Merge remote-tracking branch 'pci/pci/gavin-window-alignment' into next
[firefly-linux-kernel-4.4.55.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         current->thread.trap_nr = signal_code;
262         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
263                         11, SIGSEGV) == NOTIFY_STOP)
264                 return;
265
266         /* Deliver the signal to userspace */
267         info.si_signo = SIGTRAP;
268         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
269         info.si_code = signal_code;
270         info.si_addr = (void __user *)address;
271         force_sig_info(SIGTRAP, &info, current);
272 }
273 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
274 void do_dabr(struct pt_regs *regs, unsigned long address,
275                     unsigned long error_code)
276 {
277         siginfo_t info;
278
279         current->thread.trap_nr = TRAP_HWBKPT;
280         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
281                         11, SIGSEGV) == NOTIFY_STOP)
282                 return;
283
284         if (debugger_dabr_match(regs))
285                 return;
286
287         /* Clear the DABR */
288         set_dabr(0, 0);
289
290         /* Deliver the signal to userspace */
291         info.si_signo = SIGTRAP;
292         info.si_errno = 0;
293         info.si_code = TRAP_HWBKPT;
294         info.si_addr = (void __user *)address;
295         force_sig_info(SIGTRAP, &info, current);
296 }
297 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
298
299 static DEFINE_PER_CPU(unsigned long, current_dabr);
300
301 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
302 /*
303  * Set the debug registers back to their default "safe" values.
304  */
305 static void set_debug_reg_defaults(struct thread_struct *thread)
306 {
307         thread->iac1 = thread->iac2 = 0;
308 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
309         thread->iac3 = thread->iac4 = 0;
310 #endif
311         thread->dac1 = thread->dac2 = 0;
312 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
313         thread->dvc1 = thread->dvc2 = 0;
314 #endif
315         thread->dbcr0 = 0;
316 #ifdef CONFIG_BOOKE
317         /*
318          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
319          */
320         thread->dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |   \
321                         DBCR1_IAC3US | DBCR1_IAC4US;
322         /*
323          * Force Data Address Compare User/Supervisor bits to be User-only
324          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
325          */
326         thread->dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
327 #else
328         thread->dbcr1 = 0;
329 #endif
330 }
331
332 static void prime_debug_regs(struct thread_struct *thread)
333 {
334         mtspr(SPRN_IAC1, thread->iac1);
335         mtspr(SPRN_IAC2, thread->iac2);
336 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
337         mtspr(SPRN_IAC3, thread->iac3);
338         mtspr(SPRN_IAC4, thread->iac4);
339 #endif
340         mtspr(SPRN_DAC1, thread->dac1);
341         mtspr(SPRN_DAC2, thread->dac2);
342 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
343         mtspr(SPRN_DVC1, thread->dvc1);
344         mtspr(SPRN_DVC2, thread->dvc2);
345 #endif
346         mtspr(SPRN_DBCR0, thread->dbcr0);
347         mtspr(SPRN_DBCR1, thread->dbcr1);
348 #ifdef CONFIG_BOOKE
349         mtspr(SPRN_DBCR2, thread->dbcr2);
350 #endif
351 }
352 /*
353  * Unless neither the old or new thread are making use of the
354  * debug registers, set the debug registers from the values
355  * stored in the new thread.
356  */
357 static void switch_booke_debug_regs(struct thread_struct *new_thread)
358 {
359         if ((current->thread.dbcr0 & DBCR0_IDM)
360                 || (new_thread->dbcr0 & DBCR0_IDM))
361                         prime_debug_regs(new_thread);
362 }
363 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
364 #ifndef CONFIG_HAVE_HW_BREAKPOINT
365 static void set_debug_reg_defaults(struct thread_struct *thread)
366 {
367         if (thread->dabr) {
368                 thread->dabr = 0;
369                 thread->dabrx = 0;
370                 set_dabr(0, 0);
371         }
372 }
373 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
374 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
375
376 int set_dabr(unsigned long dabr, unsigned long dabrx)
377 {
378         __get_cpu_var(current_dabr) = dabr;
379
380         if (ppc_md.set_dabr)
381                 return ppc_md.set_dabr(dabr, dabrx);
382
383         /* XXX should we have a CPU_FTR_HAS_DABR ? */
384 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
385         mtspr(SPRN_DAC1, dabr);
386 #ifdef CONFIG_PPC_47x
387         isync();
388 #endif
389 #elif defined(CONFIG_PPC_BOOK3S)
390         mtspr(SPRN_DABR, dabr);
391         mtspr(SPRN_DABRX, dabrx);
392 #endif
393         return 0;
394 }
395
396 #ifdef CONFIG_PPC64
397 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
398 #endif
399
400 struct task_struct *__switch_to(struct task_struct *prev,
401         struct task_struct *new)
402 {
403         struct thread_struct *new_thread, *old_thread;
404         unsigned long flags;
405         struct task_struct *last;
406 #ifdef CONFIG_PPC_BOOK3S_64
407         struct ppc64_tlb_batch *batch;
408 #endif
409
410 #ifdef CONFIG_SMP
411         /* avoid complexity of lazy save/restore of fpu
412          * by just saving it every time we switch out if
413          * this task used the fpu during the last quantum.
