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
  37 #include <asm/pgtable.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/system.h>
  40 #include <asm/io.h>
  41 #include <asm/processor.h>
  42 #include <asm/mmu.h>
  43 #include <asm/prom.h>
  44 #include <asm/machdep.h>
  45 #include <asm/time.h>
  46 #include <asm/syscalls.h>
  47 #ifdef CONFIG_PPC64
  48 #include <asm/firmware.h>
  49 #endif
  50 #include <linux/kprobes.h>
  51 #include <linux/kdebug.h>
  52
  53 extern unsigned long _get_SP(void);
  54
  55 #ifndef CONFIG_SMP
  56 struct task_struct *last_task_used_math = NULL;
  57 struct task_struct *last_task_used_altivec = NULL;
  58 struct task_struct *last_task_used_vsx = NULL;
  59 struct task_struct *last_task_used_spe = NULL;
  60 #endif
  61
  62 /*
  63  * Make sure the floating-point register state in the
  64  * the thread_struct is up to date for task tsk.
  65  */
  66 void flush_fp_to_thread(struct task_struct *tsk)
  67 {
  68         if (tsk->thread.regs) {
  69                 /*
  70                  * We need to disable preemption here because if we didn't,
  71                  * another process could get scheduled after the regs->msr
  72                  * test but before we have finished saving the FP registers
  73                  * to the thread_struct.  That process could take over the
  74                  * FPU, and then when we get scheduled again we would store
  75                  * bogus values for the remaining FP registers.
  76                  */
  77                 preempt_disable();
  78                 if (tsk->thread.regs->msr & MSR_FP) {
  79 #ifdef CONFIG_SMP
  80                         /*
  81                          * This should only ever be called for current or
  82                          * for a stopped child process.  Since we save away
  83                          * the FP register state on context switch on SMP,
  84                          * there is something wrong if a stopped child appears
  85                          * to still have its FP state in the CPU registers.
  86                          */
  87                         BUG_ON(tsk != current);
  88 #endif
  89                         giveup_fpu(tsk);
  90                 }
  91                 preempt_enable();
  92         }
  93 }
  94
  95 void enable_kernel_fp(void)
  96 {
  97         WARN_ON(preemptible());
  98
  99 #ifdef CONFIG_SMP
 100         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
 101                 giveup_fpu(current);
 102         else
 103                 giveup_fpu(NULL);       /* just enables FP for kernel */
 104 #else
 105         giveup_fpu(last_task_used_math);
 106 #endif /* CONFIG_SMP */
 107 }
 108 EXPORT_SYMBOL(enable_kernel_fp);
 109
 110 #ifdef CONFIG_ALTIVEC
 111 void enable_kernel_altivec(void)
 112 {
 113         WARN_ON(preemptible());
 114
 115 #ifdef CONFIG_SMP
 116         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 117                 giveup_altivec(current);
 118         else
 119                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
 120 #else
 121         giveup_altivec(last_task_used_altivec);
 122 #endif /* CONFIG_SMP */
 123 }
 124 EXPORT_SYMBOL(enable_kernel_altivec);
 125
 126 /*
 127  * Make sure the VMX/Altivec register state in the
 128  * the thread_struct is up to date for task tsk.
