xref: /openbmc/linux/arch/powerpc/kernel/process.c (revision 77d84ff8)
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/tm.h>
54 #include <asm/debug.h>
55 #ifdef CONFIG_PPC64
56 #include <asm/firmware.h>
57 #endif
58 #include <linux/kprobes.h>
59 #include <linux/kdebug.h>
60 
61 /* Transactional Memory debug */
62 #ifdef TM_DEBUG_SW
63 #define TM_DEBUG(x...) printk(KERN_INFO x)
64 #else
65 #define TM_DEBUG(x...) do { } while(0)
66 #endif
67 
68 extern unsigned long _get_SP(void);
69 
70 #ifndef CONFIG_SMP
71 struct task_struct *last_task_used_math = NULL;
72 struct task_struct *last_task_used_altivec = NULL;
73 struct task_struct *last_task_used_vsx = NULL;
74 struct task_struct *last_task_used_spe = NULL;
75 #endif
76 
77 #ifdef CONFIG_PPC_FPU
78 /*
79  * Make sure the floating-point register state in the
80  * the thread_struct is up to date for task tsk.
81  */
82 void flush_fp_to_thread(struct task_struct *tsk)
83 {
84 	if (tsk->thread.regs) {
85 		/*
86 		 * We need to disable preemption here because if we didn't,
87 		 * another process could get scheduled after the regs->msr
88 		 * test but before we have finished saving the FP registers
89 		 * to the thread_struct.  That process could take over the
90 		 * FPU, and then when we get scheduled again we would store
91 		 * bogus values for the remaining FP registers.
92 		 */
93 		preempt_disable();
94 		if (tsk->thread.regs->msr & MSR_FP) {
95 #ifdef CONFIG_SMP
96 			/*
97 			 * This should only ever be called for current or
98 			 * for a stopped child process.  Since we save away
99 			 * the FP register state on context switch on SMP,
100 			 * there is something wrong if a stopped child appears
101 			 * to still have its FP state in the CPU registers.
102 			 */
103 			BUG_ON(tsk != current);
104 #endif
105 			giveup_fpu(tsk);
106 		}
107 		preempt_enable();
108 	}
109 }
110 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
111 #endif
112 
113 void enable_kernel_fp(void)
114 {
115 	WARN_ON(preemptible());
116 
117 #ifdef CONFIG_SMP
118 	if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
119 		giveup_fpu(current);
120 	else
121 		giveup_fpu(NULL);	/* just enables FP for kernel */
122 #else
123 	giveup_fpu(last_task_used_math);
124 #endif /* CONFIG_SMP */
125 }
126 EXPORT_SYMBOL(enable_kernel_fp);
127 
128 #ifdef CONFIG_ALTIVEC
129 void enable_kernel_altivec(void)
130 {
131 	WARN_ON(preemptible());
132 
133 #ifdef CONFIG_SMP
134 	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
135 		giveup_altivec(current);
136 	else
137 		giveup_altivec_notask();
138 #else
139 	giveup_altivec(last_task_used_altivec);
140 #endif /* CONFIG_SMP */
141 }
142 EXPORT_SYMBOL(enable_kernel_altivec);
143 
144 /*
145  * Make sure the VMX/Altivec register state in the
146  * the thread_struct is up to date for task tsk.
147  */
148 void flush_altivec_to_thread(struct task_struct *tsk)
149 {
150 	if (tsk->thread.regs) {
151 		preempt_disable();
152 		if (tsk->thread.regs->msr & MSR_VEC) {
153 #ifdef CONFIG_SMP
154 			BUG_ON(tsk != current);
155 #endif
156 			giveup_altivec(tsk);
157 		}
158 		preempt_enable();
159 	}
160 }
161 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
162 #endif /* CONFIG_ALTIVEC */
163 
164 #ifdef CONFIG_VSX
165 #if 0
166 /* not currently used, but some crazy RAID module might want to later */
167 void enable_kernel_vsx(void)
168 {
169 	WARN_ON(preemptible());
170 
171 #ifdef CONFIG_SMP
172 	if (current->thread.regs && (current->thread.regs->msr & MSR_VSX))
173 		giveup_vsx(current);
174 	else
175 		giveup_vsx(NULL);	/* just enable vsx for kernel - force */
176 #else
177 	giveup_vsx(last_task_used_vsx);
178 #endif /* CONFIG_SMP */
179 }
180 EXPORT_SYMBOL(enable_kernel_vsx);
181 #endif
182 
183 void giveup_vsx(struct task_struct *tsk)
184 {
185 	giveup_fpu(tsk);
186 	giveup_altivec(tsk);
187 	__giveup_vsx(tsk);
188 }
189 
190 void flush_vsx_to_thread(struct task_struct *tsk)
191 {
192 	if (tsk->thread.regs) {
193 		preempt_disable();
194 		if (tsk->thread.regs->msr & MSR_VSX) {
195 #ifdef CONFIG_SMP
196 			BUG_ON(tsk != current);
197 #endif
198 			giveup_vsx(tsk);
199 		}
200 		preempt_enable();
201 	}
202 }
203 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
204 #endif /* CONFIG_VSX */
205 
206 #ifdef CONFIG_SPE
207 
208 void enable_kernel_spe(void)
209 {
210 	WARN_ON(preemptible());
211 
212 #ifdef CONFIG_SMP
213 	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
214 		giveup_spe(current);
215 	else
216 		giveup_spe(NULL);	/* just enable SPE for kernel - force */
217 #else
218 	giveup_spe(last_task_used_spe);
219 #endif /* __SMP __ */
220 }
221 EXPORT_SYMBOL(enable_kernel_spe);
222 
223 void flush_spe_to_thread(struct task_struct *tsk)
224 {
225 	if (tsk->thread.regs) {
226 		preempt_disable();
227 		if (tsk->thread.regs->msr & MSR_SPE) {
228 #ifdef CONFIG_SMP
229 			BUG_ON(tsk != current);
230 #endif
231 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
232 			giveup_spe(tsk);
233 		}
234 		preempt_enable();
235 	}
236 }
237 #endif /* CONFIG_SPE */
238 
239 #ifndef CONFIG_SMP
240 /*
241  * If we are doing lazy switching of CPU state (FP, altivec or SPE),
242  * and the current task has some state, discard it.
