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/ftrace.h>
  37 #include <linux/kernel_stat.h>
  38 #include <linux/personality.h>
  39 #include <linux/random.h>
  40
  41 #include <asm/pgtable.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/system.h>
  44 #include <asm/io.h>
  45 #include <asm/processor.h>
  46 #include <asm/mmu.h>
  47 #include <asm/prom.h>
  48 #include <asm/machdep.h>
  49 #include <asm/time.h>
  50 #include <asm/syscalls.h>
  51 #ifdef CONFIG_PPC64
  52 #include <asm/firmware.h>
  53 #endif
  54 #include <linux/kprobes.h>
  55 #include <linux/kdebug.h>
  56
  57 extern unsigned long _get_SP(void);
  58
  59 #ifndef CONFIG_SMP
  60 struct task_struct *last_task_used_math = NULL;
  61 struct task_struct *last_task_used_altivec = NULL;
  62 struct task_struct *last_task_used_vsx = NULL;
  63 struct task_struct *last_task_used_spe = NULL;
  64 #endif
  65
  66 /*
  67  * Make sure the floating-point register state in the
  68  * the thread_struct is up to date for task tsk.
  69  */
  70 void flush_fp_to_thread(struct task_struct *tsk)
  71 {
  72         if (tsk->thread.regs) {
  73                 /*
  74                  * We need to disable preemption here because if we didn't,
  75                  * another process could get scheduled after the regs->msr
  76                  * test but before we have finished saving the FP registers
  77                  * to the thread_struct.  That process could take over the
  78                  * FPU, and then when we get scheduled again we would store
  79                  * bogus values for the remaining FP registers.
  80                  */
  81                 preempt_disable();
  82                 if (tsk->thread.regs->msr & MSR_FP) {
  83 #ifdef CONFIG_SMP
  84                         /*
  85                          * This should only ever be called for current or
  86                          * for a stopped child process.  Since we save away
  87                          * the FP register state on context switch on SMP,
  88                          * there is something wrong if a stopped child appears
  89                          * to still have its FP state in the CPU registers.
  90                          */
  91                         BUG_ON(tsk != current);
  92 #endif
  93                         giveup_fpu(tsk);
  94                 }
  95                 preempt_enable();
  96         }
  97 }
  98
  99 void enable_kernel_fp(void)
 100 {
 101         WARN_ON(preemptible());
 102
 103 #ifdef CONFIG_SMP
 104         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
 105                 giveup_fpu(current);
 106         else
 107                 giveup_fpu(NULL);       /* just enables FP for kernel */
 108 #else
 109         giveup_fpu(last_task_used_math);
 110 #endif /* CONFIG_SMP */
 111 }
 112 EXPORT_SYMBOL(enable_kernel_fp);
 113
 114 #ifdef CONFIG_ALTIVEC
 115 void enable_kernel_altivec(void)
 116 {
 117         WARN_ON(preemptible());
 118
 119 #ifdef CONFIG_SMP
 120         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 121                 giveup_altivec(current);
 122         else
 123                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
 124 #else
 125         giveup_altivec(last_task_used_altivec);
 126 #endif /* CONFIG_SMP */
 127 }
 128 EXPORT_SYMBOL(enable_kernel_altivec);
 129
 130 /*
 131  * Make sure the VMX/Altivec register state in the
 132  * the thread_struct is up to date for task tsk.
