arch/powerpc/kernel/process.c

   1 /*
   2  *  Derived from "arch/i386/kernel/process.c"
   3  *    Copyright (C) 1995  Linus Torvalds
   4  *
   5  *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
   6  *  Paul Mackerras (paulus@cs.anu.edu.au)
   7  *
   8  *  PowerPC version
   9  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  10  *
  11  *  This program is free software; you can redistribute it and/or
  12  *  modify it under the terms of the GNU General Public License
  13  *  as published by the Free Software Foundation; either version
  14  *  2 of the License, or (at your option) any later version.
  15  */
  16
  17 #include <linux/errno.h>
  18 #include <linux/sched.h>
  19 #include <linux/kernel.h>
  20 #include <linux/mm.h>
  21 #include <linux/smp.h>
  22 #include <linux/stddef.h>
  23 #include <linux/unistd.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/slab.h>
  26 #include <linux/user.h>
  27 #include <linux/elf.h>
  28 #include <linux/init.h>
  29 #include <linux/prctl.h>
  30 #include <linux/init_task.h>
  31 #include <linux/module.h>
  32 #include <linux/kallsyms.h>
  33 #include <linux/mqueue.h>
  34 #include <linux/hardirq.h>
  35 #include <linux/utsname.h>
  36 #include <linux/ftrace.h>
  37 #include <linux/kernel_stat.h>
  38 #include <linux/personality.h>
  39 #include <linux/random.h>
  40
  41 #include <asm/pgtable.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/system.h>
  44 #include <asm/io.h>
  45 #include <asm/processor.h>
  46 #include <asm/mmu.h>
  47 #include <asm/prom.h>
  48 #include <asm/machdep.h>
  49 #include <asm/time.h>
  50 #include <asm/syscalls.h>
  51 #ifdef CONFIG_PPC64
  52 #include <asm/firmware.h>
  53 #endif
  54 #include <linux/kprobes.h>
  55 #include <linux/kdebug.h>
  56
  57 extern unsigned long _get_SP(void);
  58
  59 #ifndef CONFIG_SMP
  60 struct task_struct *last_task_used_math = NULL;
  61 struct task_struct *last_task_used_altivec = NULL;
  62 struct task_struct *last_task_used_vsx = NULL;
  63 struct task_struct *last_task_used_spe = NULL;
  64 #endif
  65
  66 /*
  67  * Make sure the floating-point register state in the
  68  * the thread_struct is up to date for task tsk.
  69  */
  70 void flush_fp_to_thread(struct task_struct *tsk)
  71 {
  72         if (tsk->thread.regs) {
  73                 /*
  74                  * We need to disable preemption here because if we didn't,
  75                  * another process could get scheduled after the regs->msr
  76                  * test but before we have finished saving the FP registers
  77                  * to the thread_struct.  That process could take over the
  78                  * FPU, and then when we get scheduled again we would store
  79                  * bogus values for the remaining FP registers.
  80                  */
  81                 preempt_disable();
  82                 if (tsk->thread.regs->msr & MSR_FP) {
  83 #ifdef CONFIG_SMP
  84                         /*
  85                          * This should only ever be called for current or
  86                          * for a stopped child process.  Since we save away
  87                          * the FP register state on context switch on SMP,
  88                          * there is something wrong if a stopped child appears
  89                          * to still have its FP state in the CPU registers.
  90                          */
  91                         BUG_ON(tsk != current);
  92 #endif
  93                         giveup_fpu(tsk);
  94                 }
  95                 preempt_enable();
  96         }
  97 }
  98
  99 void enable_kernel_fp(void)
 100 {
 101         WARN_ON(preemptible());
 102
 103 #ifdef CONFIG_SMP
 104         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
 105                 giveup_fpu(current);
 106         else
 107                 giveup_fpu(NULL);       /* just enables FP for kernel */
 108 #else
 109         giveup_fpu(last_task_used_math);
 110 #endif /* CONFIG_SMP */
 111 }
 112 EXPORT_SYMBOL(enable_kernel_fp);
 113
 114 #ifdef CONFIG_ALTIVEC
 115 void enable_kernel_altivec(void)
 116 {
 117         WARN_ON(preemptible());
 118
 119 #ifdef CONFIG_SMP
 120         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 121                 giveup_altivec(current);
 122         else
 123                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
 124 #else
 125         giveup_altivec(last_task_used_altivec);
 126 #endif /* CONFIG_SMP */
 127 }
 128 EXPORT_SYMBOL(enable_kernel_altivec);
 129
 130 /*
 131  * Make sure the VMX/Altivec register state in the
 132  * the thread_struct is up to date for task tsk.
