Merge branch 'stable-3.2' into pandora-3.2
[pandora-kernel.git] / 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/system.h>
45 #include <asm/io.h>
46 #include <asm/processor.h>
47 #include <asm/mmu.h>
48 #include <asm/prom.h>
49 #include <asm/machdep.h>
50 #include <asm/time.h>
51 #include <asm/syscalls.h>
52 #ifdef CONFIG_PPC64
53 #include <asm/firmware.h>
54 #endif
55 #include <linux/kprobes.h>
56 #include <linux/kdebug.h>
57
58 extern unsigned long _get_SP(void);
59
60 #ifndef CONFIG_SMP
61 struct task_struct *last_task_used_math = NULL;
62 struct task_struct *last_task_used_altivec = NULL;
63 struct task_struct *last_task_used_vsx = NULL;
64 struct task_struct *last_task_used_spe = NULL;
65 #endif
66
67 /*
68  * Make sure the floating-point register state in the
69  * the thread_struct is up to date for task tsk.
70  */
71 void flush_fp_to_thread(struct task_struct *tsk)
72 {
73         if (tsk->thread.regs) {
74                 /*
75                  * We need to disable preemption here because if we didn't,
76                  * another process could get scheduled after the regs->msr
77                  * test but before we have finished saving the FP registers
78                  * to the thread_struct.  That process could take over the
79                  * FPU, and then when we get scheduled again we would store
80                  * bogus values for the remaining FP registers.
81                  */
82                 preempt_disable();
83                 if (tsk->thread.regs->msr & MSR_FP) {
84 #ifdef CONFIG_SMP
85                         /*
86                          * This should only ever be called for current or
87                          * for a stopped child process.  Since we save away
88                          * the FP register state on context switch on SMP,
89                          * there is something wrong if a stopped child appears
90                          * to still have its FP state in the CPU registers.
91                          */
92                         BUG_ON(tsk != current);
93 #endif
94                         giveup_fpu(tsk);
95                 }
96                 preempt_enable();
97         }
98 }
99 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
100
101 void enable_kernel_fp(void)
102 {
103         WARN_ON(preemptible());
104
105 #ifdef CONFIG_SMP
106         if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
107                 giveup_fpu(current);
108         else
109                 giveup_fpu(NULL);       /* just enables FP for kernel */
110 #else
111         giveup_fpu(last_task_used_math);
112 #endif /* CONFIG_SMP */
113 }
114 EXPORT_SYMBOL(enable_kernel_fp);
115
116 #ifdef CONFIG_ALTIVEC
117 void enable_kernel_altivec(void)
118 {
119         WARN_ON(preemptible());
120
121 #ifdef CONFIG_SMP
122         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
123                 giveup_altivec(current);
124         else
125                 giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
126 #else
127         giveup_altivec(last_task_used_altivec);
128 #endif /* CONFIG_SMP */
129 }
130 EXPORT_SYMBOL(enable_kernel_altivec);
131
132 /*
133  * Make sure the VMX/Altivec register state in the
134  * the thread_struct is up to date for task tsk.
135  */
136 void flush_altivec_to_thread(struct task_struct *tsk)
137 {
138         if (tsk->thread.regs) {
139                 preempt_disable();
140                 if (tsk->thread.regs->msr & MSR_VEC) {
141 #ifdef CONFIG_SMP
142                         BUG_ON(tsk != current);
143 #endif
144                         giveup_altivec(tsk);
145                 }
146                 preempt_enable();
147         }
148 }
149 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
150 #endif /* CONFIG_ALTIVEC */
151
152 #ifdef CONFIG_VSX
153 #if 0
154 /* not currently used, but some crazy RAID module might want to later */
155 void enable_kernel_vsx(void)
156 {
157         WARN_ON(preemptible());
158
159 #ifdef CONFIG_SMP
160         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
161                 giveup_vsx(current);
162         else
163                 giveup_vsx(NULL);       /* just enable vsx for kernel - force */
164 #else
165         giveup_vsx(last_task_used_vsx);
166 #endif /* CONFIG_SMP */
167 }
168 EXPORT_SYMBOL(enable_kernel_vsx);
169 #endif
170
171 void giveup_vsx(struct task_struct *tsk)
172 {
173         giveup_fpu(tsk);
174         giveup_altivec(tsk);
175         __giveup_vsx(tsk);
176 }
177
178 void flush_vsx_to_thread(struct task_struct *tsk)
179 {
180         if (tsk->thread.regs) {
181                 preempt_disable();
182                 if (tsk->thread.regs->msr & MSR_VSX) {
183 #ifdef CONFIG_SMP
184                         BUG_ON(tsk != current);
185 #endif
186                         giveup_vsx(tsk);
187                 }
188                 preempt_enable();
189         }
190 }
191 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
192 #endif /* CONFIG_VSX */
193
194 #ifdef CONFIG_SPE
195
196 void enable_kernel_spe(void)
197 {
198         WARN_ON(preemptible());
199
200 #ifdef CONFIG_SMP
201         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
202                 giveup_spe(current);
203         else
204                 giveup_spe(NULL);       /* just enable SPE for kernel - force */
205 #else
206         giveup_spe(last_task_used_spe);
207 #endif /* __SMP __ */
208 }
209 EXPORT_SYMBOL(enable_kernel_spe);
210
211 void flush_spe_to_thread(struct task_struct *tsk)
212 {
213         if (tsk->thread.regs) {
214                 preempt_disable();
215                 if (tsk->thread.regs->msr & MSR_SPE) {
216 #ifdef CONFIG_SMP
217                         BUG_ON(tsk != current);
218 #endif
219                         tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
220                         giveup_spe(tsk);
221                 }
222                 preempt_enable();
223         }
224 }
225 #endif /* CONFIG_SPE */
226
227 #ifndef CONFIG_SMP
228 /*
229  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
230  * and the current task has some state, discard it.