 133  */
 134 void flush_altivec_to_thread(struct task_struct *tsk)
 135 {
 136         if (tsk->thread.regs) {
 137                 preempt_disable();
 138                 if (tsk->thread.regs->msr & MSR_VEC) {
 139 #ifdef CONFIG_SMP
 140                         BUG_ON(tsk != current);
 141 #endif
 142                         giveup_altivec(tsk);
 143                 }
 144                 preempt_enable();
 145         }
 146 }
 147 #endif /* CONFIG_ALTIVEC */
 148
 149 #ifdef CONFIG_VSX
 150 #if 0
 151 /* not currently used, but some crazy RAID module might want to later */
 152 void enable_kernel_vsx(void)
 153 {
 154         WARN_ON(preemptible());
 155
 156 #ifdef CONFIG_SMP
 157         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 158                 giveup_vsx(current);
 159         else
 160                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 161 #else
 162         giveup_vsx(last_task_used_vsx);
 163 #endif /* CONFIG_SMP */
 164 }
 165 EXPORT_SYMBOL(enable_kernel_vsx);
 166 #endif
 167
 168 void giveup_vsx(struct task_struct *tsk)
 169 {
 170         giveup_fpu(tsk);
 171         giveup_altivec(tsk);
 172         __giveup_vsx(tsk);
 173 }
 174
 175 void flush_vsx_to_thread(struct task_struct *tsk)
 176 {
 177         if (tsk->thread.regs) {
 178                 preempt_disable();
 179                 if (tsk->thread.regs->msr & MSR_VSX) {
 180 #ifdef CONFIG_SMP
 181                         BUG_ON(tsk != current);
 182 #endif
 183                         giveup_vsx(tsk);
 184                 }
 185                 preempt_enable();
 186         }
 187 }
 188 #endif /* CONFIG_VSX */
 189
 190 #ifdef CONFIG_SPE
 191
 192 void enable_kernel_spe(void)
 193 {
 194         WARN_ON(preemptible());
 195
 196 #ifdef CONFIG_SMP
 197         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 198                 giveup_spe(current);
 199         else
 200                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 201 #else
 202         giveup_spe(last_task_used_spe);
 203 #endif /* __SMP __ */
 204 }
 205 EXPORT_SYMBOL(enable_kernel_spe);
 206
 207 void flush_spe_to_thread(struct task_struct *tsk)
 208 {
 209         if (tsk->thread.regs) {
 210                 preempt_disable();
 211                 if (tsk->thread.regs->msr & MSR_SPE) {
 212 #ifdef CONFIG_SMP
 213                         BUG_ON(tsk != current);
 214 #endif
 215                         giveup_spe(tsk);
 216                 }
 217                 preempt_enable();
 218         }
 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_VSX
 237         if (last_task_used_vsx == current)
 238                 last_task_used_vsx = NULL;
 239 #endif /* CONFIG_VSX */
 240 #ifdef CONFIG_SPE
 241         if (last_task_used_spe == current)
 242                 last_task_used_spe = NULL;
 243 #endif
 244         preempt_enable();
 245 }
 246 #endif /* CONFIG_SMP */
 247
 248 void do_dabr(struct pt_regs *regs, unsigned long address,
 249                     unsigned long error_code)
 250 {
 251         siginfo_t info;
 252
 253         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 254                         11, SIGSEGV) == NOTIFY_STOP)
 255                 return;
 256
 257         if (debugger_dabr_match(regs))
 258                 return;
 259
 260         /* Clear the DAC and struct entries.  One shot trigger */
 261 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 262         mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~(DBSR_DAC1R | DBSR_DAC1W
 263                                                         | DBCR0_IDM));
 264 #endif
 265
 266         /* Clear the DABR */
 267         set_dabr(0);
 268
 269         /* Deliver the signal to userspace */
 270         info.si_signo = SIGTRAP;
 271         info.si_errno = 0;
 272         info.si_code = TRAP_HWBKPT;
 273         info.si_addr = (void __user *)address;
 274         force_sig_info(SIGTRAP, &info, current);
 275 }
 276
 277 static DEFINE_PER_CPU(unsigned long, current_dabr);
 278
 279 int set_dabr(unsigned long dabr)
 280 {
 281         __get_cpu_var(current_dabr) = dabr;
 282
 283         if (ppc_md.set_dabr)
 284                 return ppc_md.set_dabr(dabr);
 285
 286         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 287 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 288         mtspr(SPRN_DAC1, dabr);
 289 #elif defined(CONFIG_PPC_BOOK3S)
 290         mtspr(SPRN_DABR, dabr);
 291 #endif
 292
 293
 294         return 0;
 295 }
 296
 297 #ifdef CONFIG_PPC64
 298 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 299 #endif
 300
 301 struct task_struct *__switch_to(struct task_struct *prev,
 302         struct task_struct *new)
 303 {
 304         struct thread_struct *new_thread, *old_thread;
 305         unsigned long flags;
 306         struct task_struct *last;
 307
 308 #ifdef CONFIG_SMP
 309         /* avoid complexity of lazy save/restore of fpu
 310          * by just saving it every time we switch out if
 311          * this task used the fpu during the last quantum.
 312          *
 313          * If it tries to use the fpu again, it'll trap and
 314          * reload its fp regs.  So we don't have to do a restore
 315          * every switch, just a save.
 316          *  -- Cort
 317          */
 318         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 319                 giveup_fpu(prev);
 320 #ifdef CONFIG_ALTIVEC
 321         /*
 322          * If the previous thread used altivec in the last quantum
 323          * (thus changing altivec regs) then save them.
 324          * We used to check the VRSAVE register but not all apps
 325          * set it, so we don't rely on it now (and in fact we need
 326          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 327          *
 328          * On SMP we always save/restore altivec regs just to avoid the
 329          * complexity of changing processors.
