]> git.karo-electronics.de Git - karo-tx-linux.git/blob - arch/powerpc/kernel/process.c
Merge branch 'next/cleanup2' into HEAD
[karo-tx-linux.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/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         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
262                         11, SIGSEGV) == NOTIFY_STOP)
263                 return;
264
265         /* Deliver the signal to userspace */
266         info.si_signo = SIGTRAP;
267         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
268         info.si_code = signal_code;
269         info.si_addr = (void __user *)address;
270         force_sig_info(SIGTRAP, &info, current);
271 }
272 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
273 void do_dabr(struct pt_regs *regs, unsigned long address,
274                     unsigned long error_code)
275 {
276         siginfo_t info;
277
278         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
279                         11, SIGSEGV) == NOTIFY_STOP)
280                 return;
281
282         if (debugger_dabr_match(regs))
283                 return;
284
285         /* Clear the DABR */
286         set_dabr(0);
287
288         /* Deliver the signal to userspace */
289         info.si_signo = SIGTRAP;
290         info.si_errno = 0;
291         info.si_code = TRAP_HWBKPT;
292         info.si_addr = (void __user *)address;
293         force_sig_info(SIGTRAP, &info, current);
294 }
295 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
296
297 static DEFINE_PER_CPU(unsigned long, current_dabr);
298
299 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
300 /*
301  * Set the debug registers back to their default "safe" values.
302  */
303 static void set_debug_reg_defaults(struct thread_struct *thread)
304 {
305         thread->iac1 = thread->iac2 = 0;
306 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
307         thread->iac3 = thread->iac4 = 0;
308 #endif
309         thread->dac1 = thread->dac2 = 0;
310 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
311         thread->dvc1 = thread->dvc2 = 0;
312 #endif
313         thread->dbcr0 = 0;
314 #ifdef CONFIG_BOOKE
315         /*
316          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
317          */
318         thread->dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |   \
319                         DBCR1_IAC3US | DBCR1_IAC4US;
320         /*
321          * Force Data Address Compare User/Supervisor bits to be User-only
322          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
323          */
324         thread->dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
325 #else
326         thread->dbcr1 = 0;
327 #endif
328 }
329
330 static void prime_debug_regs(struct thread_struct *thread)
331 {
332         mtspr(SPRN_IAC1, thread->iac1);
333         mtspr(SPRN_IAC2, thread->iac2);
334 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
335         mtspr(SPRN_IAC3, thread->iac3);
336         mtspr(SPRN_IAC4, thread->iac4);
337 #endif
338         mtspr(SPRN_DAC1, thread->dac1);
339         mtspr(SPRN_DAC2, thread->dac2);
340 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
341         mtspr(SPRN_DVC1, thread->dvc1);
342         mtspr(SPRN_DVC2, thread->dvc2);
343 #endif
344         mtspr(SPRN_DBCR0, thread->dbcr0);
345         mtspr(SPRN_DBCR1, thread->dbcr1);
346 #ifdef CONFIG_BOOKE
347         mtspr(SPRN_DBCR2, thread->dbcr2);
348 #endif
349 }
350 /*
351  * Unless neither the old or new thread are making use of the
352  * debug registers, set the debug registers from the values
353  * stored in the new thread.
354  */
355 static void switch_booke_debug_regs(struct thread_struct *new_thread)
356 {
357         if ((current->thread.dbcr0 & DBCR0_IDM)
358                 || (new_thread->dbcr0 & DBCR0_IDM))
359                         prime_debug_regs(new_thread);
360 }
361 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
362 #ifndef CONFIG_HAVE_HW_BREAKPOINT
363 static void set_debug_reg_defaults(struct thread_struct *thread)
364 {
365         if (thread->dabr) {
366                 thread->dabr = 0;
367                 set_dabr(0);
368         }
369 }
370 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
371 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
372
373 int set_dabr(unsigned long dabr)
374 {
375         __get_cpu_var(current_dabr) = dabr;
376
377         if (ppc_md.set_dabr)
378                 return ppc_md.set_dabr(dabr);
379
380         /* XXX should we have a CPU_FTR_HAS_DABR ? */
381 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
382         mtspr(SPRN_DAC1, dabr);
383 #ifdef CONFIG_PPC_47x
384         isync();
385 #endif
386 #elif defined(CONFIG_PPC_BOOK3S)
387         mtspr(SPRN_DABR, dabr);
388 #endif
389
390
391         return 0;
392 }
393
394 #ifdef CONFIG_PPC64
395 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
396 #endif
397
398 struct task_struct *__switch_to(struct task_struct *prev,
399         struct task_struct *new)
400 {
401         struct thread_struct *new_thread, *old_thread;
402         unsigned long flags;
403         struct task_struct *last;
404 #ifdef CONFIG_PPC_BOOK3S_64
405         struct ppc64_tlb_batch *batch;
406 #endif
407
408 #ifdef CONFIG_SMP
409         /* avoid complexity of lazy save/restore of fpu
410          * by just saving it every time we switch out if
411          * this task used the fpu during the last quantum.
