xref: /openbmc/linux/arch/powerpc/kernel/process.c (revision 82e6fdd6)
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/sched/debug.h>
20 #include <linux/sched/task.h>
21 #include <linux/sched/task_stack.h>
22 #include <linux/kernel.h>
23 #include <linux/mm.h>
24 #include <linux/smp.h>
25 #include <linux/stddef.h>
26 #include <linux/unistd.h>
27 #include <linux/ptrace.h>
28 #include <linux/slab.h>
29 #include <linux/user.h>
30 #include <linux/elf.h>
31 #include <linux/prctl.h>
32 #include <linux/init_task.h>
33 #include <linux/export.h>
34 #include <linux/kallsyms.h>
35 #include <linux/mqueue.h>
36 #include <linux/hardirq.h>
37 #include <linux/utsname.h>
38 #include <linux/ftrace.h>
39 #include <linux/kernel_stat.h>
40 #include <linux/personality.h>
41 #include <linux/random.h>
42 #include <linux/hw_breakpoint.h>
43 #include <linux/uaccess.h>
44 #include <linux/elf-randomize.h>
45 #include <linux/pkeys.h>
46 
47 #include <asm/pgtable.h>
48 #include <asm/io.h>
49 #include <asm/processor.h>
50 #include <asm/mmu.h>
51 #include <asm/prom.h>
52 #include <asm/machdep.h>
53 #include <asm/time.h>
54 #include <asm/runlatch.h>
55 #include <asm/syscalls.h>
56 #include <asm/switch_to.h>
57 #include <asm/tm.h>
58 #include <asm/debug.h>
59 #ifdef CONFIG_PPC64
60 #include <asm/firmware.h>
61 #include <asm/hw_irq.h>
62 #endif
63 #include <asm/code-patching.h>
64 #include <asm/exec.h>
65 #include <asm/livepatch.h>
66 #include <asm/cpu_has_feature.h>
67 #include <asm/asm-prototypes.h>
68 
69 #include <linux/kprobes.h>
70 #include <linux/kdebug.h>
71 
72 /* Transactional Memory debug */
73 #ifdef TM_DEBUG_SW
74 #define TM_DEBUG(x...) printk(KERN_INFO x)
75 #else
76 #define TM_DEBUG(x...) do { } while(0)
77 #endif
78 
79 extern unsigned long _get_SP(void);
80 
81 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
82 /*
83  * Are we running in "Suspend disabled" mode? If so we have to block any
84  * sigreturn that would get us into suspended state, and we also warn in some
85  * other paths that we should never reach with suspend disabled.
86  */
87 bool tm_suspend_disabled __ro_after_init = false;
88 
89 static void check_if_tm_restore_required(struct task_struct *tsk)
90 {
91 	/*
92 	 * If we are saving the current thread's registers, and the
93 	 * thread is in a transactional state, set the TIF_RESTORE_TM
94 	 * bit so that we know to restore the registers before
95 	 * returning to userspace.
96 	 */
97 	if (tsk == current && tsk->thread.regs &&
98 	    MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
99 	    !test_thread_flag(TIF_RESTORE_TM)) {
100 		tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
101 		set_thread_flag(TIF_RESTORE_TM);
102 	}
103 }
104 
105 static inline bool msr_tm_active(unsigned long msr)
106 {
107 	return MSR_TM_ACTIVE(msr);
108 }
109 
110 static bool tm_active_with_fp(struct task_struct *tsk)
111 {
112 	return msr_tm_active(tsk->thread.regs->msr) &&
113 		(tsk->thread.ckpt_regs.msr & MSR_FP);
114 }
115 
116 static bool tm_active_with_altivec(struct task_struct *tsk)
117 {
118 	return msr_tm_active(tsk->thread.regs->msr) &&
119 		(tsk->thread.ckpt_regs.msr & MSR_VEC);
120 }
121 #else
122 static inline bool msr_tm_active(unsigned long msr) { return false; }
123 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
124 static inline bool tm_active_with_fp(struct task_struct *tsk) { return false; }
125 static inline bool tm_active_with_altivec(struct task_struct *tsk) { return false; }
126 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
127 
128 bool strict_msr_control;
129 EXPORT_SYMBOL(strict_msr_control);
130 
131 static int __init enable_strict_msr_control(char *str)
132 {
133 	strict_msr_control = true;
134 	pr_info("Enabling strict facility control\n");
135 
136 	return 0;
137 }
138 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
139 
140 unsigned long msr_check_and_set(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 	return newmsr;
156 }
157 
158 void __msr_check_and_clear(unsigned long bits)
159 {
160 	unsigned long oldmsr = mfmsr();
161 	unsigned long newmsr;
162 
163 	newmsr = oldmsr & ~bits;
164 
165 #ifdef CONFIG_VSX
