m68k: Migrate exception table users off module.h and onto extable.h
[platform/kernel/linux-exynos.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/prctl.h>
29 #include <linux/init_task.h>
30 #include <linux/export.h>
31 #include <linux/kallsyms.h>
32 #include <linux/mqueue.h>
33 #include <linux/hardirq.h>
34 #include <linux/utsname.h>
35 #include <linux/ftrace.h>
36 #include <linux/kernel_stat.h>
37 #include <linux/personality.h>
38 #include <linux/random.h>
39 #include <linux/hw_breakpoint.h>
40 #include <linux/uaccess.h>
41 #include <linux/elf-randomize.h>
42
43 #include <asm/pgtable.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/tm.h>
54 #include <asm/debug.h>
55 #ifdef CONFIG_PPC64
56 #include <asm/firmware.h>
57 #endif
58 #include <asm/code-patching.h>
59 #include <asm/exec.h>
60 #include <asm/livepatch.h>
61 #include <asm/cpu_has_feature.h>
62
63 #include <linux/kprobes.h>
64 #include <linux/kdebug.h>
65
66 /* Transactional Memory debug */
67 #ifdef TM_DEBUG_SW
68 #define TM_DEBUG(x...) printk(KERN_INFO x)
69 #else
70 #define TM_DEBUG(x...) do { } while(0)
71 #endif
72
73 extern unsigned long _get_SP(void);
74
75 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
76 static void check_if_tm_restore_required(struct task_struct *tsk)
77 {
78         /*
79          * If we are saving the current thread's registers, and the
80          * thread is in a transactional state, set the TIF_RESTORE_TM
81          * bit so that we know to restore the registers before
82          * returning to userspace.
83          */
84         if (tsk == current && tsk->thread.regs &&
85             MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
86             !test_thread_flag(TIF_RESTORE_TM)) {
87                 tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
88                 set_thread_flag(TIF_RESTORE_TM);
89         }
90 }
91 #else
92 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
93 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
94
95 bool strict_msr_control;
96 EXPORT_SYMBOL(strict_msr_control);
97
98 static int __init enable_strict_msr_control(char *str)
99 {
100         strict_msr_control = true;
101         pr_info("Enabling strict facility control\n");
102
103         return 0;
104 }
105 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
106
107 void msr_check_and_set(unsigned long bits)
108 {
109         unsigned long oldmsr = mfmsr();
110         unsigned long newmsr;
111
112         newmsr = oldmsr | bits;
113
114 #ifdef CONFIG_VSX
115         if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
116                 newmsr |= MSR_VSX;
117 #endif
118
119         if (oldmsr != newmsr)
120                 mtmsr_isync(newmsr);
121 }
122
123 void __msr_check_and_clear(unsigned long bits)
124 {
125         unsigned long oldmsr = mfmsr();
126         unsigned long newmsr;
127
128         newmsr = oldmsr & ~bits;
129
130 #ifdef CONFIG_VSX
131         if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
132                 newmsr &= ~MSR_VSX;
133 #endif
134
135         if (oldmsr != newmsr)
136                 mtmsr_isync(newmsr);
137 }
138 EXPORT_SYMBOL(__msr_check_and_clear);
139
140 #ifdef CONFIG_PPC_FPU
141 void __giveup_fpu(struct task_struct *tsk)
142 {
143         unsigned long msr;
144
145         save_fpu(tsk);
146         msr = tsk->thread.regs->msr;
147         msr &= ~MSR_FP;
148 #ifdef CONFIG_VSX
149         if (cpu_has_feature(CPU_FTR_VSX))
150                 msr &= ~MSR_VSX;
151 #endif
152         tsk->thread.regs->msr = msr;
153 }
154
155 void giveup_fpu(struct task_struct *tsk)
156 {
157         check_if_tm_restore_required(tsk);
158
159         msr_check_and_set(MSR_FP);
160         __giveup_fpu(tsk);
161         msr_check_and_clear(MSR_FP);
162 }
163 EXPORT_SYMBOL(giveup_fpu);
164
165 /*
166  * Make sure the floating-point register state in the
167  * the thread_struct is up to date for task tsk.
168  */
169 void flush_fp_to_thread(struct task_struct *tsk)
170 {
171         if (tsk->thread.regs) {
172                 /*
173                  * We need to disable preemption here because if we didn't,
174                  * another process could get scheduled after the regs->msr
175                  * test but before we have finished saving the FP registers
176                  * to the thread_struct.  That process could take over the
177                  * FPU, and then when we get scheduled again we would store
178                  * bogus values for the remaining FP registers.
179                  */
180                 preempt_disable();
181                 if (tsk->thread.regs->msr & MSR_FP) {
182                         /*
183                          * This should only ever be called for current or
184                          * for a stopped child process.  Since we save away
185                          * the FP register state on context switch,
186                          * there is something wrong if a stopped child appears
187                          * to still have its FP state in the CPU registers.
188                          */
189                         BUG_ON(tsk != current);
190                         giveup_fpu(tsk);
191                 }
192                 preempt_enable();
193         }
194 }
195 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
196
197 void enable_kernel_fp(void)
198 {
199         WARN_ON(preemptible());
200
201         msr_check_and_set(MSR_FP);
202
203         if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
204                 check_if_tm_restore_required(current);
205                 __giveup_fpu(current);
206         }
207 }
208 EXPORT_SYMBOL(enable_kernel_fp);
209
210 static int restore_fp(struct task_struct *tsk) {
211         if (tsk->thread.load_fp) {
212                 load_fp_state(&current->thread.fp_state);
213                 current->thread.load_fp++;
214                 return 1;
215         }
216         return 0;
217 }
218 #else
219 static int restore_fp(struct task_struct *tsk) { return 0; }
220 #endif /* CONFIG_PPC_FPU */
221
222 #ifdef CONFIG_ALTIVEC
223 #define loadvec(thr) ((thr).load_vec)
224
225 static void __giveup_altivec(struct task_struct *tsk)
226 {
227         unsigned long msr;
228
229         save_altivec(tsk);
230         msr = tsk->thread.regs->msr;
231         msr &= ~MSR_VEC;
232 #ifdef CONFIG_VSX
233         if (cpu_has_feature(CPU_FTR_VSX))
234                 msr &= ~MSR_VSX;
235 #endif
236         tsk->thread.regs->msr = msr;
237 }
238
239 void giveup_altivec(struct task_struct *tsk)
240 {
241         check_if_tm_restore_required(tsk);
242
243         msr_check_and_set(MSR_VEC);
244         __giveup_altivec(tsk);
245         msr_check_and_clear(MSR_VEC);
246 }
247 EXPORT_SYMBOL(giveup_altivec);
248
249 void enable_kernel_altivec(void)
250 {
251         WARN_ON(preemptible());
252
253         msr_check_and_set(MSR_VEC);
254
255         if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
256                 check_if_tm_restore_required(current);
257                 __giveup_altivec(current);
258         }
259 }
260 EXPORT_SYMBOL(enable_kernel_altivec);
261
262 /*
263  * Make sure the VMX/Altivec register state in the
264  * the thread_struct is up to date for task tsk.
