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_vsx = NULL;
  57 struct task_struct *last_task_used_spe = NULL;
  58 #endif
  59
  60 /*
  61  * Make sure the floating-point register state in the
  62  * the thread_struct is up to date for task tsk.
  63  */
  64 void flush_fp_to_thread(struct task_struct *tsk)
  65 {
  66         if (tsk->thread.regs) {
  67                 /*
  68                  * We need to disable preemption here because if we didn't,
  69                  * another process could get scheduled after the regs->msr
  70                  * test but before we have finished saving the FP registers
  71                  * to the thread_struct.  That process could take over the
  72                  * FPU, and then when we get scheduled again we would store
  73                  * bogus values for the remaining FP registers.
  74                  */
  75                 preempt_disable();
  76                 if (tsk->thread.regs->msr & MSR_FP) {
  77 #ifdef CONFIG_SMP
  78                         /*
  79                          * This should only ever be called for current or
  80                          * for a stopped child process.  Since we save away
  81                          * the FP register state on context switch on SMP,
  82                          * there is something wrong if a stopped child appears
  83                          * to still have its FP state in the CPU registers.
  84                          */
  85                         BUG_ON(tsk != current);
  86 #endif
  87                         giveup_fpu(tsk);
  88                 }
  89                 preempt_enable();
  90         }
  91 }
  92
  93 void enable_kernel_fp(void)
  94 {
  95         WARN_ON(preemptible());
  96
  97 #ifdef CONFIG_SMP
  98         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
  99                 giveup_fpu(current);
 100         else
 101                 giveup_fpu(NULL);       /* just enables FP for kernel */
 102 #else
 103         giveup_fpu(last_task_used_math);
 104 #endif /* CONFIG_SMP */
 105 }
 106 EXPORT_SYMBOL(enable_kernel_fp);
 107
 108 int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs)
 109 {
 110 #ifdef CONFIG_VSX
 111         int i;
 112         elf_fpreg_t *reg;
 113 #endif
 114
 115         if (!tsk->thread.regs)
 116                 return 0;
 117         flush_fp_to_thread(current);
 118
 119 #ifdef CONFIG_VSX
 120         reg = (elf_fpreg_t *)fpregs;
 121         for (i = 0; i < ELF_NFPREG - 1; i++, reg++)
 122                 *reg = tsk->thread.TS_FPR(i);
 123         memcpy(reg, &tsk->thread.fpscr, sizeof(elf_fpreg_t));
 124 #else
 125         memcpy(fpregs, &tsk->thread.TS_FPR(0), sizeof(*fpregs));
 126 #endif
 127
 128         return 1;
 129 }
 130
 131 #ifdef CONFIG_ALTIVEC
 132 void enable_kernel_altivec(void)
 133 {
 134         WARN_ON(preemptible());
 135
 136 #ifdef CONFIG_SMP
 137         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
 138                 giveup_altivec(current);
 139         else
 140                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
 141 #else
 142         giveup_altivec(last_task_used_altivec);
 143 #endif /* CONFIG_SMP */
 144 }
 145 EXPORT_SYMBOL(enable_kernel_altivec);
 146
 147 /*
 148  * Make sure the VMX/Altivec register state in the
 149  * the thread_struct is up to date for task tsk.