414          *
415          * If it tries to use the fpu again, it'll trap and
416          * reload its fp regs.  So we don't have to do a restore
417          * every switch, just a save.
418          *  -- Cort
419          */
420         if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
421                 giveup_fpu(prev);
422 #ifdef CONFIG_ALTIVEC
423         /*
424          * If the previous thread used altivec in the last quantum
425          * (thus changing altivec regs) then save them.
426          * We used to check the VRSAVE register but not all apps
427          * set it, so we don't rely on it now (and in fact we need
428          * to save & restore VSCR even if VRSAVE == 0).  -- paulus
429          *
430          * On SMP we always save/restore altivec regs just to avoid the
431          * complexity of changing processors.
432          *  -- Cort
433          */
434         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
435                 giveup_altivec(prev);
436 #endif /* CONFIG_ALTIVEC */
437 #ifdef CONFIG_VSX
438         if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
439                 /* VMX and FPU registers are already save here */
440                 __giveup_vsx(prev);
441 #endif /* CONFIG_VSX */
442 #ifdef CONFIG_SPE
443         /*
444          * If the previous thread used spe in the last quantum
445          * (thus changing spe regs) then save them.
446          *
447          * On SMP we always save/restore spe regs just to avoid the
448          * complexity of changing processors.
449          */
450         if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
451                 giveup_spe(prev);
452 #endif /* CONFIG_SPE */
453
454 #else  /* CONFIG_SMP */
455 #ifdef CONFIG_ALTIVEC
456         /* Avoid the trap.  On smp this this never happens since
457          * we don't set last_task_used_altivec -- Cort
458          */
459         if (new->thread.regs && last_task_used_altivec == new)
460                 new->thread.regs->msr |= MSR_VEC;
461 #endif /* CONFIG_ALTIVEC */
462 #ifdef CONFIG_VSX
463         if (new->thread.regs && last_task_used_vsx == new)
464                 new->thread.regs->msr |= MSR_VSX;
465 #endif /* CONFIG_VSX */
466 #ifdef CONFIG_SPE
467         /* Avoid the trap.  On smp this this never happens since
468          * we don't set last_task_used_spe
469          */
470         if (new->thread.regs && last_task_used_spe == new)
471                 new->thread.regs->msr |= MSR_SPE;
472 #endif /* CONFIG_SPE */
473
474 #endif /* CONFIG_SMP */
475
476 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
477         switch_booke_debug_regs(&new->thread);
478 #else
479 /*
480  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
481  * schedule DABR
482  */
483 #ifndef CONFIG_HAVE_HW_BREAKPOINT
484         if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr))
485                 set_dabr(new->thread.dabr, new->thread.dabrx);
486 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
487 #endif
488
489
490         new_thread = &new->thread;
491         old_thread = &current->thread;
492
493 #ifdef CONFIG_PPC64
494         /*
495          * Collect processor utilization data per process
496          */
497         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
498                 struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
499                 long unsigned start_tb, current_tb;
500                 start_tb = old_thread->start_tb;
501                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
502                 old_thread->accum_tb += (current_tb - start_tb);
503                 new_thread->start_tb = current_tb;
504         }
505 #endif /* CONFIG_PPC64 */
506
507 #ifdef CONFIG_PPC_BOOK3S_64
508         batch = &__get_cpu_var(ppc64_tlb_batch);
509         if (batch->active) {
510                 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
511                 if (batch->index)
512                         __flush_tlb_pending(batch);
513                 batch->active = 0;
514         }
515 #endif /* CONFIG_PPC_BOOK3S_64 */
516
517         local_irq_save(flags);
518
519         account_system_vtime(current);
520         account_process_vtime(current);
521
522         /*
523          * We can't take a PMU exception inside _switch() since there is a
524          * window where the kernel stack SLB and the kernel stack are out
525          * of sync. Hard disable here.