 330          *  -- Cort
 331          */
 332         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 333                 giveup_altivec(prev);
 334 #endif /* CONFIG_ALTIVEC */
 335 #ifdef CONFIG_VSX
 336         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 337                 /* VMX and FPU registers are already save here */
 338                 __giveup_vsx(prev);
 339 #endif /* CONFIG_VSX */
 340 #ifdef CONFIG_SPE
 341         /*
 342          * If the previous thread used spe in the last quantum
 343          * (thus changing spe regs) then save them.
 344          *
 345          * On SMP we always save/restore spe regs just to avoid the
 346          * complexity of changing processors.
 347          */
 348         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 349                 giveup_spe(prev);
 350 #endif /* CONFIG_SPE */
 351
 352 #else  /* CONFIG_SMP */
 353 #ifdef CONFIG_ALTIVEC
 354         /* Avoid the trap.  On smp this this never happens since
 355          * we don't set last_task_used_altivec -- Cort
 356          */
 357         if (new->thread.regs && last_task_used_altivec == new)
 358                 new->thread.regs->msr |= MSR_VEC;
 359 #endif /* CONFIG_ALTIVEC */
 360 #ifdef CONFIG_VSX
 361         if (new->thread.regs && last_task_used_vsx == new)
 362                 new->thread.regs->msr |= MSR_VSX;
 363 #endif /* CONFIG_VSX */
 364 #ifdef CONFIG_SPE
 365         /* Avoid the trap.  On smp this this never happens since
 366          * we don't set last_task_used_spe
 367          */
 368         if (new->thread.regs && last_task_used_spe == new)
 369                 new->thread.regs->msr |= MSR_SPE;
 370 #endif /* CONFIG_SPE */
 371
 372 #endif /* CONFIG_SMP */
 373
 374 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 375         /* If new thread DAC (HW breakpoint) is the same then leave it */
 376         if (new->thread.dabr)
 377                 set_dabr(new->thread.dabr);
 378 #else
 379         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 380                 set_dabr(new->thread.dabr);
 381 #endif
 382
 383
 384         new_thread = &new->thread;
 385         old_thread = &current->thread;
 386
 387 #ifdef CONFIG_PPC64
 388         /*
 389          * Collect processor utilization data per process
 390          */
 391         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 392                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 393                 long unsigned start_tb, current_tb;
 394                 start_tb = old_thread->start_tb;
 395                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 396                 old_thread->accum_tb += (current_tb - start_tb);
 397                 new_thread->start_tb = current_tb;
 398         }
 399 #endif
 400
 401         local_irq_save(flags);
 402
 403         account_system_vtime(current);
 404         account_process_vtime(current);
 405         calculate_steal_time();
 406
 407         /*
 408          * We can't take a PMU exception inside _switch() since there is a
 409          * window where the kernel stack SLB and the kernel stack are out
 410          * of sync. Hard disable here.
 411          */
 412         hard_irq_disable();
 413         last = _switch(old_thread, new_thread);
 414
 415         local_irq_restore(flags);
 416
 417         return last;
 418 }
 419
 420 static int instructions_to_print = 16;
 421
 422 static void show_instructions(struct pt_regs *regs)
 423 {
 424         int i;
 425         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 426                         sizeof(int));
 427
 428         printk("Instruction dump:");
 429
 430         for (i = 0; i < instructions_to_print; i++) {
 431                 int instr;
 432
 433                 if (!(i % 8))
 434                         printk("\n");
 435
 436 #if !defined(CONFIG_BOOKE)
 437                 /* If executing with the IMMU off, adjust pc rather
 438                  * than print XXXXXXXX.
 439                  */
 440                 if (!(regs->msr & MSR_IR))
 441                         pc = (unsigned long)phys_to_virt(pc);
 442 #endif
 443
 444                 /* We use __get_user here *only* to avoid an OOPS on a
 445                  * bad address because the pc *should* only be a
 446                  * kernel address.
