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/export.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 #include <linux/hw_breakpoint.h>
  41
  42 #include <asm/pgtable.h>
  43 #include <asm/uaccess.h>
  44 #include <asm/system.h>
  45 #include <asm/io.h>
  46 #include <asm/processor.h>
  47 #include <asm/mmu.h>
  48 #include <asm/prom.h>
  49 #include <asm/machdep.h>
  50 #include <asm/time.h>
  51 #include <asm/syscalls.h>
  52 #ifdef CONFIG_PPC64
  53 #include <asm/firmware.h>
  54 #endif
  55 #include <linux/kprobes.h>
  56 #include <linux/kdebug.h>
  57
  58 extern unsigned long _get_SP(void);
  59
  60 #ifndef CONFIG_SMP
  61 struct task_struct *last_task_used_math = NULL;
  62 struct task_struct *last_task_used_altivec = NULL;
  63 struct task_struct *last_task_used_vsx = NULL;
  64 struct task_struct *last_task_used_spe = NULL;
  65 #endif
  66
  67 /*
  68  * Make sure the floating-point register state in the
  69  * the thread_struct is up to date for task tsk.
  70  */
  71 void flush_fp_to_thread(struct task_struct *tsk)
  72 {
  73         if (tsk->thread.regs) {
  74                 /*
  75                  * We need to disable preemption here because if we didn't,
  76                  * another process could get scheduled after the regs->msr
  77                  * test but before we have finished saving the FP registers
  78                  * to the thread_struct.  That process could take over the
  79                  * FPU, and then when we get scheduled again we would store
  80                  * bogus values for the remaining FP registers.
  81                  */
  82                 preempt_disable();
  83                 if (tsk->thread.regs->msr & MSR_FP) {
  84 #ifdef CONFIG_SMP
  85                         /*
  86                          * This should only ever be called for current or
  87                          * for a stopped child process.  Since we save away
  88                          * the FP register state on context switch on SMP,
  89                          * there is something wrong if a stopped child appears
  90                          * to still have its FP state in the CPU registers.
  91                          */
  92                         BUG_ON(tsk != current);
  93 #endif
  94                         giveup_fpu(tsk);
  95                 }
  96                 preempt_enable();
  97         }
  98 }
  99 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
 100
 101 void enable_kernel_fp(void)
 102 {
 103         WARN_ON(preemptible());
 104
 105 #ifdef CONFIG_SMP
 106         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
 107                 giveup_fpu(current);
 108         else
 109                 giveup_fpu(NULL);       /* just enables FP for kernel */
 110 #else
 111         giveup_fpu(last_task_used_math);
 112 #endif /* CONFIG_SMP */
 113 }
 114 EXPORT_SYMBOL(enable_kernel_fp);
 115
 116 #ifdef CONFIG_ALTIVEC
 117 void enable_kernel_altivec(void)
 118 {
 119         WARN_ON(preemptible());
 120
 121 #ifdef CONFIG_SMP
 122         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 123                 giveup_altivec(current);
 124         else
 125                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
 126 #else
 127         giveup_altivec(last_task_used_altivec);
 128 #endif /* CONFIG_SMP */
 129 }
 130 EXPORT_SYMBOL(enable_kernel_altivec);
 131
 132 /*
 133  * Make sure the VMX/Altivec register state in the
 134  * the thread_struct is up to date for task tsk.
 135  */
 136 void flush_altivec_to_thread(struct task_struct *tsk)
 137 {
 138         if (tsk->thread.regs) {
 139                 preempt_disable();
 140                 if (tsk->thread.regs->msr & MSR_VEC) {
 141 #ifdef CONFIG_SMP
 142                         BUG_ON(tsk != current);
 143 #endif
 144                         giveup_altivec(tsk);
 145                 }
 146                 preempt_enable();
 147         }
 148 }
 149 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
 150 #endif /* CONFIG_ALTIVEC */
 151
 152 #ifdef CONFIG_VSX
 153 #if 0
 154 /* not currently used, but some crazy RAID module might want to later */
 155 void enable_kernel_vsx(void)
 156 {
 157         WARN_ON(preemptible());
 158
 159 #ifdef CONFIG_SMP
 160         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 161                 giveup_vsx(current);
 162         else
 163                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 164 #else
 165         giveup_vsx(last_task_used_vsx);
 166 #endif /* CONFIG_SMP */
 167 }
 168 EXPORT_SYMBOL(enable_kernel_vsx);
 169 #endif
 170
 171 void giveup_vsx(struct task_struct *tsk)
 172 {
 173         giveup_fpu(tsk);
 174         giveup_altivec(tsk);
 175         __giveup_vsx(tsk);
 176 }
 177
 178 void flush_vsx_to_thread(struct task_struct *tsk)
 179 {
 180         if (tsk->thread.regs) {
 181                 preempt_disable();
 182                 if (tsk->thread.regs->msr & MSR_VSX) {
 183 #ifdef CONFIG_SMP
 184                         BUG_ON(tsk != current);
 185 #endif
 186                         giveup_vsx(tsk);
 187                 }
 188                 preempt_enable();
 189         }
 190 }
 191 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
 192 #endif /* CONFIG_VSX */
 193
 194 #ifdef CONFIG_SPE
 195
 196 void enable_kernel_spe(void)
 197 {
 198         WARN_ON(preemptible());
 199
 200 #ifdef CONFIG_SMP
 201         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 202                 giveup_spe(current);
 203         else
 204                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 205 #else
 206         giveup_spe(last_task_used_spe);
 207 #endif /* __SMP __ */
 208 }
 209 EXPORT_SYMBOL(enable_kernel_spe);
 210
 211 void flush_spe_to_thread(struct task_struct *tsk)
 212 {
 213         if (tsk->thread.regs) {
 214                 preempt_disable();
 215                 if (tsk->thread.regs->msr & MSR_SPE) {
 216 #ifdef CONFIG_SMP
 217                         BUG_ON(tsk != current);
 218 #endif
 219                         tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
 220                         giveup_spe(tsk);
 221                 }
 222                 preempt_enable();
 223         }
 224 }
 225 #endif /* CONFIG_SPE */
 226
 227 #ifndef CONFIG_SMP
 228 /*
 229  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 230  * and the current task has some state, discard it.