 150  */
 151 void flush_altivec_to_thread(struct task_struct *tsk)
 152 {
 153         if (tsk->thread.regs) {
 154                 preempt_disable();
 155                 if (tsk->thread.regs->msr & MSR_VEC) {
 156 #ifdef CONFIG_SMP
 157                         BUG_ON(tsk != current);
 158 #endif
 159                         giveup_altivec(tsk);
 160                 }
 161                 preempt_enable();
 162         }
 163 }
 164
 165 int dump_task_altivec(struct task_struct *tsk, elf_vrregset_t *vrregs)
 166 {
 167         /* ELF_NVRREG includes the VSCR and VRSAVE which we need to save
 168          * separately, see below */
 169         const int nregs = ELF_NVRREG - 2;
 170         elf_vrreg_t *reg;
 171         u32 *dest;
 172
 173         if (tsk == current)
 174                 flush_altivec_to_thread(tsk);
 175
 176         reg = (elf_vrreg_t *)vrregs;
 177
 178         /* copy the 32 vr registers */
 179         memcpy(reg, &tsk->thread.vr[0], nregs * sizeof(*reg));
 180         reg += nregs;
 181
 182         /* copy the vscr */
 183         memcpy(reg, &tsk->thread.vscr, sizeof(*reg));
 184         reg++;
 185
 186         /* vrsave is stored in the high 32bit slot of the final 128bits */
 187         memset(reg, 0, sizeof(*reg));
 188         dest = (u32 *)reg;
 189         *dest = tsk->thread.vrsave;
 190
 191         return 1;
 192 }
 193 #endif /* CONFIG_ALTIVEC */
 194
 195 #ifdef CONFIG_VSX
 196 #if 0
 197 /* not currently used, but some crazy RAID module might want to later */
 198 void enable_kernel_vsx(void)
 199 {
 200         WARN_ON(preemptible());
 201
 202 #ifdef CONFIG_SMP
 203         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 204                 giveup_vsx(current);
 205         else
 206                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 207 #else
 208         giveup_vsx(last_task_used_vsx);
 209 #endif /* CONFIG_SMP */
 210 }
 211 EXPORT_SYMBOL(enable_kernel_vsx);
 212 #endif
 213
 214 void flush_vsx_to_thread(struct task_struct *tsk)
 215 {
 216         if (tsk->thread.regs) {
 217                 preempt_disable();
 218                 if (tsk->thread.regs->msr & MSR_VSX) {
 219 #ifdef CONFIG_SMP
 220                         BUG_ON(tsk != current);
 221 #endif
 222                         giveup_vsx(tsk);
 223                 }
 224                 preempt_enable();
 225         }
 226 }
 227
 228 /*
 229  * This dumps the lower half 64bits of the first 32 VSX registers.
 230  * This needs to be called with dump_task_fp and dump_task_altivec to
 231  * get all the VSX state.
 232  */
 233 int dump_task_vsx(struct task_struct *tsk, elf_vrreg_t *vrregs)
 234 {
 235         elf_vrreg_t *reg;
 236         double buf[32];
 237         int i;
 238
 239         if (tsk == current)
 240                 flush_vsx_to_thread(tsk);
 241
 242         reg = (elf_vrreg_t *)vrregs;
 243
 244         for (i = 0; i < 32 ; i++)
 245                 buf[i] = current->thread.fpr[i][TS_VSRLOWOFFSET];
 246         memcpy(reg, buf, sizeof(buf));
 247
 248         return 1;
 249 }
 250 #endif /* CONFIG_VSX */
 251
 252 #ifdef CONFIG_SPE
 253
 254 void enable_kernel_spe(void)
 255 {
 256         WARN_ON(preemptible());
 257
 258 #ifdef CONFIG_SMP
 259         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 260                 giveup_spe(current);
 261         else
 262                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 263 #else
 264         giveup_spe(last_task_used_spe);
 265 #endif /* __SMP __ */
 266 }
 267 EXPORT_SYMBOL(enable_kernel_spe);
 268
 269 void flush_spe_to_thread(struct task_struct *tsk)
 270 {
 271         if (tsk->thread.regs) {
 272                 preempt_disable();
 273                 if (tsk->thread.regs->msr & MSR_SPE) {
 274 #ifdef CONFIG_SMP
 275                         BUG_ON(tsk != current);
 276 #endif
 277                         giveup_spe(tsk);
 278                 }
 279                 preempt_enable();
 280         }
 281 }
 282
 283 int dump_spe(struct pt_regs *regs, elf_vrregset_t *evrregs)
 284 {
 285         flush_spe_to_thread(current);
 286         /* We copy u32 evr[32] + u64 acc + u32 spefscr -> 35 */
 287         memcpy(evrregs, &current->thread.evr[0], sizeof(u32) * 35);
 288         return 1;
 289 }
 290 #endif /* CONFIG_SPE */
 291
 292 #ifndef CONFIG_SMP
 293 /*
 294  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 295  * and the current task has some state, discard it.
