xref: /openbmc/linux/arch/powerpc/kernel/process.c (revision fb574682)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Derived from "arch/i386/kernel/process.c"
4  *    Copyright (C) 1995  Linus Torvalds
5  *
6  *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
7  *  Paul Mackerras (paulus@cs.anu.edu.au)
8  *
9  *  PowerPC version
10  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
11  */
12 
13 #include <linux/errno.h>
14 #include <linux/sched.h>
15 #include <linux/sched/debug.h>
16 #include <linux/sched/task.h>
17 #include <linux/sched/task_stack.h>
18 #include <linux/kernel.h>
19 #include <linux/mm.h>
20 #include <linux/smp.h>
21 #include <linux/stddef.h>
22 #include <linux/unistd.h>
23 #include <linux/ptrace.h>
24 #include <linux/slab.h>
25 #include <linux/user.h>
26 #include <linux/elf.h>
27 #include <linux/prctl.h>
28 #include <linux/init_task.h>
29 #include <linux/export.h>
30 #include <linux/kallsyms.h>
31 #include <linux/mqueue.h>
32 #include <linux/hardirq.h>
33 #include <linux/utsname.h>
34 #include <linux/ftrace.h>
35 #include <linux/kernel_stat.h>
36 #include <linux/personality.h>
37 #include <linux/random.h>
38 #include <linux/hw_breakpoint.h>
39 #include <linux/uaccess.h>
40 #include <linux/elf-randomize.h>
41 #include <linux/pkeys.h>
42 #include <linux/seq_buf.h>
43 
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 #include <asm/hw_irq.h>
58 #endif
59 #include <asm/code-patching.h>
60 #include <asm/exec.h>
61 #include <asm/livepatch.h>
62 #include <asm/cpu_has_feature.h>
63 #include <asm/asm-prototypes.h>
64 #include <asm/stacktrace.h>
65 #include <asm/hw_breakpoint.h>
66 
67 #include <linux/kprobes.h>
68 #include <linux/kdebug.h>
69 
70 /* Transactional Memory debug */
71 #ifdef TM_DEBUG_SW
72 #define TM_DEBUG(x...) printk(KERN_INFO x)
73 #else
74 #define TM_DEBUG(x...) do { } while(0)
75 #endif
76 
77 extern unsigned long _get_SP(void);
78 
79 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
80 /*
81  * Are we running in "Suspend disabled" mode? If so we have to block any
82  * sigreturn that would get us into suspended state, and we also warn in some
83  * other paths that we should never reach with suspend disabled.
84  */
85 bool tm_suspend_disabled __ro_after_init = false;
86 
87 static void check_if_tm_restore_required(struct task_struct *tsk)
88 {
89 	/*
90 	 * If we are saving the current thread's registers, and the
91 	 * thread is in a transactional state, set the TIF_RESTORE_TM
92 	 * bit so that we know to restore the registers before
93 	 * returning to userspace.
94 	 */
95 	if (tsk == current && tsk->thread.regs &&
96 	    MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
97 	    !test_thread_flag(TIF_RESTORE_TM)) {
98 		tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
99 		set_thread_flag(TIF_RESTORE_TM);
100 	}
101 }
102 
103 #else
104 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
105 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
106 
107 bool strict_msr_control;
108 EXPORT_SYMBOL(strict_msr_control);
109 
110 static int __init enable_strict_msr_control(char *str)
111 {
112 	strict_msr_control = true;
113 	pr_info("Enabling strict facility control\n");
114 
115 	return 0;
116 }
117 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
118 
119 /* notrace because it's called by restore_math */
120 unsigned long notrace msr_check_and_set(unsigned long bits)
121 {
122 	unsigned long oldmsr = mfmsr();
123 	unsigned long newmsr;
124 
125 	newmsr = oldmsr | bits;
126 
127 #ifdef CONFIG_VSX
128 	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
129 		newmsr |= MSR_VSX;
130 #endif
131 
132 	if (oldmsr != newmsr)
133 		mtmsr_isync(newmsr);
134 
135 	return newmsr;
136 }
137 EXPORT_SYMBOL_GPL(msr_check_and_set);
138 
139 /* notrace because it's called by restore_math */
140 void notrace __msr_check_and_clear(unsigned long bits)
141 {
142 	unsigned long oldmsr = mfmsr();
143 	unsigned long newmsr;
144 
145 	newmsr = oldmsr & ~bits;
146 
147 #ifdef CONFIG_VSX
148 	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
149 		newmsr &= ~MSR_VSX;
150 #endif
151 
152 	if (oldmsr != newmsr)
153 		mtmsr_isync(newmsr);
154 }
155 EXPORT_SYMBOL(__msr_check_and_clear);
156 
157 #ifdef CONFIG_PPC_FPU
158 static void __giveup_fpu(struct task_struct *tsk)
159 {
160 	unsigned long msr;
161 
162 	save_fpu(tsk);
163 	msr = tsk->thread.regs->msr;
164 	msr &= ~(MSR_FP|MSR_FE0|MSR_FE1);
165 #ifdef CONFIG_VSX
166 	if (cpu_has_feature(CPU_FTR_VSX))
167 		msr &= ~MSR_VSX;
168 #endif
169 	tsk->thread.regs->msr = msr;
170 }
171 
172 void giveup_fpu(struct task_struct *tsk)
173 {
174 	check_if_tm_restore_required(tsk);
175 
176 	msr_check_and_set(MSR_FP);
177 	__giveup_fpu(tsk);
178 	msr_check_and_clear(MSR_FP);
179 }
180 EXPORT_SYMBOL(giveup_fpu);
181 
182 /*
183  * Make sure the floating-point register state in the
184  * the thread_struct is up to date for task tsk.
185  */
186 void flush_fp_to_thread(struct task_struct *tsk)
187 {
188 	if (tsk->thread.regs) {
189 		/*
190 		 * We need to disable preemption here because if we didn't,
191 		 * another process could get scheduled after the regs->msr
192 		 * test but before we have finished saving the FP registers
193 		 * to the thread_struct.  That process could take over the
194 		 * FPU, and then when we get scheduled again we would store
195 		 * bogus values for the remaining FP registers.
196 		 */
197 		preempt_disable();
198 		if (tsk->thread.regs->msr & MSR_FP) {
199 			/*
200 			 * This should only ever be called for current or
201 			 * for a stopped child process.  Since we save away
202 			 * the FP register state on context switch,
203 			 * there is something wrong if a stopped child appears
204 			 * to still have its FP state in the CPU registers.
205 			 */
206 			BUG_ON(tsk != current);
207 			giveup_fpu(tsk);
208 		}
209 		preempt_enable();
210 	}
211 }
212 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
213 
214 void enable_kernel_fp(void)
215 {
216 	unsigned long cpumsr;
217 
218 	WARN_ON(preemptible());
219 
220 	cpumsr = msr_check_and_set(MSR_FP);
221 
222 	if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
223 		check_if_tm_restore_required(current);
224 		/*
225 		 * If a thread has already been reclaimed then the
226 		 * checkpointed registers are on the CPU but have definitely
227 		 * been saved by the reclaim code. Don't need to and *cannot*
228 		 * giveup as this would save  to the 'live' structure not the
229 		 * checkpointed structure.
230 		 */
231 		if (!MSR_TM_ACTIVE(cpumsr) &&
232 		     MSR_TM_ACTIVE(current->thread.regs->msr))
233 			return;
234 		__giveup_fpu(current);
235 	}
236 }
237 EXPORT_SYMBOL(enable_kernel_fp);
238 #endif /* CONFIG_PPC_FPU */
239 
240 #ifdef CONFIG_ALTIVEC
241 static void __giveup_altivec(struct task_struct *tsk)
242 {
243 	unsigned long msr;
244 
245 	save_altivec(tsk);
246 	msr = tsk->thread.regs->msr;
247 	msr &= ~MSR_VEC;
248 #ifdef CONFIG_VSX
249 	if (cpu_has_feature(CPU_FTR_VSX))
250 		msr &= ~MSR_VSX;
251 #endif
252 	tsk->thread.regs->msr = msr;
253 }
254 
255 void giveup_altivec(struct task_struct *tsk)
256 {
257 	check_if_tm_restore_required(tsk);
258 
259 	msr_check_and_set(MSR_VEC);
260 	__giveup_altivec(tsk);
261 	msr_check_and_clear(MSR_VEC);
262 }
263 EXPORT_SYMBOL(giveup_altivec);
264 
265 void enable_kernel_altivec(void)
266 {
267 	unsigned long cpumsr;
268 
269 	WARN_ON(preemptible());
270 
271 	cpumsr = msr_check_and_set(MSR_VEC);
272 
273 	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
274 		check_if_tm_restore_required(current);
275 		/*
276 		 * If a thread has already been reclaimed then the
277 		 * checkpointed registers are on the CPU but have definitely
278 		 * been saved by the reclaim code. Don't need to and *cannot*
279 		 * giveup as this would save  to the 'live' structure not the
280 		 * checkpointed structure.
281 		 */
282 		if (!MSR_TM_ACTIVE(cpumsr) &&
283 		     MSR_TM_ACTIVE(current->thread.regs->msr))
284 			return;
285 		__giveup_altivec(current);
286 	}
287 }
288 EXPORT_SYMBOL(enable_kernel_altivec);
289 
290 /*
291  * Make sure the VMX/Altivec register state in the
292  * the thread_struct is up to date for task tsk.