 133  */
 134 void flush_altivec_to_thread(struct task_struct *tsk)
 135 {
 136         if (tsk->thread.regs) {
 137                 preempt_disable();
 138                 if (tsk->thread.regs->msr & MSR_VEC) {
 139 #ifdef CONFIG_SMP
 140                         BUG_ON(tsk != current);
 141 #endif
 142                         giveup_altivec(tsk);
 143                 }
 144                 preempt_enable();
 145         }
 146 }
 147 #endif /* CONFIG_ALTIVEC */
 148
 149 #ifdef CONFIG_VSX
 150 #if 0
 151 /* not currently used, but some crazy RAID module might want to later */
 152 void enable_kernel_vsx(void)
 153 {
 154         WARN_ON(preemptible());
 155
 156 #ifdef CONFIG_SMP
 157         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 158                 giveup_vsx(current);
 159         else
 160                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 161 #else
 162         giveup_vsx(last_task_used_vsx);
 163 #endif /* CONFIG_SMP */
 164 }
 165 EXPORT_SYMBOL(enable_kernel_vsx);
 166 #endif
 167
 168 void giveup_vsx(struct task_struct *tsk)
 169 {
 170         giveup_fpu(tsk);
 171         giveup_altivec(tsk);
 172         __giveup_vsx(tsk);
 173 }
 174
 175 void flush_vsx_to_thread(struct task_struct *tsk)
 176 {
 177         if (tsk->thread.regs) {
 178                 preempt_disable();
 179                 if (tsk->thread.regs->msr & MSR_VSX) {
 180 #ifdef CONFIG_SMP
 181                         BUG_ON(tsk != current);
 182 #endif
 183                         giveup_vsx(tsk);
 184                 }
 185                 preempt_enable();
 186         }
 187 }
 188 #endif /* CONFIG_VSX */
 189
 190 #ifdef CONFIG_SPE
 191
 192 void enable_kernel_spe(void)
 193 {
 194         WARN_ON(preemptible());
 195
 196 #ifdef CONFIG_SMP
 197         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 198                 giveup_spe(current);
 199         else
 200                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 201 #else
 202         giveup_spe(last_task_used_spe);
 203 #endif /* __SMP __ */
 204 }
 205 EXPORT_SYMBOL(enable_kernel_spe);
 206
 207 void flush_spe_to_thread(struct task_struct *tsk)
 208 {
 209         if (tsk->thread.regs) {
 210                 preempt_disable();
 211                 if (tsk->thread.regs->msr & MSR_SPE) {
 212 #ifdef CONFIG_SMP
 213                         BUG_ON(tsk != current);
 214 #endif
 215                         giveup_spe(tsk);
 216                 }
 217                 preempt_enable();
 218         }
 219 }
 220 #endif /* CONFIG_SPE */
 221
 222 #ifndef CONFIG_SMP
 223 /*
 224  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 225  * and the current task has some state, discard it.
 226  */
 227 void discard_lazy_cpu_state(void)
 228 {
 229         preempt_disable();
 230         if (last_task_used_math == current)
 231                 last_task_used_math = NULL;
 232 #ifdef CONFIG_ALTIVEC
 233         if (last_task_used_altivec == current)
 234                 last_task_used_altivec = NULL;
 235 #endif /* CONFIG_ALTIVEC */
 236 #ifdef CONFIG_VSX
 237         if (last_task_used_vsx == current)
 238                 last_task_used_vsx = NULL;
 239 #endif /* CONFIG_VSX */
 240 #ifdef CONFIG_SPE
 241         if (last_task_used_spe == current)
 242                 last_task_used_spe = NULL;
 243 #endif
 244         preempt_enable();
 245 }
 246 #endif /* CONFIG_SMP */
 247
 248 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 249 void do_send_trap(struct pt_regs *regs, unsigned long address,
 250                   unsigned long error_code, int signal_code, int breakpt)
 251 {
 252         siginfo_t info;
 253
 254         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 255                         11, SIGSEGV) == NOTIFY_STOP)
 256                 return;
 257
 258         /* Deliver the signal to userspace */
 259         info.si_signo = SIGTRAP;
 260         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
 261         info.si_code = signal_code;
 262         info.si_addr = (void __user *)address;
 263         force_sig_info(SIGTRAP, &info, current);
 264 }
 265 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 266 void do_dabr(struct pt_regs *regs, unsigned long address,
 267                     unsigned long error_code)
 268 {
 269         siginfo_t info;
 270
 271         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 272                         11, SIGSEGV) == NOTIFY_STOP)
 273                 return;
 274
 275         if (debugger_dabr_match(regs))
 276                 return;
 277
 278         /* Clear the DABR */
 279         set_dabr(0);
 280
 281         /* Deliver the signal to userspace */
 282         info.si_signo = SIGTRAP;
 283         info.si_errno = 0;
 284         info.si_code = TRAP_HWBKPT;
 285         info.si_addr = (void __user *)address;
 286         force_sig_info(SIGTRAP, &info, current);
 287 }
 288 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 289
 290 static DEFINE_PER_CPU(unsigned long, current_dabr);
 291
 292 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 293 /*
 294  * Set the debug registers back to their default "safe" values.
