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
  51 extern unsigned long _get_SP(void);
  52
  53 #ifndef CONFIG_SMP
  54 struct task_struct *last_task_used_math = NULL;
  55 struct task_struct *last_task_used_altivec = NULL;
  56 struct task_struct *last_task_used_spe = NULL;
  57 #endif
  58
  59 /*
  60  * Make sure the floating-point register state in the
  61  * the thread_struct is up to date for task tsk.
  62  */
  63 void flush_fp_to_thread(struct task_struct *tsk)
  64 {
  65         if (tsk->thread.regs) {
  66                 /*
  67                  * We need to disable preemption here because if we didn't,
  68                  * another process could get scheduled after the regs->msr
  69                  * test but before we have finished saving the FP registers
  70                  * to the thread_struct.  That process could take over the
  71                  * FPU, and then when we get scheduled again we would store
  72                  * bogus values for the remaining FP registers.
  73                  */
  74                 preempt_disable();
  75                 if (tsk->thread.regs->msr & MSR_FP) {
  76 #ifdef CONFIG_SMP
  77                         /*
  78                          * This should only ever be called for current or
  79                          * for a stopped child process.  Since we save away
  80                          * the FP register state on context switch on SMP,
  81                          * there is something wrong if a stopped child appears
  82                          * to still have its FP state in the CPU registers.
  83                          */
  84                         BUG_ON(tsk != current);
  85 #endif
  86                         giveup_fpu(tsk);
  87                 }
  88                 preempt_enable();
  89         }
  90 }
  91
  92 void enable_kernel_fp(void)
  93 {
  94         WARN_ON(preemptible());
  95
  96 #ifdef CONFIG_SMP
  97         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
  98                 giveup_fpu(current);
  99         else
 100                 giveup_fpu(NULL);       /* just enables FP for kernel */
 101 #else
 102         giveup_fpu(last_task_used_math);
 103 #endif /* CONFIG_SMP */
 104 }
 105 EXPORT_SYMBOL(enable_kernel_fp);
 106
 107 int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs)
 108 {
 109         if (!tsk->thread.regs)
 110                 return 0;
 111         flush_fp_to_thread(current);
 112
 113         memcpy(fpregs, &tsk->thread.fpr[0], sizeof(*fpregs));
 114
 115         return 1;
 116 }
 117
 118 #ifdef CONFIG_ALTIVEC
 119 void enable_kernel_altivec(void)
 120 {
 121         WARN_ON(preemptible());
 122
 123 #ifdef CONFIG_SMP
 124         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 125                 giveup_altivec(current);
 126         else
 127                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
 128 #else
 129         giveup_altivec(last_task_used_altivec);
 130 #endif /* CONFIG_SMP */
 131 }
 132 EXPORT_SYMBOL(enable_kernel_altivec);
 133
 134 /*
 135  * Make sure the VMX/Altivec register state in the
 136  * the thread_struct is up to date for task tsk.
