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 /*
  78  * Make sure the floating-point register state in the
  79  * the thread_struct is up to date for task tsk.
  80  */
  81 void flush_fp_to_thread(struct task_struct *tsk)
  82 {
  83         if (tsk->thread.regs) {
  84                 /*
  85                  * We need to disable preemption here because if we didn't,
  86                  * another process could get scheduled after the regs->msr
  87                  * test but before we have finished saving the FP registers
  88                  * to the thread_struct.  That process could take over the
  89                  * FPU, and then when we get scheduled again we would store
  90                  * bogus values for the remaining FP registers.
  91                  */
  92                 preempt_disable();
  93                 if (tsk->thread.regs->msr & MSR_FP) {
  94 #ifdef CONFIG_SMP
  95                         /*
  96                          * This should only ever be called for current or
  97                          * for a stopped child process.  Since we save away
  98                          * the FP register state on context switch on SMP,
  99                          * there is something wrong if a stopped child appears
 100                          * to still have its FP state in the CPU registers.
 101                          */
 102                         BUG_ON(tsk != current);
 103 #endif
 104                         giveup_fpu(tsk);
 105                 }
 106                 preempt_enable();
 107         }
 108 }
 109 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
 110
 111 void enable_kernel_fp(void)
 112 {
 113         WARN_ON(preemptible());
 114
 115 #ifdef CONFIG_SMP
 116         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
 117                 giveup_fpu(current);
 118         else
 119                 giveup_fpu(NULL);       /* just enables FP for kernel */
 120 #else
 121         giveup_fpu(last_task_used_math);
 122 #endif /* CONFIG_SMP */
 123 }
 124 EXPORT_SYMBOL(enable_kernel_fp);
 125
 126 #ifdef CONFIG_ALTIVEC
 127 void enable_kernel_altivec(void)
 128 {
 129         WARN_ON(preemptible());
 130
 131 #ifdef CONFIG_SMP
 132         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 133                 giveup_altivec(current);
 134         else
 135                 giveup_altivec_notask();
 136 #else
 137         giveup_altivec(last_task_used_altivec);
 138 #endif /* CONFIG_SMP */
 139 }
 140 EXPORT_SYMBOL(enable_kernel_altivec);
 141
 142 /*
 143  * Make sure the VMX/Altivec register state in the
 144  * the thread_struct is up to date for task tsk.
 145  */
 146 void flush_altivec_to_thread(struct task_struct *tsk)
 147 {
 148         if (tsk->thread.regs) {
 149                 preempt_disable();
 150                 if (tsk->thread.regs->msr & MSR_VEC) {
 151 #ifdef CONFIG_SMP
 152                         BUG_ON(tsk != current);
 153 #endif
 154                         giveup_altivec(tsk);
 155                 }
 156                 preempt_enable();
 157         }
 158 }
 159 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
 160 #endif /* CONFIG_ALTIVEC */
 161
 162 #ifdef CONFIG_VSX
 163 #if 0
 164 /* not currently used, but some crazy RAID module might want to later */
 165 void enable_kernel_vsx(void)
 166 {
 167         WARN_ON(preemptible());
 168
 169 #ifdef CONFIG_SMP
 170         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 171                 giveup_vsx(current);
 172         else
 173                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 174 #else
 175         giveup_vsx(last_task_used_vsx);
 176 #endif /* CONFIG_SMP */
 177 }
 178 EXPORT_SYMBOL(enable_kernel_vsx);
 179 #endif
 180
 181 void giveup_vsx(struct task_struct *tsk)
 182 {
 183         giveup_fpu(tsk);
 184         giveup_altivec(tsk);
 185         __giveup_vsx(tsk);
 186 }
 187
 188 void flush_vsx_to_thread(struct task_struct *tsk)
 189 {
 190         if (tsk->thread.regs) {
 191                 preempt_disable();
 192                 if (tsk->thread.regs->msr & MSR_VSX) {
 193 #ifdef CONFIG_SMP
 194                         BUG_ON(tsk != current);
 195 #endif
 196                         giveup_vsx(tsk);
 197                 }
 198                 preempt_enable();
 199         }
 200 }
 201 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
 202 #endif /* CONFIG_VSX */
 203
 204 #ifdef CONFIG_SPE
 205
 206 void enable_kernel_spe(void)
 207 {
 208         WARN_ON(preemptible());
 209
 210 #ifdef CONFIG_SMP
 211         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 212                 giveup_spe(current);
 213         else
 214                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 215 #else
 216         giveup_spe(last_task_used_spe);
 217 #endif /* __SMP __ */
 218 }
 219 EXPORT_SYMBOL(enable_kernel_spe);
 220
 221 void flush_spe_to_thread(struct task_struct *tsk)
 222 {
 223         if (tsk->thread.regs) {
 224                 preempt_disable();
 225                 if (tsk->thread.regs->msr & MSR_SPE) {
 226 #ifdef CONFIG_SMP
 227                         BUG_ON(tsk != current);
 228 #endif
 229                         tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
 230                         giveup_spe(tsk);
 231                 }
 232                 preempt_enable();
 233         }
 234 }
 235 #endif /* CONFIG_SPE */
 236
 237 #ifndef CONFIG_SMP
 238 /*
 239  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 240  * and the current task has some state, discard it.