 295  */
 296 static void set_debug_reg_defaults(struct thread_struct *thread)
 297 {
 298         thread->iac1 = thread->iac2 = 0;
 299 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 300         thread->iac3 = thread->iac4 = 0;
 301 #endif
 302         thread->dac1 = thread->dac2 = 0;
 303 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 304         thread->dvc1 = thread->dvc2 = 0;
 305 #endif
 306         thread->dbcr0 = 0;
 307 #ifdef CONFIG_BOOKE
 308         /*
 309          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
 310          */
 311         thread->dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |   \
 312                         DBCR1_IAC3US | DBCR1_IAC4US;
 313         /*
 314          * Force Data Address Compare User/Supervisor bits to be User-only
 315          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
 316          */
 317         thread->dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
 318 #else
 319         thread->dbcr1 = 0;
 320 #endif
 321 }
 322
 323 static void prime_debug_regs(struct thread_struct *thread)
 324 {
 325         mtspr(SPRN_IAC1, thread->iac1);
 326         mtspr(SPRN_IAC2, thread->iac2);
 327 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 328         mtspr(SPRN_IAC3, thread->iac3);
 329         mtspr(SPRN_IAC4, thread->iac4);
 330 #endif
 331         mtspr(SPRN_DAC1, thread->dac1);
 332         mtspr(SPRN_DAC2, thread->dac2);
 333 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 334         mtspr(SPRN_DVC1, thread->dvc1);
 335         mtspr(SPRN_DVC2, thread->dvc2);
 336 #endif
 337         mtspr(SPRN_DBCR0, thread->dbcr0);
 338         mtspr(SPRN_DBCR1, thread->dbcr1);
 339 #ifdef CONFIG_BOOKE
 340         mtspr(SPRN_DBCR2, thread->dbcr2);
 341 #endif
 342 }
 343 /*
 344  * Unless neither the old or new thread are making use of the
 345  * debug registers, set the debug registers from the values
 346  * stored in the new thread.
 347  */
 348 static void switch_booke_debug_regs(struct thread_struct *new_thread)
 349 {
 350         if ((current->thread.dbcr0 & DBCR0_IDM)
 351                 || (new_thread->dbcr0 & DBCR0_IDM))
 352                         prime_debug_regs(new_thread);
 353 }
 354 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 355 static void set_debug_reg_defaults(struct thread_struct *thread)
 356 {
 357         if (thread->dabr) {
 358                 thread->dabr = 0;
 359                 set_dabr(0);
 360         }
 361 }
 362 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 363
 364 int set_dabr(unsigned long dabr)
 365 {
 366         __get_cpu_var(current_dabr) = dabr;
 367
 368         if (ppc_md.set_dabr)
 369                 return ppc_md.set_dabr(dabr);
 370
 371         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 372 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 373         mtspr(SPRN_DAC1, dabr);
 374 #elif defined(CONFIG_PPC_BOOK3S)
 375         mtspr(SPRN_DABR, dabr);
 376 #endif
 377
 378
 379         return 0;
 380 }
 381
 382 #ifdef CONFIG_PPC64
 383 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 384 #endif
 385
 386 struct task_struct *__switch_to(struct task_struct *prev,
 387         struct task_struct *new)
 388 {
 389         struct thread_struct *new_thread, *old_thread;
 390         unsigned long flags;
 391         struct task_struct *last;
 392
 393 #ifdef CONFIG_SMP
 394         /* avoid complexity of lazy save/restore of fpu
 395          * by just saving it every time we switch out if
 396          * this task used the fpu during the last quantum.
 397          *
 398          * If it tries to use the fpu again, it'll trap and
 399          * reload its fp regs.  So we don't have to do a restore
 400          * every switch, just a save.
 401          *  -- Cort
 402          */
 403         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 404                 giveup_fpu(prev);
 405 #ifdef CONFIG_ALTIVEC
 406         /*
 407          * If the previous thread used altivec in the last quantum
 408          * (thus changing altivec regs) then save them.
 409          * We used to check the VRSAVE register but not all apps
 410          * set it, so we don't rely on it now (and in fact we need
 411          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 412          *
 413          * On SMP we always save/restore altivec regs just to avoid the
 414          * complexity of changing processors.
 415          *  -- Cort
 416          */
 417         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 418                 giveup_altivec(prev);
 419 #endif /* CONFIG_ALTIVEC */
 420 #ifdef CONFIG_VSX
 421         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 422                 /* VMX and FPU registers are already save here */
 423                 __giveup_vsx(prev);
 424 #endif /* CONFIG_VSX */
 425 #ifdef CONFIG_SPE
 426         /*
 427          * If the previous thread used spe in the last quantum
 428          * (thus changing spe regs) then save them.
 429          *
 430          * On SMP we always save/restore spe regs just to avoid the
 431          * complexity of changing processors.