 137  */
 138 void flush_altivec_to_thread(struct task_struct *tsk)
 139 {
 140         if (tsk->thread.regs) {
 141                 preempt_disable();
 142                 if (tsk->thread.regs->msr & MSR_VEC) {
 143 #ifdef CONFIG_SMP
 144                         BUG_ON(tsk != current);
 145 #endif
 146                         giveup_altivec(tsk);
 147                 }
 148                 preempt_enable();
 149         }
 150 }
 151
 152 int dump_task_altivec(struct pt_regs *regs, elf_vrregset_t *vrregs)
 153 {
 154         flush_altivec_to_thread(current);
 155         memcpy(vrregs, &current->thread.vr[0], sizeof(*vrregs));
 156         return 1;
 157 }
 158 #endif /* CONFIG_ALTIVEC */
 159
 160 #ifdef CONFIG_SPE
 161
 162 void enable_kernel_spe(void)
 163 {
 164         WARN_ON(preemptible());
 165
 166 #ifdef CONFIG_SMP
 167         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 168                 giveup_spe(current);
 169         else
 170                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 171 #else
 172         giveup_spe(last_task_used_spe);
 173 #endif /* __SMP __ */
 174 }
 175 EXPORT_SYMBOL(enable_kernel_spe);
 176
 177 void flush_spe_to_thread(struct task_struct *tsk)
 178 {
 179         if (tsk->thread.regs) {
 180                 preempt_disable();
 181                 if (tsk->thread.regs->msr & MSR_SPE) {
 182 #ifdef CONFIG_SMP
 183                         BUG_ON(tsk != current);
 184 #endif
 185                         giveup_spe(tsk);
 186                 }
 187                 preempt_enable();
 188         }
 189 }
 190
 191 int dump_spe(struct pt_regs *regs, elf_vrregset_t *evrregs)
 192 {
 193         flush_spe_to_thread(current);
 194         /* We copy u32 evr[32] + u64 acc + u32 spefscr -> 35 */
 195         memcpy(evrregs, &current->thread.evr[0], sizeof(u32) * 35);
 196         return 1;
 197 }
 198 #endif /* CONFIG_SPE */
 199
 200 #ifndef CONFIG_SMP
 201 /*
 202  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 203  * and the current task has some state, discard it.
 204  */
 205 void discard_lazy_cpu_state(void)
 206 {
 207         preempt_disable();
 208         if (last_task_used_math == current)
 209                 last_task_used_math = NULL;
 210 #ifdef CONFIG_ALTIVEC
 211         if (last_task_used_altivec == current)
 212                 last_task_used_altivec = NULL;
 213 #endif /* CONFIG_ALTIVEC */
 214 #ifdef CONFIG_SPE
 215         if (last_task_used_spe == current)
 216                 last_task_used_spe = NULL;
 217 #endif
 218         preempt_enable();
 219 }
 220 #endif /* CONFIG_SMP */
 221
 222 int set_dabr(unsigned long dabr)
 223 {
 224 #ifdef CONFIG_PPC_MERGE         /* XXX for now */
 225         if (ppc_md.set_dabr)
 226                 return ppc_md.set_dabr(dabr);
 227 #endif
 228
 229         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 230 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
 231         mtspr(SPRN_DABR, dabr);
 232 #endif
 233         return 0;
 234 }
 235
 236 #ifdef CONFIG_PPC64
 237 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 238 #endif
 239
 240 static DEFINE_PER_CPU(unsigned long, current_dabr);
 241
 242 struct task_struct *__switch_to(struct task_struct *prev,
 243         struct task_struct *new)
 244 {
 245         struct thread_struct *new_thread, *old_thread;
 246         unsigned long flags;
 247         struct task_struct *last;
 248
 249 #ifdef CONFIG_SMP
 250         /* avoid complexity of lazy save/restore of fpu
 251          * by just saving it every time we switch out if
 252          * this task used the fpu during the last quantum.
 253          *
 254          * If it tries to use the fpu again, it'll trap and
 255          * reload its fp regs.  So we don't have to do a restore
 256          * every switch, just a save.
 257          *  -- Cort
 258          */
 259         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 260                 giveup_fpu(prev);
 261 #ifdef CONFIG_ALTIVEC
 262         /*
 263          * If the previous thread used altivec in the last quantum
 264          * (thus changing altivec regs) then save them.
 265          * We used to check the VRSAVE register but not all apps
 266          * set it, so we don't rely on it now (and in fact we need
 267          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 268          *
 269          * On SMP we always save/restore altivec regs just to avoid the
 270          * complexity of changing processors.
 271          *  -- Cort
 272          */
 273         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 274                 giveup_altivec(prev);
 275 #endif /* CONFIG_ALTIVEC */
 276 #ifdef CONFIG_SPE
 277         /*
 278          * If the previous thread used spe in the last quantum
 279          * (thus changing spe regs) then save them.
 280          *
 281          * On SMP we always save/restore spe regs just to avoid the
 282          * complexity of changing processors.
