Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6
[linux-2.6] / arch / powerpc / kernel / process.c
1 /*
2  *  Derived from "arch/i386/kernel/process.c"
3  *    Copyright (C) 1995  Linus Torvalds
4  *
5  *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
6  *  Paul Mackerras (paulus@cs.anu.edu.au)
7  *
8  *  PowerPC version
9  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
10  *
11  *  This program is free software; you can redistribute it and/or
12  *  modify it under the terms of the GNU General Public License
13  *  as published by the Free Software Foundation; either version
14  *  2 of the License, or (at your option) any later version.
15  */
16
17 #include <linux/errno.h>
18 #include <linux/sched.h>
19 #include <linux/kernel.h>
20 #include <linux/mm.h>
21 #include <linux/smp.h>
22 #include <linux/stddef.h>
23 #include <linux/unistd.h>
24 #include <linux/ptrace.h>
25 #include <linux/slab.h>
26 #include <linux/user.h>
27 #include <linux/elf.h>
28 #include <linux/init.h>
29 #include <linux/prctl.h>
30 #include <linux/init_task.h>
31 #include <linux/module.h>
32 #include <linux/kallsyms.h>
33 #include <linux/mqueue.h>
34 #include <linux/hardirq.h>
35 #include <linux/utsname.h>
36
37 #include <asm/pgtable.h>
38 #include <asm/uaccess.h>
39 #include <asm/system.h>
40 #include <asm/io.h>
41 #include <asm/processor.h>
42 #include <asm/mmu.h>
43 #include <asm/prom.h>
44 #include <asm/machdep.h>
45 #include <asm/time.h>
46 #include <asm/syscalls.h>
47 #ifdef CONFIG_PPC64
48 #include <asm/firmware.h>
49 #endif
50 #include <linux/kprobes.h>
51 #include <linux/kdebug.h>
52
53 extern unsigned long _get_SP(void);
54
55 #ifndef CONFIG_SMP
56 struct task_struct *last_task_used_math = NULL;
57 struct task_struct *last_task_used_altivec = NULL;
58 struct task_struct *last_task_used_vsx = NULL;
59 struct task_struct *last_task_used_spe = NULL;
60 #endif
61
62 /*
63  * Make sure the floating-point register state in the
64  * the thread_struct is up to date for task tsk.
65  */
66 void flush_fp_to_thread(struct task_struct *tsk)
67 {
68         if (tsk->thread.regs) {
69                 /*
70                  * We need to disable preemption here because if we didn't,
71                  * another process could get scheduled after the regs->msr
72                  * test but before we have finished saving the FP registers
73                  * to the thread_struct.  That process could take over the
74                  * FPU, and then when we get scheduled again we would store
75                  * bogus values for the remaining FP registers.
76                  */
77                 preempt_disable();
78                 if (tsk->thread.regs->msr & MSR_FP) {
79 #ifdef CONFIG_SMP
80                         /*
81                          * This should only ever be called for current or
82                          * for a stopped child process.  Since we save away
83                          * the FP register state on context switch on SMP,
84                          * there is something wrong if a stopped child appears
85                          * to still have its FP state in the CPU registers.
86                          */
87                         BUG_ON(tsk != current);
88 #endif
89                         giveup_fpu(tsk);
90                 }
91                 preempt_enable();
92         }
93 }
94
95 void enable_kernel_fp(void)
96 {
97         WARN_ON(preemptible());
98
99 #ifdef CONFIG_SMP
100         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
101                 giveup_fpu(current);
102         else
103                 giveup_fpu(NULL);       /* just enables FP for kernel */
104 #else
105         giveup_fpu(last_task_used_math);
106 #endif /* CONFIG_SMP */
107 }
108 EXPORT_SYMBOL(enable_kernel_fp);
109
110 #ifdef CONFIG_ALTIVEC
111 void enable_kernel_altivec(void)
112 {
113         WARN_ON(preemptible());
114
115 #ifdef CONFIG_SMP
116         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
117                 giveup_altivec(current);
118         else
119                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
120 #else
121         giveup_altivec(last_task_used_altivec);
122 #endif /* CONFIG_SMP */
123 }
124 EXPORT_SYMBOL(enable_kernel_altivec);
125
126 /*
127  * Make sure the VMX/Altivec register state in the
128  * the thread_struct is up to date for task tsk.
