_spe
 323          */
 324         if (new->thread.regs && last_task_used_spe == new)
 325                 new->thread.regs->msr |= MSR_SPE;
 326 #endif /* CONFIG_SPE */
 327
 328 #endif /* CONFIG_SMP */
 329
 330         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 331                 set_dabr(new->thread.dabr);
 332
 333         new_thread = &new->thread;
 334         old_thread = &current->thread;
 335
 336 #ifdef CONFIG_PPC64
 337         /*
 338          * Collect processor utilization data per process
 339          */
 340         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 341                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 342                 long unsigned start_tb, current_tb;
 343                 start_tb = old_thread->start_tb;
 344                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 345                 old_thread->accum_tb += (current_tb - start_tb);
 346                 new_thread->start_tb = current_tb;
 347         }
 348 #endif
 349
 350         local_irq_save(flags);
 351
 352         account_system_vtime(current);
 353         account_process_vtime(current);
 354         calculate_steal_time();
 355
 356         /*
 357          * We can't take a PMU exception inside _switch() since there is a
 358          * window where the kernel stack SLB and the kernel stack are out
 359          * of sync. Hard disable here.
 360          */
 361         hard_irq_disable();
 362         last = _switch(old_thread, new_thread);
 363
 364         local_irq_restore(flags);
 365
 366         return last;
 367 }
 368
 369 static int instructions_to_print = 16;
 370
 371 static void show_instructions(struct pt_regs *regs)
 372 {
 373         int i;
 374         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 375                         sizeof(int));
 376
 377         printk("Instruction dump:");
 378
 379         for (i = 0; i < instructions_to_print; i++) {
 380                 int instr;
 381
 382                 if (!(i % 8))
 383                         printk("\n");
 384
 385 #if !defined(CONFIG_BOOKE)
 386                 /* If executing with the IMMU off, adjust pc rather
 387                  * than print XXXXXXXX.
 388                  */
 389                 if (!(regs->msr & MSR_IR))
 390                         pc = (unsigned long)phys_to_virt(pc);
 391 #endif
 392
 393                 /* We use __get_user here *only* to avoid an OOPS on a
 394                  * bad address because the pc *should* only be a
 395                  * kernel address.
 396                  */
 397                 if (!__kernel_text_address(pc) ||
 398                      __get_user(instr, (unsigned int __user *)pc)) {
 399                         printk("XXXXXXXX ");
 400                 } else {
 401                         if (regs->nip == pc)
 402                                 printk("<%08x> ", instr);
 403                         else
 404                                 printk("%08x ", instr);
 405                 }
 406
 407                 pc += sizeof(int);
 408         }
 409
 410         printk("\n");
 411 }
 412
 413 static struct regbit {
 414         unsigned long bit;
 415         const char *name;
 416 } msr_bits[] = {
 417         {MSR_EE,        "EE"},
 418         {MSR_PR,        "PR"},
 419         {MSR_FP,        "FP"},
 420         {MSR_ME,        "ME"},
 421         {MSR_IR,        "IR"},
 422         {MSR_DR,        "DR"},
 423         {0,             NULL}
 424 };
 425
 426 static void printbits(unsigned long val, struct regbit *bits)
 427 {
 428         const char *sep = "";
 429
 430         printk("<");
 431         for (; bits->bit; ++bits)
 432                 if (val & bits->bit) {
 433                         printk("%s%s", sep, bits->name);
 434                         sep = ",";
 435                 }
 436         printk(">");
 437 }
 438
 439 #ifdef CONFIG_PPC64
 440 #define REG             "%016lx"
 441 #define REGS_PER_LINE   4
 442 #define LAST_VOLATILE   13
 443 #else
 444 #define REG             "%08lx"
 445 #define REGS_PER_LINE   8
 446 #define LAST_VOLATILE   12
 447 #endif
 448
 449 void show_regs(struct pt_regs * regs)
 450 {
 451         int i, trap;
 452
 453         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 454                regs->nip, regs->link, regs->ctr);
 455         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 456                regs, regs->trap, print_tainted(), init_utsname()->release);
 457         printk("MSR: "REG" ", regs->msr);