166 	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
167 		newmsr &= ~MSR_VSX;
168 #endif
169 
170 	if (oldmsr != newmsr)
171 		mtmsr_isync(newmsr);
172 }
173 EXPORT_SYMBOL(__msr_check_and_clear);
174 
175 #ifdef CONFIG_PPC_FPU
176 void __giveup_fpu(struct task_struct *tsk)
177 {
178 	unsigned long msr;
179 
180 	save_fpu(tsk);
181 	msr = tsk->thread.regs->msr;
182 	msr &= ~MSR_FP;
183 #ifdef CONFIG_VSX
184 	if (cpu_has_feature(CPU_FTR_VSX))
185 		msr &= ~MSR_VSX;
186 #endif
187 	tsk->thread.regs->msr = msr;
188 }
189 
190 void giveup_fpu(struct task_struct *tsk)
191 {
192 	check_if_tm_restore_required(tsk);
193 
194 	msr_check_and_set(MSR_FP);
195 	__giveup_fpu(tsk);
196 	msr_check_and_clear(MSR_FP);
197 }
198 EXPORT_SYMBOL(giveup_fpu);
199 
200 /*
201  * Make sure the floating-point register state in the
202  * the thread_struct is up to date for task tsk.
203  */
204 void flush_fp_to_thread(struct task_struct *tsk)
205 {
206 	if (tsk->thread.regs) {
207 		/*
208 		 * We need to disable preemption here because if we didn't,
209 		 * another process could get scheduled after the regs->msr
210 		 * test but before we have finished saving the FP registers
211 		 * to the thread_struct.  That process could take over the
212 		 * FPU, and then when we get scheduled again we would store
213 		 * bogus values for the remaining FP registers.
214 		 */
215 		preempt_disable();
216 		if (tsk->thread.regs->msr & MSR_FP) {
217 			/*
218 			 * This should only ever be called for current or
219 			 * for a stopped child process.  Since we save away
220 			 * the FP register state on context switch,
221 			 * there is something wrong if a stopped child appears
222 			 * to still have its FP state in the CPU registers.
223 			 */
224 			BUG_ON(tsk != current);
225 			giveup_fpu(tsk);
226 		}
227 		preempt_enable();
228 	}
229 }
230 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
231 
232 void enable_kernel_fp(void)
233 {
234 	unsigned long cpumsr;
235 
236 	WARN_ON(preemptible());
237 
238 	cpumsr = msr_check_and_set(MSR_FP);
239 
240 	if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
241 		check_if_tm_restore_required(current);
242 		/*
243 		 * If a thread has already been reclaimed then the
244 		 * checkpointed registers are on the CPU but have definitely
245 		 * been saved by the reclaim code. Don't need to and *cannot*
246 		 * giveup as this would save  to the 'live' structure not the
247 		 * checkpointed structure.
248 		 */
249 		if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
250 			return;
251 		__giveup_fpu(current);
252 	}
253 }
254 EXPORT_SYMBOL(enable_kernel_fp);
255 
256 static int restore_fp(struct task_struct *tsk)
257 {
258 	if (tsk->thread.load_fp || tm_active_with_fp(tsk)) {
259 		load_fp_state(&current->thread.fp_state);
260 		current->thread.load_fp++;
261 		return 1;
262 	}
263 	return 0;
264 }
265 #else
266 static int restore_fp(struct task_struct *tsk) { return 0; }
267 #endif /* CONFIG_PPC_FPU */
268 
269 #ifdef CONFIG_ALTIVEC
270 #define loadvec(thr) ((thr).load_vec)
271 
272 static void __giveup_altivec(struct task_struct *tsk)
273 {
274 	unsigned long msr;
275 
276 	save_altivec(tsk);
277 	msr = tsk->thread.regs->msr;
278 	msr &= ~MSR_VEC;
279 #ifdef CONFIG_VSX
280 	if (cpu_has_feature(CPU_FTR_VSX))
281 		msr &= ~MSR_VSX;
282 #endif
283 	tsk->thread.regs->msr = msr;
284 }
285 
286 void giveup_altivec(struct task_struct *tsk)
287 {
288 	check_if_tm_restore_required(tsk);
289 
290 	msr_check_and_set(MSR_VEC);
291 	__giveup_altivec(tsk);
292 	msr_check_and_clear(MSR_VEC);
293 }
294 EXPORT_SYMBOL(giveup_altivec);
295 
296 void enable_kernel_altivec(void)
297 {
298 	unsigned long cpumsr;
299 
300 	WARN_ON(preemptible());
301 
302 	cpumsr = msr_check_and_set(MSR_VEC);
303 
304 	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
305 		check_if_tm_restore_required(current);
306 		/*
307 		 * If a thread has already been reclaimed then the
308 		 * checkpointed registers are on the CPU but have definitely
309 		 * been saved by the reclaim code. Don't need to and *cannot*
310 		 * giveup as this would save  to the 'live' structure not the
311 		 * checkpointed structure.