129  */
130 void flush_altivec_to_thread(struct task_struct *tsk)
131 {
132         if (tsk->thread.regs) {
133                 preempt_disable();
134                 if (tsk->thread.regs->msr & MSR_VEC) {
135 #ifdef CONFIG_SMP
136                         BUG_ON(tsk != current);
137 #endif
138                         giveup_altivec(tsk);
139                 }
140                 preempt_enable();
141         }
142 }
143 #endif /* CONFIG_ALTIVEC */
144
145 #ifdef CONFIG_VSX
146 #if 0
147 /* not currently used, but some crazy RAID module might want to later */
148 void enable_kernel_vsx(void)
149 {
150         WARN_ON(preemptible());
151
152 #ifdef CONFIG_SMP
153         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
154                 giveup_vsx(current);
155         else
156                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
157 #else
158         giveup_vsx(last_task_used_vsx);
159 #endif /* CONFIG_SMP */
160 }
161 EXPORT_SYMBOL(enable_kernel_vsx);
162 #endif
163
164 void giveup_vsx(struct task_struct *tsk)
165 {
166         giveup_fpu(tsk);
167         giveup_altivec(tsk);
168         __giveup_vsx(tsk);
169 }
170
171 void flush_vsx_to_thread(struct task_struct *tsk)
172 {
173         if (tsk->thread.regs) {
174                 preempt_disable();
175                 if (tsk->thread.regs->msr & MSR_VSX) {
176 #ifdef CONFIG_SMP
177                         BUG_ON(tsk != current);
178 #endif
179                         giveup_vsx(tsk);
180                 }
181                 preempt_enable();
182         }
183 }
184 #endif /* CONFIG_VSX */
185
186 #ifdef CONFIG_SPE
187
188 void enable_kernel_spe(void)
189 {
190         WARN_ON(preemptible());
191
192 #ifdef CONFIG_SMP
193         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
194                 giveup_spe(current);
195         else
196                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
197 #else
198         giveup_spe(last_task_used_spe);
199 #endif /* __SMP __ */
200 }
201 EXPORT_SYMBOL(enable_kernel_spe);
202
203 void flush_spe_to_thread(struct task_struct *tsk)
204 {
205         if (tsk->thread.regs) {
206                 preempt_disable();
207                 if (tsk->thread.regs->msr & MSR_SPE) {
208 #ifdef CONFIG_SMP
209                         BUG_ON(tsk != current);
210 #endif
211                         giveup_spe(tsk);
212                 }
213                 preempt_enable();
214         }
215 }
216 #endif /* CONFIG_SPE */
217
218 #ifndef CONFIG_SMP
219 /*
220  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
221  * and the current task has some state, discard it.
222  */
223 void discard_lazy_cpu_state(void)
224 {
225         preempt_disable();
226         if (last_task_used_math == current)
227                 last_task_used_math = NULL;
228 #ifdef CONFIG_ALTIVEC
229         if (last_task_used_altivec == current)
230                 last_task_used_altivec = NULL;
231 #endif /* CONFIG_ALTIVEC */
232 #ifdef CONFIG_VSX
233         if (last_task_used_vsx == current)
234                 last_task_used_vsx = NULL;
235 #endif /* CONFIG_VSX */
236 #ifdef CONFIG_SPE
237         if (last_task_used_spe == current)
238                 last_task_used_spe = NULL;
239 #endif
240         preempt_enable();
241 }
242 #endif /* CONFIG_SMP */
243
244 void do_dabr(struct pt_regs *regs, unsigned long address,
245                     unsigned long error_code)
246 {
247         siginfo_t info;
248
249         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
250                         11, SIGSEGV) == NOTIFY_STOP)
251                 return;
252
253         if (debugger_dabr_match(regs))
254                 return;
255
256         /* Clear the DAC and struct entries.  One shot trigger */
257 #if defined(CONFIG_BOOKE)
258         mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~(DBSR_DAC1R | DBSR_DAC1W
259                                                         | DBCR0_IDM));
260 #endif
261
262         /* Clear the DABR */
263         set_dabr(0);
264
265         /* Deliver the signal to userspace */
266         info.si_signo = SIGTRAP;
267         info.si_errno = 0;
268         info.si_code = TRAP_HWBKPT;
269         info.si_addr = (void __user *)address;
270         force_sig_info(SIGTRAP, &info, current);
271 }
272
273 static DEFINE_PER_CPU(unsigned long, current_dabr);
274
275 int set_dabr(unsigned long dabr)
276 {
277         __get_cpu_var(current_dabr) = dabr;
278
279 #ifdef CONFIG_PPC_MERGE         /* XXX for now */
280         if (ppc_md.set_dabr)
281                 return ppc_md.set_dabr(dabr);
282 #endif
283
284         /* XXX should we have a CPU_FTR_HAS_DABR ? */
285 #if defined(CONFIG_PPC64) || defined(CONFIG_6xx)
286         mtspr(SPRN_DABR, dabr);
287 #endif
288
289 #if defined(CONFIG_BOOKE)
290         mtspr(SPRN_DAC1, dabr);
291 #endif
292
293         return 0;
294 }
295
296 #ifdef CONFIG_PPC64
297 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
298 #endif
299
300 struct task_struct *__switch_to(struct task_struct *prev,
301         struct task_struct *new)
302 {
303         struct thread_struct *new_thread, *old_thread;
304         unsigned long flags;
305         struct task_struct *last;
306
307 #ifdef CONFIG_SMP
308         /* avoid complexity of lazy save/restore of fpu
309          * by just saving it every time we switch out if
310          * this task used the fpu during the last quantum.