 447                  */
 448                 if (!__kernel_text_address(pc) ||
 449                      __get_user(instr, (unsigned int __user *)pc)) {
 450                         printk("XXXXXXXX ");
 451                 } else {
 452                         if (regs->nip == pc)
 453                                 printk("<%08x> ", instr);
 454                         else
 455                                 printk("%08x ", instr);
 456                 }
 457
 458                 pc += sizeof(int);
 459         }
 460
 461         printk("\n");
 462 }
 463
 464 static struct regbit {
 465         unsigned long bit;
 466         const char *name;
 467 } msr_bits[] = {
 468         {MSR_EE,        "EE"},
 469         {MSR_PR,        "PR"},
 470         {MSR_FP,        "FP"},
 471         {MSR_VEC,       "VEC"},
 472         {MSR_VSX,       "VSX"},
 473         {MSR_ME,        "ME"},
 474         {MSR_CE,        "CE"},
 475         {MSR_DE,        "DE"},
 476         {MSR_IR,        "IR"},
 477         {MSR_DR,        "DR"},
 478         {0,             NULL}
 479 };
 480
 481 static void printbits(unsigned long val, struct regbit *bits)
 482 {
 483         const char *sep = "";
 484
 485         printk("<");
 486         for (; bits->bit; ++bits)
 487                 if (val & bits->bit) {
 488                         printk("%s%s", sep, bits->name);
 489                         sep = ",";
 490                 }
 491         printk(">");
 492 }
 493
 494 #ifdef CONFIG_PPC64
 495 #define REG             "%016lx"
 496 #define REGS_PER_LINE   4
 497 #define LAST_VOLATILE   13
 498 #else
 499 #define REG             "%08lx"
 500 #define REGS_PER_LINE   8
 501 #define LAST_VOLATILE   12
 502 #endif
 503
 504 void show_regs(struct pt_regs * regs)
 505 {
 506         int i, trap;
 507
 508         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 509                regs->nip, regs->link, regs->ctr);
 510         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 511                regs, regs->trap, print_tainted(), init_utsname()->release);
 512         printk("MSR: "REG" ", regs->msr);
 513         printbits(regs->msr, msr_bits);
 514         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 515         trap = TRAP(regs);
 516         if (trap == 0x300 || trap == 0x600)
 517 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 518                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 519 #else
 520                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 521 #endif
 522         printk("TASK = %p[%d] '%s' THREAD: %p",
 523                current, task_pid_nr(current), current->comm, task_thread_info(current));
 524
 525 #ifdef CONFIG_SMP
 526         printk(" CPU: %d", raw_smp_processor_id());
 527 #endif /* CONFIG_SMP */
 528
 529         for (i = 0;  i < 32;  i++) {
 530                 if ((i % REGS_PER_LINE) == 0)
 531                         printk("\nGPR%02d: ", i);
 532                 printk(REG " ", regs->gpr[i]);
 533                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 534                         break;
 535         }
 536         printk("\n");
 537 #ifdef CONFIG_KALLSYMS
 538         /*
 539          * Lookup NIP late so we have the best change of getting the
 540          * above info out without failing
 541          */
 542         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 543         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 544 #endif
 545         show_stack(current, (unsigned long *) regs->gpr[1]);
 546         if (!user_mode(regs))
 547                 show_instructions(regs);
 548 }
 549
 550 void exit_thread(void)
 551 {
 552         discard_lazy_cpu_state();
 553 }
 554
 555 void flush_thread(void)
 556 {
 557         discard_lazy_cpu_state();
 558
 559         if (current->thread.dabr) {
 560                 current->thread.dabr = 0;
 561                 set_dabr(0);
 562
 563 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 564                 current->thread.dbcr0 &= ~(DBSR_DAC1R | DBSR_DAC1W);
 565 #endif
 566         }
 567 }
 568
 569 void
 570 release_thread(struct task_struct *t)
 571 {
 572 }
 573
 574 /*
 575  * This gets called before we allocate a new thread and copy
 576  * the current task into it.
 577  */
 578 void prepare_to_copy(struct task_struct *tsk)
 579 {
 580         flush_fp_to_thread(current);
 581         flush_altivec_to_thread(current);
 582         flush_vsx_to_thread(current);
 583         flush_spe_to_thread(current);
 584 }
 585
 586 /*
 587  * Copy a thread..
 588  */
 589 int copy_thread(unsigned long clone_flags, unsigned long usp,
 590                 unsigned long unused, struct task_struct *p,
 591                 struct pt_regs *regs)
 592 {
 593         struct pt_regs *childregs, *kregs;
 594         extern void ret_from_fork(void);
 595         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 596
 597         CHECK_FULL_REGS(regs);
 598         /* Copy registers */
 599         sp -= sizeof(struct pt_regs);
 600         childregs = (struct pt_regs *) sp;
 601         *childregs = *regs;
 602         if ((childregs->msr & MSR_PR) == 0) {
 603                 /* for kernel thread, set `current' and stackptr in new task */
 604                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 605 #ifdef CONFIG_PPC32
 606                 childregs->gpr[2] = (unsigned long) p;
 607 #else
 608                 clear_tsk_thread_flag(p, TIF_32BIT);
 609 #endif
 610                 p->thread.regs = NULL;  /* no user register state */
 611         } else {
 612                 childregs->gpr[1] = usp;
 613                 p->thread.regs = childregs;
 614                 if (clone_flags & CLONE_SETTLS) {
 615 #ifdef CONFIG_PPC64
 616                         if (!test_thread_flag(TIF_32BIT))
 617                                 childregs->gpr[13] = childregs->gpr[6];
 618                         else
 619 #endif
 620                                 childregs->gpr[2] = childregs->gpr[6];
 621                 }
 622         }
 623         childregs->gpr[3] = 0;  /* Result from fork() */
 624         sp -= STACK_FRAME_OVERHEAD;
 625
 626         /*
 627          * The way this works is that at some point in the future
 628          * some task will call _switch to switch to the new task.
