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/export.h>
  32 #include <linux/kallsyms.h>
  33 #include <linux/mqueue.h>
  34 #include <linux/hardirq.h>
  35 #include <linux/utsname.h>
  36 #include <linux/ftrace.h>
  37 #include <linux/kernel_stat.h>
  38 #include <linux/personality.h>
  39 #include <linux/random.h>
  40 #include <linux/hw_breakpoint.h>
  41
  42 #include <asm/pgtable.h>
  43 #include <asm/uaccess.h>
  44 #include <asm/io.h>
  45 #include <asm/processor.h>
  46 #include <asm/mmu.h>
  47 #include <asm/prom.h>
  48 #include <asm/machdep.h>
  49 #include <asm/time.h>
  50 #include <asm/runlatch.h>
  51 #include <asm/syscalls.h>
  52 #include <asm/switch_to.h>
  53 #include <asm/debug.h>
  54 #ifdef CONFIG_PPC64
  55 #include <asm/firmware.h>
  56 #endif
  57 #include <linux/kprobes.h>
  58 #include <linux/kdebug.h>
  59
  60 extern unsigned long _get_SP(void);
  61
  62 #ifndef CONFIG_SMP
  63 struct task_struct *last_task_used_math = NULL;
  64 struct task_struct *last_task_used_altivec = NULL;
  65 struct task_struct *last_task_used_vsx = NULL;
  66 struct task_struct *last_task_used_spe = NULL;
  67 #endif
  68
  69 /*
  70  * Make sure the floating-point register state in the
  71  * the thread_struct is up to date for task tsk.
  72  */
  73 void flush_fp_to_thread(struct task_struct *tsk)
  74 {
  75         if (tsk->thread.regs) {
  76                 /*
  77                  * We need to disable preemption here because if we didn't,
  78                  * another process could get scheduled after the regs->msr
  79                  * test but before we have finished saving the FP registers
  80                  * to the thread_struct.  That process could take over the
  81                  * FPU, and then when we get scheduled again we would store
  82                  * bogus values for the remaining FP registers.
  83                  */
  84                 preempt_disable();
  85                 if (tsk->thread.regs->msr & MSR_FP) {
  86 #ifdef CONFIG_SMP
  87                         /*
  88                          * This should only ever be called for current or
  89                          * for a stopped child process.  Since we save away
  90                          * the FP register state on context switch on SMP,
  91                          * there is something wrong if a stopped child appears
  92                          * to still have its FP state in the CPU registers.
  93                          */
  94                         BUG_ON(tsk != current);
  95 #endif
  96                         giveup_fpu(tsk);
  97                 }
  98                 preempt_enable();
  99         }
 100 }
 101 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
 102
 103 void enable_kernel_fp(void)
 104 {
 105         WARN_ON(preemptible());
 106
 107 #ifdef CONFIG_SMP
 108         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
 109                 giveup_fpu(current);
 110         else
 111                 giveup_fpu(NULL);       /* just enables FP for kernel */
 112 #else
 113         giveup_fpu(last_task_used_math);
 114 #endif /* CONFIG_SMP */
 115 }
 116 EXPORT_SYMBOL(enable_kernel_fp);
 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_notask();
 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 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
 152 #endif /* CONFIG_ALTIVEC */
 153
 154 #ifdef CONFIG_VSX
 155 #if 0
 156 /* not currently used, but some crazy RAID module might want to later */
 157 void enable_kernel_vsx(void)
 158 {
 159         WARN_ON(preemptible());
 160
 161 #ifdef CONFIG_SMP
 162         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
 163                 giveup_vsx(current);
 164         else
 165                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
 166 #else
 167         giveup_vsx(last_task_used_vsx);
 168 #endif /* CONFIG_SMP */
 169 }
 170 EXPORT_SYMBOL(enable_kernel_vsx);
 171 #endif
 172
 173 void giveup_vsx(struct task_struct *tsk)
 174 {
 175         giveup_fpu(tsk);
 176         giveup_altivec(tsk);
 177         __giveup_vsx(tsk);
 178 }
 179
 180 void flush_vsx_to_thread(struct task_struct *tsk)
 181 {
 182         if (tsk->thread.regs) {
 183                 preempt_disable();
 184                 if (tsk->thread.regs->msr & MSR_VSX) {
 185 #ifdef CONFIG_SMP
 186                         BUG_ON(tsk != current);
 187 #endif
 188                         giveup_vsx(tsk);
 189                 }
 190                 preempt_enable();
 191         }
 192 }
 193 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
 194 #endif /* CONFIG_VSX */
 195
 196 #ifdef CONFIG_SPE
 197
 198 void enable_kernel_spe(void)
 199 {
 200         WARN_ON(preemptible());
 201
 202 #ifdef CONFIG_SMP
 203         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
 204                 giveup_spe(current);
 205         else
 206                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
 207 #else
 208         giveup_spe(last_task_used_spe);
 209 #endif /* __SMP __ */
 210 }
 211 EXPORT_SYMBOL(enable_kernel_spe);
 212
 213 void flush_spe_to_thread(struct task_struct *tsk)
 214 {
 215         if (tsk->thread.regs) {
 216                 preempt_disable();
 217                 if (tsk->thread.regs->msr & MSR_SPE) {
 218 #ifdef CONFIG_SMP
 219                         BUG_ON(tsk != current);
 220 #endif
 221                         tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
 222                         giveup_spe(tsk);
 223                 }
 224                 preempt_enable();
 225         }
 226 }
 227 #endif /* CONFIG_SPE */
 228
 229 #ifndef CONFIG_SMP
 230 /*
 231  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
 232  * and the current task has some state, discard it.