 458         printbits(regs->msr, msr_bits);
 459         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 460         trap = TRAP(regs);
 461         if (trap == 0x300 || trap == 0x600)
 462 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 463                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 464 #else
 465                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 466 #endif
 467         printk("TASK = %p[%d] '%s' THREAD: %p",
 468                current, task_pid_nr(current), current->comm, task_thread_info(current));
 469
 470 #ifdef CONFIG_SMP
 471         printk(" CPU: %d", raw_smp_processor_id());
 472 #endif /* CONFIG_SMP */
 473
 474         for (i = 0;  i < 32;  i++) {
 475                 if ((i % REGS_PER_LINE) == 0)
 476                         printk("\n" KERN_INFO "GPR%02d: ", i);
 477                 printk(REG " ", regs->gpr[i]);
 478                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 479                         break;
 480         }
 481         printk("\n");
 482 #ifdef CONFIG_KALLSYMS
 483         /*
 484          * Lookup NIP late so we have the best change of getting the
 485          * above info out without failing
 486          */
 487         printk("NIP ["REG"] ", regs->nip);
 488         print_symbol("%s\n", regs->nip);
 489         printk("LR ["REG"] ", regs->link);
 490         print_symbol("%s\n", regs->link);
 491 #endif
 492         show_stack(current, (unsigned long *) regs->gpr[1]);
 493         if (!user_mode(regs))
 494                 show_instructions(regs);
 495 }
 496
 497 void exit_thread(void)
 498 {
 499         discard_lazy_cpu_state();
 500 }
 501
 502 void flush_thread(void)
 503 {
 504 #ifdef CONFIG_PPC64
 505         struct thread_info *t = current_thread_info();
 506
 507         if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
 508                 clear_ti_thread_flag(t, TIF_ABI_PENDING);
 509                 if (test_ti_thread_flag(t, TIF_32BIT))
 510                         clear_ti_thread_flag(t, TIF_32BIT);
 511                 else
 512                         set_ti_thread_flag(t, TIF_32BIT);
 513         }
 514 #endif
 515
 516         discard_lazy_cpu_state();
 517
 518         if (current->thread.dabr) {
 519                 current->thread.dabr = 0;
 520                 set_dabr(0);
 521         }
 522 }
 523
 524 void
 525 release_thread(struct task_struct *t)
 526 {
 527 }
 528
 529 /*
 530  * This gets called before we allocate a new thread and copy
 531  * the current task into it.
 532  */
 533 void prepare_to_copy(struct task_struct *tsk)
 534 {
 535         flush_fp_to_thread(current);
 536         flush_altivec_to_thread(current);
 537         flush_spe_to_thread(current);
 538 }
 539
 540 /*
 541  * Copy a thread..
 542  */
 543 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 544                 unsigned long unused, struct task_struct *p,
 545                 struct pt_regs *regs)
 546 {
 547         struct pt_regs *childregs, *kregs;
 548         extern void ret_from_fork(void);
 549         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 550
 551         CHECK_FULL_REGS(regs);
 552         /* Copy registers */
 553         sp -= sizeof(struct pt_regs);
 554         childregs = (struct pt_regs *) sp;
 555         *childregs = *regs;
 556         if ((childregs->msr & MSR_PR) == 0) {
 557                 /* for kernel thread, set `current' and stackptr in new task */
 558                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 559 #ifdef CONFIG_PPC32
 560                 childregs->gpr[2] = (unsigned long) p;
 561 #else
 562                 clear_tsk_thread_flag(p, TIF_32BIT);
 563 #endif
 564                 p->thread.regs = NULL;  /* no user register state */
 565         } else {
 566                 childregs->gpr[1] = usp;
 567                 p->thread.regs = childregs;
 568                 if (clone_flags & CLONE_SETTLS) {
 569 #ifdef CONFIG_PPC64
 570                         if (!test_thread_flag(TIF_32BIT))
 571                                 childregs->gpr[13] = childregs->gpr[6];
 572                         else
 573 #endif
 574                                 childregs->gpr[2] = childregs->gpr[6];
 575                 }
 576         }
 577         childregs->gpr[3] = 0;  /* Result from fork() */
 578         sp -= STACK_FRAME_OVERHEAD;
 579
 580         /*
 581          * The way this works is that at some point in the future
 582          * some task will call _switch to switch to the new task.
