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/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/runlatch.h>
  51 #include <asm/syscalls.h>
  52 #include <asm/switch_to.h>
  53 #include <asm/tm.h>
  54 #include <asm/debug.h>
  55 #ifdef CONFIG_PPC64
  56 #include <asm/firmware.h>
  57 #endif
  58 #include <linux/kprobes.h>
  59 #include <linux/kdebug.h>
  60
  61 /* Transactional Memory debug */
  62 #ifdef TM_DEBUG_SW
  63 #define TM_DEBUG(x...) printk(KERN_INFO x)
  64 #else
  65 #define TM_DEBUG(x...) do { } while(0)
  66 #endif
  67
  68 extern unsigned long _get_SP(void);
  69
  70 #ifndef CONFIG_SMP
  71 struct task_struct *last_task_used_math = NULL;
  72 struct task_struct *last_task_used_altivec = NULL;
  73 struct task_struct *last_task_used_vsx = NULL;
  74 struct task_struct *last_task_used_spe = NULL;
  75 #endif
  76
  77 #ifdef CONFIG_PPC_FPU
  78 /*
  79  * Make sure the floating-point register state in the
  80  * the thread_struct is up to date for task tsk.
  81  */
  82 void flush_fp_to_thread(struct task_struct *tsk)
  83 {
  84         if (tsk->thread.regs) {
  85                 /*
  86                  * We need to disable preemption here because if we didn't,
  87                  * another process could get scheduled after the regs->msr
  88                  * test but before we have finished saving the FP registers
  89                  * to the thread_struct.  That process could take over the
  90                  * FPU, and then when we get scheduled again we would store
  91                  * bogus values for the remaining FP registers.
  92                  */
  93                 preempt_disable();
  94                 if (tsk->thread.regs->msr & MSR_FP) {
  95 #ifdef CONFIG_SMP
  96                         /*
  97                          * This should only ever be called for current or
  98                          * for a stopped child process.  Since we save away
  99                          * the FP register state on context switch on SMP,
 100                          * there is something wrong if a stopped child appears
 101                          * to still have its FP state in the CPU registers.
 102                          */
 103                         BUG_ON(tsk != current);
 104 #endif
 105                         giveup_fpu(tsk);
 106                 }
 107                 preempt_enable();
 108         }
 109 }
 110 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
 111 #endif
 112
 113 void enable_kernel_fp(void)
 114 {
 115         WARN_ON(preemptible());
 116
 117 #ifdef CONFIG_SMP
 118         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
 119                 giveup_fpu(current);
 120         else
 121                 giveup_fpu(NULL);       /* just enables FP for kernel */
 122 #else
 123         giveup_fpu(last_task_used_math);
 124 #endif /* CONFIG_SMP */
 125 }
 126 EXPORT_SYMBOL(enable_kernel_fp);
 127
 128 #ifdef CONFIG_ALTIVEC
 129 void enable_kernel_altivec(void)
 130 {
 131         WARN_ON(preemptible());
 132
 133 #ifdef CONFIG_SMP
 134         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 135                 giveup_altivec(current);
 136         else
 137                 giveup_altivec_notask();
 138 #else
 139         giveup_altivec(last_task_used_altivec);
 140 #endif /* CONFIG_SMP */
 141 }
 142 EXPORT_SYMBOL(enable_kernel_altivec);
 143
 144 /*
 145  * Make sure the VMX/Altivec register state in the
 146  * the thread_struct is up to date for task tsk.
 147  */
 148 void flush_altivec_to_thread(struct task_struct *tsk)
 149 {
 150         if (tsk->thread.regs) {
 151                 preempt_disable();
 152                 if (tsk->thread.regs->msr & MSR_VEC) {
 153 #ifdef CONFIG_SMP
 154                         BUG_ON(tsk != current);
 155 #endif
 156                         giveup_altivec(tsk);
 157                 }
 158                 preempt_enable();
 159         }
 160 }
 161 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
 162 #endif /* CONFIG_ALTIVEC */
 163
 164 #ifdef CONFIG_VSX
 165 #if 0
 166 /* not currently used, but some crazy RAID module might want to later */
 167 void enable_kernel_vsx(void)
 168 {
 169         WARN_ON(preemptible());
 170
 171 #ifdef CONFIG_SMP
 172         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 173                 giveup_vsx(current);
 174         else
 175                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 176 #else
 177         giveup_vsx(last_task_used_vsx);
 178 #endif /* CONFIG_SMP */
 179 }
 180 EXPORT_SYMBOL(enable_kernel_vsx);
 181 #endif
 182
 183 void giveup_vsx(struct task_struct *tsk)
 184 {
 185         giveup_fpu(tsk);
 186         giveup_altivec(tsk);
 187         __giveup_vsx(tsk);
 188 }
 189
 190 void flush_vsx_to_thread(struct task_struct *tsk)
 191 {
 192         if (tsk->thread.regs) {
 193                 preempt_disable();
 194                 if (tsk->thread.regs->msr & MSR_VSX) {
 195 #ifdef CONFIG_SMP
 196                         BUG_ON(tsk != current);
 197 #endif
 198                         giveup_vsx(tsk);
 199                 }
 200                 preempt_enable();
 201         }
 202 }
 203 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
 204 #endif /* CONFIG_VSX */
 205
 206 #ifdef CONFIG_SPE
 207
 208 void enable_kernel_spe(void)
 209 {
 210         WARN_ON(preemptible());
 211
 212 #ifdef CONFIG_SMP
 213         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 214                 giveup_spe(current);
 215         else
 216                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 217 #else
 218         giveup_spe(last_task_used_spe);
 219 #endif /* __SMP __ */
 220 }
 221 EXPORT_SYMBOL(enable_kernel_spe);
 222
 223 void flush_spe_to_thread(struct task_struct *tsk)
 224 {
 225         if (tsk->thread.regs) {
 226                 preempt_disable();
 227                 if (tsk->thread.regs->msr & MSR_SPE) {
 228 #ifdef CONFIG_SMP
 229                         BUG_ON(tsk != current);
 230 #endif
 231                         tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
 232                         giveup_spe(tsk);
 233                 }
 234                 preempt_enable();
 235         }
 236 }
 237 #endif /* CONFIG_SPE */
 238
 239 #ifndef CONFIG_SMP
 240 /*
 241  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 242  * and the current task has some state, discard it.