 283          */
 284         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 285                 giveup_spe(prev);
 286 #endif /* CONFIG_SPE */
 287
 288 #else  /* CONFIG_SMP */
 289 #ifdef CONFIG_ALTIVEC
 290         /* Avoid the trap.  On smp this this never happens since
 291          * we don't set last_task_used_altivec -- Cort
 292          */
 293         if (new->thread.regs && last_task_used_altivec == new)
 294                 new->thread.regs->msr |= MSR_VEC;
 295 #endif /* CONFIG_ALTIVEC */
 296 #ifdef CONFIG_SPE
 297         /* Avoid the trap.  On smp this this never happens since
 298          * we don't set last_task_used_spe
 299          */
 300         if (new->thread.regs && last_task_used_spe == new)
 301                 new->thread.regs->msr |= MSR_SPE;
 302 #endif /* CONFIG_SPE */
 303
 304 #endif /* CONFIG_SMP */
 305
 306         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr)) {
 307                 set_dabr(new->thread.dabr);
 308                 __get_cpu_var(current_dabr) = new->thread.dabr;
 309         }
 310
 311         new_thread = &new->thread;
 312         old_thread = &current->thread;
 313
 314 #ifdef CONFIG_PPC64
 315         /*
 316          * Collect processor utilization data per process
 317          */
 318         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 319                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 320                 long unsigned start_tb, current_tb;
 321                 start_tb = old_thread->start_tb;
 322                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 323                 old_thread->accum_tb += (current_tb - start_tb);
 324                 new_thread->start_tb = current_tb;
 325         }
 326 #endif
 327
 328         local_irq_save(flags);
 329
 330         account_system_vtime(current);
 331         account_process_vtime(current);
 332         calculate_steal_time();
 333
 334         last = _switch(old_thread, new_thread);
 335
 336         local_irq_restore(flags);
 337
 338         return last;
 339 }
 340
 341 static int instructions_to_print = 16;
 342
 343 static void show_instructions(struct pt_regs *regs)
 344 {
 345         int i;
 346         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 347                         sizeof(int));
 348
 349         printk("Instruction dump:");
 350
 351         for (i = 0; i < instructions_to_print; i++) {
 352                 int instr;
 353
 354                 if (!(i % 8))
 355                         printk("\n");
 356
 357                 /* We use __get_user here *only* to avoid an OOPS on a
 358                  * bad address because the pc *should* only be a
 359                  * kernel address.
 360                  */
 361                 if (!__kernel_text_address(pc) ||
 362                      __get_user(instr, (unsigned int __user *)pc)) {
 363                         printk("XXXXXXXX ");
 364                 } else {
 365                         if (regs->nip == pc)
 366                                 printk("<%08x> ", instr);
 367                         else
 368                                 printk("%08x ", instr);
 369                 }
 370
 371                 pc += sizeof(int);
 372         }
 373
 374         printk("\n");
 375 }
 376
 377 static struct regbit {
 378         unsigned long bit;
 379         const char *name;
 380 } msr_bits[] = {
 381         {MSR_EE,        "EE"},
 382         {MSR_PR,        "PR"},
 383         {MSR_FP,        "FP"},
 384         {MSR_ME,        "ME"},
 385         {MSR_IR,        "IR"},
 386         {MSR_DR,        "DR"},
 387         {0,             NULL}
 388 };
 389
 390 static void printbits(unsigned long val, struct regbit *bits)
 391 {
 392         const char *sep = "";
 393
 394         printk("<");
 395         for (; bits->bit; ++bits)
 396                 if (val & bits->bit) {
 397                         printk("%s%s", sep, bits->name);
 398                         sep = ",";
 399                 }
 400         printk(">");
 401 }
 402
 403 #ifdef CONFIG_PPC64
 404 #define REG             "%016lx"
 405 #define REGS_PER_LINE   4
 406 #define LAST_VOLATILE   13
 407 #else
 408 #define REG             "%08lx"
 409 #define REGS_PER_LINE   8
 410 #define LAST_VOLATILE   12
 411 #endif
 412
 413 void show_regs(struct pt_regs * regs)
 414 {
 415         int i, trap;
 416
 417         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 418                regs->nip, regs->link, regs->ctr);
 419         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 420                regs, regs->trap, print_tainted(), init_utsname()->release);