243  */
244 void discard_lazy_cpu_state(void)
245 {
246 	preempt_disable();
247 	if (last_task_used_math == current)
248 		last_task_used_math = NULL;
249 #ifdef CONFIG_ALTIVEC
250 	if (last_task_used_altivec == current)
251 		last_task_used_altivec = NULL;
252 #endif /* CONFIG_ALTIVEC */
253 #ifdef CONFIG_VSX
254 	if (last_task_used_vsx == current)
255 		last_task_used_vsx = NULL;
256 #endif /* CONFIG_VSX */
257 #ifdef CONFIG_SPE
258 	if (last_task_used_spe == current)
259 		last_task_used_spe = NULL;
260 #endif
261 	preempt_enable();
262 }
263 #endif /* CONFIG_SMP */
264 
265 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
266 void do_send_trap(struct pt_regs *regs, unsigned long address,
267 		  unsigned long error_code, int signal_code, int breakpt)
268 {
269 	siginfo_t info;
270 
271 	current->thread.trap_nr = signal_code;
272 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
273 			11, SIGSEGV) == NOTIFY_STOP)
274 		return;
275 
276 	/* Deliver the signal to userspace */
277 	info.si_signo = SIGTRAP;
278 	info.si_errno = breakpt;	/* breakpoint or watchpoint id */
279 	info.si_code = signal_code;
280 	info.si_addr = (void __user *)address;
281 	force_sig_info(SIGTRAP, &info, current);
282 }
283 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
284 void do_break (struct pt_regs *regs, unsigned long address,
285 		    unsigned long error_code)
286 {
287 	siginfo_t info;
288 
289 	current->thread.trap_nr = TRAP_HWBKPT;
290 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
291 			11, SIGSEGV) == NOTIFY_STOP)
292 		return;
293 
294 	if (debugger_break_match(regs))
295 		return;
296 
297 	/* Clear the breakpoint */
298 	hw_breakpoint_disable();
299 
300 	/* Deliver the signal to userspace */
301 	info.si_signo = SIGTRAP;
302 	info.si_errno = 0;
303 	info.si_code = TRAP_HWBKPT;
304 	info.si_addr = (void __user *)address;
305 	force_sig_info(SIGTRAP, &info, current);
306 }
307 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
308 
309 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
310 
311 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
312 /*
313  * Set the debug registers back to their default "safe" values.
314  */
315 static void set_debug_reg_defaults(struct thread_struct *thread)
316 {
317 	thread->debug.iac1 = thread->debug.iac2 = 0;
318 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
319 	thread->debug.iac3 = thread->debug.iac4 = 0;
320 #endif
321 	thread->debug.dac1 = thread->debug.dac2 = 0;
322 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
323 	thread->debug.dvc1 = thread->debug.dvc2 = 0;
324 #endif
325 	thread->debug.dbcr0 = 0;
326 #ifdef CONFIG_BOOKE
327 	/*
328 	 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
329 	 */
330 	thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
331 			DBCR1_IAC3US | DBCR1_IAC4US;
332 	/*
333 	 * Force Data Address Compare User/Supervisor bits to be User-only
334 	 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
335 	 */
336 	thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
337 #else
338 	thread->debug.dbcr1 = 0;
339 #endif
340 }
341 
342 static void prime_debug_regs(struct thread_struct *thread)
343 {
344 	/*
345 	 * We could have inherited MSR_DE from userspace, since
346 	 * it doesn't get cleared on exception entry.  Make sure
347 	 * MSR_DE is clear before we enable any debug events.
348 	 */
349 	mtmsr(mfmsr() & ~MSR_DE);
350 
351 	mtspr(SPRN_IAC1, thread->debug.iac1);
352 	mtspr(SPRN_IAC2, thread->debug.iac2);
353 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
354 	mtspr(SPRN_IAC3, thread->debug.iac3);
355 	mtspr(SPRN_IAC4, thread->debug.iac4);
356 #endif
357 	mtspr(SPRN_DAC1, thread->debug.dac1);
358 	mtspr(SPRN_DAC2, thread->debug.dac2);
359 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
360 	mtspr(SPRN_DVC1, thread->debug.dvc1);
361 	mtspr(SPRN_DVC2, thread->debug.dvc2);
362 #endif
363 	mtspr(SPRN_DBCR0, thread->debug.dbcr0);
364 	mtspr(SPRN_DBCR1, thread->debug.dbcr1);
365 #ifdef CONFIG_BOOKE
366 	mtspr(SPRN_DBCR2, thread->debug.dbcr2);
367 #endif
368 }
369 /*
370  * Unless neither the old or new thread are making use of the
371  * debug registers, set the debug registers from the values
372  * stored in the new thread.
373  */
374 void switch_booke_debug_regs(struct thread_struct *new_thread)
375 {
376 	if ((current->thread.debug.dbcr0 & DBCR0_IDM)
377 		|| (new_thread->debug.dbcr0 & DBCR0_IDM))
378 			prime_debug_regs(new_thread);
379 }
380 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
381 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
382 #ifndef CONFIG_HAVE_HW_BREAKPOINT
383 static void set_debug_reg_defaults(struct thread_struct *thread)
384 {
385 	thread->hw_brk.address = 0;
386 	thread->hw_brk.type = 0;
387 	set_breakpoint(&thread->hw_brk);
388 }
389 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
390 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
391 
392 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
393 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
394 {
395 	mtspr(SPRN_DAC1, dabr);
396 #ifdef CONFIG_PPC_47x
397 	isync();
398 #endif
399 	return 0;
400 }
401 #elif defined(CONFIG_PPC_BOOK3S)
402 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
403 {
404 	mtspr(SPRN_DABR, dabr);
405 	if (cpu_has_feature(CPU_FTR_DABRX))
406 		mtspr(SPRN_DABRX, dabrx);
407 	return 0;
408 }
409 #else
410 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
411 {
412 	return -EINVAL;
413 }
414 #endif
415 
416 static inline int set_dabr(struct arch_hw_breakpoint *brk)
417 {
418 	unsigned long dabr, dabrx;
419 
420 	dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
421 	dabrx = ((brk->type >> 3) & 0x7);
422 
423 	if (ppc_md.set_dabr)
424 		return ppc_md.set_dabr(dabr, dabrx);
425 
426 	return __set_dabr(dabr, dabrx);
427 }
428 
429 static inline int set_dawr(struct arch_hw_breakpoint *brk)
430 {
431 	unsigned long dawr, dawrx, mrd;
432 
433 	dawr = brk->address;
434 
435 	dawrx  = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
436 		                   << (63 - 58); //* read/write bits */
437 	dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
438 		                   << (63 - 59); //* translate */
439 	dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
440 		                   >> 3; //* PRIM bits */
441 	/* dawr length is stored in field MDR bits 48:53.  Matches range in
442 	   doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
443 	   0b111111=64DW.