412          *
413          * If it tries to use the fpu again, it'll trap and
414          * reload its fp regs.  So we don't have to do a restore
415          * every switch, just a save.
416          *  -- Cort
417          */
418         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
419                 giveup_fpu(prev);
420 #ifdef CONFIG_ALTIVEC
421         /*
422          * If the previous thread used altivec in the last quantum
423          * (thus changing altivec regs) then save them.
424          * We used to check the VRSAVE register but not all apps
425          * set it, so we don't rely on it now (and in fact we need
426          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
427          *
428          * On SMP we always save/restore altivec regs just to avoid the
429          * complexity of changing processors.
430          *  -- Cort
431          */
432         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
433                 giveup_altivec(prev);
434 #endif /* CONFIG_ALTIVEC */
435 #ifdef CONFIG_VSX
436         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
437                 /* VMX and FPU registers are already save here */
438                 __giveup_vsx(prev);
439 #endif /* CONFIG_VSX */
440 #ifdef CONFIG_SPE
441         /*
442          * If the previous thread used spe in the last quantum
443          * (thus changing spe regs) then save them.
444          *
445          * On SMP we always save/restore spe regs just to avoid the
446          * complexity of changing processors.
447          */
448         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
449                 giveup_spe(prev);
450 #endif /* CONFIG_SPE */
451
452 #else  /* CONFIG_SMP */
453 #ifdef CONFIG_ALTIVEC
454         /* Avoid the trap.  On smp this this never happens since
455          * we don't set last_task_used_altivec -- Cort
456          */
457         if (new->thread.regs && last_task_used_altivec == new)
458                 new->thread.regs->msr |= MSR_VEC;
459 #endif /* CONFIG_ALTIVEC */
460 #ifdef CONFIG_VSX
461         if (new->thread.regs && last_task_used_vsx == new)
462                 new->thread.regs->msr |= MSR_VSX;
463 #endif /* CONFIG_VSX */
464 #ifdef CONFIG_SPE
465         /* Avoid the trap.  On smp this this never happens since
466          * we don't set last_task_used_spe
467          */
468         if (new->thread.regs && last_task_used_spe == new)
469                 new->thread.regs->msr |= MSR_SPE;
470 #endif /* CONFIG_SPE */
471
472 #endif /* CONFIG_SMP */
473
474 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
475         switch_booke_debug_regs(&new->thread);
476 #else
477 /*
478  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
479  * schedule DABR
480  */
481 #ifndef CONFIG_HAVE_HW_BREAKPOINT
482         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
483                 set_dabr(new->thread.dabr);
484 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
485 #endif
486
487
488         new_thread = &new->thread;
489         old_thread = &current->thread;
490
491 #ifdef CONFIG_PPC64
492         /*
493          * Collect processor utilization data per process
494          */
495         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
496                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
497                 long unsigned start_tb, current_tb;
498                 start_tb = old_thread->start_tb;
499                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
500                 old_thread->accum_tb += (current_tb - start_tb);
501                 new_thread->start_tb = current_tb;
502         }
503 #endif /* CONFIG_PPC64 */
504
505 #ifdef CONFIG_PPC_BOOK3S_64
506         batch = &__get_cpu_var(ppc64_tlb_batch);
507         if (batch->active) {
508                 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
509                 if (batch->index)
510                         __flush_tlb_pending(batch);
511                 batch->active = 0;
512         }
513 #endif /* CONFIG_PPC_BOOK3S_64 */
514
515         local_irq_save(flags);
516
517         /*
518          * We can't take a PMU exception inside _switch() since there is a
519          * window where the kernel stack SLB and the kernel stack are out
520          * of sync. Hard disable here.