 231  */
 232 void discard_lazy_cpu_state(void)
 233 {
 234         preempt_disable();
 235         if (last_task_used_math == current)
 236                 last_task_used_math = NULL;
 237 #ifdef CONFIG_ALTIVEC
 238         if (last_task_used_altivec == current)
 239                 last_task_used_altivec = NULL;
 240 #endif /* CONFIG_ALTIVEC */
 241 #ifdef CONFIG_VSX
 242         if (last_task_used_vsx == current)
 243                 last_task_used_vsx = NULL;
 244 #endif /* CONFIG_VSX */
 245 #ifdef CONFIG_SPE
 246         if (last_task_used_spe == current)
 247                 last_task_used_spe = NULL;
 248 #endif
 249         preempt_enable();
 250 }
 251 #endif /* CONFIG_SMP */
 252
 253 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 254 void do_send_trap(struct pt_regs *regs, unsigned long address,
 255                   unsigned long error_code, int signal_code, int breakpt)
 256 {
 257         siginfo_t info;
 258
 259         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 260                         11, SIGSEGV) == NOTIFY_STOP)
 261                 return;
 262
 263         /* Deliver the signal to userspace */
 264         info.si_signo = SIGTRAP;
 265         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
 266         info.si_code = signal_code;
 267         info.si_addr = (void __user *)address;
 268         force_sig_info(SIGTRAP, &info, current);
 269 }
 270 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 271 void do_dabr(struct pt_regs *regs, unsigned long address,
 272                     unsigned long error_code)
 273 {
 274         siginfo_t info;
 275
 276         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 277                         11, SIGSEGV) == NOTIFY_STOP)
 278                 return;
 279
 280         if (debugger_dabr_match(regs))
 281                 return;
 282
 283         /* Clear the DABR */
 284         set_dabr(0);
 285
 286         /* Deliver the signal to userspace */
 287         info.si_signo = SIGTRAP;
 288         info.si_errno = 0;
 289         info.si_code = TRAP_HWBKPT;
 290         info.si_addr = (void __user *)address;
 291         force_sig_info(SIGTRAP, &info, current);
 292 }
 293 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 294
 295 static DEFINE_PER_CPU(unsigned long, current_dabr);
 296
 297 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 298 /*
 299  * Set the debug registers back to their default "safe" values.
 300  */
 301 static void set_debug_reg_defaults(struct thread_struct *thread)
 302 {
 303         thread->iac1 = thread->iac2 = 0;
 304 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 305         thread->iac3 = thread->iac4 = 0;
 306 #endif
 307         thread->dac1 = thread->dac2 = 0;
 308 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 309         thread->dvc1 = thread->dvc2 = 0;
 310 #endif
 311         thread->dbcr0 = 0;
 312 #ifdef CONFIG_BOOKE
 313         /*
 314          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
 315          */
 316         thread->dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |   \
 317                         DBCR1_IAC3US | DBCR1_IAC4US;
 318         /*
 319          * Force Data Address Compare User/Supervisor bits to be User-only
 320          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
 321          */
 322         thread->dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
 323 #else
 324         thread->dbcr1 = 0;
 325 #endif
 326 }
 327
 328 static void prime_debug_regs(struct thread_struct *thread)
 329 {
 330         mtspr(SPRN_IAC1, thread->iac1);
 331         mtspr(SPRN_IAC2, thread->iac2);
 332 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 333         mtspr(SPRN_IAC3, thread->iac3);
 334         mtspr(SPRN_IAC4, thread->iac4);
 335 #endif
 336         mtspr(SPRN_DAC1, thread->dac1);
 337         mtspr(SPRN_DAC2, thread->dac2);
 338 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 339         mtspr(SPRN_DVC1, thread->dvc1);
 340         mtspr(SPRN_DVC2, thread->dvc2);
 341 #endif
 342         mtspr(SPRN_DBCR0, thread->dbcr0);
 343         mtspr(SPRN_DBCR1, thread->dbcr1);
 344 #ifdef CONFIG_BOOKE
 345         mtspr(SPRN_DBCR2, thread->dbcr2);
 346 #endif
 347 }
 348 /*
 349  * Unless neither the old or new thread are making use of the
 350  * debug registers, set the debug registers from the values
 351  * stored in the new thread.
