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 int set_dabr(unsigned long dabr)
 245 {
 246 #ifdef CONFIG_PPC_MERGE         /* XXX for now */
 247         if (ppc_md.set_dabr)
 248                 return ppc_md.set_dabr(dabr);
 249 #endif
 250
 251         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 252 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
 253         mtspr(SPRN_DABR, dabr);
 254 #endif
 255         return 0;
 256 }
 257
 258 #ifdef CONFIG_PPC64
 259 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 260 #endif
 261
 262 static DEFINE_PER_CPU(unsigned long, current_dabr);
 263
 264 struct task_struct *__switch_to(struct task_struct *prev,
 265         struct task_struct *new)
 266 {
 267         struct thread_struct *new_thread, *old_thread;
 268         unsigned long flags;
 269         struct task_struct *last;
 270
 271 #ifdef CONFIG_SMP
 272         /* avoid complexity of lazy save/restore of fpu
 273          * by just saving it every time we switch out if
 274          * this task used the fpu during the last quantum.
 275          *
 276          * If it tries to use the fpu again, it'll trap and
 277          * reload its fp regs.  So we don't have to do a restore
 278          * every switch, just a save.
 279          *  -- Cort
 280          */
 281         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 282                 giveup_fpu(prev);
 283 #ifdef CONFIG_ALTIVEC
 284         /*
 285          * If the previous thread used altivec in the last quantum
 286          * (thus changing altivec regs) then save them.
 287          * We used to check the VRSAVE register but not all apps
 288          * set it, so we don't rely on it now (and in fact we need
 289          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 290          *
 291          * On SMP we always save/restore altivec regs just to avoid the
 292          * complexity of changing processors.
 293          *  -- Cort
 294          */
 295         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 296                 giveup_altivec(prev);
 297 #endif /* CONFIG_ALTIVEC */
 298 #ifdef CONFIG_SPE
 299         /*
 300          * If the previous thread used spe in the last quantum
 301          * (thus changing spe regs) then save them.
 302          *
 303          * On SMP we always save/restore spe regs just to avoid the
 304          * complexity of changing processors.
 305          */
 306         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 307                 giveup_spe(prev);
 308 #endif /* CONFIG_SPE */
 309
 310 #else  /* CONFIG_SMP */
 311 #ifdef CONFIG_ALTIVEC
 312         /* Avoid the trap.  On smp this this never happens since
 313          * we don't set last_task_used_altivec -- Cort
 314          */
 315         if (new->thread.regs && last_task_used_altivec == new)
 316                 new->thread.regs->msr |= MSR_VEC;
 317 #endif /* CONFIG_ALTIVEC */
 318 #ifdef CONFIG_SPE
 319         /* Avoid the trap.  On smp this this never happens since
 320          * we don't set last_task_used_spe
 321          */
 322         if (new->thread.regs && last_task_used_spe == new)
 323                 new->thread.regs->msr |= MSR_SPE;
 324 #endif /* CONFIG_SPE */
 325
 326 #endif /* CONFIG_SMP */
 327
 328         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr)) {
 329                 set_dabr(new->thread.dabr);
 330                 __get_cpu_var(current_dabr) = new->thread.dabr;
 331         }
 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         last = _switch(old_thread, new_thread);
 357
 358         local_irq_restore(flags);
 359
 360         return last;
 361 }
 362
 363 static int instructions_to_print = 16;
 364
 365 static void show_instructions(struct pt_regs *regs)
 366 {
 367         int i;
 368         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 369                         sizeof(int));
 370
 371         printk("Instruction dump:");
 372
 373         for (i = 0; i < instructions_to_print; i++) {
 374                 int instr;
 375
 376                 if (!(i % 8))
 377                         printk("\n");
 378
 379 #if !defined(CONFIG_BOOKE)
 380                 /* If executing with the IMMU off, adjust pc rather
 381                  * than print XXXXXXXX.
 382                  */
 383                 if (!(regs->msr & MSR_IR))
 384                         pc = (unsigned long)phys_to_virt(pc);
 385 #endif
 386
 387                 /* We use __get_user here *only* to avoid an OOPS on a
 388                  * bad address because the pc *should* only be a
 389                  * kernel address.
