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 #ifdef CONFIG_PPC_47x
 375         isync();
 376 #endif
 377 #elif defined(CONFIG_PPC_BOOK3S)
 378         mtspr(SPRN_DABR, dabr);
 379 #endif
 380
 381
 382         return 0;
 383 }
 384
 385 #ifdef CONFIG_PPC64
 386 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 387 #endif
 388
 389 struct task_struct *__switch_to(struct task_struct *prev,
 390         struct task_struct *new)
 391 {
 392         struct thread_struct *new_thread, *old_thread;
 393         unsigned long flags;
 394         struct task_struct *last;
 395
 396 #ifdef CONFIG_SMP
 397         /* avoid complexity of lazy save/restore of fpu
 398          * by just saving it every time we switch out if
 399          * this task used the fpu during the last quantum.
 400          *
 401          * If it tries to use the fpu again, it'll trap and
 402          * reload its fp regs.  So we don't have to do a restore
 403          * every switch, just a save.
 404          *  -- Cort
 405          */
 406         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 407                 giveup_fpu(prev);
 408 #ifdef CONFIG_ALTIVEC
 409         /*
 410          * If the previous thread used altivec in the last quantum
 411          * (thus changing altivec regs) then save them.
 412          * We used to check the VRSAVE register but not all apps
 413          * set it, so we don't rely on it now (and in fact we need
 414          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 415          *
 416          * On SMP we always save/restore altivec regs just to avoid the
 417          * complexity of changing processors.
 418          *  -- Cort
 419          */
 420         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 421                 giveup_altivec(prev);
 422 #endif /* CONFIG_ALTIVEC */
 423 #ifdef CONFIG_VSX
 424         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 425                 /* VMX and FPU registers are already save here */
 426                 __giveup_vsx(prev);
 427 #endif /* CONFIG_VSX */
 428 #ifdef CONFIG_SPE
 429         /*
 430          * If the previous thread used spe in the last quantum
 431          * (thus changing spe regs) then save them.
 432          *
 433          * On SMP we always save/restore spe regs just to avoid the
 434          * complexity of changing processors.
 435          */
 436         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 437                 giveup_spe(prev);
 438 #endif /* CONFIG_SPE */
 439
 440 #else  /* CONFIG_SMP */
 441 #ifdef CONFIG_ALTIVEC
 442         /* Avoid the trap.  On smp this this never happens since
 443          * we don't set last_task_used_altivec -- Cort
 444          */
 445         if (new->thread.regs && last_task_used_altivec == new)
 446                 new->thread.regs->msr |= MSR_VEC;
 447 #endif /* CONFIG_ALTIVEC */
 448 #ifdef CONFIG_VSX
 449         if (new->thread.regs && last_task_used_vsx == new)
 450                 new->thread.regs->msr |= MSR_VSX;
 451 #endif /* CONFIG_VSX */
 452 #ifdef CONFIG_SPE
 453         /* Avoid the trap.  On smp this this never happens since
 454          * we don't set last_task_used_spe
 455          */
 456         if (new->thread.regs && last_task_used_spe == new)
 457                 new->thread.regs->msr |= MSR_SPE;
 458 #endif /* CONFIG_SPE */
 459
 460 #endif /* CONFIG_SMP */
 461
 462 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 463         switch_booke_debug_regs(&new->thread);
 464 #else
 465         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 466                 set_dabr(new->thread.dabr);
 467 #endif
 468
 469
 470         new_thread = &new->thread;
 471         old_thread = &current->thread;
 472
 473 #ifdef CONFIG_PPC64
 474         /*
 475          * Collect processor utilization data per process
 476          */
 477         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 478                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 479                 long unsigned start_tb, current_tb;
 480                 start_tb = old_thread->start_tb;
 481                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 482                 old_thread->accum_tb += (current_tb - start_tb);
 483                 new_thread->start_tb = current_tb;
 484         }
 485 #endif
 486
 487         local_irq_save(flags);
 488
 489         account_system_vtime(current);
 490         account_process_vtime(current);
 491         calculate_steal_time();
 492
 493         /*
 494          * We can't take a PMU exception inside _switch() since there is a
 495          * window where the kernel stack SLB and the kernel stack are out
 496          * of sync. Hard disable here.