526          */
527         hard_irq_disable();
528         last = _switch(old_thread, new_thread);
529
530 #ifdef CONFIG_PPC_BOOK3S_64
531         if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
532                 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
533                 batch = &__get_cpu_var(ppc64_tlb_batch);
534                 batch->active = 1;
535         }
536 #endif /* CONFIG_PPC_BOOK3S_64 */
537
538         local_irq_restore(flags);
539
540         return last;
541 }
542
543 static int instructions_to_print = 16;
544
545 static void show_instructions(struct pt_regs *regs)
546 {
547         int i;
548         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
549                         sizeof(int));
550
551         printk("Instruction dump:");
552
553         for (i = 0; i < instructions_to_print; i++) {
554                 int instr;
555
556                 if (!(i % 8))
557                         printk("\n");
558
559 #if !defined(CONFIG_BOOKE)
560                 /* If executing with the IMMU off, adjust pc rather
561                  * than print XXXXXXXX.
562                  */
563                 if (!(regs->msr & MSR_IR))
564                         pc = (unsigned long)phys_to_virt(pc);
565 #endif
566
567                 /* We use __get_user here *only* to avoid an OOPS on a
568                  * bad address because the pc *should* only be a
569                  * kernel address.
570                  */
571                 if (!__kernel_text_address(pc) ||
572                      __get_user(instr, (unsigned int __user *)pc)) {
573                         printk(KERN_CONT "XXXXXXXX ");
574                 } else {
575                         if (regs->nip == pc)
576                                 printk(KERN_CONT "<%08x> ", instr);
577                         else
578                                 printk(KERN_CONT "%08x ", instr);
579                 }
580
581                 pc += sizeof(int);
582         }
583
584         printk("\n");
585 }
586
587 static struct regbit {
588         unsigned long bit;
589         const char *name;
590 } msr_bits[] = {
591 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
592         {MSR_SF,        "SF"},
593         {MSR_HV,        "HV"},
594 #endif
595         {MSR_VEC,       "VEC"},
596         {MSR_VSX,       "VSX"},
597 #ifdef CONFIG_BOOKE
598         {MSR_CE,        "CE"},
599 #endif
600         {MSR_EE,        "EE"},
601         {MSR_PR,        "PR"},
602         {MSR_FP,        "FP"},
603         {MSR_ME,        "ME"},
604 #ifdef CONFIG_BOOKE
605         {MSR_DE,        "DE"},
606 #else
607         {MSR_SE,        "SE"},
608         {MSR_BE,        "BE"},
609 #endif
610         {MSR_IR,        "IR"},
611         {MSR_DR,        "DR"},
612         {MSR_PMM,       "PMM"},
613 #ifndef CONFIG_BOOKE
614         {MSR_RI,        "RI"},
615         {MSR_LE,        "LE"},
616 #endif
617         {0,             NULL}
618 };
619
620 static void printbits(unsigned long val, struct regbit *bits)
621 {
622         const char *sep = "";
623
624         printk("<");
625         for (; bits->bit; ++bits)
626                 if (val & bits->bit) {
627                         printk("%s%s", sep, bits->name);
628                         sep = ",";
629                 }
630         printk(">");
631 }
632
633 #ifdef CONFIG_PPC64
634 #define REG             "%016lx"
635 #define REGS_PER_LINE   4
636 #define LAST_VOLATILE   13
637 #else
638 #define REG             "%08lx"
639 #define REGS_PER_LINE   8
640 #define LAST_VOLATILE   12
641 #endif
642
643 void show_regs(struct pt_regs * regs)
644 {
645         int i, trap;
646
647         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
648                regs->nip, regs->link, regs->ctr);
649         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
650                regs, regs->trap, print_tainted(), init_utsname()->release);
651         printk("MSR: "REG" ", regs->msr);
652         printbits(regs->msr, msr_bits);
653         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
654 #ifdef CONFIG_PPC64
655         printk("SOFTE: %ld\n", regs->softe);
656 #endif
657         trap = TRAP(regs);
658         if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