 421         printk("MSR: "REG" ", regs->msr);
 422         printbits(regs->msr, msr_bits);
 423         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 424         trap = TRAP(regs);
 425         if (trap == 0x300 || trap == 0x600)
 426 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 427                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 428 #else
 429                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 430 #endif
 431         printk("TASK = %p[%d] '%s' THREAD: %p",
 432                current, current->pid, current->comm, task_thread_info(current));
 433
 434 #ifdef CONFIG_SMP
 435         printk(" CPU: %d", smp_processor_id());
 436 #endif /* CONFIG_SMP */
 437
 438         for (i = 0;  i < 32;  i++) {
 439                 if ((i % REGS_PER_LINE) == 0)
 440                         printk("\n" KERN_INFO "GPR%02d: ", i);
 441                 printk(REG " ", regs->gpr[i]);
 442                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 443                         break;
 444         }
 445         printk("\n");
 446 #ifdef CONFIG_KALLSYMS
 447         /*
 448          * Lookup NIP late so we have the best change of getting the
 449          * above info out without failing
 450          */
 451         printk("NIP ["REG"] ", regs->nip);
 452         print_symbol("%s\n", regs->nip);
 453         printk("LR ["REG"] ", regs->link);
 454         print_symbol("%s\n", regs->link);
 455 #endif
 456         show_stack(current, (unsigned long *) regs->gpr[1]);
 457         if (!user_mode(regs))
 458                 show_instructions(regs);
 459 }
 460
 461 void exit_thread(void)
 462 {
 463         discard_lazy_cpu_state();
 464 }
 465
 466 void flush_thread(void)
 467 {
 468 #ifdef CONFIG_PPC64
 469         struct thread_info *t = current_thread_info();
 470
 471         if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
 472                 clear_ti_thread_flag(t, TIF_ABI_PENDING);
 473                 if (test_ti_thread_flag(t, TIF_32BIT))
 474                         clear_ti_thread_flag(t, TIF_32BIT);
 475                 else
 476                         set_ti_thread_flag(t, TIF_32BIT);
 477         }
 478 #endif
 479
 480         discard_lazy_cpu_state();
 481
 482         if (current->thread.dabr) {
 483                 current->thread.dabr = 0;
 484                 set_dabr(0);
 485         }
 486 }
 487
 488 void
 489 release_thread(struct task_struct *t)
 490 {
 491 }
 492
 493 /*
 494  * This gets called before we allocate a new thread and copy
 495  * the current task into it.
 496  */
 497 void prepare_to_copy(struct task_struct *tsk)
 498 {
 499         flush_fp_to_thread(current);
 500         flush_altivec_to_thread(current);
 501         flush_spe_to_thread(current);
 502 }
 503
 504 /*
 505  * Copy a thread..
 506  */
 507 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 508                 unsigned long unused, struct task_struct *p,
 509                 struct pt_regs *regs)
 510 {
 511         struct pt_regs *childregs, *kregs;
 512         extern void ret_from_fork(void);
 513         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 514
 515         CHECK_FULL_REGS(regs);
 516         /* Copy registers */
 517         sp -= sizeof(struct pt_regs);
 518         childregs = (struct pt_regs *) sp;
 519         *childregs = *regs;
 520         if ((childregs->msr & MSR_PR) == 0) {
 521                 /* for kernel thread, set `current' and stackptr in new task */
 522                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 523 #ifdef CONFIG_PPC32
 524                 childregs->gpr[2] = (unsigned long) p;
 525 #else
 526                 clear_tsk_thread_flag(p, TIF_32BIT);
 527 #endif
 528                 p->thread.regs = NULL;  /* no user register state */
 529         } else {
 530                 childregs->gpr[1] = usp;
 531                 p->thread.regs = childregs;
 532                 if (clone_flags & CLONE_SETTLS) {
 533 #ifdef CONFIG_PPC64
 534                         if (!test_thread_flag(TIF_32BIT))
 535                                 childregs->gpr[13] = childregs->gpr[6];
 536                         else
 537 #endif
 538                                 childregs->gpr[2] = childregs->gpr[6];
 539                 }
 540         }
 541         childregs->gpr[3] = 0;  /* Result from fork() */
 542         sp -= STACK_FRAME_OVERHEAD;
 543
 544         /*
 545          * The way this works is that at some point in the future
 546          * some task will call _switch to switch to the new task.