265  */
266 void flush_altivec_to_thread(struct task_struct *tsk)
267 {
268         if (tsk->thread.regs) {
269                 preempt_disable();
270                 if (tsk->thread.regs->msr & MSR_VEC) {
271                         BUG_ON(tsk != current);
272                         giveup_altivec(tsk);
273                 }
274                 preempt_enable();
275         }
276 }
277 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
278
279 static int restore_altivec(struct task_struct *tsk)
280 {
281         if (cpu_has_feature(CPU_FTR_ALTIVEC) && tsk->thread.load_vec) {
282                 load_vr_state(&tsk->thread.vr_state);
283                 tsk->thread.used_vr = 1;
284                 tsk->thread.load_vec++;
285
286                 return 1;
287         }
288         return 0;
289 }
290 #else
291 #define loadvec(thr) 0
292 static inline int restore_altivec(struct task_struct *tsk) { return 0; }
293 #endif /* CONFIG_ALTIVEC */
294
295 #ifdef CONFIG_VSX
296 static void __giveup_vsx(struct task_struct *tsk)
297 {
298         if (tsk->thread.regs->msr & MSR_FP)
299                 __giveup_fpu(tsk);
300         if (tsk->thread.regs->msr & MSR_VEC)
301                 __giveup_altivec(tsk);
302         tsk->thread.regs->msr &= ~MSR_VSX;
303 }
304
305 static void giveup_vsx(struct task_struct *tsk)
306 {
307         check_if_tm_restore_required(tsk);
308
309         msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
310         __giveup_vsx(tsk);
311         msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
312 }
313
314 static void save_vsx(struct task_struct *tsk)
315 {
316         if (tsk->thread.regs->msr & MSR_FP)
317                 save_fpu(tsk);
318         if (tsk->thread.regs->msr & MSR_VEC)
319                 save_altivec(tsk);
320 }
321
322 void enable_kernel_vsx(void)
323 {
324         WARN_ON(preemptible());
325
326         msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
327
328         if (current->thread.regs && (current->thread.regs->msr & MSR_VSX)) {
329                 check_if_tm_restore_required(current);
330                 if (current->thread.regs->msr & MSR_FP)
331                         __giveup_fpu(current);
332                 if (current->thread.regs->msr & MSR_VEC)
333                         __giveup_altivec(current);
334                 __giveup_vsx(current);
335         }
336 }
337 EXPORT_SYMBOL(enable_kernel_vsx);
338
339 void flush_vsx_to_thread(struct task_struct *tsk)
340 {
341         if (tsk->thread.regs) {
342                 preempt_disable();
343                 if (tsk->thread.regs->msr & MSR_VSX) {
344                         BUG_ON(tsk != current);
345                         giveup_vsx(tsk);
346                 }
347                 preempt_enable();
348         }
349 }
350 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
351
352 static int restore_vsx(struct task_struct *tsk)
353 {
354         if (cpu_has_feature(CPU_FTR_VSX)) {
355                 tsk->thread.used_vsr = 1;
356                 return 1;
357         }
358
359         return 0;
360 }
361 #else
362 static inline int restore_vsx(struct task_struct *tsk) { return 0; }
363 static inline void save_vsx(struct task_struct *tsk) { }
364 #endif /* CONFIG_VSX */
365
366 #ifdef CONFIG_SPE
367 void giveup_spe(struct task_struct *tsk)
368 {
369         check_if_tm_restore_required(tsk);
370
371         msr_check_and_set(MSR_SPE);
372         __giveup_spe(tsk);
373         msr_check_and_clear(MSR_SPE);
374 }
375 EXPORT_SYMBOL(giveup_spe);
376
377 void enable_kernel_spe(void)
378 {
379         WARN_ON(preemptible());
380
381         msr_check_and_set(MSR_SPE);
382
383         if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
384                 check_if_tm_restore_required(current);
385                 __giveup_spe(current);
386         }
387 }
388 EXPORT_SYMBOL(enable_kernel_spe);
389
390 void flush_spe_to_thread(struct task_struct *tsk)
391 {
392         if (tsk->thread.regs) {
393                 preempt_disable();
394                 if (tsk->thread.regs->msr & MSR_SPE) {
395                         BUG_ON(tsk != current);
396                         tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
397                         giveup_spe(tsk);
398                 }
399                 preempt_enable();
400         }
401 }
402 #endif /* CONFIG_SPE */
403
404 static unsigned long msr_all_available;
405
406 static int __init init_msr_all_available(void)
407 {
408 #ifdef CONFIG_PPC_FPU
409         msr_all_available |= MSR_FP;
410 #endif
411 #ifdef CONFIG_ALTIVEC
412         if (cpu_has_feature(CPU_FTR_ALTIVEC))
413                 msr_all_available |= MSR_VEC;
414 #endif
415 #ifdef CONFIG_VSX
416         if (cpu_has_feature(CPU_FTR_VSX))
417                 msr_all_available |= MSR_VSX;
418 #endif
419 #ifdef CONFIG_SPE
420         if (cpu_has_feature(CPU_FTR_SPE))
421                 msr_all_available |= MSR_SPE;
422 #endif
423
424         return 0;
425 }
426 early_initcall(init_msr_all_available);
427
428 void giveup_all(struct task_struct *tsk)
429 {
430         unsigned long usermsr;
431
432         if (!tsk->thread.regs)
433                 return;
434
435         usermsr = tsk->thread.regs->msr;
436
437         if ((usermsr & msr_all_available) == 0)
438                 return;
439
440         msr_check_and_set(msr_all_available);
441
442 #ifdef CONFIG_PPC_FPU
443         if (usermsr & MSR_FP)
444                 __giveup_fpu(tsk);
445 #endif
446 #ifdef CONFIG_ALTIVEC
447         if (usermsr & MSR_VEC)
448                 __giveup_altivec(tsk);
449 #endif
450 #ifdef CONFIG_VSX
451         if (usermsr & MSR_VSX)
452                 __giveup_vsx(tsk);
453 #endif
454 #ifdef CONFIG_SPE
455         if (usermsr & MSR_SPE)
456                 __giveup_spe(tsk);
457 #endif
458
459         msr_check_and_clear(msr_all_available);
460 }
461 EXPORT_SYMBOL(giveup_all);
462
463 void restore_math(struct pt_regs *regs)
464 {
465         unsigned long msr;
466
467         if (!current->thread.load_fp && !loadvec(current->thread))
468                 return;
469
470         msr = regs->msr;
471         msr_check_and_set(msr_all_available);
472
473         /*
474          * Only reload if the bit is not set in the user MSR, the bit BEING set
475          * indicates that the registers are hot
476          */
477         if ((!(msr & MSR_FP)) && restore_fp(current))
478                 msr |= MSR_FP | current->thread.fpexc_mode;
479
480         if ((!(msr & MSR_VEC)) && restore_altivec(current))
481                 msr |= MSR_VEC;
482
483         if ((msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC) &&
484                         restore_vsx(current)) {
485                 msr |= MSR_VSX;
486         }
487
488         msr_check_and_clear(msr_all_available);
489
490         regs->msr = msr;
491 }
492
493 void save_all(struct task_struct *tsk)
494 {
495         unsigned long usermsr;
496
497         if (!tsk->thread.regs)
498                 return;
499
500         usermsr = tsk->thread.regs->msr;
501
502         if ((usermsr & msr_all_available) == 0)
503                 return;
504
505         msr_check_and_set(msr_all_available);
506
507         /*
508          * Saving the way the register space is in hardware, save_vsx boils
509          * down to a save_fpu() and save_altivec()
510          */
511         if (usermsr & MSR_VSX) {
512                 save_vsx(tsk);
513         } else {
514                 if (usermsr & MSR_FP)
515                         save_fpu(tsk);
516
517                 if (usermsr & MSR_VEC)
518                         save_altivec(tsk);
519         }
520
521         if (usermsr & MSR_SPE)
522                 __giveup_spe(tsk);
523
524         msr_check_and_clear(msr_all_available);
525 }
526
527 void flush_all_to_thread(struct task_struct *tsk)
528 {
529         if (tsk->thread.regs) {
530                 preempt_disable();
531                 BUG_ON(tsk != current);
532                 save_all(tsk);
533
534 #ifdef CONFIG_SPE
535                 if (tsk->thread.regs->msr & MSR_SPE)
536                         tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
537 #endif
538
539                 preempt_enable();
540         }
541 }
542 EXPORT_SYMBOL(flush_all_to_thread);
543
544 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
545 void do_send_trap(struct pt_regs *regs, unsigned long address,
546                   unsigned long error_code, int signal_code, int breakpt)
547 {
548         siginfo_t info;
549
550         current->thread.trap_nr = signal_code;
551         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
552                         11, SIGSEGV) == NOTIFY_STOP)
553                 return;
554
555         /* Deliver the signal to userspace */
556         info.si_signo = SIGTRAP;
557         info.si_errno = breakpt;        /* breakpoint or watchpoint id */
558         info.si_code = signal_code;
559         info.si_addr = (void __user *)address;
560         force_sig_info(SIGTRAP, &info, current);
561 }
562 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
563 void do_break (struct pt_regs *regs, unsigned long address,
564                     unsigned long error_code)
565 {
566         siginfo_t info;
567
568         current->thread.trap_nr = TRAP_HWBKPT;
569         if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
570                         11, SIGSEGV) == NOTIFY_STOP)
571                 return;
572
573         if (debugger_break_match(regs))
574                 return;
575
576         /* Clear the breakpoint */
577         hw_breakpoint_disable();
578
579         /* Deliver the signal to userspace */
580         info.si_signo = SIGTRAP;
581         info.si_errno = 0;
582         info.si_code = TRAP_HWBKPT;
583         info.si_addr = (void __user *)address;
584         force_sig_info(SIGTRAP, &info, current);
585 }
586 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
587
588 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
589
590 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
591 /*
592  * Set the debug registers back to their default "safe" values.