 352  */
 353 static void switch_booke_debug_regs(struct thread_struct *new_thread)
 354 {
 355         if ((current->thread.dbcr0 & DBCR0_IDM)
 356                 || (new_thread->dbcr0 & DBCR0_IDM))
 357                         prime_debug_regs(new_thread);
 358 }
 359 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 360 #ifndef CONFIG_HAVE_HW_BREAKPOINT
 361 static void set_debug_reg_defaults(struct thread_struct *thread)
 362 {
 363         if (thread->dabr) {
 364                 thread->dabr = 0;
 365                 set_dabr(0);
 366         }
 367 }
 368 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
 369 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 370
 371 int set_dabr(unsigned long dabr)
 372 {
 373         __get_cpu_var(current_dabr) = dabr;
 374
 375         if (ppc_md.set_dabr)
 376                 return ppc_md.set_dabr(dabr);
 377
 378         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 379 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 380         mtspr(SPRN_DAC1, dabr);
 381 #ifdef CONFIG_PPC_47x
 382         isync();
 383 #endif
 384 #elif defined(CONFIG_PPC_BOOK3S)
 385         mtspr(SPRN_DABR, dabr);
 386 #endif
 387
 388
 389         return 0;
 390 }
 391
 392 #ifdef CONFIG_PPC64
 393 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 394 #endif
 395
 396 struct task_struct *__switch_to(struct task_struct *prev,
 397         struct task_struct *new)
 398 {
 399         struct thread_struct *new_thread, *old_thread;
 400         unsigned long flags;
 401         struct task_struct *last;
 402 #ifdef CONFIG_PPC_BOOK3S_64
 403         struct ppc64_tlb_batch *batch;
 404 #endif
 405
 406 #ifdef CONFIG_SMP
 407         /* avoid complexity of lazy save/restore of fpu
 408          * by just saving it every time we switch out if
 409          * this task used the fpu during the last quantum.
 410          *
 411          * If it tries to use the fpu again, it'll trap and
 412          * reload its fp regs.  So we don't have to do a restore
 413          * every switch, just a save.
 414          *  -- Cort
 415          */
 416         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 417                 giveup_fpu(prev);
 418 #ifdef CONFIG_ALTIVEC
 419         /*
 420          * If the previous thread used altivec in the last quantum
 421          * (thus changing altivec regs) then save them.
 422          * We used to check the VRSAVE register but not all apps
 423          * set it, so we don't rely on it now (and in fact we need
 424          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 425          *
 426          * On SMP we always save/restore altivec regs just to avoid the
 427          * complexity of changing processors.
 428          *  -- Cort
 429          */
 430         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 431                 giveup_altivec(prev);
 432 #endif /* CONFIG_ALTIVEC */
 433 #ifdef CONFIG_VSX
 434         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 435                 /* VMX and FPU registers are already save here */
 436                 __giveup_vsx(prev);
 437 #endif /* CONFIG_VSX */
 438 #ifdef CONFIG_SPE
 439         /*
 440          * If the previous thread used spe in the last quantum
 441          * (thus changing spe regs) then save them.
 442          *
 443          * On SMP we always save/restore spe regs just to avoid the
 444          * complexity of changing processors.
 445          */
 446         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 447                 giveup_spe(prev);
 448 #endif /* CONFIG_SPE */
 449
 450 #else  /* CONFIG_SMP */
 451 #ifdef CONFIG_ALTIVEC
 452         /* Avoid the trap.  On smp this this never happens since
 453          * we don't set last_task_used_altivec -- Cort
 454          */
 455         if (new->thread.regs && last_task_used_altivec == new)
 456                 new->thread.regs->msr |= MSR_VEC;
 457 #endif /* CONFIG_ALTIVEC */
 458 #ifdef CONFIG_VSX
 459         if (new->thread.regs && last_task_used_vsx == new)
 460                 new->thread.regs->msr |= MSR_VSX;
 461 #endif /* CONFIG_VSX */
 462 #ifdef CONFIG_SPE
 463         /* Avoid the trap.  On smp this this never happens since
 464          * we don't set last_task_used_spe
 465          */
 466         if (new->thread.regs && last_task_used_spe == new)
 467                 new->thread.regs->msr |= MSR_SPE;
 468 #endif /* CONFIG_SPE */
 469
 470 #endif /* CONFIG_SMP */
 471
 472 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 473         switch_booke_debug_regs(&new->thread);
 474 #else
 475 /*
 476  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
 477  * schedule DABR
 478  */
 479 #ifndef CONFIG_HAVE_HW_BREAKPOINT
 480         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 481                 set_dabr(new->thread.dabr);
 482 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 483 #endif
 484
 485
 486         new_thread = &new->thread;
 487         old_thread = &current->thread;
 488
 489 #if defined(CONFIG_PPC_BOOK3E_64)
 490         /* XXX Current Book3E code doesn't deal with kernel side DBCR0,
 491          * we always hold the user values, so we set it now.
 492          *
 493          * However, we ensure the kernel MSR:DE is appropriately cleared too
 494          * to avoid spurrious single step exceptions in the kernel.