 243  */
 244 void discard_lazy_cpu_state(void)
 245 {
 246         preempt_disable();
 247         if (last_task_used_math == current)
 248                 last_task_used_math = NULL;
 249 #ifdef CONFIG_ALTIVEC
 250         if (last_task_used_altivec == current)
 251                 last_task_used_altivec = NULL;
 252 #endif /* CONFIG_ALTIVEC */
 253 #ifdef CONFIG_VSX
 254         if (last_task_used_vsx == current)
 255                 last_task_used_vsx = NULL;
 256 #endif /* CONFIG_VSX */
 257 #ifdef CONFIG_SPE
 258         if (last_task_used_spe == current)
 259                 last_task_used_spe = NULL;
 260 #endif
 261         preempt_enable();
 262 }
 263 #endif /* CONFIG_SMP */
 264
 265 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 266 void do_send_trap(struct pt_regs *regs, unsigned long address,
 267                   unsigned long error_code, int signal_code, int breakpt)
 268 {
 269         siginfo_t info;
 270
 271         current->thread.trap_nr = signal_code;
 272         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 273                         11, SIGSEGV) == NOTIFY_STOP)
 274                 return;
 275
 276         /* Deliver the signal to userspace */
 277         info.si_signo = SIGTRAP;
 278         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
 279         info.si_code = signal_code;
 280         info.si_addr = (void __user *)address;
 281         force_sig_info(SIGTRAP, &info, current);
 282 }
 283 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 284 void do_break (struct pt_regs *regs, unsigned long address,
 285                     unsigned long error_code)
 286 {
 287         siginfo_t info;
 288
 289         current->thread.trap_nr = TRAP_HWBKPT;
 290         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 291                         11, SIGSEGV) == NOTIFY_STOP)
 292                 return;
 293
 294         if (debugger_break_match(regs))
 295                 return;
 296
 297         /* Clear the breakpoint */
 298         hw_breakpoint_disable();
 299
 300         /* Deliver the signal to userspace */
 301         info.si_signo = SIGTRAP;
 302         info.si_errno = 0;
 303         info.si_code = TRAP_HWBKPT;
 304         info.si_addr = (void __user *)address;
 305         force_sig_info(SIGTRAP, &info, current);
 306 }
 307 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 308
 309 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
 310
 311 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 312 /*
 313  * Set the debug registers back to their default "safe" values.
 314  */
 315 static void set_debug_reg_defaults(struct thread_struct *thread)
 316 {
 317         thread->debug.iac1 = thread->debug.iac2 = 0;
 318 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 319         thread->debug.iac3 = thread->debug.iac4 = 0;
 320 #endif
 321         thread->debug.dac1 = thread->debug.dac2 = 0;
 322 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 323         thread->debug.dvc1 = thread->debug.dvc2 = 0;
 324 #endif
 325         thread->debug.dbcr0 = 0;
 326 #ifdef CONFIG_BOOKE
 327         /*
 328          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
 329          */
 330         thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
 331                         DBCR1_IAC3US | DBCR1_IAC4US;
 332         /*
 333          * Force Data Address Compare User/Supervisor bits to be User-only
 334          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
 335          */
 336         thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
 337 #else
 338         thread->debug.dbcr1 = 0;
 339 #endif
 340 }
 341
 342 static void prime_debug_regs(struct thread_struct *thread)
 343 {
 344         /*
 345          * We could have inherited MSR_DE from userspace, since
 346          * it doesn't get cleared on exception entry.  Make sure
 347          * MSR_DE is clear before we enable any debug events.
 348          */
 349         mtmsr(mfmsr() & ~MSR_DE);
 350
 351         mtspr(SPRN_IAC1, thread->debug.iac1);
 352         mtspr(SPRN_IAC2, thread->debug.iac2);
 353 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 354         mtspr(SPRN_IAC3, thread->debug.iac3);
 355         mtspr(SPRN_IAC4, thread->debug.iac4);
 356 #endif
 357         mtspr(SPRN_DAC1, thread->debug.dac1);
 358         mtspr(SPRN_DAC2, thread->debug.dac2);
 359 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 360         mtspr(SPRN_DVC1, thread->debug.dvc1);
 361         mtspr(SPRN_DVC2, thread->debug.dvc2);
 362 #endif
 363         mtspr(SPRN_DBCR0, thread->debug.dbcr0);
 364         mtspr(SPRN_DBCR1, thread->debug.dbcr1);
 365 #ifdef CONFIG_BOOKE
 366         mtspr(SPRN_DBCR2, thread->debug.dbcr2);
 367 #endif
 368 }
 369 /*
 370  * Unless neither the old or new thread are making use of the
 371  * debug registers, set the debug registers from the values
 372  * stored in the new thread.