 583          * That will pop off the stack frame created below and start
 584          * the new task running at ret_from_fork.  The new task will
 585          * do some house keeping and then return from the fork or clone
 586          * system call, using the stack frame created above.
 587          */
 588         sp -= sizeof(struct pt_regs);
 589         kregs = (struct pt_regs *) sp;
 590         sp -= STACK_FRAME_OVERHEAD;
 591         p->thread.ksp = sp;
 592         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 593                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 594
 595 #ifdef CONFIG_PPC64
 596         if (cpu_has_feature(CPU_FTR_SLB)) {
 597                 unsigned long sp_vsid;
 598                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 599
 600                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
 601                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 602                                 << SLB_VSID_SHIFT_1T;
 603                 else
 604                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 605                                 << SLB_VSID_SHIFT;
 606                 sp_vsid |= SLB_VSID_KERNEL | llp;
 607                 p->thread.ksp_vsid = sp_vsid;
 608         }
 609
 610         /*
 611          * The PPC64 ABI makes use of a TOC to contain function
 612          * pointers.  The function (ret_from_except) is actually a pointer
 613          * to the TOC entry.  The first entry is a pointer to the actual
 614          * function.
 615          */
 616         kregs->nip = *((unsigned long *)ret_from_fork);
 617 #else
 618         kregs->nip = (unsigned long)ret_from_fork;
 619 #endif
 620
 621         return 0;
 622 }
 623
 624 /*
 625  * Set up a thread for executing a new program
 626  */
 627 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 628 {
 629 #ifdef CONFIG_PPC64
 630         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 631 #endif
 632
 633         set_fs(USER_DS);
 634
 635         /*
 636          * If we exec out of a kernel thread then thread.regs will not be
 637          * set.  Do it now.
 638          */
 639         if (!current->thread.regs) {
 640                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 641                 current->thread.regs = regs - 1;
 642         }
 643
 644         memset(regs->gpr, 0, sizeof(regs->gpr));
 645         regs->ctr = 0;
 646         regs->link = 0;
 647         regs->xer = 0;
 648         regs->ccr = 0;
 649         regs->gpr[1] = sp;
 650
 651         /*
 652          * We have just cleared all the nonvolatile GPRs, so make
 653          * FULL_REGS(regs) return true.  This is necessary to allow
 654          * ptrace to examine the thread immediately after exec.
 655          */
 656         regs->trap &= ~1UL;
 657
 658 #ifdef CONFIG_PPC32
 659         regs->mq = 0;
 660         regs->nip = start;
 661         regs->msr = MSR_USER;
 662 #else
 663         if (!test_thread_flag(TIF_32BIT)) {
 664                 unsigned long entry, toc;
 665
 666                 /* start is a relocated pointer to the function descriptor for
 667                  * the elf _start routine.  The first entry in the function
 668                  * descriptor is the entry address of _start and the second
 669                  * entry is the TOC value we need to use.
 670                  */
 671                 __get_user(entry, (unsigned long __user *)start);
 672                 __get_user(toc, (unsigned long __user *)start+1);
 673
 674                 /* Check whether the e_entry function descriptor entries
 675                  * need to be relocated before we can use them.