 241  */
 242 void discard_lazy_cpu_state(void)
 243 {
 244         preempt_disable();
 245         if (last_task_used_math == current)
 246                 last_task_used_math = NULL;
 247 #ifdef CONFIG_ALTIVEC
 248         if (last_task_used_altivec == current)
 249                 last_task_used_altivec = NULL;
 250 #endif /* CONFIG_ALTIVEC */
 251 #ifdef CONFIG_VSX
 252         if (last_task_used_vsx == current)
 253                 last_task_used_vsx = NULL;
 254 #endif /* CONFIG_VSX */
 255 #ifdef CONFIG_SPE
 256         if (last_task_used_spe == current)
 257                 last_task_used_spe = NULL;
 258 #endif
 259         preempt_enable();
 260 }
 261 #endif /* CONFIG_SMP */
 262
 263 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 264 void do_send_trap(struct pt_regs *regs, unsigned long address,
 265                   unsigned long error_code, int signal_code, int breakpt)
 266 {
 267         siginfo_t info;
 268
 269         current->thread.trap_nr = signal_code;
 270         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 271                         11, SIGSEGV) == NOTIFY_STOP)
 272                 return;
 273
 274         /* Deliver the signal to userspace */
 275         info.si_signo = SIGTRAP;
 276         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
 277         info.si_code = signal_code;
 278         info.si_addr = (void __user *)address;
 279         force_sig_info(SIGTRAP, &info, current);
 280 }
 281 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 282 void do_break (struct pt_regs *regs, unsigned long address,
 283                     unsigned long error_code)
 284 {
 285         siginfo_t info;
 286
 287         current->thread.trap_nr = TRAP_HWBKPT;
 288         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 289                         11, SIGSEGV) == NOTIFY_STOP)
 290                 return;
 291
 292         if (debugger_break_match(regs))
 293                 return;
 294
 295         /* Clear the breakpoint */
 296         hw_breakpoint_disable();
 297
 298         /* Deliver the signal to userspace */
 299         info.si_signo = SIGTRAP;
 300         info.si_errno = 0;
 301         info.si_code = TRAP_HWBKPT;
 302         info.si_addr = (void __user *)address;
 303         force_sig_info(SIGTRAP, &info, current);
 304 }
 305 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 306
 307 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
 308
 309 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 310 /*
 311  * Set the debug registers back to their default "safe" values.
 312  */
 313 static void set_debug_reg_defaults(struct thread_struct *thread)
 314 {
 315         thread->iac1 = thread->iac2 = 0;
 316 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 317         thread->iac3 = thread->iac4 = 0;
 318 #endif
 319         thread->dac1 = thread->dac2 = 0;
 320 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 321         thread->dvc1 = thread->dvc2 = 0;
 322 #endif
 323         thread->dbcr0 = 0;
 324 #ifdef CONFIG_BOOKE
 325         /*
 326          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
 327          */
 328         thread->dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |   \
 329                         DBCR1_IAC3US | DBCR1_IAC4US;
 330         /*
 331          * Force Data Address Compare User/Supervisor bits to be User-only
 332          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
 333          */
 334         thread->dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
 335 #else
 336         thread->dbcr1 = 0;
 337 #endif
 338 }
 339
 340 static void prime_debug_regs(struct thread_struct *thread)
 341 {
 342         /*
 343          * We could have inherited MSR_DE from userspace, since
 344          * it doesn't get cleared on exception entry.  Make sure
 345          * MSR_DE is clear before we enable any debug events.
 346          */
 347         mtmsr(mfmsr() & ~MSR_DE);
 348
 349         mtspr(SPRN_IAC1, thread->iac1);
 350         mtspr(SPRN_IAC2, thread->iac2);
 351 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 352         mtspr(SPRN_IAC3, thread->iac3);
 353         mtspr(SPRN_IAC4, thread->iac4);
 354 #endif
 355         mtspr(SPRN_DAC1, thread->dac1);
 356         mtspr(SPRN_DAC2, thread->dac2);
 357 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 358         mtspr(SPRN_DVC1, thread->dvc1);
 359         mtspr(SPRN_DVC2, thread->dvc2);
 360 #endif
 361         mtspr(SPRN_DBCR0, thread->dbcr0);
 362         mtspr(SPRN_DBCR1, thread->dbcr1);
 363 #ifdef CONFIG_BOOKE
 364         mtspr(SPRN_DBCR2, thread->dbcr2);
 365 #endif
 366 }
 367 /*
 368  * Unless neither the old or new thread are making use of the
 369  * debug registers, set the debug registers from the values
 370  * stored in the new thread.
 371  */
 372 static void switch_booke_debug_regs(struct thread_struct *new_thread)
 373 {
 374         if ((current->thread.dbcr0 & DBCR0_IDM)
 375                 || (new_thread->dbcr0 & DBCR0_IDM))
 376                         prime_debug_regs(new_thread);
 377 }
 378 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 379 #ifndef CONFIG_HAVE_HW_BREAKPOINT
 380 static void set_debug_reg_defaults(struct thread_struct *thread)
 381 {
 382         thread->hw_brk.address = 0;
 383         thread->hw_brk.type = 0;
 384         set_breakpoint(&thread->hw_brk);
 385 }
 386 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
 387 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 388
 389 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 390 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 391 {
 392         mtspr(SPRN_DAC1, dabr);
 393 #ifdef CONFIG_PPC_47x
 394         isync();
 395 #endif
 396         return 0;
 397 }
 398 #elif defined(CONFIG_PPC_BOOK3S)
 399 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 400 {
 401         mtspr(SPRN_DABR, dabr);
 402         if (cpu_has_feature(CPU_FTR_DABRX))
 403                 mtspr(SPRN_DABRX, dabrx);
 404         return 0;
 405 }
 406 #else
 407 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 408 {
 409         return -EINVAL;
 410 }
 411 #endif
 412
 413 static inline int set_dabr(struct arch_hw_breakpoint *brk)
 414 {
 415         unsigned long dabr, dabrx;
 416
 417         dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
 418         dabrx = ((brk->type >> 3) & 0x7);
 419
 420         if (ppc_md.set_dabr)
 421                 return ppc_md.set_dabr(dabr, dabrx);
 422
 423         return __set_dabr(dabr, dabrx);
 424 }
 425
 426 static inline int set_dawr(struct arch_hw_breakpoint *brk)
 427 {
 428         unsigned long dawr, dawrx, mrd;
 429
 430         dawr = brk->address;
 431
 432         dawrx  = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
 433                                    << (63 - 58); //* read/write bits */
 434         dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
 435                                    << (63 - 59); //* translate */
 436         dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
 437                                    >> 3; //* PRIM bits */
 438         /* dawr length is stored in field MDR bits 48:53.  Matches range in
 439            doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
 440            0b111111=64DW.