 373  */
 374 void switch_booke_debug_regs(struct thread_struct *new_thread)
 375 {
 376         if ((current->thread.debug.dbcr0 & DBCR0_IDM)
 377                 || (new_thread->debug.dbcr0 & DBCR0_IDM))
 378                         prime_debug_regs(new_thread);
 379 }
 380 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
 381 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 382 #ifndef CONFIG_HAVE_HW_BREAKPOINT
 383 static void set_debug_reg_defaults(struct thread_struct *thread)
 384 {
 385         thread->hw_brk.address = 0;
 386         thread->hw_brk.type = 0;
 387         set_breakpoint(&thread->hw_brk);
 388 }
 389 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
 390 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 391
 392 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 393 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 394 {
 395         mtspr(SPRN_DAC1, dabr);
 396 #ifdef CONFIG_PPC_47x
 397         isync();
 398 #endif
 399         return 0;
 400 }
 401 #elif defined(CONFIG_PPC_BOOK3S)
 402 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 403 {
 404         mtspr(SPRN_DABR, dabr);
 405         if (cpu_has_feature(CPU_FTR_DABRX))
 406                 mtspr(SPRN_DABRX, dabrx);
 407         return 0;
 408 }
 409 #else
 410 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 411 {
 412         return -EINVAL;
 413 }
 414 #endif
 415
 416 static inline int set_dabr(struct arch_hw_breakpoint *brk)
 417 {
 418         unsigned long dabr, dabrx;
 419
 420         dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
 421         dabrx = ((brk->type >> 3) & 0x7);
 422
 423         if (ppc_md.set_dabr)
 424                 return ppc_md.set_dabr(dabr, dabrx);
 425
 426         return __set_dabr(dabr, dabrx);
 427 }
 428
 429 static inline int set_dawr(struct arch_hw_breakpoint *brk)
 430 {
 431         unsigned long dawr, dawrx, mrd;
 432
 433         dawr = brk->address;
 434
 435         dawrx  = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
 436                                    << (63 - 58); //* read/write bits */
 437         dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
 438                                    << (63 - 59); //* translate */
 439         dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
 440                                    >> 3; //* PRIM bits */
 441         /* dawr length is stored in field MDR bits 48:53.  Matches range in
 442            doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
 443            0b111111=64DW.
 444            brk->len is in bytes.
 445            This aligns up to double word size, shifts and does the bias.
 446         */
 447         mrd = ((brk->len + 7) >> 3) - 1;
 448         dawrx |= (mrd & 0x3f) << (63 - 53);
 449
 450         if (ppc_md.set_dawr)
 451                 return ppc_md.set_dawr(dawr, dawrx);
 452         mtspr(SPRN_DAWR, dawr);
 453         mtspr(SPRN_DAWRX, dawrx);
 454         return 0;
 455 }
 456
 457 int set_breakpoint(struct arch_hw_breakpoint *brk)
 458 {
 459         __get_cpu_var(current_brk) = *brk;
 460
 461         if (cpu_has_feature(CPU_FTR_DAWR))
 462                 return set_dawr(brk);
 463
 464         return set_dabr(brk);
 465 }
 466
 467 #ifdef CONFIG_PPC64
 468 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 469 #endif
 470
 471 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
 472                               struct arch_hw_breakpoint *b)
 473 {
 474         if (a->address != b->address)
 475                 return false;
 476         if (a->type != b->type)
 477                 return false;
 478         if (a->len != b->len)
 479                 return false;
 480         return true;
 481 }
 482 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 483 static inline void tm_reclaim_task(struct task_struct *tsk)
 484 {
 485         /* We have to work out if we're switching from/to a task that's in the
 486          * middle of a transaction.
 487          *
 488          * In switching we need to maintain a 2nd register state as
 489          * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
 490          * checkpointed (tbegin) state in ckpt_regs and saves the transactional
 491          * (current) FPRs into oldtask->thread.transact_fpr[].
 492          *
 493          * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
 494          */
 495         struct thread_struct *thr = &tsk->thread;
 496
 497         if (!thr->regs)
 498                 return;
 499
 500         if (!MSR_TM_ACTIVE(thr->regs->msr))
 501                 goto out_and_saveregs;
 502
 503         /* Stash the original thread MSR, as giveup_fpu et al will
 504          * modify it.  We hold onto it to see whether the task used
 505          * FP & vector regs.
 506          */
 507         thr->tm_orig_msr = thr->regs->msr;
 508
 509         TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
 510                  "ccr=%lx, msr=%lx, trap=%lx)\n",
 511                  tsk->pid, thr->regs->nip,
 512                  thr->regs->ccr, thr->regs->msr,
 513                  thr->regs->trap);
 514
 515         tm_reclaim(thr, thr->regs->msr, TM_CAUSE_RESCHED);
 516
 517         TM_DEBUG("--- tm_reclaim on pid %d complete\n",
 518                  tsk->pid);
 519
 520 out_and_saveregs:
 521         /* Always save the regs here, even if a transaction's not active.
 522          * This context-switches a thread's TM info SPRs.  We do it here to
 523          * be consistent with the restore path (in recheckpoint) which
 524          * cannot happen later in _switch().
 525          */
 526         tm_save_sprs(thr);
 527 }
 528
 529 static inline void tm_recheckpoint_new_task(struct task_struct *new)
 530 {
 531         unsigned long msr;
 532
 533         if (!cpu_has_feature(CPU_FTR_TM))
 534                 return;
 535
 536         /* Recheckpoint the registers of the thread we're about to switch to.
 537          *
 538          * If the task was using FP, we non-lazily reload both the original and
 539          * the speculative FP register states.  This is because the kernel
 540          * doesn't see if/when a TM rollback occurs, so if we take an FP
 541          * unavoidable later, we are unable to determine which set of FP regs
 542          * need to be restored.
 543          */
 544         if (!new->thread.regs)
 545                 return;
 546
 547         /* The TM SPRs are restored here, so that TEXASR.FS can be set
 548          * before the trecheckpoint and no explosion occurs.
 549          */
 550         tm_restore_sprs(&new->thread);
 551
 552         if (!MSR_TM_ACTIVE(new->thread.regs->msr))
 553                 return;
 554         msr = new->thread.tm_orig_msr;
 555         /* Recheckpoint to restore original checkpointed register state. */
 556         TM_DEBUG("*** tm_recheckpoint of pid %d "
 557                  "(new->msr 0x%lx, new->origmsr 0x%lx)\n",
 558                  new->pid, new->thread.regs->msr, msr);
 559
 560         /* This loads the checkpointed FP/VEC state, if used */
 561         tm_recheckpoint(&new->thread, msr);
 562
 563         /* This loads the speculative FP/VEC state, if used */
 564         if (msr & MSR_FP) {
 565                 do_load_up_transact_fpu(&new->thread);
 566                 new->thread.regs->msr |=
 567                         (MSR_FP | new->thread.fpexc_mode);
 568         }
 569 #ifdef CONFIG_ALTIVEC
 570         if (msr & MSR_VEC) {
 571                 do_load_up_transact_altivec(&new->thread);
 572                 new->thread.regs->msr |= MSR_VEC;
 573         }
 574 #endif
 575         /* We may as well turn on VSX too since all the state is restored now */
 576         if (msr & MSR_VSX)
 577                 new->thread.regs->msr |= MSR_VSX;
 578
 579         TM_DEBUG("*** tm_recheckpoint of pid %d complete "
 580                  "(kernel msr 0x%lx)\n",
 581                  new->pid, mfmsr());
 582 }
 583
 584 static inline void __switch_to_tm(struct task_struct *prev)
 585 {
 586         if (cpu_has_feature(CPU_FTR_TM)) {
 587                 tm_enable();
 588                 tm_reclaim_task(prev);
 589         }
 590 }
 591 #else
 592 #define tm_recheckpoint_new_task(new)
 593 #define __switch_to_tm(prev)
 594 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
 595
 596 struct task_struct *__switch_to(struct task_struct *prev,
 597         struct task_struct *new)
 598 {
 599         struct thread_struct *new_thread, *old_thread;
 600         unsigned long flags;
 601         struct task_struct *last;
 602 #ifdef CONFIG_PPC_BOOK3S_64
 603         struct ppc64_tlb_batch *batch;
 604 #endif
 605
 606         /* Back up the TAR across context switches.