 129  */
 130 void flush_altivec_to_thread(struct task_struct *tsk)
 131 {
 132         if (tsk->thread.regs) {
 133                 preempt_disable();
 134                 if (tsk->thread.regs->msr & MSR_VEC) {
 135 #ifdef CONFIG_SMP
 136                         BUG_ON(tsk != current);
 137 #endif
 138                         giveup_altivec(tsk);
 139                 }
 140                 preempt_enable();
 141         }
 142 }
 143 #endif /* CONFIG_ALTIVEC */
 144
 145 #ifdef CONFIG_VSX
 146 #if 0
 147 /* not currently used, but some crazy RAID module might want to later */
 148 void enable_kernel_vsx(void)
 149 {
 150         WARN_ON(preemptible());
 151
 152 #ifdef CONFIG_SMP
 153         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 154                 giveup_vsx(current);
 155         else
 156                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 157 #else
 158         giveup_vsx(last_task_used_vsx);
 159 #endif /* CONFIG_SMP */
 160 }
 161 EXPORT_SYMBOL(enable_kernel_vsx);
 162 #endif
 163
 164 void giveup_vsx(struct task_struct *tsk)
 165 {
 166         giveup_fpu(tsk);
 167         giveup_altivec(tsk);
 168         __giveup_vsx(tsk);
 169 }
 170
 171 void flush_vsx_to_thread(struct task_struct *tsk)
 172 {
 173         if (tsk->thread.regs) {
 174                 preempt_disable();
 175                 if (tsk->thread.regs->msr & MSR_VSX) {
 176 #ifdef CONFIG_SMP
 177                         BUG_ON(tsk != current);
 178 #endif
 179                         giveup_vsx(tsk);
 180                 }
 181                 preempt_enable();
 182         }
 183 }
 184 #endif /* CONFIG_VSX */
 185
 186 #ifdef CONFIG_SPE
 187
 188 void enable_kernel_spe(void)
 189 {
 190         WARN_ON(preemptible());
 191
 192 #ifdef CONFIG_SMP
 193         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 194                 giveup_spe(current);
 195         else
 196                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 197 #else
 198         giveup_spe(last_task_used_spe);
 199 #endif /* __SMP __ */
 200 }
 201 EXPORT_SYMBOL(enable_kernel_spe);
 202
 203 void flush_spe_to_thread(struct task_struct *tsk)
 204 {
 205         if (tsk->thread.regs) {
 206                 preempt_disable();
 207                 if (tsk->thread.regs->msr & MSR_SPE) {
 208 #ifdef CONFIG_SMP
 209                         BUG_ON(tsk != current);
 210 #endif
 211                         giveup_spe(tsk);
 212                 }
 213                 preempt_enable();
 214         }
 215 }
 216 #endif /* CONFIG_SPE */
 217
 218 #ifndef CONFIG_SMP
 219 /*
 220  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 221  * and the current task has some state, discard it.
 222  */
 223 void discard_lazy_cpu_state(void)
 224 {
 225         preempt_disable();
 226         if (last_task_used_math == current)
 227                 last_task_used_math = NULL;
 228 #ifdef CONFIG_ALTIVEC
 229         if (last_task_used_altivec == current)
 230                 last_task_used_altivec = NULL;
 231 #endif /* CONFIG_ALTIVEC */
 232 #ifdef CONFIG_VSX
 233         if (last_task_used_vsx == current)
 234                 last_task_used_vsx = NULL;
 235 #endif /* CONFIG_VSX */
 236 #ifdef CONFIG_SPE
 237         if (last_task_used_spe == current)
 238                 last_task_used_spe = NULL;
 239 #endif
 240         preempt_enable();
 241 }
 242 #endif /* CONFIG_SMP */
 243
 244 void do_dabr(struct pt_regs *regs, unsigned long address,
 245                     unsigned long error_code)
 246 {
 247         siginfo_t info;
 248
 249         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 250                         11, SIGSEGV) == NOTIFY_STOP)
 251                 return;
 252
 253         if (debugger_dabr_match(regs))
 254                 return;
 255
 256         /* Clear the DAC and struct entries.  One shot trigger */
 257 #if defined(CONFIG_BOOKE)
 258         mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~(DBSR_DAC1R | DBSR_DAC1W
 259                                                         | DBCR0_IDM));
 260 #endif
 261
 262         /* Clear the DABR */
 263         set_dabr(0);
 264
 265         /* Deliver the signal to userspace */
 266         info.si_signo = SIGTRAP;
 267         info.si_errno = 0;
 268         info.si_code = TRAP_HWBKPT;
 269         info.si_addr = (void __user *)address;
 270         force_sig_info(SIGTRAP, &info, current);
 271 }
 272
 273 static DEFINE_PER_CPU(unsigned long, current_dabr);
 274
 275 int set_dabr(unsigned long dabr)
 276 {
 277         __get_cpu_var(current_dabr) = dabr;
 278
 279         if (ppc_md.set_dabr)
 280                 return ppc_md.set_dabr(dabr);
 281
 282         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 283 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
 284         mtspr(SPRN_DABR, dabr);
 285 #endif
 286
 287 #if defined(CONFIG_BOOKE)
 288         mtspr(SPRN_DAC1, dabr);
 289 #endif
 290
 291         return 0;
 292 }
 293
 294 #ifdef CONFIG_PPC64
 295 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 296 #endif
 297
 298 struct task_struct *__switch_to(struct task_struct *prev,
 299         struct task_struct *new)
 300 {
 301         struct thread_struct *new_thread, *old_thread;
 302         unsigned long flags;
 303         struct task_struct *last;
 304
 305 #ifdef CONFIG_SMP
 306         /* avoid complexity of lazy save/restore of fpu
 307          * by just saving it every time we switch out if
 308          * this task used the fpu during the last quantum.