 296  */
 297 void discard_lazy_cpu_state(void)
 298 {
 299         preempt_disable();
 300         if (last_task_used_math == current)
 301                 last_task_used_math = NULL;
 302 #ifdef CONFIG_ALTIVEC
 303         if (last_task_used_altivec == current)
 304                 last_task_used_altivec = NULL;
 305 #endif /* CONFIG_ALTIVEC */
 306 #ifdef CONFIG_VSX
 307         if (last_task_used_vsx == current)
 308                 last_task_used_vsx = NULL;
 309 #endif /* CONFIG_VSX */
 310 #ifdef CONFIG_SPE
 311         if (last_task_used_spe == current)
 312                 last_task_used_spe = NULL;
 313 #endif
 314         preempt_enable();
 315 }
 316 #endif /* CONFIG_SMP */
 317
 318 static DEFINE_PER_CPU(unsigned long, current_dabr);
 319
 320 int set_dabr(unsigned long dabr)
 321 {
 322         __get_cpu_var(current_dabr) = dabr;
 323
 324 #ifdef CONFIG_PPC_MERGE         /* XXX for now */
 325         if (ppc_md.set_dabr)
 326                 return ppc_md.set_dabr(dabr);
 327 #endif
 328
 329         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 330 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
 331         mtspr(SPRN_DABR, dabr);
 332 #endif
 333         return 0;
 334 }
 335
 336 #ifdef CONFIG_PPC64
 337 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 338 #endif
 339
 340 struct task_struct *__switch_to(struct task_struct *prev,
 341         struct task_struct *new)
 342 {
 343         struct thread_struct *new_thread, *old_thread;
 344         unsigned long flags;
 345         struct task_struct *last;
 346
 347 #ifdef CONFIG_SMP
 348         /* avoid complexity of lazy save/restore of fpu
 349          * by just saving it every time we switch out if
 350          * this task used the fpu during the last quantum.
 351          *
 352          * If it tries to use the fpu again, it'll trap and
 353          * reload its fp regs.  So we don't have to do a restore
 354          * every switch, just a save.
 355          *  -- Cort
 356          */
 357         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 358                 giveup_fpu(prev);
 359 #ifdef CONFIG_ALTIVEC
 360         /*
 361          * If the previous thread used altivec in the last quantum
 362          * (thus changing altivec regs) then save them.
 363          * We used to check the VRSAVE register but not all apps
 364          * set it, so we don't rely on it now (and in fact we need
 365          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 366          *
 367          * On SMP we always save/restore altivec regs just to avoid the
 368          * complexity of changing processors.
 369          *  -- Cort
 370          */
 371         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 372                 giveup_altivec(prev);
 373 #endif /* CONFIG_ALTIVEC */
 374 #ifdef CONFIG_VSX
 375         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 376                 giveup_vsx(prev);
 377 #endif /* CONFIG_VSX */
 378 #ifdef CONFIG_SPE
 379         /*
 380          * If the previous thread used spe in the last quantum
 381          * (thus changing spe regs) then save them.
 382          *
 383          * On SMP we always save/restore spe regs just to avoid the
 384          * complexity of changing processors.
 385          */
 386         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 387                 giveup_spe(prev);
 388 #endif /* CONFIG_SPE */
 389
 390 #else  /* CONFIG_SMP */
 391 #ifdef CONFIG_ALTIVEC
 392         /* Avoid the trap.  On smp this this never happens since
 393          * we don't set last_task_used_altivec -- Cort
 394          */
 395         if (new->thread.regs && last_task_used_altivec == new)
 396                 new->thread.regs->msr |= MSR_VEC;
 397 #endif /* CONFIG_ALTIVEC */
 398 #ifdef CONFIG_VSX
 399         if (new->thread.regs && last_task_used_vsx == new)
 400                 new->thread.regs->msr |= MSR_VSX;
 401 #endif /* CONFIG_VSX */
 402 #ifdef CONFIG_SPE
 403         /* Avoid the trap.  On smp this this never happens since
 404          * we don't set last_task_used_spe
 405          */
 406         if (new->thread.regs && last_task_used_spe == new)
 407                 new->thread.regs->msr |= MSR_SPE;
 408 #endif /* CONFIG_SPE */
 409
 410 #endif /* CONFIG_SMP */
 411
 412         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 413                 set_dabr(new->thread.dabr);
 414
 415         new_thread = &new->thread;
 416         old_thread = &current->thread;
 417
 418 #ifdef CONFIG_PPC64
 419         /*
 420          * Collect processor utilization data per process
 421          */
 422         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 423                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 424                 long unsigned start_tb, current_tb;
 425                 start_tb = old_thread->start_tb;
 426                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 427                 old_thread->accum_tb += (current_tb - start_tb);
 428                 new_thread->start_tb = current_tb;
 429         }
 430 #endif
 431
 432         local_irq_save(flags);
 433
 434         account_system_vtime(current);
 435         account_process_vtime(current);
 436         calculate_steal_time();
 437
 438         /*
 439          * We can't take a PMU exception inside _switch() since there is a
 440          * window where the kernel stack SLB and the kernel stack are out
 441          * of sync. Hard disable here.