521          */
522         hard_irq_disable();
523         last = _switch(old_thread, new_thread);
524
525 #ifdef CONFIG_PPC_BOOK3S_64
526         if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
527                 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
528                 batch = &__get_cpu_var(ppc64_tlb_batch);
529                 batch->active = 1;
530         }
531 #endif /* CONFIG_PPC_BOOK3S_64 */
532
533         local_irq_restore(flags);
534
535         return last;
536 }
537
538 static int instructions_to_print = 16;
539
540 static void show_instructions(struct pt_regs *regs)
541 {
542         int i;
543         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
544                         sizeof(int));
545
546         printk("Instruction dump:");
547
548         for (i = 0; i < instructions_to_print; i++) {
549                 int instr;
550
551                 if (!(i % 8))
552                         printk("\n");
553
554 #if !defined(CONFIG_BOOKE)
555                 /* If executing with the IMMU off, adjust pc rather
556                  * than print XXXXXXXX.
557                  */
558                 if (!(regs->msr & MSR_IR))
559                         pc = (unsigned long)phys_to_virt(pc);
560 #endif
561
562                 /* We use __get_user here *only* to avoid an OOPS on a
563                  * bad address because the pc *should* only be a
564                  * kernel address.
565                  */
566                 if (!__kernel_text_address(pc) ||
567                      __get_user(instr, (unsigned int __user *)pc)) {
568                         printk(KERN_CONT "XXXXXXXX ");
569                 } else {
570                         if (regs->nip == pc)
571                                 printk(KERN_CONT "<%08x> ", instr);
572                         else
573                                 printk(KERN_CONT "%08x ", instr);
574                 }
575
576                 pc += sizeof(int);
577         }
578
579         printk("\n");
580 }
581
582 static struct regbit {
583         unsigned long bit;
584         const char *name;
585 } msr_bits[] = {
586 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
587         {MSR_SF,        "SF"},
588         {MSR_HV,        "HV"},
589 #endif
590         {MSR_VEC,       "VEC"},
591         {MSR_VSX,       "VSX"},
592 #ifdef CONFIG_BOOKE
593         {MSR_CE,        "CE"},
594 #endif
595         {MSR_EE,        "EE"},
596         {MSR_PR,        "PR"},
597         {MSR_FP,        "FP"},
598         {MSR_ME,        "ME"},
599 #ifdef CONFIG_BOOKE
600         {MSR_DE,        "DE"},
601 #else
602         {MSR_SE,        "SE"},
603         {MSR_BE,        "BE"},
604 #endif
605         {MSR_IR,        "IR"},
606         {MSR_DR,        "DR"},
607         {MSR_PMM,       "PMM"},
608 #ifndef CONFIG_BOOKE
609         {MSR_RI,        "RI"},
610         {MSR_LE,        "LE"},
611 #endif
612         {0,             NULL}
613 };
614
615 static void printbits(unsigned long val, struct regbit *bits)
616 {
617         const char *sep = "";
618
619         printk("<");
620         for (; bits->bit; ++bits)
621                 if (val & bits->bit) {
622                         printk("%s%s", sep, bits->name);
623                         sep = ",";
624                 }
625         printk(">");
626 }
627
628 #ifdef CONFIG_PPC64
629 #define REG             "%016lx"
630 #define REGS_PER_LINE   4
631 #define LAST_VOLATILE   13
632 #else
633 #define REG             "%08lx"
634 #define REGS_PER_LINE   8
635 #define LAST_VOLATILE   12
636 #endif
637
638 void show_regs(struct pt_regs * regs)
639 {
640         int i, trap;
641
642         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
643                regs->nip, regs->link, regs->ctr);
644         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
645                regs, regs->trap, print_tainted(), init_utsname()->release);
646         printk("MSR: "REG" ", regs->msr);
647         printbits(regs->msr, msr_bits);
648         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
649 #ifdef CONFIG_PPC64
650         printk("SOFTE: %ld\n", regs->softe);
651 #endif
652         trap = TRAP(regs);
653         if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