659                 printk("CFAR: "REG"\n", regs->orig_gpr3);
660         if (trap == 0x300 || trap == 0x600)
661 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
662                 printk("DEAR: "REG", ESR: "REG"\n", regs->dar, regs->dsisr);
663 #else
664                 printk("DAR: "REG", DSISR: %08lx\n", regs->dar, regs->dsisr);
665 #endif
666         printk("TASK = %p[%d] '%s' THREAD: %p",
667                current, task_pid_nr(current), current->comm, task_thread_info(current));
668
669 #ifdef CONFIG_SMP
670         printk(" CPU: %d", raw_smp_processor_id());
671 #endif /* CONFIG_SMP */
672
673         for (i = 0;  i < 32;  i++) {
674                 if ((i % REGS_PER_LINE) == 0)
675                         printk("\nGPR%02d: ", i);
676                 printk(REG " ", regs->gpr[i]);
677                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
678                         break;
679         }
680         printk("\n");
681 #ifdef CONFIG_KALLSYMS
682         /*
683          * Lookup NIP late so we have the best change of getting the
684          * above info out without failing
685          */
686         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
687         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
688 #endif
689         show_stack(current, (unsigned long *) regs->gpr[1]);
690         if (!user_mode(regs))
691                 show_instructions(regs);
692 }
693
694 void exit_thread(void)
695 {
696         discard_lazy_cpu_state();
697 }
698
699 void flush_thread(void)
700 {
701         discard_lazy_cpu_state();
702
703 #ifdef CONFIG_HAVE_HW_BREAKPOINT
704         flush_ptrace_hw_breakpoint(current);
705 #else /* CONFIG_HAVE_HW_BREAKPOINT */
706         set_debug_reg_defaults(&current->thread);
707 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
708 }
709
710 void
711 release_thread(struct task_struct *t)
712 {
713 }
714
715 /*
716  * this gets called so that we can store coprocessor state into memory and
717  * copy the current task into the new thread.
718  */
719 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
720 {
721         flush_fp_to_thread(src);
722         flush_altivec_to_thread(src);
723         flush_vsx_to_thread(src);
724         flush_spe_to_thread(src);
725 #ifdef CONFIG_HAVE_HW_BREAKPOINT
726         flush_ptrace_hw_breakpoint(src);
727 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
728
729         *dst = *src;
730         return 0;
731 }
732
733 /*
734  * Copy a thread..
735  */
736 extern unsigned long dscr_default; /* defined in arch/powerpc/kernel/sysfs.c */
737
738 int copy_thread(unsigned long clone_flags, unsigned long usp,
739                 unsigned long unused, struct task_struct *p,
740                 struct pt_regs *regs)
741 {
742         struct pt_regs *childregs, *kregs;
743         extern void ret_from_fork(void);
744         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
745
746         CHECK_FULL_REGS(regs);
747         /* Copy registers */
748         sp -= sizeof(struct pt_regs);
749         childregs = (struct pt_regs *) sp;
750         *childregs = *regs;
751         if ((childregs->msr & MSR_PR) == 0) {
752                 /* for kernel thread, set `current' and stackptr in new task */
753                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
754 #ifdef CONFIG_PPC32
755                 childregs->gpr[2] = (unsigned long) p;
756 #else
757                 clear_tsk_thread_flag(p, TIF_32BIT);
758 #endif
759                 p->thread.regs = NULL;  /* no user register state */
760         } else {
761                 childregs->gpr[1] = usp;
762                 p->thread.regs = childregs;
763                 if (clone_flags & CLONE_SETTLS) {
764 #ifdef CONFIG_PPC64
765                         if (!is_32bit_task())
766                                 childregs->gpr[13] = childregs->gpr[6];
767                         else
768 #endif
769                                 childregs->gpr[2] = childregs->gpr[6];
770                 }
771         }
772         childregs->gpr[3] = 0;  /* Result from fork() */
773         sp -= STACK_FRAME_OVERHEAD;
774
775         /*
776          * The way this works is that at some point in the future
777          * some task will call _switch to switch to the new task.