 442          */
 443         hard_irq_disable();
 444         last = _switch(old_thread, new_thread);
 445
 446         local_irq_restore(flags);
 447
 448         return last;
 449 }
 450
 451 static int instructions_to_print = 16;
 452
 453 static void show_instructions(struct pt_regs *regs)
 454 {
 455         int i;
 456         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 457                         sizeof(int));
 458
 459         printk("Instruction dump:");
 460
 461         for (i = 0; i < instructions_to_print; i++) {
 462                 int instr;
 463
 464                 if (!(i % 8))
 465                         printk("\n");
 466
 467 #if !defined(CONFIG_BOOKE)
 468                 /* If executing with the IMMU off, adjust pc rather
 469                  * than print XXXXXXXX.
 470                  */
 471                 if (!(regs->msr & MSR_IR))
 472                         pc = (unsigned long)phys_to_virt(pc);
 473 #endif
 474
 475                 /* We use __get_user here *only* to avoid an OOPS on a
 476                  * bad address because the pc *should* only be a
 477                  * kernel address.
 478                  */
 479                 if (!__kernel_text_address(pc) ||
 480                      __get_user(instr, (unsigned int __user *)pc)) {
 481                         printk("XXXXXXXX ");
 482                 } else {
 483                         if (regs->nip == pc)
 484                                 printk("<%08x> ", instr);
 485                         else
 486                                 printk("%08x ", instr);
 487                 }
 488
 489                 pc += sizeof(int);
 490         }
 491
 492         printk("\n");
 493 }
 494
 495 static struct regbit {
 496         unsigned long bit;
 497         const char *name;
 498 } msr_bits[] = {
 499         {MSR_EE,        "EE"},
 500         {MSR_PR,        "PR"},
 501         {MSR_FP,        "FP"},
 502         {MSR_VEC,       "VEC"},
 503         {MSR_VSX,       "VSX"},
 504         {MSR_ME,        "ME"},
 505         {MSR_IR,        "IR"},
 506         {MSR_DR,        "DR"},
 507         {0,             NULL}
 508 };
 509
 510 static void printbits(unsigned long val, struct regbit *bits)
 511 {
 512         const char *sep = "";
 513
 514         printk("<");
 515         for (; bits->bit; ++bits)
 516                 if (val & bits->bit) {
 517                         printk("%s%s", sep, bits->name);
 518                         sep = ",";
 519                 }
 520         printk(">");
 521 }
 522
 523 #ifdef CONFIG_PPC64
 524 #define REG             "%016lx"
 525 #define REGS_PER_LINE   4
 526 #define LAST_VOLATILE   13
 527 #else
 528 #define REG             "%08lx"
 529 #define REGS_PER_LINE   8
 530 #define LAST_VOLATILE   12
 531 #endif
 532
 533 void show_regs(struct pt_regs * regs)
 534 {
 535         int i, trap;
 536
 537         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 538                regs->nip, regs->link, regs->ctr);
 539         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 540                regs, regs->trap, print_tainted(), init_utsname()->release);
 541         printk("MSR: "REG" ", regs->msr);
 542         printbits(regs->msr, msr_bits);
 543         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 544         trap = TRAP(regs);
 545         if (trap == 0x300 || trap == 0x600)
 546 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 547                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 548 #else
 549                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
 550 #endif
 551         printk("TASK = %p[%d] '%s' THREAD: %p",
 552                current, task_pid_nr(current), current->comm, task_thread_info(current));
 553
 554 #ifdef CONFIG_SMP
 555         printk(" CPU: %d", raw_smp_processor_id());
 556 #endif /* CONFIG_SMP */
 557
 558         for (i = 0;  i < 32;  i++) {
 559                 if ((i % REGS_PER_LINE) == 0)
 560                         printk("\n" KERN_INFO "GPR%02d: ", i);
 561                 printk(REG " ", regs->gpr[i]);
 562                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 563                         break;
 564         }
 565         printk("\n");
 566 #ifdef CONFIG_KALLSYMS
 567         /*
 568          * Lookup NIP late so we have the best change of getting the
 569          * above info out without failing
 570          */
 571         printk("NIP ["REG"] ", regs->nip);
 572         print_symbol("%s\n", regs->nip);
 573         printk("LR ["REG"] ", regs->link);
 574         print_symbol("%s\n", regs->link);
 575 #endif
 576         show_stack(current, (unsigned long *) regs->gpr[1]);
 577         if (!user_mode(regs))
 578                 show_instructions(regs);
 579 }
 580
 581 void exit_thread(void)
 582 {
 583         discard_lazy_cpu_state();
 584 }
 585
 586 void flush_thread(void)
 587 {
 588 #ifdef CONFIG_PPC64
 589         struct thread_info *t = current_thread_info();
 590
 591         if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
 592                 clear_ti_thread_flag(t, TIF_ABI_PENDING);
 593                 if (test_ti_thread_flag(t, TIF_32BIT))
 594                         clear_ti_thread_flag(t, TIF_32BIT);
 595                 else
 596                         set_ti_thread_flag(t, TIF_32BIT);
 597         }
 598 #endif
 599
 600         discard_lazy_cpu_state();
 601
 602         if (current->thread.dabr) {
 603                 current->thread.dabr = 0;
 604                 set_dabr(0);
 605         }
 606 }
 607
 608 void
 609 release_thread(struct task_struct *t)
 610 {
 611 }
 612
 613 /*
 614  * This gets called before we allocate a new thread and copy
 615  * the current task into it.