231  */
232 void discard_lazy_cpu_state(void)
233 {
234         preempt_disable();
235         if (last_task_used_math == current)
236                 last_task_used_math = NULL;
237 #ifdef CONFIG_ALTIVEC
238         if (last_task_used_altivec == current)
239                 last_task_used_altivec = NULL;
240 #endif /* CONFIG_ALTIVEC */
241 #ifdef CONFIG_VSX
242         if (last_task_used_vsx == current)
243                 last_task_used_vsx = NULL;
244 #endif /* CONFIG_VSX */
245 #ifdef CONFIG_SPE
246         if (last_task_used_spe == current)
247                 last_task_used_spe = NULL;
248 #endif
249         preempt_enable();
250 }
251 #endif /* CONFIG_SMP */
252
253 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
254 void do_send_trap(struct pt_regs *regs, unsigned long address,
255                   unsigned long error_code, int signal_code, int breakpt)
256 {
257         siginfo_t info;
258
259         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
260                         11, SIGSEGV) == NOTIFY_STOP)
261                 return;
262
263         /* Deliver the signal to userspace */
264         info.si_signo = SIGTRAP;
265         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
266         info.si_code = signal_code;
267         info.si_addr = (void __user *)address;
268         force_sig_info(SIGTRAP, &info, current);
269 }
270 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
271 void do_dabr(struct pt_regs *regs, unsigned long address,
272                     unsigned long error_code)
273 {
274         siginfo_t info;
275
276         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
277                         11, SIGSEGV) == NOTIFY_STOP)
278                 return;
279
280         if (debugger_dabr_match(regs))
281                 return;
282
283         /* Clear the DABR */
284         set_dabr(0);
285
286         /* Deliver the signal to userspace */
287         info.si_signo = SIGTRAP;
288         info.si_errno = 0;
289         info.si_code = TRAP_HWBKPT;
290         info.si_addr = (void __user *)address;
291         force_sig_info(SIGTRAP, &info, current);
292 }
293 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
294
295 static DEFINE_PER_CPU(unsigned long, current_dabr);
296
297 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
298 /*
299  * Set the debug registers back to their default "safe" values.
300  */
301 static void set_debug_reg_defaults(struct thread_struct *thread)
302 {
303         thread->iac1 = thread->iac2 = 0;
304 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
305         thread->iac3 = thread->iac4 = 0;
306 #endif
307         thread->dac1 = thread->dac2 = 0;
308 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
309         thread->dvc1 = thread->dvc2 = 0;
310 #endif
311         thread->dbcr0 = 0;
312 #ifdef CONFIG_BOOKE
313         /*
314          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
315          */
316         thread->dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |   \
317                         DBCR1_IAC3US | DBCR1_IAC4US;
318         /*
319          * Force Data Address Compare User/Supervisor bits to be User-only
320          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
321          */
322         thread->dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
323 #else
324         thread->dbcr1 = 0;
325 #endif
326 }
327
328 static void prime_debug_regs(struct thread_struct *thread)
329 {
330         mtspr(SPRN_IAC1, thread->iac1);
331         mtspr(SPRN_IAC2, thread->iac2);
332 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
333         mtspr(SPRN_IAC3, thread->iac3);
334         mtspr(SPRN_IAC4, thread->iac4);
335 #endif
336         mtspr(SPRN_DAC1, thread->dac1);
337         mtspr(SPRN_DAC2, thread->dac2);
338 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
339         mtspr(SPRN_DVC1, thread->dvc1);
340         mtspr(SPRN_DVC2, thread->dvc2);
341 #endif
342         mtspr(SPRN_DBCR0, thread->dbcr0);
343         mtspr(SPRN_DBCR1, thread->dbcr1);
344 #ifdef CONFIG_BOOKE
345         mtspr(SPRN_DBCR2, thread->dbcr2);
346 #endif
347 }
348 /*
349  * Unless neither the old or new thread are making use of the
350  * debug registers, set the debug registers from the values
351  * stored in the new thread.