 432          */
 433         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 434                 giveup_spe(prev);
 435 #endif /* CONFIG_SPE */
 436
 437 #else  /* CONFIG_SMP */
 438 #ifdef CONFIG_ALTIVEC
 439         /* Avoid the trap.  On smp this this never happens since
 440          * we don't set last_task_used_altivec -- Cort
 441          */
 442         if (new->thread.regs && last_task_used_altivec == new)
 443                 new->thread.regs->msr |= MSR_VEC;
 444 #endif /* CONFIG_ALTIVEC */
 445 #ifdef CONFIG_VSX
 446         if (new->thread.regs && last_task_used_vsx == new)
 447                 new->thread.regs->msr |= MSR_VSX;
 448 #endif /* CONFIG_VSX */
 449 #ifdef CONFIG_SPE
 450         /* Avoid the trap.  On smp this this never happens since
 451          * we don't set last_task_used_spe
 452          */
 453         if (new->thread.regs && last_task_used_spe == new)
 454                 new->thread.regs->msr |= MSR_SPE;
 455 #endif /* CONFIG_SPE */
 456
 457 #endif /* CONFIG_SMP */
 458
 459 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 460         switch_booke_debug_regs(&new->thread);
 461 #else
 462         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 463                 set_dabr(new->thread.dabr);
 464 #endif
 465
 466
 467         new_thread = &new->thread;
 468         old_thread = &current->thread;
 469
 470 #ifdef CONFIG_PPC64
 471         /*
 472          * Collect processor utilization data per process
 473          */
 474         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 475                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 476                 long unsigned start_tb, current_tb;
 477                 start_tb = old_thread->start_tb;
 478                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 479                 old_thread->accum_tb += (current_tb - start_tb);
 480                 new_thread->start_tb = current_tb;
 481         }
 482 #endif
 483
 484         local_irq_save(flags);
 485
 486         account_system_vtime(current);
 487         account_process_vtime(current);
 488         calculate_steal_time();
 489
 490         /*
 491          * We can't take a PMU exception inside _switch() since there is a
 492          * window where the kernel stack SLB and the kernel stack are out
 493          * of sync. Hard disable here.
 494          */
 495         hard_irq_disable();
 496         last = _switch(old_thread, new_thread);
 497
 498         local_irq_restore(flags);
 499
 500         return last;
 501 }
 502
 503 static int instructions_to_print = 16;
 504
 505 static void show_instructions(struct pt_regs *regs)
 506 {
 507         int i;
 508         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 509                         sizeof(int));
 510
 511         printk("Instruction dump:");
 512
 513         for (i = 0; i < instructions_to_print; i++) {
 514                 int instr;
 515
 516                 if (!(i % 8))
 517                         printk("\n");
 518
 519 #if !defined(CONFIG_BOOKE)
 520                 /* If executing with the IMMU off, adjust pc rather
 521                  * than print XXXXXXXX.
 522                  */
 523                 if (!(regs->msr & MSR_IR))
 524                         pc = (unsigned long)phys_to_virt(pc);
 525 #endif
 526
 527                 /* We use __get_user here *only* to avoid an OOPS on a
 528                  * bad address because the pc *should* only be a
 529                  * kernel address.
 530                  */
 531                 if (!__kernel_text_address(pc) ||
 532                      __get_user(instr, (unsigned int __user *)pc)) {
 533                         printk("XXXXXXXX ");
 534                 } else {
 535                         if (regs->nip == pc)
 536                                 printk("<%08x> ", instr);
 537                         else
 538                                 printk("%08x ", instr);
 539                 }
 540
 541                 pc += sizeof(int);
 542         }
 543
 544         printk("\n");
 545 }
 546
 547 static struct regbit {
 548         unsigned long bit;
 549         const char *name;
 550 } msr_bits[] = {
 551         {MSR_EE,        "EE"},
 552         {MSR_PR,        "PR"},
 553         {MSR_FP,        "FP"},
 554         {MSR_VEC,       "VEC"},
 555         {MSR_VSX,       "VSX"},
 556         {MSR_ME,        "ME"},
 557         {MSR_CE,        "CE"},
 558         {MSR_DE,        "DE"},
 559         {MSR_IR,        "IR"},
 560         {MSR_DR,        "DR"},
 561         {0,             NULL}
 562 };
 563
 564 static void printbits(unsigned long val, struct regbit *bits)
 565 {
 566         const char *sep = "";
 567
 568         printk("<");
 569         for (; bits->bit; ++bits)
 570                 if (val & bits->bit) {
 571                         printk("%s%s", sep, bits->name);
 572                         sep = ",";
 573                 }
 574         printk(">");
 575 }
 576
 577 #ifdef CONFIG_PPC64
 578 #define REG             "%016lx"
 579 #define REGS_PER_LINE   4
 580 #define LAST_VOLATILE   13
 581 #else
 582 #define REG             "%08lx"
 583 #define REGS_PER_LINE   8
 584 #define LAST_VOLATILE   12
 585 #endif
 586
 587 void show_regs(struct pt_regs * regs)
 588 {
 589         int i, trap;
 590
 591         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 592                regs->nip, regs->link, regs->ctr);
 593         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 594                regs, regs->trap, print_tainted(), init_utsname()->release);
 595         printk("MSR: "REG" ", regs->msr);
 596         printbits(regs->msr, msr_bits);
 597         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 598         trap = TRAP(regs);
 599         if (trap == 0x300 || trap == 0x600)
 600 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 601                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 602 #else
 603                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 604 #endif
 605         printk("TASK = %p[%d] '%s' THREAD: %p",
 606                current, task_pid_nr(current), current->comm, task_thread_info(current));
 607
 608 #ifdef CONFIG_SMP
 609         printk(" CPU: %d", raw_smp_processor_id());
 610 #endif /* CONFIG_SMP */
 611
 612         for (i = 0;  i < 32;  i++) {
 613                 if ((i % REGS_PER_LINE) == 0)
 614                         printk("\nGPR%02d: ", i);
 615                 printk(REG " ", regs->gpr[i]);
 616                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 617                         break;
 618         }
 619         printk("\n");
 620 #ifdef CONFIG_KALLSYMS
 621         /*
 622          * Lookup NIP late so we have the best change of getting the
 623          * above info out without failing
 624          */
 625         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 626         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 627 #endif
 628         show_stack(current, (unsigned long *) regs->gpr[1]);
 629         if (!user_mode(regs))
 630                 show_instructions(regs);
 631 }
 632
 633 void exit_thread(void)
 634 {
 635         discard_lazy_cpu_state();
 636 }
 637
 638 void flush_thread(void)
 639 {
 640         discard_lazy_cpu_state();
 641
 642         set_debug_reg_defaults(&current->thread);
 643 }
 644
 645 void
 646 release_thread(struct task_struct *t)
 647 {
 648 }
 649
 650 /*
 651  * This gets called before we allocate a new thread and copy
 652  * the current task into it.