 497          */
 498         hard_irq_disable();
 499         last = _switch(old_thread, new_thread);
 500
 501         local_irq_restore(flags);
 502
 503         return last;
 504 }
 505
 506 static int instructions_to_print = 16;
 507
 508 static void show_instructions(struct pt_regs *regs)
 509 {
 510         int i;
 511         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 512                         sizeof(int));
 513
 514         printk("Instruction dump:");
 515
 516         for (i = 0; i < instructions_to_print; i++) {
 517                 int instr;
 518
 519                 if (!(i % 8))
 520                         printk("\n");
 521
 522 #if !defined(CONFIG_BOOKE)
 523                 /* If executing with the IMMU off, adjust pc rather
 524                  * than print XXXXXXXX.
 525                  */
 526                 if (!(regs->msr & MSR_IR))
 527                         pc = (unsigned long)phys_to_virt(pc);
 528 #endif
 529
 530                 /* We use __get_user here *only* to avoid an OOPS on a
 531                  * bad address because the pc *should* only be a
 532                  * kernel address.
 533                  */
 534                 if (!__kernel_text_address(pc) ||
 535                      __get_user(instr, (unsigned int __user *)pc)) {
 536                         printk("XXXXXXXX ");
 537                 } else {
 538                         if (regs->nip == pc)
 539                                 printk("<%08x> ", instr);
 540                         else
 541                                 printk("%08x ", instr);
 542                 }
 543
 544                 pc += sizeof(int);
 545         }
 546
 547         printk("\n");
 548 }
 549
 550 static struct regbit {
 551         unsigned long bit;
 552         const char *name;
 553 } msr_bits[] = {
 554         {MSR_EE,        "EE"},
 555         {MSR_PR,        "PR"},
 556         {MSR_FP,        "FP"},
 557         {MSR_VEC,       "VEC"},
 558         {MSR_VSX,       "VSX"},
 559         {MSR_ME,        "ME"},
 560         {MSR_CE,        "CE"},
 561         {MSR_DE,        "DE"},
 562         {MSR_IR,        "IR"},
 563         {MSR_DR,        "DR"},
 564         {0,             NULL}
 565 };
 566
 567 static void printbits(unsigned long val, struct regbit *bits)
 568 {
 569         const char *sep = "";
 570
 571         printk("<");
 572         for (; bits->bit; ++bits)
 573                 if (val & bits->bit) {
 574                         printk("%s%s", sep, bits->name);
 575                         sep = ",";
 576                 }
 577         printk(">");
 578 }
 579
 580 #ifdef CONFIG_PPC64
 581 #define REG             "%016lx"
 582 #define REGS_PER_LINE   4
 583 #define LAST_VOLATILE   13
 584 #else
 585 #define REG             "%08lx"
 586 #define REGS_PER_LINE   8
 587 #define LAST_VOLATILE   12
 588 #endif
 589
 590 void show_regs(struct pt_regs * regs)
 591 {
 592         int i, trap;
 593
 594         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 595                regs->nip, regs->link, regs->ctr);
 596         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 597                regs, regs->trap, print_tainted(), init_utsname()->release);
 598         printk("MSR: "REG" ", regs->msr);
 599         printbits(regs->msr, msr_bits);
 600         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 601         trap = TRAP(regs);
 602         if (trap == 0x300 || trap == 0x600)
 603 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 604                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 605 #else
 606                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 607 #endif
 608         printk("TASK = %p[%d] '%s' THREAD: %p",
 609                current, task_pid_nr(current), current->comm, task_thread_info(current));
 610
 611 #ifdef CONFIG_SMP
 612         printk(" CPU: %d", raw_smp_processor_id());
 613 #endif /* CONFIG_SMP */
 614
 615         for (i = 0;  i < 32;  i++) {
 616                 if ((i % REGS_PER_LINE) == 0)
 617                         printk("\nGPR%02d: ", i);
 618                 printk(REG " ", regs->gpr[i]);
 619                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 620                         break;
 621         }
 622         printk("\n");
 623 #ifdef CONFIG_KALLSYMS
 624         /*
 625          * Lookup NIP late so we have the best change of getting the
 626          * above info out without failing
 627          */
 628         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 629         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 630 #endif
 631         show_stack(current, (unsigned long *) regs->gpr[1]);
 632         if (!user_mode(regs))
 633                 show_instructions(regs);
 634 }
 635
 636 void exit_thread(void)
 637 {
 638         discard_lazy_cpu_state();
 639 }
 640
 641 void flush_thread(void)
 642 {
 643         discard_lazy_cpu_state();
 644
 645         set_debug_reg_defaults(&current->thread);
 646 }
 647
 648 void
 649 release_thread(struct task_struct *t)
 650 {
 651 }
 652
 653 /*
 654  * This gets called before we allocate a new thread and copy
 655  * the current task into it.