 390                  */
 391                 if (!__kernel_text_address(pc) ||
 392                      __get_user(instr, (unsigned int __user *)pc)) {
 393                         printk("XXXXXXXX ");
 394                 } else {
 395                         if (regs->nip == pc)
 396                                 printk("<%08x> ", instr);
 397                         else
 398                                 printk("%08x ", instr);
 399                 }
 400
 401                 pc += sizeof(int);
 402         }
 403
 404         printk("\n");
 405 }
 406
 407 static struct regbit {
 408         unsigned long bit;
 409         const char *name;
 410 } msr_bits[] = {
 411         {MSR_EE,        "EE"},
 412         {MSR_PR,        "PR"},
 413         {MSR_FP,        "FP"},
 414         {MSR_ME,        "ME"},
 415         {MSR_IR,        "IR"},
 416         {MSR_DR,        "DR"},
 417         {0,             NULL}
 418 };
 419
 420 static void printbits(unsigned long val, struct regbit *bits)
 421 {
 422         const char *sep = "";
 423
 424         printk("<");
 425         for (; bits->bit; ++bits)
 426                 if (val & bits->bit) {
 427                         printk("%s%s", sep, bits->name);
 428                         sep = ",";
 429                 }
 430         printk(">");
 431 }
 432
 433 #ifdef CONFIG_PPC64
 434 #define REG             "%016lx"
 435 #define REGS_PER_LINE   4
 436 #define LAST_VOLATILE   13
 437 #else
 438 #define REG             "%08lx"
 439 #define REGS_PER_LINE   8
 440 #define LAST_VOLATILE   12
 441 #endif
 442
 443 void show_regs(struct pt_regs * regs)
 444 {
 445         int i, trap;
 446
 447         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 448                regs->nip, regs->link, regs->ctr);
 449         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 450                regs, regs->trap, print_tainted(), init_utsname()->release);
 451         printk("MSR: "REG" ", regs->msr);
 452         printbits(regs->msr, msr_bits);
 453         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 454         trap = TRAP(regs);
 455         if (trap == 0x300 || trap == 0x600)
 456 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 457                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 458 #else
 459                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 460 #endif
 461         printk("TASK = %p[%d] '%s' THREAD: %p",
 462                current, task_pid_nr(current), current->comm, task_thread_info(current));
 463
 464 #ifdef CONFIG_SMP
 465         printk(" CPU: %d", smp_processor_id());
 466 #endif /* CONFIG_SMP */
 467
 468         for (i = 0;  i < 32;  i++) {
 469                 if ((i % REGS_PER_LINE) == 0)
 470                         printk("\n" KERN_INFO "GPR%02d: ", i);
 471                 printk(REG " ", regs->gpr[i]);
 472                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 473                         break;
 474         }
 475         printk("\n");
 476 #ifdef CONFIG_KALLSYMS
 477         /*
 478          * Lookup NIP late so we have the best change of getting the
 479          * above info out without failing
 480          */
 481         printk("NIP ["REG"] ", regs->nip);
 482         print_symbol("%s\n", regs->nip);
 483         printk("LR ["REG"] ", regs->link);
 484         print_symbol("%s\n", regs->link);
 485 #endif
 486         show_stack(current, (unsigned long *) regs->gpr[1]);
 487         if (!user_mode(regs))
 488                 show_instructions(regs);
 489 }
 490
 491 void exit_thread(void)
 492 {
 493         discard_lazy_cpu_state();
 494 }
 495
 496 void flush_thread(void)
 497 {
 498 #ifdef CONFIG_PPC64
 499         struct thread_info *t = current_thread_info();
 500
 501         if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
 502                 clear_ti_thread_flag(t, TIF_ABI_PENDING);
 503                 if (test_ti_thread_flag(t, TIF_32BIT))
 504                         clear_ti_thread_flag(t, TIF_32BIT);
 505                 else
 506                         set_ti_thread_flag(t, TIF_32BIT);
 507         }
 508 #endif
 509
 510         discard_lazy_cpu_state();
 511
 512         if (current->thread.dabr) {
 513                 current->thread.dabr = 0;
 514                 set_dabr(0);
 515         }
 516 }
 517
 518 void
 519 release_thread(struct task_struct *t)
 520 {
 521 }
 522
 523 /*
 524  * This gets called before we allocate a new thread and copy
 525  * the current task into it.
 526  */
 527 void prepare_to_copy(struct task_struct *tsk)
 528 {
 529         flush_fp_to_thread(current);
 530         flush_altivec_to_thread(current);
 531         flush_spe_to_thread(current);
 532 }
 533
 534 /*
 535  * Copy a thread..