 629          * That will pop off the stack frame created below and start
 630          * the new task running at ret_from_fork.  The new task will
 631          * do some house keeping and then return from the fork or clone
 632          * system call, using the stack frame created above.
 633          */
 634         sp -= sizeof(struct pt_regs);
 635         kregs = (struct pt_regs *) sp;
 636         sp -= STACK_FRAME_OVERHEAD;
 637         p->thread.ksp = sp;
 638         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 639                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 640
 641 #ifdef CONFIG_PPC_STD_MMU_64
 642         if (cpu_has_feature(CPU_FTR_SLB)) {
 643                 unsigned long sp_vsid;
 644                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 645
 646                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
 647                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 648                                 << SLB_VSID_SHIFT_1T;
 649                 else
 650                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 651                                 << SLB_VSID_SHIFT;
 652                 sp_vsid |= SLB_VSID_KERNEL | llp;
 653                 p->thread.ksp_vsid = sp_vsid;
 654         }
 655 #endif /* CONFIG_PPC_STD_MMU_64 */
 656
 657         /*
 658          * The PPC64 ABI makes use of a TOC to contain function
 659          * pointers.  The function (ret_from_except) is actually a pointer
 660          * to the TOC entry.  The first entry is a pointer to the actual
 661          * function.
 662          */
 663 #ifdef CONFIG_PPC64
 664         kregs->nip = *((unsigned long *)ret_from_fork);
 665 #else
 666         kregs->nip = (unsigned long)ret_from_fork;
 667 #endif
 668
 669         return 0;
 670 }
 671
 672 /*
 673  * Set up a thread for executing a new program
 674  */
 675 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 676 {
 677 #ifdef CONFIG_PPC64
 678         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 679 #endif
 680
 681         set_fs(USER_DS);
 682
 683         /*
 684          * If we exec out of a kernel thread then thread.regs will not be
 685          * set.  Do it now.
 686          */
 687         if (!current->thread.regs) {
 688                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 689                 current->thread.regs = regs - 1;
 690         }
 691
 692         memset(regs->gpr, 0, sizeof(regs->gpr));
 693         regs->ctr = 0;
 694         regs->link = 0;
 695         regs->xer = 0;
 696         regs->ccr = 0;
 697         regs->gpr[1] = sp;
 698
 699         /*
 700          * We have just cleared all the nonvolatile GPRs, so make
 701          * FULL_REGS(regs) return true.  This is necessary to allow
 702          * ptrace to examine the thread immediately after exec.
 703          */
 704         regs->trap &= ~1UL;
 705
 706 #ifdef CONFIG_PPC32
 707         regs->mq = 0;
 708         regs->nip = start;
 709         regs->msr = MSR_USER;
 710 #else
 711         if (!test_thread_flag(TIF_32BIT)) {
 712                 unsigned long entry, toc;
 713
 714                 /* start is a relocated pointer to the function descriptor for
 715                  * the elf _start routine.  The first entry in the function
 716                  * descriptor is the entry address of _start and the second
 717                  * entry is the TOC value we need to use.
 718                  */
 719                 __get_user(entry, (unsigned long __user *)start);
 720                 __get_user(toc, (unsigned long __user *)start+1);
 721
 722                 /* Check whether the e_entry function descriptor entries
 723                  * need to be relocated before we can use them.