593  */
594 static void set_debug_reg_defaults(struct thread_struct *thread)
595 {
596         thread->debug.iac1 = thread->debug.iac2 = 0;
597 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
598         thread->debug.iac3 = thread->debug.iac4 = 0;
599 #endif
600         thread->debug.dac1 = thread->debug.dac2 = 0;
601 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
602         thread->debug.dvc1 = thread->debug.dvc2 = 0;
603 #endif
604         thread->debug.dbcr0 = 0;
605 #ifdef CONFIG_BOOKE
606         /*
607          * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
608          */
609         thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
610                         DBCR1_IAC3US | DBCR1_IAC4US;
611         /*
612          * Force Data Address Compare User/Supervisor bits to be User-only
613          * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
614          */
615         thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
616 #else
617         thread->debug.dbcr1 = 0;
618 #endif
619 }
620
621 static void prime_debug_regs(struct debug_reg *debug)
622 {
623         /*
624          * We could have inherited MSR_DE from userspace, since
625          * it doesn't get cleared on exception entry.  Make sure
626          * MSR_DE is clear before we enable any debug events.
627          */
628         mtmsr(mfmsr() & ~MSR_DE);
629
630         mtspr(SPRN_IAC1, debug->iac1);
631         mtspr(SPRN_IAC2, debug->iac2);
632 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
633         mtspr(SPRN_IAC3, debug->iac3);
634         mtspr(SPRN_IAC4, debug->iac4);
635 #endif
636         mtspr(SPRN_DAC1, debug->dac1);
637         mtspr(SPRN_DAC2, debug->dac2);
638 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
639         mtspr(SPRN_DVC1, debug->dvc1);
640         mtspr(SPRN_DVC2, debug->dvc2);
641 #endif
642         mtspr(SPRN_DBCR0, debug->dbcr0);
643         mtspr(SPRN_DBCR1, debug->dbcr1);
644 #ifdef CONFIG_BOOKE
645         mtspr(SPRN_DBCR2, debug->dbcr2);
646 #endif
647 }
648 /*
649  * Unless neither the old or new thread are making use of the
650  * debug registers, set the debug registers from the values
651  * stored in the new thread.
652  */
653 void switch_booke_debug_regs(struct debug_reg *new_debug)
654 {
655         if ((current->thread.debug.dbcr0 & DBCR0_IDM)
656                 || (new_debug->dbcr0 & DBCR0_IDM))
657                         prime_debug_regs(new_debug);
658 }
659 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
660 #else   /* !CONFIG_PPC_ADV_DEBUG_REGS */
661 #ifndef CONFIG_HAVE_HW_BREAKPOINT
662 static void set_debug_reg_defaults(struct thread_struct *thread)
663 {
664         thread->hw_brk.address = 0;
665         thread->hw_brk.type = 0;
666         set_breakpoint(&thread->hw_brk);
667 }
668 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
669 #endif  /* CONFIG_PPC_ADV_DEBUG_REGS */
670
671 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
672 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
673 {
674         mtspr(SPRN_DAC1, dabr);
675 #ifdef CONFIG_PPC_47x
676         isync();
677 #endif
678         return 0;
679 }
680 #elif defined(CONFIG_PPC_BOOK3S)
681 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
682 {
683         mtspr(SPRN_DABR, dabr);
684         if (cpu_has_feature(CPU_FTR_DABRX))
685                 mtspr(SPRN_DABRX, dabrx);
686         return 0;
687 }
688 #else
689 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
690 {
691         return -EINVAL;
692 }
693 #endif
694
695 static inline int set_dabr(struct arch_hw_breakpoint *brk)
696 {
697         unsigned long dabr, dabrx;
698
699         dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
700         dabrx = ((brk->type >> 3) & 0x7);
701
702         if (ppc_md.set_dabr)
703                 return ppc_md.set_dabr(dabr, dabrx);
704
705         return __set_dabr(dabr, dabrx);
706 }
707
708 static inline int set_dawr(struct arch_hw_breakpoint *brk)
709 {
710         unsigned long dawr, dawrx, mrd;
711
712         dawr = brk->address;
713
714         dawrx  = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
715                                    << (63 - 58); //* read/write bits */
716         dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
717                                    << (63 - 59); //* translate */
718         dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
719                                    >> 3; //* PRIM bits */
720         /* dawr length is stored in field MDR bits 48:53.  Matches range in
721            doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
722            0b111111=64DW.
723            brk->len is in bytes.
724            This aligns up to double word size, shifts and does the bias.
725         */
726         mrd = ((brk->len + 7) >> 3) - 1;
727         dawrx |= (mrd & 0x3f) << (63 - 53);
728
729         if (ppc_md.set_dawr)
730                 return ppc_md.set_dawr(dawr, dawrx);
731         mtspr(SPRN_DAWR, dawr);
732         mtspr(SPRN_DAWRX, dawrx);
733         return 0;
734 }
735
736 void __set_breakpoint(struct arch_hw_breakpoint *brk)
737 {
738         memcpy(this_cpu_ptr(&current_brk), brk, sizeof(*brk));
739
740         if (cpu_has_feature(CPU_FTR_DAWR))
741                 set_dawr(brk);
742         else
743                 set_dabr(brk);
744 }
745
746 void set_breakpoint(struct arch_hw_breakpoint *brk)
747 {
748         preempt_disable();
749         __set_breakpoint(brk);
750         preempt_enable();
751 }
752
753 #ifdef CONFIG_PPC64
754 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
755 #endif
756
757 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
758                               struct arch_hw_breakpoint *b)
759 {
760         if (a->address != b->address)
761                 return false;
762         if (a->type != b->type)
763                 return false;
764         if (a->len != b->len)
765                 return false;
766         return true;
767 }
768
769 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
770 static void tm_reclaim_thread(struct thread_struct *thr,
771                               struct thread_info *ti, uint8_t cause)
772 {
773         unsigned long msr_diff = 0;
774
775         /*
776          * If FP/VSX registers have been already saved to the
777          * thread_struct, move them to the transact_fp array.
778          * We clear the TIF_RESTORE_TM bit since after the reclaim
779          * the thread will no longer be transactional.
780          */
781         if (test_ti_thread_flag(ti, TIF_RESTORE_TM)) {
782                 msr_diff = thr->ckpt_regs.msr & ~thr->regs->msr;
783                 if (msr_diff & MSR_FP)
784                         memcpy(&thr->transact_fp, &thr->fp_state,
785                                sizeof(struct thread_fp_state));
786                 if (msr_diff & MSR_VEC)
787                         memcpy(&thr->transact_vr, &thr->vr_state,
788                                sizeof(struct thread_vr_state));
789                 clear_ti_thread_flag(ti, TIF_RESTORE_TM);
790                 msr_diff &= MSR_FP | MSR_VEC | MSR_VSX | MSR_FE0 | MSR_FE1;
791         }
792
793         /*
794          * Use the current MSR TM suspended bit to track if we have
795          * checkpointed state outstanding.
796          * On signal delivery, we'd normally reclaim the checkpointed
797          * state to obtain stack pointer (see:get_tm_stackpointer()).
798          * This will then directly return to userspace without going
799          * through __switch_to(). However, if the stack frame is bad,
800          * we need to exit this thread which calls __switch_to() which
801          * will again attempt to reclaim the already saved tm state.
802          * Hence we need to check that we've not already reclaimed
803          * this state.