654                 printk("CFAR: "REG"\n", regs->orig_gpr3);
655         if (trap == 0x300 || trap == 0x600)
656 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
657                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
658 #else
659                 printk("DAR: "REG", DSISR: %08lx\n", regs->dar, regs->dsisr);
660 #endif
661         printk("TASK = %p[%d] '%s' THREAD: %p",
662                current, task_pid_nr(current), current->comm, task_thread_info(current));
663
664 #ifdef CONFIG_SMP
665         printk(" CPU: %d", raw_smp_processor_id());
666 #endif /* CONFIG_SMP */
667
668         for (i = 0;  i < 32;  i++) {
669                 if ((i % REGS_PER_LINE) == 0)
670                         printk("\nGPR%02d: ", i);
671                 printk(REG " ", regs->gpr[i]);
672                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
673                         break;
674         }
675         printk("\n");
676 #ifdef CONFIG_KALLSYMS
677         /*
678          * Lookup NIP late so we have the best change of getting the
679          * above info out without failing
680          */
681         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
682         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
683 #endif
684         show_stack(current, (unsigned long *) regs->gpr[1]);
685         if (!user_mode(regs))
686                 show_instructions(regs);
687 }
688
689 void exit_thread(void)
690 {
691         discard_lazy_cpu_state();
692 }
693
694 void flush_thread(void)
695 {
696         discard_lazy_cpu_state();
697
698 #ifdef CONFIG_HAVE_HW_BREAKPOINT
699         flush_ptrace_hw_breakpoint(current);
700 #else /* CONFIG_HAVE_HW_BREAKPOINT */
701         set_debug_reg_defaults(&current->thread);
702 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
703 }
704
705 void
706 release_thread(struct task_struct *t)
707 {
708 }
709
710 /*
711  * this gets called so that we can store coprocessor state into memory and
712  * copy the current task into the new thread.
713  */
714 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
715 {
716         flush_fp_to_thread(src);
717         flush_altivec_to_thread(src);
718         flush_vsx_to_thread(src);
719         flush_spe_to_thread(src);
720 #ifdef CONFIG_HAVE_HW_BREAKPOINT
721         flush_ptrace_hw_breakpoint(src);
722 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
723
724         *dst = *src;
725         return 0;
726 }
727
728 /*
729  * Copy a thread..
730  */
731 extern unsigned long dscr_default; /* defined in arch/powerpc/kernel/sysfs.c */
732
733 int copy_thread(unsigned long clone_flags, unsigned long usp,
734                 unsigned long unused, struct task_struct *p,
735                 struct pt_regs *regs)
736 {
737         struct pt_regs *childregs, *kregs;
738         extern void ret_from_fork(void);
739         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
740
741         CHECK_FULL_REGS(regs);
742         /* Copy registers */
743         sp -= sizeof(struct pt_regs);
744         childregs = (struct pt_regs *) sp;
745         *childregs = *regs;
746         if ((childregs->msr & MSR_PR) == 0) {
747                 /* for kernel thread, set `current' and stackptr in new task */
748                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
749 #ifdef CONFIG_PPC32
750                 childregs->gpr[2] = (unsigned long) p;
751 #else
752                 clear_tsk_thread_flag(p, TIF_32BIT);
753 #endif
754                 p->thread.regs = NULL;  /* no user register state */
755         } else {
756                 childregs->gpr[1] = usp;
757                 p->thread.regs = childregs;
758                 if (clone_flags & CLONE_SETTLS) {
759 #ifdef CONFIG_PPC64
760                         if (!is_32bit_task())
761                                 childregs->gpr[13] = childregs->gpr[6];
762                         else
763 #endif
764                                 childregs->gpr[2] = childregs->gpr[6];
765                 }
766         }
767         childregs->gpr[3] = 0;  /* Result from fork() */
768         sp -= STACK_FRAME_OVERHEAD;
769
770         /*
771          * The way this works is that at some point in the future
772          * some task will call _switch to switch to the new task.