352  */
353 static void switch_booke_debug_regs(struct thread_struct *new_thread)
354 {
355         if ((current->thread.dbcr0 & DBCR0_IDM)
356                 || (new_thread->dbcr0 & DBCR0_IDM))
357                         prime_debug_regs(new_thread);
358 }
359 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
360 #ifndef CONFIG_HAVE_HW_BREAKPOINT
361 static void set_debug_reg_defaults(struct thread_struct *thread)
362 {
363         if (thread->dabr) {
364                 thread->dabr = 0;
365                 set_dabr(0);
366         }
367 }
368 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
369 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
370
371 int set_dabr(unsigned long dabr)
372 {
373         __get_cpu_var(current_dabr) = dabr;
374
375         if (ppc_md.set_dabr)
376                 return ppc_md.set_dabr(dabr);
377
378         /* XXX should we have a CPU_FTR_HAS_DABR ? */
379 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
380         mtspr(SPRN_DAC1, dabr);
381 #ifdef CONFIG_PPC_47x
382         isync();
383 #endif
384 #elif defined(CONFIG_PPC_BOOK3S)
385         mtspr(SPRN_DABR, dabr);
386 #endif
387
388
389         return 0;
390 }
391
392 #ifdef CONFIG_PPC64
393 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
394 #endif
395
396 struct task_struct *__switch_to(struct task_struct *prev,
397         struct task_struct *new)
398 {
399         struct thread_struct *new_thread, *old_thread;
400         unsigned long flags;
401         struct task_struct *last;
402 #ifdef CONFIG_PPC_BOOK3S_64
403         struct ppc64_tlb_batch *batch;
404 #endif
405
406 #ifdef CONFIG_SMP
407         /* avoid complexity of lazy save/restore of fpu
408          * by just saving it every time we switch out if
409          * this task used the fpu during the last quantum.
410          *
411          * If it tries to use the fpu again, it'll trap and
412          * reload its fp regs.  So we don't have to do a restore
413          * every switch, just a save.
414          *  -- Cort
415          */
416         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
417                 giveup_fpu(prev);
418 #ifdef CONFIG_ALTIVEC
419         /*
420          * If the previous thread used altivec in the last quantum
421          * (thus changing altivec regs) then save them.
422          * We used to check the VRSAVE register but not all apps
423          * set it, so we don't rely on it now (and in fact we need
424          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
425          *
426          * On SMP we always save/restore altivec regs just to avoid the
427          * complexity of changing processors.
428          *  -- Cort
429          */
430         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
431                 giveup_altivec(prev);
432 #endif /* CONFIG_ALTIVEC */
433 #ifdef CONFIG_VSX
434         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
435                 /* VMX and FPU registers are already save here */
436                 __giveup_vsx(prev);
437 #endif /* CONFIG_VSX */
438 #ifdef CONFIG_SPE
439         /*
440          * If the previous thread used spe in the last quantum
441          * (thus changing spe regs) then save them.
442          *
443          * On SMP we always save/restore spe regs just to avoid the
444          * complexity of changing processors.
445          */
446         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
447                 giveup_spe(prev);
448 #endif /* CONFIG_SPE */
449
450 #else  /* CONFIG_SMP */
451 #ifdef CONFIG_ALTIVEC
452         /* Avoid the trap.  On smp this this never happens since
453          * we don't set last_task_used_altivec -- Cort
454          */
455         if (new->thread.regs && last_task_used_altivec == new)
456                 new->thread.regs->msr |= MSR_VEC;
457 #endif /* CONFIG_ALTIVEC */
458 #ifdef CONFIG_VSX
459         if (new->thread.regs && last_task_used_vsx == new)
460                 new->thread.regs->msr |= MSR_VSX;
461 #endif /* CONFIG_VSX */
462 #ifdef CONFIG_SPE
463         /* Avoid the trap.  On smp this this never happens since
464          * we don't set last_task_used_spe
465          */
466         if (new->thread.regs && last_task_used_spe == new)
467                 new->thread.regs->msr |= MSR_SPE;
468 #endif /* CONFIG_SPE */
469
470 #endif /* CONFIG_SMP */
471
472 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
473         switch_booke_debug_regs(&new->thread);
474 #else
475 /*
476  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
477  * schedule DABR
478  */
479 #ifndef CONFIG_HAVE_HW_BREAKPOINT
480         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
481                 set_dabr(new->thread.dabr);
482 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
483 #endif
484
485
486         new_thread = &new->thread;
487         old_thread = &current->thread;
488
489 #ifdef CONFIG_PPC64
490         /*
491          * Collect processor utilization data per process
492          */
493         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
494                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
495                 long unsigned start_tb, current_tb;
496                 start_tb = old_thread->start_tb;
497                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
498                 old_thread->accum_tb += (current_tb - start_tb);
499                 new_thread->start_tb = current_tb;
500         }
501 #endif /* CONFIG_PPC64 */
502
503 #ifdef CONFIG_PPC_BOOK3S_64
504         batch = &__get_cpu_var(ppc64_tlb_batch);
505         if (batch->active) {
506                 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
507                 if (batch->index)
508                         __flush_tlb_pending(batch);
509                 batch->active = 0;
510         }
511 #endif /* CONFIG_PPC_BOOK3S_64 */
512
513         local_irq_save(flags);
514
515         account_system_vtime(current);
516         account_process_vtime(current);
517
518         /*
519          * We can't take a PMU exception inside _switch() since there is a
520          * window where the kernel stack SLB and the kernel stack are out
521          * of sync. Hard disable here.