 309          *
 310          * If it tries to use the fpu again, it'll trap and
 311          * reload its fp regs.  So we don't have to do a restore
 312          * every switch, just a save.
 313          *  -- Cort
 314          */
 315         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 316                 giveup_fpu(prev);
 317 #ifdef CONFIG_ALTIVEC
 318         /*
 319          * If the previous thread used altivec in the last quantum
 320          * (thus changing altivec regs) then save them.
 321          * We used to check the VRSAVE register but not all apps
 322          * set it, so we don't rely on it now (and in fact we need
 323          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 324          *
 325          * On SMP we always save/restore altivec regs just to avoid the
 326          * complexity of changing processors.
 327          *  -- Cort
 328          */
 329         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 330                 giveup_altivec(prev);
 331 #endif /* CONFIG_ALTIVEC */
 332 #ifdef CONFIG_VSX
 333         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 334                 /* VMX and FPU registers are already save here */
 335                 __giveup_vsx(prev);
 336 #endif /* CONFIG_VSX */
 337 #ifdef CONFIG_SPE
 338         /*
 339          * If the previous thread used spe in the last quantum
 340          * (thus changing spe regs) then save them.
 341          *
 342          * On SMP we always save/restore spe regs just to avoid the
 343          * complexity of changing processors.
 344          */
 345         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 346                 giveup_spe(prev);
 347 #endif /* CONFIG_SPE */
 348
 349 #else  /* CONFIG_SMP */
 350 #ifdef CONFIG_ALTIVEC
 351         /* Avoid the trap.  On smp this this never happens since
 352          * we don't set last_task_used_altivec -- Cort
 353          */
 354         if (new->thread.regs && last_task_used_altivec == new)
 355                 new->thread.regs->msr |= MSR_VEC;
 356 #endif /* CONFIG_ALTIVEC */
 357 #ifdef CONFIG_VSX
 358         if (new->thread.regs && last_task_used_vsx == new)
 359                 new->thread.regs->msr |= MSR_VSX;
 360 #endif /* CONFIG_VSX */
 361 #ifdef CONFIG_SPE
 362         /* Avoid the trap.  On smp this this never happens since
 363          * we don't set last_task_used_spe
 364          */
 365         if (new->thread.regs && last_task_used_spe == new)
 366                 new->thread.regs->msr |= MSR_SPE;
 367 #endif /* CONFIG_SPE */
 368
 369 #endif /* CONFIG_SMP */
 370
 371         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 372                 set_dabr(new->thread.dabr);
 373
 374 #if defined(CONFIG_BOOKE)
 375         /* If new thread DAC (HW breakpoint) is the same then leave it */
 376         if (new->thread.dabr)
 377                 set_dabr(new->thread.dabr);
 378 #endif
 379
 380         new_thread = &new->thread;
 381         old_thread = &current->thread;
 382
 383 #ifdef CONFIG_PPC64
 384         /*
 385          * Collect processor utilization data per process
 386          */
 387         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 388                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 389                 long unsigned start_tb, current_tb;
 390                 start_tb = old_thread->start_tb;
 391                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 392                 old_thread->accum_tb += (current_tb - start_tb);
 393                 new_thread->start_tb = current_tb;
 394         }
 395 #endif
 396
 397         local_irq_save(flags);
 398
 399         account_system_vtime(current);
 400         account_process_vtime(current);
 401         calculate_steal_time();
 402
 403         /*
 404          * We can't take a PMU exception inside _switch() since there is a
 405          * window where the kernel stack SLB and the kernel stack are out
 406          * of sync. Hard disable here.
 407          */
 408         hard_irq_disable();
 409         last = _switch(old_thread, new_thread);
 410
 411         local_irq_restore(flags);
 412
 413         return last;
 414 }
 415
 416 static int instructions_to_print = 16;
 417
 418 static void show_instructions(struct pt_regs *regs)
 419 {
 420         int i;
 421         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 422                         sizeof(int));
 423
 424         printk("Instruction dump:");
 425
 426         for (i = 0; i < instructions_to_print; i++) {
 427                 int instr;
 428
 429                 if (!(i % 8))
 430                         printk("\n");
 431
 432 #if !defined(CONFIG_BOOKE)
 433                 /* If executing with the IMMU off, adjust pc rather
 434                  * than print XXXXXXXX.