804          * We do this using the current MSR, rather tracking it in
805          * some specific thread_struct bit, as it has the additional
806          * benefit of checking for a potential TM bad thing exception.
807          */
808         if (!MSR_TM_SUSPENDED(mfmsr()))
809                 return;
810
811         tm_reclaim(thr, thr->regs->msr, cause);
812
813         /* Having done the reclaim, we now have the checkpointed
814          * FP/VSX values in the registers.  These might be valid
815          * even if we have previously called enable_kernel_fp() or
816          * flush_fp_to_thread(), so update thr->regs->msr to
817          * indicate their current validity.
818          */
819         thr->regs->msr |= msr_diff;
820 }
821
822 void tm_reclaim_current(uint8_t cause)
823 {
824         tm_enable();
825         tm_reclaim_thread(&current->thread, current_thread_info(), cause);
826 }
827
828 static inline void tm_reclaim_task(struct task_struct *tsk)
829 {
830         /* We have to work out if we're switching from/to a task that's in the
831          * middle of a transaction.
832          *
833          * In switching we need to maintain a 2nd register state as
834          * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
835          * checkpointed (tbegin) state in ckpt_regs and saves the transactional
836          * (current) FPRs into oldtask->thread.transact_fpr[].
837          *
838          * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
839          */
840         struct thread_struct *thr = &tsk->thread;
841
842         if (!thr->regs)
843                 return;
844
845         if (!MSR_TM_ACTIVE(thr->regs->msr))
846                 goto out_and_saveregs;
847
848         /* Stash the original thread MSR, as giveup_fpu et al will
849          * modify it.  We hold onto it to see whether the task used
850          * FP & vector regs.  If the TIF_RESTORE_TM flag is set,
851          * ckpt_regs.msr is already set.
852          */
853         if (!test_ti_thread_flag(task_thread_info(tsk), TIF_RESTORE_TM))
854                 thr->ckpt_regs.msr = thr->regs->msr;
855
856         TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
857                  "ccr=%lx, msr=%lx, trap=%lx)\n",
858                  tsk->pid, thr->regs->nip,
859                  thr->regs->ccr, thr->regs->msr,
860                  thr->regs->trap);
861
862         tm_reclaim_thread(thr, task_thread_info(tsk), TM_CAUSE_RESCHED);
863
864         TM_DEBUG("--- tm_reclaim on pid %d complete\n",
865                  tsk->pid);
866
867 out_and_saveregs:
868         /* Always save the regs here, even if a transaction's not active.
869          * This context-switches a thread's TM info SPRs.  We do it here to
870          * be consistent with the restore path (in recheckpoint) which
871          * cannot happen later in _switch().
872          */
873         tm_save_sprs(thr);
874 }
875
876 extern void __tm_recheckpoint(struct thread_struct *thread,
877                               unsigned long orig_msr);
878
879 void tm_recheckpoint(struct thread_struct *thread,
880                      unsigned long orig_msr)
881 {
882         unsigned long flags;
883
884         /* We really can't be interrupted here as the TEXASR registers can't
885          * change and later in the trecheckpoint code, we have a userspace R1.
886          * So let's hard disable over this region.
887          */
888         local_irq_save(flags);
889         hard_irq_disable();
890
891         /* The TM SPRs are restored here, so that TEXASR.FS can be set
892          * before the trecheckpoint and no explosion occurs.
893          */
894         tm_restore_sprs(thread);
895
896         __tm_recheckpoint(thread, orig_msr);
897
898         local_irq_restore(flags);
899 }
900
901 static inline void tm_recheckpoint_new_task(struct task_struct *new)
902 {
903         unsigned long msr;
904
905         if (!cpu_has_feature(CPU_FTR_TM))
906                 return;
907
908         /* Recheckpoint the registers of the thread we're about to switch to.
909          *
910          * If the task was using FP, we non-lazily reload both the original and
911          * the speculative FP register states.  This is because the kernel
912          * doesn't see if/when a TM rollback occurs, so if we take an FP
913          * unavoidable later, we are unable to determine which set of FP regs
914          * need to be restored.
915          */
916         if (!new->thread.regs)
917                 return;
918
919         if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
920                 tm_restore_sprs(&new->thread);
921                 return;
922         }
923         msr = new->thread.ckpt_regs.msr;
924         /* Recheckpoint to restore original checkpointed register state. */
925         TM_DEBUG("*** tm_recheckpoint of pid %d "
926                  "(new->msr 0x%lx, new->origmsr 0x%lx)\n",
927                  new->pid, new->thread.regs->msr, msr);
928
929         /* This loads the checkpointed FP/VEC state, if used */
930         tm_recheckpoint(&new->thread, msr);
931
932         /* This loads the speculative FP/VEC state, if used */
933         if (msr & MSR_FP) {
934                 do_load_up_transact_fpu(&new->thread);
935                 new->thread.regs->msr |=
936                         (MSR_FP | new->thread.fpexc_mode);
937         }
938 #ifdef CONFIG_ALTIVEC
939         if (msr & MSR_VEC) {
940                 do_load_up_transact_altivec(&new->thread);
941                 new->thread.regs->msr |= MSR_VEC;
942         }
943 #endif
944         /* We may as well turn on VSX too since all the state is restored now */
945         if (msr & MSR_VSX)
946                 new->thread.regs->msr |= MSR_VSX;
947
948         TM_DEBUG("*** tm_recheckpoint of pid %d complete "
949                  "(kernel msr 0x%lx)\n",
950                  new->pid, mfmsr());
951 }
952
953 static inline void __switch_to_tm(struct task_struct *prev)
954 {
955         if (cpu_has_feature(CPU_FTR_TM)) {
956                 tm_enable();
957                 tm_reclaim_task(prev);
958         }
959 }
960
961 /*
962  * This is called if we are on the way out to userspace and the
963  * TIF_RESTORE_TM flag is set.  It checks if we need to reload
964  * FP and/or vector state and does so if necessary.
965  * If userspace is inside a transaction (whether active or
966  * suspended) and FP/VMX/VSX instructions have ever been enabled
967  * inside that transaction, then we have to keep them enabled
968  * and keep the FP/VMX/VSX state loaded while ever the transaction
969  * continues.  The reason is that if we didn't, and subsequently
970  * got a FP/VMX/VSX unavailable interrupt inside a transaction,
971  * we don't know whether it's the same transaction, and thus we
972  * don't know which of the checkpointed state and the transactional
973  * state to use.
974  */
975 void restore_tm_state(struct pt_regs *regs)
976 {
977         unsigned long msr_diff;
978
979         clear_thread_flag(TIF_RESTORE_TM);
980         if (!MSR_TM_ACTIVE(regs->msr))
981                 return;
982
983         msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
984         msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
985
986         restore_math(regs);
987
988         regs->msr |= msr_diff;
989 }
990
991 #else
992 #define tm_recheckpoint_new_task(new)
993 #define __switch_to_tm(prev)
994 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
995
996 static inline void save_sprs(struct thread_struct *t)
997 {
998 #ifdef CONFIG_ALTIVEC
999         if (cpu_has_feature(CPU_FTR_ALTIVEC))
1000                 t->vrsave = mfspr(SPRN_VRSAVE);
1001 #endif
1002 #ifdef CONFIG_PPC_BOOK3S_64
1003         if (cpu_has_feature(CPU_FTR_DSCR))
1004                 t->dscr = mfspr(SPRN_DSCR);
1005
1006         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1007                 t->bescr = mfspr(SPRN_BESCR);
1008                 t->ebbhr = mfspr(SPRN_EBBHR);
1009                 t->ebbrr = mfspr(SPRN_EBBRR);
1010
1011                 t->fscr = mfspr(SPRN_FSCR);
1012
1013                 /*
1014                  * Note that the TAR is not available for use in the kernel.
1015                  * (To provide this, the TAR should be backed up/restored on
1016                  * exception entry/exit instead, and be in pt_regs.  FIXME,
1017                  * this should be in pt_regs anyway (for debug).)