 724                  */
 725                 if (load_addr != 0) {
 726                         entry += load_addr;
 727                         toc   += load_addr;
 728                 }
 729                 regs->nip = entry;
 730                 regs->gpr[2] = toc;
 731                 regs->msr = MSR_USER64;
 732         } else {
 733                 regs->nip = start;
 734                 regs->gpr[2] = 0;
 735                 regs->msr = MSR_USER32;
 736         }
 737 #endif
 738
 739         discard_lazy_cpu_state();
 740 #ifdef CONFIG_VSX
 741         current->thread.used_vsr = 0;
 742 #endif
 743         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 744         current->thread.fpscr.val = 0;
 745 #ifdef CONFIG_ALTIVEC
 746         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 747         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 748         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 749         current->thread.vrsave = 0;
 750         current->thread.used_vr = 0;
 751 #endif /* CONFIG_ALTIVEC */
 752 #ifdef CONFIG_SPE
 753         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 754         current->thread.acc = 0;
 755         current->thread.spefscr = 0;
 756         current->thread.used_spe = 0;
 757 #endif /* CONFIG_SPE */
 758 }
 759
 760 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 761                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 762
 763 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 764 {
 765         struct pt_regs *regs = tsk->thread.regs;
 766
 767         /* This is a bit hairy.  If we are an SPE enabled  processor
 768          * (have embedded fp) we store the IEEE exception enable flags in
 769          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 770          * mode (asyn, precise, disabled) for 'Classic' FP. */
 771         if (val & PR_FP_EXC_SW_ENABLE) {
 772 #ifdef CONFIG_SPE
 773                 if (cpu_has_feature(CPU_FTR_SPE)) {
 774                         tsk->thread.fpexc_mode = val &
 775                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 776                         return 0;
 777                 } else {
 778                         return -EINVAL;
 779                 }
 780 #else
 781                 return -EINVAL;
 782 #endif
 783         }
 784
 785         /* on a CONFIG_SPE this does not hurt us.  The bits that
 786          * __pack_fe01 use do not overlap with bits used for
 787          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 788          * on CONFIG_SPE implementations are reserved so writing to
 789          * them does not change anything */
 790         if (val > PR_FP_EXC_PRECISE)
 791                 return -EINVAL;
 792         tsk->thread.fpexc_mode = __pack_fe01(val);
 793         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 794                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 795                         | tsk->thread.fpexc_mode;
 796         return 0;
 797 }
 798
 799 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 800 {
 801         unsigned int val;
 802
 803         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 804 #ifdef CONFIG_SPE
 805                 if (cpu_has_feature(CPU_FTR_SPE))
 806                         val = tsk->thread.fpexc_mode;
 807                 else
 808                         return -EINVAL;
 809 #else
 810                 return -EINVAL;
 811 #endif
 812         else
 813                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 814         return put_user(val, (unsigned int __user *) adr);
 815 }
 816
 817 int set_endian(struct task_struct *tsk, unsigned int val)
 818 {
 819         struct pt_regs *regs = tsk->thread.regs;
 820
 821         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 822             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 823                 return -EINVAL;
 824
 825         if (regs == NULL)
 826                 return -EINVAL;
 827
 828         if (val == PR_ENDIAN_BIG)
 829                 regs->msr &= ~MSR_LE;
 830         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 831                 regs->msr |= MSR_LE;
 832         else
 833                 return -EINVAL;
 834
 835         return 0;
 836 }
 837
 838 int get_endian(struct task_struct *tsk, unsigned long adr)
 839 {
 840         struct pt_regs *regs = tsk->thread.regs;
 841         unsigned int val;
 842
 843         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 844             !cpu_has_feature(CPU_FTR_REAL_LE))
 845                 return -EINVAL;
 846
 847         if (regs == NULL)
 848                 return -EINVAL;
 849
 850         if (regs->msr & MSR_LE) {
 851                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 852                         val = PR_ENDIAN_LITTLE;
 853                 else
 854                         val = PR_ENDIAN_PPC_LITTLE;
 855         } else
 856                 val = PR_ENDIAN_BIG;
 857
 858         return put_user(val, (unsigned int __user *)adr);
 859 }
 860
 861 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 862 {
 863         tsk->thread.align_ctl = val;
 864         return 0;
 865 }
 866
 867 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 868 {
 869         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 870 }
 871
 872 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 873
 874 int sys_clone(unsigned long clone_flags, unsigned long usp,
 875               int __user *parent_tidp, void __user *child_threadptr,
 876               int __user *child_tidp, int p6,
 877               struct pt_regs *regs)
 878 {
 879         CHECK_FULL_REGS(regs);
 880         if (usp == 0)
 881                 usp = regs->gpr[1];     /* stack pointer for child */
 882 #ifdef CONFIG_PPC64
 883         if (test_thread_flag(TIF_32BIT)) {
 884                 parent_tidp = TRUNC_PTR(parent_tidp);
 885                 child_tidp = TRUNC_PTR(child_tidp);
 886         }
 887 #endif
 888         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 889 }