 435                  */
 436                 if (!(regs->msr & MSR_IR))
 437                         pc = (unsigned long)phys_to_virt(pc);
 438 #endif
 439
 440                 /* We use __get_user here *only* to avoid an OOPS on a
 441                  * bad address because the pc *should* only be a
 442                  * kernel address.
 443                  */
 444                 if (!__kernel_text_address(pc) ||
 445                      __get_user(instr, (unsigned int __user *)pc)) {
 446                         printk("XXXXXXXX ");
 447                 } else {
 448                         if (regs->nip == pc)
 449                                 printk("<%08x> ", instr);
 450                         else
 451                                 printk("%08x ", instr);
 452                 }
 453
 454                 pc += sizeof(int);
 455         }
 456
 457         printk("\n");
 458 }
 459
 460 static struct regbit {
 461         unsigned long bit;
 462         const char *name;
 463 } msr_bits[] = {
 464         {MSR_EE,        "EE"},
 465         {MSR_PR,        "PR"},
 466         {MSR_FP,        "FP"},
 467         {MSR_VEC,       "VEC"},
 468         {MSR_VSX,       "VSX"},
 469         {MSR_ME,        "ME"},
 470         {MSR_IR,        "IR"},
 471         {MSR_DR,        "DR"},
 472         {0,             NULL}
 473 };
 474
 475 static void printbits(unsigned long val, struct regbit *bits)
 476 {
 477         const char *sep = "";
 478
 479         printk("<");
 480         for (; bits->bit; ++bits)
 481                 if (val & bits->bit) {
 482                         printk("%s%s", sep, bits->name);
 483                         sep = ",";
 484                 }
 485         printk(">");
 486 }
 487
 488 #ifdef CONFIG_PPC64
 489 #define REG             "%016lx"
 490 #define REGS_PER_LINE   4
 491 #define LAST_VOLATILE   13
 492 #else
 493 #define REG             "%08lx"
 494 #define REGS_PER_LINE   8
 495 #define LAST_VOLATILE   12
 496 #endif
 497
 498 void show_regs(struct pt_regs * regs)
 499 {
 500         int i, trap;
 501
 502         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 503                regs->nip, regs->link, regs->ctr);
 504         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 505                regs, regs->trap, print_tainted(), init_utsname()->release);
 506         printk("MSR: "REG" ", regs->msr);
 507         printbits(regs->msr, msr_bits);
 508         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 509         trap = TRAP(regs);
 510         if (trap == 0x300 || trap == 0x600)
 511 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 512                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 513 #else
 514                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 515 #endif
 516         printk("TASK = %p[%d] '%s' THREAD: %p",
 517                current, task_pid_nr(current), current->comm, task_thread_info(current));
 518
 519 #ifdef CONFIG_SMP
 520         printk(" CPU: %d", raw_smp_processor_id());
 521 #endif /* CONFIG_SMP */
 522
 523         for (i = 0;  i < 32;  i++) {
 524                 if ((i % REGS_PER_LINE) == 0)
 525                         printk("\n" KERN_INFO "GPR%02d: ", i);
 526                 printk(REG " ", regs->gpr[i]);
 527                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 528                         break;
 529         }
 530         printk("\n");
 531 #ifdef CONFIG_KALLSYMS
 532         /*
 533          * Lookup NIP late so we have the best change of getting the
 534          * above info out without failing
 535          */
 536         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 537         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 538 #endif
 539         show_stack(current, (unsigned long *) regs->gpr[1]);
 540         if (!user_mode(regs))
 541                 show_instructions(regs);
 542 }
 543
 544 void exit_thread(void)
 545 {
 546         discard_lazy_cpu_state();
 547 }
 548
 549 void flush_thread(void)
 550 {
 551 #ifdef CONFIG_PPC64
 552         struct thread_info *t = current_thread_info();
 553
 554         if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
 555                 clear_ti_thread_flag(t, TIF_ABI_PENDING);
 556                 if (test_ti_thread_flag(t, TIF_32BIT))
 557                         clear_ti_thread_flag(t, TIF_32BIT);
 558                 else
 559                         set_ti_thread_flag(t, TIF_32BIT);
 560         }
 561 #endif
 562
 563         discard_lazy_cpu_state();
 564
 565         if (current->thread.dabr) {
 566                 current->thread.dabr = 0;
 567                 set_dabr(0);
 568
 569 #if defined(CONFIG_BOOKE)
 570                 current->thread.dbcr0 &= ~(DBSR_DAC1R | DBSR_DAC1W);
 571 #endif
 572         }
 573 }
 574
 575 void
 576 release_thread(struct task_struct *t)
 577 {
 578 }
 579
 580 /*
 581  * This gets called before we allocate a new thread and copy
 582  * the current task into it.