293  */
294 void flush_altivec_to_thread(struct task_struct *tsk)
295 {
296 	if (tsk->thread.regs) {
297 		preempt_disable();
298 		if (tsk->thread.regs->msr & MSR_VEC) {
299 			BUG_ON(tsk != current);
300 			giveup_altivec(tsk);
301 		}
302 		preempt_enable();
303 	}
304 }
305 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
306 #endif /* CONFIG_ALTIVEC */
307 
308 #ifdef CONFIG_VSX
309 static void __giveup_vsx(struct task_struct *tsk)
310 {
311 	unsigned long msr = tsk->thread.regs->msr;
312 
313 	/*
314 	 * We should never be ssetting MSR_VSX without also setting
315 	 * MSR_FP and MSR_VEC
316 	 */
317 	WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
318 
319 	/* __giveup_fpu will clear MSR_VSX */
320 	if (msr & MSR_FP)
321 		__giveup_fpu(tsk);
322 	if (msr & MSR_VEC)
323 		__giveup_altivec(tsk);
324 }
325 
326 static void giveup_vsx(struct task_struct *tsk)
327 {
328 	check_if_tm_restore_required(tsk);
329 
330 	msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
331 	__giveup_vsx(tsk);
332 	msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
333 }
334 
335 void enable_kernel_vsx(void)
336 {
337 	unsigned long cpumsr;
338 
339 	WARN_ON(preemptible());
340 
341 	cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
342 
343 	if (current->thread.regs &&
344 	    (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
345 		check_if_tm_restore_required(current);
346 		/*
347 		 * If a thread has already been reclaimed then the
348 		 * checkpointed registers are on the CPU but have definitely
349 		 * been saved by the reclaim code. Don't need to and *cannot*
350 		 * giveup as this would save  to the 'live' structure not the
351 		 * checkpointed structure.
352 		 */
353 		if (!MSR_TM_ACTIVE(cpumsr) &&
354 		     MSR_TM_ACTIVE(current->thread.regs->msr))
355 			return;
356 		__giveup_vsx(current);
357 	}
358 }
359 EXPORT_SYMBOL(enable_kernel_vsx);
360 
361 void flush_vsx_to_thread(struct task_struct *tsk)
362 {
363 	if (tsk->thread.regs) {
364 		preempt_disable();
365 		if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
366 			BUG_ON(tsk != current);
367 			giveup_vsx(tsk);
368 		}
369 		preempt_enable();
370 	}
371 }
372 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
373 #endif /* CONFIG_VSX */
374 
375 #ifdef CONFIG_SPE
376 void giveup_spe(struct task_struct *tsk)
377 {
378 	check_if_tm_restore_required(tsk);
379 
380 	msr_check_and_set(MSR_SPE);
381 	__giveup_spe(tsk);
382 	msr_check_and_clear(MSR_SPE);
383 }
384 EXPORT_SYMBOL(giveup_spe);
385 
386 void enable_kernel_spe(void)
387 {
388 	WARN_ON(preemptible());
389 
390 	msr_check_and_set(MSR_SPE);
391 
392 	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
393 		check_if_tm_restore_required(current);
394 		__giveup_spe(current);
395 	}
396 }
397 EXPORT_SYMBOL(enable_kernel_spe);
398 
399 void flush_spe_to_thread(struct task_struct *tsk)
400 {
401 	if (tsk->thread.regs) {
402 		preempt_disable();
403 		if (tsk->thread.regs->msr & MSR_SPE) {
404 			BUG_ON(tsk != current);
405 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
406 			giveup_spe(tsk);
407 		}
408 		preempt_enable();
409 	}
410 }
411 #endif /* CONFIG_SPE */
412 
413 static unsigned long msr_all_available;
414 
415 static int __init init_msr_all_available(void)
416 {
417 #ifdef CONFIG_PPC_FPU
418 	msr_all_available |= MSR_FP;
419 #endif
420 #ifdef CONFIG_ALTIVEC
421 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
422 		msr_all_available |= MSR_VEC;
423 #endif
424 #ifdef CONFIG_VSX
425 	if (cpu_has_feature(CPU_FTR_VSX))
426 		msr_all_available |= MSR_VSX;
427 #endif
428 #ifdef CONFIG_SPE
429 	if (cpu_has_feature(CPU_FTR_SPE))
430 		msr_all_available |= MSR_SPE;
431 #endif
432 
433 	return 0;
434 }
435 early_initcall(init_msr_all_available);
436 
437 void giveup_all(struct task_struct *tsk)
438 {
439 	unsigned long usermsr;
440 
441 	if (!tsk->thread.regs)
442 		return;
443 
444 	check_if_tm_restore_required(tsk);
445 
446 	usermsr = tsk->thread.regs->msr;
447 
448 	if ((usermsr & msr_all_available) == 0)
449 		return;
450 
451 	msr_check_and_set(msr_all_available);
452 
453 	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
454 
455 #ifdef CONFIG_PPC_FPU
456 	if (usermsr & MSR_FP)
457 		__giveup_fpu(tsk);
458 #endif
459 #ifdef CONFIG_ALTIVEC
460 	if (usermsr & MSR_VEC)
461 		__giveup_altivec(tsk);
462 #endif
463 #ifdef CONFIG_SPE
464 	if (usermsr & MSR_SPE)
465 		__giveup_spe(tsk);
466 #endif
467 
468 	msr_check_and_clear(msr_all_available);
469 }
470 EXPORT_SYMBOL(giveup_all);
471 
472 #ifdef CONFIG_PPC_BOOK3S_64
473 #ifdef CONFIG_PPC_FPU
474 static int restore_fp(struct task_struct *tsk)
475 {
476 	if (tsk->thread.load_fp) {
477 		load_fp_state(&current->thread.fp_state);
478 		current->thread.load_fp++;
479 		return 1;
480 	}
481 	return 0;
482 }
483 #else
484 static int restore_fp(struct task_struct *tsk) { return 0; }
485 #endif /* CONFIG_PPC_FPU */
486 
487 #ifdef CONFIG_ALTIVEC
488 #define loadvec(thr) ((thr).load_vec)
489 static int restore_altivec(struct task_struct *tsk)
490 {
491 	if (cpu_has_feature(CPU_FTR_ALTIVEC) && (tsk->thread.load_vec)) {
492 		load_vr_state(&tsk->thread.vr_state);
493 		tsk->thread.used_vr = 1;
494 		tsk->thread.load_vec++;
495 
496 		return 1;
497 	}
498 	return 0;
499 }
500 #else
501 #define loadvec(thr) 0
502 static inline int restore_altivec(struct task_struct *tsk) { return 0; }
503 #endif /* CONFIG_ALTIVEC */
504 
505 #ifdef CONFIG_VSX
506 static int restore_vsx(struct task_struct *tsk)
507 {
508 	if (cpu_has_feature(CPU_FTR_VSX)) {
509 		tsk->thread.used_vsr = 1;
510 		return 1;
511 	}
512 
513 	return 0;
514 }
515 #else
516 static inline int restore_vsx(struct task_struct *tsk) { return 0; }
517 #endif /* CONFIG_VSX */
518 
519 /*
520  * The exception exit path calls restore_math() with interrupts hard disabled
521  * but the soft irq state not "reconciled". ftrace code that calls
522  * local_irq_save/restore causes warnings.
523  *
524  * Rather than complicate the exit path, just don't trace restore_math. This
525  * could be done by having ftrace entry code check for this un-reconciled
526  * condition where MSR[EE]=0 and PACA_IRQ_HARD_DIS is not set, and
527  * temporarily fix it up for the duration of the ftrace call.
528  */
529 void notrace restore_math(struct pt_regs *regs)
530 {
531 	unsigned long msr;
532 
533 	if (!MSR_TM_ACTIVE(regs->msr) &&
534 		!current->thread.load_fp && !loadvec(current->thread))
535 		return;
536 
537 	msr = regs->msr;
538 	msr_check_and_set(msr_all_available);
539 
540 	/*
541 	 * Only reload if the bit is not set in the user MSR, the bit BEING set
542 	 * indicates that the registers are hot
543 	 */
544 	if ((!(msr & MSR_FP)) && restore_fp(current))
545 		msr |= MSR_FP | current->thread.fpexc_mode;
546 
547 	if ((!(msr & MSR_VEC)) && restore_altivec(current))
548 		msr |= MSR_VEC;
549 
550 	if ((msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC) &&
551 			restore_vsx(current)) {
552 		msr |= MSR_VSX;
553 	}
554 
555 	msr_check_and_clear(msr_all_available);
556 
557 	regs->msr = msr;
558 }
559 #endif
560 
561 static void save_all(struct task_struct *tsk)
562 {
563 	unsigned long usermsr;
564 
565 	if (!tsk->thread.regs)
566 		return;
567 
568 	usermsr = tsk->thread.regs->msr;
569 
570 	if ((usermsr & msr_all_available) == 0)
571 		return;
572 
573 	msr_check_and_set(msr_all_available);
574 
575 	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
576 
577 	if (usermsr & MSR_FP)
578 		save_fpu(tsk);
579 
580 	if (usermsr & MSR_VEC)
581 		save_altivec(tsk);
582 
583 	if (usermsr & MSR_SPE)
584 		__giveup_spe(tsk);
585 
586 	msr_check_and_clear(msr_all_available);
587 	thread_pkey_regs_save(&tsk->thread);
588 }
589 
590 void flush_all_to_thread(struct task_struct *tsk)
591 {
592 	if (tsk->thread.regs) {
593 		preempt_disable();
594 		BUG_ON(tsk != current);
595 #ifdef CONFIG_SPE
596 		if (tsk->thread.regs->msr & MSR_SPE)
597 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
598 #endif
599 		save_all(tsk);
600 
601 		preempt_enable();
602 	}
603 }
604 EXPORT_SYMBOL(flush_all_to_thread);
605 
606 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
607 void do_send_trap(struct pt_regs *regs, unsigned long address,
608 		  unsigned long error_code, int breakpt)
609 {
610 	current->thread.trap_nr = TRAP_HWBKPT;
611 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
612 			11, SIGSEGV) == NOTIFY_STOP)
613 		return;
614 
615 	/* Deliver the signal to userspace */
616 	force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
617 				    (void __user *)address);
618 }
619 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
620 void do_break (struct pt_regs *regs, unsigned long address,
621 		    unsigned long error_code)
622 {
623 	current->thread.trap_nr = TRAP_HWBKPT;
624 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
625 			11, SIGSEGV) == NOTIFY_STOP)
626 		return;
627 
628 	if (debugger_break_match(regs))
629 		return;
630 
631 	/* Deliver the signal to userspace */
632 	force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)address);
633 }
634 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
635 
636 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk[HBP_NUM_MAX]);
637 
638 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
639 /*
640  * Set the debug registers back to their default "safe" values.