 656  */
 657 void prepare_to_copy(struct task_struct *tsk)
 658 {
 659         flush_fp_to_thread(current);
 660         flush_altivec_to_thread(current);
 661         flush_vsx_to_thread(current);
 662         flush_spe_to_thread(current);
 663 }
 664
 665 /*
 666  * Copy a thread..
 667  */
 668 int copy_thread(unsigned long clone_flags, unsigned long usp,
 669                 unsigned long unused, struct task_struct *p,
 670                 struct pt_regs *regs)
 671 {
 672         struct pt_regs *childregs, *kregs;
 673         extern void ret_from_fork(void);
 674         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 675
 676         CHECK_FULL_REGS(regs);
 677         /* Copy registers */
 678         sp -= sizeof(struct pt_regs);
 679         childregs = (struct pt_regs *) sp;
 680         *childregs = *regs;
 681         if ((childregs->msr & MSR_PR) == 0) {
 682                 /* for kernel thread, set `current' and stackptr in new task */
 683                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 684 #ifdef CONFIG_PPC32
 685                 childregs->gpr[2] = (unsigned long) p;
 686 #else
 687                 clear_tsk_thread_flag(p, TIF_32BIT);
 688 #endif
 689                 p->thread.regs = NULL;  /* no user register state */
 690         } else {
 691                 childregs->gpr[1] = usp;
 692                 p->thread.regs = childregs;
 693                 if (clone_flags & CLONE_SETTLS) {
 694 #ifdef CONFIG_PPC64
 695                         if (!test_thread_flag(TIF_32BIT))
 696                                 childregs->gpr[13] = childregs->gpr[6];
 697                         else
 698 #endif
 699                                 childregs->gpr[2] = childregs->gpr[6];
 700                 }
 701         }
 702         childregs->gpr[3] = 0;  /* Result from fork() */
 703         sp -= STACK_FRAME_OVERHEAD;
 704
 705         /*
 706          * The way this works is that at some point in the future
 707          * some task will call _switch to switch to the new task.
 708          * That will pop off the stack frame created below and start
 709          * the new task running at ret_from_fork.  The new task will
 710          * do some house keeping and then return from the fork or clone
 711          * system call, using the stack frame created above.
 712          */
 713         sp -= sizeof(struct pt_regs);
 714         kregs = (struct pt_regs *) sp;
 715         sp -= STACK_FRAME_OVERHEAD;
 716         p->thread.ksp = sp;
 717         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 718                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 719
 720 #ifdef CONFIG_PPC_STD_MMU_64
 721         if (cpu_has_feature(CPU_FTR_SLB)) {
 722                 unsigned long sp_vsid;
 723                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 724
 725                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
 726                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 727                                 << SLB_VSID_SHIFT_1T;
 728                 else
 729                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 730                                 << SLB_VSID_SHIFT;
 731                 sp_vsid |= SLB_VSID_KERNEL | llp;
 732                 p->thread.ksp_vsid = sp_vsid;
 733         }
 734 #endif /* CONFIG_PPC_STD_MMU_64 */
 735
 736         /*
 737          * The PPC64 ABI makes use of a TOC to contain function
 738          * pointers.  The function (ret_from_except) is actually a pointer
 739          * to the TOC entry.  The first entry is a pointer to the actual
 740          * function.