 495          *
 496          * This will have to change to merge with the ppc32 code at some point,
 497          * but I don't like much what ppc32 is doing today so there's some
 498          * thinking needed there
 499          */
 500         if ((new_thread->dbcr0 | old_thread->dbcr0) & DBCR0_IDM) {
 501                 u32 dbcr0;
 502
 503                 mtmsr(mfmsr() & ~MSR_DE);
 504                 isync();
 505                 dbcr0 = mfspr(SPRN_DBCR0);
 506                 dbcr0 = (dbcr0 & DBCR0_EDM) | new_thread->dbcr0;
 507                 mtspr(SPRN_DBCR0, dbcr0);
 508         }
 509 #endif /* CONFIG_PPC64_BOOK3E */
 510
 511 #ifdef CONFIG_PPC64
 512         /*
 513          * Collect processor utilization data per process
 514          */
 515         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 516                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 517                 long unsigned start_tb, current_tb;
 518                 start_tb = old_thread->start_tb;
 519                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 520                 old_thread->accum_tb += (current_tb - start_tb);
 521                 new_thread->start_tb = current_tb;
 522         }
 523 #endif /* CONFIG_PPC64 */
 524
 525 #ifdef CONFIG_PPC_BOOK3S_64
 526         batch = &__get_cpu_var(ppc64_tlb_batch);
 527         if (batch->active) {
 528                 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
 529                 if (batch->index)
 530                         __flush_tlb_pending(batch);
 531                 batch->active = 0;
 532         }
 533 #endif /* CONFIG_PPC_BOOK3S_64 */
 534
 535         local_irq_save(flags);
 536
 537         account_system_vtime(current);
 538         account_process_vtime(current);
 539
 540         /*
 541          * We can't take a PMU exception inside _switch() since there is a
 542          * window where the kernel stack SLB and the kernel stack are out
 543          * of sync. Hard disable here.
 544          */
 545         hard_irq_disable();
 546         last = _switch(old_thread, new_thread);
 547
 548 #ifdef CONFIG_PPC_BOOK3S_64
 549         if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
 550                 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
 551                 batch = &__get_cpu_var(ppc64_tlb_batch);
 552                 batch->active = 1;
 553         }
 554 #endif /* CONFIG_PPC_BOOK3S_64 */
 555
 556         local_irq_restore(flags);
 557
 558         return last;
 559 }
 560
 561 static int instructions_to_print = 16;
 562
 563 static void show_instructions(struct pt_regs *regs)
 564 {
 565         int i;
 566         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 567                         sizeof(int));
 568
 569         printk("Instruction dump:");
 570
 571         for (i = 0; i < instructions_to_print; i++) {
 572                 int instr;
 573
 574                 if (!(i % 8))
 575                         printk("\n");
 576
 577 #if !defined(CONFIG_BOOKE)
 578                 /* If executing with the IMMU off, adjust pc rather
 579                  * than print XXXXXXXX.
 580                  */
 581                 if (!(regs->msr & MSR_IR))
 582                         pc = (unsigned long)phys_to_virt(pc);
 583 #endif
 584
 585                 /* We use __get_user here *only* to avoid an OOPS on a
 586                  * bad address because the pc *should* only be a
 587                  * kernel address.
 588                  */
 589                 if (!__kernel_text_address(pc) ||
 590                      __get_user(instr, (unsigned int __user *)pc)) {
 591                         printk("XXXXXXXX ");
 592                 } else {
 593                         if (regs->nip == pc)
 594                                 printk("<%08x> ", instr);
 595                         else
 596                                 printk("%08x ", instr);
 597                 }
 598
 599                 pc += sizeof(int);
 600         }
 601
 602         printk("\n");
 603 }
 604
 605 static struct regbit {
 606         unsigned long bit;
 607         const char *name;
 608 } msr_bits[] = {
 609         {MSR_EE,        "EE"},
 610         {MSR_PR,        "PR"},
 611         {MSR_FP,        "FP"},
 612         {MSR_VEC,       "VEC"},
 613         {MSR_VSX,       "VSX"},
 614         {MSR_ME,        "ME"},
 615         {MSR_CE,        "CE"},
 616         {MSR_DE,        "DE"},
 617         {MSR_IR,        "IR"},
 618         {MSR_DR,        "DR"},
 619         {0,             NULL}
 620 };
 621
 622 static void printbits(unsigned long val, struct regbit *bits)
 623 {
 624         const char *sep = "";
 625
 626         printk("<");
 627         for (; bits->bit; ++bits)
 628                 if (val & bits->bit) {
 629                         printk("%s%s", sep, bits->name);
 630                         sep = ",";
 631                 }
 632         printk(">");
 633 }
 634
 635 #ifdef CONFIG_PPC64
 636 #define REG             "%016lx"
 637 #define REGS_PER_LINE   4
 638 #define LAST_VOLATILE   13
 639 #else
 640 #define REG             "%08lx"
 641 #define REGS_PER_LINE   8
 642 #define LAST_VOLATILE   12
 643 #endif
 644
 645 void show_regs(struct pt_regs * regs)
 646 {
 647         int i, trap;
 648
 649         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 650                regs->nip, regs->link, regs->ctr);
 651         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 652                regs, regs->trap, print_tainted(), init_utsname()->release);
 653         printk("MSR: "REG" ", regs->msr);
 654         printbits(regs->msr, msr_bits);
 655         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 656         trap = TRAP(regs);
 657         if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
 658                 printk("CFAR: "REG"\n", regs->orig_gpr3);
 659         if (trap == 0x300 || trap == 0x600)
 660 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 661                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 662 #else
 663                 printk("DAR: "REG", DSISR: %08lx\n", regs->dar, regs->dsisr);
 664 #endif
 665         printk("TASK = %p[%d] '%s' THREAD: %p",
 666                current, task_pid_nr(current), current->comm, task_thread_info(current));
 667
 668 #ifdef CONFIG_SMP
 669         printk(" CPU: %d", raw_smp_processor_id());
 670 #endif /* CONFIG_SMP */
 671
 672         for (i = 0;  i < 32;  i++) {
 673                 if ((i % REGS_PER_LINE) == 0)
 674                         printk("\nGPR%02d: ", i);
 675                 printk(REG " ", regs->gpr[i]);
 676                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 677                         break;
 678         }
 679         printk("\n");
 680 #ifdef CONFIG_KALLSYMS
 681         /*
 682          * Lookup NIP late so we have the best change of getting the
 683          * above info out without failing
 684          */
 685         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 686         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 687 #endif
 688         show_stack(current, (unsigned long *) regs->gpr[1]);
 689         if (!user_mode(regs))
 690                 show_instructions(regs);
 691 }
 692
 693 void exit_thread(void)
 694 {
 695         discard_lazy_cpu_state();
 696 }
 697
 698 void flush_thread(void)
 699 {
 700         discard_lazy_cpu_state();
 701
 702 #ifdef CONFIG_HAVE_HW_BREAKPOINT
 703         flush_ptrace_hw_breakpoint(current);
 704 #else /* CONFIG_HAVE_HW_BREAKPOINT */
 705         set_debug_reg_defaults(&current->thread);
 706 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 707 }
 708
 709 void
 710 release_thread(struct task_struct *t)
 711 {
 712 }
 713
 714 /*
 715  * This gets called before we allocate a new thread and copy
 716  * the current task into it.