778          * That will pop off the stack frame created below and start
779          * the new task running at ret_from_fork.  The new task will
780          * do some house keeping and then return from the fork or clone
781          * system call, using the stack frame created above.
782          */
783         sp -= sizeof(struct pt_regs);
784         kregs = (struct pt_regs *) sp;
785         sp -= STACK_FRAME_OVERHEAD;
786         p->thread.ksp = sp;
787         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
788                                 _ALIGN_UP(sizeof(struct thread_info), 16);
789
790 #ifdef CONFIG_PPC_STD_MMU_64
791         if (mmu_has_feature(MMU_FTR_SLB)) {
792                 unsigned long sp_vsid;
793                 unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
794
795                 if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
796                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
797                                 << SLB_VSID_SHIFT_1T;
798                 else
799                         sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
800                                 << SLB_VSID_SHIFT;
801                 sp_vsid |= SLB_VSID_KERNEL | llp;
802                 p->thread.ksp_vsid = sp_vsid;
803         }
804 #endif /* CONFIG_PPC_STD_MMU_64 */
805 #ifdef CONFIG_PPC64 
806         if (cpu_has_feature(CPU_FTR_DSCR)) {
807                 p->thread.dscr_inherit = current->thread.dscr_inherit;
808                 p->thread.dscr = current->thread.dscr;
809         }
810 #endif
811
812         /*
813          * The PPC64 ABI makes use of a TOC to contain function 
814          * pointers.  The function (ret_from_except) is actually a pointer
815          * to the TOC entry.  The first entry is a pointer to the actual
816          * function.
817          */
818 #ifdef CONFIG_PPC64
819         kregs->nip = *((unsigned long *)ret_from_fork);
820 #else
821         kregs->nip = (unsigned long)ret_from_fork;
822 #endif
823
824         return 0;
825 }
826
827 /*
828  * Set up a thread for executing a new program
829  */
830 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
831 {
832 #ifdef CONFIG_PPC64
833         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
834 #endif
835
836         /*
837          * If we exec out of a kernel thread then thread.regs will not be
838          * set.  Do it now.
839          */
840         if (!current->thread.regs) {
841                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
842                 current->thread.regs = regs - 1;
843         }
844
845         memset(regs->gpr, 0, sizeof(regs->gpr));
846         regs->ctr = 0;
847         regs->link = 0;
848         regs->xer = 0;
849         regs->ccr = 0;
850         regs->gpr[1] = sp;
851
852         /*
853          * We have just cleared all the nonvolatile GPRs, so make
854          * FULL_REGS(regs) return true.  This is necessary to allow
855          * ptrace to examine the thread immediately after exec.
856          */
857         regs->trap &= ~1UL;
858
859 #ifdef CONFIG_PPC32
860         regs->mq = 0;
861         regs->nip = start;
862         regs->msr = MSR_USER;
863 #else
864         if (!is_32bit_task()) {
865                 unsigned long entry, toc;
866
867                 /* start is a relocated pointer to the function descriptor for
868                  * the elf _start routine.  The first entry in the function
869                  * descriptor is the entry address of _start and the second
870                  * entry is the TOC value we need to use.
871                  */
872                 __get_user(entry, (unsigned long __user *)start);
873                 __get_user(toc, (unsigned long __user *)start+1);
874
875                 /* Check whether the e_entry function descriptor entries
876                  * need to be relocated before we can use them.