 441            brk->len is in bytes.
 442            This aligns up to double word size, shifts and does the bias.
 443         */
 444         mrd = ((brk->len + 7) >> 3) - 1;
 445         dawrx |= (mrd & 0x3f) << (63 - 53);
 446
 447         if (ppc_md.set_dawr)
 448                 return ppc_md.set_dawr(dawr, dawrx);
 449         mtspr(SPRN_DAWR, dawr);
 450         mtspr(SPRN_DAWRX, dawrx);
 451         return 0;
 452 }
 453
 454 int set_breakpoint(struct arch_hw_breakpoint *brk)
 455 {
 456         __get_cpu_var(current_brk) = *brk;
 457
 458         if (cpu_has_feature(CPU_FTR_DAWR))
 459                 return set_dawr(brk);
 460
 461         return set_dabr(brk);
 462 }
 463
 464 #ifdef CONFIG_PPC64
 465 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 466 #endif
 467
 468 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
 469                               struct arch_hw_breakpoint *b)
 470 {
 471         if (a->address != b->address)
 472                 return false;
 473         if (a->type != b->type)
 474                 return false;
 475         if (a->len != b->len)
 476                 return false;
 477         return true;
 478 }
 479 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 480 static inline void tm_reclaim_task(struct task_struct *tsk)
 481 {
 482         /* We have to work out if we're switching from/to a task that's in the
 483          * middle of a transaction.
 484          *
 485          * In switching we need to maintain a 2nd register state as
 486          * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
 487          * checkpointed (tbegin) state in ckpt_regs and saves the transactional
 488          * (current) FPRs into oldtask->thread.transact_fpr[].
 489          *
 490          * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
 491          */
 492         struct thread_struct *thr = &tsk->thread;
 493
 494         if (!thr->regs)
 495                 return;
 496
 497         if (!MSR_TM_ACTIVE(thr->regs->msr))
 498                 goto out_and_saveregs;
 499
 500         /* Stash the original thread MSR, as giveup_fpu et al will
 501          * modify it.  We hold onto it to see whether the task used
 502          * FP & vector regs.
 503          */
 504         thr->tm_orig_msr = thr->regs->msr;
 505
 506         TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
 507                  "ccr=%lx, msr=%lx, trap=%lx)\n",
 508                  tsk->pid, thr->regs->nip,
 509                  thr->regs->ccr, thr->regs->msr,
 510                  thr->regs->trap);
 511
 512         tm_reclaim(thr, thr->regs->msr, TM_CAUSE_RESCHED);
 513
 514         TM_DEBUG("--- tm_reclaim on pid %d complete\n",
 515                  tsk->pid);
 516
 517 out_and_saveregs:
 518         /* Always save the regs here, even if a transaction's not active.
 519          * This context-switches a thread's TM info SPRs.  We do it here to
 520          * be consistent with the restore path (in recheckpoint) which
 521          * cannot happen later in _switch().
 522          */
 523         tm_save_sprs(thr);
 524 }
 525
 526 extern void __tm_recheckpoint(struct thread_struct *thread,
 527                               unsigned long orig_msr);
 528
 529 void tm_recheckpoint(struct thread_struct *thread,
 530                      unsigned long orig_msr)
 531 {
 532         unsigned long flags;
 533
 534         /* We really can't be interrupted here as the TEXASR registers can't
 535          * change and later in the trecheckpoint code, we have a userspace R1.
 536          * So let's hard disable over this region.
 537          */
 538         local_irq_save(flags);
 539         hard_irq_disable();
 540
 541         /* The TM SPRs are restored here, so that TEXASR.FS can be set
 542          * before the trecheckpoint and no explosion occurs.
 543          */
 544         tm_restore_sprs(thread);
 545
 546         __tm_recheckpoint(thread, orig_msr);
 547
 548         local_irq_restore(flags);
 549 }
 550
 551 static inline void tm_recheckpoint_new_task(struct task_struct *new)
 552 {
 553         unsigned long msr;
 554
 555         if (!cpu_has_feature(CPU_FTR_TM))
 556                 return;
 557
 558         /* Recheckpoint the registers of the thread we're about to switch to.
 559          *
 560          * If the task was using FP, we non-lazily reload both the original and
 561          * the speculative FP register states.  This is because the kernel
 562          * doesn't see if/when a TM rollback occurs, so if we take an FP
 563          * unavoidable later, we are unable to determine which set of FP regs
 564          * need to be restored.
 565          */
 566         if (!new->thread.regs)
 567                 return;
 568
 569         if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
 570                 tm_restore_sprs(&new->thread);
 571                 return;
 572         }
 573         msr = new->thread.tm_orig_msr;
 574         /* Recheckpoint to restore original checkpointed register state. */
 575         TM_DEBUG("*** tm_recheckpoint of pid %d "
 576                  "(new->msr 0x%lx, new->origmsr 0x%lx)\n",
 577                  new->pid, new->thread.regs->msr, msr);
 578
 579         /* This loads the checkpointed FP/VEC state, if used */
 580         tm_recheckpoint(&new->thread, msr);
 581
 582         /* This loads the speculative FP/VEC state, if used */
 583         if (msr & MSR_FP) {
 584                 do_load_up_transact_fpu(&new->thread);
 585                 new->thread.regs->msr |=
 586                         (MSR_FP | new->thread.fpexc_mode);
 587         }
 588 #ifdef CONFIG_ALTIVEC
 589         if (msr & MSR_VEC) {
 590                 do_load_up_transact_altivec(&new->thread);
 591                 new->thread.regs->msr |= MSR_VEC;
 592         }
 593 #endif
 594         /* We may as well turn on VSX too since all the state is restored now */
 595         if (msr & MSR_VSX)
 596                 new->thread.regs->msr |= MSR_VSX;
 597
 598         TM_DEBUG("*** tm_recheckpoint of pid %d complete "
 599                  "(kernel msr 0x%lx)\n",
 600                  new->pid, mfmsr());
 601 }
 602
 603 static inline void __switch_to_tm(struct task_struct *prev)
 604 {
 605         if (cpu_has_feature(CPU_FTR_TM)) {
 606                 tm_enable();
 607                 tm_reclaim_task(prev);
 608         }
 609 }
 610 #else
 611 #define tm_recheckpoint_new_task(new)
 612 #define __switch_to_tm(prev)
 613 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
 614
 615 struct task_struct *__switch_to(struct task_struct *prev,
 616         struct task_struct *new)
 617 {
 618         struct thread_struct *new_thread, *old_thread;
 619         unsigned long flags;
 620         struct task_struct *last;
 621 #ifdef CONFIG_PPC_BOOK3S_64
 622         struct ppc64_tlb_batch *batch;
 623 #endif
 624
 625         /* Back up the TAR across context switches.