 607          * Note that the TAR is not available for use in the kernel.  (To
 608          * provide this, the TAR should be backed up/restored on exception
 609          * entry/exit instead, and be in pt_regs.  FIXME, this should be in
 610          * pt_regs anyway (for debug).)
 611          * Save the TAR here before we do treclaim/trecheckpoint as these
 612          * will change the TAR.
 613          */
 614         save_tar(&prev->thread);
 615
 616         __switch_to_tm(prev);
 617
 618 #ifdef CONFIG_SMP
 619         /* avoid complexity of lazy save/restore of fpu
 620          * by just saving it every time we switch out if
 621          * this task used the fpu during the last quantum.
 622          *
 623          * If it tries to use the fpu again, it'll trap and
 624          * reload its fp regs.  So we don't have to do a restore
 625          * every switch, just a save.
 626          *  -- Cort
 627          */
 628         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 629                 giveup_fpu(prev);
 630 #ifdef CONFIG_ALTIVEC
 631         /*
 632          * If the previous thread used altivec in the last quantum
 633          * (thus changing altivec regs) then save them.
 634          * We used to check the VRSAVE register but not all apps
 635          * set it, so we don't rely on it now (and in fact we need
 636          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 637          *
 638          * On SMP we always save/restore altivec regs just to avoid the
 639          * complexity of changing processors.
 640          *  -- Cort
 641          */
 642         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 643                 giveup_altivec(prev);
 644 #endif /* CONFIG_ALTIVEC */
 645 #ifdef CONFIG_VSX
 646         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 647                 /* VMX and FPU registers are already save here */
 648                 __giveup_vsx(prev);
 649 #endif /* CONFIG_VSX */
 650 #ifdef CONFIG_SPE
 651         /*
 652          * If the previous thread used spe in the last quantum
 653          * (thus changing spe regs) then save them.
 654          *
 655          * On SMP we always save/restore spe regs just to avoid the
 656          * complexity of changing processors.
 657          */
 658         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 659                 giveup_spe(prev);
 660 #endif /* CONFIG_SPE */
 661
 662 #else  /* CONFIG_SMP */
 663 #ifdef CONFIG_ALTIVEC
 664         /* Avoid the trap.  On smp this this never happens since
 665          * we don't set last_task_used_altivec -- Cort
 666          */
 667         if (new->thread.regs && last_task_used_altivec == new)
 668                 new->thread.regs->msr |= MSR_VEC;
 669 #endif /* CONFIG_ALTIVEC */
 670 #ifdef CONFIG_VSX
 671         if (new->thread.regs && last_task_used_vsx == new)
 672                 new->thread.regs->msr |= MSR_VSX;
 673 #endif /* CONFIG_VSX */
 674 #ifdef CONFIG_SPE
 675         /* Avoid the trap.  On smp this this never happens since
 676          * we don't set last_task_used_spe
 677          */
 678         if (new->thread.regs && last_task_used_spe == new)
 679                 new->thread.regs->msr |= MSR_SPE;
 680 #endif /* CONFIG_SPE */
 681
 682 #endif /* CONFIG_SMP */
 683
 684 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 685         switch_booke_debug_regs(&new->thread);
 686 #else
 687 /*
 688  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
 689  * schedule DABR
 690  */
 691 #ifndef CONFIG_HAVE_HW_BREAKPOINT
 692         if (unlikely(hw_brk_match(&__get_cpu_var(current_brk), &new->thread.hw_brk)))
 693                 set_breakpoint(&new->thread.hw_brk);
 694 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 695 #endif
 696
 697
 698         new_thread = &new->thread;
 699         old_thread = &current->thread;
 700
 701 #ifdef CONFIG_PPC64
 702         /*
 703          * Collect processor utilization data per process
 704          */
 705         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 706                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 707                 long unsigned start_tb, current_tb;
 708                 start_tb = old_thread->start_tb;
 709                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 710                 old_thread->accum_tb += (current_tb - start_tb);
 711                 new_thread->start_tb = current_tb;
 712         }
 713 #endif /* CONFIG_PPC64 */
 714
 715 #ifdef CONFIG_PPC_BOOK3S_64
 716         batch = &__get_cpu_var(ppc64_tlb_batch);
 717         if (batch->active) {
 718                 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
 719                 if (batch->index)
 720                         __flush_tlb_pending(batch);
 721                 batch->active = 0;
 722         }
 723 #endif /* CONFIG_PPC_BOOK3S_64 */
 724
 725         local_irq_save(flags);
 726
 727         /*
 728          * We can't take a PMU exception inside _switch() since there is a
 729          * window where the kernel stack SLB and the kernel stack are out
 730          * of sync. Hard disable here.