 676                  */
 677                 if (load_addr != 0) {
 678                         entry += load_addr;
 679                         toc   += load_addr;
 680                 }
 681                 regs->nip = entry;
 682                 regs->gpr[2] = toc;
 683                 regs->msr = MSR_USER64;
 684         } else {
 685                 regs->nip = start;
 686                 regs->gpr[2] = 0;
 687                 regs->msr = MSR_USER32;
 688         }
 689 #endif
 690
 691         discard_lazy_cpu_state();
 692         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 693         current->thread.fpscr.val = 0;
 694 #ifdef CONFIG_ALTIVEC
 695         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 696         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 697         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 698         current->thread.vrsave = 0;
 699         current->thread.used_vr = 0;
 700 #endif /* CONFIG_ALTIVEC */
 701 #ifdef CONFIG_SPE
 702         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 703         current->thread.acc = 0;
 704         current->thread.spefscr = 0;
 705         current->thread.used_spe = 0;
 706 #endif /* CONFIG_SPE */
 707 }
 708
 709 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 710                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 711
 712 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 713 {
 714         struct pt_regs *regs = tsk->thread.regs;
 715
 716         /* This is a bit hairy.  If we are an SPE enabled  processor
 717          * (have embedded fp) we store the IEEE exception enable flags in
 718          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 719          * mode (asyn, precise, disabled) for 'Classic' FP. */
 720         if (val & PR_FP_EXC_SW_ENABLE) {
 721 #ifdef CONFIG_SPE
 722                 if (cpu_has_feature(CPU_FTR_SPE)) {
 723                         tsk->thread.fpexc_mode = val &
 724                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 725                         return 0;
 726                 } else {
 727                         return -EINVAL;
 728                 }
 729 #else
 730                 return -EINVAL;
 731 #endif
 732         }
 733
 734         /* on a CONFIG_SPE this does not hurt us.  The bits that
 735          * __pack_fe01 use do not overlap with bits used for
 736          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 737          * on CONFIG_SPE implementations are reserved so writing to
 738          * them does not change anything */
 739         if (val > PR_FP_EXC_PRECISE)
 740                 return -EINVAL;
 741         tsk->thread.fpexc_mode = __pack_fe01(val);
 742         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 743                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 744                         | tsk->thread.fpexc_mode;
 745         return 0;
 746 }
 747
 748 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 749 {
 750         unsigned int val;
 751
 752         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 753 #ifdef CONFIG_SPE
 754                 if (cpu_has_feature(CPU_FTR_SPE))
 755                         val = tsk->thread.fpexc_mode;
 756                 else
 757                         return -EINVAL;
 758 #else
 759                 return -EINVAL;
 760 #endif
 761         else
 762                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 763         return put_user(val, (unsigned int __user *) adr);
 764 }
 765
 766 int set_endian(struct task_struct *tsk, unsigned int val)
 767 {
 768         struct pt_regs *regs = tsk->thread.regs;
 769
 770         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 771             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 772                 return -EINVAL;
 773
 774         if (regs == NULL)
 775                 return -EINVAL;
 776
 777         if (val == PR_ENDIAN_BIG)
 778                 regs->msr &= ~MSR_LE;
 779         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 780                 regs->msr |= MSR_LE;
 781         else
 782                 return -EINVAL;
 783
 784         return 0;
 785 }
 786
 787 int get_endian(struct task_struct *tsk, unsigned long adr)
 788 {
 789         struct pt_regs *regs = tsk->thread.regs;
 790         unsigned int val;
 791
 792         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 793             !cpu_has_feature(CPU_FTR_REAL_LE))
 794                 return -EINVAL;
 795
 796         if (regs == NULL)
 797                 return -EINVAL;
 798