312 		 */
313 		if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
314 			return;
315 		__giveup_altivec(current);
316 	}
317 }
318 EXPORT_SYMBOL(enable_kernel_altivec);
319 
320 /*
321  * Make sure the VMX/Altivec register state in the
322  * the thread_struct is up to date for task tsk.
323  */
324 void flush_altivec_to_thread(struct task_struct *tsk)
325 {
326 	if (tsk->thread.regs) {
327 		preempt_disable();
328 		if (tsk->thread.regs->msr & MSR_VEC) {
329 			BUG_ON(tsk != current);
330 			giveup_altivec(tsk);
331 		}
332 		preempt_enable();
333 	}
334 }
335 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
336 
337 static int restore_altivec(struct task_struct *tsk)
338 {
339 	if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
340 		(tsk->thread.load_vec || tm_active_with_altivec(tsk))) {
341 		load_vr_state(&tsk->thread.vr_state);
342 		tsk->thread.used_vr = 1;
343 		tsk->thread.load_vec++;
344 
345 		return 1;
346 	}
347 	return 0;
348 }
349 #else
350 #define loadvec(thr) 0
351 static inline int restore_altivec(struct task_struct *tsk) { return 0; }
352 #endif /* CONFIG_ALTIVEC */
353 
354 #ifdef CONFIG_VSX
355 static void __giveup_vsx(struct task_struct *tsk)
356 {
357 	unsigned long msr = tsk->thread.regs->msr;
358 
359 	/*
360 	 * We should never be ssetting MSR_VSX without also setting
361 	 * MSR_FP and MSR_VEC
362 	 */
363 	WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
364 
365 	/* __giveup_fpu will clear MSR_VSX */
366 	if (msr & MSR_FP)
367 		__giveup_fpu(tsk);
368 	if (msr & MSR_VEC)
369 		__giveup_altivec(tsk);
370 }
371 
372 static void giveup_vsx(struct task_struct *tsk)
373 {
374 	check_if_tm_restore_required(tsk);
375 
376 	msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
377 	__giveup_vsx(tsk);
378 	msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
379 }
380 
381 void enable_kernel_vsx(void)
382 {
383 	unsigned long cpumsr;
384 
385 	WARN_ON(preemptible());
386 
387 	cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
388 
389 	if (current->thread.regs &&
390 	    (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
391 		check_if_tm_restore_required(current);
392 		/*
393 		 * If a thread has already been reclaimed then the
394 		 * checkpointed registers are on the CPU but have definitely
395 		 * been saved by the reclaim code. Don't need to and *cannot*
396 		 * giveup as this would save  to the 'live' structure not the
397 		 * checkpointed structure.