 717  */
 718 void prepare_to_copy(struct task_struct *tsk)
 719 {
 720         flush_fp_to_thread(current);
 721         flush_altivec_to_thread(current);
 722         flush_vsx_to_thread(current);
 723         flush_spe_to_thread(current);
 724 #ifdef CONFIG_HAVE_HW_BREAKPOINT
 725         flush_ptrace_hw_breakpoint(tsk);
 726 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 727 }
 728
 729 /*
 730  * Copy a thread..
 731  */
 732 extern unsigned long dscr_default; /* defined in arch/powerpc/kernel/sysfs.c */
 733
 734 int copy_thread(unsigned long clone_flags, unsigned long usp,
 735                 unsigned long unused, struct task_struct *p,
 736                 struct pt_regs *regs)
 737 {
 738         struct pt_regs *childregs, *kregs;
 739         extern void ret_from_fork(void);
 740         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 741
 742         CHECK_FULL_REGS(regs);
 743         /* Copy registers */
 744         sp -= sizeof(struct pt_regs);
 745         childregs = (struct pt_regs *) sp;
 746         *childregs = *regs;
 747         if ((childregs->msr & MSR_PR) == 0) {
 748                 /* for kernel thread, set `current' and stackptr in new task */
 749                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 750 #ifdef CONFIG_PPC32
 751                 childregs->gpr[2] = (unsigned long) p;
 752 #else
 753                 clear_tsk_thread_flag(p, TIF_32BIT);
 754 #endif
 755                 p->thread.regs = NULL;  /* no user register state */
 756         } else {
 757                 childregs->gpr[1] = usp;
 758                 p->thread.regs = childregs;
 759                 if (clone_flags & CLONE_SETTLS) {
 760 #ifdef CONFIG_PPC64
 761                         if (!is_32bit_task())
 762                                 childregs->gpr[13] = childregs->gpr[6];
 763                         else
 764 #endif
 765                                 childregs->gpr[2] = childregs->gpr[6];
 766                 }
 767         }
 768         childregs->gpr[3] = 0;  /* Result from fork() */
 769         sp -= STACK_FRAME_OVERHEAD;
 770
 771         /*
 772          * The way this works is that at some point in the future
 773          * some task will call _switch to switch to the new task.
 774          * That will pop off the stack frame created below and start
 775          * the new task running at ret_from_fork.  The new task will
 776          * do some house keeping and then return from the fork or clone
 777          * system call, using the stack frame created above.
 778          */
 779         sp -= sizeof(struct pt_regs);
 780         kregs = (struct pt_regs *) sp;
 781         sp -= STACK_FRAME_OVERHEAD;
 782         p->thread.ksp = sp;
 783         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 784                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 785
 786 #ifdef CONFIG_PPC_STD_MMU_64
 787         if (mmu_has_feature(MMU_FTR_SLB)) {
 788                 unsigned long sp_vsid;
 789                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 790
 791                 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
 792                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 793                                 << SLB_VSID_SHIFT_1T;
 794                 else
 795                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 796                                 << SLB_VSID_SHIFT;
 797                 sp_vsid |= SLB_VSID_KERNEL | llp;
 798                 p->thread.ksp_vsid = sp_vsid;
 799         }
 800 #endif /* CONFIG_PPC_STD_MMU_64 */
 801 #ifdef CONFIG_PPC64
 802         if (cpu_has_feature(CPU_FTR_DSCR)) {
 803                 if (current->thread.dscr_inherit) {
 804                         p->thread.dscr_inherit = 1;
 805                         p->thread.dscr = current->thread.dscr;
 806                 } else if (0 != dscr_default) {
 807                         p->thread.dscr_inherit = 1;
 808                         p->thread.dscr = dscr_default;
 809                 } else {
 810                         p->thread.dscr_inherit = 0;
 811                         p->thread.dscr = 0;
 812                 }
 813         }
 814 #endif
 815
 816         /*
 817          * The PPC64 ABI makes use of a TOC to contain function
 818          * pointers.  The function (ret_from_except) is actually a pointer
 819          * to the TOC entry.  The first entry is a pointer to the actual
 820          * function.