 233  */
 234 void discard_lazy_cpu_state(void)
 235 {
 236         preempt_disable();
 237         if (last_task_used_math == current)
 238                 last_task_used_math = NULL;
 239 #ifdef CONFIG_ALTIVEC
 240         if (last_task_used_altivec == current)
 241                 last_task_used_altivec = NULL;
 242 #endif /* CONFIG_ALTIVEC */
 243 #ifdef CONFIG_VSX
 244         if (last_task_used_vsx == current)
 245                 last_task_used_vsx = NULL;
 246 #endif /* CONFIG_VSX */
 247 #ifdef CONFIG_SPE
 248         if (last_task_used_spe == current)
 249                 last_task_used_spe = NULL;
 250 #endif
 251         preempt_enable();
 252 }
 253 #endif /* CONFIG_SMP */
 254
 255 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 256 void do_send_trap(struct pt_regs *regs, unsigned long address,
 257                   unsigned long error_code, int signal_code, int breakpt)
 258 {
 259         siginfo_t info;
 260
 261         current->thread.trap_nr = signal_code;
 262         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 263                         11, SIGSEGV) == NOTIFY_STOP)
 264                 return;
 265
 266         /* Deliver the signal to userspace */
 267         info.si_signo = SIGTRAP;
 268         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
 269         info.si_code = signal_code;
 270         info.si_addr = (void __user *)address;
 271         force_sig_info(SIGTRAP, &info, current);
 272 }
 273 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 274 void do_dabr(struct pt_regs *regs, unsigned long address,
 275                     unsigned long error_code)
 276 {
 277         siginfo_t info;
 278
 279         current->thread.trap_nr = TRAP_HWBKPT;
 280         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 281                         11, SIGSEGV) == NOTIFY_STOP)
 282                 return;
 283
 284         if (debugger_dabr_match(regs))
 285                 return;
 286
 287         /* Clear the DABR */
 288         set_dabr(0, 0);
 289
 290         /* Deliver the signal to userspace */
 291         info.si_signo = SIGTRAP;
 292         info.si_errno = 0;
 293         info.si_code = TRAP_HWBKPT;
 294         info.si_addr = (void __user *)address;
 295         force_sig_info(SIGTRAP, &info, current);
 296 }
 297 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 298
 299 static DEFINE_PER_CPU(unsigned long, current_dabr);
 300
 301 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 302 /*
 303  * Set the debug registers back to their default "safe" values.
 304  */
 305 static void set_debug_reg_defaults(struct thread_struct *thread)
 306 {
 307         thread->iac1 = thread->iac2 = 0;
 308 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 309         thread->iac3 = thread->iac4 = 0;
 310 #endif
 311         thread->dac1 = thread->dac2 = 0;
 312 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 313         thread->dvc1 = thread->dvc2 = 0;
 314 #endif
 315         thread->dbcr0 = 0;
 316 #ifdef CONFIG_BOOKE
 317         /*
 318          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
 319          */
 320         thread->dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |   \
 321                         DBCR1_IAC3US | DBCR1_IAC4US;
 322         /*
 323          * Force Data Address Compare User/Supervisor bits to be User-only
 324          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
 325          */
 326         thread->dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
 327 #else
 328         thread->dbcr1 = 0;
 329 #endif
 330 }
 331
 332 static void prime_debug_regs(struct thread_struct *thread)
 333 {
 334         mtspr(SPRN_IAC1, thread->iac1);
 335         mtspr(SPRN_IAC2, thread->iac2);
 336 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
 337         mtspr(SPRN_IAC3, thread->iac3);
 338         mtspr(SPRN_IAC4, thread->iac4);
 339 #endif
 340         mtspr(SPRN_DAC1, thread->dac1);
 341         mtspr(SPRN_DAC2, thread->dac2);
 342 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 343         mtspr(SPRN_DVC1, thread->dvc1);
 344         mtspr(SPRN_DVC2, thread->dvc2);
 345 #endif
 346         mtspr(SPRN_DBCR0, thread->dbcr0);
 347         mtspr(SPRN_DBCR1, thread->dbcr1);
 348 #ifdef CONFIG_BOOKE
 349         mtspr(SPRN_DBCR2, thread->dbcr2);
 350 #endif
 351 }
 352 /*
 353  * Unless neither the old or new thread are making use of the
 354  * debug registers, set the debug registers from the values
 355  * stored in the new thread.