 547          * That will pop off the stack frame created below and start
 548          * the new task running at ret_from_fork.  The new task will
 549          * do some house keeping and then return from the fork or clone
 550          * system call, using the stack frame created above.
 551          */
 552         sp -= sizeof(struct pt_regs);
 553         kregs = (struct pt_regs *) sp;
 554         sp -= STACK_FRAME_OVERHEAD;
 555         p->thread.ksp = sp;
 556
 557 #ifdef CONFIG_PPC64
 558         if (cpu_has_feature(CPU_FTR_SLB)) {
 559                 unsigned long sp_vsid = get_kernel_vsid(sp);
 560                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 561
 562                 sp_vsid <<= SLB_VSID_SHIFT;
 563                 sp_vsid |= SLB_VSID_KERNEL | llp;
 564                 p->thread.ksp_vsid = sp_vsid;
 565         }
 566
 567         /*
 568          * The PPC64 ABI makes use of a TOC to contain function
 569          * pointers.  The function (ret_from_except) is actually a pointer
 570          * to the TOC entry.  The first entry is a pointer to the actual
 571          * function.
 572          */
 573         kregs->nip = *((unsigned long *)ret_from_fork);
 574 #else
 575         kregs->nip = (unsigned long)ret_from_fork;
 576 #endif
 577
 578         return 0;
 579 }
 580
 581 /*
 582  * Set up a thread for executing a new program
 583  */
 584 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 585 {
 586 #ifdef CONFIG_PPC64
 587         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 588 #endif
 589
 590         set_fs(USER_DS);
 591
 592         /*
 593          * If we exec out of a kernel thread then thread.regs will not be
 594          * set.  Do it now.
 595          */
 596         if (!current->thread.regs) {
 597                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 598                 current->thread.regs = regs - 1;
 599         }
 600
 601         memset(regs->gpr, 0, sizeof(regs->gpr));
 602         regs->ctr = 0;
 603         regs->link = 0;
 604         regs->xer = 0;
 605         regs->ccr = 0;
 606         regs->gpr[1] = sp;
 607
 608 #ifdef CONFIG_PPC32
 609         regs->mq = 0;
 610         regs->nip = start;
 611         regs->msr = MSR_USER;
 612 #else
 613         if (!test_thread_flag(TIF_32BIT)) {
 614                 unsigned long entry, toc;
 615
 616                 /* start is a relocated pointer to the function descriptor for
 617                  * the elf _start routine.  The first entry in the function
 618                  * descriptor is the entry address of _start and the second
 619                  * entry is the TOC value we need to use.
 620                  */
 621                 __get_user(entry, (unsigned long __user *)start);
 622                 __get_user(toc, (unsigned long __user *)start+1);
 623
 624                 /* Check whether the e_entry function descriptor entries
 625                  * need to be relocated before we can use them.