 653  */
 654 void prepare_to_copy(struct task_struct *tsk)
 655 {
 656         flush_fp_to_thread(current);
 657         flush_altivec_to_thread(current);
 658         flush_vsx_to_thread(current);
 659         flush_spe_to_thread(current);
 660 }
 661
 662 /*
 663  * Copy a thread..
 664  */
 665 int copy_thread(unsigned long clone_flags, unsigned long usp,
 666                 unsigned long unused, struct task_struct *p,
 667                 struct pt_regs *regs)
 668 {
 669         struct pt_regs *childregs, *kregs;
 670         extern void ret_from_fork(void);
 671         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 672
 673         CHECK_FULL_REGS(regs);
 674         /* Copy registers */
 675         sp -= sizeof(struct pt_regs);
 676         childregs = (struct pt_regs *) sp;
 677         *childregs = *regs;
 678         if ((childregs->msr & MSR_PR) == 0) {
 679                 /* for kernel thread, set `current' and stackptr in new task */
 680                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 681 #ifdef CONFIG_PPC32
 682                 childregs->gpr[2] = (unsigned long) p;
 683 #else
 684                 clear_tsk_thread_flag(p, TIF_32BIT);
 685 #endif
 686                 p->thread.regs = NULL;  /* no user register state */
 687         } else {
 688                 childregs->gpr[1] = usp;
 689                 p->thread.regs = childregs;
 690                 if (clone_flags & CLONE_SETTLS) {
 691 #ifdef CONFIG_PPC64
 692                         if (!test_thread_flag(TIF_32BIT))
 693                                 childregs->gpr[13] = childregs->gpr[6];
 694                         else
 695 #endif
 696                                 childregs->gpr[2] = childregs->gpr[6];
 697                 }
 698         }
 699         childregs->gpr[3] = 0;  /* Result from fork() */
 700         sp -= STACK_FRAME_OVERHEAD;
 701
 702         /*
 703          * The way this works is that at some point in the future
 704          * some task will call _switch to switch to the new task.
 705          * That will pop off the stack frame created below and start
 706          * the new task running at ret_from_fork.  The new task will
 707          * do some house keeping and then return from the fork or clone
 708          * system call, using the stack frame created above.
 709          */
 710         sp -= sizeof(struct pt_regs);
 711         kregs = (struct pt_regs *) sp;
 712         sp -= STACK_FRAME_OVERHEAD;
 713         p->thread.ksp = sp;
 714         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 715                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 716
 717 #ifdef CONFIG_PPC_STD_MMU_64
 718         if (cpu_has_feature(CPU_FTR_SLB)) {
 719                 unsigned long sp_vsid;
 720                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 721
 722                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
 723                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 724                                 << SLB_VSID_SHIFT_1T;
 725                 else
 726                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 727                                 << SLB_VSID_SHIFT;
 728                 sp_vsid |= SLB_VSID_KERNEL | llp;
 729                 p->thread.ksp_vsid = sp_vsid;
 730         }
 731 #endif /* CONFIG_PPC_STD_MMU_64 */
 732
 733         /*
 734          * The PPC64 ABI makes use of a TOC to contain function
 735          * pointers.  The function (ret_from_except) is actually a pointer
 736          * to the TOC entry.  The first entry is a pointer to the actual
 737          * function.
 738          */
 739 #ifdef CONFIG_PPC64
 740         kregs->nip = *((unsigned long *)ret_from_fork);
 741 #else
 742         kregs->nip = (unsigned long)ret_from_fork;
 743 #endif
 744
 745         return 0;
 746 }
 747
 748 /*
 749  * Set up a thread for executing a new program
 750  */
 751 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 752 {
 753 #ifdef CONFIG_PPC64
 754         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 755 #endif
 756
 757         set_fs(USER_DS);
 758
 759         /*
 760          * If we exec out of a kernel thread then thread.regs will not be
 761          * set.  Do it now.