 731          */
 732         hard_irq_disable();
 733
 734         tm_recheckpoint_new_task(new);
 735
 736         last = _switch(old_thread, new_thread);
 737
 738 #ifdef CONFIG_PPC_BOOK3S_64
 739         if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
 740                 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
 741                 batch = &__get_cpu_var(ppc64_tlb_batch);
 742                 batch->active = 1;
 743         }
 744 #endif /* CONFIG_PPC_BOOK3S_64 */
 745
 746         local_irq_restore(flags);
 747
 748         return last;
 749 }
 750
 751 static int instructions_to_print = 16;
 752
 753 static void show_instructions(struct pt_regs *regs)
 754 {
 755         int i;
 756         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 757                         sizeof(int));
 758
 759         printk("Instruction dump:");
 760
 761         for (i = 0; i < instructions_to_print; i++) {
 762                 int instr;
 763
 764                 if (!(i % 8))
 765                         printk("\n");
 766
 767 #if !defined(CONFIG_BOOKE)
 768                 /* If executing with the IMMU off, adjust pc rather
 769                  * than print XXXXXXXX.
 770                  */
 771                 if (!(regs->msr & MSR_IR))
 772                         pc = (unsigned long)phys_to_virt(pc);
 773 #endif
 774
 775                 /* We use __get_user here *only* to avoid an OOPS on a
 776                  * bad address because the pc *should* only be a
 777                  * kernel address.
 778                  */
 779                 if (!__kernel_text_address(pc) ||
 780                      __get_user(instr, (unsigned int __user *)pc)) {
 781                         printk(KERN_CONT "XXXXXXXX ");
 782                 } else {
 783                         if (regs->nip == pc)
 784                                 printk(KERN_CONT "<%08x> ", instr);
 785                         else
 786                                 printk(KERN_CONT "%08x ", instr);
 787                 }
 788
 789                 pc += sizeof(int);
 790         }
 791
 792         printk("\n");
 793 }
 794
 795 static struct regbit {
 796         unsigned long bit;
 797         const char *name;
 798 } msr_bits[] = {
 799 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
 800         {MSR_SF,        "SF"},
 801         {MSR_HV,        "HV"},
 802 #endif
 803         {MSR_VEC,       "VEC"},
 804         {MSR_VSX,       "VSX"},
 805 #ifdef CONFIG_BOOKE
 806         {MSR_CE,        "CE"},
 807 #endif
 808         {MSR_EE,        "EE"},
 809         {MSR_PR,        "PR"},
 810         {MSR_FP,        "FP"},
 811         {MSR_ME,        "ME"},
 812 #ifdef CONFIG_BOOKE
 813         {MSR_DE,        "DE"},
 814 #else
 815         {MSR_SE,        "SE"},
 816         {MSR_BE,        "BE"},
 817 #endif
 818         {MSR_IR,        "IR"},
 819         {MSR_DR,        "DR"},
 820         {MSR_PMM,       "PMM"},
 821 #ifndef CONFIG_BOOKE
 822         {MSR_RI,        "RI"},
 823         {MSR_LE,        "LE"},
 824 #endif
 825         {0,             NULL}
 826 };
 827
 828 static void printbits(unsigned long val, struct regbit *bits)
 829 {
 830         const char *sep = "";
 831
 832         printk("<");
 833         for (; bits->bit; ++bits)
 834                 if (val & bits->bit) {
 835                         printk("%s%s", sep, bits->name);
 836                         sep = ",";
 837                 }
 838         printk(">");
 839 }
 840
 841 #ifdef CONFIG_PPC64
 842 #define REG             "%016lx"
 843 #define REGS_PER_LINE   4
 844 #define LAST_VOLATILE   13
 845 #else
 846 #define REG             "%08lx"
 847 #define REGS_PER_LINE   8
 848 #define LAST_VOLATILE   12
 849 #endif
 850
 851 void show_regs(struct pt_regs * regs)
 852 {
 853         int i, trap;
 854
 855         show_regs_print_info(KERN_DEFAULT);
 856
 857         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 858                regs->nip, regs->link, regs->ctr);
 859         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 860                regs, regs->trap, print_tainted(), init_utsname()->release);
 861         printk("MSR: "REG" ", regs->msr);
 862         printbits(regs->msr, msr_bits);
 863         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 864 #ifdef CONFIG_PPC64
 865         printk("SOFTE: %ld\n", regs->softe);
 866 #endif
 867         trap = TRAP(regs);
 868         if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
 869                 printk("CFAR: "REG"\n", regs->orig_gpr3);
 870         if (trap == 0x300 || trap == 0x600)
 871 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 872                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 873 #else
 874                 printk("DAR: "REG", DSISR: %08lx\n", regs->dar, regs->dsisr);
 875 #endif
 876
 877         for (i = 0;  i < 32;  i++) {
 878                 if ((i % REGS_PER_LINE) == 0)
 879                         printk("\nGPR%02d: ", i);
 880                 printk(REG " ", regs->gpr[i]);
 881                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 882                         break;
 883         }
 884         printk("\n");
 885 #ifdef CONFIG_KALLSYMS
 886         /*
 887          * Lookup NIP late so we have the best change of getting the
 888          * above info out without failing
 889          */
 890         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 891         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 892 #endif
 893 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 894         printk("PACATMSCRATCH [%llx]\n", get_paca()->tm_scratch);
 895 #endif
 896         show_stack(current, (unsigned long *) regs->gpr[1]);
 897         if (!user_mode(regs))
 898                 show_instructions(regs);
 899 }
 900
 901 void exit_thread(void)
 902 {
 903         discard_lazy_cpu_state();
 904 }
 905
 906 void flush_thread(void)
 907 {
 908         discard_lazy_cpu_state();
 909
 910 #ifdef CONFIG_HAVE_HW_BREAKPOINT
 911         flush_ptrace_hw_breakpoint(current);
 912 #else /* CONFIG_HAVE_HW_BREAKPOINT */
 913         set_debug_reg_defaults(&current->thread);
 914 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 915 }
 916
 917 void
 918 release_thread(struct task_struct *t)
 919 {
 920 }
 921
 922 /*
 923  * this gets called so that we can store coprocessor state into memory and
 924  * copy the current task into the new thread.