 821          */
 822 #ifdef CONFIG_PPC64
 823         kregs->nip = *((unsigned long *)ret_from_fork);
 824 #else
 825         kregs->nip = (unsigned long)ret_from_fork;
 826 #endif
 827
 828         return 0;
 829 }
 830
 831 /*
 832  * Set up a thread for executing a new program
 833  */
 834 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 835 {
 836 #ifdef CONFIG_PPC64
 837         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 838 #endif
 839
 840         /*
 841          * If we exec out of a kernel thread then thread.regs will not be
 842          * set.  Do it now.
 843          */
 844         if (!current->thread.regs) {
 845                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 846                 current->thread.regs = regs - 1;
 847         }
 848
 849         memset(regs->gpr, 0, sizeof(regs->gpr));
 850         regs->ctr = 0;
 851         regs->link = 0;
 852         regs->xer = 0;
 853         regs->ccr = 0;
 854         regs->gpr[1] = sp;
 855
 856         /*
 857          * We have just cleared all the nonvolatile GPRs, so make
 858          * FULL_REGS(regs) return true.  This is necessary to allow
 859          * ptrace to examine the thread immediately after exec.
 860          */
 861         regs->trap &= ~1UL;
 862
 863 #ifdef CONFIG_PPC32
 864         regs->mq = 0;
 865         regs->nip = start;
 866         regs->msr = MSR_USER;
 867 #else
 868         if (!is_32bit_task()) {
 869                 unsigned long entry, toc;
 870
 871                 /* start is a relocated pointer to the function descriptor for
 872                  * the elf _start routine.  The first entry in the function
 873                  * descriptor is the entry address of _start and the second
 874                  * entry is the TOC value we need to use.
 875                  */
 876                 __get_user(entry, (unsigned long __user *)start);
 877                 __get_user(toc, (unsigned long __user *)start+1);
 878
 879                 /* Check whether the e_entry function descriptor entries
 880                  * need to be relocated before we can use them.
 881                  */
 882                 if (load_addr != 0) {
 883                         entry += load_addr;
 884                         toc   += load_addr;
 885                 }
 886                 regs->nip = entry;
 887                 regs->gpr[2] = toc;
 888                 regs->msr = MSR_USER64;
 889         } else {
 890                 regs->nip = start;
 891                 regs->gpr[2] = 0;
 892                 regs->msr = MSR_USER32;
 893         }
 894 #endif
 895
 896         discard_lazy_cpu_state();
 897 #ifdef CONFIG_VSX
 898         current->thread.used_vsr = 0;
 899 #endif
 900         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 901         current->thread.fpscr.val = 0;
 902 #ifdef CONFIG_ALTIVEC
 903         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 904         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 905         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 906         current->thread.vrsave = 0;
 907         current->thread.used_vr = 0;
 908 #endif /* CONFIG_ALTIVEC */
 909 #ifdef CONFIG_SPE
 910         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 911         current->thread.acc = 0;
 912         current->thread.spefscr = 0;
 913         current->thread.used_spe = 0;
 914 #endif /* CONFIG_SPE */
 915 }
 916
 917 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 918                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 919
 920 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 921 {
 922         struct pt_regs *regs = tsk->thread.regs;
 923
 924         /* This is a bit hairy.  If we are an SPE enabled  processor
 925          * (have embedded fp) we store the IEEE exception enable flags in
 926          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 927          * mode (asyn, precise, disabled) for 'Classic' FP. */
 928         if (val & PR_FP_EXC_SW_ENABLE) {
 929 #ifdef CONFIG_SPE
 930                 if (cpu_has_feature(CPU_FTR_SPE)) {
 931                         tsk->thread.fpexc_mode = val &
 932                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 933                         return 0;
 934                 } else {
 935                         return -EINVAL;
 936                 }
 937 #else
 938                 return -EINVAL;
 939 #endif
 940         }
 941
 942         /* on a CONFIG_SPE this does not hurt us.  The bits that
 943          * __pack_fe01 use do not overlap with bits used for
 944          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 945          * on CONFIG_SPE implementations are reserved so writing to
 946          * them does not change anything */
 947         if (val > PR_FP_EXC_PRECISE)
 948                 return -EINVAL;
 949         tsk->thread.fpexc_mode = __pack_fe01(val);
 950         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 951                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 952                         | tsk->thread.fpexc_mode;
 953         return 0;
 954 }
 955
 956 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 957 {
 958         unsigned int val;
 959
 960         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 961 #ifdef CONFIG_SPE
 962                 if (cpu_has_feature(CPU_FTR_SPE))
 963                         val = tsk->thread.fpexc_mode;
 964                 else
 965                         return -EINVAL;
 966 #else
 967                 return -EINVAL;
 968 #endif
 969         else
 970                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 971         return put_user(val, (unsigned int __user *) adr);
 972 }
 973
 974 int set_endian(struct task_struct *tsk, unsigned int val)
 975 {
 976         struct pt_regs *regs = tsk->thread.regs;
 977
 978         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 979             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 980                 return -EINVAL;
 981
 982         if (regs == NULL)
 983                 return -EINVAL;
 984
 985         if (val == PR_ENDIAN_BIG)
 986                 regs->msr &= ~MSR_LE;
 987         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 988                 regs->msr |= MSR_LE;
 989         else
 990                 return -EINVAL;
 991
 992         return 0;
 993 }
 994
 995 int get_endian(struct task_struct *tsk, unsigned long adr)
 996 {
 997         struct pt_regs *regs = tsk->thread.regs;
 998         unsigned int val;
 999
1000         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1001             !cpu_has_feature(CPU_FTR_REAL_LE))
1002                 return -EINVAL;
1003
1004         if (regs == NULL)
1005                 return -EINVAL;
1006
1007         if (regs->msr & MSR_LE) {
1008                 if (cpu_has_feature(CPU_FTR_REAL_LE))
1009                         val = PR_ENDIAN_LITTLE;
1010                 else
1011                         val = PR_ENDIAN_PPC_LITTLE;
1012         } else
1013                 val = PR_ENDIAN_BIG;