444 	   brk->len is in bytes.
445 	   This aligns up to double word size, shifts and does the bias.
446 	*/
447 	mrd = ((brk->len + 7) >> 3) - 1;
448 	dawrx |= (mrd & 0x3f) << (63 - 53);
449 
450 	if (ppc_md.set_dawr)
451 		return ppc_md.set_dawr(dawr, dawrx);
452 	mtspr(SPRN_DAWR, dawr);
453 	mtspr(SPRN_DAWRX, dawrx);
454 	return 0;
455 }
456 
457 int set_breakpoint(struct arch_hw_breakpoint *brk)
458 {
459 	__get_cpu_var(current_brk) = *brk;
460 
461 	if (cpu_has_feature(CPU_FTR_DAWR))
462 		return set_dawr(brk);
463 
464 	return set_dabr(brk);
465 }
466 
467 #ifdef CONFIG_PPC64
468 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
469 #endif
470 
471 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
472 			      struct arch_hw_breakpoint *b)
473 {
474 	if (a->address != b->address)
475 		return false;
476 	if (a->type != b->type)
477 		return false;
478 	if (a->len != b->len)
479 		return false;
480 	return true;
481 }
482 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
483 static inline void tm_reclaim_task(struct task_struct *tsk)
484 {
485 	/* We have to work out if we're switching from/to a task that's in the
486 	 * middle of a transaction.
487 	 *
488 	 * In switching we need to maintain a 2nd register state as
489 	 * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
490 	 * checkpointed (tbegin) state in ckpt_regs and saves the transactional
491 	 * (current) FPRs into oldtask->thread.transact_fpr[].
492 	 *
493 	 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
494 	 */
495 	struct thread_struct *thr = &tsk->thread;
496 
497 	if (!thr->regs)
498 		return;
499 
500 	if (!MSR_TM_ACTIVE(thr->regs->msr))
501 		goto out_and_saveregs;
502 
503 	/* Stash the original thread MSR, as giveup_fpu et al will
504 	 * modify it.  We hold onto it to see whether the task used
505 	 * FP & vector regs.
506 	 */
507 	thr->tm_orig_msr = thr->regs->msr;
508 
509 	TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
510 		 "ccr=%lx, msr=%lx, trap=%lx)\n",
511 		 tsk->pid, thr->regs->nip,
512 		 thr->regs->ccr, thr->regs->msr,
513 		 thr->regs->trap);
514 
515 	tm_reclaim(thr, thr->regs->msr, TM_CAUSE_RESCHED);
516 
517 	TM_DEBUG("--- tm_reclaim on pid %d complete\n",
518 		 tsk->pid);
519 
520 out_and_saveregs:
521 	/* Always save the regs here, even if a transaction's not active.
522 	 * This context-switches a thread's TM info SPRs.  We do it here to
523 	 * be consistent with the restore path (in recheckpoint) which
524 	 * cannot happen later in _switch().
525 	 */
526 	tm_save_sprs(thr);
527 }
528 
529 static inline void tm_recheckpoint_new_task(struct task_struct *new)
530 {
531 	unsigned long msr;
532 
533 	if (!cpu_has_feature(CPU_FTR_TM))
534 		return;
535 
536 	/* Recheckpoint the registers of the thread we're about to switch to.
537 	 *
538 	 * If the task was using FP, we non-lazily reload both the original and
539 	 * the speculative FP register states.  This is because the kernel
540 	 * doesn't see if/when a TM rollback occurs, so if we take an FP
541 	 * unavoidable later, we are unable to determine which set of FP regs
542 	 * need to be restored.
543 	 */
544 	if (!new->thread.regs)
545 		return;
546 
547 	/* The TM SPRs are restored here, so that TEXASR.FS can be set
548 	 * before the trecheckpoint and no explosion occurs.
549 	 */
550 	tm_restore_sprs(&new->thread);
551 
552 	if (!MSR_TM_ACTIVE(new->thread.regs->msr))
553 		return;
554 	msr = new->thread.tm_orig_msr;
555 	/* Recheckpoint to restore original checkpointed register state. */
556 	TM_DEBUG("*** tm_recheckpoint of pid %d "
557 		 "(new->msr 0x%lx, new->origmsr 0x%lx)\n",
558 		 new->pid, new->thread.regs->msr, msr);
559 
560 	/* This loads the checkpointed FP/VEC state, if used */
561 	tm_recheckpoint(&new->thread, msr);
562 
563 	/* This loads the speculative FP/VEC state, if used */
564 	if (msr & MSR_FP) {
565 		do_load_up_transact_fpu(&new->thread);
566 		new->thread.regs->msr |=
567 			(MSR_FP | new->thread.fpexc_mode);
568 	}
569 #ifdef CONFIG_ALTIVEC
570 	if (msr & MSR_VEC) {
571 		do_load_up_transact_altivec(&new->thread);
572 		new->thread.regs->msr |= MSR_VEC;
573 	}
574 #endif
575 	/* We may as well turn on VSX too since all the state is restored now */
576 	if (msr & MSR_VSX)
577 		new->thread.regs->msr |= MSR_VSX;
578 
579 	TM_DEBUG("*** tm_recheckpoint of pid %d complete "
580 		 "(kernel msr 0x%lx)\n",
581 		 new->pid, mfmsr());
582 }
583 
584 static inline void __switch_to_tm(struct task_struct *prev)
585 {
586 	if (cpu_has_feature(CPU_FTR_TM)) {
587 		tm_enable();
588 		tm_reclaim_task(prev);
589 	}
590 }
591 #else
592 #define tm_recheckpoint_new_task(new)
593 #define __switch_to_tm(prev)
594 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
595 
596 struct task_struct *__switch_to(struct task_struct *prev,
597 	struct task_struct *new)
598 {
599 	struct thread_struct *new_thread, *old_thread;
600 	struct task_struct *last;
601 #ifdef CONFIG_PPC_BOOK3S_64
602 	struct ppc64_tlb_batch *batch;
603 #endif
604 
605 	WARN_ON(!irqs_disabled());
606 
607 	/* Back up the TAR across context switches.