773          * That will pop off the stack frame created below and start
774          * the new task running at ret_from_fork.  The new task will
775          * do some house keeping and then return from the fork or clone
776          * system call, using the stack frame created above.
777          */
778         sp -= sizeof(struct pt_regs);
779         kregs = (struct pt_regs *) sp;
780         sp -= STACK_FRAME_OVERHEAD;
781         p->thread.ksp = sp;
782         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
783                                 _ALIGN_UP(sizeof(struct thread_info), 16);
784
785 #ifdef CONFIG_PPC_STD_MMU_64
786         if (mmu_has_feature(MMU_FTR_SLB)) {
787                 unsigned long sp_vsid;
788                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
789
790                 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
791                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
792                                 << SLB_VSID_SHIFT_1T;
793                 else
794                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
795                                 << SLB_VSID_SHIFT;
796                 sp_vsid |= SLB_VSID_KERNEL | llp;
797                 p->thread.ksp_vsid = sp_vsid;
798         }
799 #endif /* CONFIG_PPC_STD_MMU_64 */
800 #ifdef CONFIG_PPC64 
801         if (cpu_has_feature(CPU_FTR_DSCR)) {
802                 p->thread.dscr_inherit = current->thread.dscr_inherit;
803                 p->thread.dscr = current->thread.dscr;
804         }
805 #endif
806
807         /*
808          * The PPC64 ABI makes use of a TOC to contain function 
809          * pointers.  The function (ret_from_except) is actually a pointer
810          * to the TOC entry.  The first entry is a pointer to the actual
811          * function.
812          */
813 #ifdef CONFIG_PPC64
814         kregs->nip = *((unsigned long *)ret_from_fork);
815 #else
816         kregs->nip = (unsigned long)ret_from_fork;
817 #endif
818
819         return 0;
820 }
821
822 /*
823  * Set up a thread for executing a new program
824  */
825 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
826 {
827 #ifdef CONFIG_PPC64
828         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
829 #endif
830
831         /*
832          * If we exec out of a kernel thread then thread.regs will not be
833          * set.  Do it now.
834          */
835         if (!current->thread.regs) {
836                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
837                 current->thread.regs = regs - 1;
838         }
839
840         memset(regs->gpr, 0, sizeof(regs->gpr));
841         regs->ctr = 0;
842         regs->link = 0;
843         regs->xer = 0;
844         regs->ccr = 0;
845         regs->gpr[1] = sp;
846
847         /*
848          * We have just cleared all the nonvolatile GPRs, so make
849          * FULL_REGS(regs) return true.  This is necessary to allow
850          * ptrace to examine the thread immediately after exec.
851          */
852         regs->trap &= ~1UL;
853
854 #ifdef CONFIG_PPC32
855         regs->mq = 0;
856         regs->nip = start;
857         regs->msr = MSR_USER;
858 #else
859         if (!is_32bit_task()) {
860                 unsigned long entry, toc;
861
862                 /* start is a relocated pointer to the function descriptor for
863                  * the elf _start routine.  The first entry in the function
864                  * descriptor is the entry address of _start and the second
865                  * entry is the TOC value we need to use.
866                  */
867                 __get_user(entry, (unsigned long __user *)start);
868                 __get_user(toc, (unsigned long __user *)start+1);
869
870                 /* Check whether the e_entry function descriptor entries
871                  * need to be relocated before we can use them.