 762          */
 763         if (!current->thread.regs) {
 764                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 765                 current->thread.regs = regs - 1;
 766         }
 767
 768         memset(regs->gpr, 0, sizeof(regs->gpr));
 769         regs->ctr = 0;
 770         regs->link = 0;
 771         regs->xer = 0;
 772         regs->ccr = 0;
 773         regs->gpr[1] = sp;
 774
 775         /*
 776          * We have just cleared all the nonvolatile GPRs, so make
 777          * FULL_REGS(regs) return true.  This is necessary to allow
 778          * ptrace to examine the thread immediately after exec.
 779          */
 780         regs->trap &= ~1UL;
 781
 782 #ifdef CONFIG_PPC32
 783         regs->mq = 0;
 784         regs->nip = start;
 785         regs->msr = MSR_USER;
 786 #else
 787         if (!test_thread_flag(TIF_32BIT)) {
 788                 unsigned long entry, toc;
 789
 790                 /* start is a relocated pointer to the function descriptor for
 791                  * the elf _start routine.  The first entry in the function
 792                  * descriptor is the entry address of _start and the second
 793                  * entry is the TOC value we need to use.
 794                  */
 795                 __get_user(entry, (unsigned long __user *)start);
 796                 __get_user(toc, (unsigned long __user *)start+1);
 797
 798                 /* Check whether the e_entry function descriptor entries
 799                  * need to be relocated before we can use them.
 800                  */
 801                 if (load_addr != 0) {
 802                         entry += load_addr;
 803                         toc   += load_addr;
 804                 }
 805                 regs->nip = entry;
 806                 regs->gpr[2] = toc;
 807                 regs->msr = MSR_USER64;
 808         } else {
 809                 regs->nip = start;
 810                 regs->gpr[2] = 0;
 811                 regs->msr = MSR_USER32;
 812         }
 813 #endif
 814
 815         discard_lazy_cpu_state();
 816 #ifdef CONFIG_VSX
 817         current->thread.used_vsr = 0;
 818 #endif
 819         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 820         current->thread.fpscr.val = 0;
 821 #ifdef CONFIG_ALTIVEC
 822         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 823         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 824         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 825         current->thread.vrsave = 0;
 826         current->thread.used_vr = 0;
 827 #endif /* CONFIG_ALTIVEC */
 828 #ifdef CONFIG_SPE
 829         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 830         current->thread.acc = 0;
 831         current->thread.spefscr = 0;
 832         current->thread.used_spe = 0;
 833 #endif /* CONFIG_SPE */
 834 }
 835
 836 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 837                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 838
 839 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 840 {
 841         struct pt_regs *regs = tsk->thread.regs;
 842
 843         /* This is a bit hairy.  If we are an SPE enabled  processor
 844          * (have embedded fp) we store the IEEE exception enable flags in
 845          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 846          * mode (asyn, precise, disabled) for 'Classic' FP. */
 847         if (val & PR_FP_EXC_SW_ENABLE) {
 848 #ifdef CONFIG_SPE
 849                 if (cpu_has_feature(CPU_FTR_SPE)) {
 850                         tsk->thread.fpexc_mode = val &
 851                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 852                         return 0;
 853                 } else {
 854                         return -EINVAL;
 855                 }
 856 #else
 857                 return -EINVAL;
 858 #endif
 859         }
 860
 861         /* on a CONFIG_SPE this does not hurt us.  The bits that
 862          * __pack_fe01 use do not overlap with bits used for
 863          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 864          * on CONFIG_SPE implementations are reserved so writing to
 865          * them does not change anything */
 866         if (val > PR_FP_EXC_PRECISE)
 867                 return -EINVAL;
 868         tsk->thread.fpexc_mode = __pack_fe01(val);
 869         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 870                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 871                         | tsk->thread.fpexc_mode;
 872         return 0;
 873 }
 874
 875 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 876 {
 877         unsigned int val;
 878
 879         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 880 #ifdef CONFIG_SPE
 881                 if (cpu_has_feature(CPU_FTR_SPE))
 882                         val = tsk->thread.fpexc_mode;
 883                 else
 884                         return -EINVAL;
 885 #else
 886                 return -EINVAL;
 887 #endif
 888         else
 889                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 890         return put_user(val, (unsigned int __user *) adr);
 891 }
 892
 893 int set_endian(struct task_struct *tsk, unsigned int val)
 894 {
 895         struct pt_regs *regs = tsk->thread.regs;
 896
 897         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 898             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 899                 return -EINVAL;
 900
 901         if (regs == NULL)
 902                 return -EINVAL;
 903
 904         if (val == PR_ENDIAN_BIG)
 905                 regs->msr &= ~MSR_LE;
 906         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 907                 regs->msr |= MSR_LE;
 908         else
 909                 return -EINVAL;
 910
 911         return 0;
 912 }
 913
 914 int get_endian(struct task_struct *tsk, unsigned long adr)
 915 {
 916         struct pt_regs *regs = tsk->thread.regs;
 917         unsigned int val;
 918
 919         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 920             !cpu_has_feature(CPU_FTR_REAL_LE))
 921                 return -EINVAL;
 922
 923         if (regs == NULL)
 924                 return -EINVAL;
 925
 926         if (regs->msr & MSR_LE) {
 927                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 928                         val = PR_ENDIAN_LITTLE;
 929                 else
 930                         val = PR_ENDIAN_PPC_LITTLE;
 931         } else
 932                 val = PR_ENDIAN_BIG;
 933
 934         return put_user(val, (unsigned int __user *)adr);
 935 }
 936
 937 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 938 {
 939         tsk->thread.align_ctl = val;
 940         return 0;
 941 }
 942
 943 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 944 {