608 	 * Note that the TAR is not available for use in the kernel.  (To
609 	 * provide this, the TAR should be backed up/restored on exception
610 	 * entry/exit instead, and be in pt_regs.  FIXME, this should be in
611 	 * pt_regs anyway (for debug).)
612 	 * Save the TAR here before we do treclaim/trecheckpoint as these
613 	 * will change the TAR.
614 	 */
615 	save_tar(&prev->thread);
616 
617 	__switch_to_tm(prev);
618 
619 #ifdef CONFIG_SMP
620 	/* avoid complexity of lazy save/restore of fpu
621 	 * by just saving it every time we switch out if
622 	 * this task used the fpu during the last quantum.
623 	 *
624 	 * If it tries to use the fpu again, it'll trap and
625 	 * reload its fp regs.  So we don't have to do a restore
626 	 * every switch, just a save.
627 	 *  -- Cort
628 	 */
629 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
630 		giveup_fpu(prev);
631 #ifdef CONFIG_ALTIVEC
632 	/*
633 	 * If the previous thread used altivec in the last quantum
634 	 * (thus changing altivec regs) then save them.
635 	 * We used to check the VRSAVE register but not all apps
636 	 * set it, so we don't rely on it now (and in fact we need
637 	 * to save & restore VSCR even if VRSAVE == 0).  -- paulus
638 	 *
639 	 * On SMP we always save/restore altivec regs just to avoid the
640 	 * complexity of changing processors.
641 	 *  -- Cort
642 	 */
643 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
644 		giveup_altivec(prev);
645 #endif /* CONFIG_ALTIVEC */
646 #ifdef CONFIG_VSX
647 	if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX))
648 		/* VMX and FPU registers are already save here */
649 		__giveup_vsx(prev);
650 #endif /* CONFIG_VSX */
651 #ifdef CONFIG_SPE
652 	/*
653 	 * If the previous thread used spe in the last quantum
654 	 * (thus changing spe regs) then save them.
655 	 *
656 	 * On SMP we always save/restore spe regs just to avoid the
657 	 * complexity of changing processors.
658 	 */
659 	if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
660 		giveup_spe(prev);
661 #endif /* CONFIG_SPE */
662 
663 #else  /* CONFIG_SMP */
664 #ifdef CONFIG_ALTIVEC
665 	/* Avoid the trap.  On smp this this never happens since
666 	 * we don't set last_task_used_altivec -- Cort
667 	 */
668 	if (new->thread.regs && last_task_used_altivec == new)
669 		new->thread.regs->msr |= MSR_VEC;
670 #endif /* CONFIG_ALTIVEC */
671 #ifdef CONFIG_VSX
672 	if (new->thread.regs && last_task_used_vsx == new)
673 		new->thread.regs->msr |= MSR_VSX;
674 #endif /* CONFIG_VSX */
675 #ifdef CONFIG_SPE
676 	/* Avoid the trap.  On smp this this never happens since
677 	 * we don't set last_task_used_spe
678 	 */
679 	if (new->thread.regs && last_task_used_spe == new)
680 		new->thread.regs->msr |= MSR_SPE;
681 #endif /* CONFIG_SPE */
682 
683 #endif /* CONFIG_SMP */
684 
685 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
686 	switch_booke_debug_regs(&new->thread);
687 #else
688 /*
689  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
690  * schedule DABR
691  */
692 #ifndef CONFIG_HAVE_HW_BREAKPOINT
693 	if (unlikely(hw_brk_match(&__get_cpu_var(current_brk), &new->thread.hw_brk)))
694 		set_breakpoint(&new->thread.hw_brk);
695 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
696 #endif
697 
698 
699 	new_thread = &new->thread;
700 	old_thread = &current->thread;
701 
702 #ifdef CONFIG_PPC64
703 	/*
704 	 * Collect processor utilization data per process
705 	 */
706 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
707 		struct cpu_usage *cu = &__get_cpu_var(cpu_usage_array);
708 		long unsigned start_tb, current_tb;
709 		start_tb = old_thread->start_tb;
710 		cu->current_tb = current_tb = mfspr(SPRN_PURR);
711 		old_thread->accum_tb += (current_tb - start_tb);
712 		new_thread->start_tb = current_tb;
713 	}
714 #endif /* CONFIG_PPC64 */
715 
716 #ifdef CONFIG_PPC_BOOK3S_64
717 	batch = &__get_cpu_var(ppc64_tlb_batch);
718 	if (batch->active) {
719 		current_thread_info()->local_flags |= _TLF_LAZY_MMU;
720 		if (batch->index)
721 			__flush_tlb_pending(batch);
722 		batch->active = 0;
723 	}
724 #endif /* CONFIG_PPC_BOOK3S_64 */
725 
726 	/*
727 	 * We can't take a PMU exception inside _switch() since there is a
728 	 * window where the kernel stack SLB and the kernel stack are out
729 	 * of sync. Hard disable here.
730 	 */
731 	hard_irq_disable();
732 
733 	tm_recheckpoint_new_task(new);
734 
735 	last = _switch(old_thread, new_thread);
736 
737 #ifdef CONFIG_PPC_BOOK3S_64
738 	if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
739 		current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
740 		batch = &__get_cpu_var(ppc64_tlb_batch);
741 		batch->active = 1;
742 	}
743 #endif /* CONFIG_PPC_BOOK3S_64 */
744 
745 	return last;
746 }
747 
748 static int instructions_to_print = 16;
749 
750 static void show_instructions(struct pt_regs *regs)
751 {
752 	int i;
753 	unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
754 			sizeof(int));
755 
756 	printk("Instruction dump:");
757 
758 	for (i = 0; i < instructions_to_print; i++) {
759 		int instr;
760 
761 		if (!(i % 8))
762 			printk("\n");
763 
764 #if !defined(CONFIG_BOOKE)
765 		/* If executing with the IMMU off, adjust pc rather
766 		 * than print XXXXXXXX.