 356  */
 357 static void switch_booke_debug_regs(struct thread_struct *new_thread)
 358 {
 359         if ((current->thread.dbcr0 & DBCR0_IDM)
 360                 || (new_thread->dbcr0 & DBCR0_IDM))
 361                         prime_debug_regs(new_thread);
 362 }
 363 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
 364 #ifndef CONFIG_HAVE_HW_BREAKPOINT
 365 static void set_debug_reg_defaults(struct thread_struct *thread)
 366 {
 367         if (thread->dabr) {
 368                 thread->dabr = 0;
 369                 thread->dabrx = 0;
 370                 set_dabr(0, 0);
 371         }
 372 }
 373 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
 374 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
 375
 376 int set_dabr(unsigned long dabr, unsigned long dabrx)
 377 {
 378         __get_cpu_var(current_dabr) = dabr;
 379
 380         if (ppc_md.set_dabr)
 381                 return ppc_md.set_dabr(dabr, dabrx);
 382
 383         /* XXX should we have a CPU_FTR_HAS_DABR ? */
 384 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 385         mtspr(SPRN_DAC1, dabr);
 386 #ifdef CONFIG_PPC_47x
 387         isync();
 388 #endif
 389 #elif defined(CONFIG_PPC_BOOK3S)
 390         mtspr(SPRN_DABR, dabr);
 391         mtspr(SPRN_DABRX, dabrx);
 392 #endif
 393         return 0;
 394 }
 395
 396 #ifdef CONFIG_PPC64
 397 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
 398 #endif
 399
 400 struct task_struct *__switch_to(struct task_struct *prev,
 401         struct task_struct *new)
 402 {
 403         struct thread_struct *new_thread, *old_thread;
 404         unsigned long flags;
 405         struct task_struct *last;
 406 #ifdef CONFIG_PPC_BOOK3S_64
 407         struct ppc64_tlb_batch *batch;
 408 #endif
 409
 410 #ifdef CONFIG_SMP
 411         /* avoid complexity of lazy save/restore of fpu
 412          * by just saving it every time we switch out if
 413          * this task used the fpu during the last quantum.
 414          *
 415          * If it tries to use the fpu again, it'll trap and
 416          * reload its fp regs.  So we don't have to do a restore
 417          * every switch, just a save.
 418          *  -- Cort
 419          */
 420         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
 421                 giveup_fpu(prev);
 422 #ifdef CONFIG_ALTIVEC
 423         /*
 424          * If the previous thread used altivec in the last quantum
 425          * (thus changing altivec regs) then save them.
 426          * We used to check the VRSAVE register but not all apps
 427          * set it, so we don't rely on it now (and in fact we need
 428          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
 429          *
 430          * On SMP we always save/restore altivec regs just to avoid the
 431          * complexity of changing processors.
 432          *  -- Cort
 433          */
 434         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
 435                 giveup_altivec(prev);
 436 #endif /* CONFIG_ALTIVEC */
 437 #ifdef CONFIG_VSX
 438         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
 439                 /* VMX and FPU registers are already save here */
 440                 __giveup_vsx(prev);
 441 #endif /* CONFIG_VSX */
 442 #ifdef CONFIG_SPE
 443         /*
 444          * If the previous thread used spe in the last quantum
 445          * (thus changing spe regs) then save them.
 446          *
 447          * On SMP we always save/restore spe regs just to avoid the
 448          * complexity of changing processors.
 449          */
 450         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
 451                 giveup_spe(prev);
 452 #endif /* CONFIG_SPE */
 453
 454 #else  /* CONFIG_SMP */
 455 #ifdef CONFIG_ALTIVEC
 456         /* Avoid the trap.  On smp this this never happens since
 457          * we don't set last_task_used_altivec -- Cort
 458          */
 459         if (new->thread.regs && last_task_used_altivec == new)
 460                 new->thread.regs->msr |= MSR_VEC;
 461 #endif /* CONFIG_ALTIVEC */
 462 #ifdef CONFIG_VSX
 463         if (new->thread.regs && last_task_used_vsx == new)
 464                 new->thread.regs->msr |= MSR_VSX;
 465 #endif /* CONFIG_VSX */
 466 #ifdef CONFIG_SPE
 467         /* Avoid the trap.  On smp this this never happens since
 468          * we don't set last_task_used_spe
 469          */
 470         if (new->thread.regs && last_task_used_spe == new)
 471                 new->thread.regs->msr |= MSR_SPE;
 472 #endif /* CONFIG_SPE */
 473
 474 #endif /* CONFIG_SMP */
 475
 476 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
 477         switch_booke_debug_regs(&new->thread);
 478 #else
 479 /*
 480  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
 481  * schedule DABR
 482  */
 483 #ifndef CONFIG_HAVE_HW_BREAKPOINT
 484         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
 485                 set_dabr(new->thread.dabr, new->thread.dabrx);
 486 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 487 #endif
 488
 489
 490         new_thread = &new->thread;
 491         old_thread = &current->thread;
 492
 493 #ifdef CONFIG_PPC64
 494         /*
 495          * Collect processor utilization data per process
 496          */
 497         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 498                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
 499                 long unsigned start_tb, current_tb;
 500                 start_tb = old_thread->start_tb;
 501                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
 502                 old_thread->accum_tb += (current_tb - start_tb);
 503                 new_thread->start_tb = current_tb;
 504         }
 505 #endif /* CONFIG_PPC64 */
 506
 507 #ifdef CONFIG_PPC_BOOK3S_64
 508         batch = &__get_cpu_var(ppc64_tlb_batch);
 509         if (batch->active) {
 510                 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
 511                 if (batch->index)
 512                         __flush_tlb_pending(batch);
 513                 batch->active = 0;
 514         }
 515 #endif /* CONFIG_PPC_BOOK3S_64 */
 516
 517         local_irq_save(flags);
 518
 519         /*
 520          * We can't take a PMU exception inside _switch() since there is a
 521          * window where the kernel stack SLB and the kernel stack are out
 522          * of sync. Hard disable here.