 925  */
 926 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 927 {
 928         flush_fp_to_thread(src);
 929         flush_altivec_to_thread(src);
 930         flush_vsx_to_thread(src);
 931         flush_spe_to_thread(src);
 932
 933         *dst = *src;
 934
 935         clear_task_ebb(dst);
 936
 937         return 0;
 938 }
 939
 940 /*
 941  * Copy a thread..
 942  */
 943 extern unsigned long dscr_default; /* defined in arch/powerpc/kernel/sysfs.c */
 944
 945 int copy_thread(unsigned long clone_flags, unsigned long usp,
 946                 unsigned long arg, struct task_struct *p)
 947 {
 948         struct pt_regs *childregs, *kregs;
 949         extern void ret_from_fork(void);
 950         extern void ret_from_kernel_thread(void);
 951         void (*f)(void);
 952         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 953
 954         /* Copy registers */
 955         sp -= sizeof(struct pt_regs);
 956         childregs = (struct pt_regs *) sp;
 957         if (unlikely(p->flags & PF_KTHREAD)) {
 958                 struct thread_info *ti = (void *)task_stack_page(p);
 959                 memset(childregs, 0, sizeof(struct pt_regs));
 960                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 961                 childregs->gpr[14] = usp;       /* function */
 962 #ifdef CONFIG_PPC64
 963                 clear_tsk_thread_flag(p, TIF_32BIT);
 964                 childregs->softe = 1;
 965 #endif
 966                 childregs->gpr[15] = arg;
 967                 p->thread.regs = NULL;  /* no user register state */
 968                 ti->flags |= _TIF_RESTOREALL;
 969                 f = ret_from_kernel_thread;
 970         } else {
 971                 struct pt_regs *regs = current_pt_regs();
 972                 CHECK_FULL_REGS(regs);
 973                 *childregs = *regs;
 974                 if (usp)
 975                         childregs->gpr[1] = usp;
 976                 p->thread.regs = childregs;
 977                 childregs->gpr[3] = 0;  /* Result from fork() */
 978                 if (clone_flags & CLONE_SETTLS) {
 979 #ifdef CONFIG_PPC64
 980                         if (!is_32bit_task())
 981                                 childregs->gpr[13] = childregs->gpr[6];
 982                         else
 983 #endif
 984                                 childregs->gpr[2] = childregs->gpr[6];
 985                 }
 986
 987                 f = ret_from_fork;
 988         }
 989         sp -= STACK_FRAME_OVERHEAD;
 990
 991         /*
 992          * The way this works is that at some point in the future
 993          * some task will call _switch to switch to the new task.
 994          * That will pop off the stack frame created below and start
 995          * the new task running at ret_from_fork.  The new task will
 996          * do some house keeping and then return from the fork or clone
 997          * system call, using the stack frame created above.
 998          */
 999         ((unsigned long *)sp)[0] = 0;
1000         sp -= sizeof(struct pt_regs);
1001         kregs = (struct pt_regs *) sp;
1002         sp -= STACK_FRAME_OVERHEAD;
1003         p->thread.ksp = sp;
1004 #ifdef CONFIG_PPC32
1005         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1006                                 _ALIGN_UP(sizeof(struct thread_info), 16);
1007 #endif
1008 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1009         p->thread.ptrace_bps[0] = NULL;
1010 #endif
1011
1012         p->thread.fp_save_area = NULL;
1013 #ifdef CONFIG_ALTIVEC
1014         p->thread.vr_save_area = NULL;
1015 #endif
1016
1017 #ifdef CONFIG_PPC_STD_MMU_64
1018         if (mmu_has_feature(MMU_FTR_SLB)) {
1019                 unsigned long sp_vsid;
1020                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1021
1022                 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1023                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1024                                 << SLB_VSID_SHIFT_1T;
1025                 else
1026                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1027                                 << SLB_VSID_SHIFT;
1028                 sp_vsid |= SLB_VSID_KERNEL | llp;
1029                 p->thread.ksp_vsid = sp_vsid;
1030         }
1031 #endif /* CONFIG_PPC_STD_MMU_64 */
1032 #ifdef CONFIG_PPC64
1033         if (cpu_has_feature(CPU_FTR_DSCR)) {
1034                 p->thread.dscr_inherit = current->thread.dscr_inherit;
1035                 p->thread.dscr = current->thread.dscr;
1036         }
1037         if (cpu_has_feature(CPU_FTR_HAS_PPR))
1038                 p->thread.ppr = INIT_PPR;
1039 #endif
1040         /*
1041          * The PPC64 ABI makes use of a TOC to contain function
1042          * pointers.  The function (ret_from_except) is actually a pointer
1043          * to the TOC entry.  The first entry is a pointer to the actual
1044          * function.
1045          */
1046 #ifdef CONFIG_PPC64
1047         kregs->nip = *((unsigned long *)f);
1048 #else
1049         kregs->nip = (unsigned long)f;
1050 #endif
1051         return 0;
1052 }
1053
1054 /*
1055  * Set up a thread for executing a new program
1056  */
1057 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1058 {
1059 #ifdef CONFIG_PPC64
1060         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1061 #endif
1062
1063         /*
1064          * If we exec out of a kernel thread then thread.regs will not be
1065          * set.  Do it now.
1066          */
1067         if (!current->thread.regs) {
1068                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1069                 current->thread.regs = regs - 1;
1070         }
1071
1072         memset(regs->gpr, 0, sizeof(regs->gpr));
1073         regs->ctr = 0;
1074         regs->link = 0;
1075         regs->xer = 0;
1076         regs->ccr = 0;
1077         regs->gpr[1] = sp;
1078
1079         /*
1080          * We have just cleared all the nonvolatile GPRs, so make
1081          * FULL_REGS(regs) return true.  This is necessary to allow
1082          * ptrace to examine the thread immediately after exec.
1083          */
1084         regs->trap &= ~1UL;
1085
1086 #ifdef CONFIG_PPC32
1087         regs->mq = 0;
1088         regs->nip = start;
1089         regs->msr = MSR_USER;
1090 #else
1091         if (!is_32bit_task()) {
1092                 unsigned long entry, toc;
1093
1094                 /* start is a relocated pointer to the function descriptor for
1095                  * the elf _start routine.  The first entry in the function
1096                  * descriptor is the entry address of _start and the second
1097                  * entry is the TOC value we need to use.