 626                  */
 627                 if (load_addr != 0) {
 628                         entry += load_addr;
 629                         toc   += load_addr;
 630                 }
 631                 regs->nip = entry;
 632                 regs->gpr[2] = toc;
 633                 regs->msr = MSR_USER64;
 634         } else {
 635                 regs->nip = start;
 636                 regs->gpr[2] = 0;
 637                 regs->msr = MSR_USER32;
 638         }
 639 #endif
 640
 641         discard_lazy_cpu_state();
 642         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 643         current->thread.fpscr.val = 0;
 644 #ifdef CONFIG_ALTIVEC
 645         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 646         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 647         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 648         current->thread.vrsave = 0;
 649         current->thread.used_vr = 0;
 650 #endif /* CONFIG_ALTIVEC */
 651 #ifdef CONFIG_SPE
 652         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 653         current->thread.acc = 0;
 654         current->thread.spefscr = 0;
 655         current->thread.used_spe = 0;
 656 #endif /* CONFIG_SPE */
 657 }
 658
 659 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 660                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 661
 662 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 663 {
 664         struct pt_regs *regs = tsk->thread.regs;
 665
 666         /* This is a bit hairy.  If we are an SPE enabled  processor
 667          * (have embedded fp) we store the IEEE exception enable flags in
 668          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 669          * mode (asyn, precise, disabled) for 'Classic' FP. */
 670         if (val & PR_FP_EXC_SW_ENABLE) {
 671 #ifdef CONFIG_SPE
 672                 tsk->thread.fpexc_mode = val &
 673                         (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 674                 return 0;
 675 #else
 676                 return -EINVAL;
 677 #endif
 678         }
 679
 680         /* on a CONFIG_SPE this does not hurt us.  The bits that
 681          * __pack_fe01 use do not overlap with bits used for
 682          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 683          * on CONFIG_SPE implementations are reserved so writing to
 684          * them does not change anything */
 685         if (val > PR_FP_EXC_PRECISE)
 686                 return -EINVAL;
 687         tsk->thread.fpexc_mode = __pack_fe01(val);
 688         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 689                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 690                         | tsk->thread.fpexc_mode;
 691         return 0;
 692 }
 693
 694 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 695 {
 696         unsigned int val;
 697
 698         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 699 #ifdef CONFIG_SPE
 700                 val = tsk->thread.fpexc_mode;
 701 #else
 702                 return -EINVAL;
 703 #endif
 704         else
 705                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 706         return put_user(val, (unsigned int __user *) adr);
 707 }
 708
 709 int set_endian(struct task_struct *tsk, unsigned int val)
 710 {
 711         struct pt_regs *regs = tsk->thread.regs;
 712
 713         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 714             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 715                 return -EINVAL;
 716
 717         if (regs == NULL)
 718                 return -EINVAL;
 719
 720         if (val == PR_ENDIAN_BIG)
 721                 regs->msr &= ~MSR_LE;
 722         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 723                 regs->msr |= MSR_LE;
 724         else
 725                 return -EINVAL;
 726
 727         return 0;
 728 }
 729
 730 int get_endian(struct task_struct *tsk, unsigned long adr)
 731 {
 732         struct pt_regs *regs = tsk->thread.regs;
 733         unsigned int val;
 734
 735         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 736             !cpu_has_feature(CPU_FTR_REAL_LE))
 737                 return -EINVAL;
 738
 739         if (regs == NULL)
 740                 return -EINVAL;
 741
 742         if (regs->msr & MSR_LE) {
 743                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 744                         val = PR_ENDIAN_LITTLE;