 626          * Note that the TAR is not available for use in the kernel.  (To
 627          * provide this, the TAR should be backed up/restored on exception
 628          * entry/exit instead, and be in pt_regs.  FIXME, this should be in
 629          * pt_regs anyway (for debug).)
 630          * Save the TAR here before we do treclaim/trecheckpoint as these
 631          * will change the TAR.
 632          */
 633         save_tar(&prev->thread);
 634
 635         __switch_to_tm(prev);
 636
 637 #ifdef CONFIG_SMP
 638         /* avoid complexity of lazy save/restore of fpu
 639          * by just saving it every time we switch out if
 640          * this task used the fpu during the last quantum.
 641          *
 642          * If it tries to use the fpu again, it'll trap and
 643          * reload its fp regs.  So we don't have to do a restore
 644          * every switch, just a save.
 645          *  -- Cort
 646          */
 647         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 648                 giveup_fpu(prev);
 649 #ifdef CONFIG_ALTIVEC
 650         /*
 651          * If the previous thread used altivec in the last quantum
 652          * (thus changing altivec regs) then save them.
 653          * We used to check the VRSAVE register but not all apps
 654          * set it, so we don't rely on it now (and in fact we need
 655          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 656          *
 657          * On SMP we always save/restore altivec regs just to avoid the
 658          * complexity of changing processors.
 659          *  -- Cort
 660          */
 661         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 662                 giveup_altivec(prev);
 663 #endif /* CONFIG_ALTIVEC */
 664 #ifdef CONFIG_VSX
 665         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 666                 /* VMX and FPU registers are already save here */
 667                 __giveup_vsx(prev);
 668 #endif /* CONFIG_VSX */
 669 #ifdef CONFIG_SPE
 670         /*
 671          * If the previous thread used spe in the last quantum
 672          * (thus changing spe regs) then save them.
 673          *
 674          * On SMP we always save/restore spe regs just to avoid the
 675          * complexity of changing processors.
 676          */
 677         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 678                 giveup_spe(prev);
 679 #endif /* CONFIG_SPE */
 680
 681 #else  /* CONFIG_SMP */
 682 #ifdef CONFIG_ALTIVEC
 683         /* Avoid the trap.  On smp this this never happens since
 684          * we don't set last_task_used_altivec -- Cort
 685          */
 686         if (new->thread.regs && last_task_used_altivec == new)
 687                 new->thread.regs->msr |= MSR_VEC;
 688 #endif /* CONFIG_ALTIVEC */
 689 #ifdef CONFIG_VSX
 690         if (new->thread.regs && last_task_used_vsx == new)
 691                 new->thread.regs->msr |= MSR_VSX;
 692 #endif /* CONFIG_VSX */
 693 #ifdef CONFIG_SPE
 694         /* Avoid the trap.  On smp this this never happens since
 695          * we don't set last_task_used_spe
 696          */
 697         if (new->thread.regs && last_task_used_spe == new)
 698                 new->thread.regs->msr |= MSR_SPE;
 699 #endif /* CONFIG_SPE */
 700
 701 #endif /* CONFIG_SMP */
 702
 703 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 704         switch_booke_debug_regs(&new->thread);
 705 #else
 706 /*
 707  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
 708  * schedule DABR
 709  */
 710 #ifndef CONFIG_HAVE_HW_BREAKPOINT
 711         if (unlikely(hw_brk_match(&__get_cpu_var(current_brk), &new->thread.hw_brk)))
 712                 set_breakpoint(&new->thread.hw_brk);
 713 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 714 #endif
 715
 716
 717         new_thread = &new->thread;
 718         old_thread = &current->thread;
 719
 720 #ifdef CONFIG_PPC64
 721         /*
 722          * Collect processor utilization data per process
 723          */
 724         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 725                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 726                 long unsigned start_tb, current_tb;
 727                 start_tb = old_thread->start_tb;
 728                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 729                 old_thread->accum_tb += (current_tb - start_tb);
 730                 new_thread->start_tb = current_tb;
 731         }
 732 #endif /* CONFIG_PPC64 */
 733
 734 #ifdef CONFIG_PPC_BOOK3S_64
 735         batch = &__get_cpu_var(ppc64_tlb_batch);
 736         if (batch->active) {
 737                 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
 738                 if (batch->index)
 739                         __flush_tlb_pending(batch);
 740                 batch->active = 0;
 741         }
 742 #endif /* CONFIG_PPC_BOOK3S_64 */
 743
 744         local_irq_save(flags);
 745
 746         /*
 747          * We can't take a PMU exception inside _switch() since there is a
 748          * window where the kernel stack SLB and the kernel stack are out
 749          * of sync. Hard disable here.