522          */
523         hard_irq_disable();
524         last = _switch(old_thread, new_thread);
525
526 #ifdef CONFIG_PPC_BOOK3S_64
527         if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
528                 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
529                 batch = &__get_cpu_var(ppc64_tlb_batch);
530                 batch->active = 1;
531         }
532 #endif /* CONFIG_PPC_BOOK3S_64 */
533
534         local_irq_restore(flags);
535
536         return last;
537 }
538
539 static int instructions_to_print = 16;
540
541 static void show_instructions(struct pt_regs *regs)
542 {
543         int i;
544         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
545                         sizeof(int));
546
547         printk("Instruction dump:");
548
549         for (i = 0; i < instructions_to_print; i++) {
550                 int instr;
551
552                 if (!(i % 8))
553                         printk("\n");
554
555 #if !defined(CONFIG_BOOKE)
556                 /* If executing with the IMMU off, adjust pc rather
557                  * than print XXXXXXXX.
558                  */
559                 if (!(regs->msr & MSR_IR))
560                         pc = (unsigned long)phys_to_virt(pc);
561 #endif
562
563                 /* We use __get_user here *only* to avoid an OOPS on a
564                  * bad address because the pc *should* only be a
565                  * kernel address.
566                  */
567                 if (!__kernel_text_address(pc) ||
568                      __get_user(instr, (unsigned int __user *)pc)) {
569                         printk("XXXXXXXX ");
570                 } else {
571                         if (regs->nip == pc)
572                                 printk("<%08x> ", instr);
573                         else
574                                 printk("%08x ", instr);
575                 }
576
577                 pc += sizeof(int);
578         }
579
580         printk("\n");
581 }
582
583 static struct regbit {
584         unsigned long bit;
585         const char *name;
586 } msr_bits[] = {
587         {MSR_EE,        "EE"},
588         {MSR_PR,        "PR"},
589         {MSR_FP,        "FP"},
590         {MSR_VEC,       "VEC"},
591         {MSR_VSX,       "VSX"},
592         {MSR_ME,        "ME"},
593         {MSR_CE,        "CE"},
594         {MSR_DE,        "DE"},
595         {MSR_IR,        "IR"},
596         {MSR_DR,        "DR"},
597         {0,             NULL}
598 };
599
600 static void printbits(unsigned long val, struct regbit *bits)
601 {
602         const char *sep = "";
603
604         printk("<");
605         for (; bits->bit; ++bits)
606                 if (val & bits->bit) {
607                         printk("%s%s", sep, bits->name);
608                         sep = ",";
609                 }
610         printk(">");
611 }
612
613 #ifdef CONFIG_PPC64
614 #define REG             "%016lx"
615 #define REGS_PER_LINE   4
616 #define LAST_VOLATILE   13
617 #else
618 #define REG             "%08lx"
619 #define REGS_PER_LINE   8
620 #define LAST_VOLATILE   12
621 #endif
622
623 void show_regs(struct pt_regs * regs)
624 {
625         int i, trap;
626
627         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
628                regs->nip, regs->link, regs->ctr);
629         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
630                regs, regs->trap, print_tainted(), init_utsname()->release);
631         printk("MSR: "REG" ", regs->msr);
632         printbits(regs->msr, msr_bits);
633         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
634         trap = TRAP(regs);
635         if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
636                 printk("CFAR: "REG"\n", regs->orig_gpr3);
637         if (trap == 0x300 || trap == 0x600)
638 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
639                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
640 #else
641                 printk("DAR: "REG", DSISR: %08lx\n", regs->dar, regs->dsisr);
642 #endif
643         printk("TASK = %p[%d] '%s' THREAD: %p",
644                current, task_pid_nr(current), current->comm, task_thread_info(current));
645
646 #ifdef CONFIG_SMP
647         printk(" CPU: %d", raw_smp_processor_id());
648 #endif /* CONFIG_SMP */
649
650         for (i = 0;  i < 32;  i++) {
651                 if ((i % REGS_PER_LINE) == 0)
652                         printk("\nGPR%02d: ", i);
653                 printk(REG " ", regs->gpr[i]);
654                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
655                         break;
656         }
657         printk("\n");
658 #ifdef CONFIG_KALLSYMS
659         /*
660          * Lookup NIP late so we have the best change of getting the
661          * above info out without failing
662          */
663         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
664         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
665 #endif
666         show_stack(current, (unsigned long *) regs->gpr[1]);
667         if (!user_mode(regs))
668                 show_instructions(regs);
669 }
670
671 void exit_thread(void)
672 {
673         discard_lazy_cpu_state();
674 }
675
676 void flush_thread(void)
677 {
678         discard_lazy_cpu_state();
679
680 #ifdef CONFIG_HAVE_HW_BREAKPOINT
681         flush_ptrace_hw_breakpoint(current);
682 #else /* CONFIG_HAVE_HW_BREAKPOINT */
683         set_debug_reg_defaults(&current->thread);
684 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
685 }
686
687 void
688 release_thread(struct task_struct *t)
689 {
690 }
691
692 /*
693  * This gets called before we allocate a new thread and copy
694  * the current task into it.