 799         if (regs->msr & MSR_LE) {
 800                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 801                         val = PR_ENDIAN_LITTLE;
 802                 else
 803                         val = PR_ENDIAN_PPC_LITTLE;
 804         } else
 805                 val = PR_ENDIAN_BIG;
 806
 807         return put_user(val, (unsigned int __user *)adr);
 808 }
 809
 810 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 811 {
 812         tsk->thread.align_ctl = val;
 813         return 0;
 814 }
 815
 816 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 817 {
 818         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 819 }
 820
 821 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 822
 823 int sys_clone(unsigned long clone_flags, unsigned long usp,
 824               int __user *parent_tidp, void __user *child_threadptr,
 825               int __user *child_tidp, int p6,
 826               struct pt_regs *regs)
 827 {
 828         CHECK_FULL_REGS(regs);
 829         if (usp == 0)
 830                 usp = regs->gpr[1];     /* stack pointer for child */
 831 #ifdef CONFIG_PPC64
 832         if (test_thread_flag(TIF_32BIT)) {
 833                 parent_tidp = TRUNC_PTR(parent_tidp);
 834                 child_tidp = TRUNC_PTR(child_tidp);
 835         }
 836 #endif
 837         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 838 }
 839
 840 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 841              unsigned long p4, unsigned long p5, unsigned long p6,
 842              struct pt_regs *regs)
 843 {
 844         CHECK_FULL_REGS(regs);
 845         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 846 }
 847
 848 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
 849               unsigned long p4, unsigned long p5, unsigned long p6,
 850               struct pt_regs *regs)
 851 {
 852         CHECK_FULL_REGS(regs);
 853         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 854                         regs, 0, NULL, NULL);
 855 }
 856
 857 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 858                unsigned long a3, unsigned long a4, unsigned long a5,
 859                struct pt_regs *regs)
 860 {
 861         int error;
 862         char *filename;
 863
 864         filename = getname((char __user *) a0);
 865         error = PTR_ERR(filename);
 866         if (IS_ERR(filename))
 867                 goto out;
 868         flush_fp_to_thread(current);
 869         flush_altivec_to_thread(current);
 870         flush_spe_to_thread(current);
 871         error = do_execve(filename, (char __user * __user *) a1,
 872                           (char __user * __user *) a2, regs);
 873         putname(filename);
 874 out:
 875         return error;
 876 }
 877
 878 #ifdef CONFIG_IRQSTACKS
 879 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
 880                                   unsigned long nbytes)
 881 {
 882         unsigned long stack_page;
 883         unsigned long cpu = task_cpu(p);
 884
 885         /*
 886          * Avoid crashing if the stack has overflowed and corrupted
 887          * task_cpu(p), which is in the thread_info struct.
 888          */
 889         if (cpu < NR_CPUS && cpu_possible(cpu)) {
 890                 stack_page = (unsigned long) hardirq_ctx[cpu];
 891                 if (sp >= stack_page + sizeof(struct thread_struct)
 892                     && sp <= stack_page + THREAD_SIZE - nbytes)
 893                         return 1;
 894
 895                 stack_page = (unsigned long) softirq_ctx[cpu];
 896                 if (sp >= stack_page + sizeof(struct thread_struct)
 897                     && sp <= stack_page + THREAD_SIZE - nbytes)
 898                         return 1;
 899         }
 900         return 0;
 901 }
 902
 903 #else
 904 #define valid_irq_stack(sp, p, nb)      0
 905 #endif /* CONFIG_IRQSTACKS */
 906
 907 int validate_sp(unsigned long sp, struct task_struct *p,
 908                        unsigned long nbytes)
 909 {
 910         unsigned long stack_page = (unsigned long)task_stack_page(p);
 911
 912         if (sp >= stack_page + sizeof(struct thread_struct)
 913             && sp <= stack_page + THREAD_SIZE - nbytes)
 914                 return 1;
 915
 916         return valid_irq_stack(sp, p, nbytes);
 917 }
 918
 919 EXPORT_SYMBOL(validate_sp);
 920
 921 unsigned long get_wchan(struct task_struct *p)
 922 {
 923         unsigned long ip, sp;
 924         int count = 0;
 925
 926         if (!p || p == current || p->state == TASK_RUNNING)
 927                 return 0;
 928
 929         sp = p->thread.ksp;