872                  */
873                 if (load_addr != 0) {
874                         entry += load_addr;
875                         toc   += load_addr;
876                 }
877                 regs->nip = entry;
878                 regs->gpr[2] = toc;
879                 regs->msr = MSR_USER64;
880         } else {
881                 regs->nip = start;
882                 regs->gpr[2] = 0;
883                 regs->msr = MSR_USER32;
884         }
885 #endif
886
887         discard_lazy_cpu_state();
888 #ifdef CONFIG_VSX
889         current->thread.used_vsr = 0;
890 #endif
891         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
892         current->thread.fpscr.val = 0;
893 #ifdef CONFIG_ALTIVEC
894         memset(current->thread.vr, 0, sizeof(current->thread.vr));
895         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
896         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
897         current->thread.vrsave = 0;
898         current->thread.used_vr = 0;
899 #endif /* CONFIG_ALTIVEC */
900 #ifdef CONFIG_SPE
901         memset(current->thread.evr, 0, sizeof(current->thread.evr));
902         current->thread.acc = 0;
903         current->thread.spefscr = 0;
904         current->thread.used_spe = 0;
905 #endif /* CONFIG_SPE */
906 }
907
908 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
909                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
910
911 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
912 {
913         struct pt_regs *regs = tsk->thread.regs;
914
915         /* This is a bit hairy.  If we are an SPE enabled  processor
916          * (have embedded fp) we store the IEEE exception enable flags in
917          * fpexc_mode.  fpexc_mode is also used for setting FP exception
918          * mode (asyn, precise, disabled) for 'Classic' FP. */
919         if (val & PR_FP_EXC_SW_ENABLE) {
920 #ifdef CONFIG_SPE
921                 if (cpu_has_feature(CPU_FTR_SPE)) {
922                         tsk->thread.fpexc_mode = val &
923                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
924                         return 0;
925                 } else {
926                         return -EINVAL;
927                 }
928 #else
929                 return -EINVAL;
930 #endif
931         }
932
933         /* on a CONFIG_SPE this does not hurt us.  The bits that
934          * __pack_fe01 use do not overlap with bits used for
935          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
936          * on CONFIG_SPE implementations are reserved so writing to
937          * them does not change anything */
938         if (val > PR_FP_EXC_PRECISE)
939                 return -EINVAL;
940         tsk->thread.fpexc_mode = __pack_fe01(val);
941         if (regs != NULL && (regs->msr & MSR_FP) != 0)
942                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
943                         | tsk->thread.fpexc_mode;
944         return 0;
945 }
946
947 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
948 {
949         unsigned int val;
950
951         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
952 #ifdef CONFIG_SPE
953                 if (cpu_has_feature(CPU_FTR_SPE))
954                         val = tsk->thread.fpexc_mode;
955                 else
956                         return -EINVAL;
957 #else
958                 return -EINVAL;
959 #endif
960         else
961                 val = __unpack_fe01(tsk->thread.fpexc_mode);
962         return put_user(val, (unsigned int __user *) adr);
963 }
964
965 int set_endian(struct task_struct *tsk, unsigned int val)
966 {
967         struct pt_regs *regs = tsk->thread.regs;
968
969         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
970             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
971                 return -EINVAL;
972
973         if (regs == NULL)
974                 return -EINVAL;
975
976         if (val == PR_ENDIAN_BIG)
977                 regs->msr &= ~MSR_LE;
978         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
979                 regs->msr |= MSR_LE;
980         else
981                 return -EINVAL;
982
983         return 0;
984 }
985
986 int get_endian(struct task_struct *tsk, unsigned long adr)
987 {
988         struct pt_regs *regs = tsk->thread.regs;
989         unsigned int val;
990
991         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
992             !cpu_has_feature(CPU_FTR_REAL_LE))
993                 return -EINVAL;
994
995         if (regs == NULL)
996                 return -EINVAL;
997
998         if (regs->msr & MSR_LE) {
999                 if (cpu_has_feature(CPU_FTR_REAL_LE))
1000                         val = PR_ENDIAN_LITTLE;
1001                 else
1002                         val = PR_ENDIAN_PPC_LITTLE;
1003         } else
1004                 val = PR_ENDIAN_BIG;
1005
1006         return put_user(val, (unsigned int __user *)adr);
1007 }
1008
1009 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1010 {
1011         tsk->thread.align_ctl = val;
1012         return 0;
1013 }
1014
1015 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1016 {
1017         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1018 }
1019
1020 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
1021
1022 int sys_clone(unsigned long clone_flags, unsigned long usp,
1023               int __user *parent_tidp, void __user *child_threadptr,
1024               int __user *child_tidp, int p6,
1025               struct pt_regs *regs)
1026 {
1027         CHECK_FULL_REGS(regs);
1028         if (usp == 0)
1029                 usp = regs->gpr[1];     /* stack pointer for child */
1030 #ifdef CONFIG_PPC64
1031         if (is_32bit_task()) {
1032                 parent_tidp = TRUNC_PTR(parent_tidp);
1033                 child_tidp = TRUNC_PTR(child_tidp);
1034         }
1035 #endif
1036         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
1037 }
1038