 750          */
 751         hard_irq_disable();
 752
 753         tm_recheckpoint_new_task(new);
 754
 755         last = _switch(old_thread, new_thread);
 756
 757 #ifdef CONFIG_PPC_BOOK3S_64
 758         if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
 759                 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
 760                 batch = &__get_cpu_var(ppc64_tlb_batch);
 761                 batch->active = 1;
 762         }
 763 #endif /* CONFIG_PPC_BOOK3S_64 */
 764
 765         local_irq_restore(flags);
 766
 767         return last;
 768 }
 769
 770 static int instructions_to_print = 16;
 771
 772 static void show_instructions(struct pt_regs *regs)
 773 {
 774         int i;
 775         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 776                         sizeof(int));
 777
 778         printk("Instruction dump:");
 779
 780         for (i = 0; i < instructions_to_print; i++) {
 781                 int instr;
 782
 783                 if (!(i % 8))
 784                         printk("\n");
 785
 786 #if !defined(CONFIG_BOOKE)
 787                 /* If executing with the IMMU off, adjust pc rather
 788                  * than print XXXXXXXX.
 789                  */
 790                 if (!(regs->msr & MSR_IR))
 791                         pc = (unsigned long)phys_to_virt(pc);
 792 #endif
 793
 794                 /* We use __get_user here *only* to avoid an OOPS on a
 795                  * bad address because the pc *should* only be a
 796                  * kernel address.
 797                  */
 798                 if (!__kernel_text_address(pc) ||
 799                      __get_user(instr, (unsigned int __user *)pc)) {
 800                         printk(KERN_CONT "XXXXXXXX ");
 801                 } else {
 802                         if (regs->nip == pc)
 803                                 printk(KERN_CONT "<%08x> ", instr);
 804                         else
 805                                 printk(KERN_CONT "%08x ", instr);
 806                 }
 807
 808                 pc += sizeof(int);
 809         }
 810
 811         printk("\n");
 812 }
 813
 814 static struct regbit {
 815         unsigned long bit;
 816         const char *name;
 817 } msr_bits[] = {
 818 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
 819         {MSR_SF,        "SF"},
 820         {MSR_HV,        "HV"},
 821 #endif
 822         {MSR_VEC,       "VEC"},
 823         {MSR_VSX,       "VSX"},
 824 #ifdef CONFIG_BOOKE
 825         {MSR_CE,        "CE"},
 826 #endif
 827         {MSR_EE,        "EE"},
 828         {MSR_PR,        "PR"},
 829         {MSR_FP,        "FP"},
 830         {MSR_ME,        "ME"},
 831 #ifdef CONFIG_BOOKE
 832         {MSR_DE,        "DE"},
 833 #else
 834         {MSR_SE,        "SE"},
 835         {MSR_BE,        "BE"},
 836 #endif
 837         {MSR_IR,        "IR"},
 838         {MSR_DR,        "DR"},
 839         {MSR_PMM,       "PMM"},
 840 #ifndef CONFIG_BOOKE
 841         {MSR_RI,        "RI"},
 842         {MSR_LE,        "LE"},
 843 #endif
 844         {0,             NULL}
 845 };
 846
 847 static void printbits(unsigned long val, struct regbit *bits)
 848 {
 849         const char *sep = "";
 850
 851         printk("<");
 852         for (; bits->bit; ++bits)
 853                 if (val & bits->bit) {
 854                         printk("%s%s", sep, bits->name);
 855                         sep = ",";
 856                 }
 857         printk(">");
 858 }
 859
 860 #ifdef CONFIG_PPC64
 861 #define REG             "%016lx"
 862 #define REGS_PER_LINE   4
 863 #define LAST_VOLATILE   13
 864 #else
 865 #define REG             "%08lx"
 866 #define REGS_PER_LINE   8
 867 #define LAST_VOLATILE   12
 868 #endif
 869
 870 void show_regs(struct pt_regs * regs)
 871 {
 872         int i, trap;
 873
 874         show_regs_print_info(KERN_DEFAULT);
 875
 876         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 877                regs->nip, regs->link, regs->ctr);
 878         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 879                regs, regs->trap, print_tainted(), init_utsname()->release);
 880         printk("MSR: "REG" ", regs->msr);
 881         printbits(regs->msr, msr_bits);
 882         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 883 #ifdef CONFIG_PPC64
 884         printk("SOFTE: %ld\n", regs->softe);
 885 #endif
 886         trap = TRAP(regs);
 887         if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
 888                 printk("CFAR: "REG"\n", regs->orig_gpr3);
 889         if (trap == 0x300 || trap == 0x600)
 890 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 891                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 892 #else
 893                 printk("DAR: "REG", DSISR: %08lx\n", regs->dar, regs->dsisr);
 894 #endif
 895
 896         for (i = 0;  i < 32;  i++) {
 897                 if ((i % REGS_PER_LINE) == 0)
 898                         printk("\nGPR%02d: ", i);
 899                 printk(REG " ", regs->gpr[i]);
 900                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 901                         break;
 902         }
 903         printk("\n");
 904 #ifdef CONFIG_KALLSYMS
 905         /*
 906          * Lookup NIP late so we have the best change of getting the
 907          * above info out without failing
 908          */
 909         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 910         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 911 #endif
 912 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 913         printk("PACATMSCRATCH [%llx]\n", get_paca()->tm_scratch);
 914 #endif
 915         show_stack(current, (unsigned long *) regs->gpr[1]);
 916         if (!user_mode(regs))
 917                 show_instructions(regs);
 918 }
 919
 920 void exit_thread(void)
 921 {
 922         discard_lazy_cpu_state();
 923 }
 924
 925 void flush_thread(void)
 926 {
 927         discard_lazy_cpu_state();
 928
 929 #ifdef CONFIG_HAVE_HW_BREAKPOINT
 930         flush_ptrace_hw_breakpoint(current);
 931 #else /* CONFIG_HAVE_HW_BREAKPOINT */
 932         set_debug_reg_defaults(&current->thread);
 933 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 934 }
 935
 936 void
 937 release_thread(struct task_struct *t)
 938 {
 939 }
 940
 941 /*
 942  * this gets called so that we can store coprocessor state into memory and
 943  * copy the current task into the new thread.
 944  */
 945 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 946 {
 947         flush_fp_to_thread(src);
 948         flush_altivec_to_thread(src);
 949         flush_vsx_to_thread(src);
 950         flush_spe_to_thread(src);
 951         *dst = *src;
 952         return 0;
 953 }
 954
 955 /*
 956  * Copy a thread..