695  */
696 void prepare_to_copy(struct task_struct *tsk)
697 {
698         flush_fp_to_thread(current);
699         flush_altivec_to_thread(current);
700         flush_vsx_to_thread(current);
701         flush_spe_to_thread(current);
702 #ifdef CONFIG_HAVE_HW_BREAKPOINT
703         flush_ptrace_hw_breakpoint(tsk);
704 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
705 }
706
707 /*
708  * Copy a thread..
709  */
710 extern unsigned long dscr_default; /* defined in arch/powerpc/kernel/sysfs.c */
711
712 int copy_thread(unsigned long clone_flags, unsigned long usp,
713                 unsigned long unused, struct task_struct *p,
714                 struct pt_regs *regs)
715 {
716         struct pt_regs *childregs, *kregs;
717         extern void ret_from_fork(void);
718         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
719
720         CHECK_FULL_REGS(regs);
721         /* Copy registers */
722         sp -= sizeof(struct pt_regs);
723         childregs = (struct pt_regs *) sp;
724         *childregs = *regs;
725         if ((childregs->msr & MSR_PR) == 0) {
726                 /* for kernel thread, set `current' and stackptr in new task */
727                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
728 #ifdef CONFIG_PPC32
729                 childregs->gpr[2] = (unsigned long) p;
730 #else
731                 clear_tsk_thread_flag(p, TIF_32BIT);
732 #endif
733                 p->thread.regs = NULL;  /* no user register state */
734         } else {
735                 childregs->gpr[1] = usp;
736                 p->thread.regs = childregs;
737                 if (clone_flags & CLONE_SETTLS) {
738 #ifdef CONFIG_PPC64
739                         if (!is_32bit_task())
740                                 childregs->gpr[13] = childregs->gpr[6];
741                         else
742 #endif
743                                 childregs->gpr[2] = childregs->gpr[6];
744                 }
745         }
746         childregs->gpr[3] = 0;  /* Result from fork() */
747         sp -= STACK_FRAME_OVERHEAD;
748
749         /*
750          * The way this works is that at some point in the future
751          * some task will call _switch to switch to the new task.
752          * That will pop off the stack frame created below and start
753          * the new task running at ret_from_fork.  The new task will
754          * do some house keeping and then return from the fork or clone
755          * system call, using the stack frame created above.
756          */
757         sp -= sizeof(struct pt_regs);
758         kregs = (struct pt_regs *) sp;
759         sp -= STACK_FRAME_OVERHEAD;
760         p->thread.ksp = sp;
761         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
762                                 _ALIGN_UP(sizeof(struct thread_info), 16);
763
764 #ifdef CONFIG_PPC_STD_MMU_64
765         if (mmu_has_feature(MMU_FTR_SLB)) {
766                 unsigned long sp_vsid;
767                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
768
769                 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
770                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
771                                 << SLB_VSID_SHIFT_1T;
772                 else
773                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
774                                 << SLB_VSID_SHIFT;
775                 sp_vsid |= SLB_VSID_KERNEL | llp;
776                 p->thread.ksp_vsid = sp_vsid;
777         }
778 #endif /* CONFIG_PPC_STD_MMU_64 */
779 #ifdef CONFIG_PPC64 
780         if (cpu_has_feature(CPU_FTR_DSCR)) {
781                 p->thread.dscr_inherit = current->thread.dscr_inherit;
782                 p->thread.dscr = current->thread.dscr;
783         }
784 #endif
785
786         /*
787          * The PPC64 ABI makes use of a TOC to contain function 
788          * pointers.  The function (ret_from_except) is actually a pointer
789          * to the TOC entry.  The first entry is a pointer to the actual
790          * function.
791          */
792 #ifdef CONFIG_PPC64
793         kregs->nip = *((unsigned long *)ret_from_fork);
794 #else
795         kregs->nip = (unsigned long)ret_from_fork;
796 #endif
797
798         return 0;
799 }
800
801 /*
802  * Set up a thread for executing a new program
803  */
804 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
805 {
806 #ifdef CONFIG_PPC64
807         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
808 #endif
809
810         /*
811          * If we exec out of a kernel thread then thread.regs will not be
812          * set.  Do it now.
813          */
814         if (!current->thread.regs) {
815                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
816                 current->thread.regs = regs - 1;
817         }
818
819         memset(regs->gpr, 0, sizeof(regs->gpr));
820         regs->ctr = 0;
821         regs->link = 0;
822         regs->xer = 0;
823         regs->ccr = 0;
824         regs->gpr[1] = sp;
825
826         /*
827          * We have just cleared all the nonvolatile GPRs, so make
828          * FULL_REGS(regs) return true.  This is necessary to allow
829          * ptrace to examine the thread immediately after exec.