398 		 */
399 		if(!msr_tm_active(cpumsr) && msr_tm_active(current->thread.regs->msr))
400 			return;
401 		__giveup_vsx(current);
402 	}
403 }
404 EXPORT_SYMBOL(enable_kernel_vsx);
405 
406 void flush_vsx_to_thread(struct task_struct *tsk)
407 {
408 	if (tsk->thread.regs) {
409 		preempt_disable();
410 		if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
411 			BUG_ON(tsk != current);
412 			giveup_vsx(tsk);
413 		}
414 		preempt_enable();
415 	}
416 }
417 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
418 
419 static int restore_vsx(struct task_struct *tsk)
420 {
421 	if (cpu_has_feature(CPU_FTR_VSX)) {
422 		tsk->thread.used_vsr = 1;
423 		return 1;
424 	}
425 
426 	return 0;
427 }
428 #else
429 static inline int restore_vsx(struct task_struct *tsk) { return 0; }
430 #endif /* CONFIG_VSX */
431 
432 #ifdef CONFIG_SPE
433 void giveup_spe(struct task_struct *tsk)
434 {
435 	check_if_tm_restore_required(tsk);
436 
437 	msr_check_and_set(MSR_SPE);
438 	__giveup_spe(tsk);
439 	msr_check_and_clear(MSR_SPE);
440 }
441 EXPORT_SYMBOL(giveup_spe);
442 
443 void enable_kernel_spe(void)
444 {
445 	WARN_ON(preemptible());
446 
447 	msr_check_and_set(MSR_SPE);
448 
449 	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
450 		check_if_tm_restore_required(current);
451 		__giveup_spe(current);
452 	}
453 }
454 EXPORT_SYMBOL(enable_kernel_spe);
455 
456 void flush_spe_to_thread(struct task_struct *tsk)
457 {
458 	if (tsk->thread.regs) {
459 		preempt_disable();
460 		if (tsk->thread.regs->msr & MSR_SPE) {
461 			BUG_ON(tsk != current);
462 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
463 			giveup_spe(tsk);
464 		}
465 		preempt_enable();
466 	}
467 }
468 #endif /* CONFIG_SPE */
469 
470 static unsigned long msr_all_available;
471 
472 static int __init init_msr_all_available(void)
473 {
474 #ifdef CONFIG_PPC_FPU
475 	msr_all_available |= MSR_FP;
476 #endif
477 #ifdef CONFIG_ALTIVEC
478 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
479 		msr_all_available |= MSR_VEC;
480 #endif
481 #ifdef CONFIG_VSX
482 	if (cpu_has_feature(CPU_FTR_VSX))
483 		msr_all_available |= MSR_VSX;
484 #endif
485 #ifdef CONFIG_SPE
486 	if (cpu_has_feature(CPU_FTR_SPE))
487 		msr_all_available |= MSR_SPE;
488 #endif
489 
490 	return 0;
491 }
492 early_initcall(init_msr_all_available);
493 
494 void giveup_all(struct task_struct *tsk)
495 {
496 	unsigned long usermsr;
497 
498 	if (!tsk->thread.regs)
499 		return;
500 
501 	usermsr = tsk->thread.regs->msr;
502 
503 	if ((usermsr & msr_all_available) == 0)
504 		return;
505 
506 	msr_check_and_set(msr_all_available);
507 	check_if_tm_restore_required(tsk);
508 
509 	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
510 
511 #ifdef CONFIG_PPC_FPU
512 	if (usermsr & MSR_FP)
513 		__giveup_fpu(tsk);
514 #endif
515 #ifdef CONFIG_ALTIVEC
516 	if (usermsr & MSR_VEC)
517 		__giveup_altivec(tsk);
518 #endif
519 #ifdef CONFIG_SPE
520 	if (usermsr & MSR_SPE)
521 		__giveup_spe(tsk);
522 #endif
523 
524 	msr_check_and_clear(msr_all_available);
525 }
526 EXPORT_SYMBOL(giveup_all);
527 
528 void restore_math(struct pt_regs *regs)
529 {
530 	unsigned long msr;
531 
532 	if (!msr_tm_active(regs->msr) &&
533 		!current->thread.load_fp && !loadvec(current->thread))
534 		return;
535 
536 	msr = regs->msr;
537 	msr_check_and_set(msr_all_available);
538 
539 	/*
540 	 * Only reload if the bit is not set in the user MSR, the bit BEING set
541 	 * indicates that the registers are hot
542 	 */
543 	if ((!(msr & MSR_FP)) && restore_fp(current))
544 		msr |= MSR_FP | current->thread.fpexc_mode;
545 
546 	if ((!(msr & MSR_VEC)) && restore_altivec(current))
547 		msr |= MSR_VEC;
548 
549 	if ((msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC) &&
550 			restore_vsx(current)) {
551 		msr |= MSR_VSX;
552 	}
553 
554 	msr_check_and_clear(msr_all_available);
555 
556 	regs->msr = msr;
557 }
558 
559 void save_all(struct task_struct *tsk)
560 {
561 	unsigned long usermsr;
562 
563 	if (!tsk->thread.regs)
564 		return;
565 
566 	usermsr = tsk->thread.regs->msr;
567 
568 	if ((usermsr & msr_all_available) == 0)