767 		 */
768 		if (!(regs->msr & MSR_IR))
769 			pc = (unsigned long)phys_to_virt(pc);
770 #endif
771 
772 		/* We use __get_user here *only* to avoid an OOPS on a
773 		 * bad address because the pc *should* only be a
774 		 * kernel address.
775 		 */
776 		if (!__kernel_text_address(pc) ||
777 		     __get_user(instr, (unsigned int __user *)pc)) {
778 			printk(KERN_CONT "XXXXXXXX ");
779 		} else {
780 			if (regs->nip == pc)
781 				printk(KERN_CONT "<%08x> ", instr);
782 			else
783 				printk(KERN_CONT "%08x ", instr);
784 		}
785 
786 		pc += sizeof(int);
787 	}
788 
789 	printk("\n");
790 }
791 
792 static struct regbit {
793 	unsigned long bit;
794 	const char *name;
795 } msr_bits[] = {
796 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
797 	{MSR_SF,	"SF"},
798 	{MSR_HV,	"HV"},
799 #endif
800 	{MSR_VEC,	"VEC"},
801 	{MSR_VSX,	"VSX"},
802 #ifdef CONFIG_BOOKE
803 	{MSR_CE,	"CE"},
804 #endif
805 	{MSR_EE,	"EE"},
806 	{MSR_PR,	"PR"},
807 	{MSR_FP,	"FP"},
808 	{MSR_ME,	"ME"},
809 #ifdef CONFIG_BOOKE
810 	{MSR_DE,	"DE"},
811 #else
812 	{MSR_SE,	"SE"},
813 	{MSR_BE,	"BE"},
814 #endif
815 	{MSR_IR,	"IR"},
816 	{MSR_DR,	"DR"},
817 	{MSR_PMM,	"PMM"},
818 #ifndef CONFIG_BOOKE
819 	{MSR_RI,	"RI"},
820 	{MSR_LE,	"LE"},
821 #endif
822 	{0,		NULL}
823 };
824 
825 static void printbits(unsigned long val, struct regbit *bits)
826 {
827 	const char *sep = "";
828 
829 	printk("<");
830 	for (; bits->bit; ++bits)
831 		if (val & bits->bit) {
832 			printk("%s%s", sep, bits->name);
833 			sep = ",";
834 		}
835 	printk(">");
836 }
837 
838 #ifdef CONFIG_PPC64
839 #define REG		"%016lx"
840 #define REGS_PER_LINE	4
841 #define LAST_VOLATILE	13
842 #else
843 #define REG		"%08lx"
844 #define REGS_PER_LINE	8
845 #define LAST_VOLATILE	12
846 #endif
847 
848 void show_regs(struct pt_regs * regs)
849 {
850 	int i, trap;
851 
852 	show_regs_print_info(KERN_DEFAULT);
853 
854 	printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
855 	       regs->nip, regs->link, regs->ctr);
856 	printk("REGS: %p TRAP: %04lx   %s  (%s)\n",
857 	       regs, regs->trap, print_tainted(), init_utsname()->release);
858 	printk("MSR: "REG" ", regs->msr);
859 	printbits(regs->msr, msr_bits);
860 	printk("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
861 	trap = TRAP(regs);
862 	if ((regs->trap != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
863 		printk("CFAR: "REG" ", regs->orig_gpr3);
864 	if (trap == 0x200 || trap == 0x300 || trap == 0x600)
865 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
866 		printk("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
867 #else
868 		printk("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
869 #endif
870 #ifdef CONFIG_PPC64
871 	printk("SOFTE: %ld ", regs->softe);
872 #endif
873 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
874 	if (MSR_TM_ACTIVE(regs->msr))
875 		printk("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
876 #endif
877 
878 	for (i = 0;  i < 32;  i++) {
879 		if ((i % REGS_PER_LINE) == 0)
880 			printk("\nGPR%02d: ", i);
881 		printk(REG " ", regs->gpr[i]);
882 		if (i == LAST_VOLATILE && !FULL_REGS(regs))
883 			break;
884 	}
885 	printk("\n");
886 #ifdef CONFIG_KALLSYMS
887 	/*
888 	 * Lookup NIP late so we have the best change of getting the
889 	 * above info out without failing
890 	 */
891 	printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
892 	printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
893 #endif
894 	show_stack(current, (unsigned long *) regs->gpr[1]);
895 	if (!user_mode(regs))
896 		show_instructions(regs);
897 }
898 
899 void exit_thread(void)
900 {
901 	discard_lazy_cpu_state();
902 }
903 
904 void flush_thread(void)
905 {
906 	discard_lazy_cpu_state();
907 
908 #ifdef CONFIG_HAVE_HW_BREAKPOINT
909 	flush_ptrace_hw_breakpoint(current);
910 #else /* CONFIG_HAVE_HW_BREAKPOINT */
911 	set_debug_reg_defaults(&current->thread);
912 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
913 }
914 
915 void
916 release_thread(struct task_struct *t)
917 {
918 }
919 
920 /*
921  * this gets called so that we can store coprocessor state into memory and
922  * copy the current task into the new thread.
923  */
924 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
925 {
926 	flush_fp_to_thread(src);
927 	flush_altivec_to_thread(src);
928 	flush_vsx_to_thread(src);
929 	flush_spe_to_thread(src);
930 
931 	*dst = *src;
932 
933 	clear_task_ebb(dst);
934 
935 	return 0;
936 }
937 
938 /*
939  * Copy a thread..