1098                  */
1099                 __get_user(entry, (unsigned long __user *)start);
1100                 __get_user(toc, (unsigned long __user *)start+1);
1101
1102                 /* Check whether the e_entry function descriptor entries
1103                  * need to be relocated before we can use them.
1104                  */
1105                 if (load_addr != 0) {
1106                         entry += load_addr;
1107                         toc   += load_addr;
1108                 }
1109                 regs->nip = entry;
1110                 regs->gpr[2] = toc;
1111                 regs->msr = MSR_USER64;
1112         } else {
1113                 regs->nip = start;
1114                 regs->gpr[2] = 0;
1115                 regs->msr = MSR_USER32;
1116         }
1117 #endif
1118         discard_lazy_cpu_state();
1119 #ifdef CONFIG_VSX
1120         current->thread.used_vsr = 0;
1121 #endif
1122         memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1123         current->thread.fp_save_area = NULL;
1124 #ifdef CONFIG_ALTIVEC
1125         memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1126         current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1127         current->thread.vr_save_area = NULL;
1128         current->thread.vrsave = 0;
1129         current->thread.used_vr = 0;
1130 #endif /* CONFIG_ALTIVEC */
1131 #ifdef CONFIG_SPE
1132         memset(current->thread.evr, 0, sizeof(current->thread.evr));
1133         current->thread.acc = 0;
1134         current->thread.spefscr = 0;
1135         current->thread.used_spe = 0;
1136 #endif /* CONFIG_SPE */
1137 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1138         if (cpu_has_feature(CPU_FTR_TM))
1139                 regs->msr |= MSR_TM;
1140         current->thread.tm_tfhar = 0;
1141         current->thread.tm_texasr = 0;
1142         current->thread.tm_tfiar = 0;
1143 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1144 }
1145
1146 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1147                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
1148
1149 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1150 {
1151         struct pt_regs *regs = tsk->thread.regs;
1152
1153         /* This is a bit hairy.  If we are an SPE enabled  processor
1154          * (have embedded fp) we store the IEEE exception enable flags in
1155          * fpexc_mode.  fpexc_mode is also used for setting FP exception
1156          * mode (asyn, precise, disabled) for 'Classic' FP. */
1157         if (val & PR_FP_EXC_SW_ENABLE) {
1158 #ifdef CONFIG_SPE
1159                 if (cpu_has_feature(CPU_FTR_SPE)) {
1160                         tsk->thread.fpexc_mode = val &
1161                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1162                         return 0;
1163                 } else {
1164                         return -EINVAL;
1165                 }
1166 #else
1167                 return -EINVAL;
1168 #endif
1169         }
1170
1171         /* on a CONFIG_SPE this does not hurt us.  The bits that
1172          * __pack_fe01 use do not overlap with bits used for
1173          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1174          * on CONFIG_SPE implementations are reserved so writing to
1175          * them does not change anything */
1176         if (val > PR_FP_EXC_PRECISE)
1177                 return -EINVAL;
1178         tsk->thread.fpexc_mode = __pack_fe01(val);
1179         if (regs != NULL && (regs->msr & MSR_FP) != 0)
1180                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1181                         | tsk->thread.fpexc_mode;
1182         return 0;
1183 }
1184
1185 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1186 {
1187         unsigned int val;
1188
1189         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1190 #ifdef CONFIG_SPE
1191                 if (cpu_has_feature(CPU_FTR_SPE))
1192                         val = tsk->thread.fpexc_mode;
1193                 else
1194                         return -EINVAL;
1195 #else
1196                 return -EINVAL;
1197 #endif
1198         else
1199                 val = __unpack_fe01(tsk->thread.fpexc_mode);
1200         return put_user(val, (unsigned int __user *) adr);
1201 }
1202
1203 int set_endian(struct task_struct *tsk, unsigned int val)
1204 {
1205         struct pt_regs *regs = tsk->thread.regs;
1206
1207         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1208             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1209                 return -EINVAL;
1210
1211         if (regs == NULL)
1212                 return -EINVAL;
1213
1214         if (val == PR_ENDIAN_BIG)
1215                 regs->msr &= ~MSR_LE;
1216         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1217                 regs->msr |= MSR_LE;
1218         else
1219                 return -EINVAL;
1220
1221         return 0;
1222 }
1223
1224 int get_endian(struct task_struct *tsk, unsigned long adr)
1225 {
1226         struct pt_regs *regs = tsk->thread.regs;
1227         unsigned int val;
1228
1229         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1230             !cpu_has_feature(CPU_FTR_REAL_LE))
1231                 return -EINVAL;
1232
1233         if (regs == NULL)
1234                 return -EINVAL;
1235
1236         if (regs->msr & MSR_LE) {
1237                 if (cpu_has_feature(CPU_FTR_REAL_LE))
1238                         val = PR_ENDIAN_LITTLE;
1239                 else
1240                         val = PR_ENDIAN_PPC_LITTLE;
1241         } else
1242                 val = PR_ENDIAN_BIG;
1243
1244         return put_user(val, (unsigned int __user *)adr);
1245 }
1246
1247 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1248 {
1249         tsk->thread.align_ctl = val;
1250         return 0;
1251 }
1252
1253 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1254 {
1255         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1256 }
1257
1258 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1259                                   unsigned long nbytes)
1260 {
1261         unsigned long stack_page;
1262         unsigned long cpu = task_cpu(p);
1263
1264         /*
1265          * Avoid crashing if the stack has overflowed and corrupted
1266          * task_cpu(p), which is in the thread_info struct.