 745                 else
 746                         val = PR_ENDIAN_PPC_LITTLE;
 747         } else
 748                 val = PR_ENDIAN_BIG;
 749
 750         return put_user(val, (unsigned int __user *)adr);
 751 }
 752
 753 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 754 {
 755         tsk->thread.align_ctl = val;
 756         return 0;
 757 }
 758
 759 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 760 {
 761         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 762 }
 763
 764 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 765
 766 int sys_clone(unsigned long clone_flags, unsigned long usp,
 767               int __user *parent_tidp, void __user *child_threadptr,
 768               int __user *child_tidp, int p6,
 769               struct pt_regs *regs)
 770 {
 771         CHECK_FULL_REGS(regs);
 772         if (usp == 0)
 773                 usp = regs->gpr[1];     /* stack pointer for child */
 774 #ifdef CONFIG_PPC64
 775         if (test_thread_flag(TIF_32BIT)) {
 776                 parent_tidp = TRUNC_PTR(parent_tidp);
 777                 child_tidp = TRUNC_PTR(child_tidp);
 778         }
 779 #endif
 780         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 781 }
 782
 783 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 784              unsigned long p4, unsigned long p5, unsigned long p6,
 785              struct pt_regs *regs)
 786 {
 787         CHECK_FULL_REGS(regs);
 788         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 789 }
 790
 791 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
 792               unsigned long p4, unsigned long p5, unsigned long p6,
 793               struct pt_regs *regs)
 794 {
 795         CHECK_FULL_REGS(regs);
 796         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 797                         regs, 0, NULL, NULL);
 798 }
 799
 800 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 801                unsigned long a3, unsigned long a4, unsigned long a5,
 802                struct pt_regs *regs)
 803 {
 804         int error;
 805         char *filename;
 806
 807         filename = getname((char __user *) a0);
 808         error = PTR_ERR(filename);
 809         if (IS_ERR(filename))
 810                 goto out;
 811         flush_fp_to_thread(current);
 812         flush_altivec_to_thread(current);
 813         flush_spe_to_thread(current);
 814         error = do_execve(filename, (char __user * __user *) a1,
 815                           (char __user * __user *) a2, regs);
 816         if (error == 0) {
 817                 task_lock(current);
 818                 current->ptrace &= ~PT_DTRACE;
 819                 task_unlock(current);
 820         }
 821         putname(filename);
 822 out:
 823         return error;
 824 }
 825
 826 #ifdef CONFIG_IRQSTACKS
 827 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
 828                                   unsigned long nbytes)
 829 {
 830         unsigned long stack_page;
 831         unsigned long cpu = task_cpu(p);
 832
 833         /*
 834          * Avoid crashing if the stack has overflowed and corrupted
 835          * task_cpu(p), which is in the thread_info struct.
 836          */
 837         if (cpu < NR_CPUS && cpu_possible(cpu)) {
 838                 stack_page = (unsigned long) hardirq_ctx[cpu];
 839                 if (sp >= stack_page + sizeof(struct thread_struct)
 840                     && sp <= stack_page + THREAD_SIZE - nbytes)
 841                         return 1;
 842
 843                 stack_page = (unsigned long) softirq_ctx[cpu];
 844                 if (sp >= stack_page + sizeof(struct thread_struct)
 845                     && sp <= stack_page + THREAD_SIZE - nbytes)
 846                         return 1;
 847         }
 848         return 0;
 849 }
 850
 851 #else
 852 #define valid_irq_stack(sp, p, nb)      0
 853 #endif /* CONFIG_IRQSTACKS */
 854
 855 int validate_sp(unsigned long sp, struct task_struct *p,
 856                        unsigned long nbytes)
 857 {
 858         unsigned long stack_page = (unsigned long)task_stack_page(p);
 859
 860         if (sp >= stack_page + sizeof(struct thread_struct)
 861             && sp <= stack_page + THREAD_SIZE - nbytes)
 862                 return 1;
 863
 864         return valid_irq_stack(sp, p, nbytes);
 865 }
 866
 867 #ifdef CONFIG_PPC64
 868 #define MIN_STACK_FRAME 112     /* same as STACK_FRAME_OVERHEAD, in fact */