877                  */
878                 if (load_addr != 0) {
879                         entry += load_addr;
880                         toc   += load_addr;
881                 }
882                 regs->nip = entry;
883                 regs->gpr[2] = toc;
884                 regs->msr = MSR_USER64;
885         } else {
886                 regs->nip = start;
887                 regs->gpr[2] = 0;
888                 regs->msr = MSR_USER32;
889         }
890 #endif
891
892         discard_lazy_cpu_state();
893 #ifdef CONFIG_VSX
894         current->thread.used_vsr = 0;
895 #endif
896         memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
897         current->thread.fpscr.val = 0;
898 #ifdef CONFIG_ALTIVEC
899         memset(current->thread.vr, 0, sizeof(current->thread.vr));
900         memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
901         current->thread.vscr.u[3] = 0x00010000; /* Java mode disabled */
902         current->thread.vrsave = 0;
903         current->thread.used_vr = 0;
904 #endif /* CONFIG_ALTIVEC */
905 #ifdef CONFIG_SPE
906         memset(current->thread.evr, 0, sizeof(current->thread.evr));
907         current->thread.acc = 0;
908         current->thread.spefscr = 0;
909         current->thread.used_spe = 0;
910 #endif /* CONFIG_SPE */
911 }
912
913 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
914                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
915
916 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
917 {
918         struct pt_regs *regs = tsk->thread.regs;
919
920         /* This is a bit hairy.  If we are an SPE enabled  processor
921          * (have embedded fp) we store the IEEE exception enable flags in
922          * fpexc_mode.  fpexc_mode is also used for setting FP exception
923          * mode (asyn, precise, disabled) for 'Classic' FP. */
924         if (val & PR_FP_EXC_SW_ENABLE) {
925 #ifdef CONFIG_SPE
926                 if (cpu_has_feature(CPU_FTR_SPE)) {
927                         tsk->thread.fpexc_mode = val &
928                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
929                         return 0;
930                 } else {
931                         return -EINVAL;
932                 }
933 #else
934                 return -EINVAL;
935 #endif
936         }
937
938         /* on a CONFIG_SPE this does not hurt us.  The bits that
939          * __pack_fe01 use do not overlap with bits used for
940          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
941          * on CONFIG_SPE implementations are reserved so writing to
942          * them does not change anything */
943         if (val > PR_FP_EXC_PRECISE)
944                 return -EINVAL;
945         tsk->thread.fpexc_mode = __pack_fe01(val);
946         if (regs != NULL && (regs->msr & MSR_FP) != 0)
947                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
948                         | tsk->thread.fpexc_mode;
949         return 0;
950 }
951
952 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
953 {
954         unsigned int val;
955
956         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
957 #ifdef CONFIG_SPE
958                 if (cpu_has_feature(CPU_FTR_SPE))
959                         val = tsk->thread.fpexc_mode;
960                 else
961                         return -EINVAL;
962 #else
963                 return -EINVAL;
964 #endif
965         else
966                 val = __unpack_fe01(tsk->thread.fpexc_mode);
967         return put_user(val, (unsigned int __user *) adr);
968 }
969
970 int set_endian(struct task_struct *tsk, unsigned int val)
971 {
972         struct pt_regs *regs = tsk->thread.regs;
973
974         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
975             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
976                 return -EINVAL;
977
978         if (regs == NULL)
979                 return -EINVAL;
980
981         if (val == PR_ENDIAN_BIG)
982                 regs->msr &= ~MSR_LE;
983         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
984                 regs->msr |= MSR_LE;
985         else
986                 return -EINVAL;
987
988         return 0;
989 }
990
991 int get_endian(struct task_struct *tsk, unsigned long adr)
992 {
993         struct pt_regs *regs = tsk->thread.regs;
994         unsigned int val;
995
996         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
997             !cpu_has_feature(CPU_FTR_REAL_LE))
998                 return -EINVAL;
999
1000         if (regs == NULL)
1001                 return -EINVAL;
1002
1003         if (regs->msr & MSR_LE) {
1004                 if (cpu_has_feature(CPU_FTR_REAL_LE))
1005                         val = PR_ENDIAN_LITTLE;
1006                 else
1007                         val = PR_ENDIAN_PPC_LITTLE;
1008         } else
1009                 val = PR_ENDIAN_BIG;
1010
1011         return put_user(val, (unsigned int __user *)adr);
1012 }
1013
1014 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1015 {
1016         tsk->thread.align_ctl = val;
1017         return 0;
1018 }
1019
1020 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1021 {
1022         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1023 }
1024
1025 #define TRUNC_PTR(x)    ((typeof(x))(((unsigned long)(x)) & 0xffffffff))
1026
1027 int sys_clone(unsigned long clone_flags, unsigned long usp,
1028               int __user *parent_tidp, void __user *child_threadptr,
1029               int __user *child_tidp, int p6,
1030               struct pt_regs *regs)
1031 {
1032         CHECK_FULL_REGS(regs);
1033         if (usp == 0)
1034                 usp = regs->gpr[1];     /* stack pointer for child */
1035 #ifdef CONFIG_PPC64
1036         if (is_32bit_task()) {
1037                 parent_tidp = TRUNC_PTR(parent_tidp);
1038                 child_tidp = TRUNC_PTR(child_tidp);
1039         }
1040 #endif
1041         return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
1042 }
1043