830          */
831         regs->trap &= ~1UL;
832
833 #ifdef CONFIG_PPC32
834         regs->mq = 0;
835         regs->nip = start;
836         regs->msr = MSR_USER;
837 #else
838         if (!is_32bit_task()) {
839                 unsigned long entry, toc;
840
841                 /* start is a relocated pointer to the function descriptor for
842                  * the elf _start routine.  The first entry in the function
843                  * descriptor is the entry address of _start and the second
844                  * entry is the TOC value we need to use.
845                  */
846                 __get_user(entry, (unsigned long __user *)start);
847                 __get_user(toc, (unsigned long __user *)start+1);
848
849                 /* Check whether the e_entry function descriptor entries
850                  * need to be relocated before we can use them.
851                  */
852                 if (load_addr != 0) {
853                         entry += load_addr;
854                         toc   += load_addr;
855                 }
856                 regs->nip = entry;
857                 regs->gpr[2] = toc;
858                 regs->msr = MSR_USER64;
859         } else {
860                 regs->nip = start;
861                 regs->gpr[2] = 0;
862                 regs->msr = MSR_USER32;
863         }
864 #endif
865
866         discard_lazy_cpu_state();
867 #ifdef CONFIG_VSX
868         current->thread.used_vsr = 0;
869 #endif
870         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
871         current->thread.fpscr.val = 0;
872 #ifdef CONFIG_ALTIVEC
873         memset(current->thread.vr, 0, sizeof(current->thread.vr));
874         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
875         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
876         current->thread.vrsave = 0;
877         current->thread.used_vr = 0;
878 #endif /* CONFIG_ALTIVEC */
879 #ifdef CONFIG_SPE
880         memset(current->thread.evr, 0, sizeof(current->thread.evr));
881         current->thread.acc = 0;
882         current->thread.spefscr = 0;
883         current->thread.used_spe = 0;
884 #endif /* CONFIG_SPE */
885 }
886
887 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
888                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
889
890 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
891 {
892         struct pt_regs *regs = tsk->thread.regs;
893
894         /* This is a bit hairy.  If we are an SPE enabled  processor
895          * (have embedded fp) we store the IEEE exception enable flags in
896          * fpexc_mode.  fpexc_mode is also used for setting FP exception
897          * mode (asyn, precise, disabled) for 'Classic' FP. */
898         if (val & PR_FP_EXC_SW_ENABLE) {
899 #ifdef CONFIG_SPE
900                 if (cpu_has_feature(CPU_FTR_SPE)) {
901                         tsk->thread.fpexc_mode = val &
902                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
903                         return 0;
904                 } else {
905                         return -EINVAL;
906                 }
907 #else
908                 return -EINVAL;
909 #endif
910         }
911
912         /* on a CONFIG_SPE this does not hurt us.  The bits that
913          * __pack_fe01 use do not overlap with bits used for
914          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
915          * on CONFIG_SPE implementations are reserved so writing to
916          * them does not change anything */
917         if (val > PR_FP_EXC_PRECISE)
918                 return -EINVAL;
919         tsk->thread.fpexc_mode = __pack_fe01(val);
920         if (regs != NULL && (regs->msr & MSR_FP) != 0)
921                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
922                         | tsk->thread.fpexc_mode;
923         return 0;
924 }
925
926 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
927 {
928         unsigned int val;
929
930         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
931 #ifdef CONFIG_SPE
932                 if (cpu_has_feature(CPU_FTR_SPE))
933                         val = tsk->thread.fpexc_mode;
934                 else
935                         return -EINVAL;
936 #else
937                 return -EINVAL;
938 #endif
939         else
940                 val = __unpack_fe01(tsk->thread.fpexc_mode);
941         return put_user(val, (unsigned int __user *) adr);
942 }
943
944 int set_endian(struct task_struct *tsk, unsigned int val)
945 {
946         struct pt_regs *regs = tsk->thread.regs;
947
948         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
949             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
950                 return -EINVAL;
951
952         if (regs == NULL)
953                 return -EINVAL;
954
955         if (val == PR_ENDIAN_BIG)
956                 regs->msr &= ~MSR_LE;
957         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
958                 regs->msr |= MSR_LE;
959         else
960                 return -EINVAL;
961
962         return 0;
963 }
964
965 int get_endian(struct task_struct *tsk, unsigned long adr)
966 {
967         struct pt_regs *regs = tsk->thread.regs;
968         unsigned int val;
969
970         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
971             !cpu_has_feature(CPU_FTR_REAL_LE))
972                 return -EINVAL;
973
974         if (regs == NULL)
975                 return -EINVAL;
976
977         if (regs->msr & MSR_LE) {
978                 if (cpu_has_feature(CPU_FTR_REAL_LE))
979                         val = PR_ENDIAN_LITTLE;
980                 else
981                         val = PR_ENDIAN_PPC_LITTLE;
982         } else
983                 val = PR_ENDIAN_BIG;
984
985         return put_user(val, (unsigned int __user *)adr);
986 }
987
988 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
989 {
990         tsk->thread.align_ctl = val;
991         return 0;
992 }
993
994 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
995 {
996         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
997 }
998
999 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
1000
1001 int sys_clone(unsigned long clone_flags, unsigned long usp,
1002               int __user *parent_tidp, void __user *child_threadptr,
1003               int __user *child_tidp, int p6,
1004               struct pt_regs *regs)
1005 {