311          *
312          * If it tries to use the fpu again, it'll trap and
313          * reload its fp regs.  So we don't have to do a restore
314          * every switch, just a save.
315          *  -- Cort
316          */
317         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
318                 giveup_fpu(prev);
319 #ifdef CONFIG_ALTIVEC
320         /*
321          * If the previous thread used altivec in the last quantum
322          * (thus changing altivec regs) then save them.
323          * We used to check the VRSAVE register but not all apps
324          * set it, so we don't rely on it now (and in fact we need
325          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
326          *
327          * On SMP we always save/restore altivec regs just to avoid the
328          * complexity of changing processors.
329          *  -- Cort
330          */
331         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
332                 giveup_altivec(prev);
333 #endif /* CONFIG_ALTIVEC */
334 #ifdef CONFIG_VSX
335         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
336                 /* VMX and FPU registers are already save here */
337                 __giveup_vsx(prev);
338 #endif /* CONFIG_VSX */
339 #ifdef CONFIG_SPE
340         /*
341          * If the previous thread used spe in the last quantum
342          * (thus changing spe regs) then save them.
343          *
344          * On SMP we always save/restore spe regs just to avoid the
345          * complexity of changing processors.
346          */
347         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
348                 giveup_spe(prev);
349 #endif /* CONFIG_SPE */
350
351 #else  /* CONFIG_SMP */
352 #ifdef CONFIG_ALTIVEC
353         /* Avoid the trap.  On smp this this never happens since
354          * we don't set last_task_used_altivec -- Cort
355          */
356         if (new->thread.regs && last_task_used_altivec == new)
357                 new->thread.regs->msr |= MSR_VEC;
358 #endif /* CONFIG_ALTIVEC */
359 #ifdef CONFIG_VSX
360         if (new->thread.regs && last_task_used_vsx == new)
361                 new->thread.regs->msr |= MSR_VSX;
362 #endif /* CONFIG_VSX */
363 #ifdef CONFIG_SPE
364         /* Avoid the trap.  On smp this this never happens since
365          * we don't set last_task_used_spe
366          */
367         if (new->thread.regs && last_task_used_spe == new)
368                 new->thread.regs->msr |= MSR_SPE;
369 #endif /* CONFIG_SPE */
370
371 #endif /* CONFIG_SMP */
372
373         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
374                 set_dabr(new->thread.dabr);
375
376 #if defined(CONFIG_BOOKE)
377         /* If new thread DAC (HW breakpoint) is the same then leave it */
378         if (new->thread.dabr)
379                 set_dabr(new->thread.dabr);
380 #endif
381
382         new_thread = &new->thread;
383         old_thread = &current->thread;
384
385 #ifdef CONFIG_PPC64
386         /*
387          * Collect processor utilization data per process
388          */
389         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
390                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
391                 long unsigned start_tb, current_tb;
392                 start_tb = old_thread->start_tb;
393                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
394                 old_thread->accum_tb += (current_tb - start_tb);
395                 new_thread->start_tb = current_tb;
396         }
397 #endif
398
399         local_irq_save(flags);
400
401         account_system_vtime(current);
402         account_process_vtime(current);
403         calculate_steal_time();
404
405         /*
406          * We can't take a PMU exception inside _switch() since there is a
407          * window where the kernel stack SLB and the kernel stack are out
408          * of sync. Hard disable here.