 957  */
 958 extern unsigned long dscr_default; /* defined in arch/powerpc/kernel/sysfs.c */
 959
 960 int copy_thread(unsigned long clone_flags, unsigned long usp,
 961                 unsigned long arg, struct task_struct *p)
 962 {
 963         struct pt_regs *childregs, *kregs;
 964         extern void ret_from_fork(void);
 965         extern void ret_from_kernel_thread(void);
 966         void (*f)(void);
 967         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 968
 969         /* Copy registers */
 970         sp -= sizeof(struct pt_regs);
 971         childregs = (struct pt_regs *) sp;
 972         if (unlikely(p->flags & PF_KTHREAD)) {
 973                 struct thread_info *ti = (void *)task_stack_page(p);
 974                 memset(childregs, 0, sizeof(struct pt_regs));
 975                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 976                 childregs->gpr[14] = usp;       /* function */
 977 #ifdef CONFIG_PPC64
 978                 clear_tsk_thread_flag(p, TIF_32BIT);
 979                 childregs->softe = 1;
 980 #endif
 981                 childregs->gpr[15] = arg;
 982                 p->thread.regs = NULL;  /* no user register state */
 983                 ti->flags |= _TIF_RESTOREALL;
 984                 f = ret_from_kernel_thread;
 985         } else {
 986                 struct pt_regs *regs = current_pt_regs();
 987                 CHECK_FULL_REGS(regs);
 988                 *childregs = *regs;
 989                 if (usp)
 990                         childregs->gpr[1] = usp;
 991                 p->thread.regs = childregs;
 992                 childregs->gpr[3] = 0;  /* Result from fork() */
 993                 if (clone_flags & CLONE_SETTLS) {
 994 #ifdef CONFIG_PPC64
 995                         if (!is_32bit_task())
 996                                 childregs->gpr[13] = childregs->gpr[6];
 997                         else
 998 #endif
 999                                 childregs->gpr[2] = childregs->gpr[6];
1000                 }
1001
1002                 f = ret_from_fork;
1003         }
1004         sp -= STACK_FRAME_OVERHEAD;
1005
1006         /*
1007          * The way this works is that at some point in the future
1008          * some task will call _switch to switch to the new task.
1009          * That will pop off the stack frame created below and start
1010          * the new task running at ret_from_fork.  The new task will
1011          * do some house keeping and then return from the fork or clone
1012          * system call, using the stack frame created above.
1013          */
1014         ((unsigned long *)sp)[0] = 0;
1015         sp -= sizeof(struct pt_regs);
1016         kregs = (struct pt_regs *) sp;
1017         sp -= STACK_FRAME_OVERHEAD;
1018         p->thread.ksp = sp;
1019         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1020                                 _ALIGN_UP(sizeof(struct thread_info), 16);
1021
1022 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1023         p->thread.ptrace_bps[0] = NULL;
1024 #endif
1025
1026 #ifdef CONFIG_PPC_STD_MMU_64
1027         if (mmu_has_feature(MMU_FTR_SLB)) {
1028                 unsigned long sp_vsid;
1029                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1030
1031                 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1032                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1033                                 << SLB_VSID_SHIFT_1T;
1034                 else
1035                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1036                                 << SLB_VSID_SHIFT;
1037                 sp_vsid |= SLB_VSID_KERNEL | llp;
1038                 p->thread.ksp_vsid = sp_vsid;
1039         }
1040 #endif /* CONFIG_PPC_STD_MMU_64 */
1041 #ifdef CONFIG_PPC64
1042         if (cpu_has_feature(CPU_FTR_DSCR)) {
1043                 p->thread.dscr_inherit = current->thread.dscr_inherit;
1044                 p->thread.dscr = current->thread.dscr;
1045         }
1046         if (cpu_has_feature(CPU_FTR_HAS_PPR))
1047                 p->thread.ppr = INIT_PPR;
1048 #endif
1049         /*
1050          * The PPC64 ABI makes use of a TOC to contain function
1051          * pointers.  The function (ret_from_except) is actually a pointer
1052          * to the TOC entry.  The first entry is a pointer to the actual
1053          * function.
1054          */
1055 #ifdef CONFIG_PPC64
1056         kregs->nip = *((unsigned long *)f);
1057 #else
1058         kregs->nip = (unsigned long)f;
1059 #endif
1060         return 0;
1061 }
1062
1063 /*
1064  * Set up a thread for executing a new program
1065  */
1066 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1067 {
1068 #ifdef CONFIG_PPC64
1069         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1070 #endif
1071
1072         /*
1073          * If we exec out of a kernel thread then thread.regs will not be
1074          * set.  Do it now.
1075          */
1076         if (!current->thread.regs) {
1077                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1078                 current->thread.regs = regs - 1;
1079         }
1080
1081         memset(regs->gpr, 0, sizeof(regs->gpr));
1082         regs->ctr = 0;
1083         regs->link = 0;
1084         regs->xer = 0;
1085         regs->ccr = 0;
1086         regs->gpr[1] = sp;
1087
1088         /*
1089          * We have just cleared all the nonvolatile GPRs, so make
1090          * FULL_REGS(regs) return true.  This is necessary to allow
1091          * ptrace to examine the thread immediately after exec.
1092          */
1093         regs->trap &= ~1UL;
1094
1095 #ifdef CONFIG_PPC32
1096         regs->mq = 0;
1097         regs->nip = start;
1098         regs->msr = MSR_USER;
1099 #else
1100         if (!is_32bit_task()) {
1101                 unsigned long entry, toc;
1102
1103                 /* start is a relocated pointer to the function descriptor for
1104                  * the elf _start routine.  The first entry in the function
1105                  * descriptor is the entry address of _start and the second
1106                  * entry is the TOC value we need to use.