641  */
642 static void set_debug_reg_defaults(struct thread_struct *thread)
643 {
644 	thread->debug.iac1 = thread->debug.iac2 = 0;
645 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
646 	thread->debug.iac3 = thread->debug.iac4 = 0;
647 #endif
648 	thread->debug.dac1 = thread->debug.dac2 = 0;
649 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
650 	thread->debug.dvc1 = thread->debug.dvc2 = 0;
651 #endif
652 	thread->debug.dbcr0 = 0;
653 #ifdef CONFIG_BOOKE
654 	/*
655 	 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
656 	 */
657 	thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
658 			DBCR1_IAC3US | DBCR1_IAC4US;
659 	/*
660 	 * Force Data Address Compare User/Supervisor bits to be User-only
661 	 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
662 	 */
663 	thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
664 #else
665 	thread->debug.dbcr1 = 0;
666 #endif
667 }
668 
669 static void prime_debug_regs(struct debug_reg *debug)
670 {
671 	/*
672 	 * We could have inherited MSR_DE from userspace, since
673 	 * it doesn't get cleared on exception entry.  Make sure
674 	 * MSR_DE is clear before we enable any debug events.
675 	 */
676 	mtmsr(mfmsr() & ~MSR_DE);
677 
678 	mtspr(SPRN_IAC1, debug->iac1);
679 	mtspr(SPRN_IAC2, debug->iac2);
680 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
681 	mtspr(SPRN_IAC3, debug->iac3);
682 	mtspr(SPRN_IAC4, debug->iac4);
683 #endif
684 	mtspr(SPRN_DAC1, debug->dac1);
685 	mtspr(SPRN_DAC2, debug->dac2);
686 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
687 	mtspr(SPRN_DVC1, debug->dvc1);
688 	mtspr(SPRN_DVC2, debug->dvc2);
689 #endif
690 	mtspr(SPRN_DBCR0, debug->dbcr0);
691 	mtspr(SPRN_DBCR1, debug->dbcr1);
692 #ifdef CONFIG_BOOKE
693 	mtspr(SPRN_DBCR2, debug->dbcr2);
694 #endif
695 }
696 /*
697  * Unless neither the old or new thread are making use of the
698  * debug registers, set the debug registers from the values
699  * stored in the new thread.
700  */
701 void switch_booke_debug_regs(struct debug_reg *new_debug)
702 {
703 	if ((current->thread.debug.dbcr0 & DBCR0_IDM)
704 		|| (new_debug->dbcr0 & DBCR0_IDM))
705 			prime_debug_regs(new_debug);
706 }
707 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
708 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
709 #ifndef CONFIG_HAVE_HW_BREAKPOINT
710 static void set_breakpoint(int i, struct arch_hw_breakpoint *brk)
711 {
712 	preempt_disable();
713 	__set_breakpoint(i, brk);
714 	preempt_enable();
715 }
716 
717 static void set_debug_reg_defaults(struct thread_struct *thread)
718 {
719 	int i;
720 	struct arch_hw_breakpoint null_brk = {0};
721 
722 	for (i = 0; i < nr_wp_slots(); i++) {
723 		thread->hw_brk[i] = null_brk;
724 		if (ppc_breakpoint_available())
725 			set_breakpoint(i, &thread->hw_brk[i]);
726 	}
727 }
728 
729 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
730 				struct arch_hw_breakpoint *b)
731 {
732 	if (a->address != b->address)
733 		return false;
734 	if (a->type != b->type)
735 		return false;
736 	if (a->len != b->len)
737 		return false;
738 	/* no need to check hw_len. it's calculated from address and len */
739 	return true;
740 }
741 
742 static void switch_hw_breakpoint(struct task_struct *new)
743 {
744 	int i;
745 
746 	for (i = 0; i < nr_wp_slots(); i++) {
747 		if (likely(hw_brk_match(this_cpu_ptr(&current_brk[i]),
748 					&new->thread.hw_brk[i])))
749 			continue;
750 
751 		__set_breakpoint(i, &new->thread.hw_brk[i]);
752 	}
753 }
754 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
755 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
756 
757 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
758 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
759 {
760 	mtspr(SPRN_DAC1, dabr);
761 #ifdef CONFIG_PPC_47x
762 	isync();
763 #endif
764 	return 0;
765 }
766 #elif defined(CONFIG_PPC_BOOK3S)
767 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
768 {
769 	mtspr(SPRN_DABR, dabr);
770 	if (cpu_has_feature(CPU_FTR_DABRX))
771 		mtspr(SPRN_DABRX, dabrx);
772 	return 0;
773 }
774 #else
775 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
776 {
777 	return -EINVAL;
778 }
779 #endif
780 
781 static inline int set_dabr(struct arch_hw_breakpoint *brk)
782 {
783 	unsigned long dabr, dabrx;
784 
785 	dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
786 	dabrx = ((brk->type >> 3) & 0x7);
787 
788 	if (ppc_md.set_dabr)
789 		return ppc_md.set_dabr(dabr, dabrx);
790 
791 	return __set_dabr(dabr, dabrx);
792 }
793 
794 static inline int set_breakpoint_8xx(struct arch_hw_breakpoint *brk)
795 {
796 	unsigned long lctrl1 = LCTRL1_CTE_GT | LCTRL1_CTF_LT | LCTRL1_CRWE_RW |
797 			       LCTRL1_CRWF_RW;
798 	unsigned long lctrl2 = LCTRL2_LW0EN | LCTRL2_LW0LADC | LCTRL2_SLW0EN;
799 	unsigned long start_addr = ALIGN_DOWN(brk->address, HW_BREAKPOINT_SIZE);
800 	unsigned long end_addr = ALIGN(brk->address + brk->len, HW_BREAKPOINT_SIZE);
801 
802 	if (start_addr == 0)
803 		lctrl2 |= LCTRL2_LW0LA_F;
804 	else if (end_addr == 0)
805 		lctrl2 |= LCTRL2_LW0LA_E;
806 	else
807 		lctrl2 |= LCTRL2_LW0LA_EandF;
808 
809 	mtspr(SPRN_LCTRL2, 0);
810 
811 	if ((brk->type & HW_BRK_TYPE_RDWR) == 0)
812 		return 0;
813 
814 	if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
815 		lctrl1 |= LCTRL1_CRWE_RO | LCTRL1_CRWF_RO;
816 	if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
817 		lctrl1 |= LCTRL1_CRWE_WO | LCTRL1_CRWF_WO;
818 
819 	mtspr(SPRN_CMPE, start_addr - 1);
820 	mtspr(SPRN_CMPF, end_addr);
821 	mtspr(SPRN_LCTRL1, lctrl1);
822 	mtspr(SPRN_LCTRL2, lctrl2);
823 
824 	return 0;
825 }
826 
827 void __set_breakpoint(int nr, struct arch_hw_breakpoint *brk)
828 {
829 	memcpy(this_cpu_ptr(&current_brk[nr]), brk, sizeof(*brk));
830 
831 	if (dawr_enabled())
832 		// Power8 or later
833 		set_dawr(nr, brk);
834 	else if (IS_ENABLED(CONFIG_PPC_8xx))
835 		set_breakpoint_8xx(brk);
836 	else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
837 		// Power7 or earlier
838 		set_dabr(brk);
839 	else
840 		// Shouldn't happen due to higher level checks
841 		WARN_ON_ONCE(1);
842 }
843 
844 /* Check if we have DAWR or DABR hardware */
845 bool ppc_breakpoint_available(void)
846 {
847 	if (dawr_enabled())
848 		return true; /* POWER8 DAWR or POWER9 forced DAWR */
849 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
850 		return false; /* POWER9 with DAWR disabled */
851 	/* DABR: Everything but POWER8 and POWER9 */
852 	return true;
853 }
854 EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
855 
856 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
857 
858 static inline bool tm_enabled(struct task_struct *tsk)
859 {
860 	return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
861 }
862 
863 static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
864 {
865 	/*
866 	 * Use the current MSR TM suspended bit to track if we have
867 	 * checkpointed state outstanding.
868 	 * On signal delivery, we'd normally reclaim the checkpointed
869 	 * state to obtain stack pointer (see:get_tm_stackpointer()).