409          */
410         hard_irq_disable();
411         last = _switch(old_thread, new_thread);
412
413         local_irq_restore(flags);
414
415         return last;
416 }
417
418 static int instructions_to_print = 16;
419
420 static void show_instructions(struct pt_regs *regs)
421 {
422         int i;
423         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
424                         sizeof(int));
425
426         printk("Instruction dump:");
427
428         for (i = 0; i < instructions_to_print; i++) {
429                 int instr;
430
431                 if (!(i % 8))
432                         printk("\n");
433
434 #if !defined(CONFIG_BOOKE)
435                 /* If executing with the IMMU off, adjust pc rather
436                  * than print XXXXXXXX.
437                  */
438                 if (!(regs->msr & MSR_IR))
439                         pc = (unsigned long)phys_to_virt(pc);
440 #endif
441
442                 /* We use __get_user here *only* to avoid an OOPS on a
443                  * bad address because the pc *should* only be a
444                  * kernel address.
445                  */
446                 if (!__kernel_text_address(pc) ||
447                      __get_user(instr, (unsigned int __user *)pc)) {
448                         printk("XXXXXXXX ");
449                 } else {
450                         if (regs->nip == pc)
451                                 printk("<%08x> ", instr);
452                         else
453                                 printk("%08x ", instr);
454                 }
455
456                 pc += sizeof(int);
457         }
458
459         printk("\n");
460 }
461
462 static struct regbit {
463         unsigned long bit;
464         const char *name;
465 } msr_bits[] = {
466         {MSR_EE,        "EE"},
467         {MSR_PR,        "PR"},
468         {MSR_FP,        "FP"},
469         {MSR_VEC,       "VEC"},
470         {MSR_VSX,       "VSX"},
471         {MSR_ME,        "ME"},
472         {MSR_IR,        "IR"},
473         {MSR_DR,        "DR"},
474         {0,             NULL}
475 };
476
477 static void printbits(unsigned long val, struct regbit *bits)
478 {
479         const char *sep = "";
480
481         printk("<");
482         for (; bits->bit; ++bits)
483                 if (val & bits->bit) {
484                         printk("%s%s", sep, bits->name);
485                         sep = ",";
486                 }
487         printk(">");
488 }
489
490 #ifdef CONFIG_PPC64
491 #define REG             "%016lx"
492 #define REGS_PER_LINE   4
493 #define LAST_VOLATILE   13
494 #else
495 #define REG             "%08lx"
496 #define REGS_PER_LINE   8
497 #define LAST_VOLATILE   12
498 #endif
499
500 void show_regs(struct pt_regs * regs)
501 {
502         int i, trap;
503
504         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
505                regs->nip, regs->link, regs->ctr);
506         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
507                regs, regs->trap, print_tainted(), init_utsname()->release);
508         printk("MSR: "REG" ", regs->msr);
509         printbits(regs->msr, msr_bits);
510         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
511         trap = TRAP(regs);
512         if (trap == 0x300 || trap == 0x600)
513 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
514                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
515 #else
516                 printk("DAR: "REG", DSISR: "REG"\n", regs->dar, regs->dsisr);
517 #endif
518         printk("TASK = %p[%d] '%s' THREAD: %p",
519                current, task_pid_nr(current), current->comm, task_thread_info(current));
520
521 #ifdef CONFIG_SMP
522         printk(" CPU: %d", raw_smp_processor_id());
523 #endif /* CONFIG_SMP */
524
525         for (i = 0;  i < 32;  i++) {
526                 if ((i % REGS_PER_LINE) == 0)
527                         printk("\n" KERN_INFO "GPR%02d: ", i);
528                 printk(REG " ", regs->gpr[i]);
529                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
530                         break;
531         }
532         printk("\n");
533 #ifdef CONFIG_KALLSYMS
534         /*
535          * Lookup NIP late so we have the best change of getting the
536          * above info out without failing
537          */
538         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
539         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
540 #endif
541         show_stack(current, (unsigned long *) regs->gpr[1]);
542         if (!user_mode(regs))
543                 show_instructions(regs);
544 }
545
546 void exit_thread(void)
547 {
548         discard_lazy_cpu_state();
549 }
550
551 void flush_thread(void)
552 {
553 #ifdef CONFIG_PPC64
554         struct thread_info *t = current_thread_info();
555
556         if (test_ti_thread_flag(t, TIF_ABI_PENDING)) {
557                 clear_ti_thread_flag(t, TIF_ABI_PENDING);
558                 if (test_ti_thread_flag(t, TIF_32BIT))
559                         clear_ti_thread_flag(t, TIF_32BIT);
560                 else
561                         set_ti_thread_flag(t, TIF_32BIT);
562         }
563 #endif
564
565         discard_lazy_cpu_state();
566
567         if (current->thread.dabr) {
568                 current->thread.dabr = 0;
569                 set_dabr(0);
570
571 #if defined(CONFIG_BOOKE)
572                 current->thread.dbcr0 &= ~(DBSR_DAC1R | DBSR_DAC1W);
573 #endif
574         }
575 }
576
577 void
578 release_thread(struct task_struct *t)
579 {
580 }
581
582 /*
583  * This gets called before we allocate a new thread and copy
584  * the current task into it.