940  */
941 extern unsigned long dscr_default; /* defined in arch/powerpc/kernel/sysfs.c */
942 
943 int copy_thread(unsigned long clone_flags, unsigned long usp,
944 		unsigned long arg, struct task_struct *p)
945 {
946 	struct pt_regs *childregs, *kregs;
947 	extern void ret_from_fork(void);
948 	extern void ret_from_kernel_thread(void);
949 	void (*f)(void);
950 	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
951 
952 	/* Copy registers */
953 	sp -= sizeof(struct pt_regs);
954 	childregs = (struct pt_regs *) sp;
955 	if (unlikely(p->flags & PF_KTHREAD)) {
956 		struct thread_info *ti = (void *)task_stack_page(p);
957 		memset(childregs, 0, sizeof(struct pt_regs));
958 		childregs->gpr[1] = sp + sizeof(struct pt_regs);
959 		childregs->gpr[14] = usp;	/* function */
960 #ifdef CONFIG_PPC64
961 		clear_tsk_thread_flag(p, TIF_32BIT);
962 		childregs->softe = 1;
963 #endif
964 		childregs->gpr[15] = arg;
965 		p->thread.regs = NULL;	/* no user register state */
966 		ti->flags |= _TIF_RESTOREALL;
967 		f = ret_from_kernel_thread;
968 	} else {
969 		struct pt_regs *regs = current_pt_regs();
970 		CHECK_FULL_REGS(regs);
971 		*childregs = *regs;
972 		if (usp)
973 			childregs->gpr[1] = usp;
974 		p->thread.regs = childregs;
975 		childregs->gpr[3] = 0;  /* Result from fork() */
976 		if (clone_flags & CLONE_SETTLS) {
977 #ifdef CONFIG_PPC64
978 			if (!is_32bit_task())
979 				childregs->gpr[13] = childregs->gpr[6];
980 			else
981 #endif
982 				childregs->gpr[2] = childregs->gpr[6];
983 		}
984 
985 		f = ret_from_fork;
986 	}
987 	sp -= STACK_FRAME_OVERHEAD;
988 
989 	/*
990 	 * The way this works is that at some point in the future
991 	 * some task will call _switch to switch to the new task.
992 	 * That will pop off the stack frame created below and start
993 	 * the new task running at ret_from_fork.  The new task will
994 	 * do some house keeping and then return from the fork or clone
995 	 * system call, using the stack frame created above.
996 	 */
997 	((unsigned long *)sp)[0] = 0;
998 	sp -= sizeof(struct pt_regs);
999 	kregs = (struct pt_regs *) sp;
1000 	sp -= STACK_FRAME_OVERHEAD;
1001 	p->thread.ksp = sp;
1002 #ifdef CONFIG_PPC32
1003 	p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1004 				_ALIGN_UP(sizeof(struct thread_info), 16);
1005 #endif
1006 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1007 	p->thread.ptrace_bps[0] = NULL;
1008 #endif
1009 
1010 	p->thread.fp_save_area = NULL;
1011 #ifdef CONFIG_ALTIVEC
1012 	p->thread.vr_save_area = NULL;
1013 #endif
1014 
1015 #ifdef CONFIG_PPC_STD_MMU_64
1016 	if (mmu_has_feature(MMU_FTR_SLB)) {
1017 		unsigned long sp_vsid;
1018 		unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1019 
1020 		if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1021 			sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1022 				<< SLB_VSID_SHIFT_1T;
1023 		else
1024 			sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1025 				<< SLB_VSID_SHIFT;
1026 		sp_vsid |= SLB_VSID_KERNEL | llp;
1027 		p->thread.ksp_vsid = sp_vsid;
1028 	}
1029 #endif /* CONFIG_PPC_STD_MMU_64 */
1030 #ifdef CONFIG_PPC64
1031 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1032 		p->thread.dscr_inherit = current->thread.dscr_inherit;
1033 		p->thread.dscr = current->thread.dscr;
1034 	}
1035 	if (cpu_has_feature(CPU_FTR_HAS_PPR))
1036 		p->thread.ppr = INIT_PPR;
1037 #endif
1038 	/*
1039 	 * The PPC64 ABI makes use of a TOC to contain function
1040 	 * pointers.  The function (ret_from_except) is actually a pointer
1041 	 * to the TOC entry.  The first entry is a pointer to the actual
1042 	 * function.
1043 	 */
1044 #ifdef CONFIG_PPC64
1045 	kregs->nip = *((unsigned long *)f);
1046 #else
1047 	kregs->nip = (unsigned long)f;
1048 #endif
1049 	return 0;
1050 }
1051 
1052 /*
1053  * Set up a thread for executing a new program
1054  */
1055 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1056 {
1057 #ifdef CONFIG_PPC64
1058 	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
1059 #endif
1060 
1061 	/*
1062 	 * If we exec out of a kernel thread then thread.regs will not be
1063 	 * set.  Do it now.
1064 	 */
1065 	if (!current->thread.regs) {
1066 		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1067 		current->thread.regs = regs - 1;
1068 	}
1069 
1070 	memset(regs->gpr, 0, sizeof(regs->gpr));
1071 	regs->ctr = 0;
1072 	regs->link = 0;
1073 	regs->xer = 0;
1074 	regs->ccr = 0;
1075 	regs->gpr[1] = sp;
1076 
1077 	/*
1078 	 * We have just cleared all the nonvolatile GPRs, so make
1079 	 * FULL_REGS(regs) return true.  This is necessary to allow
1080 	 * ptrace to examine the thread immediately after exec.
1081 	 */
1082 	regs->trap &= ~1UL;
1083 
1084 #ifdef CONFIG_PPC32
1085 	regs->mq = 0;
1086 	regs->nip = start;
1087 	regs->msr = MSR_USER;
1088 #else
1089 	if (!is_32bit_task()) {
1090 		unsigned long entry;
1091 
1092 		if (is_elf2_task()) {
1093 			/* Look ma, no function descriptors! */
1094 			entry = start;
1095 
1096 			/*
1097 			 * Ulrich says:
1098 			 *   The latest iteration of the ABI requires that when
1099 			 *   calling a function (at its global entry point),
1100 			 *   the caller must ensure r12 holds the entry point
1101 			 *   address (so that the function can quickly
1102 			 *   establish addressability).
1103 			 */
1104 			regs->gpr[12] = start;
1105 			/* Make sure that's restored on entry to userspace. */
1106 			set_thread_flag(TIF_RESTOREALL);
1107 		} else {
1108 			unsigned long toc;
1109 
1110 			/* start is a relocated pointer to the function
1111 			 * descriptor for the elf _start routine.  The first
1112 			 * entry in the function descriptor is the entry
1113 			 * address of _start and the second entry is the TOC
1114 			 * value we need to use.
1115 			 */
1116 			__get_user(entry, (unsigned long __user *)start);
1117 			__get_user(toc, (unsigned long __user *)start+1);
1118 
1119 			/* Check whether the e_entry function descriptor entries
1120 			 * need to be relocated before we can use them.