1039 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
1040              unsigned long p4, unsigned long p5, unsigned long p6,
1041              struct pt_regs *regs)
1042 {
1043         CHECK_FULL_REGS(regs);
1044         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
1045 }
1046
1047 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
1048               unsigned long p4, unsigned long p5, unsigned long p6,
1049               struct pt_regs *regs)
1050 {
1051         CHECK_FULL_REGS(regs);
1052         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
1053                         regs, 0, NULL, NULL);
1054 }
1055
1056 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
1057                unsigned long a3, unsigned long a4, unsigned long a5,
1058                struct pt_regs *regs)
1059 {
1060         int error;
1061         char *filename;
1062
1063         filename = getname((const char __user *) a0);
1064         error = PTR_ERR(filename);
1065         if (IS_ERR(filename))
1066                 goto out;
1067         flush_fp_to_thread(current);
1068         flush_altivec_to_thread(current);
1069         flush_spe_to_thread(current);
1070         error = do_execve(filename,
1071                           (const char __user *const __user *) a1,
1072                           (const char __user *const __user *) a2, regs);
1073         putname(filename);
1074 out:
1075         return error;
1076 }
1077
1078 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1079                                   unsigned long nbytes)
1080 {
1081         unsigned long stack_page;
1082         unsigned long cpu = task_cpu(p);
1083
1084         /*
1085          * Avoid crashing if the stack has overflowed and corrupted
1086          * task_cpu(p), which is in the thread_info struct.
1087          */
1088         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1089                 stack_page = (unsigned long) hardirq_ctx[cpu];
1090                 if (sp >= stack_page + sizeof(struct thread_struct)
1091                     && sp <= stack_page + THREAD_SIZE - nbytes)
1092                         return 1;
1093
1094                 stack_page = (unsigned long) softirq_ctx[cpu];
1095                 if (sp >= stack_page + sizeof(struct thread_struct)
1096                     && sp <= stack_page + THREAD_SIZE - nbytes)
1097                         return 1;
1098         }
1099         return 0;
1100 }
1101
1102 int validate_sp(unsigned long sp, struct task_struct *p,
1103                        unsigned long nbytes)
1104 {
1105         unsigned long stack_page = (unsigned long)task_stack_page(p);
1106
1107         if (sp >= stack_page + sizeof(struct thread_struct)
1108             && sp <= stack_page + THREAD_SIZE - nbytes)
1109                 return 1;
1110
1111         return valid_irq_stack(sp, p, nbytes);
1112 }
1113
1114 EXPORT_SYMBOL(validate_sp);
1115
1116 unsigned long get_wchan(struct task_struct *p)
1117 {
1118         unsigned long ip, sp;
1119         int count = 0;
1120
1121         if (!p || p == current || p->state == TASK_RUNNING)
1122                 return 0;
1123
1124         sp = p->thread.ksp;
1125         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1126                 return 0;
1127
1128         do {
1129                 sp = *(unsigned long *)sp;
1130                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1131                         return 0;
1132                 if (count > 0) {
1133                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1134                         if (!in_sched_functions(ip))
1135                                 return ip;
1136                 }
1137         } while (count++ < 16);
1138         return 0;
1139 }
1140
1141 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1142
1143 void show_stack(struct task_struct *tsk, unsigned long *stack)
1144 {
1145         unsigned long sp, ip, lr, newsp;
1146         int count = 0;
1147         int firstframe = 1;
1148 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1149         int curr_frame = current->curr_ret_stack;
1150         extern void return_to_handler(void);
1151         unsigned long rth = (unsigned long)return_to_handler;
1152         unsigned long mrth = -1;
1153 #ifdef CONFIG_PPC64
1154         extern void mod_return_to_handler(void);
1155         rth = *(unsigned long *)rth;
1156         mrth = (unsigned long)mod_return_to_handler;
1157         mrth = *(unsigned long *)mrth;
1158 #endif
1159 #endif
1160
1161         sp = (unsigned long) stack;
1162         if (tsk == NULL)
1163                 tsk = current;
1164         if (sp == 0) {
1165                 if (tsk == current)
1166                         asm("mr %0,1" : "=r" (sp));
1167                 else
1168                         sp = tsk->thread.ksp;
1169         }
1170
1171         lr = 0;
1172         printk("Call Trace:\n");
1173         do {
1174                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1175                         return;
1176
1177                 stack = (unsigned long *) sp;
1178                 newsp = stack[0];
1179                 ip = stack[STACK_FRAME_LR_SAVE];
1180                 if (!firstframe || ip != lr) {
1181                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1182 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1183                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1184                                 printk(" (%pS)",
1185                                        (void *)current->ret_stack[curr_frame].ret);
1186                                 curr_frame--;
1187                         }
1188 #endif
1189                         if (firstframe)
1190                                 printk(" (unreliable)");
1191                         printk("\n");
1192                 }
1193                 firstframe = 0;
1194
1195                 /*
1196                  * See if this is an exception frame.