585  */
586 void prepare_to_copy(struct task_struct *tsk)
587 {
588         flush_fp_to_thread(current);
589         flush_altivec_to_thread(current);
590         flush_vsx_to_thread(current);
591         flush_spe_to_thread(current);
592 }
593
594 /*
595  * Copy a thread..
596  */
597 int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
598                 unsigned long unused, struct task_struct *p,
599                 struct pt_regs *regs)
600 {
601         struct pt_regs *childregs, *kregs;
602         extern void ret_from_fork(void);
603         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
604
605         CHECK_FULL_REGS(regs);
606         /* Copy registers */
607         sp -= sizeof(struct pt_regs);
608         childregs = (struct pt_regs *) sp;
609         *childregs = *regs;
610         if ((childregs->msr & MSR_PR) == 0) {
611                 /* for kernel thread, set `current' and stackptr in new task */
612                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
613 #ifdef CONFIG_PPC32
614                 childregs->gpr[2] = (unsigned long) p;
615 #else
616                 clear_tsk_thread_flag(p, TIF_32BIT);
617 #endif
618                 p->thread.regs = NULL;  /* no user register state */
619         } else {
620                 childregs->gpr[1] = usp;
621                 p->thread.regs = childregs;
622                 if (clone_flags & CLONE_SETTLS) {
623 #ifdef CONFIG_PPC64
624                         if (!test_thread_flag(TIF_32BIT))
625                                 childregs->gpr[13] = childregs->gpr[6];
626                         else
627 #endif
628                                 childregs->gpr[2] = childregs->gpr[6];
629                 }
630         }
631         childregs->gpr[3] = 0;  /* Result from fork() */
632         sp -= STACK_FRAME_OVERHEAD;
633
634         /*
635          * The way this works is that at some point in the future
636          * some task will call _switch to switch to the new task.
637          * That will pop off the stack frame created below and start
638          * the new task running at ret_from_fork.  The new task will
639          * do some house keeping and then return from the fork or clone
640          * system call, using the stack frame created above.
641          */
642         sp -= sizeof(struct pt_regs);
643         kregs = (struct pt_regs *) sp;
644         sp -= STACK_FRAME_OVERHEAD;
645         p->thread.ksp = sp;
646         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
647                                 _ALIGN_UP(sizeof(struct thread_info), 16);
648
649 #ifdef CONFIG_PPC64
650         if (cpu_has_feature(CPU_FTR_SLB)) {
651                 unsigned long sp_vsid;
652                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
653
654                 if (cpu_has_feature(CPU_FTR_1T_SEGMENT))
655                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
656                                 << SLB_VSID_SHIFT_1T;
657                 else
658                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
659                                 << SLB_VSID_SHIFT;
660                 sp_vsid |= SLB_VSID_KERNEL | llp;
661                 p->thread.ksp_vsid = sp_vsid;
662         }
663
664         /*
665          * The PPC64 ABI makes use of a TOC to contain function 
666          * pointers.  The function (ret_from_except) is actually a pointer
667          * to the TOC entry.  The first entry is a pointer to the actual
668          * function.
669          */
670         kregs->nip = *((unsigned long *)ret_from_fork);
671 #else
672         kregs->nip = (unsigned long)ret_from_fork;
673 #endif
674
675         return 0;
676 }
677
678 /*
679  * Set up a thread for executing a new program
680  */
681 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
682 {
683 #ifdef CONFIG_PPC64
684         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
685 #endif
686
687         set_fs(USER_DS);
688
689         /*
690          * If we exec out of a kernel thread then thread.regs will not be
691          * set.  Do it now.