 616  */
 617 void prepare_to_copy(struct task_struct *tsk)
 618 {
 619         flush_fp_to_thread(current);
 620         flush_altivec_to_thread(current);
 621         flush_vsx_to_thread(current);
 622         flush_spe_to_thread(current);
 623 }
 624
 625 /*
 626  * Copy a thread..
 627  */
 628 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 629                 unsigned long unused, struct task_struct *p,
 630                 struct pt_regs *regs)
 631 {
 632         struct pt_regs *childregs, *kregs;
 633         extern void ret_from_fork(void);
 634         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 635
 636         CHECK_FULL_REGS(regs);
 637         /* Copy registers */
 638         sp -= sizeof(struct pt_regs);
 639         childregs = (struct pt_regs *) sp;
 640         *childregs = *regs;
 641         if ((childregs->msr & MSR_PR) == 0) {
 642                 /* for kernel thread, set `current' and stackptr in new task */
 643                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 644 #ifdef CONFIG_PPC32
 645                 childregs->gpr[2] = (unsigned long) p;
 646 #else
 647                 clear_tsk_thread_flag(p, TIF_32BIT);
 648 #endif
 649                 p->thread.regs = NULL;  /* no user register state */
 650         } else {
 651                 childregs->gpr[1] = usp;
 652                 p->thread.regs = childregs;
 653                 if (clone_flags & CLONE_SETTLS) {
 654 #ifdef CONFIG_PPC64
 655                         if (!test_thread_flag(TIF_32BIT))
 656                                 childregs->gpr[13] = childregs->gpr[6];
 657                         else
 658 #endif
 659                                 childregs->gpr[2] = childregs->gpr[6];
 660                 }
 661         }
 662         childregs->gpr[3] = 0;  /* Result from fork() */
 663         sp -= STACK_FRAME_OVERHEAD;
 664
 665         /*
 666          * The way this works is that at some point in the future
 667          * some task will call _switch to switch to the new task.
 668          * That will pop off the stack frame created below and start
 669          * the new task running at ret_from_fork.  The new task will
 670          * do some house keeping and then return from the fork or clone
 671          * system call, using the stack frame created above.
 672          */
 673         sp -= sizeof(struct pt_regs);
 674         kregs = (struct pt_regs *) sp;
 675         sp -= STACK_FRAME_OVERHEAD;
 676         p->thread.ksp = sp;
 677         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 678                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 679
 680 #ifdef CONFIG_PPC64
 681         if (cpu_has_feature(CPU_FTR_SLB)) {
 682                 unsigned long sp_vsid;
 683                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 684
 685                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
 686                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 687                                 << SLB_VSID_SHIFT_1T;
 688                 else
 689                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 690                                 << SLB_VSID_SHIFT;
 691                 sp_vsid |= SLB_VSID_KERNEL | llp;
 692                 p->thread.ksp_vsid = sp_vsid;
 693         }
 694
 695         /*
 696          * The PPC64 ABI makes use of a TOC to contain function
 697          * pointers.  The function (ret_from_except) is actually a pointer
 698          * to the TOC entry.  The first entry is a pointer to the actual
 699          * function.
 700          */
 701         kregs->nip = *((unsigned long *)ret_from_fork);
 702 #else
 703         kregs->nip = (unsigned long)ret_from_fork;
 704 #endif
 705
 706         return 0;
 707 }
 708
 709 /*
 710  * Set up a thread for executing a new program
 711  */
 712 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 713 {
 714 #ifdef CONFIG_PPC64
 715         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 716 #endif
 717
 718         set_fs(USER_DS);
 719
 720         /*
 721          * If we exec out of a kernel thread then thread.regs will not be
 722          * set.  Do it now.