 536  */
 537 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 538                 unsigned long unused, struct task_struct *p,
 539                 struct pt_regs *regs)
 540 {
 541         struct pt_regs *childregs, *kregs;
 542         extern void ret_from_fork(void);
 543         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 544
 545         CHECK_FULL_REGS(regs);
 546         /* Copy registers */
 547         sp -= sizeof(struct pt_regs);
 548         childregs = (struct pt_regs *) sp;
 549         *childregs = *regs;
 550         if ((childregs->msr & MSR_PR) == 0) {
 551                 /* for kernel thread, set `current' and stackptr in new task */
 552                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 553 #ifdef CONFIG_PPC32
 554                 childregs->gpr[2] = (unsigned long) p;
 555 #else
 556                 clear_tsk_thread_flag(p, TIF_32BIT);
 557 #endif
 558                 p->thread.regs = NULL;  /* no user register state */
 559         } else {
 560                 childregs->gpr[1] = usp;
 561                 p->thread.regs = childregs;
 562                 if (clone_flags & CLONE_SETTLS) {
 563 #ifdef CONFIG_PPC64
 564                         if (!test_thread_flag(TIF_32BIT))
 565                                 childregs->gpr[13] = childregs->gpr[6];
 566                         else
 567 #endif
 568                                 childregs->gpr[2] = childregs->gpr[6];
 569                 }
 570         }
 571         childregs->gpr[3] = 0;  /* Result from fork() */
 572         sp -= STACK_FRAME_OVERHEAD;
 573
 574         /*
 575          * The way this works is that at some point in the future
 576          * some task will call _switch to switch to the new task.
 577          * That will pop off the stack frame created below and start
 578          * the new task running at ret_from_fork.  The new task will
 579          * do some house keeping and then return from the fork or clone
 580          * system call, using the stack frame created above.
 581          */
 582         sp -= sizeof(struct pt_regs);
 583         kregs = (struct pt_regs *) sp;
 584         sp -= STACK_FRAME_OVERHEAD;
 585         p->thread.ksp = sp;
 586
 587 #ifdef CONFIG_PPC64
 588         if (cpu_has_feature(CPU_FTR_SLB)) {
 589                 unsigned long sp_vsid;
 590                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 591
 592                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
 593                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 594                                 << SLB_VSID_SHIFT_1T;
 595                 else
 596                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 597                                 << SLB_VSID_SHIFT;
 598                 sp_vsid |= SLB_VSID_KERNEL | llp;
 599                 p->thread.ksp_vsid = sp_vsid;
 600         }
 601
 602         /*
 603          * The PPC64 ABI makes use of a TOC to contain function
 604          * pointers.  The function (ret_from_except) is actually a pointer
 605          * to the TOC entry.  The first entry is a pointer to the actual
 606          * function.
 607          */
 608         kregs->nip = *((unsigned long *)ret_from_fork);
 609 #else
 610         kregs->nip = (unsigned long)ret_from_fork;
 611 #endif
 612
 613         return 0;
 614 }
 615
 616 /*
 617  * Set up a thread for executing a new program
 618  */
 619 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 620 {
 621 #ifdef CONFIG_PPC64
 622         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 623 #endif
 624
 625         set_fs(USER_DS);
 626
 627         /*
 628          * If we exec out of a kernel thread then thread.regs will not be
 629          * set.  Do it now.
 630          */
 631         if (!current->thread.regs) {
 632                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 633                 current->thread.regs = regs - 1;
 634         }
 635
 636         memset(regs->gpr, 0, sizeof(regs->gpr));
 637         regs->ctr = 0;
 638         regs->link = 0;
 639         regs->xer = 0;
 640         regs->ccr = 0;
 641         regs->gpr[1] = sp;
 642
 643         /*
 644          * We have just cleared all the nonvolatile GPRs, so make
 645          * FULL_REGS(regs) return true.  This is necessary to allow
 646          * ptrace to examine the thread immediately after exec.
 647          */
 648         regs->trap &= ~1UL;
 649
 650 #ifdef CONFIG_PPC32
 651         regs->mq = 0;
 652         regs->nip = start;
 653         regs->msr = MSR_USER;
 654 #else
 655         if (!test_thread_flag(TIF_32BIT)) {
 656                 unsigned long entry, toc;
 657
 658                 /* start is a relocated pointer to the function descriptor for
 659                  * the elf _start routine.  The first entry in the function
 660                  * descriptor is the entry address of _start and the second
 661                  * entry is the TOC value we need to use.