 890
 891 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 892              unsigned long p4, unsigned long p5, unsigned long p6,
 893              struct pt_regs *regs)
 894 {
 895         CHECK_FULL_REGS(regs);
 896         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 897 }
 898
 899 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
 900               unsigned long p4, unsigned long p5, unsigned long p6,
 901               struct pt_regs *regs)
 902 {
 903         CHECK_FULL_REGS(regs);
 904         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 905                         regs, 0, NULL, NULL);
 906 }
 907
 908 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 909                unsigned long a3, unsigned long a4, unsigned long a5,
 910                struct pt_regs *regs)
 911 {
 912         int error;
 913         char *filename;
 914
 915         filename = getname((char __user *) a0);
 916         error = PTR_ERR(filename);
 917         if (IS_ERR(filename))
 918                 goto out;
 919         flush_fp_to_thread(current);
 920         flush_altivec_to_thread(current);
 921         flush_spe_to_thread(current);
 922         error = do_execve(filename, (char __user * __user *) a1,
 923                           (char __user * __user *) a2, regs);
 924         putname(filename);
 925 out:
 926         return error;
 927 }
 928
 929 #ifdef CONFIG_IRQSTACKS
 930 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
 931                                   unsigned long nbytes)
 932 {
 933         unsigned long stack_page;
 934         unsigned long cpu = task_cpu(p);
 935
 936         /*
 937          * Avoid crashing if the stack has overflowed and corrupted
 938          * task_cpu(p), which is in the thread_info struct.
 939          */
 940         if (cpu < NR_CPUS && cpu_possible(cpu)) {
 941                 stack_page = (unsigned long) hardirq_ctx[cpu];
 942                 if (sp >= stack_page + sizeof(struct thread_struct)
 943                     && sp <= stack_page + THREAD_SIZE - nbytes)
 944                         return 1;
 945
 946                 stack_page = (unsigned long) softirq_ctx[cpu];
 947                 if (sp >= stack_page + sizeof(struct thread_struct)
 948                     && sp <= stack_page + THREAD_SIZE - nbytes)
 949                         return 1;
 950         }
 951         return 0;
 952 }
 953
 954 #else
 955 #define valid_irq_stack(sp, p, nb)      0
 956 #endif /* CONFIG_IRQSTACKS */
 957
 958 int validate_sp(unsigned long sp, struct task_struct *p,
 959                        unsigned long nbytes)
 960 {
 961         unsigned long stack_page = (unsigned long)task_stack_page(p);
 962
 963         if (sp >= stack_page + sizeof(struct thread_struct)
 964             && sp <= stack_page + THREAD_SIZE - nbytes)
 965                 return 1;
 966
 967         return valid_irq_stack(sp, p, nbytes);
 968 }
 969
 970 EXPORT_SYMBOL(validate_sp);
 971
 972 unsigned long get_wchan(struct task_struct *p)
 973 {
 974         unsigned long ip, sp;
 975         int count = 0;
 976
 977         if (!p || p == current || p->state == TASK_RUNNING)
 978                 return 0;
 979
 980         sp = p->thread.ksp;
 981         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 982                 return 0;
 983
 984         do {
 985                 sp = *(unsigned long *)sp;
 986                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 987                         return 0;
 988                 if (count > 0) {
 989                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
 990                         if (!in_sched_functions(ip))
 991                                 return ip;
 992                 }
 993         } while (count++ < 16);
 994         return 0;
 995 }
 996
 997 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
 998
 999 void show_stack(struct task_struct *tsk, unsigned long *stack)
1000 {
1001         unsigned long sp, ip, lr, newsp;
1002         int count = 0;
1003         int firstframe = 1;
1004 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1005         int curr_frame = current->curr_ret_stack;
1006         extern void return_to_handler(void);
1007         unsigned long rth = (unsigned long)return_to_handler;
1008         unsigned long mrth = -1;
1009 #ifdef CONFIG_PPC64
1010         extern void mod_return_to_handler(void);
1011         rth = *(unsigned long *)rth;
1012         mrth = (unsigned long)mod_return_to_handler;
1013         mrth = *(unsigned long *)mrth;
1014 #endif
1015 #endif
1016
1017         sp = (unsigned long) stack;
1018         if (tsk == NULL)
1019                 tsk = current;
1020         if (sp == 0) {
1021                 if (tsk == current)
1022                         asm("mr %0,1" : "=r" (sp));
1023                 else
1024                         sp = tsk->thread.ksp;
1025         }
1026
1027         lr = 0;
1028         printk("Call Trace:\n");
1029         do {
1030                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1031                         return;
1032
1033                 stack = (unsigned long *) sp;
1034                 newsp = stack[0];
1035                 ip = stack[STACK_FRAME_LR_SAVE];
1036                 if (!firstframe || ip != lr) {
1037                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1038 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1039                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1040                                 printk(" (%pS)",
1041                                        (void *)current->ret_stack[curr_frame].ret);
1042                                 curr_frame--;
1043                         }
1044 #endif
1045                         if (firstframe)
1046                                 printk(" (unreliable)");
1047                         printk("\n");
1048                 }
1049                 firstframe = 0;
1050
1051                 /*
1052                  * See if this is an exception frame.