 523          */
 524         hard_irq_disable();
 525         last = _switch(old_thread, new_thread);
 526
 527 #ifdef CONFIG_PPC_BOOK3S_64
 528         if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
 529                 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
 530                 batch = &__get_cpu_var(ppc64_tlb_batch);
 531                 batch->active = 1;
 532         }
 533 #endif /* CONFIG_PPC_BOOK3S_64 */
 534
 535         local_irq_restore(flags);
 536
 537         return last;
 538 }
 539
 540 static int instructions_to_print = 16;
 541
 542 static void show_instructions(struct pt_regs *regs)
 543 {
 544         int i;
 545         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
 546                         sizeof(int));
 547
 548         printk("Instruction dump:");
 549
 550         for (i = 0; i < instructions_to_print; i++) {
 551                 int instr;
 552
 553                 if (!(i % 8))
 554                         printk("\n");
 555
 556 #if !defined(CONFIG_BOOKE)
 557                 /* If executing with the IMMU off, adjust pc rather
 558                  * than print XXXXXXXX.
 559                  */
 560                 if (!(regs->msr & MSR_IR))
 561                         pc = (unsigned long)phys_to_virt(pc);
 562 #endif
 563
 564                 /* We use __get_user here *only* to avoid an OOPS on a
 565                  * bad address because the pc *should* only be a
 566                  * kernel address.
 567                  */
 568                 if (!__kernel_text_address(pc) ||
 569                      __get_user(instr, (unsigned int __user *)pc)) {
 570                         printk(KERN_CONT "XXXXXXXX ");
 571                 } else {
 572                         if (regs->nip == pc)
 573                                 printk(KERN_CONT "<%08x> ", instr);
 574                         else
 575                                 printk(KERN_CONT "%08x ", instr);
 576                 }
 577
 578                 pc += sizeof(int);
 579         }
 580
 581         printk("\n");
 582 }
 583
 584 static struct regbit {
 585         unsigned long bit;
 586         const char *name;
 587 } msr_bits[] = {
 588 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
 589         {MSR_SF,        "SF"},
 590         {MSR_HV,        "HV"},
 591 #endif
 592         {MSR_VEC,       "VEC"},
 593         {MSR_VSX,       "VSX"},
 594 #ifdef CONFIG_BOOKE
 595         {MSR_CE,        "CE"},
 596 #endif
 597         {MSR_EE,        "EE"},
 598         {MSR_PR,        "PR"},
 599         {MSR_FP,        "FP"},
 600         {MSR_ME,        "ME"},
 601 #ifdef CONFIG_BOOKE
 602         {MSR_DE,        "DE"},
 603 #else
 604         {MSR_SE,        "SE"},
 605         {MSR_BE,        "BE"},
 606 #endif
 607         {MSR_IR,        "IR"},
 608         {MSR_DR,        "DR"},
 609         {MSR_PMM,       "PMM"},
 610 #ifndef CONFIG_BOOKE
 611         {MSR_RI,        "RI"},
 612         {MSR_LE,        "LE"},
 613 #endif
 614         {0,             NULL}
 615 };
 616
 617 static void printbits(unsigned long val, struct regbit *bits)
 618 {
 619         const char *sep = "";
 620
 621         printk("<");
 622         for (; bits->bit; ++bits)
 623                 if (val & bits->bit) {
 624                         printk("%s%s", sep, bits->name);
 625                         sep = ",";
 626                 }
 627         printk(">");
 628 }
 629
 630 #ifdef CONFIG_PPC64
 631 #define REG             "%016lx"
 632 #define REGS_PER_LINE   4
 633 #define LAST_VOLATILE   13
 634 #else
 635 #define REG             "%08lx"
 636 #define REGS_PER_LINE   8
 637 #define LAST_VOLATILE   12
 638 #endif
 639
 640 void show_regs(struct pt_regs * regs)
 641 {
 642         int i, trap;
 643
 644         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
 645                regs->nip, regs->link, regs->ctr);
 646         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
 647                regs, regs->trap, print_tainted(), init_utsname()->release);
 648         printk("MSR: "REG" ", regs->msr);
 649         printbits(regs->msr, msr_bits);
 650         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
 651 #ifdef CONFIG_PPC64
 652         printk("SOFTE: %ld\n", regs->softe);
 653 #endif
 654         trap = TRAP(regs);
 655         if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
 656                 printk("CFAR: "REG"\n", regs->orig_gpr3);
 657         if (trap == 0x300 || trap == 0x600)
 658 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 659                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
 660 #else
 661                 printk("DAR: "REG", DSISR: %08lx\n", regs->dar, regs->dsisr);
 662 #endif