 930         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 931                 return 0;
 932
 933         do {
 934                 sp = *(unsigned long *)sp;
 935                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 936                         return 0;
 937                 if (count > 0) {
 938                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
 939                         if (!in_sched_functions(ip))
 940                                 return ip;
 941                 }
 942         } while (count++ < 16);
 943         return 0;
 944 }
 945
 946 static int kstack_depth_to_print = 64;
 947
 948 void show_stack(struct task_struct *tsk, unsigned long *stack)
 949 {
 950         unsigned long sp, ip, lr, newsp;
 951         int count = 0;
 952         int firstframe = 1;
 953
 954         sp = (unsigned long) stack;
 955         if (tsk == NULL)
 956                 tsk = current;
 957         if (sp == 0) {
 958                 if (tsk == current)
 959                         asm("mr %0,1" : "=r" (sp));
 960                 else
 961                         sp = tsk->thread.ksp;
 962         }
 963
 964         lr = 0;
 965         printk("Call Trace:\n");
 966         do {
 967                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
 968                         return;
 969
 970                 stack = (unsigned long *) sp;
 971                 newsp = stack[0];
 972                 ip = stack[STACK_FRAME_LR_SAVE];
 973                 if (!firstframe || ip != lr) {
 974                         printk("["REG"] ["REG"] ", sp, ip);
 975                         print_symbol("%s", ip);
 976                         if (firstframe)
 977                                 printk(" (unreliable)");
 978                         printk("\n");
 979                 }
 980                 firstframe = 0;
 981
 982                 /*
 983                  * See if this is an exception frame.
 984                  * We look for the "regshere" marker in the current frame.
 985                  */
 986                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
 987                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
 988                         struct pt_regs *regs = (struct pt_regs *)
 989                                 (sp + STACK_FRAME_OVERHEAD);
 990                         printk("--- Exception: %lx", regs->trap);
 991                         print_symbol(" at %s\n", regs->nip);
 992                         lr = regs->link;
 993                         print_symbol("    LR = %s\n", lr);
 994                         firstframe = 1;
 995                 }
 996
 997                 sp = newsp;
 998         } while (count++ < kstack_depth_to_print);
 999 }
1000
1001 void dump_stack(void)
1002 {
1003         show_stack(current, NULL);
1004 }
1005 EXPORT_SYMBOL(dump_stack);
1006
1007 #ifdef CONFIG_PPC64
1008 void ppc64_runlatch_on(void)
1009 {
1010         unsigned long ctrl;
1011
1012         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1013                 HMT_medium();
1014
1015                 ctrl = mfspr(SPRN_CTRLF);
1016                 ctrl |= CTRL_RUNLATCH;
1017                 mtspr(SPRN_CTRLT, ctrl);
1018
1019                 set_thread_flag(TIF_RUNLATCH);
1020         }
1021 }
1022
1023 void ppc64_runlatch_off(void)
1024 {
1025         unsigned long ctrl;
1026
1027         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1028                 HMT_medium();
1029
1030                 clear_thread_flag(TIF_RUNLATCH);
1031
1032                 ctrl = mfspr(SPRN_CTRLF);
1033                 ctrl &= ~CTRL_RUNLATCH;
1034                 mtspr(SPRN_CTRLT, ctrl);
1035         }
1036 }
1037 #endif
1038
1039 #if THREAD_SHIFT < PAGE_SHIFT
1040
1041 static struct kmem_cache *thread_info_cache;
1042
1043 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1044 {
1045         struct thread_info *ti;
1046
1047         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1048         if (unlikely(ti == NULL))
1049                 return NULL;
1050 #ifdef CONFIG_DEBUG_STACK_USAGE
1051         memset(ti, 0, THREAD_SIZE);
1052 #endif
1053         return ti;
1054 }
1055
1056 void free_thread_info(struct thread_info *ti)
1057 {
1058         kmem_cache_free(thread_info_cache, ti);
1059 }
1060
1061 void thread_info_cache_init(void)
1062 {
1063         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1064                                               THREAD_SIZE, 0, NULL);
1065         BUG_ON(thread_info_cache == NULL);
1066 }
1067
1068 #endif /* THREAD_SHIFT < PAGE_SHIFT */