692          */
693         if (!current->thread.regs) {
694                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
695                 current->thread.regs = regs - 1;
696         }
697
698         memset(regs->gpr, 0, sizeof(regs->gpr));
699         regs->ctr = 0;
700         regs->link = 0;
701         regs->xer = 0;
702         regs->ccr = 0;
703         regs->gpr[1] = sp;
704
705         /*
706          * We have just cleared all the nonvolatile GPRs, so make
707          * FULL_REGS(regs) return true.  This is necessary to allow
708          * ptrace to examine the thread immediately after exec.
709          */
710         regs->trap &= ~1UL;
711
712 #ifdef CONFIG_PPC32
713         regs->mq = 0;
714         regs->nip = start;
715         regs->msr = MSR_USER;
716 #else
717         if (!test_thread_flag(TIF_32BIT)) {
718                 unsigned long entry, toc;
719
720                 /* start is a relocated pointer to the function descriptor for
721                  * the elf _start routine.  The first entry in the function
722                  * descriptor is the entry address of _start and the second
723                  * entry is the TOC value we need to use.
724                  */
725                 __get_user(entry, (unsigned long __user *)start);
726                 __get_user(toc, (unsigned long __user *)start+1);
727
728                 /* Check whether the e_entry function descriptor entries
729                  * need to be relocated before we can use them.
730                  */
731                 if (load_addr != 0) {
732                         entry += load_addr;
733                         toc   += load_addr;
734                 }
735                 regs->nip = entry;
736                 regs->gpr[2] = toc;
737                 regs->msr = MSR_USER64;
738         } else {
739                 regs->nip = start;
740                 regs->gpr[2] = 0;
741                 regs->msr = MSR_USER32;
742         }
743 #endif
744
745         discard_lazy_cpu_state();
746 #ifdef CONFIG_VSX
747         current->thread.used_vsr = 0;
748 #endif
749         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
750         current->thread.fpscr.val = 0;
751 #ifdef CONFIG_ALTIVEC
752         memset(current->thread.vr, 0, sizeof(current->thread.vr));
753         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
754         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
755         current->thread.vrsave = 0;
756         current->thread.used_vr = 0;
757 #endif /* CONFIG_ALTIVEC */
758 #ifdef CONFIG_SPE
759         memset(current->thread.evr, 0, sizeof(current->thread.evr));
760         current->thread.acc = 0;
761         current->thread.spefscr = 0;
762         current->thread.used_spe = 0;
763 #endif /* CONFIG_SPE */
764 }
765
766 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
767                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
768
769 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
770 {
771         struct pt_regs *regs = tsk->thread.regs;
772
773         /* This is a bit hairy.  If we are an SPE enabled  processor
774          * (have embedded fp) we store the IEEE exception enable flags in
775          * fpexc_mode.  fpexc_mode is also used for setting FP exception
776          * mode (asyn, precise, disabled) for 'Classic' FP. */
777         if (val & PR_FP_EXC_SW_ENABLE) {
778 #ifdef CONFIG_SPE
779                 if (cpu_has_feature(CPU_FTR_SPE)) {
780                         tsk->thread.fpexc_mode = val &
781                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
782                         return 0;
783                 } else {
784                         return -EINVAL;
785                 }
786 #else
787                 return -EINVAL;
788 #endif
789         }
790
791         /* on a CONFIG_SPE this does not hurt us.  The bits that
792          * __pack_fe01 use do not overlap with bits used for
793          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
794          * on CONFIG_SPE implementations are reserved so writing to
795          * them does not change anything */
796         if (val > PR_FP_EXC_PRECISE)
797                 return -EINVAL;
798         tsk->thread.fpexc_mode = __pack_fe01(val);
799         if (regs != NULL && (regs->msr & MSR_FP) != 0)
800                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
801                         | tsk->thread.fpexc_mode;
802         return 0;
803 }
804
805 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
806 {
807         unsigned int val;
808