870 	 * This will then directly return to userspace without going
871 	 * through __switch_to(). However, if the stack frame is bad,
872 	 * we need to exit this thread which calls __switch_to() which
873 	 * will again attempt to reclaim the already saved tm state.
874 	 * Hence we need to check that we've not already reclaimed
875 	 * this state.
876 	 * We do this using the current MSR, rather tracking it in
877 	 * some specific thread_struct bit, as it has the additional
878 	 * benefit of checking for a potential TM bad thing exception.
879 	 */
880 	if (!MSR_TM_SUSPENDED(mfmsr()))
881 		return;
882 
883 	giveup_all(container_of(thr, struct task_struct, thread));
884 
885 	tm_reclaim(thr, cause);
886 
887 	/*
888 	 * If we are in a transaction and FP is off then we can't have
889 	 * used FP inside that transaction. Hence the checkpointed
890 	 * state is the same as the live state. We need to copy the
891 	 * live state to the checkpointed state so that when the
892 	 * transaction is restored, the checkpointed state is correct
893 	 * and the aborted transaction sees the correct state. We use
894 	 * ckpt_regs.msr here as that's what tm_reclaim will use to
895 	 * determine if it's going to write the checkpointed state or
896 	 * not. So either this will write the checkpointed registers,
897 	 * or reclaim will. Similarly for VMX.
898 	 */
899 	if ((thr->ckpt_regs.msr & MSR_FP) == 0)
900 		memcpy(&thr->ckfp_state, &thr->fp_state,
901 		       sizeof(struct thread_fp_state));
902 	if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
903 		memcpy(&thr->ckvr_state, &thr->vr_state,
904 		       sizeof(struct thread_vr_state));
905 }
906 
907 void tm_reclaim_current(uint8_t cause)
908 {
909 	tm_enable();
910 	tm_reclaim_thread(&current->thread, cause);
911 }
912 
913 static inline void tm_reclaim_task(struct task_struct *tsk)
914 {
915 	/* We have to work out if we're switching from/to a task that's in the
916 	 * middle of a transaction.
917 	 *
918 	 * In switching we need to maintain a 2nd register state as
919 	 * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
920 	 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
921 	 * ckvr_state
922 	 *
923 	 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
924 	 */
925 	struct thread_struct *thr = &tsk->thread;
926 
927 	if (!thr->regs)
928 		return;
929 
930 	if (!MSR_TM_ACTIVE(thr->regs->msr))
931 		goto out_and_saveregs;
932 
933 	WARN_ON(tm_suspend_disabled);
934 
935 	TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
936 		 "ccr=%lx, msr=%lx, trap=%lx)\n",
937 		 tsk->pid, thr->regs->nip,
938 		 thr->regs->ccr, thr->regs->msr,
939 		 thr->regs->trap);
940 
941 	tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
942 
943 	TM_DEBUG("--- tm_reclaim on pid %d complete\n",
944 		 tsk->pid);
945 
946 out_and_saveregs:
947 	/* Always save the regs here, even if a transaction's not active.
948 	 * This context-switches a thread's TM info SPRs.  We do it here to
949 	 * be consistent with the restore path (in recheckpoint) which
950 	 * cannot happen later in _switch().
951 	 */
952 	tm_save_sprs(thr);
953 }
954 
955 extern void __tm_recheckpoint(struct thread_struct *thread);
956 
957 void tm_recheckpoint(struct thread_struct *thread)
958 {
959 	unsigned long flags;
960 
961 	if (!(thread->regs->msr & MSR_TM))
962 		return;
963 
964 	/* We really can't be interrupted here as the TEXASR registers can't
965 	 * change and later in the trecheckpoint code, we have a userspace R1.
966 	 * So let's hard disable over this region.
967 	 */
968 	local_irq_save(flags);
969 	hard_irq_disable();
970 
971 	/* The TM SPRs are restored here, so that TEXASR.FS can be set
972 	 * before the trecheckpoint and no explosion occurs.
973 	 */
974 	tm_restore_sprs(thread);
975 
976 	__tm_recheckpoint(thread);
977 
978 	local_irq_restore(flags);
979 }
980 
981 static inline void tm_recheckpoint_new_task(struct task_struct *new)
982 {
983 	if (!cpu_has_feature(CPU_FTR_TM))
984 		return;
985 
986 	/* Recheckpoint the registers of the thread we're about to switch to.
987 	 *
988 	 * If the task was using FP, we non-lazily reload both the original and
989 	 * the speculative FP register states.  This is because the kernel
990 	 * doesn't see if/when a TM rollback occurs, so if we take an FP
991 	 * unavailable later, we are unable to determine which set of FP regs
992 	 * need to be restored.
993 	 */
994 	if (!tm_enabled(new))
995 		return;
996 
997 	if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
998 		tm_restore_sprs(&new->thread);
999 		return;
1000 	}
1001 	/* Recheckpoint to restore original checkpointed register state. */
1002 	TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
1003 		 new->pid, new->thread.regs->msr);
1004 
1005 	tm_recheckpoint(&new->thread);
1006 
1007 	/*
1008 	 * The checkpointed state has been restored but the live state has
1009 	 * not, ensure all the math functionality is turned off to trigger
1010 	 * restore_math() to reload.
1011 	 */
1012 	new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1013 
1014 	TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1015 		 "(kernel msr 0x%lx)\n",
1016 		 new->pid, mfmsr());
1017 }
1018 
1019 static inline void __switch_to_tm(struct task_struct *prev,
1020 		struct task_struct *new)
1021 {
1022 	if (cpu_has_feature(CPU_FTR_TM)) {
1023 		if (tm_enabled(prev) || tm_enabled(new))
1024 			tm_enable();
1025 
1026 		if (tm_enabled(prev)) {
1027 			prev->thread.load_tm++;
1028 			tm_reclaim_task(prev);
1029 			if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1030 				prev->thread.regs->msr &= ~MSR_TM;
1031 		}
1032 
1033 		tm_recheckpoint_new_task(new);
1034 	}
1035 }
1036 
1037 /*
1038  * This is called if we are on the way out to userspace and the
1039  * TIF_RESTORE_TM flag is set.  It checks if we need to reload
1040  * FP and/or vector state and does so if necessary.
1041  * If userspace is inside a transaction (whether active or
1042  * suspended) and FP/VMX/VSX instructions have ever been enabled
1043  * inside that transaction, then we have to keep them enabled
1044  * and keep the FP/VMX/VSX state loaded while ever the transaction
1045  * continues.  The reason is that if we didn't, and subsequently
1046  * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1047  * we don't know whether it's the same transaction, and thus we
1048  * don't know which of the checkpointed state and the transactional
1049  * state to use.
1050  */
1051 void restore_tm_state(struct pt_regs *regs)
1052 {
1053 	unsigned long msr_diff;
1054 
1055 	/*
1056 	 * This is the only moment we should clear TIF_RESTORE_TM as
1057 	 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1058 	 * again, anything else could lead to an incorrect ckpt_msr being
1059 	 * saved and therefore incorrect signal contexts.
1060 	 */
1061 	clear_thread_flag(TIF_RESTORE_TM);
1062 	if (!MSR_TM_ACTIVE(regs->msr))
1063 		return;
1064 
1065 	msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1066 	msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1067 
1068 	/* Ensure that restore_math() will restore */
1069 	if (msr_diff & MSR_FP)
1070 		current->thread.load_fp = 1;
1071 #ifdef CONFIG_ALTIVEC
1072 	if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1073 		current->thread.load_vec = 1;
1074 #endif
1075 	restore_math(regs);
1076 
1077 	regs->msr |= msr_diff;
1078 }
1079 
1080 #else
1081 #define tm_recheckpoint_new_task(new)
1082 #define __switch_to_tm(prev, new)
1083 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1084 
1085 static inline void save_sprs(struct thread_struct *t)
1086 {
1087 #ifdef CONFIG_ALTIVEC
1088 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
1089 		t->vrsave = mfspr(SPRN_VRSAVE);
1090 #endif
1091 #ifdef CONFIG_PPC_BOOK3S_64
1092 	if (cpu_has_feature(CPU_FTR_DSCR))
1093 		t->dscr = mfspr(SPRN_DSCR);
1094 
1095 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1096 		t->bescr = mfspr(SPRN_BESCR);
1097 		t->ebbhr = mfspr(SPRN_EBBHR);
1098 		t->ebbrr = mfspr(SPRN_EBBRR);
1099 
1100 		t->fscr = mfspr(SPRN_FSCR);
1101 
1102 		/*
1103 		 * Note that the TAR is not available for use in the kernel.
1104 		 * (To provide this, the TAR should be backed up/restored on
1105 		 * exception entry/exit instead, and be in pt_regs.  FIXME,
1106 		 * this should be in pt_regs anyway (for debug).)