1121 			 */
1122 			if (load_addr != 0) {
1123 				entry += load_addr;
1124 				toc   += load_addr;
1125 			}
1126 			regs->gpr[2] = toc;
1127 		}
1128 		regs->nip = entry;
1129 		regs->msr = MSR_USER64;
1130 	} else {
1131 		regs->nip = start;
1132 		regs->gpr[2] = 0;
1133 		regs->msr = MSR_USER32;
1134 	}
1135 #endif
1136 	discard_lazy_cpu_state();
1137 #ifdef CONFIG_VSX
1138 	current->thread.used_vsr = 0;
1139 #endif
1140 	memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1141 	current->thread.fp_save_area = NULL;
1142 #ifdef CONFIG_ALTIVEC
1143 	memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1144 	current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1145 	current->thread.vr_save_area = NULL;
1146 	current->thread.vrsave = 0;
1147 	current->thread.used_vr = 0;
1148 #endif /* CONFIG_ALTIVEC */
1149 #ifdef CONFIG_SPE
1150 	memset(current->thread.evr, 0, sizeof(current->thread.evr));
1151 	current->thread.acc = 0;
1152 	current->thread.spefscr = 0;
1153 	current->thread.used_spe = 0;
1154 #endif /* CONFIG_SPE */
1155 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1156 	if (cpu_has_feature(CPU_FTR_TM))
1157 		regs->msr |= MSR_TM;
1158 	current->thread.tm_tfhar = 0;
1159 	current->thread.tm_texasr = 0;
1160 	current->thread.tm_tfiar = 0;
1161 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1162 }
1163 
1164 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1165 		| PR_FP_EXC_RES | PR_FP_EXC_INV)
1166 
1167 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1168 {
1169 	struct pt_regs *regs = tsk->thread.regs;
1170 
1171 	/* This is a bit hairy.  If we are an SPE enabled  processor
1172 	 * (have embedded fp) we store the IEEE exception enable flags in
1173 	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
1174 	 * mode (asyn, precise, disabled) for 'Classic' FP. */
1175 	if (val & PR_FP_EXC_SW_ENABLE) {
1176 #ifdef CONFIG_SPE
1177 		if (cpu_has_feature(CPU_FTR_SPE)) {
1178 			tsk->thread.fpexc_mode = val &
1179 				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1180 			return 0;
1181 		} else {
1182 			return -EINVAL;
1183 		}
1184 #else
1185 		return -EINVAL;
1186 #endif
1187 	}
1188 
1189 	/* on a CONFIG_SPE this does not hurt us.  The bits that
1190 	 * __pack_fe01 use do not overlap with bits used for
1191 	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1192 	 * on CONFIG_SPE implementations are reserved so writing to
1193 	 * them does not change anything */
1194 	if (val > PR_FP_EXC_PRECISE)
1195 		return -EINVAL;
1196 	tsk->thread.fpexc_mode = __pack_fe01(val);
1197 	if (regs != NULL && (regs->msr & MSR_FP) != 0)
1198 		regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1199 			| tsk->thread.fpexc_mode;
1200 	return 0;
1201 }
1202 
1203 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1204 {
1205 	unsigned int val;
1206 
1207 	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1208 #ifdef CONFIG_SPE
1209 		if (cpu_has_feature(CPU_FTR_SPE))
1210 			val = tsk->thread.fpexc_mode;
1211 		else
1212 			return -EINVAL;
1213 #else
1214 		return -EINVAL;
1215 #endif
1216 	else
1217 		val = __unpack_fe01(tsk->thread.fpexc_mode);
1218 	return put_user(val, (unsigned int __user *) adr);
1219 }
1220 
1221 int set_endian(struct task_struct *tsk, unsigned int val)
1222 {
1223 	struct pt_regs *regs = tsk->thread.regs;
1224 
1225 	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1226 	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1227 		return -EINVAL;
1228 
1229 	if (regs == NULL)
1230 		return -EINVAL;
1231 
1232 	if (val == PR_ENDIAN_BIG)
1233 		regs->msr &= ~MSR_LE;
1234 	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1235 		regs->msr |= MSR_LE;
1236 	else
1237 		return -EINVAL;
1238 
1239 	return 0;
1240 }
1241 
1242 int get_endian(struct task_struct *tsk, unsigned long adr)
1243 {
1244 	struct pt_regs *regs = tsk->thread.regs;
1245 	unsigned int val;
1246 
1247 	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1248 	    !cpu_has_feature(CPU_FTR_REAL_LE))
1249 		return -EINVAL;
1250 
1251 	if (regs == NULL)
1252 		return -EINVAL;
1253 
1254 	if (regs->msr & MSR_LE) {
1255 		if (cpu_has_feature(CPU_FTR_REAL_LE))
1256 			val = PR_ENDIAN_LITTLE;
1257 		else
1258 			val = PR_ENDIAN_PPC_LITTLE;
1259 	} else
1260 		val = PR_ENDIAN_BIG;
1261 
1262 	return put_user(val, (unsigned int __user *)adr);
1263 }
1264 
1265 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1266 {
1267 	tsk->thread.align_ctl = val;
1268 	return 0;
1269 }
1270 
1271 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1272 {
1273 	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1274 }
1275 
1276 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1277 				  unsigned long nbytes)
1278 {
1279 	unsigned long stack_page;
1280 	unsigned long cpu = task_cpu(p);
1281 
1282 	/*
1283 	 * Avoid crashing if the stack has overflowed and corrupted
1284 	 * task_cpu(p), which is in the thread_info struct.