 945         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 946 }
 947
 948 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 949
 950 int sys_clone(unsigned long clone_flags, unsigned long usp,
 951               int __user *parent_tidp, void __user *child_threadptr,
 952               int __user *child_tidp, int p6,
 953               struct pt_regs *regs)
 954 {
 955         CHECK_FULL_REGS(regs);
 956         if (usp == 0)
 957                 usp = regs->gpr[1];     /* stack pointer for child */
 958 #ifdef CONFIG_PPC64
 959         if (test_thread_flag(TIF_32BIT)) {
 960                 parent_tidp = TRUNC_PTR(parent_tidp);
 961                 child_tidp = TRUNC_PTR(child_tidp);
 962         }
 963 #endif
 964         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 965 }
 966
 967 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 968              unsigned long p4, unsigned long p5, unsigned long p6,
 969              struct pt_regs *regs)
 970 {
 971         CHECK_FULL_REGS(regs);
 972         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 973 }
 974
 975 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
 976               unsigned long p4, unsigned long p5, unsigned long p6,
 977               struct pt_regs *regs)
 978 {
 979         CHECK_FULL_REGS(regs);
 980         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 981                         regs, 0, NULL, NULL);
 982 }
 983
 984 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 985                unsigned long a3, unsigned long a4, unsigned long a5,
 986                struct pt_regs *regs)
 987 {
 988         int error;
 989         char *filename;
 990
 991         filename = getname((char __user *) a0);
 992         error = PTR_ERR(filename);
 993         if (IS_ERR(filename))
 994                 goto out;
 995         flush_fp_to_thread(current);
 996         flush_altivec_to_thread(current);
 997         flush_spe_to_thread(current);
 998         error = do_execve(filename, (char __user * __user *) a1,
 999                           (char __user * __user *) a2, regs);
1000         putname(filename);
1001 out:
1002         return error;
1003 }
1004
1005 #ifdef CONFIG_IRQSTACKS
1006 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1007                                   unsigned long nbytes)
1008 {
1009         unsigned long stack_page;
1010         unsigned long cpu = task_cpu(p);
1011
1012         /*
1013          * Avoid crashing if the stack has overflowed and corrupted
1014          * task_cpu(p), which is in the thread_info struct.
1015          */
1016         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1017                 stack_page = (unsigned long) hardirq_ctx[cpu];
1018                 if (sp >= stack_page + sizeof(struct thread_struct)
1019                     && sp <= stack_page + THREAD_SIZE - nbytes)
1020                         return 1;
1021
1022                 stack_page = (unsigned long) softirq_ctx[cpu];
1023                 if (sp >= stack_page + sizeof(struct thread_struct)
1024                     && sp <= stack_page + THREAD_SIZE - nbytes)
1025                         return 1;
1026         }
1027         return 0;
1028 }
1029
1030 #else
1031 #define valid_irq_stack(sp, p, nb)      0
1032 #endif /* CONFIG_IRQSTACKS */
1033
1034 int validate_sp(unsigned long sp, struct task_struct *p,
1035                        unsigned long nbytes)
1036 {
1037         unsigned long stack_page = (unsigned long)task_stack_page(p);
1038
1039         if (sp >= stack_page + sizeof(struct thread_struct)
1040             && sp <= stack_page + THREAD_SIZE - nbytes)
1041                 return 1;
1042
1043         return valid_irq_stack(sp, p, nbytes);
1044 }
1045
1046 EXPORT_SYMBOL(validate_sp);
1047
1048 unsigned long get_wchan(struct task_struct *p)
1049 {
1050         unsigned long ip, sp;
1051         int count = 0;
1052
1053         if (!p || p == current || p->state == TASK_RUNNING)
1054                 return 0;
1055
1056         sp = p->thread.ksp;
1057         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1058                 return 0;
1059
1060         do {
1061                 sp = *(unsigned long *)sp;
1062                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1063                         return 0;
1064                 if (count > 0) {
1065                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1066                         if (!in_sched_functions(ip))
1067                                 return ip;
1068                 }
1069         } while (count++ < 16);
1070         return 0;
1071 }
1072
1073 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1074
1075 void show_stack(struct task_struct *tsk, unsigned long *stack)
1076 {
1077         unsigned long sp, ip, lr, newsp;
1078         int count = 0;
1079         int firstframe = 1;
1080 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1081         int curr_frame = current->curr_ret_stack;
1082         extern void return_to_handler(void);
1083         unsigned long rth = (unsigned long)return_to_handler;
1084         unsigned long mrth = -1;
1085 #ifdef CONFIG_PPC64
1086         extern void mod_return_to_handler(void);
1087         rth = *(unsigned long *)rth;
1088         mrth = (unsigned long)mod_return_to_handler;
1089         mrth = *(unsigned long *)mrth;
1090 #endif
1091 #endif
1092
1093         sp = (unsigned long) stack;
1094         if (tsk == NULL)
1095                 tsk = current;
1096         if (sp == 0) {
1097                 if (tsk == current)
1098                         asm("mr %0,1" : "=r" (sp));
1099                 else
1100                         sp = tsk->thread.ksp;
1101         }
1102
1103         lr = 0;
1104         printk("Call Trace:\n");
1105         do {
1106                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1107                         return;
1108
1109                 stack = (unsigned long *) sp;
1110                 newsp = stack[0];
1111                 ip = stack[STACK_FRAME_LR_SAVE];
1112                 if (!firstframe || ip != lr) {
1113                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1114 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1115                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1116                                 printk(" (%pS)",
1117                                        (void *)current->ret_stack[curr_frame].ret);
1118                                 curr_frame--;
1119                         }
1120 #endif
1121                         if (firstframe)
1122                                 printk(" (unreliable)");
1123                         printk("\n");
1124                 }
1125                 firstframe = 0;
1126
1127                 /*
1128                  * See if this is an exception frame.