 662                  */
 663                 __get_user(entry, (unsigned long __user *)start);
 664                 __get_user(toc, (unsigned long __user *)start+1);
 665
 666                 /* Check whether the e_entry function descriptor entries
 667                  * need to be relocated before we can use them.
 668                  */
 669                 if (load_addr != 0) {
 670                         entry += load_addr;
 671                         toc   += load_addr;
 672                 }
 673                 regs->nip = entry;
 674                 regs->gpr[2] = toc;
 675                 regs->msr = MSR_USER64;
 676         } else {
 677                 regs->nip = start;
 678                 regs->gpr[2] = 0;
 679                 regs->msr = MSR_USER32;
 680         }
 681 #endif
 682
 683         discard_lazy_cpu_state();
 684         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 685         current->thread.fpscr.val = 0;
 686 #ifdef CONFIG_ALTIVEC
 687         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 688         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 689         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 690         current->thread.vrsave = 0;
 691         current->thread.used_vr = 0;
 692 #endif /* CONFIG_ALTIVEC */
 693 #ifdef CONFIG_SPE
 694         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 695         current->thread.acc = 0;
 696         current->thread.spefscr = 0;
 697         current->thread.used_spe = 0;
 698 #endif /* CONFIG_SPE */
 699 }
 700
 701 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 702                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 703
 704 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 705 {
 706         struct pt_regs *regs = tsk->thread.regs;
 707
 708         /* This is a bit hairy.  If we are an SPE enabled  processor
 709          * (have embedded fp) we store the IEEE exception enable flags in
 710          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 711          * mode (asyn, precise, disabled) for 'Classic' FP. */
 712         if (val & PR_FP_EXC_SW_ENABLE) {
 713 #ifdef CONFIG_SPE
 714                 if (cpu_has_feature(CPU_FTR_SPE)) {
 715                         tsk->thread.fpexc_mode = val &
 716                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 717                         return 0;
 718                 } else {
 719                         return -EINVAL;
 720                 }
 721 #else
 722                 return -EINVAL;
 723 #endif
 724         }
 725
 726         /* on a CONFIG_SPE this does not hurt us.  The bits that
 727          * __pack_fe01 use do not overlap with bits used for
 728          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 729          * on CONFIG_SPE implementations are reserved so writing to
 730          * them does not change anything */
 731         if (val > PR_FP_EXC_PRECISE)
 732                 return -EINVAL;
 733         tsk->thread.fpexc_mode = __pack_fe01(val);
 734         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 735                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 736                         | tsk->thread.fpexc_mode;
 737         return 0;
 738 }
 739
 740 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 741 {
 742         unsigned int val;
 743
 744         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 745 #ifdef CONFIG_SPE
 746                 if (cpu_has_feature(CPU_FTR_SPE))
 747                         val = tsk->thread.fpexc_mode;
 748                 else
 749                         return -EINVAL;
 750 #else
 751                 return -EINVAL;
 752 #endif
 753         else
 754                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 755         return put_user(val, (unsigned int __user *) adr);
 756 }
 757
 758 int set_endian(struct task_struct *tsk, unsigned int val)
 759 {
 760         struct pt_regs *regs = tsk->thread.regs;
 761
 762         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 763             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 764                 return -EINVAL;
 765
 766         if (regs == NULL)
 767                 return -EINVAL;
 768
 769         if (val == PR_ENDIAN_BIG)
 770                 regs->msr &= ~MSR_LE;
 771         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 772                 regs->msr |= MSR_LE;
 773         else
 774                 return -EINVAL;
 775
 776         return 0;
 777 }
 778
 779 int get_endian(struct task_struct *tsk, unsigned long adr)
 780 {