1014
1015         return put_user(val, (unsigned int __user *)adr);
1016 }
1017
1018 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1019 {
1020         tsk->thread.align_ctl = val;
1021         return 0;
1022 }
1023
1024 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1025 {
1026         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1027 }
1028
1029 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
1030
1031 int sys_clone(unsigned long clone_flags, unsigned long usp,
1032               int __user *parent_tidp, void __user *child_threadptr,
1033               int __user *child_tidp, int p6,
1034               struct pt_regs *regs)
1035 {
1036         CHECK_FULL_REGS(regs);
1037         if (usp == 0)
1038                 usp = regs->gpr[1];     /* stack pointer for child */
1039 #ifdef CONFIG_PPC64
1040         if (is_32bit_task()) {
1041                 parent_tidp = TRUNC_PTR(parent_tidp);
1042                 child_tidp = TRUNC_PTR(child_tidp);
1043         }
1044 #endif
1045         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
1046 }
1047
1048 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
1049              unsigned long p4, unsigned long p5, unsigned long p6,
1050              struct pt_regs *regs)
1051 {
1052         CHECK_FULL_REGS(regs);
1053         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
1054 }
1055
1056 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
1057               unsigned long p4, unsigned long p5, unsigned long p6,
1058               struct pt_regs *regs)
1059 {
1060         CHECK_FULL_REGS(regs);
1061         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
1062                         regs, 0, NULL, NULL);
1063 }
1064
1065 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
1066                unsigned long a3, unsigned long a4, unsigned long a5,
1067                struct pt_regs *regs)
1068 {
1069         int error;
1070         char *filename;
1071
1072         filename = getname((const char __user *) a0);
1073         error = PTR_ERR(filename);
1074         if (IS_ERR(filename))
1075                 goto out;
1076         flush_fp_to_thread(current);
1077         flush_altivec_to_thread(current);
1078         flush_spe_to_thread(current);
1079         error = do_execve(filename,
1080                           (const char __user *const __user *) a1,
1081                           (const char __user *const __user *) a2, regs);
1082         putname(filename);
1083 out:
1084         return error;
1085 }
1086
1087 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1088                                   unsigned long nbytes)
1089 {
1090         unsigned long stack_page;
1091         unsigned long cpu = task_cpu(p);
1092
1093         /*
1094          * Avoid crashing if the stack has overflowed and corrupted
1095          * task_cpu(p), which is in the thread_info struct.
1096          */
1097         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1098                 stack_page = (unsigned long) hardirq_ctx[cpu];
1099                 if (sp >= stack_page + sizeof(struct thread_struct)
1100                     && sp <= stack_page + THREAD_SIZE - nbytes)
1101                         return 1;
1102
1103                 stack_page = (unsigned long) softirq_ctx[cpu];
1104                 if (sp >= stack_page + sizeof(struct thread_struct)
1105                     && sp <= stack_page + THREAD_SIZE - nbytes)
1106                         return 1;
1107         }
1108         return 0;
1109 }
1110
1111 int validate_sp(unsigned long sp, struct task_struct *p,
1112                        unsigned long nbytes)
1113 {
1114         unsigned long stack_page = (unsigned long)task_stack_page(p);
1115
1116         if (sp >= stack_page + sizeof(struct thread_struct)
1117             && sp <= stack_page + THREAD_SIZE - nbytes)
1118                 return 1;
1119
1120         return valid_irq_stack(sp, p, nbytes);
1121 }
1122
1123 EXPORT_SYMBOL(validate_sp);
1124
1125 unsigned long get_wchan(struct task_struct *p)
1126 {
1127         unsigned long ip, sp;
1128         int count = 0;
1129
1130         if (!p || p == current || p->state == TASK_RUNNING)
1131                 return 0;
1132
1133         sp = p->thread.ksp;
1134         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1135                 return 0;
1136
1137         do {
1138                 sp = *(unsigned long *)sp;
1139                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1140                         return 0;
1141                 if (count > 0) {
1142                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1143                         if (!in_sched_functions(ip))
1144                                 return ip;
1145                 }
1146         } while (count++ < 16);
1147         return 0;
1148 }
1149
1150 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1151
1152 void show_stack(struct task_struct *tsk, unsigned long *stack)
1153 {
1154         unsigned long sp, ip, lr, newsp;
1155         int count = 0;
1156         int firstframe = 1;
1157 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1158         int curr_frame = current->curr_ret_stack;
1159         extern void return_to_handler(void);
1160         unsigned long rth = (unsigned long)return_to_handler;
1161         unsigned long mrth = -1;
1162 #ifdef CONFIG_PPC64
1163         extern void mod_return_to_handler(void);
1164         rth = *(unsigned long *)rth;
1165         mrth = (unsigned long)mod_return_to_handler;
1166         mrth = *(unsigned long *)mrth;
1167 #endif
1168 #endif
1169
1170         sp = (unsigned long) stack;
1171         if (tsk == NULL)
1172                 tsk = current;
1173         if (sp == 0) {
1174                 if (tsk == current)
1175                         asm("mr %0,1" : "=r" (sp));
1176                 else
1177                         sp = tsk->thread.ksp;
1178         }
1179
1180         lr = 0;
1181         printk("Call Trace:\n");
1182         do {
1183                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1184                         return;
1185
1186                 stack = (unsigned long *) sp;
1187                 newsp = stack[0];
1188                 ip = stack[STACK_FRAME_LR_SAVE];
1189                 if (!firstframe || ip != lr) {
1190                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1191 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1192                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1193                                 printk(" (%pS)",
1194                                        (void *)current->ret_stack[curr_frame].ret);
1195                                 curr_frame--;
1196                         }
1197 #endif
1198                         if (firstframe)
1199                                 printk(" (unreliable)");
1200                         printk("\n");
1201                 }
1202                 firstframe = 0;
1203
1204                 /*
1205                  * See if this is an exception frame.