1285 	 */
1286 	if (cpu < NR_CPUS && cpu_possible(cpu)) {
1287 		stack_page = (unsigned long) hardirq_ctx[cpu];
1288 		if (sp >= stack_page + sizeof(struct thread_struct)
1289 		    && sp <= stack_page + THREAD_SIZE - nbytes)
1290 			return 1;
1291 
1292 		stack_page = (unsigned long) softirq_ctx[cpu];
1293 		if (sp >= stack_page + sizeof(struct thread_struct)
1294 		    && sp <= stack_page + THREAD_SIZE - nbytes)
1295 			return 1;
1296 	}
1297 	return 0;
1298 }
1299 
1300 int validate_sp(unsigned long sp, struct task_struct *p,
1301 		       unsigned long nbytes)
1302 {
1303 	unsigned long stack_page = (unsigned long)task_stack_page(p);
1304 
1305 	if (sp >= stack_page + sizeof(struct thread_struct)
1306 	    && sp <= stack_page + THREAD_SIZE - nbytes)
1307 		return 1;
1308 
1309 	return valid_irq_stack(sp, p, nbytes);
1310 }
1311 
1312 EXPORT_SYMBOL(validate_sp);
1313 
1314 unsigned long get_wchan(struct task_struct *p)
1315 {
1316 	unsigned long ip, sp;
1317 	int count = 0;
1318 
1319 	if (!p || p == current || p->state == TASK_RUNNING)
1320 		return 0;
1321 
1322 	sp = p->thread.ksp;
1323 	if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1324 		return 0;
1325 
1326 	do {
1327 		sp = *(unsigned long *)sp;
1328 		if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1329 			return 0;
1330 		if (count > 0) {
1331 			ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
1332 			if (!in_sched_functions(ip))
1333 				return ip;
1334 		}
1335 	} while (count++ < 16);
1336 	return 0;
1337 }
1338 
1339 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
1340 
1341 void show_stack(struct task_struct *tsk, unsigned long *stack)
1342 {
1343 	unsigned long sp, ip, lr, newsp;
1344 	int count = 0;
1345 	int firstframe = 1;
1346 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1347 	int curr_frame = current->curr_ret_stack;
1348 	extern void return_to_handler(void);
1349 	unsigned long rth = (unsigned long)return_to_handler;
1350 	unsigned long mrth = -1;
1351 #ifdef CONFIG_PPC64
1352 	extern void mod_return_to_handler(void);
1353 	rth = *(unsigned long *)rth;
1354 	mrth = (unsigned long)mod_return_to_handler;
1355 	mrth = *(unsigned long *)mrth;
1356 #endif
1357 #endif
1358 
1359 	sp = (unsigned long) stack;
1360 	if (tsk == NULL)
1361 		tsk = current;
1362 	if (sp == 0) {
1363 		if (tsk == current)
1364 			asm("mr %0,1" : "=r" (sp));
1365 		else
1366 			sp = tsk->thread.ksp;
1367 	}
1368 
1369 	lr = 0;
1370 	printk("Call Trace:\n");
1371 	do {
1372 		if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
1373 			return;
1374 
1375 		stack = (unsigned long *) sp;
1376 		newsp = stack[0];
1377 		ip = stack[STACK_FRAME_LR_SAVE];
1378 		if (!firstframe || ip != lr) {
1379 			printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
1380 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1381 			if ((ip == rth || ip == mrth) && curr_frame >= 0) {
1382 				printk(" (%pS)",
1383 				       (void *)current->ret_stack[curr_frame].ret);
1384 				curr_frame--;
1385 			}
1386 #endif
1387 			if (firstframe)
1388 				printk(" (unreliable)");
1389 			printk("\n");
1390 		}
1391 		firstframe = 0;
1392 
1393 		/*
1394 		 * See if this is an exception frame.
1395 		 * We look for the "regshere" marker in the current frame.
1396 		 */
1397 		if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
1398 		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
1399 			struct pt_regs *regs = (struct pt_regs *)
1400 				(sp + STACK_FRAME_OVERHEAD);
1401 			lr = regs->link;
1402 			printk("--- Exception: %lx at %pS\n    LR = %pS\n",
1403 			       regs->trap, (void *)regs->nip, (void *)lr);
1404 			firstframe = 1;
1405 		}
1406 
1407 		sp = newsp;
1408 	} while (count++ < kstack_depth_to_print);
1409 }
1410 
1411 #ifdef CONFIG_PPC64
1412 /* Called with hard IRQs off */
1413 void notrace __ppc64_runlatch_on(void)
1414 {
1415 	struct thread_info *ti = current_thread_info();
1416 	unsigned long ctrl;
1417 
1418 	ctrl = mfspr(SPRN_CTRLF);
1419 	ctrl |= CTRL_RUNLATCH;
1420 	mtspr(SPRN_CTRLT, ctrl);
1421 
1422 	ti->local_flags |= _TLF_RUNLATCH;
1423 }
1424 
1425 /* Called with hard IRQs off */
1426 void notrace __ppc64_runlatch_off(void)
1427 {
1428 	struct thread_info *ti = current_thread_info();
1429 	unsigned long ctrl;
1430 
1431 	ti->local_flags &= ~_TLF_RUNLATCH;
1432 
1433 	ctrl = mfspr(SPRN_CTRLF);
1434 	ctrl &= ~CTRL_RUNLATCH;
1435 	mtspr(SPRN_CTRLT, ctrl);
1436 }
1437 #endif /* CONFIG_PPC64 */
1438 
1439 unsigned long arch_align_stack(unsigned long sp)
1440 {
1441 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1442 		sp -= get_random_int() & ~PAGE_MASK;
1443 	return sp & ~0xf;
1444 }
1445 
1446 static inline unsigned long brk_rnd(void)
1447 {
1448         unsigned long rnd = 0;
1449 
1450 	/* 8MB for 32bit, 1GB for 64bit */
1451 	if (is_32bit_task())
1452 		rnd = (long)(get_random_int() % (1<<(23-PAGE_SHIFT)));
1453 	else
1454 		rnd = (long)(get_random_int() % (1<<(30-PAGE_SHIFT)));
1455 
1456 	return rnd << PAGE_SHIFT;
1457 }
1458 
1459 unsigned long arch_randomize_brk(struct mm_struct *mm)
1460 {
1461 	unsigned long base = mm->brk;
1462 	unsigned long ret;
1463 
1464 #ifdef CONFIG_PPC_STD_MMU_64
1465 	/*
1466 	 * If we are using 1TB segments and we are allowed to randomise
1467 	 * the heap, we can put it above 1TB so it is backed by a 1TB
1468 	 * segment. Otherwise the heap will be in the bottom 1TB
1469 	 * which always uses 256MB segments and this may result in a
1470 	 * performance penalty.
1471 	 */
1472 	if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
1473 		base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
1474 #endif
1475 
1476 	ret = PAGE_ALIGN(base + brk_rnd());
1477 
1478 	if (ret < mm->brk)
1479 		return mm->brk;
1480 
1481 	return ret;
1482 }
1483 
1484 unsigned long randomize_et_dyn(unsigned long base)
1485 {
1486 	unsigned long ret = PAGE_ALIGN(base + brk_rnd());
1487 
1488 	if (ret < base)
1489 		return base;
1490 
1491 	return ret;
1492 }
1493