1006         CHECK_FULL_REGS(regs);
1007         if (usp == 0)
1008                 usp = regs->gpr[1];     /* stack pointer for child */
1009 #ifdef CONFIG_PPC64
1010         if (is_32bit_task()) {
1011                 parent_tidp = TRUNC_PTR(parent_tidp);
1012                 child_tidp = TRUNC_PTR(child_tidp);
1013         }
1014 #endif
1015         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
1016 }
1017
1018 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
1019              unsigned long p4, unsigned long p5, unsigned long p6,
1020              struct pt_regs *regs)
1021 {
1022         CHECK_FULL_REGS(regs);
1023         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
1024 }
1025
1026 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
1027               unsigned long p4, unsigned long p5, unsigned long p6,
1028               struct pt_regs *regs)
1029 {
1030         CHECK_FULL_REGS(regs);
1031         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
1032                         regs, 0, NULL, NULL);
1033 }
1034
1035 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
1036                unsigned long a3, unsigned long a4, unsigned long a5,
1037                struct pt_regs *regs)
1038 {
1039         int error;
1040         char *filename;
1041
1042         filename = getname((const char __user *) a0);
1043         error = PTR_ERR(filename);
1044         if (IS_ERR(filename))
1045                 goto out;
1046         flush_fp_to_thread(current);
1047         flush_altivec_to_thread(current);
1048         flush_spe_to_thread(current);
1049         error = do_execve(filename,
1050                           (const char __user *const __user *) a1,
1051                           (const char __user *const __user *) a2, regs);
1052         putname(filename);
1053 out:
1054         return error;
1055 }
1056
1057 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1058                                   unsigned long nbytes)
1059 {
1060         unsigned long stack_page;
1061         unsigned long cpu = task_cpu(p);
1062
1063         /*
1064          * Avoid crashing if the stack has overflowed and corrupted
1065          * task_cpu(p), which is in the thread_info struct.
1066          */
1067         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1068                 stack_page = (unsigned long) hardirq_ctx[cpu];
1069                 if (sp >= stack_page + sizeof(struct thread_struct)
1070                     && sp <= stack_page + THREAD_SIZE - nbytes)
1071                         return 1;
1072
1073                 stack_page = (unsigned long) softirq_ctx[cpu];
1074                 if (sp >= stack_page + sizeof(struct thread_struct)
1075                     && sp <= stack_page + THREAD_SIZE - nbytes)
1076                         return 1;
1077         }
1078         return 0;
1079 }
1080
1081 int validate_sp(unsigned long sp, struct task_struct *p,
1082                        unsigned long nbytes)
1083 {
1084         unsigned long stack_page = (unsigned long)task_stack_page(p);
1085
1086         if (sp >= stack_page + sizeof(struct thread_struct)
1087             && sp <= stack_page + THREAD_SIZE - nbytes)
1088                 return 1;
1089
1090         return valid_irq_stack(sp, p, nbytes);
1091 }
1092
1093 EXPORT_SYMBOL(validate_sp);
1094
1095 unsigned long get_wchan(struct task_struct *p)
1096 {
1097         unsigned long ip, sp;
1098         int count = 0;
1099
1100         if (!p || p == current || p->state == TASK_RUNNING)
1101                 return 0;
1102
1103         sp = p->thread.ksp;
1104         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1105                 return 0;
1106
1107         do {
1108                 sp = *(unsigned long *)sp;
1109                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1110                         return 0;
1111                 if (count > 0) {
1112                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1113                         if (!in_sched_functions(ip))
1114                                 return ip;
1115                 }
1116         } while (count++ < 16);
1117         return 0;
1118 }
1119
1120 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1121
1122 void show_stack(struct task_struct *tsk, unsigned long *stack)
1123 {
1124         unsigned long sp, ip, lr, newsp;
1125         int count = 0;
1126         int firstframe = 1;
1127 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1128         int curr_frame = current->curr_ret_stack;
1129         extern void return_to_handler(void);
1130         unsigned long rth = (unsigned long)return_to_handler;
1131         unsigned long mrth = -1;
1132 #ifdef CONFIG_PPC64
1133         extern void mod_return_to_handler(void);
1134         rth = *(unsigned long *)rth;
1135         mrth = (unsigned long)mod_return_to_handler;
1136         mrth = *(unsigned long *)mrth;
1137 #endif
1138 #endif
1139
1140         sp = (unsigned long) stack;
1141         if (tsk == NULL)
1142                 tsk = current;
1143         if (sp == 0) {
1144                 if (tsk == current)
1145                         asm("mr %0,1" : "=r" (sp));
1146                 else
1147                         sp = tsk->thread.ksp;
1148         }
1149
1150         lr = 0;
1151         printk("Call Trace:\n");
1152         do {
1153                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1154                         return;
1155
1156                 stack = (unsigned long *) sp;
1157                 newsp = stack[0];
1158                 ip = stack[STACK_FRAME_LR_SAVE];
1159                 if (!firstframe || ip != lr) {
1160                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1161 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1162                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1163                                 printk(" (%pS)",
1164                                        (void *)current->ret_stack[curr_frame].ret);
1165                                 curr_frame--;
1166                         }
1167 #endif
1168                         if (firstframe)
1169                                 printk(" (unreliable)");
1170                         printk("\n");
1171                 }
1172                 firstframe = 0;
1173
1174                 /*
1175                  * See if this is an exception frame.