 869 #define FRAME_LR_SAVE   2
 870 #define INT_FRAME_SIZE  (sizeof(struct pt_regs) + STACK_FRAME_OVERHEAD + 288)
 871 #define REGS_MARKER     0x7265677368657265ul
 872 #define FRAME_MARKER    12
 873 #else
 874 #define MIN_STACK_FRAME 16
 875 #define FRAME_LR_SAVE   1
 876 #define INT_FRAME_SIZE  (sizeof(struct pt_regs) + STACK_FRAME_OVERHEAD)
 877 #define REGS_MARKER     0x72656773ul
 878 #define FRAME_MARKER    2
 879 #endif
 880
 881 EXPORT_SYMBOL(validate_sp);
 882
 883 unsigned long get_wchan(struct task_struct *p)
 884 {
 885         unsigned long ip, sp;
 886         int count = 0;
 887
 888         if (!p || p == current || p->state == TASK_RUNNING)
 889                 return 0;
 890
 891         sp = p->thread.ksp;
 892         if (!validate_sp(sp, p, MIN_STACK_FRAME))
 893                 return 0;
 894
 895         do {
 896                 sp = *(unsigned long *)sp;
 897                 if (!validate_sp(sp, p, MIN_STACK_FRAME))
 898                         return 0;
 899                 if (count > 0) {
 900                         ip = ((unsigned long *)sp)[FRAME_LR_SAVE];
 901                         if (!in_sched_functions(ip))
 902                                 return ip;
 903                 }
 904         } while (count++ < 16);
 905         return 0;
 906 }
 907
 908 static int kstack_depth_to_print = 64;
 909
 910 void show_stack(struct task_struct *tsk, unsigned long *stack)
 911 {
 912         unsigned long sp, ip, lr, newsp;
 913         int count = 0;
 914         int firstframe = 1;
 915
 916         sp = (unsigned long) stack;
 917         if (tsk == NULL)
 918                 tsk = current;
 919         if (sp == 0) {
 920                 if (tsk == current)
 921                         asm("mr %0,1" : "=r" (sp));
 922                 else
 923                         sp = tsk->thread.ksp;
 924         }
 925
 926         lr = 0;
 927         printk("Call Trace:\n");
 928         do {
 929                 if (!validate_sp(sp, tsk, MIN_STACK_FRAME))
 930                         return;
 931
 932                 stack = (unsigned long *) sp;
 933                 newsp = stack[0];
 934                 ip = stack[FRAME_LR_SAVE];
 935                 if (!firstframe || ip != lr) {
 936                         printk("["REG"] ["REG"] ", sp, ip);
 937                         print_symbol("%s", ip);
 938                         if (firstframe)
 939                                 printk(" (unreliable)");
 940                         printk("\n");
 941                 }
 942                 firstframe = 0;
 943
 944                 /*
 945                  * See if this is an exception frame.
 946                  * We look for the "regshere" marker in the current frame.
 947                  */
 948                 if (validate_sp(sp, tsk, INT_FRAME_SIZE)
 949                     && stack[FRAME_MARKER] == REGS_MARKER) {
 950                         struct pt_regs *regs = (struct pt_regs *)
 951                                 (sp + STACK_FRAME_OVERHEAD);
 952                         printk("--- Exception: %lx", regs->trap);
 953                         print_symbol(" at %s\n", regs->nip);
 954                         lr = regs->link;
 955                         print_symbol("    LR = %s\n", lr);
 956                         firstframe = 1;
 957                 }
 958
 959                 sp = newsp;
 960         } while (count++ < kstack_depth_to_print);
 961 }
 962
 963 void dump_stack(void)
 964 {
 965         show_stack(current, NULL);
 966 }
 967 EXPORT_SYMBOL(dump_stack);
 968
 969 #ifdef CONFIG_PPC64
 970 void ppc64_runlatch_on(void)
 971 {
 972         unsigned long ctrl;
 973
 974         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
 975                 HMT_medium();
 976
 977                 ctrl = mfspr(SPRN_CTRLF);
 978                 ctrl |= CTRL_RUNLATCH;
 979                 mtspr(SPRN_CTRLT, ctrl);
 980
 981                 set_thread_flag(TIF_RUNLATCH);
 982         }
 983 }
 984
 985 void ppc64_runlatch_off(void)
 986 {
 987         unsigned long ctrl;
 988
 989         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
 990                 HMT_medium();
 991
 992                 clear_thread_flag(TIF_RUNLATCH);
 993
 994                 ctrl = mfspr(SPRN_CTRLF);
 995                 ctrl &= ~CTRL_RUNLATCH;
 996                 mtspr(SPRN_CTRLT, ctrl);
 997         }
 998 }
 999 #endif