1107                  */
1108                 __get_user(entry, (unsigned long __user *)start);
1109                 __get_user(toc, (unsigned long __user *)start+1);
1110
1111                 /* Check whether the e_entry function descriptor entries
1112                  * need to be relocated before we can use them.
1113                  */
1114                 if (load_addr != 0) {
1115                         entry += load_addr;
1116                         toc   += load_addr;
1117                 }
1118                 regs->nip = entry;
1119                 regs->gpr[2] = toc;
1120                 regs->msr = MSR_USER64;
1121         } else {
1122                 regs->nip = start;
1123                 regs->gpr[2] = 0;
1124                 regs->msr = MSR_USER32;
1125         }
1126 #endif
1127         discard_lazy_cpu_state();
1128 #ifdef CONFIG_VSX
1129         current->thread.used_vsr = 0;
1130 #endif
1131         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
1132         current->thread.fpscr.val = 0;
1133 #ifdef CONFIG_ALTIVEC
1134         memset(current->thread.vr, 0, sizeof(current->thread.vr));
1135         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
1136         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
1137         current->thread.vrsave = 0;
1138         current->thread.used_vr = 0;
1139 #endif /* CONFIG_ALTIVEC */
1140 #ifdef CONFIG_SPE
1141         memset(current->thread.evr, 0, sizeof(current->thread.evr));
1142         current->thread.acc = 0;
1143         current->thread.spefscr = 0;
1144         current->thread.used_spe = 0;
1145 #endif /* CONFIG_SPE */
1146 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1147         if (cpu_has_feature(CPU_FTR_TM))
1148                 regs->msr |= MSR_TM;
1149         current->thread.tm_tfhar = 0;
1150         current->thread.tm_texasr = 0;
1151         current->thread.tm_tfiar = 0;
1152 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1153 }
1154
1155 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1156                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
1157
1158 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1159 {
1160         struct pt_regs *regs = tsk->thread.regs;
1161
1162         /* This is a bit hairy.  If we are an SPE enabled  processor
1163          * (have embedded fp) we store the IEEE exception enable flags in
1164          * fpexc_mode.  fpexc_mode is also used for setting FP exception
1165          * mode (asyn, precise, disabled) for 'Classic' FP. */
1166         if (val & PR_FP_EXC_SW_ENABLE) {
1167 #ifdef CONFIG_SPE
1168                 if (cpu_has_feature(CPU_FTR_SPE)) {
1169                         tsk->thread.fpexc_mode = val &
1170                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1171                         return 0;
1172                 } else {
1173                         return -EINVAL;
1174                 }
1175 #else
1176                 return -EINVAL;
1177 #endif
1178         }
1179
1180         /* on a CONFIG_SPE this does not hurt us.  The bits that
1181          * __pack_fe01 use do not overlap with bits used for
1182          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1183          * on CONFIG_SPE implementations are reserved so writing to
1184          * them does not change anything */
1185         if (val > PR_FP_EXC_PRECISE)
1186                 return -EINVAL;
1187         tsk->thread.fpexc_mode = __pack_fe01(val);
1188         if (regs != NULL && (regs->msr & MSR_FP) != 0)
1189                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1190                         | tsk->thread.fpexc_mode;
1191         return 0;
1192 }
1193
1194 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1195 {
1196         unsigned int val;
1197
1198         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1199 #ifdef CONFIG_SPE
1200                 if (cpu_has_feature(CPU_FTR_SPE))
1201                         val = tsk->thread.fpexc_mode;
1202                 else
1203                         return -EINVAL;
1204 #else
1205                 return -EINVAL;
1206 #endif
1207         else
1208                 val = __unpack_fe01(tsk->thread.fpexc_mode);
1209         return put_user(val, (unsigned int __user *) adr);
1210 }
1211
1212 int set_endian(struct task_struct *tsk, unsigned int val)
1213 {
1214         struct pt_regs *regs = tsk->thread.regs;
1215
1216         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1217             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1218                 return -EINVAL;
1219
1220         if (regs == NULL)
1221                 return -EINVAL;
1222
1223         if (val == PR_ENDIAN_BIG)
1224                 regs->msr &= ~MSR_LE;
1225         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1226                 regs->msr |= MSR_LE;
1227         else
1228                 return -EINVAL;
1229
1230         return 0;
1231 }
1232
1233 int get_endian(struct task_struct *tsk, unsigned long adr)
1234 {
1235         struct pt_regs *regs = tsk->thread.regs;
1236         unsigned int val;
1237
1238         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1239             !cpu_has_feature(CPU_FTR_REAL_LE))
1240                 return -EINVAL;
1241
1242         if (regs == NULL)
1243                 return -EINVAL;
1244
1245         if (regs->msr & MSR_LE) {
1246                 if (cpu_has_feature(CPU_FTR_REAL_LE))
1247                         val = PR_ENDIAN_LITTLE;
1248                 else
1249                         val = PR_ENDIAN_PPC_LITTLE;
1250         } else
1251                 val = PR_ENDIAN_BIG;
1252
1253         return put_user(val, (unsigned int __user *)adr);
1254 }
1255
1256 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1257 {
1258         tsk->thread.align_ctl = val;
1259         return 0;
1260 }
1261
1262 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1263 {
1264         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1265 }
1266
1267 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1268                                   unsigned long nbytes)
1269 {
1270         unsigned long stack_page;
1271         unsigned long cpu = task_cpu(p);
1272
1273         /*
1274          * Avoid crashing if the stack has overflowed and corrupted
1275          * task_cpu(p), which is in the thread_info struct.