1206                  * We look for the "regshere" marker in the current frame.
1207                  */
1208                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1209                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1210                         struct pt_regs *regs = (struct pt_regs *)
1211                                 (sp + STACK_FRAME_OVERHEAD);
1212                         lr = regs->link;
1213                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1214                                regs->trap, (void *)regs->nip, (void *)lr);
1215                         firstframe = 1;
1216                 }
1217
1218                 sp = newsp;
1219         } while (count++ < kstack_depth_to_print);
1220 }
1221
1222 void dump_stack(void)
1223 {
1224         show_stack(current, NULL);
1225 }
1226 EXPORT_SYMBOL(dump_stack);
1227
1228 #ifdef CONFIG_PPC64
1229 void ppc64_runlatch_on(void)
1230 {
1231         unsigned long ctrl;
1232
1233         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1234                 HMT_medium();
1235
1236                 ctrl = mfspr(SPRN_CTRLF);
1237                 ctrl |= CTRL_RUNLATCH;
1238                 mtspr(SPRN_CTRLT, ctrl);
1239
1240                 set_thread_flag(TIF_RUNLATCH);
1241         }
1242 }
1243
1244 void __ppc64_runlatch_off(void)
1245 {
1246         unsigned long ctrl;
1247
1248         HMT_medium();
1249
1250         clear_thread_flag(TIF_RUNLATCH);
1251
1252         ctrl = mfspr(SPRN_CTRLF);
1253         ctrl &= ~CTRL_RUNLATCH;
1254         mtspr(SPRN_CTRLT, ctrl);
1255 }
1256 #endif
1257
1258 #if THREAD_SHIFT < PAGE_SHIFT
1259
1260 static struct kmem_cache *thread_info_cache;
1261
1262 struct thread_info *alloc_thread_info_node(struct task_struct *tsk, int node)
1263 {
1264         struct thread_info *ti;
1265
1266         ti = kmem_cache_alloc_node(thread_info_cache, GFP_KERNEL, node);
1267         if (unlikely(ti == NULL))
1268                 return NULL;
1269 #ifdef CONFIG_DEBUG_STACK_USAGE
1270         memset(ti, 0, THREAD_SIZE);
1271 #endif
1272         return ti;
1273 }
1274
1275 void free_thread_info(struct thread_info *ti)
1276 {
1277         kmem_cache_free(thread_info_cache, ti);
1278 }
1279
1280 void thread_info_cache_init(void)
1281 {
1282         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1283                                               THREAD_SIZE, 0, NULL);
1284         BUG_ON(thread_info_cache == NULL);
1285 }
1286
1287 #endif /* THREAD_SHIFT < PAGE_SHIFT */
1288
1289 unsigned long arch_align_stack(unsigned long sp)
1290 {
1291         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1292                 sp -= get_random_int() & ~PAGE_MASK;
1293         return sp & ~0xf;
1294 }
1295
1296 static inline unsigned long brk_rnd(void)
1297 {
1298         unsigned long rnd = 0;
1299
1300         /* 8MB for 32bit, 1GB for 64bit */
1301         if (is_32bit_task())
1302                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1303         else
1304                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1305
1306         return rnd << PAGE_SHIFT;
1307 }
1308
1309 unsigned long arch_randomize_brk(struct mm_struct *mm)
1310 {
1311         unsigned long base = mm->brk;
1312         unsigned long ret;
1313
1314 #ifdef CONFIG_PPC_STD_MMU_64
1315         /*
1316          * If we are using 1TB segments and we are allowed to randomise
1317          * the heap, we can put it above 1TB so it is backed by a 1TB
1318          * segment. Otherwise the heap will be in the bottom 1TB
1319          * which always uses 256MB segments and this may result in a
1320          * performance penalty.
1321          */
1322         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1323                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1324 #endif
1325
1326         ret = PAGE_ALIGN(base + brk_rnd());
1327
1328         if (ret < mm->brk)
1329                 return mm->brk;
1330
1331         return ret;
1332 }
1333
1334 unsigned long randomize_et_dyn(unsigned long base)
1335 {
1336         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1337
1338         if (ret < base)
1339                 return base;
1340
1341         return ret;
1342 }