 583  */
 584 void prepare_to_copy(struct task_struct *tsk)
 585 {
 586         flush_fp_to_thread(current);
 587         flush_altivec_to_thread(current);
 588         flush_vsx_to_thread(current);
 589         flush_spe_to_thread(current);
 590 }
 591
 592 /*
 593  * Copy a thread..
 594  */
 595 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 596                 unsigned long unused, struct task_struct *p,
 597                 struct pt_regs *regs)
 598 {
 599         struct pt_regs *childregs, *kregs;
 600         extern void ret_from_fork(void);
 601         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 602
 603         CHECK_FULL_REGS(regs);
 604         /* Copy registers */
 605         sp -= sizeof(struct pt_regs);
 606         childregs = (struct pt_regs *) sp;
 607         *childregs = *regs;
 608         if ((childregs->msr & MSR_PR) == 0) {
 609                 /* for kernel thread, set `current' and stackptr in new task */
 610                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 611 #ifdef CONFIG_PPC32
 612                 childregs->gpr[2] = (unsigned long) p;
 613 #else
 614                 clear_tsk_thread_flag(p, TIF_32BIT);
 615 #endif
 616                 p->thread.regs = NULL;  /* no user register state */
 617         } else {
 618                 childregs->gpr[1] = usp;
 619                 p->thread.regs = childregs;
 620                 if (clone_flags & CLONE_SETTLS) {
 621 #ifdef CONFIG_PPC64
 622                         if (!test_thread_flag(TIF_32BIT))
 623                                 childregs->gpr[13] = childregs->gpr[6];
 624                         else
 625 #endif
 626                                 childregs->gpr[2] = childregs->gpr[6];
 627                 }
 628         }
 629         childregs->gpr[3] = 0;  /* Result from fork() */
 630         sp -= STACK_FRAME_OVERHEAD;
 631
 632         /*
 633          * The way this works is that at some point in the future
 634          * some task will call _switch to switch to the new task.
 635          * That will pop off the stack frame created below and start
 636          * the new task running at ret_from_fork.  The new task will
 637          * do some house keeping and then return from the fork or clone
 638          * system call, using the stack frame created above.
 639          */
 640         sp -= sizeof(struct pt_regs);
 641         kregs = (struct pt_regs *) sp;
 642         sp -= STACK_FRAME_OVERHEAD;
 643         p->thread.ksp = sp;
 644         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 645                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 646
 647 #ifdef CONFIG_PPC64
 648         if (cpu_has_feature(CPU_FTR_SLB)) {
 649                 unsigned long sp_vsid;
 650                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 651
 652                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
 653                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 654                                 << SLB_VSID_SHIFT_1T;
 655                 else
 656                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 657                                 << SLB_VSID_SHIFT;
 658                 sp_vsid |= SLB_VSID_KERNEL | llp;
 659                 p->thread.ksp_vsid = sp_vsid;
 660         }
 661
 662         /*
 663          * The PPC64 ABI makes use of a TOC to contain function
 664          * pointers.  The function (ret_from_except) is actually a pointer
 665          * to the TOC entry.  The first entry is a pointer to the actual
 666          * function.
 667          */
 668         kregs->nip = *((unsigned long *)ret_from_fork);
 669 #else
 670         kregs->nip = (unsigned long)ret_from_fork;
 671 #endif
 672
 673         return 0;
 674 }
 675
 676 /*
 677  * Set up a thread for executing a new program
 678  */
 679 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 680 {
 681 #ifdef CONFIG_PPC64
 682         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 683 #endif
 684
 685         set_fs(USER_DS);
 686
 687         /*
 688          * If we exec out of a kernel thread then thread.regs will not be
 689          * set.  Do it now.