1044 int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
1045              unsigned long p4, unsigned long p5, unsigned long p6,
1046              struct pt_regs *regs)
1047 {
1048         CHECK_FULL_REGS(regs);
1049         return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
1050 }
1051
1052 int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
1053               unsigned long p4, unsigned long p5, unsigned long p6,
1054               struct pt_regs *regs)
1055 {
1056         CHECK_FULL_REGS(regs);
1057         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
1058                         regs, 0, NULL, NULL);
1059 }
1060
1061 int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
1062                unsigned long a3, unsigned long a4, unsigned long a5,
1063                struct pt_regs *regs)
1064 {
1065         int error;
1066         char *filename;
1067
1068         filename = getname((const char __user *) a0);
1069         error = PTR_ERR(filename);
1070         if (IS_ERR(filename))
1071                 goto out;
1072         flush_fp_to_thread(current);
1073         flush_altivec_to_thread(current);
1074         flush_spe_to_thread(current);
1075         error = do_execve(filename,
1076                           (const char __user *const __user *) a1,
1077                           (const char __user *const __user *) a2, regs);
1078         putname(filename);
1079 out:
1080         return error;
1081 }
1082
1083 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1084                                   unsigned long nbytes)
1085 {
1086         unsigned long stack_page;
1087         unsigned long cpu = task_cpu(p);
1088
1089         /*
1090          * Avoid crashing if the stack has overflowed and corrupted
1091          * task_cpu(p), which is in the thread_info struct.
1092          */
1093         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1094                 stack_page = (unsigned long) hardirq_ctx[cpu];
1095                 if (sp >= stack_page + sizeof(struct thread_struct)
1096                     && sp <= stack_page + THREAD_SIZE - nbytes)
1097                         return 1;
1098
1099                 stack_page = (unsigned long) softirq_ctx[cpu];
1100                 if (sp >= stack_page + sizeof(struct thread_struct)
1101                     && sp <= stack_page + THREAD_SIZE - nbytes)
1102                         return 1;
1103         }
1104         return 0;
1105 }
1106
1107 int validate_sp(unsigned long sp, struct task_struct *p,
1108                        unsigned long nbytes)
1109 {
1110         unsigned long stack_page = (unsigned long)task_stack_page(p);
1111
1112         if (sp >= stack_page + sizeof(struct thread_struct)
1113             && sp <= stack_page + THREAD_SIZE - nbytes)
1114                 return 1;
1115
1116         return valid_irq_stack(sp, p, nbytes);
1117 }
1118
1119 EXPORT_SYMBOL(validate_sp);
1120
1121 unsigned long get_wchan(struct task_struct *p)
1122 {
1123         unsigned long ip, sp;
1124         int count = 0;
1125
1126         if (!p || p == current || p->state == TASK_RUNNING)
1127                 return 0;
1128
1129         sp = p->thread.ksp;
1130         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1131                 return 0;
1132
1133         do {
1134                 sp = *(unsigned long *)sp;
1135                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1136                         return 0;
1137                 if (count > 0) {
1138                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1139                         if (!in_sched_functions(ip))
1140                                 return ip;
1141                 }
1142         } while (count++ < 16);
1143         return 0;
1144 }
1145
1146 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1147
1148 void show_stack(struct task_struct *tsk, unsigned long *stack)
1149 {
1150         unsigned long sp, ip, lr, newsp;
1151         int count = 0;
1152         int firstframe = 1;
1153 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1154         int curr_frame = current->curr_ret_stack;
1155         extern void return_to_handler(void);
1156         unsigned long rth = (unsigned long)return_to_handler;
1157         unsigned long mrth = -1;
1158 #ifdef CONFIG_PPC64
1159         extern void mod_return_to_handler(void);
1160         rth = *(unsigned long *)rth;
1161         mrth = (unsigned long)mod_return_to_handler;
1162         mrth = *(unsigned long *)mrth;
1163 #endif
1164 #endif
1165
1166         sp = (unsigned long) stack;
1167         if (tsk == NULL)
1168                 tsk = current;
1169         if (sp == 0) {
1170                 if (tsk == current)
1171                         asm("mr %0,1" : "=r" (sp));
1172                 else
1173                         sp = tsk->thread.ksp;
1174         }
1175
1176         lr = 0;
1177         printk("Call Trace:\n");
1178         do {
1179                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1180                         return;
1181
1182                 stack = (unsigned long *) sp;
1183                 newsp = stack[0];
1184                 ip = stack[STACK_FRAME_LR_SAVE];
1185                 if (!firstframe || ip != lr) {
1186                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1187 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1188                         if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1189                                 printk(" (%pS)",
1190                                        (void *)current->ret_stack[curr_frame].ret);
1191                                 curr_frame--;
1192                         }
1193 #endif
1194                         if (firstframe)
1195                                 printk(" (unreliable)");
1196                         printk("\n");
1197                 }
1198                 firstframe = 0;
1199
1200                 /*
1201                  * See if this is an exception frame.