809         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
810 #ifdef CONFIG_SPE
811                 if (cpu_has_feature(CPU_FTR_SPE))
812                         val = tsk->thread.fpexc_mode;
813                 else
814                         return -EINVAL;
815 #else
816                 return -EINVAL;
817 #endif
818         else
819                 val = __unpack_fe01(tsk->thread.fpexc_mode);
820         return put_user(val, (unsigned int __user *) adr);
821 }
822
823 int set_endian(struct task_struct *tsk, unsigned int val)
824 {
825         struct pt_regs *regs = tsk->thread.regs;
826
827         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
828             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
829                 return -EINVAL;
830
831         if (regs == NULL)
832                 return -EINVAL;
833
834         if (val == PR_ENDIAN_BIG)
835                 regs->msr &= ~MSR_LE;
836         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
837                 regs->msr |= MSR_LE;
838         else
839                 return -EINVAL;
840
841         return 0;
842 }
843
844 int get_endian(struct task_struct *tsk, unsigned long adr)
845 {
846         struct pt_regs *regs = tsk->thread.regs;
847         unsigned int val;
848
849         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
850             !cpu_has_feature(CPU_FTR_REAL_LE))
851                 return -EINVAL;
852
853         if (regs == NULL)
854                 return -EINVAL;
855
856         if (regs->msr & MSR_LE) {
857                 if (cpu_has_feature(CPU_FTR_REAL_LE))
858                         val = PR_ENDIAN_LITTLE;
859                 else
860                         val = PR_ENDIAN_PPC_LITTLE;
861         } else
862                 val = PR_ENDIAN_BIG;
863
864         return put_user(val, (unsigned int __user *)adr);
865 }
866
867 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
868 {
869         tsk->thread.align_ctl = val;
870         return 0;
871 }
872
873 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
874 {
875         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
876 }
877
878 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
879
880 int sys_clone(unsigned long clone_flags, unsigned long usp,
881               int __user *parent_tidp, void __user *child_threadptr,
882               int __user *child_tidp, int p6,
883               struct pt_regs *regs)
884 {
885         CHECK_FULL_REGS(regs);
886         if (usp == 0)
887                 usp = regs->gpr[1];     /* stack pointer for child */
888 #ifdef CONFIG_PPC64
889         if (test_thread_flag(TIF_32BIT)) {
890                 parent_tidp = TRUNC_PTR(parent_tidp);
891                 child_tidp = TRUNC_PTR(child_tidp);
892         }
893 #endif
894         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
895 }
896
897 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
898              unsigned long p4, unsigned long p5, unsigned long p6,
899              struct pt_regs *regs)
900 {
901         CHECK_FULL_REGS(regs);
902         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
903 }
904
905 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
906               unsigned long p4, unsigned long p5, unsigned long p6,
907               struct pt_regs *regs)
908 {
909         CHECK_FULL_REGS(regs);
910         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
911                         regs, 0, NULL, NULL);
912 }
913
914 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
915                unsigned long a3, unsigned long a4, unsigned long a5,
916                struct pt_regs *regs)
917 {
918         int error;
919         char *filename;
920
921         filename = getname((char __user *) a0);
922         error = PTR_ERR(filename);
923         if (IS_ERR(filename))
924                 goto out;
925         flush_fp_to_thread(current);
926         flush_altivec_to_thread(current);
927         flush_spe_to_thread(current);
928         error = do_execve(filename, (char __user * __user *) a1,
929                           (char __user * __user *) a2, regs);
930         putname(filename);
931 out:
932         return error;
933 }
934
935 #ifdef CONFIG_IRQSTACKS
936 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
937                                   unsigned long nbytes)
938 {
939         unsigned long stack_page;
940         unsigned long cpu = task_cpu(p);
941
942         /*
943          * Avoid crashing if the stack has overflowed and corrupted
944          * task_cpu(p), which is in the thread_info struct.