1018                  */
1019                 t->tar = mfspr(SPRN_TAR);
1020         }
1021
1022         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1023                 /* Conditionally save Load Monitor registers, if enabled */
1024                 if (t->fscr & FSCR_LM) {
1025                         t->lmrr = mfspr(SPRN_LMRR);
1026                         t->lmser = mfspr(SPRN_LMSER);
1027                 }
1028         }
1029 #endif
1030 }
1031
1032 static inline void restore_sprs(struct thread_struct *old_thread,
1033                                 struct thread_struct *new_thread)
1034 {
1035 #ifdef CONFIG_ALTIVEC
1036         if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1037             old_thread->vrsave != new_thread->vrsave)
1038                 mtspr(SPRN_VRSAVE, new_thread->vrsave);
1039 #endif
1040 #ifdef CONFIG_PPC_BOOK3S_64
1041         if (cpu_has_feature(CPU_FTR_DSCR)) {
1042                 u64 dscr = get_paca()->dscr_default;
1043                 if (new_thread->dscr_inherit)
1044                         dscr = new_thread->dscr;
1045
1046                 if (old_thread->dscr != dscr)
1047                         mtspr(SPRN_DSCR, dscr);
1048         }
1049
1050         if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1051                 if (old_thread->bescr != new_thread->bescr)
1052                         mtspr(SPRN_BESCR, new_thread->bescr);
1053                 if (old_thread->ebbhr != new_thread->ebbhr)
1054                         mtspr(SPRN_EBBHR, new_thread->ebbhr);
1055                 if (old_thread->ebbrr != new_thread->ebbrr)
1056                         mtspr(SPRN_EBBRR, new_thread->ebbrr);
1057
1058                 if (old_thread->fscr != new_thread->fscr)
1059                         mtspr(SPRN_FSCR, new_thread->fscr);
1060
1061                 if (old_thread->tar != new_thread->tar)
1062                         mtspr(SPRN_TAR, new_thread->tar);
1063         }
1064
1065         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1066                 /* Conditionally restore Load Monitor registers, if enabled */
1067                 if (new_thread->fscr & FSCR_LM) {
1068                         if (old_thread->lmrr != new_thread->lmrr)
1069                                 mtspr(SPRN_LMRR, new_thread->lmrr);
1070                         if (old_thread->lmser != new_thread->lmser)
1071                                 mtspr(SPRN_LMSER, new_thread->lmser);
1072                 }
1073         }
1074 #endif
1075 }
1076
1077 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1078 void flush_tmregs_to_thread(struct task_struct *tsk)
1079 {
1080         /*
1081          * Process self tracing is not yet supported through
1082          * ptrace interface. Ptrace generic code should have
1083          * prevented this from happening in the first place.
1084          * Warn once here with the message, if some how it
1085          * is attempted.
1086          */
1087         WARN_ONCE(tsk == current,
1088                 "Not expecting ptrace on self: TM regs may be incorrect\n");
1089
1090         /*
1091          * If task is not current, it should have been flushed
1092          * already to it's thread_struct during __switch_to().
1093          */
1094 }
1095 #endif
1096
1097 struct task_struct *__switch_to(struct task_struct *prev,
1098         struct task_struct *new)
1099 {
1100         struct thread_struct *new_thread, *old_thread;
1101         struct task_struct *last;
1102 #ifdef CONFIG_PPC_BOOK3S_64
1103         struct ppc64_tlb_batch *batch;
1104 #endif
1105
1106         new_thread = &new->thread;
1107         old_thread = &current->thread;
1108
1109         WARN_ON(!irqs_disabled());
1110
1111 #ifdef CONFIG_PPC64
1112         /*
1113          * Collect processor utilization data per process
1114          */
1115         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
1116                 struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
1117                 long unsigned start_tb, current_tb;
1118                 start_tb = old_thread->start_tb;
1119                 cu->current_tb = current_tb = mfspr(SPRN_PURR);
1120                 old_thread->accum_tb += (current_tb - start_tb);
1121                 new_thread->start_tb = current_tb;
1122         }
1123 #endif /* CONFIG_PPC64 */
1124
1125 #ifdef CONFIG_PPC_STD_MMU_64
1126         batch = this_cpu_ptr(&ppc64_tlb_batch);
1127         if (batch->active) {
1128                 current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1129                 if (batch->index)
1130                         __flush_tlb_pending(batch);
1131                 batch->active = 0;
1132         }
1133 #endif /* CONFIG_PPC_STD_MMU_64 */
1134
1135 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1136         switch_booke_debug_regs(&new->thread.debug);
1137 #else
1138 /*
1139  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1140  * schedule DABR
1141  */
1142 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1143         if (unlikely(!hw_brk_match(this_cpu_ptr(&current_brk), &new->thread.hw_brk)))
1144                 __set_breakpoint(&new->thread.hw_brk);
1145 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1146 #endif
1147
1148         /*
1149          * We need to save SPRs before treclaim/trecheckpoint as these will
1150          * change a number of them.
1151          */
1152         save_sprs(&prev->thread);
1153
1154         __switch_to_tm(prev);
1155
1156         /* Save FPU, Altivec, VSX and SPE state */
1157         giveup_all(prev);
1158
1159         /*
1160          * We can't take a PMU exception inside _switch() since there is a
1161          * window where the kernel stack SLB and the kernel stack are out
1162          * of sync. Hard disable here.
1163          */
1164         hard_irq_disable();
1165
1166         tm_recheckpoint_new_task(new);
1167
1168         /*
1169          * Call restore_sprs() before calling _switch(). If we move it after
1170          * _switch() then we miss out on calling it for new tasks. The reason
1171          * for this is we manually create a stack frame for new tasks that
1172          * directly returns through ret_from_fork() or
1173          * ret_from_kernel_thread(). See copy_thread() for details.
1174          */
1175         restore_sprs(old_thread, new_thread);
1176
1177         last = _switch(old_thread, new_thread);
1178
1179 #ifdef CONFIG_PPC_STD_MMU_64
1180         if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1181                 current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1182                 batch = this_cpu_ptr(&ppc64_tlb_batch);
1183                 batch->active = 1;
1184         }
1185
1186         if (current_thread_info()->task->thread.regs)
1187                 restore_math(current_thread_info()->task->thread.regs);
1188 #endif /* CONFIG_PPC_STD_MMU_64 */
1189
1190         return last;
1191 }
1192
1193 static int instructions_to_print = 16;
1194
1195 static void show_instructions(struct pt_regs *regs)
1196 {
1197         int i;
1198         unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
1199                         sizeof(int));
1200
1201         printk("Instruction dump:");
1202
1203         for (i = 0; i < instructions_to_print; i++) {
1204                 int instr;
1205
1206                 if (!(i % 8))
1207                         printk("\n");
1208
1209 #if !defined(CONFIG_BOOKE)
1210                 /* If executing with the IMMU off, adjust pc rather
1211                  * than print XXXXXXXX.
1212                  */
1213                 if (!(regs->msr & MSR_IR))
1214                         pc = (unsigned long)phys_to_virt(pc);
1215 #endif
1216
1217                 if (!__kernel_text_address(pc) ||
1218                      probe_kernel_address((unsigned int __user *)pc, instr)) {
1219                         printk(KERN_CONT "XXXXXXXX ");
1220                 } else {
1221                         if (regs->nip == pc)
1222                                 printk(KERN_CONT "<%08x> ", instr);
1223                         else
1224                                 printk(KERN_CONT "%08x ", instr);
1225                 }
1226
1227                 pc += sizeof(int);
1228         }
1229
1230         printk("\n");
1231 }
1232
1233 struct regbit {
1234         unsigned long bit;
1235         const char *name;
1236 };
1237
1238 static struct regbit msr_bits[] = {
1239 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1240         {MSR_SF,        "SF"},
1241         {MSR_HV,        "HV"},
1242 #endif
1243         {MSR_VEC,       "VEC"},
1244         {MSR_VSX,       "VSX"},
1245 #ifdef CONFIG_BOOKE
1246         {MSR_CE,        "CE"},
1247 #endif
1248         {MSR_EE,        "EE"},
1249         {MSR_PR,        "PR"},
1250         {MSR_FP,        "FP"},
1251         {MSR_ME,        "ME"},
1252 #ifdef CONFIG_BOOKE
1253         {MSR_DE,        "DE"},
1254 #else
1255         {MSR_SE,        "SE"},
1256         {MSR_BE,        "BE"},
1257 #endif
1258         {MSR_IR,        "IR"},
1259         {MSR_DR,        "DR"},
1260         {MSR_PMM,       "PMM"},
1261 #ifndef CONFIG_BOOKE
1262         {MSR_RI,        "RI"},
1263         {MSR_LE,        "LE"},
1264 #endif
1265         {0,             NULL}
1266 };
1267
1268 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1269 {
1270         const char *s = "";
1271
1272         for (; bits->bit; ++bits)
1273                 if (val & bits->bit) {
1274                         printk("%s%s", s, bits->name);
1275                         s = sep;
1276                 }
1277 }
1278
1279 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1280 static struct regbit msr_tm_bits[] = {
1281         {MSR_TS_T,      "T"},
1282         {MSR_TS_S,      "S"},
1283         {MSR_TM,        "E"},
1284         {0,             NULL}
1285 };
1286
1287 static void print_tm_bits(unsigned long val)
1288 {
1289 /*
1290  * This only prints something if at least one of the TM bit is set.