1107 		 */
1108 		t->tar = mfspr(SPRN_TAR);
1109 	}
1110 #endif
1111 
1112 	thread_pkey_regs_save(t);
1113 }
1114 
1115 static inline void restore_sprs(struct thread_struct *old_thread,
1116 				struct thread_struct *new_thread)
1117 {
1118 #ifdef CONFIG_ALTIVEC
1119 	if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1120 	    old_thread->vrsave != new_thread->vrsave)
1121 		mtspr(SPRN_VRSAVE, new_thread->vrsave);
1122 #endif
1123 #ifdef CONFIG_PPC_BOOK3S_64
1124 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1125 		u64 dscr = get_paca()->dscr_default;
1126 		if (new_thread->dscr_inherit)
1127 			dscr = new_thread->dscr;
1128 
1129 		if (old_thread->dscr != dscr)
1130 			mtspr(SPRN_DSCR, dscr);
1131 	}
1132 
1133 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1134 		if (old_thread->bescr != new_thread->bescr)
1135 			mtspr(SPRN_BESCR, new_thread->bescr);
1136 		if (old_thread->ebbhr != new_thread->ebbhr)
1137 			mtspr(SPRN_EBBHR, new_thread->ebbhr);
1138 		if (old_thread->ebbrr != new_thread->ebbrr)
1139 			mtspr(SPRN_EBBRR, new_thread->ebbrr);
1140 
1141 		if (old_thread->fscr != new_thread->fscr)
1142 			mtspr(SPRN_FSCR, new_thread->fscr);
1143 
1144 		if (old_thread->tar != new_thread->tar)
1145 			mtspr(SPRN_TAR, new_thread->tar);
1146 	}
1147 
1148 	if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1149 	    old_thread->tidr != new_thread->tidr)
1150 		mtspr(SPRN_TIDR, new_thread->tidr);
1151 #endif
1152 
1153 	thread_pkey_regs_restore(new_thread, old_thread);
1154 }
1155 
1156 struct task_struct *__switch_to(struct task_struct *prev,
1157 	struct task_struct *new)
1158 {
1159 	struct thread_struct *new_thread, *old_thread;
1160 	struct task_struct *last;
1161 #ifdef CONFIG_PPC_BOOK3S_64
1162 	struct ppc64_tlb_batch *batch;
1163 #endif
1164 
1165 	new_thread = &new->thread;
1166 	old_thread = &current->thread;
1167 
1168 	WARN_ON(!irqs_disabled());
1169 
1170 #ifdef CONFIG_PPC_BOOK3S_64
1171 	batch = this_cpu_ptr(&ppc64_tlb_batch);
1172 	if (batch->active) {
1173 		current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1174 		if (batch->index)
1175 			__flush_tlb_pending(batch);
1176 		batch->active = 0;
1177 	}
1178 #endif /* CONFIG_PPC_BOOK3S_64 */
1179 
1180 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1181 	switch_booke_debug_regs(&new->thread.debug);
1182 #else
1183 /*
1184  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1185  * schedule DABR
1186  */
1187 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1188 	switch_hw_breakpoint(new);
1189 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1190 #endif
1191 
1192 	/*
1193 	 * We need to save SPRs before treclaim/trecheckpoint as these will
1194 	 * change a number of them.
1195 	 */
1196 	save_sprs(&prev->thread);
1197 
1198 	/* Save FPU, Altivec, VSX and SPE state */
1199 	giveup_all(prev);
1200 
1201 	__switch_to_tm(prev, new);
1202 
1203 	if (!radix_enabled()) {
1204 		/*
1205 		 * We can't take a PMU exception inside _switch() since there
1206 		 * is a window where the kernel stack SLB and the kernel stack
1207 		 * are out of sync. Hard disable here.
1208 		 */
1209 		hard_irq_disable();
1210 	}
1211 
1212 	/*
1213 	 * Call restore_sprs() before calling _switch(). If we move it after
1214 	 * _switch() then we miss out on calling it for new tasks. The reason
1215 	 * for this is we manually create a stack frame for new tasks that
1216 	 * directly returns through ret_from_fork() or
1217 	 * ret_from_kernel_thread(). See copy_thread() for details.
1218 	 */
1219 	restore_sprs(old_thread, new_thread);
1220 
1221 	last = _switch(old_thread, new_thread);
1222 
1223 #ifdef CONFIG_PPC_BOOK3S_64
1224 	if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1225 		current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1226 		batch = this_cpu_ptr(&ppc64_tlb_batch);
1227 		batch->active = 1;
1228 	}
1229 
1230 	if (current->thread.regs) {
1231 		restore_math(current->thread.regs);
1232 
1233 		/*
1234 		 * The copy-paste buffer can only store into foreign real
1235 		 * addresses, so unprivileged processes can not see the
1236 		 * data or use it in any way unless they have foreign real
1237 		 * mappings. If the new process has the foreign real address
1238 		 * mappings, we must issue a cp_abort to clear any state and
1239 		 * prevent snooping, corruption or a covert channel.
1240 		 */
1241 		if (current->mm &&
1242 			atomic_read(&current->mm->context.vas_windows))
1243 			asm volatile(PPC_CP_ABORT);
1244 	}
1245 #endif /* CONFIG_PPC_BOOK3S_64 */
1246 
1247 	return last;
1248 }
1249 
1250 #define NR_INSN_TO_PRINT	16
1251 
1252 static void show_instructions(struct pt_regs *regs)
1253 {
1254 	int i;
1255 	unsigned long pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1256 
1257 	printk("Instruction dump:");
1258 
1259 	for (i = 0; i < NR_INSN_TO_PRINT; i++) {
1260 		int instr;
1261 
1262 		if (!(i % 8))
1263 			pr_cont("\n");
1264 
1265 #if !defined(CONFIG_BOOKE)
1266 		/* If executing with the IMMU off, adjust pc rather
1267 		 * than print XXXXXXXX.
1268 		 */
1269 		if (!(regs->msr & MSR_IR))
1270 			pc = (unsigned long)phys_to_virt(pc);
1271 #endif
1272 
1273 		if (!__kernel_text_address(pc) ||
1274 		    probe_kernel_address((const void *)pc, instr)) {
1275 			pr_cont("XXXXXXXX ");
1276 		} else {
1277 			if (regs->nip == pc)
1278 				pr_cont("<%08x> ", instr);
1279 			else
1280 				pr_cont("%08x ", instr);
1281 		}
1282 
1283 		pc += sizeof(int);
1284 	}
1285 
1286 	pr_cont("\n");
1287 }
1288 
1289 void show_user_instructions(struct pt_regs *regs)
1290 {
1291 	unsigned long pc;
1292 	int n = NR_INSN_TO_PRINT;
1293 	struct seq_buf s;
1294 	char buf[96]; /* enough for 8 times 9 + 2 chars */
1295 
1296 	pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1297 
1298 	seq_buf_init(&s, buf, sizeof(buf));
1299 
1300 	while (n) {
1301 		int i;
1302 
1303 		seq_buf_clear(&s);
1304 
1305 		for (i = 0; i < 8 && n; i++, n--, pc += sizeof(int)) {
1306 			int instr;
1307 
1308 			if (probe_user_read(&instr, (void __user *)pc, sizeof(instr))) {
1309 				seq_buf_printf(&s, "XXXXXXXX ");
1310 				continue;
1311 			}
1312 			seq_buf_printf(&s, regs->nip == pc ? "<%08x> " : "%08x ", instr);
1313 		}
1314 
1315 		if (!seq_buf_has_overflowed(&s))
1316 			pr_info("%s[%d]: code: %s\n", current->comm,
1317 				current->pid, s.buffer);
1318 	}
1319 }
1320 
1321 struct regbit {
1322 	unsigned long bit;
1323 	const char *name;
1324 };
1325 
1326 static struct regbit msr_bits[] = {
1327 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1328 	{MSR_SF,	"SF"},
1329 	{MSR_HV,	"HV"},
1330 #endif
1331 	{MSR_VEC,	"VEC"},
1332 	{MSR_VSX,	"VSX"},
1333 #ifdef CONFIG_BOOKE
1334 	{MSR_CE,	"CE"},
1335 #endif
1336 	{MSR_EE,	"EE"},
1337 	{MSR_PR,	"PR"},
1338 	{MSR_FP,	"FP"},
1339 	{MSR_ME,	"ME"},
1340 #ifdef CONFIG_BOOKE
1341 	{MSR_DE,	"DE"},
1342 #else
1343 	{MSR_SE,	"SE"},
1344 	{MSR_BE,	"BE"},
1345 #endif
1346 	{MSR_IR,	"IR"},
1347 	{MSR_DR,	"DR"},
1348 	{MSR_PMM,	"PMM"},
1349 #ifndef CONFIG_BOOKE
1350 	{MSR_RI,	"RI"},
1351 	{MSR_LE,	"LE"},
1352 #endif
1353 	{0,		NULL}
1354 };
1355 
1356 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1357 {
1358 	const char *s = "";
1359 
1360 	for (; bits->bit; ++bits)
1361 		if (val & bits->bit) {
1362 			pr_cont("%s%s", s, bits->name);
1363 			s = sep;
1364 		}
1365 }
1366 
1367 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1368 static struct regbit msr_tm_bits[] = {
1369 	{MSR_TS_T,	"T"},
1370 	{MSR_TS_S,	"S"},
1371 	{MSR_TM,	"E"},
1372 	{0,		NULL}
1373 };
1374 
1375 static void print_tm_bits(unsigned long val)
1376 {
1377 /*
1378  * This only prints something if at least one of the TM bit is set.