 690          */
 691         if (!current->thread.regs) {
 692                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 693                 current->thread.regs = regs - 1;
 694         }
 695
 696         memset(regs->gpr, 0, sizeof(regs->gpr));
 697         regs->ctr = 0;
 698         regs->link = 0;
 699         regs->xer = 0;
 700         regs->ccr = 0;
 701         regs->gpr[1] = sp;
 702
 703         /*
 704          * We have just cleared all the nonvolatile GPRs, so make
 705          * FULL_REGS(regs) return true.  This is necessary to allow
 706          * ptrace to examine the thread immediately after exec.
 707          */
 708         regs->trap &= ~1UL;
 709
 710 #ifdef CONFIG_PPC32
 711         regs->mq = 0;
 712         regs->nip = start;
 713         regs->msr = MSR_USER;
 714 #else
 715         if (!test_thread_flag(TIF_32BIT)) {
 716                 unsigned long entry, toc;
 717
 718                 /* start is a relocated pointer to the function descriptor for
 719                  * the elf _start routine.  The first entry in the function
 720                  * descriptor is the entry address of _start and the second
 721                  * entry is the TOC value we need to use.
 722                  */
 723                 __get_user(entry, (unsigned long __user *)start);
 724                 __get_user(toc, (unsigned long __user *)start+1);
 725
 726                 /* Check whether the e_entry function descriptor entries
 727                  * need to be relocated before we can use them.
 728                  */
 729                 if (load_addr != 0) {
 730                         entry += load_addr;
 731                         toc   += load_addr;
 732                 }
 733                 regs->nip = entry;
 734                 regs->gpr[2] = toc;
 735                 regs->msr = MSR_USER64;
 736         } else {
 737                 regs->nip = start;
 738                 regs->gpr[2] = 0;
 739                 regs->msr = MSR_USER32;
 740         }
 741 #endif
 742
 743         discard_lazy_cpu_state();
 744 #ifdef CONFIG_VSX
 745         current->thread.used_vsr = 0;
 746 #endif
 747         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 748         current->thread.fpscr.val = 0;
 749 #ifdef CONFIG_ALTIVEC
 750         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 751         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 752         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 753         current->thread.vrsave = 0;
 754         current->thread.used_vr = 0;
 755 #endif /* CONFIG_ALTIVEC */
 756 #ifdef CONFIG_SPE
 757         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 758         current->thread.acc = 0;
 759         current->thread.spefscr = 0;
 760         current->thread.used_spe = 0;
 761 #endif /* CONFIG_SPE */
 762 }
 763
 764 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 765                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 766
 767 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 768 {
 769         struct pt_regs *regs = tsk->thread.regs;
 770
 771         /* This is a bit hairy.  If we are an SPE enabled  processor
 772          * (have embedded fp) we store the IEEE exception enable flags in
 773          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 774          * mode (asyn, precise, disabled) for 'Classic' FP. */
 775         if (val & PR_FP_EXC_SW_ENABLE) {
 776 #ifdef CONFIG_SPE
 777                 if (cpu_has_feature(CPU_FTR_SPE)) {
 778                         tsk->thread.fpexc_mode = val &
 779                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 780                         return 0;
 781                 } else {
 782                         return -EINVAL;
 783                 }
 784 #else
 785                 return -EINVAL;
 786 #endif
 787         }
 788
 789         /* on a CONFIG_SPE this does not hurt us.  The bits that
 790          * __pack_fe01 use do not overlap with bits used for
 791          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 792          * on CONFIG_SPE implementations are reserved so writing to
 793          * them does not change anything */
 794         if (val > PR_FP_EXC_PRECISE)
 795                 return -EINVAL;
 796         tsk->thread.fpexc_mode = __pack_fe01(val);
 797         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 798                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 799                         | tsk->thread.fpexc_mode;
 800         return 0;
 801 }
 802
 803 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 804 {
 805         unsigned int val;
 806