945          */
946         if (cpu < NR_CPUS && cpu_possible(cpu)) {
947                 stack_page = (unsigned long) hardirq_ctx[cpu];
948                 if (sp >= stack_page + sizeof(struct thread_struct)
949                     && sp <= stack_page + THREAD_SIZE - nbytes)
950                         return 1;
951
952                 stack_page = (unsigned long) softirq_ctx[cpu];
953                 if (sp >= stack_page + sizeof(struct thread_struct)
954                     && sp <= stack_page + THREAD_SIZE - nbytes)
955                         return 1;
956         }
957         return 0;
958 }
959
960 #else
961 #define valid_irq_stack(sp, p, nb)      0
962 #endif /* CONFIG_IRQSTACKS */
963
964 int validate_sp(unsigned long sp, struct task_struct *p,
965                        unsigned long nbytes)
966 {
967         unsigned long stack_page = (unsigned long)task_stack_page(p);
968
969         if (sp >= stack_page + sizeof(struct thread_struct)
970             && sp <= stack_page + THREAD_SIZE - nbytes)
971                 return 1;
972
973         return valid_irq_stack(sp, p, nbytes);
974 }
975
976 EXPORT_SYMBOL(validate_sp);
977
978 unsigned long get_wchan(struct task_struct *p)
979 {
980         unsigned long ip, sp;
981         int count = 0;
982
983         if (!p || p == current || p->state == TASK_RUNNING)
984                 return 0;
985
986         sp = p->thread.ksp;
987         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
988                 return 0;
989
990         do {
991                 sp = *(unsigned long *)sp;
992                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
993                         return 0;
994                 if (count > 0) {
995                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
996                         if (!in_sched_functions(ip))
997                                 return ip;
998                 }
999         } while (count++ < 16);
1000         return 0;
1001 }
1002
1003 static int kstack_depth_to_print = 64;
1004
1005 void show_stack(struct task_struct *tsk, unsigned long *stack)
1006 {
1007         unsigned long sp, ip, lr, newsp;
1008         int count = 0;
1009         int firstframe = 1;
1010
1011         sp = (unsigned long) stack;
1012         if (tsk == NULL)
1013                 tsk = current;
1014         if (sp == 0) {
1015                 if (tsk == current)
1016                         asm("mr %0,1" : "=r" (sp));
1017                 else
1018                         sp = tsk->thread.ksp;
1019         }
1020
1021         lr = 0;
1022         printk("Call Trace:\n");
1023         do {
1024                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1025                         return;
1026
1027                 stack = (unsigned long *) sp;
1028                 newsp = stack[0];
1029                 ip = stack[STACK_FRAME_LR_SAVE];
1030                 if (!firstframe || ip != lr) {
1031                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1032                         if (firstframe)
1033                                 printk(" (unreliable)");
1034                         printk("\n");
1035                 }
1036                 firstframe = 0;
1037
1038                 /*
1039                  * See if this is an exception frame.
1040                  * We look for the "regshere" marker in the current frame.
1041                  */
1042                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1043                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1044                         struct pt_regs *regs = (struct pt_regs *)
1045                                 (sp + STACK_FRAME_OVERHEAD);
1046                         lr = regs->link;
1047                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1048                                regs->trap, (void *)regs->nip, (void *)lr);
1049                         firstframe = 1;
1050                 }
1051
1052                 sp = newsp;
1053         } while (count++ < kstack_depth_to_print);
1054 }
1055
1056 void dump_stack(void)
1057 {
1058         show_stack(current, NULL);
1059 }
1060 EXPORT_SYMBOL(dump_stack);
1061
1062 #ifdef CONFIG_PPC64
1063 void ppc64_runlatch_on(void)
1064 {
1065         unsigned long ctrl;
1066
1067         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1068                 HMT_medium();
1069
1070                 ctrl = mfspr(SPRN_CTRLF);
1071                 ctrl |= CTRL_RUNLATCH;
1072                 mtspr(SPRN_CTRLT, ctrl);
1073
1074                 set_thread_flag(TIF_RUNLATCH);
1075         }
1076 }
1077
1078 void ppc64_runlatch_off(void)
1079 {
1080         unsigned long ctrl;
1081
1082         if (cpu_has_feature(CPU_FTR_CTRL) && test_thread_flag(TIF_RUNLATCH)) {
1083                 HMT_medium();
1084
1085                 clear_thread_flag(TIF_RUNLATCH);
1086
1087                 ctrl = mfspr(SPRN_CTRLF);
1088                 ctrl &= ~CTRL_RUNLATCH;
1089                 mtspr(SPRN_CTRLT, ctrl);
1090         }
1091 }
1092 #endif
1093
1094 #if THREAD_SHIFT < PAGE_SHIFT
1095
1096 static struct kmem_cache *thread_info_cache;
1097
1098 struct thread_info *alloc_thread_info(struct task_struct *tsk)
1099 {
1100         struct thread_info *ti;
1101
1102         ti = kmem_cache_alloc(thread_info_cache, GFP_KERNEL);
1103         if (unlikely(ti == NULL))
1104                 return NULL;
1105 #ifdef CONFIG_DEBUG_STACK_USAGE
1106         memset(ti, 0, THREAD_SIZE);
1107 #endif
1108         return ti;
1109 }
1110
1111 void free_thread_info(struct thread_info *ti)
1112 {
1113         kmem_cache_free(thread_info_cache, ti);
1114 }
1115
1116 void thread_info_cache_init(void)
1117 {
1118         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1119                                               THREAD_SIZE, 0, NULL);
1120         BUG_ON(thread_info_cache == NULL);
1121 }
1122
1123 #endif /* THREAD_SHIFT < PAGE_SHIFT */