1379  * Inside the TM[], the output means:
1380  *   E: Enabled		(bit 32)
1381  *   S: Suspended	(bit 33)
1382  *   T: Transactional	(bit 34)
1383  */
1384 	if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1385 		pr_cont(",TM[");
1386 		print_bits(val, msr_tm_bits, "");
1387 		pr_cont("]");
1388 	}
1389 }
1390 #else
1391 static void print_tm_bits(unsigned long val) {}
1392 #endif
1393 
1394 static void print_msr_bits(unsigned long val)
1395 {
1396 	pr_cont("<");
1397 	print_bits(val, msr_bits, ",");
1398 	print_tm_bits(val);
1399 	pr_cont(">");
1400 }
1401 
1402 #ifdef CONFIG_PPC64
1403 #define REG		"%016lx"
1404 #define REGS_PER_LINE	4
1405 #define LAST_VOLATILE	13
1406 #else
1407 #define REG		"%08lx"
1408 #define REGS_PER_LINE	8
1409 #define LAST_VOLATILE	12
1410 #endif
1411 
1412 void show_regs(struct pt_regs * regs)
1413 {
1414 	int i, trap;
1415 
1416 	show_regs_print_info(KERN_DEFAULT);
1417 
1418 	printk("NIP:  "REG" LR: "REG" CTR: "REG"\n",
1419 	       regs->nip, regs->link, regs->ctr);
1420 	printk("REGS: %px TRAP: %04lx   %s  (%s)\n",
1421 	       regs, regs->trap, print_tainted(), init_utsname()->release);
1422 	printk("MSR:  "REG" ", regs->msr);
1423 	print_msr_bits(regs->msr);
1424 	pr_cont("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1425 	trap = TRAP(regs);
1426 	if (!trap_is_syscall(regs) && cpu_has_feature(CPU_FTR_CFAR))
1427 		pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1428 	if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1429 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1430 		pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1431 #else
1432 		pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1433 #endif
1434 #ifdef CONFIG_PPC64
1435 	pr_cont("IRQMASK: %lx ", regs->softe);
1436 #endif
1437 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1438 	if (MSR_TM_ACTIVE(regs->msr))
1439 		pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1440 #endif
1441 
1442 	for (i = 0;  i < 32;  i++) {
1443 		if ((i % REGS_PER_LINE) == 0)
1444 			pr_cont("\nGPR%02d: ", i);
1445 		pr_cont(REG " ", regs->gpr[i]);
1446 		if (i == LAST_VOLATILE && !FULL_REGS(regs))
1447 			break;
1448 	}
1449 	pr_cont("\n");
1450 #ifdef CONFIG_KALLSYMS
1451 	/*
1452 	 * Lookup NIP late so we have the best change of getting the
1453 	 * above info out without failing
1454 	 */
1455 	printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1456 	printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1457 #endif
1458 	show_stack(current, (unsigned long *) regs->gpr[1], KERN_DEFAULT);
1459 	if (!user_mode(regs))
1460 		show_instructions(regs);
1461 }
1462 
1463 void flush_thread(void)
1464 {
1465 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1466 	flush_ptrace_hw_breakpoint(current);
1467 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1468 	set_debug_reg_defaults(&current->thread);
1469 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1470 }
1471 
1472 #ifdef CONFIG_PPC_BOOK3S_64
1473 void arch_setup_new_exec(void)
1474 {
1475 	if (radix_enabled())
1476 		return;
1477 	hash__setup_new_exec();
1478 }
1479 #endif
1480 
1481 #ifdef CONFIG_PPC64
1482 /**
1483  * Assign a TIDR (thread ID) for task @t and set it in the thread
1484  * structure. For now, we only support setting TIDR for 'current' task.
1485  *
1486  * Since the TID value is a truncated form of it PID, it is possible
1487  * (but unlikely) for 2 threads to have the same TID. In the unlikely event
1488  * that 2 threads share the same TID and are waiting, one of the following
1489  * cases will happen:
1490  *
1491  * 1. The correct thread is running, the wrong thread is not
1492  * In this situation, the correct thread is woken and proceeds to pass it's
1493  * condition check.
1494  *
1495  * 2. Neither threads are running
1496  * In this situation, neither thread will be woken. When scheduled, the waiting
1497  * threads will execute either a wait, which will return immediately, followed
1498  * by a condition check, which will pass for the correct thread and fail
1499  * for the wrong thread, or they will execute the condition check immediately.
1500  *
1501  * 3. The wrong thread is running, the correct thread is not
1502  * The wrong thread will be woken, but will fail it's condition check and
1503  * re-execute wait. The correct thread, when scheduled, will execute either
1504  * it's condition check (which will pass), or wait, which returns immediately
1505  * when called the first time after the thread is scheduled, followed by it's
1506  * condition check (which will pass).
1507  *
1508  * 4. Both threads are running
1509  * Both threads will be woken. The wrong thread will fail it's condition check
1510  * and execute another wait, while the correct thread will pass it's condition
1511  * check.
1512  *
1513  * @t: the task to set the thread ID for
1514  */
1515 int set_thread_tidr(struct task_struct *t)
1516 {
1517 	if (!cpu_has_feature(CPU_FTR_P9_TIDR))
1518 		return -EINVAL;
1519 
1520 	if (t != current)
1521 		return -EINVAL;
1522 
1523 	if (t->thread.tidr)
1524 		return 0;
1525 
1526 	t->thread.tidr = (u16)task_pid_nr(t);
1527 	mtspr(SPRN_TIDR, t->thread.tidr);
1528 
1529 	return 0;
1530 }
1531 EXPORT_SYMBOL_GPL(set_thread_tidr);
1532 
1533 #endif /* CONFIG_PPC64 */
1534 
1535 void
1536 release_thread(struct task_struct *t)
1537 {
1538 }
1539 
1540 /*
1541  * this gets called so that we can store coprocessor state into memory and
1542  * copy the current task into the new thread.
1543  */
1544 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1545 {
1546 	flush_all_to_thread(src);
1547 	/*
1548 	 * Flush TM state out so we can copy it.  __switch_to_tm() does this
1549 	 * flush but it removes the checkpointed state from the current CPU and
1550 	 * transitions the CPU out of TM mode.  Hence we need to call
1551 	 * tm_recheckpoint_new_task() (on the same task) to restore the
1552 	 * checkpointed state back and the TM mode.
1553 	 *
1554 	 * Can't pass dst because it isn't ready. Doesn't matter, passing
1555 	 * dst is only important for __switch_to()
1556 	 */
1557 	__switch_to_tm(src, src);
1558 
1559 	*dst = *src;
1560 
1561 	clear_task_ebb(dst);
1562 
1563 	return 0;
1564 }
1565 
1566 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1567 {
1568 #ifdef CONFIG_PPC_BOOK3S_64
1569 	unsigned long sp_vsid;
1570 	unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1571 
1572 	if (radix_enabled())
1573 		return;
1574 
1575 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1576 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1577 			<< SLB_VSID_SHIFT_1T;
1578 	else
1579 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1580 			<< SLB_VSID_SHIFT;
1581 	sp_vsid |= SLB_VSID_KERNEL | llp;
1582 	p->thread.ksp_vsid = sp_vsid;
1583 #endif
1584 }
1585 
1586 /*
1587  * Copy a thread..
1588  */
1589 
1590 /*
1591  * Copy architecture-specific thread state
1592  */
1593 int copy_thread_tls(unsigned long clone_flags, unsigned long usp,
1594 		unsigned long kthread_arg, struct task_struct *p,
1595 		unsigned long tls)
1596 {
1597 	struct pt_regs *childregs, *kregs;
1598 	extern void ret_from_fork(void);
1599 	extern void ret_from_kernel_thread(void);
1600 	void (*f)(void);
1601 	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1602 	struct thread_info *ti = task_thread_info(p);
1603 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1604 	int i;
1605 #endif
1606 
1607 	klp_init_thread_info(p);
1608 
1609 	/* Copy registers */
1610 	sp -= sizeof(struct pt_regs);
1611 	childregs = (struct pt_regs *) sp;
1612 	if (unlikely(p->flags & PF_KTHREAD)) {
1613 		/* kernel thread */
1614 		memset(childregs, 0, sizeof(struct pt_regs));
1615 		childregs->gpr[1] = sp + sizeof(struct pt_regs);
1616 		/* function */
1617 		if (usp)
1618 			childregs->gpr[14] = ppc_function_entry((void *)usp);
1619 #ifdef CONFIG_PPC64
1620 		clear_tsk_thread_flag(p, TIF_32BIT);
1621 		childregs->softe = IRQS_ENABLED;
1622 #endif
1623 		childregs->gpr[15] = kthread_arg;
1624 		p->thread.regs = NULL;	/* no user register state */
1625 		ti->flags |= _TIF_RESTOREALL;
1626 		f = ret_from_kernel_thread;
1627 	} else {
1628 		/* user thread */
1629 		struct pt_regs *regs = current_pt_regs();
1630 		CHECK_FULL_REGS(regs);
1631 		*childregs = *regs;
1632 		if (usp)
1633 			childregs->gpr[1] = usp;
1634 		p->thread.regs = childregs;
1635 		childregs->gpr[3] = 0;  /* Result from fork() */
1636 		if (clone_flags & CLONE_SETTLS) {
1637 			if (!is_32bit_task())
1638 				childregs->gpr[13] = tls;
1639 			else
1640 				childregs->gpr[2] = tls;
1641 		}
1642 
1643 		f = ret_from_fork;
1644 	}
1645 	childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1646 	sp -= STACK_FRAME_OVERHEAD;
1647 
1648 	/*
1649 	 * The way this works is that at some point in the future
1650 	 * some task will call _switch to switch to the new task.
1651 	 * That will pop off the stack frame created below and start
1652 	 * the new task running at ret_from_fork.  The new task will
1653 	 * do some house keeping and then return from the fork or clone
1654 	 * system call, using the stack frame created above.