 807         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 808 #ifdef CONFIG_SPE
 809                 if (cpu_has_feature(CPU_FTR_SPE))
 810                         val = tsk->thread.fpexc_mode;
 811                 else
 812                         return -EINVAL;
 813 #else
 814                 return -EINVAL;
 815 #endif
 816         else
 817                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 818         return put_user(val, (unsigned int __user *) adr);
 819 }
 820
 821 int set_endian(struct task_struct *tsk, unsigned int val)
 822 {
 823         struct pt_regs *regs = tsk->thread.regs;
 824
 825         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 826             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 827                 return -EINVAL;
 828
 829         if (regs == NULL)
 830                 return -EINVAL;
 831
 832         if (val == PR_ENDIAN_BIG)
 833                 regs->msr &= ~MSR_LE;
 834         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 835                 regs->msr |= MSR_LE;
 836         else
 837                 return -EINVAL;
 838
 839         return 0;
 840 }
 841
 842 int get_endian(struct task_struct *tsk, unsigned long adr)
 843 {
 844         struct pt_regs *regs = tsk->thread.regs;
 845         unsigned int val;
 846
 847         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 848             !cpu_has_feature(CPU_FTR_REAL_LE))
 849                 return -EINVAL;
 850
 851         if (regs == NULL)
 852                 return -EINVAL;
 853
 854         if (regs->msr & MSR_LE) {
 855                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 856                         val = PR_ENDIAN_LITTLE;
 857                 else
 858                         val = PR_ENDIAN_PPC_LITTLE;
 859         } else
 860                 val = PR_ENDIAN_BIG;
 861
 862         return put_user(val, (unsigned int __user *)adr);
 863 }
 864
 865 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 866 {
 867         tsk->thread.align_ctl = val;
 868         return 0;
 869 }
 870
 871 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 872 {
 873         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 874 }
 875
 876 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 877
 878 int sys_clone(unsigned long clone_flags, unsigned long usp,
 879               int __user *parent_tidp, void __user *child_threadptr,
 880               int __user *child_tidp, int p6,
 881               struct pt_regs *regs)
 882 {
 883         CHECK_FULL_REGS(regs);
 884         if (usp == 0)
 885                 usp = regs->gpr[1];     /* stack pointer for child */
 886 #ifdef CONFIG_PPC64
 887         if (test_thread_flag(TIF_32BIT)) {
 888                 parent_tidp = TRUNC_PTR(parent_tidp);
 889                 child_tidp = TRUNC_PTR(child_tidp);
 890         }
 891 #endif
 892         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 893 }
 894
 895 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 896              unsigned long p4, unsigned long p5, unsigned long p6,
 897              struct pt_regs *regs)
 898 {
 899         CHECK_FULL_REGS(regs);
 900         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 901 }
 902
 903 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
 904               unsigned long p4, unsigned long p5, unsigned long p6,
 905               struct pt_regs *regs)
 906 {
 907         CHECK_FULL_REGS(regs);
 908         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 909                         regs, 0, NULL, NULL);
 910 }
 911
 912 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 913                unsigned long a3, unsigned long a4, unsigned long a5,
 914                struct pt_regs *regs)
 915 {
 916         int error;
 917         char *filename;
 918
 919         filename = getname((char __user *) a0);
 920         error = PTR_ERR(filename);
 921         if (IS_ERR(filename))
 922                 goto out;
 923         flush_fp_to_thread(current);
 924         flush_altivec_to_thread(current);
 925         flush_spe_to_thread(current);
 926         error = do_execve(filename, (char __user * __user *) a1,
 927                           (char __user * __user *) a2, regs);
 928         putname(filename);
 929 out:
 930         return error;
 931 }
 932
 933 #ifdef CONFIG_IRQSTACKS
 934 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
 935                                   unsigned long nbytes)
 936 {
 937         unsigned long stack_page;
 938         unsigned long cpu = task_cpu(p);
 939
 940         /*
 941          * Avoid crashing if the stack has overflowed and corrupted
 942          * task_cpu(p), which is in the thread_info struct.