 741          */
 742 #ifdef CONFIG_PPC64
 743         kregs->nip = *((unsigned long *)ret_from_fork);
 744 #else
 745         kregs->nip = (unsigned long)ret_from_fork;
 746 #endif
 747
 748         return 0;
 749 }
 750
 751 /*
 752  * Set up a thread for executing a new program
 753  */
 754 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 755 {
 756 #ifdef CONFIG_PPC64
 757         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 758 #endif
 759
 760         set_fs(USER_DS);
 761
 762         /*
 763          * If we exec out of a kernel thread then thread.regs will not be
 764          * set.  Do it now.
 765          */
 766         if (!current->thread.regs) {
 767                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 768                 current->thread.regs = regs - 1;
 769         }
 770
 771         memset(regs->gpr, 0, sizeof(regs->gpr));
 772         regs->ctr = 0;
 773         regs->link = 0;
 774         regs->xer = 0;
 775         regs->ccr = 0;
 776         regs->gpr[1] = sp;
 777
 778         /*
 779          * We have just cleared all the nonvolatile GPRs, so make
 780          * FULL_REGS(regs) return true.  This is necessary to allow
 781          * ptrace to examine the thread immediately after exec.
 782          */
 783         regs->trap &= ~1UL;
 784
 785 #ifdef CONFIG_PPC32
 786         regs->mq = 0;
 787         regs->nip = start;
 788         regs->msr = MSR_USER;
 789 #else
 790         if (!test_thread_flag(TIF_32BIT)) {
 791                 unsigned long entry, toc;
 792
 793                 /* start is a relocated pointer to the function descriptor for
 794                  * the elf _start routine.  The first entry in the function
 795                  * descriptor is the entry address of _start and the second
 796                  * entry is the TOC value we need to use.
 797                  */
 798                 __get_user(entry, (unsigned long __user *)start);
 799                 __get_user(toc, (unsigned long __user *)start+1);
 800
 801                 /* Check whether the e_entry function descriptor entries
 802                  * need to be relocated before we can use them.
 803                  */
 804                 if (load_addr != 0) {
 805                         entry += load_addr;
 806                         toc   += load_addr;
 807                 }
 808                 regs->nip = entry;
 809                 regs->gpr[2] = toc;
 810                 regs->msr = MSR_USER64;
 811         } else {
 812                 regs->nip = start;
 813                 regs->gpr[2] = 0;
 814                 regs->msr = MSR_USER32;
 815         }
 816 #endif
 817
 818         discard_lazy_cpu_state();
 819 #ifdef CONFIG_VSX
 820         current->thread.used_vsr = 0;
 821 #endif
 822         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 823         current->thread.fpscr.val = 0;
 824 #ifdef CONFIG_ALTIVEC
 825         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 826         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 827         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 828         current->thread.vrsave = 0;
 829         current->thread.used_vr = 0;
 830 #endif /* CONFIG_ALTIVEC */
 831 #ifdef CONFIG_SPE
 832         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 833         current->thread.acc = 0;
 834         current->thread.spefscr = 0;
 835         current->thread.used_spe = 0;
 836 #endif /* CONFIG_SPE */
 837 }
 838
 839 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 840                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 841
 842 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 843 {
 844         struct pt_regs *regs = tsk->thread.regs;
 845
 846         /* This is a bit hairy.  If we are an SPE enabled  processor
 847          * (have embedded fp) we store the IEEE exception enable flags in
 848          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 849          * mode (asyn, precise, disabled) for 'Classic' FP. */
 850         if (val & PR_FP_EXC_SW_ENABLE) {
 851 #ifdef CONFIG_SPE
 852                 if (cpu_has_feature(CPU_FTR_SPE)) {
 853                         tsk->thread.fpexc_mode = val &
 854                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 855                         return 0;
 856                 } else {
 857                         return -EINVAL;
 858                 }
 859 #else
 860                 return -EINVAL;
 861 #endif
 862         }
 863
 864         /* on a CONFIG_SPE this does not hurt us.  The bits that
 865          * __pack_fe01 use do not overlap with bits used for
 866          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 867          * on CONFIG_SPE implementations are reserved so writing to
 868          * them does not change anything */
 869         if (val > PR_FP_EXC_PRECISE)
 870                 return -EINVAL;
 871         tsk->thread.fpexc_mode = __pack_fe01(val);