1291  * Inside the TM[], the output means:
1292  *   E: Enabled         (bit 32)
1293  *   S: Suspended       (bit 33)
1294  *   T: Transactional   (bit 34)
1295  */
1296         if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1297                 printk(",TM[");
1298                 print_bits(val, msr_tm_bits, "");
1299                 printk("]");
1300         }
1301 }
1302 #else
1303 static void print_tm_bits(unsigned long val) {}
1304 #endif
1305
1306 static void print_msr_bits(unsigned long val)
1307 {
1308         printk("<");
1309         print_bits(val, msr_bits, ",");
1310         print_tm_bits(val);
1311         printk(">");
1312 }
1313
1314 #ifdef CONFIG_PPC64
1315 #define REG             "%016lx"
1316 #define REGS_PER_LINE   4
1317 #define LAST_VOLATILE   13
1318 #else
1319 #define REG             "%08lx"
1320 #define REGS_PER_LINE   8
1321 #define LAST_VOLATILE   12
1322 #endif
1323
1324 void show_regs(struct pt_regs * regs)
1325 {
1326         int i, trap;
1327
1328         show_regs_print_info(KERN_DEFAULT);
1329
1330         printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
1331                regs->nip, regs->link, regs->ctr);
1332         printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
1333                regs, regs->trap, print_tainted(), init_utsname()->release);
1334         printk("MSR: "REG" ", regs->msr);
1335         print_msr_bits(regs->msr);
1336         printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1337         trap = TRAP(regs);
1338         if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
1339                 printk("CFAR: "REG" ", regs->orig_gpr3);
1340         if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1341 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1342                 printk("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1343 #else
1344                 printk("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1345 #endif
1346 #ifdef CONFIG_PPC64
1347         printk("SOFTE: %ld ", regs->softe);
1348 #endif
1349 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1350         if (MSR_TM_ACTIVE(regs->msr))
1351                 printk("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1352 #endif
1353
1354         for (i = 0;  i < 32;  i++) {
1355                 if ((i % REGS_PER_LINE) == 0)
1356                         printk("\nGPR%02d: ", i);
1357                 printk(REG " ", regs->gpr[i]);
1358                 if (i == LAST_VOLATILE && !FULL_REGS(regs))
1359                         break;
1360         }
1361         printk("\n");
1362 #ifdef CONFIG_KALLSYMS
1363         /*
1364          * Lookup NIP late so we have the best change of getting the
1365          * above info out without failing
1366          */
1367         printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1368         printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1369 #endif
1370         show_stack(current, (unsigned long *) regs->gpr[1]);
1371         if (!user_mode(regs))
1372                 show_instructions(regs);
1373 }
1374
1375 void flush_thread(void)
1376 {
1377 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1378         flush_ptrace_hw_breakpoint(current);
1379 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1380         set_debug_reg_defaults(&current->thread);
1381 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1382 }
1383
1384 void
1385 release_thread(struct task_struct *t)
1386 {
1387 }
1388
1389 /*
1390  * this gets called so that we can store coprocessor state into memory and
1391  * copy the current task into the new thread.
1392  */
1393 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1394 {
1395         flush_all_to_thread(src);
1396         /*
1397          * Flush TM state out so we can copy it.  __switch_to_tm() does this
1398          * flush but it removes the checkpointed state from the current CPU and
1399          * transitions the CPU out of TM mode.  Hence we need to call
1400          * tm_recheckpoint_new_task() (on the same task) to restore the
1401          * checkpointed state back and the TM mode.
1402          */
1403         __switch_to_tm(src);
1404         tm_recheckpoint_new_task(src);
1405
1406         *dst = *src;
1407
1408         clear_task_ebb(dst);
1409
1410         return 0;
1411 }
1412
1413 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1414 {
1415 #ifdef CONFIG_PPC_STD_MMU_64
1416         unsigned long sp_vsid;
1417         unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1418
1419         if (radix_enabled())
1420                 return;
1421
1422         if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1423                 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1424                         << SLB_VSID_SHIFT_1T;
1425         else
1426                 sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1427                         << SLB_VSID_SHIFT;
1428         sp_vsid |= SLB_VSID_KERNEL | llp;
1429         p->thread.ksp_vsid = sp_vsid;
1430 #endif
1431 }
1432
1433 /*
1434  * Copy a thread..
1435  */
1436
1437 /*
1438  * Copy architecture-specific thread state
1439  */
1440 int copy_thread(unsigned long clone_flags, unsigned long usp,
1441                 unsigned long kthread_arg, struct task_struct *p)
1442 {
1443         struct pt_regs *childregs, *kregs;
1444         extern void ret_from_fork(void);
1445         extern void ret_from_kernel_thread(void);
1446         void (*f)(void);
1447         unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1448         struct thread_info *ti = task_thread_info(p);
1449
1450         klp_init_thread_info(ti);
1451
1452         /* Copy registers */
1453         sp -= sizeof(struct pt_regs);
1454         childregs = (struct pt_regs *) sp;
1455         if (unlikely(p->flags & PF_KTHREAD)) {
1456                 /* kernel thread */
1457                 memset(childregs, 0, sizeof(struct pt_regs));
1458                 childregs->gpr[1] = sp + sizeof(struct pt_regs);
1459                 /* function */
1460                 if (usp)
1461                         childregs->gpr[14] = ppc_function_entry((void *)usp);
1462 #ifdef CONFIG_PPC64
1463                 clear_tsk_thread_flag(p, TIF_32BIT);
1464                 childregs->softe = 1;
1465 #endif
1466                 childregs->gpr[15] = kthread_arg;
1467                 p->thread.regs = NULL;  /* no user register state */
1468                 ti->flags |= _TIF_RESTOREALL;
1469                 f = ret_from_kernel_thread;
1470         } else {
1471                 /* user thread */
1472                 struct pt_regs *regs = current_pt_regs();
1473                 CHECK_FULL_REGS(regs);
1474                 *childregs = *regs;
1475                 if (usp)
1476                         childregs->gpr[1] = usp;
1477                 p->thread.regs = childregs;
1478                 childregs->gpr[3] = 0;  /* Result from fork() */
1479                 if (clone_flags & CLONE_SETTLS) {
1480 #ifdef CONFIG_PPC64
1481                         if (!is_32bit_task())
1482                                 childregs->gpr[13] = childregs->gpr[6];
1483                         else
1484 #endif
1485                                 childregs->gpr[2] = childregs->gpr[6];
1486                 }
1487
1488                 f = ret_from_fork;
1489         }
1490         childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1491         sp -= STACK_FRAME_OVERHEAD;
1492
1493         /*
1494          * The way this works is that at some point in the future
1495          * some task will call _switch to switch to the new task.
1496          * That will pop off the stack frame created below and start
1497          * the new task running at ret_from_fork.  The new task will
1498          * do some house keeping and then return from the fork or clone
1499          * system call, using the stack frame created above.