1129                  * We look for the "regshere" marker in the current frame.
1130                  */
1131                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1132                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1133                         struct pt_regs *regs = (struct pt_regs *)
1134                                 (sp + STACK_FRAME_OVERHEAD);
1135                         lr = regs->link;
1136                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1137                                regs->trap, (void *)regs->nip, (void *)lr);
1138                         firstframe = 1;
1139                 }
1140
1141                 sp = newsp;
1142         } while (count++ < kstack_depth_to_print);
1143 }
1144
1145 void dump_stack(void)
1146 {
1147         show_stack(current, NULL);
1148 }
1149 EXPORT_SYMBOL(dump_stack);
1150
1151 #ifdef CONFIG_PPC64
1152 void ppc64_runlatch_on(void)
1153 {
1154         unsigned long ctrl;
1155
1156         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1157                 HMT_medium();
1158
1159                 ctrl = mfspr(SPRN_CTRLF);
1160                 ctrl |= CTRL_RUNLATCH;
1161                 mtspr(SPRN_CTRLT, ctrl);
1162
1163                 set_thread_flag(TIF_RUNLATCH);
1164         }
1165 }
1166
1167 void ppc64_runlatch_off(void)
1168 {
1169         unsigned long ctrl;
1170
1171         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1172                 HMT_medium();
1173
1174                 clear_thread_flag(TIF_RUNLATCH);
1175
1176                 ctrl = mfspr(SPRN_CTRLF);
1177                 ctrl &= ~CTRL_RUNLATCH;
1178                 mtspr(SPRN_CTRLT, ctrl);
1179         }
1180 }
1181 #endif
1182
1183 #if THREAD_SHIFT < PAGE_SHIFT
1184
1185 static struct kmem_cache *thread_info_cache;
1186
1187 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1188 {
1189         struct thread_info *ti;
1190
1191         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1192         if (unlikely(ti == NULL))
1193                 return NULL;
1194 #ifdef CONFIG_DEBUG_STACK_USAGE
1195         memset(ti, 0, THREAD_SIZE);
1196 #endif
1197         return ti;
1198 }
1199
1200 void free_thread_info(struct thread_info *ti)
1201 {
1202         kmem_cache_free(thread_info_cache, ti);
1203 }
1204
1205 void thread_info_cache_init(void)
1206 {
1207         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1208                                               THREAD_SIZE, 0, NULL);
1209         BUG_ON(thread_info_cache == NULL);
1210 }
1211
1212 #endif /* THREAD_SHIFT < PAGE_SHIFT */
1213
1214 unsigned long arch_align_stack(unsigned long sp)
1215 {
1216         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1217                 sp -= get_random_int() & ~PAGE_MASK;
1218         return sp & ~0xf;
1219 }
1220
1221 static inline unsigned long brk_rnd(void)
1222 {
1223         unsigned long rnd = 0;
1224
1225         /* 8MB for 32bit, 1GB for 64bit */
1226         if (is_32bit_task())
1227                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1228         else
1229                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1230
1231         return rnd << PAGE_SHIFT;
1232 }
1233
1234 unsigned long arch_randomize_brk(struct mm_struct *mm)
1235 {
1236         unsigned long base = mm->brk;
1237         unsigned long ret;
1238
1239 #ifdef CONFIG_PPC_STD_MMU_64
1240         /*
1241          * If we are using 1TB segments and we are allowed to randomise
1242          * the heap, we can put it above 1TB so it is backed by a 1TB
1243          * segment. Otherwise the heap will be in the bottom 1TB
1244          * which always uses 256MB segments and this may result in a
1245          * performance penalty.
1246          */
1247         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1248                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1249 #endif
1250
1251         ret = PAGE_ALIGN(base + brk_rnd());
1252
1253         if (ret < mm->brk)
1254                 return mm->brk;
1255
1256         return ret;
1257 }
1258
1259 unsigned long randomize_et_dyn(unsigned long base)
1260 {
1261         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1262
1263         if (ret < base)
1264                 return base;
1265
1266         return ret;
1267 }