 872         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 873                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 874                         | tsk->thread.fpexc_mode;
 875         return 0;
 876 }
 877
 878 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 879 {
 880         unsigned int val;
 881
 882         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 883 #ifdef CONFIG_SPE
 884                 if (cpu_has_feature(CPU_FTR_SPE))
 885                         val = tsk->thread.fpexc_mode;
 886                 else
 887                         return -EINVAL;
 888 #else
 889                 return -EINVAL;
 890 #endif
 891         else
 892                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 893         return put_user(val, (unsigned int __user *) adr);
 894 }
 895
 896 int set_endian(struct task_struct *tsk, unsigned int val)
 897 {
 898         struct pt_regs *regs = tsk->thread.regs;
 899
 900         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 901             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 902                 return -EINVAL;
 903
 904         if (regs == NULL)
 905                 return -EINVAL;
 906
 907         if (val == PR_ENDIAN_BIG)
 908                 regs->msr &= ~MSR_LE;
 909         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 910                 regs->msr |= MSR_LE;
 911         else
 912                 return -EINVAL;
 913
 914         return 0;
 915 }
 916
 917 int get_endian(struct task_struct *tsk, unsigned long adr)
 918 {
 919         struct pt_regs *regs = tsk->thread.regs;
 920         unsigned int val;
 921
 922         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 923             !cpu_has_feature(CPU_FTR_REAL_LE))
 924                 return -EINVAL;
 925
 926         if (regs == NULL)
 927                 return -EINVAL;
 928
 929         if (regs->msr & MSR_LE) {
 930                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 931                         val = PR_ENDIAN_LITTLE;
 932                 else
 933                         val = PR_ENDIAN_PPC_LITTLE;
 934         } else
 935                 val = PR_ENDIAN_BIG;
 936
 937         return put_user(val, (unsigned int __user *)adr);
 938 }
 939
 940 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 941 {
 942         tsk->thread.align_ctl = val;
 943         return 0;
 944 }
 945
 946 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 947 {
 948         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 949 }
 950
 951 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 952
 953 int sys_clone(unsigned long clone_flags, unsigned long usp,
 954               int __user *parent_tidp, void __user *child_threadptr,
 955               int __user *child_tidp, int p6,
 956               struct pt_regs *regs)
 957 {
 958         CHECK_FULL_REGS(regs);
 959         if (usp == 0)
 960                 usp = regs->gpr[1];     /* stack pointer for child */
 961 #ifdef CONFIG_PPC64
 962         if (test_thread_flag(TIF_32BIT)) {
 963                 parent_tidp = TRUNC_PTR(parent_tidp);
 964                 child_tidp = TRUNC_PTR(child_tidp);
 965         }
 966 #endif
 967         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 968 }
 969
 970 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 971              unsigned long p4, unsigned long p5, unsigned long p6,
 972              struct pt_regs *regs)
 973 {
 974         CHECK_FULL_REGS(regs);
 975         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 976 }
 977
 978 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
 979               unsigned long p4, unsigned long p5, unsigned long p6,
 980               struct pt_regs *regs)
 981 {
 982         CHECK_FULL_REGS(regs);
 983         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 984                         regs, 0, NULL, NULL);
 985 }
 986
 987 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 988                unsigned long a3, unsigned long a4, unsigned long a5,
 989                struct pt_regs *regs)
 990 {
 991         int error;
 992         char *filename;
 993
 994         filename = getname((char __user *) a0);
 995         error = PTR_ERR(filename);
 996         if (IS_ERR(filename))
 997                 goto out;
 998         flush_fp_to_thread(current);
 999         flush_altivec_to_thread(current);
1000         flush_spe_to_thread(current);
1001         error = do_execve(filename, (char __user * __user *) a1,
1002                           (char __user * __user *) a2, regs);
1003         putname(filename);
1004 out:
1005         return error;
1006 }
1007
1008 #ifdef CONFIG_IRQSTACKS
1009 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1010                                   unsigned long nbytes)
1011 {
1012         unsigned long stack_page;
1013         unsigned long cpu = task_cpu(p);
1014
1015         /*
1016          * Avoid crashing if the stack has overflowed and corrupted
1017          * task_cpu(p), which is in the thread_info struct.