1197                  * We look for the "regshere" marker in the current frame.
1198                  */
1199                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1200                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1201                         struct pt_regs *regs = (struct pt_regs *)
1202                                 (sp + STACK_FRAME_OVERHEAD);
1203                         lr = regs->link;
1204                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1205                                regs->trap, (void *)regs->nip, (void *)lr);
1206                         firstframe = 1;
1207                 }
1208
1209                 sp = newsp;
1210         } while (count++ < kstack_depth_to_print);
1211 }
1212
1213 void dump_stack(void)
1214 {
1215         show_stack(current, NULL);
1216 }
1217 EXPORT_SYMBOL(dump_stack);
1218
1219 #ifdef CONFIG_PPC64
1220 /* Called with hard IRQs off */
1221 void __ppc64_runlatch_on(void)
1222 {
1223         struct thread_info *ti = current_thread_info();
1224         unsigned long ctrl;
1225
1226         ctrl = mfspr(SPRN_CTRLF);
1227         ctrl |= CTRL_RUNLATCH;
1228         mtspr(SPRN_CTRLT, ctrl);
1229
1230         ti->local_flags |= _TLF_RUNLATCH;
1231 }
1232
1233 /* Called with hard IRQs off */
1234 void __ppc64_runlatch_off(void)
1235 {
1236         struct thread_info *ti = current_thread_info();
1237         unsigned long ctrl;
1238
1239         ti->local_flags &= ~_TLF_RUNLATCH;
1240
1241         ctrl = mfspr(SPRN_CTRLF);
1242         ctrl &= ~CTRL_RUNLATCH;
1243         mtspr(SPRN_CTRLT, ctrl);
1244 }
1245 #endif /* CONFIG_PPC64 */
1246
1247 unsigned long arch_align_stack(unsigned long sp)
1248 {
1249         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1250                 sp -= get_random_int() & ~PAGE_MASK;
1251         return sp & ~0xf;
1252 }
1253
1254 static inline unsigned long brk_rnd(void)
1255 {
1256         unsigned long rnd = 0;
1257
1258         /* 8MB for 32bit, 1GB for 64bit */
1259         if (is_32bit_task())
1260                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1261         else
1262                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1263
1264         return rnd << PAGE_SHIFT;
1265 }
1266
1267 unsigned long arch_randomize_brk(struct mm_struct *mm)
1268 {
1269         unsigned long base = mm->brk;
1270         unsigned long ret;
1271
1272 #ifdef CONFIG_PPC_STD_MMU_64
1273         /*
1274          * If we are using 1TB segments and we are allowed to randomise
1275          * the heap, we can put it above 1TB so it is backed by a 1TB
1276          * segment. Otherwise the heap will be in the bottom 1TB
1277          * which always uses 256MB segments and this may result in a
1278          * performance penalty.
1279          */
1280         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1281                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1282 #endif
1283
1284         ret = PAGE_ALIGN(base + brk_rnd());
1285
1286         if (ret < mm->brk)
1287                 return mm->brk;
1288
1289         return ret;
1290 }
1291
1292 unsigned long randomize_et_dyn(unsigned long base)
1293 {
1294         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1295
1296         if (ret < base)
1297                 return base;
1298
1299         return ret;
1300 }