 663         printk("TASK = %p[%d] '%s' THREAD: %p",
 664                current, task_pid_nr(current), current->comm, task_thread_info(current));
 665
 666 #ifdef CONFIG_SMP
 667         printk(" CPU: %d", raw_smp_processor_id());
 668 #endif /* CONFIG_SMP */
 669
 670         for (i = 0;  i < 32;  i++) {
 671                 if ((i % REGS_PER_LINE) == 0)
 672                         printk("\nGPR%02d: ", i);
 673                 printk(REG " ", regs->gpr[i]);
 674                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
 675                         break;
 676         }
 677         printk("\n");
 678 #ifdef CONFIG_KALLSYMS
 679         /*
 680          * Lookup NIP late so we have the best change of getting the
 681          * above info out without failing
 682          */
 683         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
 684         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
 685 #endif
 686         show_stack(current, (unsigned long *) regs->gpr[1]);
 687         if (!user_mode(regs))
 688                 show_instructions(regs);
 689 }
 690
 691 void exit_thread(void)
 692 {
 693         discard_lazy_cpu_state();
 694 }
 695
 696 void flush_thread(void)
 697 {
 698         discard_lazy_cpu_state();
 699
 700 #ifdef CONFIG_HAVE_HW_BREAKPOINT
 701         flush_ptrace_hw_breakpoint(current);
 702 #else /* CONFIG_HAVE_HW_BREAKPOINT */
 703         set_debug_reg_defaults(&current->thread);
 704 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 705 }
 706
 707 void
 708 release_thread(struct task_struct *t)
 709 {
 710 }
 711
 712 /*
 713  * this gets called so that we can store coprocessor state into memory and
 714  * copy the current task into the new thread.
 715  */
 716 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 717 {
 718         flush_fp_to_thread(src);
 719         flush_altivec_to_thread(src);
 720         flush_vsx_to_thread(src);
 721         flush_spe_to_thread(src);
 722 #ifdef CONFIG_HAVE_HW_BREAKPOINT
 723         flush_ptrace_hw_breakpoint(src);
 724 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
 725
 726         *dst = *src;
 727         return 0;
 728 }
 729
 730 /*
 731  * Copy a thread..
 732  */
 733 extern unsigned long dscr_default; /* defined in arch/powerpc/kernel/sysfs.c */
 734
 735 int copy_thread(unsigned long clone_flags, unsigned long usp,
 736                 unsigned long arg, struct task_struct *p)
 737 {
 738         struct pt_regs *childregs, *kregs;
 739         extern void ret_from_fork(void);
 740         extern void ret_from_kernel_thread(void);
 741         void (*f)(void);
 742         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 743
 744         /* Copy registers */
 745         sp -= sizeof(struct pt_regs);
 746         childregs = (struct pt_regs *) sp;
 747         if (unlikely(p->flags & PF_KTHREAD)) {
 748                 struct thread_info *ti = (void *)task_stack_page(p);
 749                 memset(childregs, 0, sizeof(struct pt_regs));
 750                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
 751                 childregs->gpr[14] = usp;       /* function */
 752 #ifdef CONFIG_PPC64
 753                 clear_tsk_thread_flag(p, TIF_32BIT);
 754                 childregs->softe = 1;
 755 #endif
 756                 childregs->gpr[15] = arg;
 757                 p->thread.regs = NULL;  /* no user register state */
 758                 ti->flags |= _TIF_RESTOREALL;
 759                 f = ret_from_kernel_thread;
 760         } else {
 761                 struct pt_regs *regs = current_pt_regs();
 762                 CHECK_FULL_REGS(regs);
 763                 *childregs = *regs;
 764                 if (usp)
 765                         childregs->gpr[1] = usp;
 766                 p->thread.regs = childregs;
 767                 childregs->gpr[3] = 0;  /* Result from fork() */
 768                 if (clone_flags & CLONE_SETTLS) {
 769 #ifdef CONFIG_PPC64
 770                         if (!is_32bit_task())
 771                                 childregs->gpr[13] = childregs->gpr[6];
 772                         else
 773 #endif
 774                                 childregs->gpr[2] = childregs->gpr[6];
 775                 }
 776
 777                 f = ret_from_fork;
 778         }
 779         sp -= STACK_FRAME_OVERHEAD;
 780
 781         /*
 782          * The way this works is that at some point in the future
 783          * some task will call _switch to switch to the new task.
 784          * That will pop off the stack frame created below and start
 785          * the new task running at ret_from_fork.  The new task will
 786          * do some house keeping and then return from the fork or clone
 787          * system call, using the stack frame created above.