 723          */
 724         if (!current->thread.regs) {
 725                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 726                 current->thread.regs = regs - 1;
 727         }
 728
 729         memset(regs->gpr, 0, sizeof(regs->gpr));
 730         regs->ctr = 0;
 731         regs->link = 0;
 732         regs->xer = 0;
 733         regs->ccr = 0;
 734         regs->gpr[1] = sp;
 735
 736         /*
 737          * We have just cleared all the nonvolatile GPRs, so make
 738          * FULL_REGS(regs) return true.  This is necessary to allow
 739          * ptrace to examine the thread immediately after exec.
 740          */
 741         regs->trap &= ~1UL;
 742
 743 #ifdef CONFIG_PPC32
 744         regs->mq = 0;
 745         regs->nip = start;
 746         regs->msr = MSR_USER;
 747 #else
 748         if (!test_thread_flag(TIF_32BIT)) {
 749                 unsigned long entry, toc;
 750
 751                 /* start is a relocated pointer to the function descriptor for
 752                  * the elf _start routine.  The first entry in the function
 753                  * descriptor is the entry address of _start and the second
 754                  * entry is the TOC value we need to use.
 755                  */
 756                 __get_user(entry, (unsigned long __user *)start);
 757                 __get_user(toc, (unsigned long __user *)start+1);
 758
 759                 /* Check whether the e_entry function descriptor entries
 760                  * need to be relocated before we can use them.
 761                  */
 762                 if (load_addr != 0) {
 763                         entry += load_addr;
 764                         toc   += load_addr;
 765                 }
 766                 regs->nip = entry;
 767                 regs->gpr[2] = toc;
 768                 regs->msr = MSR_USER64;
 769         } else {
 770                 regs->nip = start;
 771                 regs->gpr[2] = 0;
 772                 regs->msr = MSR_USER32;
 773         }
 774 #endif
 775
 776         discard_lazy_cpu_state();
 777 #ifdef CONFIG_VSX
 778         current->thread.used_vsr = 0;
 779 #endif
 780         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 781         current->thread.fpscr.val = 0;
 782 #ifdef CONFIG_ALTIVEC
 783         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 784         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 785         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 786         current->thread.vrsave = 0;
 787         current->thread.used_vr = 0;
 788 #endif /* CONFIG_ALTIVEC */
 789 #ifdef CONFIG_SPE
 790         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 791         current->thread.acc = 0;
 792         current->thread.spefscr = 0;
 793         current->thread.used_spe = 0;
 794 #endif /* CONFIG_SPE */
 795 }
 796
 797 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 798                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 799
 800 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 801 {
 802         struct pt_regs *regs = tsk->thread.regs;
 803
 804         /* This is a bit hairy.  If we are an SPE enabled  processor
 805          * (have embedded fp) we store the IEEE exception enable flags in
 806          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 807          * mode (asyn, precise, disabled) for 'Classic' FP. */
 808         if (val & PR_FP_EXC_SW_ENABLE) {
 809 #ifdef CONFIG_SPE
 810                 if (cpu_has_feature(CPU_FTR_SPE)) {
 811                         tsk->thread.fpexc_mode = val &
 812                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 813                         return 0;
 814                 } else {
 815                         return -EINVAL;
 816                 }
 817 #else
 818                 return -EINVAL;
 819 #endif
 820         }
 821
 822         /* on a CONFIG_SPE this does not hurt us.  The bits that
 823          * __pack_fe01 use do not overlap with bits used for
 824          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 825          * on CONFIG_SPE implementations are reserved so writing to
 826          * them does not change anything */
 827         if (val > PR_FP_EXC_PRECISE)
 828                 return -EINVAL;
 829         tsk->thread.fpexc_mode = __pack_fe01(val);
 830         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 831                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 832                         | tsk->thread.fpexc_mode;
 833         return 0;
 834 }
 835
 836 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 837 {
 838         unsigned int val;
 839