 943          */
 944         if (cpu < NR_CPUS && cpu_possible(cpu)) {
 945                 stack_page = (unsigned long) hardirq_ctx[cpu];
 946                 if (sp >= stack_page + sizeof(struct thread_struct)
 947                     && sp <= stack_page + THREAD_SIZE - nbytes)
 948                         return 1;
 949
 950                 stack_page = (unsigned long) softirq_ctx[cpu];
 951                 if (sp >= stack_page + sizeof(struct thread_struct)
 952                     && sp <= stack_page + THREAD_SIZE - nbytes)
 953                         return 1;
 954         }
 955         return 0;
 956 }
 957
 958 #else
 959 #define valid_irq_stack(sp, p, nb)      0
 960 #endif /* CONFIG_IRQSTACKS */
 961
 962 int validate_sp(unsigned long sp, struct task_struct *p,
 963                        unsigned long nbytes)
 964 {
 965         unsigned long stack_page = (unsigned long)task_stack_page(p);
 966
 967         if (sp >= stack_page + sizeof(struct thread_struct)
 968             && sp <= stack_page + THREAD_SIZE - nbytes)
 969                 return 1;
 970
 971         return valid_irq_stack(sp, p, nbytes);
 972 }
 973
 974 EXPORT_SYMBOL(validate_sp);
 975
 976 unsigned long get_wchan(struct task_struct *p)
 977 {
 978         unsigned long ip, sp;
 979         int count = 0;
 980
 981         if (!p || p == current || p->state == TASK_RUNNING)
 982                 return 0;
 983
 984         sp = p->thread.ksp;
 985         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 986                 return 0;
 987
 988         do {
 989                 sp = *(unsigned long *)sp;
 990                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
 991                         return 0;
 992                 if (count > 0) {
 993                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
 994                         if (!in_sched_functions(ip))
 995                                 return ip;
 996                 }
 997         } while (count++ < 16);
 998         return 0;
 999 }
1000
1001 static int kstack_depth_to_print = 64;
1002
1003 void show_stack(struct task_struct *tsk, unsigned long *stack)
1004 {
1005         unsigned long sp, ip, lr, newsp;
1006         int count = 0;
1007         int firstframe = 1;
1008
1009         sp = (unsigned long) stack;
1010         if (tsk == NULL)
1011                 tsk = current;
1012         if (sp == 0) {
1013                 if (tsk == current)
1014                         asm("mr %0,1" : "=r" (sp));
1015                 else
1016                         sp = tsk->thread.ksp;
1017         }
1018
1019         lr = 0;
1020         printk("Call Trace:\n");
1021         do {
1022                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1023                         return;
1024
1025                 stack = (unsigned long *) sp;
1026                 newsp = stack[0];
1027                 ip = stack[STACK_FRAME_LR_SAVE];
1028                 if (!firstframe || ip != lr) {
1029                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1030                         if (firstframe)
1031                                 printk(" (unreliable)");
1032                         printk("\n");
1033                 }
1034                 firstframe = 0;
1035
1036                 /*
1037                  * See if this is an exception frame.
1038                  * We look for the "regshere" marker in the current frame.
1039                  */
1040                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1041                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1042                         struct pt_regs *regs = (struct pt_regs *)
1043                                 (sp + STACK_FRAME_OVERHEAD);
1044                         lr = regs->link;
1045                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1046                                regs->trap, (void *)regs->nip, (void *)lr);
1047                         firstframe = 1;
1048                 }
1049
1050                 sp = newsp;
1051         } while (count++ < kstack_depth_to_print);
1052 }
1053
1054 void dump_stack(void)
1055 {
1056         show_stack(current, NULL);
1057 }
1058 EXPORT_SYMBOL(dump_stack);
1059
1060 #ifdef CONFIG_PPC64
1061 void ppc64_runlatch_on(void)
1062 {
1063         unsigned long ctrl;
1064
1065         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1066                 HMT_medium();
1067
1068                 ctrl = mfspr(SPRN_CTRLF);
1069                 ctrl |= CTRL_RUNLATCH;
1070                 mtspr(SPRN_CTRLT, ctrl);
1071
1072                 set_thread_flag(TIF_RUNLATCH);
1073         }
1074 }
1075
1076 void ppc64_runlatch_off(void)
1077 {
1078         unsigned long ctrl;
1079
1080         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1081                 HMT_medium();
1082
1083                 clear_thread_flag(TIF_RUNLATCH);
1084
1085                 ctrl = mfspr(SPRN_CTRLF);
1086                 ctrl &= ~CTRL_RUNLATCH;
1087                 mtspr(SPRN_CTRLT, ctrl);
1088         }
1089 }
1090 #endif
1091
1092 #if THREAD_SHIFT < PAGE_SHIFT
1093
1094 static struct kmem_cache *thread_info_cache;
1095
1096 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1097 {
1098         struct thread_info *ti;
1099
1100         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1101         if (unlikely(ti == NULL))
1102                 return NULL;
1103 #ifdef CONFIG_DEBUG_STACK_USAGE
1104         memset(ti, 0, THREAD_SIZE);
1105 #endif
1106         return ti;
1107 }
1108
1109 void free_thread_info(struct thread_info *ti)
1110 {
1111         kmem_cache_free(thread_info_cache, ti);
1112 }
1113
1114 void thread_info_cache_init(void)
1115 {
1116         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1117                                               THREAD_SIZE, 0, NULL);
1118         BUG_ON(thread_info_cache == NULL);
1119 }
1120
1121 #endif /* THREAD_SHIFT < PAGE_SHIFT */