 788          */
 789         sp -= sizeof(struct pt_regs);
 790         kregs = (struct pt_regs *) sp;
 791         sp -= STACK_FRAME_OVERHEAD;
 792         p->thread.ksp = sp;
 793         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
 794                                 _ALIGN_UP(sizeof(struct thread_info), 16);
 795
 796 #ifdef CONFIG_PPC_STD_MMU_64
 797         if (mmu_has_feature(MMU_FTR_SLB)) {
 798                 unsigned long sp_vsid;
 799                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
 800
 801                 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
 802                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
 803                                 << SLB_VSID_SHIFT_1T;
 804                 else
 805                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
 806                                 << SLB_VSID_SHIFT;
 807                 sp_vsid |= SLB_VSID_KERNEL | llp;
 808                 p->thread.ksp_vsid = sp_vsid;
 809         }
 810 #endif /* CONFIG_PPC_STD_MMU_64 */
 811 #ifdef CONFIG_PPC64
 812         if (cpu_has_feature(CPU_FTR_DSCR)) {
 813                 p->thread.dscr_inherit = current->thread.dscr_inherit;
 814                 p->thread.dscr = current->thread.dscr;
 815         }
 816 #endif
 817         /*
 818          * The PPC64 ABI makes use of a TOC to contain function
 819          * pointers.  The function (ret_from_except) is actually a pointer
 820          * to the TOC entry.  The first entry is a pointer to the actual
 821          * function.
 822          */
 823 #ifdef CONFIG_PPC64
 824         kregs->nip = *((unsigned long *)f);
 825 #else
 826         kregs->nip = (unsigned long)f;
 827 #endif
 828         return 0;
 829 }
 830
 831 /*
 832  * Set up a thread for executing a new program
 833  */
 834 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 835 {
 836 #ifdef CONFIG_PPC64
 837         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
 838 #endif
 839
 840         /*
 841          * If we exec out of a kernel thread then thread.regs will not be
 842          * set.  Do it now.
 843          */
 844         if (!current->thread.regs) {
 845                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
 846                 current->thread.regs = regs - 1;
 847         }
 848
 849         memset(regs->gpr, 0, sizeof(regs->gpr));
 850         regs->ctr = 0;
 851         regs->link = 0;
 852         regs->xer = 0;
 853         regs->ccr = 0;
 854         regs->gpr[1] = sp;
 855
 856         /*
 857          * We have just cleared all the nonvolatile GPRs, so make
 858          * FULL_REGS(regs) return true.  This is necessary to allow
 859          * ptrace to examine the thread immediately after exec.
 860          */
 861         regs->trap &= ~1UL;
 862
 863 #ifdef CONFIG_PPC32
 864         regs->mq = 0;
 865         regs->nip = start;
 866         regs->msr = MSR_USER;
 867 #else
 868         if (!is_32bit_task()) {
 869                 unsigned long entry, toc;
 870
 871                 /* start is a relocated pointer to the function descriptor for
 872                  * the elf _start routine.  The first entry in the function
 873                  * descriptor is the entry address of _start and the second
 874                  * entry is the TOC value we need to use.
 875                  */
 876                 __get_user(entry, (unsigned long __user *)start);
 877                 __get_user(toc, (unsigned long __user *)start+1);
 878
 879                 /* Check whether the e_entry function descriptor entries
 880                  * need to be relocated before we can use them.
 881                  */
 882                 if (load_addr != 0) {
 883                         entry += load_addr;
 884                         toc   += load_addr;
 885                 }
 886                 regs->nip = entry;
 887                 regs->gpr[2] = toc;
 888                 regs->msr = MSR_USER64;
 889         } else {
 890                 regs->nip = start;
 891                 regs->gpr[2] = 0;
 892                 regs->msr = MSR_USER32;
 893         }
 894 #endif
 895
 896         discard_lazy_cpu_state();
 897 #ifdef CONFIG_VSX
 898         current->thread.used_vsr = 0;
 899 #endif
 900         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
 901         current->thread.fpscr.val = 0;
 902 #ifdef CONFIG_ALTIVEC
 903         memset(current->thread.vr, 0, sizeof(current->thread.vr));
 904         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
 905         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
 906         current->thread.vrsave = 0;
 907         current->thread.used_vr = 0;
 908 #endif /* CONFIG_ALTIVEC */
 909 #ifdef CONFIG_SPE
 910         memset(current->thread.evr, 0, sizeof(current->thread.evr));
 911         current->thread.acc = 0;
 912         current->thread.spefscr = 0;
 913         current->thread.used_spe = 0;
 914 #endif /* CONFIG_SPE */
 915 }
 916
 917 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
 918                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
 919
 920 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
 921 {
 922         struct pt_regs *regs = tsk->thread.regs;
 923
 924         /* This is a bit hairy.  If we are an SPE enabled  processor
 925          * (have embedded fp) we store the IEEE exception enable flags in
 926          * fpexc_mode.  fpexc_mode is also used for setting FP exception
 927          * mode (asyn, precise, disabled) for 'Classic' FP. */
 928         if (val & PR_FP_EXC_SW_ENABLE) {
 929 #ifdef CONFIG_SPE