 840         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 841 #ifdef CONFIG_SPE
 842                 if (cpu_has_feature(CPU_FTR_SPE))
 843                         val = tsk->thread.fpexc_mode;
 844                 else
 845                         return -EINVAL;
 846 #else
 847                 return -EINVAL;
 848 #endif
 849         else
 850                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 851         return put_user(val, (unsigned int __user *) adr);
 852 }
 853
 854 int set_endian(struct task_struct *tsk, unsigned int val)
 855 {
 856         struct pt_regs *regs = tsk->thread.regs;
 857
 858         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 859             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 860                 return -EINVAL;
 861
 862         if (regs == NULL)
 863                 return -EINVAL;
 864
 865         if (val == PR_ENDIAN_BIG)
 866                 regs->msr &= ~MSR_LE;
 867         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 868                 regs->msr |= MSR_LE;
 869         else
 870                 return -EINVAL;
 871
 872         return 0;
 873 }
 874
 875 int get_endian(struct task_struct *tsk, unsigned long adr)
 876 {
 877         struct pt_regs *regs = tsk->thread.regs;
 878         unsigned int val;
 879
 880         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
 881             !cpu_has_feature(CPU_FTR_REAL_LE))
 882                 return -EINVAL;
 883
 884         if (regs == NULL)
 885                 return -EINVAL;
 886
 887         if (regs->msr & MSR_LE) {
 888                 if (cpu_has_feature(CPU_FTR_REAL_LE))
 889                         val = PR_ENDIAN_LITTLE;
 890                 else
 891                         val = PR_ENDIAN_PPC_LITTLE;
 892         } else
 893                 val = PR_ENDIAN_BIG;
 894
 895         return put_user(val, (unsigned int __user *)adr);
 896 }
 897
 898 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
 899 {
 900         tsk->thread.align_ctl = val;
 901         return 0;
 902 }
 903
 904 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
 905 {
 906         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
 907 }
 908
 909 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
 910
 911 int sys_clone(unsigned long clone_flags, unsigned long usp,
 912               int __user *parent_tidp, void __user *child_threadptr,
 913               int __user *child_tidp, int p6,
 914               struct pt_regs *regs)
 915 {
 916         CHECK_FULL_REGS(regs);
 917         if (usp == 0)
 918                 usp = regs->gpr[1];     /* stack pointer for child */
 919 #ifdef CONFIG_PPC64
 920         if (test_thread_flag(TIF_32BIT)) {
 921                 parent_tidp = TRUNC_PTR(parent_tidp);
 922                 child_tidp = TRUNC_PTR(child_tidp);
 923         }
 924 #endif
 925         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
 926 }
 927
 928 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
 929              unsigned long p4, unsigned long p5, unsigned long p6,
 930              struct pt_regs *regs)
 931 {
 932         CHECK_FULL_REGS(regs);
 933         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
 934 }
 935
 936 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
 937               unsigned long p4, unsigned long p5, unsigned long p6,
 938               struct pt_regs *regs)
 939 {
 940         CHECK_FULL_REGS(regs);
 941         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
 942                         regs, 0, NULL, NULL);
 943 }
 944
 945 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
 946                unsigned long a3, unsigned long a4, unsigned long a5,
 947                struct pt_regs *regs)
 948 {
 949         int error;
 950         char *filename;
 951
 952         filename = getname((char __user *) a0);
 953         error = PTR_ERR(filename);
 954         if (IS_ERR(filename))
 955                 goto out;
 956         flush_fp_to_thread(current);
 957         flush_altivec_to_thread(current);
 958         flush_spe_to_thread(current);
 959         error = do_execve(filename, (char __user * __user *) a1,
 960                           (char __user * __user *) a2, regs);
 961         putname(filename);
 962 out:
 963         return error;
 964 }
 965
 966 #ifdef CONFIG_IRQSTACKS
 967 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
 968                                   unsigned long nbytes)
 969 {
 970         unsigned long stack_page;
 971         unsigned long cpu = task_cpu(p);
 972
 973         /*
 974          * Avoid crashing if the stack has overflowed and corrupted
 975          * task_cpu(p), which is in the thread_info struct.