1500          */
1501         ((unsigned long *)sp)[0] = 0;
1502         sp -= sizeof(struct pt_regs);
1503         kregs = (struct pt_regs *) sp;
1504         sp -= STACK_FRAME_OVERHEAD;
1505         p->thread.ksp = sp;
1506 #ifdef CONFIG_PPC32
1507         p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1508                                 _ALIGN_UP(sizeof(struct thread_info), 16);
1509 #endif
1510 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1511         p->thread.ptrace_bps[0] = NULL;
1512 #endif
1513
1514         p->thread.fp_save_area = NULL;
1515 #ifdef CONFIG_ALTIVEC
1516         p->thread.vr_save_area = NULL;
1517 #endif
1518
1519         setup_ksp_vsid(p, sp);
1520
1521 #ifdef CONFIG_PPC64 
1522         if (cpu_has_feature(CPU_FTR_DSCR)) {
1523                 p->thread.dscr_inherit = current->thread.dscr_inherit;
1524                 p->thread.dscr = mfspr(SPRN_DSCR);
1525         }
1526         if (cpu_has_feature(CPU_FTR_HAS_PPR))
1527                 p->thread.ppr = INIT_PPR;
1528 #endif
1529         kregs->nip = ppc_function_entry(f);
1530         return 0;
1531 }
1532
1533 /*
1534  * Set up a thread for executing a new program
1535  */
1536 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1537 {
1538 #ifdef CONFIG_PPC64
1539         unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
1540 #endif
1541
1542         /*
1543          * If we exec out of a kernel thread then thread.regs will not be
1544          * set.  Do it now.
1545          */
1546         if (!current->thread.regs) {
1547                 struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1548                 current->thread.regs = regs - 1;
1549         }
1550
1551 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1552         /*
1553          * Clear any transactional state, we're exec()ing. The cause is
1554          * not important as there will never be a recheckpoint so it's not
1555          * user visible.
1556          */
1557         if (MSR_TM_SUSPENDED(mfmsr()))
1558                 tm_reclaim_current(0);
1559 #endif
1560
1561         memset(regs->gpr, 0, sizeof(regs->gpr));
1562         regs->ctr = 0;
1563         regs->link = 0;
1564         regs->xer = 0;
1565         regs->ccr = 0;
1566         regs->gpr[1] = sp;
1567
1568         /*
1569          * We have just cleared all the nonvolatile GPRs, so make
1570          * FULL_REGS(regs) return true.  This is necessary to allow
1571          * ptrace to examine the thread immediately after exec.
1572          */
1573         regs->trap &= ~1UL;
1574
1575 #ifdef CONFIG_PPC32
1576         regs->mq = 0;
1577         regs->nip = start;
1578         regs->msr = MSR_USER;
1579 #else
1580         if (!is_32bit_task()) {
1581                 unsigned long entry;
1582
1583                 if (is_elf2_task()) {
1584                         /* Look ma, no function descriptors! */
1585                         entry = start;
1586
1587                         /*
1588                          * Ulrich says:
1589                          *   The latest iteration of the ABI requires that when
1590                          *   calling a function (at its global entry point),
1591                          *   the caller must ensure r12 holds the entry point
1592                          *   address (so that the function can quickly
1593                          *   establish addressability).
1594                          */
1595                         regs->gpr[12] = start;
1596                         /* Make sure that's restored on entry to userspace. */
1597                         set_thread_flag(TIF_RESTOREALL);
1598                 } else {
1599                         unsigned long toc;
1600
1601                         /* start is a relocated pointer to the function
1602                          * descriptor for the elf _start routine.  The first
1603                          * entry in the function descriptor is the entry
1604                          * address of _start and the second entry is the TOC
1605                          * value we need to use.
1606                          */
1607                         __get_user(entry, (unsigned long __user *)start);
1608                         __get_user(toc, (unsigned long __user *)start+1);
1609
1610                         /* Check whether the e_entry function descriptor entries
1611                          * need to be relocated before we can use them.
1612                          */
1613                         if (load_addr != 0) {
1614                                 entry += load_addr;
1615                                 toc   += load_addr;
1616                         }
1617                         regs->gpr[2] = toc;
1618                 }
1619                 regs->nip = entry;
1620                 regs->msr = MSR_USER64;
1621         } else {
1622                 regs->nip = start;
1623                 regs->gpr[2] = 0;
1624                 regs->msr = MSR_USER32;
1625         }
1626 #endif
1627 #ifdef CONFIG_VSX
1628         current->thread.used_vsr = 0;
1629 #endif
1630         memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1631         current->thread.fp_save_area = NULL;
1632 #ifdef CONFIG_ALTIVEC
1633         memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1634         current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1635         current->thread.vr_save_area = NULL;
1636         current->thread.vrsave = 0;
1637         current->thread.used_vr = 0;
1638 #endif /* CONFIG_ALTIVEC */
1639 #ifdef CONFIG_SPE
1640         memset(current->thread.evr, 0, sizeof(current->thread.evr));
1641         current->thread.acc = 0;
1642         current->thread.spefscr = 0;
1643         current->thread.used_spe = 0;
1644 #endif /* CONFIG_SPE */
1645 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1646         if (cpu_has_feature(CPU_FTR_TM))
1647                 regs->msr |= MSR_TM;
1648         current->thread.tm_tfhar = 0;
1649         current->thread.tm_texasr = 0;
1650         current->thread.tm_tfiar = 0;
1651 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1652 }
1653 EXPORT_SYMBOL(start_thread);
1654
1655 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1656                 | PR_FP_EXC_RES | PR_FP_EXC_INV)
1657
1658 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1659 {
1660         struct pt_regs *regs = tsk->thread.regs;
1661
1662         /* This is a bit hairy.  If we are an SPE enabled  processor
1663          * (have embedded fp) we store the IEEE exception enable flags in
1664          * fpexc_mode.  fpexc_mode is also used for setting FP exception
1665          * mode (asyn, precise, disabled) for 'Classic' FP. */
1666         if (val & PR_FP_EXC_SW_ENABLE) {
1667 #ifdef CONFIG_SPE
1668                 if (cpu_has_feature(CPU_FTR_SPE)) {
1669                         /*
1670                          * When the sticky exception bits are set
1671                          * directly by userspace, it must call prctl
1672                          * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1673                          * in the existing prctl settings) or
1674                          * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1675                          * the bits being set).  <fenv.h> functions
1676                          * saving and restoring the whole
1677                          * floating-point environment need to do so
1678                          * anyway to restore the prctl settings from
1679                          * the saved environment.
1680                          */
1681                         tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1682                         tsk->thread.fpexc_mode = val &
1683                                 (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1684                         return 0;
1685                 } else {
1686                         return -EINVAL;
1687                 }
1688 #else
1689                 return -EINVAL;
1690 #endif
1691         }
1692
1693         /* on a CONFIG_SPE this does not hurt us.  The bits that
1694          * __pack_fe01 use do not overlap with bits used for
1695          * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1696          * on CONFIG_SPE implementations are reserved so writing to
1697          * them does not change anything */
1698         if (val > PR_FP_EXC_PRECISE)
1699                 return -EINVAL;
1700         tsk->thread.fpexc_mode = __pack_fe01(val);
1701         if (regs != NULL && (regs->msr & MSR_FP) != 0)
1702                 regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1703                         | tsk->thread.fpexc_mode;
1704         return 0;
1705 }
1706
1707 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1708 {
1709         unsigned int val;
1710
1711         if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1712 #ifdef CONFIG_SPE
1713                 if (cpu_has_feature(CPU_FTR_SPE)) {
1714                         /*
1715                          * When the sticky exception bits are set
1716                          * directly by userspace, it must call prctl
1717                          * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1718                          * in the existing prctl settings) or
1719                          * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1720                          * the bits being set).  <fenv.h> functions
1721                          * saving and restoring the whole
1722                          * floating-point environment need to do so
1723                          * anyway to restore the prctl settings from
1724                          * the saved environment.