1267          */
1268         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1269                 stack_page = (unsigned long) hardirq_ctx[cpu];
1270                 if (sp >= stack_page + sizeof(struct thread_struct)
1271                     && sp <= stack_page + THREAD_SIZE - nbytes)
1272                         return 1;
1273
1274                 stack_page = (unsigned long) softirq_ctx[cpu];
1275                 if (sp >= stack_page + sizeof(struct thread_struct)
1276                     && sp <= stack_page + THREAD_SIZE - nbytes)
1277                         return 1;
1278         }
1279         return 0;
1280 }
1281
1282 int validate_sp(unsigned long sp, struct task_struct *p,
1283                        unsigned long nbytes)
1284 {
1285         unsigned long stack_page = (unsigned long)task_stack_page(p);
1286
1287         if (sp >= stack_page + sizeof(struct thread_struct)
1288             && sp <= stack_page + THREAD_SIZE - nbytes)
1289                 return 1;
1290
1291         return valid_irq_stack(sp, p, nbytes);
1292 }
1293
1294 EXPORT_SYMBOL(validate_sp);
1295
1296 unsigned long get_wchan(struct task_struct *p)
1297 {
1298         unsigned long ip, sp;
1299         int count = 0;
1300
1301         if (!p || p == current || p->state == TASK_RUNNING)
1302                 return 0;
1303
1304         sp = p->thread.ksp;
1305         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1306                 return 0;
1307
1308         do {
1309                 sp = *(unsigned long *)sp;
1310                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1311                         return 0;
1312                 if (count > 0) {
1313                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1314                         if (!in_sched_functions(ip))
1315                                 return ip;
1316                 }
1317         } while (count++ < 16);
1318         return 0;
1319 }
1320
1321 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1322
1323 void show_stack(struct task_struct *tsk, unsigned long *stack)
1324 {
1325         unsigned long sp, ip, lr, newsp;
1326         int count = 0;
1327         int firstframe = 1;
1328 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1329         int curr_frame = current->curr_ret_stack;
1330         extern void return_to_handler(void);
1331         unsigned long rth = (unsigned long)return_to_handler;
1332         unsigned long mrth = -1;
1333 #ifdef CONFIG_PPC64
1334         extern void mod_return_to_handler(void);
1335         rth = *(unsigned long *)rth;
1336         mrth = (unsigned long)mod_return_to_handler;
1337         mrth = *(unsigned long *)mrth;
1338 #endif
1339 #endif
1340
1341         sp = (unsigned long) stack;
1342         if (tsk == NULL)
1343                 tsk = current;
1344         if (sp == 0) {
1345                 if (tsk == current)
1346                         asm("mr %0,1" : "=r" (sp));
1347                 else
1348                         sp = tsk->thread.ksp;
1349         }
1350
1351         lr = 0;
1352         printk("Call Trace:\n");
1353         do {
1354                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1355                         return;
1356
1357                 stack = (unsigned long *) sp;
1358                 newsp = stack[0];
1359                 ip = stack[STACK_FRAME_LR_SAVE];
1360                 if (!firstframe || ip != lr) {
1361                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1362 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1363                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1364                                 printk(" (%pS)",
1365                                        (void *)current->ret_stack[curr_frame].ret);
1366                                 curr_frame--;
1367                         }
1368 #endif
1369                         if (firstframe)
1370                                 printk(" (unreliable)");
1371                         printk("\n");
1372                 }
1373                 firstframe = 0;
1374
1375                 /*
1376                  * See if this is an exception frame.
1377                  * We look for the "regshere" marker in the current frame.
1378                  */
1379                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1380                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1381                         struct pt_regs *regs = (struct pt_regs *)
1382                                 (sp + STACK_FRAME_OVERHEAD);
1383                         lr = regs->link;
1384                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1385                                regs->trap, (void *)regs->nip, (void *)lr);
1386                         firstframe = 1;
1387                 }
1388
1389                 sp = newsp;
1390         } while (count++ < kstack_depth_to_print);
1391 }
1392
1393 #ifdef CONFIG_PPC64
1394 /* Called with hard IRQs off */
1395 void notrace __ppc64_runlatch_on(void)
1396 {
1397         struct thread_info *ti = current_thread_info();
1398         unsigned long ctrl;
1399
1400         ctrl = mfspr(SPRN_CTRLF);
1401         ctrl |= CTRL_RUNLATCH;
1402         mtspr(SPRN_CTRLT, ctrl);
1403
1404         ti->local_flags |= _TLF_RUNLATCH;
1405 }
1406
1407 /* Called with hard IRQs off */
1408 void notrace __ppc64_runlatch_off(void)
1409 {
1410         struct thread_info *ti = current_thread_info();
1411         unsigned long ctrl;
1412
1413         ti->local_flags &= ~_TLF_RUNLATCH;
1414
1415         ctrl = mfspr(SPRN_CTRLF);
1416         ctrl &= ~CTRL_RUNLATCH;
1417         mtspr(SPRN_CTRLT, ctrl);
1418 }
1419 #endif /* CONFIG_PPC64 */
1420
1421 unsigned long arch_align_stack(unsigned long sp)
1422 {
1423         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1424                 sp -= get_random_int() & ~PAGE_MASK;
1425         return sp & ~0xf;
1426 }
1427
1428 static inline unsigned long brk_rnd(void)
1429 {
1430         unsigned long rnd = 0;
1431
1432         /* 8MB for 32bit, 1GB for 64bit */
1433         if (is_32bit_task())
1434                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1435         else
1436                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1437
1438         return rnd << PAGE_SHIFT;
1439 }
1440
1441 unsigned long arch_randomize_brk(struct mm_struct *mm)
1442 {
1443         unsigned long base = mm->brk;
1444         unsigned long ret;
1445
1446 #ifdef CONFIG_PPC_STD_MMU_64
1447         /*
1448          * If we are using 1TB segments and we are allowed to randomise
1449          * the heap, we can put it above 1TB so it is backed by a 1TB
1450          * segment. Otherwise the heap will be in the bottom 1TB
1451          * which always uses 256MB segments and this may result in a
1452          * performance penalty.
1453          */
1454         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1455                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1456 #endif
1457
1458         ret = PAGE_ALIGN(base + brk_rnd());
1459
1460         if (ret < mm->brk)
1461                 return mm->brk;
1462
1463         return ret;
1464 }
1465
1466 unsigned long randomize_et_dyn(unsigned long base)
1467 {
1468         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1469
1470         if (ret < base)
1471                 return base;
1472
1473         return ret;
1474 }