 976          */
 977         if (cpu < NR_CPUS && cpu_possible(cpu)) {
 978                 stack_page = (unsigned long) hardirq_ctx[cpu];
 979                 if (sp >= stack_page + sizeof(struct thread_struct)
 980                     && sp <= stack_page + THREAD_SIZE - nbytes)
 981                         return 1;
 982
 983                 stack_page = (unsigned long) softirq_ctx[cpu];
 984                 if (sp >= stack_page + sizeof(struct thread_struct)
 985                     && sp <= stack_page + THREAD_SIZE - nbytes)
 986                         return 1;
 987         }
 988         return 0;
 989 }
 990
 991 #else
 992 #define valid_irq_stack(sp, p, nb)      0
 993 #endif /* CONFIG_IRQSTACKS */
 994
 995 int validate_sp(unsigned long sp, struct task_struct *p,
 996                        unsigned long nbytes)
 997 {
 998         unsigned long stack_page = (unsigned long)task_stack_page(p);
 999
1000         if (sp >= stack_page + sizeof(struct thread_struct)
1001             && sp <= stack_page + THREAD_SIZE - nbytes)
1002                 return 1;
1003
1004         return valid_irq_stack(sp, p, nbytes);
1005 }
1006
1007 EXPORT_SYMBOL(validate_sp);
1008
1009 unsigned long get_wchan(struct task_struct *p)
1010 {
1011         unsigned long ip, sp;
1012         int count = 0;
1013
1014         if (!p || p == current || p->state == TASK_RUNNING)
1015                 return 0;
1016
1017         sp = p->thread.ksp;
1018         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1019                 return 0;
1020
1021         do {
1022                 sp = *(unsigned long *)sp;
1023                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1024                         return 0;
1025                 if (count > 0) {
1026                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1027                         if (!in_sched_functions(ip))
1028                                 return ip;
1029                 }
1030         } while (count++ < 16);
1031         return 0;
1032 }
1033
1034 static int kstack_depth_to_print = 64;
1035
1036 void show_stack(struct task_struct *tsk, unsigned long *stack)
1037 {
1038         unsigned long sp, ip, lr, newsp;
1039         int count = 0;
1040         int firstframe = 1;
1041
1042         sp = (unsigned long) stack;
1043         if (tsk == NULL)
1044                 tsk = current;
1045         if (sp == 0) {
1046                 if (tsk == current)
1047                         asm("mr %0,1" : "=r" (sp));
1048                 else
1049                         sp = tsk->thread.ksp;
1050         }
1051
1052         lr = 0;
1053         printk("Call Trace:\n");
1054         do {
1055                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1056                         return;
1057
1058                 stack = (unsigned long *) sp;
1059                 newsp = stack[0];
1060                 ip = stack[STACK_FRAME_LR_SAVE];
1061                 if (!firstframe || ip != lr) {
1062                         printk("["REG"] ["REG"] ", sp, ip);
1063                         print_symbol("%s", ip);
1064                         if (firstframe)
1065                                 printk(" (unreliable)");
1066                         printk("\n");
1067                 }
1068                 firstframe = 0;
1069
1070                 /*
1071                  * See if this is an exception frame.
1072                  * We look for the "regshere" marker in the current frame.
1073                  */
1074                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1075                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1076                         struct pt_regs *regs = (struct pt_regs *)
1077                                 (sp + STACK_FRAME_OVERHEAD);
1078                         printk("--- Exception: %lx", regs->trap);
1079                         print_symbol(" at %s\n", regs->nip);
1080                         lr = regs->link;
1081                         print_symbol("    LR = %s\n", lr);
1082                         firstframe = 1;
1083                 }
1084
1085                 sp = newsp;
1086         } while (count++ < kstack_depth_to_print);
1087 }
1088
1089 void dump_stack(void)
1090 {
1091         show_stack(current, NULL);
1092 }
1093 EXPORT_SYMBOL(dump_stack);
1094
1095 #ifdef CONFIG_PPC64
1096 void ppc64_runlatch_on(void)
1097 {
1098         unsigned long ctrl;
1099
1100         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1101                 HMT_medium();
1102
1103                 ctrl = mfspr(SPRN_CTRLF);
1104                 ctrl |= CTRL_RUNLATCH;
1105                 mtspr(SPRN_CTRLT, ctrl);
1106
1107                 set_thread_flag(TIF_RUNLATCH);
1108         }
1109 }
1110
1111 void ppc64_runlatch_off(void)
1112 {
1113         unsigned long ctrl;
1114
1115         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1116                 HMT_medium();
1117
1118                 clear_thread_flag(TIF_RUNLATCH);
1119
1120                 ctrl = mfspr(SPRN_CTRLF);
1121                 ctrl &= ~CTRL_RUNLATCH;
1122                 mtspr(SPRN_CTRLT, ctrl);
1123         }
1124 }
1125 #endif
1126
1127 #if THREAD_SHIFT < PAGE_SHIFT
1128
1129 static struct kmem_cache *thread_info_cache;
1130
1131 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1132 {
1133         struct thread_info *ti;
1134
1135         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1136         if (unlikely(ti == NULL))
1137                 return NULL;
1138 #ifdef CONFIG_DEBUG_STACK_USAGE
1139         memset(ti, 0, THREAD_SIZE);
1140 #endif
1141         return ti;
1142 }
1143
1144 void free_thread_info(struct thread_info *ti)
1145 {
1146         kmem_cache_free(thread_info_cache, ti);
1147 }
1148
1149 void thread_info_cache_init(void)
1150 {
1151         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1152                                               THREAD_SIZE, 0, NULL);
1153         BUG_ON(thread_info_cache == NULL);
1154 }
1155
1156 #endif /* THREAD_SHIFT < PAGE_SHIFT */