1018          */
1019         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1020                 stack_page = (unsigned long) hardirq_ctx[cpu];
1021                 if (sp >= stack_page + sizeof(struct thread_struct)
1022                     && sp <= stack_page + THREAD_SIZE - nbytes)
1023                         return 1;
1024
1025                 stack_page = (unsigned long) softirq_ctx[cpu];
1026                 if (sp >= stack_page + sizeof(struct thread_struct)
1027                     && sp <= stack_page + THREAD_SIZE - nbytes)
1028                         return 1;
1029         }
1030         return 0;
1031 }
1032
1033 #else
1034 #define valid_irq_stack(sp, p, nb)      0
1035 #endif /* CONFIG_IRQSTACKS */
1036
1037 int validate_sp(unsigned long sp, struct task_struct *p,
1038                        unsigned long nbytes)
1039 {
1040         unsigned long stack_page = (unsigned long)task_stack_page(p);
1041
1042         if (sp >= stack_page + sizeof(struct thread_struct)
1043             && sp <= stack_page + THREAD_SIZE - nbytes)
1044                 return 1;
1045
1046         return valid_irq_stack(sp, p, nbytes);
1047 }
1048
1049 EXPORT_SYMBOL(validate_sp);
1050
1051 unsigned long get_wchan(struct task_struct *p)
1052 {
1053         unsigned long ip, sp;
1054         int count = 0;
1055
1056         if (!p || p == current || p->state == TASK_RUNNING)
1057                 return 0;
1058
1059         sp = p->thread.ksp;
1060         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1061                 return 0;
1062
1063         do {
1064                 sp = *(unsigned long *)sp;
1065                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1066                         return 0;
1067                 if (count > 0) {
1068                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1069                         if (!in_sched_functions(ip))
1070                                 return ip;
1071                 }
1072         } while (count++ < 16);
1073         return 0;
1074 }
1075
1076 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1077
1078 void show_stack(struct task_struct *tsk, unsigned long *stack)
1079 {
1080         unsigned long sp, ip, lr, newsp;
1081         int count = 0;
1082         int firstframe = 1;
1083 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1084         int curr_frame = current->curr_ret_stack;
1085         extern void return_to_handler(void);
1086         unsigned long rth = (unsigned long)return_to_handler;
1087         unsigned long mrth = -1;
1088 #ifdef CONFIG_PPC64
1089         extern void mod_return_to_handler(void);
1090         rth = *(unsigned long *)rth;
1091         mrth = (unsigned long)mod_return_to_handler;
1092         mrth = *(unsigned long *)mrth;
1093 #endif
1094 #endif
1095
1096         sp = (unsigned long) stack;
1097         if (tsk == NULL)
1098                 tsk = current;
1099         if (sp == 0) {
1100                 if (tsk == current)
1101                         asm("mr %0,1" : "=r" (sp));
1102                 else
1103                         sp = tsk->thread.ksp;
1104         }
1105
1106         lr = 0;
1107         printk("Call Trace:\n");
1108         do {
1109                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1110                         return;
1111
1112                 stack = (unsigned long *) sp;
1113                 newsp = stack[0];
1114                 ip = stack[STACK_FRAME_LR_SAVE];
1115                 if (!firstframe || ip != lr) {
1116                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1117 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1118                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1119                                 printk(" (%pS)",
1120                                        (void *)current->ret_stack[curr_frame].ret);
1121                                 curr_frame--;
1122                         }
1123 #endif
1124                         if (firstframe)
1125                                 printk(" (unreliable)");
1126                         printk("\n");
1127                 }
1128                 firstframe = 0;
1129
1130                 /*
1131                  * See if this is an exception frame.