 930                 if (cpu_has_feature(CPU_FTR_SPE)) {
 931                         tsk->thread.fpexc_mode = val &
 932                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
 933                         return 0;
 934                 } else {
 935                         return -EINVAL;
 936                 }
 937 #else
 938                 return -EINVAL;
 939 #endif
 940         }
 941
 942         /* on a CONFIG_SPE this does not hurt us.  The bits that
 943          * __pack_fe01 use do not overlap with bits used for
 944          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
 945          * on CONFIG_SPE implementations are reserved so writing to
 946          * them does not change anything */
 947         if (val > PR_FP_EXC_PRECISE)
 948                 return -EINVAL;
 949         tsk->thread.fpexc_mode = __pack_fe01(val);
 950         if (regs != NULL && (regs->msr & MSR_FP) != 0)
 951                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
 952                         | tsk->thread.fpexc_mode;
 953         return 0;
 954 }
 955
 956 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
 957 {
 958         unsigned int val;
 959
 960         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
 961 #ifdef CONFIG_SPE
 962                 if (cpu_has_feature(CPU_FTR_SPE))
 963                         val = tsk->thread.fpexc_mode;
 964                 else
 965                         return -EINVAL;
 966 #else
 967                 return -EINVAL;
 968 #endif
 969         else
 970                 val = __unpack_fe01(tsk->thread.fpexc_mode);
 971         return put_user(val, (unsigned int __user *) adr);
 972 }
 973
 974 int set_endian(struct task_struct *tsk, unsigned int val)
 975 {
 976         struct pt_regs *regs = tsk->thread.regs;
 977
 978         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
 979             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
 980                 return -EINVAL;
 981
 982         if (regs == NULL)
 983                 return -EINVAL;
 984
 985         if (val == PR_ENDIAN_BIG)
 986                 regs->msr &= ~MSR_LE;
 987         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
 988                 regs->msr |= MSR_LE;
 989         else
 990                 return -EINVAL;
 991
 992         return 0;
 993 }
 994
 995 int get_endian(struct task_struct *tsk, unsigned long adr)
 996 {
 997         struct pt_regs *regs = tsk->thread.regs;
 998         unsigned int val;
 999
1000         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1001             !cpu_has_feature(CPU_FTR_REAL_LE))
1002                 return -EINVAL;
1003
1004         if (regs == NULL)
1005                 return -EINVAL;
1006
1007         if (regs->msr & MSR_LE) {
1008                 if (cpu_has_feature(CPU_FTR_REAL_LE))
1009                         val = PR_ENDIAN_LITTLE;
1010                 else
1011                         val = PR_ENDIAN_PPC_LITTLE;
1012         } else
1013                 val = PR_ENDIAN_BIG;
1014
1015         return put_user(val, (unsigned int __user *)adr);
1016 }
1017
1018 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1019 {
1020         tsk->thread.align_ctl = val;
1021         return 0;
1022 }
1023
1024 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1025 {
1026         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1027 }
1028
1029 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1030                                   unsigned long nbytes)
1031 {
1032         unsigned long stack_page;
1033         unsigned long cpu = task_cpu(p);
1034
1035         /*
1036          * Avoid crashing if the stack has overflowed and corrupted
1037          * task_cpu(p), which is in the thread_info struct.
1038          */
1039         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1040                 stack_page = (unsigned long) hardirq_ctx[cpu];
1041                 if (sp >= stack_page + sizeof(struct thread_struct)
1042                     && sp <= stack_page + THREAD_SIZE - nbytes)
1043                         return 1;
1044
1045                 stack_page = (unsigned long) softirq_ctx[cpu];
1046                 if (sp >= stack_page + sizeof(struct thread_struct)
1047                     && sp <= stack_page + THREAD_SIZE - nbytes)
1048                         return 1;
1049         }
1050         return 0;
1051 }
1052
1053 int validate_sp(unsigned long sp, struct task_struct *p,
1054                        unsigned long nbytes)
1055 {
1056         unsigned long stack_page = (unsigned long)task_stack_page(p);
1057
1058         if (sp >= stack_page + sizeof(struct thread_struct)
1059             && sp <= stack_page + THREAD_SIZE - nbytes)
1060                 return 1;
1061
1062         return valid_irq_stack(sp, p, nbytes);
1063 }
1064
1065 EXPORT_SYMBOL(validate_sp);
1066
1067 unsigned long get_wchan(struct task_struct *p)
1068 {
1069         unsigned long ip, sp;
1070         int count = 0;
1071
1072         if (!p || p == current || p->state == TASK_RUNNING)
1073                 return 0;
1074
1075         sp = p->thread.ksp;
1076         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1077                 return 0;
1078
1079         do {
1080                 sp = *(unsigned long *)sp;
1081                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1082                         return 0;
1083                 if (count > 0) {
1084                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1085                         if (!in_sched_functions(ip))
1086                                 return ip;
1087                 }