1725                          */
1726                         tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1727                         val = tsk->thread.fpexc_mode;
1728                 } else
1729                         return -EINVAL;
1730 #else
1731                 return -EINVAL;
1732 #endif
1733         else
1734                 val = __unpack_fe01(tsk->thread.fpexc_mode);
1735         return put_user(val, (unsigned int __user *) adr);
1736 }
1737
1738 int set_endian(struct task_struct *tsk, unsigned int val)
1739 {
1740         struct pt_regs *regs = tsk->thread.regs;
1741
1742         if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1743             (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1744                 return -EINVAL;
1745
1746         if (regs == NULL)
1747                 return -EINVAL;
1748
1749         if (val == PR_ENDIAN_BIG)
1750                 regs->msr &= ~MSR_LE;
1751         else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1752                 regs->msr |= MSR_LE;
1753         else
1754                 return -EINVAL;
1755
1756         return 0;
1757 }
1758
1759 int get_endian(struct task_struct *tsk, unsigned long adr)
1760 {
1761         struct pt_regs *regs = tsk->thread.regs;
1762         unsigned int val;
1763
1764         if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1765             !cpu_has_feature(CPU_FTR_REAL_LE))
1766                 return -EINVAL;
1767
1768         if (regs == NULL)
1769                 return -EINVAL;
1770
1771         if (regs->msr & MSR_LE) {
1772                 if (cpu_has_feature(CPU_FTR_REAL_LE))
1773                         val = PR_ENDIAN_LITTLE;
1774                 else
1775                         val = PR_ENDIAN_PPC_LITTLE;
1776         } else
1777                 val = PR_ENDIAN_BIG;
1778
1779         return put_user(val, (unsigned int __user *)adr);
1780 }
1781
1782 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1783 {
1784         tsk->thread.align_ctl = val;
1785         return 0;
1786 }
1787
1788 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1789 {
1790         return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1791 }
1792
1793 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1794                                   unsigned long nbytes)
1795 {
1796         unsigned long stack_page;
1797         unsigned long cpu = task_cpu(p);
1798
1799         /*
1800          * Avoid crashing if the stack has overflowed and corrupted
1801          * task_cpu(p), which is in the thread_info struct.
1802          */
1803         if (cpu < NR_CPUS && cpu_possible(cpu)) {
1804                 stack_page = (unsigned long) hardirq_ctx[cpu];
1805                 if (sp >= stack_page + sizeof(struct thread_struct)
1806                     && sp <= stack_page + THREAD_SIZE - nbytes)
1807                         return 1;
1808
1809                 stack_page = (unsigned long) softirq_ctx[cpu];
1810                 if (sp >= stack_page + sizeof(struct thread_struct)
1811                     && sp <= stack_page + THREAD_SIZE - nbytes)
1812                         return 1;
1813         }
1814         return 0;
1815 }
1816
1817 int validate_sp(unsigned long sp, struct task_struct *p,
1818                        unsigned long nbytes)
1819 {
1820         unsigned long stack_page = (unsigned long)task_stack_page(p);
1821
1822         if (sp >= stack_page + sizeof(struct thread_struct)
1823             && sp <= stack_page + THREAD_SIZE - nbytes)
1824                 return 1;
1825
1826         return valid_irq_stack(sp, p, nbytes);
1827 }
1828
1829 EXPORT_SYMBOL(validate_sp);
1830
1831 unsigned long get_wchan(struct task_struct *p)
1832 {
1833         unsigned long ip, sp;
1834         int count = 0;
1835
1836         if (!p || p == current || p->state == TASK_RUNNING)
1837                 return 0;
1838
1839         sp = p->thread.ksp;
1840         if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1841                 return 0;
1842
1843         do {
1844                 sp = *(unsigned long *)sp;
1845                 if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1846                         return 0;
1847                 if (count > 0) {
1848                         ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1849                         if (!in_sched_functions(ip))
1850                                 return ip;
1851                 }
1852         } while (count++ < 16);
1853         return 0;
1854 }
1855
1856 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1857
1858 void show_stack(struct task_struct *tsk, unsigned long *stack)
1859 {
1860         unsigned long sp, ip, lr, newsp;
1861         int count = 0;
1862         int firstframe = 1;
1863 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1864         int curr_frame = current->curr_ret_stack;
1865         extern void return_to_handler(void);
1866         unsigned long rth = (unsigned long)return_to_handler;
1867 #endif
1868
1869         sp = (unsigned long) stack;
1870         if (tsk == NULL)
1871                 tsk = current;
1872         if (sp == 0) {
1873                 if (tsk == current)
1874                         sp = current_stack_pointer();
1875                 else
1876                         sp = tsk->thread.ksp;
1877         }
1878
1879         lr = 0;
1880         printk("Call Trace:\n");
1881         do {
1882                 if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1883                         return;
1884
1885                 stack = (unsigned long *) sp;
1886                 newsp = stack[0];
1887                 ip = stack[STACK_FRAME_LR_SAVE];
1888                 if (!firstframe || ip != lr) {
1889                         printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1890 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1891                         if ((ip == rth) && curr_frame >= 0) {
1892                                 printk(" (%pS)",
1893                                        (void *)current->ret_stack[curr_frame].ret);
1894                                 curr_frame--;
1895                         }
1896 #endif
1897                         if (firstframe)
1898                                 printk(" (unreliable)");
1899                         printk("\n");
1900                 }
1901                 firstframe = 0;
1902
1903                 /*
1904                  * See if this is an exception frame.
1905                  * We look for the "regshere" marker in the current frame.
1906                  */
1907                 if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1908                     && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1909                         struct pt_regs *regs = (struct pt_regs *)
1910                                 (sp + STACK_FRAME_OVERHEAD);
1911                         lr = regs->link;
1912                         printk("--- interrupt: %lx at %pS\n    LR = %pS\n",
1913                                regs->trap, (void *)regs->nip, (void *)lr);
1914                         firstframe = 1;
1915                 }
1916
1917                 sp = newsp;
1918         } while (count++ < kstack_depth_to_print);
1919 }
1920
1921 #ifdef CONFIG_PPC64
1922 /* Called with hard IRQs off */
1923 void notrace __ppc64_runlatch_on(void)
1924 {
1925         struct thread_info *ti = current_thread_info();
1926         unsigned long ctrl;
1927
1928         ctrl = mfspr(SPRN_CTRLF);
1929         ctrl |= CTRL_RUNLATCH;
1930         mtspr(SPRN_CTRLT, ctrl);
1931
1932         ti->local_flags |= _TLF_RUNLATCH;
1933 }
1934
1935 /* Called with hard IRQs off */
1936 void notrace __ppc64_runlatch_off(void)
1937 {
1938         struct thread_info *ti = current_thread_info();
1939         unsigned long ctrl;
1940
1941         ti->local_flags &= ~_TLF_RUNLATCH;
1942
1943         ctrl = mfspr(SPRN_CTRLF);
1944         ctrl &= ~CTRL_RUNLATCH;
1945         mtspr(SPRN_CTRLT, ctrl);
1946 }
1947 #endif /* CONFIG_PPC64 */
1948
1949 unsigned long arch_align_stack(unsigned long sp)
1950 {
1951         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1952                 sp -= get_random_int() & ~PAGE_MASK;
1953         return sp & ~0xf;
1954 }
1955
1956 static inline unsigned long brk_rnd(void)
1957 {
1958         unsigned long rnd = 0;
1959
1960         /* 8MB for 32bit, 1GB for 64bit */
1961         if (is_32bit_task())
1962                 rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
1963         else
1964                 rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
1965
1966         return rnd << PAGE_SHIFT;
1967 }
1968
1969 unsigned long arch_randomize_brk(struct mm_struct *mm)
1970 {
1971         unsigned long base = mm->brk;
1972         unsigned long ret;
1973
1974 #ifdef CONFIG_PPC_STD_MMU_64
1975         /*
1976          * If we are using 1TB segments and we are allowed to randomise
1977          * the heap, we can put it above 1TB so it is backed by a 1TB
1978          * segment. Otherwise the heap will be in the bottom 1TB
1979          * which always uses 256MB segments and this may result in a
1980          * performance penalty. We don't need to worry about radix. For
1981          * radix, mmu_highuser_ssize remains unchanged from 256MB.
1982          */
1983         if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1984                 base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1985 #endif
1986
1987         ret = PAGE_ALIGN(base + brk_rnd());
1988
1989         if (ret < mm->brk)
1990                 return mm->brk;
1991
1992         return ret;
1993 }
1994