 781         struct pt_regs *regs = tsk->thread.regs;
 782         unsigned int val;
 783
 784         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 785             !cpu_has_feature(CPU_FTR_REAL_LE))
 786                 return -EINVAL;
 787
 788         if (regs == NULL)
 789                 return -EINVAL;
 790
 791         if (regs->msr & MSR_LE) {
 792                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 793                         val = PR_ENDIAN_LITTLE;
 794                 else
 795                         val = PR_ENDIAN_PPC_LITTLE;
 796         } else
 797                 val = PR_ENDIAN_BIG;
 798
 799         return put_user(val, (unsigned int __user *)adr);
 800 }
 801
 802 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 803 {
 804         tsk->thread.align_ctl = val;
 805         return 0;
 806 }
 807
 808 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 809 {
 810         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 811 }
 812
 813 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 814
 815 int sys_clone(unsigned long clone_flags, unsigned long usp,
 816               int __user *parent_tidp, void __user *child_threadptr,
 817               int __user *child_tidp, int p6,
 818               struct pt_regs *regs)
 819 {
 820         CHECK_FULL_REGS(regs);
 821         if (usp == 0)
 822                 usp = regs->gpr[1];     /* stack pointer for child */
 823 #ifdef CONFIG_PPC64
 824         if (test_thread_flag(TIF_32BIT)) {
 825                 parent_tidp = TRUNC_PTR(parent_tidp);
 826                 child_tidp = TRUNC_PTR(child_tidp);
 827         }
 828 #endif
 829         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 830 }
 831
 832 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 833              unsigned long p4, unsigned long p5, unsigned long p6,
 834              struct pt_regs *regs)
 835 {
 836         CHECK_FULL_REGS(regs);
 837         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 838 }
 839
 840 int sys_vfork(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(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 846                         regs, 0, NULL, NULL);
 847 }
 848
 849 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 850                unsigned long a3, unsigned long a4, unsigned long a5,
 851                struct pt_regs *regs)
 852 {
 853         int error;
 854         char *filename;
 855
 856         filename = getname((char __user *) a0);
 857         error = PTR_ERR(filename);
 858         if (IS_ERR(filename))
 859                 goto out;
 860         flush_fp_to_thread(current);
 861         flush_altivec_to_thread(current);
 862         flush_spe_to_thread(current);
 863         error = do_execve(filename, (char __user * __user *) a1,
 864                           (char __user * __user *) a2, regs);
 865         if (error == 0) {
 866                 task_lock(current);
 867                 current->ptrace &= ~PT_DTRACE;
 868                 task_unlock(current);
 869         }
 870         putname(filename);
 871 out:
 872         return error;
 873 }
 874
 875 #ifdef CONFIG_IRQSTACKS
 876 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
 877                                   unsigned long nbytes)
 878 {
 879         unsigned long stack_page;
 880         unsigned long cpu = task_cpu(p);
 881
 882         /*
 883          * Avoid crashing if the stack has overflowed and corrupted
 884          * task_cpu(p), which is in the thread_info struct.
 885          */
 886         if (cpu < NR_CPUS && cpu_possible(cpu)) {
 887                 stack_page = (unsigned long) hardirq_ctx[cpu];
 888                 if (sp >= stack_page + sizeof(struct thread_struct)
 889                     && sp <= stack_page + THREAD_SIZE - nbytes)
 890                         return 1;
 891
 892                 stack_page = (unsigned long) softirq_ctx[cpu];
 893                 if (sp >= stack_page + sizeof(struct thread_struct)
 894                     && sp <= stack_page + THREAD_SIZE - nbytes)
 895                         return 1;
 896         }
 897         return 0;
 898 }
 899
 900 #else
 901 #define valid_irq_stack(sp, p, nb)      0
 902 #endif /* CONFIG_IRQSTACKS */
 903
 904 int validate_sp(unsigned long sp, struct task_struct *p,
 905                        unsigned long nbytes)
 906 {
 907         unsigned long stack_page = (unsigned long)task_stack_page(p);
 908
 909         if (sp >= stack_page + sizeof(struct thread_struct)
 910             && sp <= stack_page + THREAD_SIZE - nbytes)
 911                 return 1;