1088         } while (count++ < 16);
1089         return 0;
1090 }
1091
1092 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1093
1094 void show_stack(struct task_struct *tsk, unsigned long *stack)
1095 {
1096         unsigned long sp, ip, lr, newsp;
1097         int count = 0;
1098         int firstframe = 1;
1099 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1100         int curr_frame = current->curr_ret_stack;
1101         extern void return_to_handler(void);
1102         unsigned long rth = (unsigned long)return_to_handler;
1103         unsigned long mrth = -1;
1104 #ifdef CONFIG_PPC64
1105         extern void mod_return_to_handler(void);
1106         rth = *(unsigned long *)rth;
1107         mrth = (unsigned long)mod_return_to_handler;
1108         mrth = *(unsigned long *)mrth;
1109 #endif
1110 #endif
1111
1112         sp = (unsigned long) stack;
1113         if (tsk == NULL)
1114                 tsk = current;
1115         if (sp == 0) {
1116                 if (tsk == current)
1117                         asm("mr %0,1" : "=r" (sp));
1118                 else
1119                         sp = tsk->thread.ksp;
1120         }
1121
1122         lr = 0;
1123         printk("Call Trace:\n");
1124         do {
1125                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1126                         return;
1127
1128                 stack = (unsigned long *) sp;
1129                 newsp = stack[0];
1130                 ip = stack[STACK_FRAME_LR_SAVE];
1131                 if (!firstframe || ip != lr) {
1132                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1133 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1134                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1135                                 printk(" (%pS)",
1136                                        (void *)current->ret_stack[curr_frame].ret);
1137                                 curr_frame--;
1138                         }
1139 #endif
1140                         if (firstframe)
1141                                 printk(" (unreliable)");
1142                         printk("\n");
1143                 }
1144                 firstframe = 0;
1145
1146                 /*
1147                  * See if this is an exception frame.
1148                  * We look for the "regshere" marker in the current frame.
1149                  */
1150                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1151                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1152                         struct pt_regs *regs = (struct pt_regs *)
1153                                 (sp + STACK_FRAME_OVERHEAD);
1154                         lr = regs->link;
1155                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1156                                regs->trap, (void *)regs->nip, (void *)lr);
1157                         firstframe = 1;
1158                 }
1159
1160                 sp = newsp;
1161         } while (count++ < kstack_depth_to_print);
1162 }
1163
1164 void dump_stack(void)
1165 {
1166         show_stack(current, NULL);
1167 }
1168 EXPORT_SYMBOL(dump_stack);
1169
1170 #ifdef CONFIG_PPC64
1171 /* Called with hard IRQs off */
1172 void __ppc64_runlatch_on(void)
1173 {
1174         struct thread_info *ti = current_thread_info();
1175         unsigned long ctrl;
1176
1177         ctrl = mfspr(SPRN_CTRLF);
1178         ctrl |= CTRL_RUNLATCH;
1179         mtspr(SPRN_CTRLT, ctrl);
1180
1181         ti->local_flags |= _TLF_RUNLATCH;
1182 }
1183
1184 /* Called with hard IRQs off */
1185 void __ppc64_runlatch_off(void)
1186 {
1187         struct thread_info *ti = current_thread_info();
1188         unsigned long ctrl;
1189
1190         ti->local_flags &= ~_TLF_RUNLATCH;
1191
1192         ctrl = mfspr(SPRN_CTRLF);
1193         ctrl &= ~CTRL_RUNLATCH;
1194         mtspr(SPRN_CTRLT, ctrl);
1195 }
1196 #endif /* CONFIG_PPC64 */
1197
1198 unsigned long arch_align_stack(unsigned long sp)
1199 {
1200         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1201                 sp -= get_random_int() & ~PAGE_MASK;
1202         return sp & ~0xf;
1203 }
1204
1205 static inline unsigned long brk_rnd(void)
1206 {
1207         unsigned long rnd = 0;
1208
1209         /* 8MB for 32bit, 1GB for 64bit */
1210         if (is_32bit_task())
1211                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1212         else
1213                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1214
1215         return rnd << PAGE_SHIFT;
1216 }
1217
1218 unsigned long arch_randomize_brk(struct mm_struct *mm)
1219 {
1220         unsigned long base = mm->brk;
1221         unsigned long ret;
1222
1223 #ifdef CONFIG_PPC_STD_MMU_64
1224         /*
1225          * If we are using 1TB segments and we are allowed to randomise
1226          * the heap, we can put it above 1TB so it is backed by a 1TB
1227          * segment. Otherwise the heap will be in the bottom 1TB
1228          * which always uses 256MB segments and this may result in a
1229          * performance penalty.
1230          */
1231         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1232                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1233 #endif
1234
1235         ret = PAGE_ALIGN(base + brk_rnd());
1236
1237         if (ret < mm->brk)
1238                 return mm->brk;
1239
1240         return ret;
1241 }
1242
1243 unsigned long randomize_et_dyn(unsigned long base)
1244 {
1245         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1246
1247         if (ret < base)
1248                 return base;
1249
1250         return ret;
1251 }