1176                  * We look for the "regshere" marker in the current frame.
1177                  */
1178                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1179                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1180                         struct pt_regs *regs = (struct pt_regs *)
1181                                 (sp + STACK_FRAME_OVERHEAD);
1182                         lr = regs->link;
1183                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1184                                regs->trap, (void *)regs->nip, (void *)lr);
1185                         firstframe = 1;
1186                 }
1187
1188                 sp = newsp;
1189         } while (count++ < kstack_depth_to_print);
1190 }
1191
1192 void dump_stack(void)
1193 {
1194         show_stack(current, NULL);
1195 }
1196 EXPORT_SYMBOL(dump_stack);
1197
1198 #ifdef CONFIG_PPC64
1199 void ppc64_runlatch_on(void)
1200 {
1201         unsigned long ctrl;
1202
1203         if (cpu_has_feature(CPU_FTR_CTRL) && !test_thread_flag(TIF_RUNLATCH)) {
1204                 HMT_medium();
1205
1206                 ctrl = mfspr(SPRN_CTRLF);
1207                 ctrl |= CTRL_RUNLATCH;
1208                 mtspr(SPRN_CTRLT, ctrl);
1209
1210                 set_thread_flag(TIF_RUNLATCH);
1211         }
1212 }
1213
1214 void __ppc64_runlatch_off(void)
1215 {
1216         unsigned long ctrl;
1217
1218         HMT_medium();
1219
1220         clear_thread_flag(TIF_RUNLATCH);
1221
1222         ctrl = mfspr(SPRN_CTRLF);
1223         ctrl &= ~CTRL_RUNLATCH;
1224         mtspr(SPRN_CTRLT, ctrl);
1225 }
1226 #endif
1227
1228 #if THREAD_SHIFT < PAGE_SHIFT
1229
1230 static struct kmem_cache *thread_info_cache;
1231
1232 struct thread_info *alloc_thread_info_node(struct task_struct *tsk, int node)
1233 {
1234         struct thread_info *ti;
1235
1236         ti = kmem_cache_alloc_node(thread_info_cache, GFP_KERNEL, node);
1237         if (unlikely(ti == NULL))
1238                 return NULL;
1239 #ifdef CONFIG_DEBUG_STACK_USAGE
1240         memset(ti, 0, THREAD_SIZE);
1241 #endif
1242         return ti;
1243 }
1244
1245 void free_thread_info(struct thread_info *ti)
1246 {
1247         kmem_cache_free(thread_info_cache, ti);
1248 }
1249
1250 void thread_info_cache_init(void)
1251 {
1252         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
1253                                               THREAD_SIZE, 0, NULL);
1254         BUG_ON(thread_info_cache == NULL);
1255 }
1256
1257 #endif /* THREAD_SHIFT < PAGE_SHIFT */
1258
1259 unsigned long arch_align_stack(unsigned long sp)
1260 {
1261         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1262                 sp -= get_random_int() & ~PAGE_MASK;
1263         return sp & ~0xf;
1264 }
1265
1266 static inline unsigned long brk_rnd(void)
1267 {
1268         unsigned long rnd = 0;
1269
1270         /* 8MB for 32bit, 1GB for 64bit */
1271         if (is_32bit_task())
1272                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1273         else
1274                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1275
1276         return rnd << PAGE_SHIFT;
1277 }
1278
1279 unsigned long arch_randomize_brk(struct mm_struct *mm)
1280 {
1281         unsigned long base = mm->brk;
1282         unsigned long ret;
1283
1284 #ifdef CONFIG_PPC_STD_MMU_64
1285         /*
1286          * If we are using 1TB segments and we are allowed to randomise
1287          * the heap, we can put it above 1TB so it is backed by a 1TB
1288          * segment. Otherwise the heap will be in the bottom 1TB
1289          * which always uses 256MB segments and this may result in a
1290          * performance penalty.
1291          */
1292         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1293                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1294 #endif
1295
1296         ret = PAGE_ALIGN(base + brk_rnd());
1297
1298         if (ret < mm->brk)
1299                 return mm->brk;
1300
1301         return ret;
1302 }
1303
1304 unsigned long randomize_et_dyn(unsigned long base)
1305 {
1306         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1307
1308         if (ret < base)
1309                 return base;
1310
1311         return ret;
1312 }