1132                  * We look for the "regshere" marker in the current frame.
1133                  */
1134                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1135                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1136                         struct pt_regs *regs = (struct pt_regs *)
1137                                 (sp + STACK_FRAME_OVERHEAD);
1138                         lr = regs->link;
1139                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1140                                regs->trap, (void *)regs->nip, (void *)lr);
1141                         firstframe = 1;
1142                 }
1143
1144                 sp = newsp;
1145         } while (count++ < kstack_depth_to_print);
1146 }
1147
1148 void dump_stack(void)
1149 {
1150         show_stack(current, NULL);
1151 }
1152 EXPORT_SYMBOL(dump_stack);
1153
1154 #ifdef CONFIG_PPC64
1155 void ppc64_runlatch_on(void)
1156 {
1157         unsigned long ctrl;
1158
1159         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1160                 HMT_medium();
1161
1162                 ctrl = mfspr(SPRN_CTRLF);
1163                 ctrl |= CTRL_RUNLATCH;
1164                 mtspr(SPRN_CTRLT, ctrl);
1165
1166                 set_thread_flag(TIF_RUNLATCH);
1167         }
1168 }
1169
1170 void ppc64_runlatch_off(void)
1171 {
1172         unsigned long ctrl;
1173
1174         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1175                 HMT_medium();
1176
1177                 clear_thread_flag(TIF_RUNLATCH);
1178
1179                 ctrl = mfspr(SPRN_CTRLF);
1180                 ctrl &= ~CTRL_RUNLATCH;
1181                 mtspr(SPRN_CTRLT, ctrl);
1182         }
1183 }
1184 #endif
1185
1186 #if THREAD_SHIFT < PAGE_SHIFT
1187
1188 static struct kmem_cache *thread_info_cache;
1189
1190 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1191 {
1192         struct thread_info *ti;
1193
1194         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1195         if (unlikely(ti == NULL))
1196                 return NULL;
1197 #ifdef CONFIG_DEBUG_STACK_USAGE
1198         memset(ti, 0, THREAD_SIZE);
1199 #endif
1200         return ti;
1201 }
1202
1203 void free_thread_info(struct thread_info *ti)
1204 {
1205         kmem_cache_free(thread_info_cache, ti);
1206 }
1207
1208 void thread_info_cache_init(void)
1209 {
1210         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1211                                               THREAD_SIZE, 0, NULL);
1212         BUG_ON(thread_info_cache == NULL);
1213 }
1214
1215 #endif /* THREAD_SHIFT < PAGE_SHIFT */
1216
1217 unsigned long arch_align_stack(unsigned long sp)
1218 {
1219         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1220                 sp -= get_random_int() & ~PAGE_MASK;
1221         return sp & ~0xf;
1222 }
1223
1224 static inline unsigned long brk_rnd(void)
1225 {
1226         unsigned long rnd = 0;
1227
1228         /* 8MB for 32bit, 1GB for 64bit */
1229         if (is_32bit_task())
1230                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1231         else
1232                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1233
1234         return rnd << PAGE_SHIFT;
1235 }
1236
1237 unsigned long arch_randomize_brk(struct mm_struct *mm)
1238 {
1239         unsigned long base = mm->brk;
1240         unsigned long ret;
1241
1242 #ifdef CONFIG_PPC_STD_MMU_64
1243         /*
1244          * If we are using 1TB segments and we are allowed to randomise
1245          * the heap, we can put it above 1TB so it is backed by a 1TB
1246          * segment. Otherwise the heap will be in the bottom 1TB
1247          * which always uses 256MB segments and this may result in a
1248          * performance penalty.
1249          */
1250         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1251                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1252 #endif
1253
1254         ret = PAGE_ALIGN(base + brk_rnd());
1255
1256         if (ret < mm->brk)
1257                 return mm->brk;
1258
1259         return ret;
1260 }
1261
1262 unsigned long randomize_et_dyn(unsigned long base)
1263 {
1264         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1265
1266         if (ret < base)
1267                 return base;
1268
1269         return ret;
1270 }