1655 	 */
1656 	((unsigned long *)sp)[0] = 0;
1657 	sp -= sizeof(struct pt_regs);
1658 	kregs = (struct pt_regs *) sp;
1659 	sp -= STACK_FRAME_OVERHEAD;
1660 	p->thread.ksp = sp;
1661 #ifdef CONFIG_PPC32
1662 	p->thread.ksp_limit = (unsigned long)end_of_stack(p);
1663 #endif
1664 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1665 	for (i = 0; i < nr_wp_slots(); i++)
1666 		p->thread.ptrace_bps[i] = NULL;
1667 #endif
1668 
1669 	p->thread.fp_save_area = NULL;
1670 #ifdef CONFIG_ALTIVEC
1671 	p->thread.vr_save_area = NULL;
1672 #endif
1673 
1674 	setup_ksp_vsid(p, sp);
1675 
1676 #ifdef CONFIG_PPC64
1677 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1678 		p->thread.dscr_inherit = current->thread.dscr_inherit;
1679 		p->thread.dscr = mfspr(SPRN_DSCR);
1680 	}
1681 	if (cpu_has_feature(CPU_FTR_HAS_PPR))
1682 		childregs->ppr = DEFAULT_PPR;
1683 
1684 	p->thread.tidr = 0;
1685 #endif
1686 	kregs->nip = ppc_function_entry(f);
1687 	return 0;
1688 }
1689 
1690 void preload_new_slb_context(unsigned long start, unsigned long sp);
1691 
1692 /*
1693  * Set up a thread for executing a new program
1694  */
1695 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1696 {
1697 #ifdef CONFIG_PPC64
1698 	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
1699 
1700 #ifdef CONFIG_PPC_BOOK3S_64
1701 	if (!radix_enabled())
1702 		preload_new_slb_context(start, sp);
1703 #endif
1704 #endif
1705 
1706 	/*
1707 	 * If we exec out of a kernel thread then thread.regs will not be
1708 	 * set.  Do it now.
1709 	 */
1710 	if (!current->thread.regs) {
1711 		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1712 		current->thread.regs = regs - 1;
1713 	}
1714 
1715 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1716 	/*
1717 	 * Clear any transactional state, we're exec()ing. The cause is
1718 	 * not important as there will never be a recheckpoint so it's not
1719 	 * user visible.
1720 	 */
1721 	if (MSR_TM_SUSPENDED(mfmsr()))
1722 		tm_reclaim_current(0);
1723 #endif
1724 
1725 	memset(regs->gpr, 0, sizeof(regs->gpr));
1726 	regs->ctr = 0;
1727 	regs->link = 0;
1728 	regs->xer = 0;
1729 	regs->ccr = 0;
1730 	regs->gpr[1] = sp;
1731 
1732 	/*
1733 	 * We have just cleared all the nonvolatile GPRs, so make
1734 	 * FULL_REGS(regs) return true.  This is necessary to allow
1735 	 * ptrace to examine the thread immediately after exec.
1736 	 */
1737 	SET_FULL_REGS(regs);
1738 
1739 #ifdef CONFIG_PPC32
1740 	regs->mq = 0;
1741 	regs->nip = start;
1742 	regs->msr = MSR_USER;
1743 #else
1744 	if (!is_32bit_task()) {
1745 		unsigned long entry;
1746 
1747 		if (is_elf2_task()) {
1748 			/* Look ma, no function descriptors! */
1749 			entry = start;
1750 
1751 			/*
1752 			 * Ulrich says:
1753 			 *   The latest iteration of the ABI requires that when
1754 			 *   calling a function (at its global entry point),
1755 			 *   the caller must ensure r12 holds the entry point
1756 			 *   address (so that the function can quickly
1757 			 *   establish addressability).
1758 			 */
1759 			regs->gpr[12] = start;
1760 			/* Make sure that's restored on entry to userspace. */
1761 			set_thread_flag(TIF_RESTOREALL);
1762 		} else {
1763 			unsigned long toc;
1764 
1765 			/* start is a relocated pointer to the function
1766 			 * descriptor for the elf _start routine.  The first
1767 			 * entry in the function descriptor is the entry
1768 			 * address of _start and the second entry is the TOC
1769 			 * value we need to use.
1770 			 */
1771 			__get_user(entry, (unsigned long __user *)start);
1772 			__get_user(toc, (unsigned long __user *)start+1);
1773 
1774 			/* Check whether the e_entry function descriptor entries
1775 			 * need to be relocated before we can use them.
1776 			 */
1777 			if (load_addr != 0) {
1778 				entry += load_addr;
1779 				toc   += load_addr;
1780 			}
1781 			regs->gpr[2] = toc;
1782 		}
1783 		regs->nip = entry;
1784 		regs->msr = MSR_USER64;
1785 	} else {
1786 		regs->nip = start;
1787 		regs->gpr[2] = 0;
1788 		regs->msr = MSR_USER32;
1789 	}
1790 #endif
1791 #ifdef CONFIG_VSX
1792 	current->thread.used_vsr = 0;
1793 #endif
1794 	current->thread.load_slb = 0;
1795 	current->thread.load_fp = 0;
1796 	memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1797 	current->thread.fp_save_area = NULL;
1798 #ifdef CONFIG_ALTIVEC
1799 	memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1800 	current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1801 	current->thread.vr_save_area = NULL;
1802 	current->thread.vrsave = 0;
1803 	current->thread.used_vr = 0;
1804 	current->thread.load_vec = 0;
1805 #endif /* CONFIG_ALTIVEC */
1806 #ifdef CONFIG_SPE
1807 	memset(current->thread.evr, 0, sizeof(current->thread.evr));
1808 	current->thread.acc = 0;
1809 	current->thread.spefscr = 0;
1810 	current->thread.used_spe = 0;
1811 #endif /* CONFIG_SPE */
1812 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1813 	current->thread.tm_tfhar = 0;
1814 	current->thread.tm_texasr = 0;
1815 	current->thread.tm_tfiar = 0;
1816 	current->thread.load_tm = 0;
1817 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1818 
1819 	thread_pkey_regs_init(&current->thread);
1820 }
1821 EXPORT_SYMBOL(start_thread);
1822 
1823 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1824 		| PR_FP_EXC_RES | PR_FP_EXC_INV)
1825 
1826 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1827 {
1828 	struct pt_regs *regs = tsk->thread.regs;
1829 
1830 	/* This is a bit hairy.  If we are an SPE enabled  processor
1831 	 * (have embedded fp) we store the IEEE exception enable flags in
1832 	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
1833 	 * mode (asyn, precise, disabled) for 'Classic' FP. */
1834 	if (val & PR_FP_EXC_SW_ENABLE) {
1835 #ifdef CONFIG_SPE
1836 		if (cpu_has_feature(CPU_FTR_SPE)) {
1837 			/*
1838 			 * When the sticky exception bits are set
1839 			 * directly by userspace, it must call prctl
1840 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1841 			 * in the existing prctl settings) or
1842 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1843 			 * the bits being set).  <fenv.h> functions
1844 			 * saving and restoring the whole
1845 			 * floating-point environment need to do so
1846 			 * anyway to restore the prctl settings from
1847 			 * the saved environment.
1848 			 */
1849 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1850 			tsk->thread.fpexc_mode = val &
1851 				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1852 			return 0;
1853 		} else {
1854 			return -EINVAL;
1855 		}
1856 #else
1857 		return -EINVAL;
1858 #endif
1859 	}
1860 
1861 	/* on a CONFIG_SPE this does not hurt us.  The bits that
1862 	 * __pack_fe01 use do not overlap with bits used for
1863 	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1864 	 * on CONFIG_SPE implementations are reserved so writing to
1865 	 * them does not change anything */
1866 	if (val > PR_FP_EXC_PRECISE)
1867 		return -EINVAL;
1868 	tsk->thread.fpexc_mode = __pack_fe01(val);
1869 	if (regs != NULL && (regs->msr & MSR_FP) != 0)
1870 		regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1871 			| tsk->thread.fpexc_mode;
1872 	return 0;
1873 }
1874 
1875 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1876 {
1877 	unsigned int val;
1878 
1879 	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1880 #ifdef CONFIG_SPE
1881 		if (cpu_has_feature(CPU_FTR_SPE)) {
1882 			/*
1883 			 * When the sticky exception bits are set
1884 			 * directly by userspace, it must call prctl
1885 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1886 			 * in the existing prctl settings) or
1887 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1888 			 * the bits being set).  <fenv.h> functions
1889 			 * saving and restoring the whole
1890 			 * floating-point environment need to do so
1891 			 * anyway to restore the prctl settings from
1892 			 * the saved environment.