1202                  * We look for the "regshere" marker in the current frame.
1203                  */
1204                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1205                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1206                         struct pt_regs *regs = (struct pt_regs *)
1207                                 (sp + STACK_FRAME_OVERHEAD);
1208                         lr = regs->link;
1209                         printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1210                                regs->trap, (void *)regs->nip, (void *)lr);
1211                         firstframe = 1;
1212                 }
1213
1214                 sp = newsp;
1215         } while (count++ < kstack_depth_to_print);
1216 }
1217
1218 void dump_stack(void)
1219 {
1220         show_stack(current, NULL);
1221 }
1222 EXPORT_SYMBOL(dump_stack);
1223
1224 #ifdef CONFIG_PPC64
1225 /* Called with hard IRQs off */
1226 void __ppc64_runlatch_on(void)
1227 {
1228         struct thread_info *ti = current_thread_info();
1229         unsigned long ctrl;
1230
1231         ctrl = mfspr(SPRN_CTRLF);
1232         ctrl |= CTRL_RUNLATCH;
1233         mtspr(SPRN_CTRLT, ctrl);
1234
1235         ti->local_flags |= _TLF_RUNLATCH;
1236 }
1237
1238 /* Called with hard IRQs off */
1239 void __ppc64_runlatch_off(void)
1240 {
1241         struct thread_info *ti = current_thread_info();
1242         unsigned long ctrl;
1243
1244         ti->local_flags &= ~_TLF_RUNLATCH;
1245
1246         ctrl = mfspr(SPRN_CTRLF);
1247         ctrl &= ~CTRL_RUNLATCH;
1248         mtspr(SPRN_CTRLT, ctrl);
1249 }
1250 #endif /* CONFIG_PPC64 */
1251
1252 unsigned long arch_align_stack(unsigned long sp)
1253 {
1254         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1255                 sp -= get_random_int() & ~PAGE_MASK;
1256         return sp & ~0xf;
1257 }
1258
1259 static inline unsigned long brk_rnd(void)
1260 {
1261         unsigned long rnd = 0;
1262
1263         /* 8MB for 32bit, 1GB for 64bit */
1264         if (is_32bit_task())
1265                 rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1266         else
1267                 rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1268
1269         return rnd << PAGE_SHIFT;
1270 }
1271
1272 unsigned long arch_randomize_brk(struct mm_struct *mm)
1273 {
1274         unsigned long base = mm->brk;
1275         unsigned long ret;
1276
1277 #ifdef CONFIG_PPC_STD_MMU_64
1278         /*
1279          * If we are using 1TB segments and we are allowed to randomise
1280          * the heap, we can put it above 1TB so it is backed by a 1TB
1281          * segment. Otherwise the heap will be in the bottom 1TB
1282          * which always uses 256MB segments and this may result in a
1283          * performance penalty.
1284          */
1285         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1286                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1287 #endif
1288
1289         ret = PAGE_ALIGN(base + brk_rnd());
1290
1291         if (ret < mm->brk)
1292                 return mm->brk;
1293
1294         return ret;
1295 }
1296
1297 unsigned long randomize_et_dyn(unsigned long base)
1298 {
1299         unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1300
1301         if (ret < base)
1302                 return base;
1303
1304         return ret;
1305 }