1053                  * We look for the "regshere" marker in the current frame.
1054                  */
1055                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1056                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1057                         struct pt_regs *regs = (struct pt_regs *)
1058                                 (sp + STACK_FRAME_OVERHEAD);
1059                         lr = regs->link;
1060                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1061                                regs->trap, (void *)regs->nip, (void *)lr);
1062                         firstframe = 1;
1063                 }
1064
1065                 sp = newsp;
1066         } while (count++ < kstack_depth_to_print);
1067 }
1068
1069 void dump_stack(void)
1070 {
1071         show_stack(current, NULL);
1072 }
1073 EXPORT_SYMBOL(dump_stack);
1074
1075 #ifdef CONFIG_PPC64
1076 void ppc64_runlatch_on(void)
1077 {
1078         unsigned long ctrl;
1079
1080         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1081                 HMT_medium();
1082
1083                 ctrl = mfspr(SPRN_CTRLF);
1084                 ctrl |= CTRL_RUNLATCH;
1085                 mtspr(SPRN_CTRLT, ctrl);
1086
1087                 set_thread_flag(TIF_RUNLATCH);
1088         }
1089 }
1090
1091 void ppc64_runlatch_off(void)
1092 {
1093         unsigned long ctrl;
1094
1095         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1096                 HMT_medium();
1097
1098                 clear_thread_flag(TIF_RUNLATCH);
1099
1100                 ctrl = mfspr(SPRN_CTRLF);
1101                 ctrl &= ~CTRL_RUNLATCH;
1102                 mtspr(SPRN_CTRLT, ctrl);
1103         }
1104 }
1105 #endif
1106
1107 #if THREAD_SHIFT < PAGE_SHIFT
1108
1109 static struct kmem_cache *thread_info_cache;
1110
1111 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1112 {
1113         struct thread_info *ti;
1114
1115         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1116         if (unlikely(ti == NULL))
1117                 return NULL;
1118 #ifdef CONFIG_DEBUG_STACK_USAGE
1119         memset(ti, 0, THREAD_SIZE);
1120 #endif
1121         return ti;
1122 }
1123
1124 void free_thread_info(struct thread_info *ti)
1125 {
1126         kmem_cache_free(thread_info_cache, ti);
1127 }
1128
1129 void thread_info_cache_init(void)
1130 {
1131         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1132                                               THREAD_SIZE, 0, NULL);
1133         BUG_ON(thread_info_cache == NULL);
1134 }
1135
1136 #endif /* THREAD_SHIFT < PAGE_SHIFT */
1137
1138 unsigned long arch_align_stack(unsigned long sp)
1139 {
1140         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1141                 sp -= get_random_int() & ~PAGE_MASK;
1142         return sp & ~0xf;
1143 }
1144
1145 static inline unsigned long brk_rnd(void)
1146 {
1147         unsigned long rnd = 0;
1148
1149         /* 8MB for 32bit, 1GB for 64bit */
1150         if (is_32bit_task())
1151                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1152         else
1153                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1154
1155         return rnd << PAGE_SHIFT;
1156 }
1157
1158 unsigned long arch_randomize_brk(struct mm_struct *mm)
1159 {
1160         unsigned long base = mm->brk;
1161         unsigned long ret;
1162
1163 #ifdef CONFIG_PPC_STD_MMU_64
1164         /*
1165          * If we are using 1TB segments and we are allowed to randomise
1166          * the heap, we can put it above 1TB so it is backed by a 1TB
1167          * segment. Otherwise the heap will be in the bottom 1TB
1168          * which always uses 256MB segments and this may result in a
1169          * performance penalty.
1170          */
1171         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1172                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1173 #endif
1174
1175         ret = PAGE_ALIGN(base + brk_rnd());
1176
1177         if (ret < mm->brk)
1178                 return mm->brk;
1179
1180         return ret;
1181 }
1182
1183 unsigned long randomize_et_dyn(unsigned long base)
1184 {
1185         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1186
1187         if (ret < base)
1188                 return base;
1189
1190         return ret;
1191 }