1893 			 */
1894 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1895 			val = tsk->thread.fpexc_mode;
1896 		} else
1897 			return -EINVAL;
1898 #else
1899 		return -EINVAL;
1900 #endif
1901 	else
1902 		val = __unpack_fe01(tsk->thread.fpexc_mode);
1903 	return put_user(val, (unsigned int __user *) adr);
1904 }
1905 
1906 int set_endian(struct task_struct *tsk, unsigned int val)
1907 {
1908 	struct pt_regs *regs = tsk->thread.regs;
1909 
1910 	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1911 	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1912 		return -EINVAL;
1913 
1914 	if (regs == NULL)
1915 		return -EINVAL;
1916 
1917 	if (val == PR_ENDIAN_BIG)
1918 		regs->msr &= ~MSR_LE;
1919 	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1920 		regs->msr |= MSR_LE;
1921 	else
1922 		return -EINVAL;
1923 
1924 	return 0;
1925 }
1926 
1927 int get_endian(struct task_struct *tsk, unsigned long adr)
1928 {
1929 	struct pt_regs *regs = tsk->thread.regs;
1930 	unsigned int val;
1931 
1932 	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1933 	    !cpu_has_feature(CPU_FTR_REAL_LE))
1934 		return -EINVAL;
1935 
1936 	if (regs == NULL)
1937 		return -EINVAL;
1938 
1939 	if (regs->msr & MSR_LE) {
1940 		if (cpu_has_feature(CPU_FTR_REAL_LE))
1941 			val = PR_ENDIAN_LITTLE;
1942 		else
1943 			val = PR_ENDIAN_PPC_LITTLE;
1944 	} else
1945 		val = PR_ENDIAN_BIG;
1946 
1947 	return put_user(val, (unsigned int __user *)adr);
1948 }
1949 
1950 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1951 {
1952 	tsk->thread.align_ctl = val;
1953 	return 0;
1954 }
1955 
1956 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1957 {
1958 	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1959 }
1960 
1961 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1962 				  unsigned long nbytes)
1963 {
1964 	unsigned long stack_page;
1965 	unsigned long cpu = task_cpu(p);
1966 
1967 	stack_page = (unsigned long)hardirq_ctx[cpu];
1968 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
1969 		return 1;
1970 
1971 	stack_page = (unsigned long)softirq_ctx[cpu];
1972 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
1973 		return 1;
1974 
1975 	return 0;
1976 }
1977 
1978 static inline int valid_emergency_stack(unsigned long sp, struct task_struct *p,
1979 					unsigned long nbytes)
1980 {
1981 #ifdef CONFIG_PPC64
1982 	unsigned long stack_page;
1983 	unsigned long cpu = task_cpu(p);
1984 
1985 	stack_page = (unsigned long)paca_ptrs[cpu]->emergency_sp - THREAD_SIZE;
1986 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
1987 		return 1;
1988 
1989 # ifdef CONFIG_PPC_BOOK3S_64
1990 	stack_page = (unsigned long)paca_ptrs[cpu]->nmi_emergency_sp - THREAD_SIZE;
1991 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
1992 		return 1;
1993 
1994 	stack_page = (unsigned long)paca_ptrs[cpu]->mc_emergency_sp - THREAD_SIZE;
1995 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
1996 		return 1;
1997 # endif
1998 #endif
1999 
2000 	return 0;
2001 }
2002 
2003 
2004 int validate_sp(unsigned long sp, struct task_struct *p,
2005 		       unsigned long nbytes)
2006 {
2007 	unsigned long stack_page = (unsigned long)task_stack_page(p);
2008 
2009 	if (sp < THREAD_SIZE)
2010 		return 0;
2011 
2012 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2013 		return 1;
2014 
2015 	if (valid_irq_stack(sp, p, nbytes))
2016 		return 1;
2017 
2018 	return valid_emergency_stack(sp, p, nbytes);
2019 }
2020 
2021 EXPORT_SYMBOL(validate_sp);
2022 
2023 static unsigned long __get_wchan(struct task_struct *p)
2024 {
2025 	unsigned long ip, sp;
2026 	int count = 0;
2027 
2028 	if (!p || p == current || p->state == TASK_RUNNING)
2029 		return 0;
2030 
2031 	sp = p->thread.ksp;
2032 	if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2033 		return 0;
2034 
2035 	do {
2036 		sp = *(unsigned long *)sp;
2037 		if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2038 		    p->state == TASK_RUNNING)
2039 			return 0;
2040 		if (count > 0) {
2041 			ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2042 			if (!in_sched_functions(ip))
2043 				return ip;
2044 		}
2045 	} while (count++ < 16);
2046 	return 0;
2047 }
2048 
2049 unsigned long get_wchan(struct task_struct *p)
2050 {
2051 	unsigned long ret;
2052 
2053 	if (!try_get_task_stack(p))
2054 		return 0;
2055 
2056 	ret = __get_wchan(p);
2057 
2058 	put_task_stack(p);
2059 
2060 	return ret;
2061 }
2062 
2063 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2064 
2065 void show_stack(struct task_struct *tsk, unsigned long *stack,
2066 		const char *loglvl)
2067 {
2068 	unsigned long sp, ip, lr, newsp;
2069 	int count = 0;
2070 	int firstframe = 1;
2071 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2072 	unsigned long ret_addr;
2073 	int ftrace_idx = 0;
2074 #endif
2075 
2076 	if (tsk == NULL)
2077 		tsk = current;
2078 
2079 	if (!try_get_task_stack(tsk))
2080 		return;
2081 
2082 	sp = (unsigned long) stack;
2083 	if (sp == 0) {
2084 		if (tsk == current)
2085 			sp = current_stack_frame();
2086 		else
2087 			sp = tsk->thread.ksp;
2088 	}
2089 
2090 	lr = 0;
2091 	printk("%sCall Trace:\n", loglvl);
2092 	do {
2093 		if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2094 			break;
2095 
2096 		stack = (unsigned long *) sp;
2097 		newsp = stack[0];
2098 		ip = stack[STACK_FRAME_LR_SAVE];
2099 		if (!firstframe || ip != lr) {
2100 			printk("%s["REG"] ["REG"] %pS",
2101 				loglvl, sp, ip, (void *)ip);
2102 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2103 			ret_addr = ftrace_graph_ret_addr(current,
2104 						&ftrace_idx, ip, stack);
2105 			if (ret_addr != ip)
2106 				pr_cont(" (%pS)", (void *)ret_addr);
2107 #endif
2108 			if (firstframe)
2109 				pr_cont(" (unreliable)");
2110 			pr_cont("\n");
2111 		}
2112 		firstframe = 0;
2113 
2114 		/*
2115 		 * See if this is an exception frame.
2116 		 * We look for the "regshere" marker in the current frame.
2117 		 */
2118 		if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
2119 		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2120 			struct pt_regs *regs = (struct pt_regs *)
2121 				(sp + STACK_FRAME_OVERHEAD);
2122 			lr = regs->link;
2123 			printk("%s--- interrupt: %lx at %pS\n    LR = %pS\n",
2124 			       loglvl, regs->trap,
2125 			       (void *)regs->nip, (void *)lr);
2126 			firstframe = 1;
2127 		}
2128 
2129 		sp = newsp;
2130 	} while (count++ < kstack_depth_to_print);
2131 
2132 	put_task_stack(tsk);
2133 }
2134 
2135 #ifdef CONFIG_PPC64
2136 /* Called with hard IRQs off */
2137 void notrace __ppc64_runlatch_on(void)
2138 {
2139 	struct thread_info *ti = current_thread_info();
2140 
2141 	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2142 		/*
2143 		 * Least significant bit (RUN) is the only writable bit of
2144 		 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2145 		 * earliest ISA where this is the case, but it's convenient.
2146 		 */
2147 		mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2148 	} else {
2149 		unsigned long ctrl;
2150 
2151 		/*
2152 		 * Some architectures (e.g., Cell) have writable fields other
2153 		 * than RUN, so do the read-modify-write.
2154 		 */
2155 		ctrl = mfspr(SPRN_CTRLF);
2156 		ctrl |= CTRL_RUNLATCH;
2157 		mtspr(SPRN_CTRLT, ctrl);
2158 	}
2159 
2160 	ti->local_flags |= _TLF_RUNLATCH;
2161 }
2162 
2163 /* Called with hard IRQs off */
2164 void notrace __ppc64_runlatch_off(void)
2165 {
2166 	struct thread_info *ti = current_thread_info();
2167 
2168 	ti->local_flags &= ~_TLF_RUNLATCH;
2169 
2170 	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2171 		mtspr(SPRN_CTRLT, 0);
2172 	} else {
2173 		unsigned long ctrl;
2174 
2175 		ctrl = mfspr(SPRN_CTRLF);
2176 		ctrl &= ~CTRL_RUNLATCH;
2177 		mtspr(SPRN_CTRLT, ctrl);
2178 	}
2179 }
2180 #endif /* CONFIG_PPC64 */
2181 
2182 unsigned long arch_align_stack(unsigned long sp)
2183 {
2184 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2185 		sp -= get_random_int() & ~PAGE_MASK;
2186 	return sp & ~0xf;
2187 }
2188 
2189 static inline unsigned long brk_rnd(void)
2190 {
2191         unsigned long rnd = 0;
2192 
2193 	/* 8MB for 32bit, 1GB for 64bit */
2194 	if (is_32bit_task())
2195 		rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2196 	else
2197 		rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2198 
2199 	return rnd << PAGE_SHIFT;
2200 }
2201 
2202 unsigned long arch_randomize_brk(struct mm_struct *mm)
2203 {
2204 	unsigned long base = mm->brk;
2205 	unsigned long ret;
2206 
2207 #ifdef CONFIG_PPC_BOOK3S_64
2208 	/*
2209 	 * If we are using 1TB segments and we are allowed to randomise
2210 	 * the heap, we can put it above 1TB so it is backed by a 1TB
2211 	 * segment. Otherwise the heap will be in the bottom 1TB
2212 	 * which always uses 256MB segments and this may result in a
2213 	 * performance penalty. We don't need to worry about radix. For
2214 	 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2215 	 */
2216 	if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2217 		base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2218 #endif
2219 
2220 	ret = PAGE_ALIGN(base + brk_rnd());
2221 
2222 	if (ret < mm->brk)
2223 		return mm->brk;
2224 
2225 	return ret;
2226 }
2227 
2228