1276          */
1277         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1278                 stack_page = (unsigned long) hardirq_ctx[cpu];
1279                 if (sp >= stack_page + sizeof(struct thread_struct)
1280                     && sp <= stack_page + THREAD_SIZE - nbytes)
1281                         return 1;
1282
1283                 stack_page = (unsigned long) softirq_ctx[cpu];
1284                 if (sp >= stack_page + sizeof(struct thread_struct)
1285                     && sp <= stack_page + THREAD_SIZE - nbytes)
1286                         return 1;
1287         }
1288         return 0;
1289 }
1290
1291 int validate_sp(unsigned long sp, struct task_struct *p,
1292                        unsigned long nbytes)
1293 {
1294         unsigned long stack_page = (unsigned long)task_stack_page(p);
1295
1296         if (sp >= stack_page + sizeof(struct thread_struct)
1297             && sp <= stack_page + THREAD_SIZE - nbytes)
1298                 return 1;
1299
1300         return valid_irq_stack(sp, p, nbytes);
1301 }
1302
1303 EXPORT_SYMBOL(validate_sp);
1304
1305 unsigned long get_wchan(struct task_struct *p)
1306 {
1307         unsigned long ip, sp;
1308         int count = 0;
1309
1310         if (!p || p == current || p->state == TASK_RUNNING)
1311                 return 0;
1312
1313         sp = p->thread.ksp;
1314         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1315                 return 0;
1316
1317         do {
1318                 sp = *(unsigned long *)sp;
1319                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1320                         return 0;
1321                 if (count > 0) {
1322                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1323                         if (!in_sched_functions(ip))
1324                                 return ip;
1325                 }
1326         } while (count++ < 16);
1327         return 0;
1328 }
1329
1330 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1331
1332 void show_stack(struct task_struct *tsk, unsigned long *stack)
1333 {
1334         unsigned long sp, ip, lr, newsp;
1335         int count = 0;
1336         int firstframe = 1;
1337 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1338         int curr_frame = current->curr_ret_stack;
1339         extern void return_to_handler(void);
1340         unsigned long rth = (unsigned long)return_to_handler;
1341         unsigned long mrth = -1;
1342 #ifdef CONFIG_PPC64
1343         extern void mod_return_to_handler(void);
1344         rth = *(unsigned long *)rth;
1345         mrth = (unsigned long)mod_return_to_handler;
1346         mrth = *(unsigned long *)mrth;
1347 #endif
1348 #endif
1349
1350         sp = (unsigned long) stack;
1351         if (tsk == NULL)
1352                 tsk = current;
1353         if (sp == 0) {
1354                 if (tsk == current)
1355                         asm("mr %0,1" : "=r" (sp));
1356                 else
1357                         sp = tsk->thread.ksp;
1358         }
1359
1360         lr = 0;
1361         printk("Call Trace:\n");
1362         do {
1363                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1364                         return;
1365
1366                 stack = (unsigned long *) sp;
1367                 newsp = stack[0];
1368                 ip = stack[STACK_FRAME_LR_SAVE];
1369                 if (!firstframe || ip != lr) {
1370                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1371 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1372                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1373                                 printk(" (%pS)",
1374                                        (void *)current->ret_stack[curr_frame].ret);
1375                                 curr_frame--;
1376                         }
1377 #endif
1378                         if (firstframe)
1379                                 printk(" (unreliable)");
1380                         printk("\n");
1381                 }
1382                 firstframe = 0;
1383
1384                 /*
1385                  * See if this is an exception frame.
1386                  * We look for the "regshere" marker in the current frame.
1387                  */
1388                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1389                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1390                         struct pt_regs *regs = (struct pt_regs *)
1391                                 (sp + STACK_FRAME_OVERHEAD);
1392                         lr = regs->link;
1393                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1394                                regs->trap, (void *)regs->nip, (void *)lr);
1395                         firstframe = 1;
1396                 }
1397
1398                 sp = newsp;
1399         } while (count++ < kstack_depth_to_print);
1400 }
1401
1402 #ifdef CONFIG_PPC64
1403 /* Called with hard IRQs off */
1404 void notrace __ppc64_runlatch_on(void)
1405 {
1406         struct thread_info *ti = current_thread_info();
1407         unsigned long ctrl;
1408
1409         ctrl = mfspr(SPRN_CTRLF);
1410         ctrl |= CTRL_RUNLATCH;
1411         mtspr(SPRN_CTRLT, ctrl);
1412
1413         ti->local_flags |= _TLF_RUNLATCH;
1414 }
1415
1416 /* Called with hard IRQs off */
1417 void notrace __ppc64_runlatch_off(void)
1418 {
1419         struct thread_info *ti = current_thread_info();
1420         unsigned long ctrl;
1421
1422         ti->local_flags &= ~_TLF_RUNLATCH;
1423
1424         ctrl = mfspr(SPRN_CTRLF);
1425         ctrl &= ~CTRL_RUNLATCH;
1426         mtspr(SPRN_CTRLT, ctrl);
1427 }
1428 #endif /* CONFIG_PPC64 */
1429
1430 unsigned long arch_align_stack(unsigned long sp)
1431 {
1432         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1433                 sp -= get_random_int() & ~PAGE_MASK;
1434         return sp & ~0xf;
1435 }
1436
1437 static inline unsigned long brk_rnd(void)
1438 {
1439         unsigned long rnd = 0;
1440
1441         /* 8MB for 32bit, 1GB for 64bit */
1442         if (is_32bit_task())
1443                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1444         else
1445                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1446
1447         return rnd << PAGE_SHIFT;
1448 }
1449
1450 unsigned long arch_randomize_brk(struct mm_struct *mm)
1451 {
1452         unsigned long base = mm->brk;
1453         unsigned long ret;
1454
1455 #ifdef CONFIG_PPC_STD_MMU_64
1456         /*
1457          * If we are using 1TB segments and we are allowed to randomise
1458          * the heap, we can put it above 1TB so it is backed by a 1TB
1459          * segment. Otherwise the heap will be in the bottom 1TB
1460          * which always uses 256MB segments and this may result in a
1461          * performance penalty.
1462          */
1463         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1464                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1465 #endif
1466
1467         ret = PAGE_ALIGN(base + brk_rnd());
1468
1469         if (ret < mm->brk)
1470                 return mm->brk;
1471
1472         return ret;
1473 }
1474
1475 unsigned long randomize_et_dyn(unsigned long base)
1476 {
1477         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1478
1479         if (ret < base)
1480                 return base;
1481
1482         return ret;
1483 }