 912
 913         return valid_irq_stack(sp, p, nbytes);
 914 }
 915
 916 #ifdef CONFIG_PPC64
 917 #define MIN_STACK_FRAME 112     /* same as STACK_FRAME_OVERHEAD, in fact */
 918 #define FRAME_LR_SAVE   2
 919 #define INT_FRAME_SIZE  (sizeof(struct pt_regs) + STACK_FRAME_OVERHEAD + 288)
 920 #define REGS_MARKER     0x7265677368657265ul
 921 #define FRAME_MARKER    12
 922 #else
 923 #define MIN_STACK_FRAME 16
 924 #define FRAME_LR_SAVE   1
 925 #define INT_FRAME_SIZE  (sizeof(struct pt_regs) + STACK_FRAME_OVERHEAD)
 926 #define REGS_MARKER     0x72656773ul
 927 #define FRAME_MARKER    2
 928 #endif
 929
 930 EXPORT_SYMBOL(validate_sp);
 931
 932 unsigned long get_wchan(struct task_struct *p)
 933 {
 934         unsigned long ip, sp;
 935         int count = 0;
 936
 937         if (!p || p == current || p->state == TASK_RUNNING)
 938                 return 0;
 939
 940         sp = p->thread.ksp;
 941         if (!validate_sp(sp, p, MIN_STACK_FRAME))
 942                 return 0;
 943
 944         do {
 945                 sp = *(unsigned long *)sp;
 946                 if (!validate_sp(sp, p, MIN_STACK_FRAME))
 947                         return 0;
 948                 if (count > 0) {
 949                         ip = ((unsigned long *)sp)[FRAME_LR_SAVE];
 950                         if (!in_sched_functions(ip))
 951                                 return ip;
 952                 }
 953         } while (count++ < 16);
 954         return 0;
 955 }
 956
 957 static int kstack_depth_to_print = 64;
 958
 959 void show_stack(struct task_struct *tsk, unsigned long *stack)
 960 {
 961         unsigned long sp, ip, lr, newsp;
 962         int count = 0;
 963         int firstframe = 1;
 964
 965         sp = (unsigned long) stack;
 966         if (tsk == NULL)
 967                 tsk = current;
 968         if (sp == 0) {
 969                 if (tsk == current)
 970                         asm("mr %0,1" : "=r" (sp));
 971                 else
 972                         sp = tsk->thread.ksp;
 973         }
 974
 975         lr = 0;
 976         printk("Call Trace:\n");
 977         do {
 978                 if (!validate_sp(sp, tsk, MIN_STACK_FRAME))
 979                         return;
 980
 981                 stack = (unsigned long *) sp;
 982                 newsp = stack[0];
 983                 ip = stack[FRAME_LR_SAVE];
 984                 if (!firstframe || ip != lr) {
 985                         printk("["REG"] ["REG"] ", sp, ip);
 986                         print_symbol("%s", ip);
 987                         if (firstframe)
 988                                 printk(" (unreliable)");
 989                         printk("\n");
 990                 }
 991                 firstframe = 0;
 992
 993                 /*
 994                  * See if this is an exception frame.
 995                  * We look for the "regshere" marker in the current frame.
 996                  */
 997                 if (validate_sp(sp, tsk, INT_FRAME_SIZE)
 998                     && stack[FRAME_MARKER] == REGS_MARKER) {
 999                         struct pt_regs *regs = (struct pt_regs *)
1000                                 (sp + STACK_FRAME_OVERHEAD);
1001                         printk("--- Exception: %lx", regs->trap);
1002                         print_symbol(" at %s\n", regs->nip);
1003                         lr = regs->link;
1004                         print_symbol("    LR = %s\n", lr);
1005                         firstframe = 1;
1006                 }
1007
1008                 sp = newsp;
1009         } while (count++ < kstack_depth_to_print);
1010 }
1011
1012 void dump_stack(void)
1013 {
1014         show_stack(current, NULL);
1015 }
1016 EXPORT_SYMBOL(dump_stack);
1017
1018 #ifdef CONFIG_PPC64
1019 void ppc64_runlatch_on(void)
1020 {
1021         unsigned long ctrl;
1022
1023         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1024                 HMT_medium();
1025
1026                 ctrl = mfspr(SPRN_CTRLF);
1027                 ctrl |= CTRL_RUNLATCH;
1028                 mtspr(SPRN_CTRLT, ctrl);
1029
1030                 set_thread_flag(TIF_RUNLATCH);
1031         }
1032 }
1033
1034 void ppc64_runlatch_off(void)
1035 {
1036         unsigned long ctrl;
1037
1038         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1039                 HMT_medium();
1040
1041                 clear_thread_flag(TIF_RUNLATCH);
1042
1043                 ctrl = mfspr(SPRN_CTRLF);
1044                 ctrl &= ~CTRL_RUNLATCH;
1045                 mtspr(SPRN_CTRLT, ctrl);
1046         }
1047 }
1048 #endif