569 		return;
570 
571 	msr_check_and_set(msr_all_available);
572 
573 	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
574 
575 	if (usermsr & MSR_FP)
576 		save_fpu(tsk);
577 
578 	if (usermsr & MSR_VEC)
579 		save_altivec(tsk);
580 
581 	if (usermsr & MSR_SPE)
582 		__giveup_spe(tsk);
583 
584 	msr_check_and_clear(msr_all_available);
585 }
586 
587 void flush_all_to_thread(struct task_struct *tsk)
588 {
589 	if (tsk->thread.regs) {
590 		preempt_disable();
591 		BUG_ON(tsk != current);
592 		save_all(tsk);
593 
594 #ifdef CONFIG_SPE
595 		if (tsk->thread.regs->msr & MSR_SPE)
596 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
597 #endif
598 
599 		preempt_enable();
600 	}
601 }
602 EXPORT_SYMBOL(flush_all_to_thread);
603 
604 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
605 void do_send_trap(struct pt_regs *regs, unsigned long address,
606 		  unsigned long error_code, int breakpt)
607 {
608 	current->thread.trap_nr = TRAP_HWBKPT;
609 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
610 			11, SIGSEGV) == NOTIFY_STOP)
611 		return;
612 
613 	/* Deliver the signal to userspace */
614 	force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
615 				    (void __user *)address);
616 }
617 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
618 void do_break (struct pt_regs *regs, unsigned long address,
619 		    unsigned long error_code)
620 {
621 	siginfo_t info;
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 	/* Clear the breakpoint */
632 	hw_breakpoint_disable();
633 
634 	/* Deliver the signal to userspace */
635 	info.si_signo = SIGTRAP;
636 	info.si_errno = 0;
637 	info.si_code = TRAP_HWBKPT;
638 	info.si_addr = (void __user *)address;
639 	force_sig_info(SIGTRAP, &info, current);
640 }
641 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
642 
643 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
644 
645 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
646 /*
647  * Set the debug registers back to their default "safe" values.
648  */
649 static void set_debug_reg_defaults(struct thread_struct *thread)
650 {
651 	thread->debug.iac1 = thread->debug.iac2 = 0;
652 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
653 	thread->debug.iac3 = thread->debug.iac4 = 0;
654 #endif
655 	thread->debug.dac1 = thread->debug.dac2 = 0;
656 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
657 	thread->debug.dvc1 = thread->debug.dvc2 = 0;
658 #endif
659 	thread->debug.dbcr0 = 0;
660 #ifdef CONFIG_BOOKE
661 	/*
662 	 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
663 	 */
664 	thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
665 			DBCR1_IAC3US | DBCR1_IAC4US;
666 	/*
667 	 * Force Data Address Compare User/Supervisor bits to be User-only
668 	 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
669 	 */
670 	thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
671 #else
672 	thread->debug.dbcr1 = 0;
673 #endif
674 }
675 
676 static void prime_debug_regs(struct debug_reg *debug)
677 {
678 	/*
679 	 * We could have inherited MSR_DE from userspace, since
680 	 * it doesn't get cleared on exception entry.  Make sure
681 	 * MSR_DE is clear before we enable any debug events.
682 	 */
683 	mtmsr(mfmsr() & ~MSR_DE);
684 
685 	mtspr(SPRN_IAC1, debug->iac1);
686 	mtspr(SPRN_IAC2, debug->iac2);
687 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
688 	mtspr(SPRN_IAC3, debug->iac3);
689 	mtspr(SPRN_IAC4, debug->iac4);
690 #endif
691 	mtspr(SPRN_DAC1, debug->dac1);
692 	mtspr(SPRN_DAC2, debug->dac2);
693 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
694 	mtspr(SPRN_DVC1, debug->dvc1);
695 	mtspr(SPRN_DVC2, debug->dvc2);
696 #endif
697 	mtspr(SPRN_DBCR0, debug->dbcr0);
698 	mtspr(SPRN_DBCR1, debug->dbcr1);
699 #ifdef CONFIG_BOOKE
700 	mtspr(SPRN_DBCR2, debug->dbcr2);
701 #endif
702 }
703 /*
704  * Unless neither the old or new thread are making use of the
705  * debug registers, set the debug registers from the values
706  * stored in the new thread.
707  */
708 void switch_booke_debug_regs(struct debug_reg *new_debug)
709 {
710 	if ((current->thread.debug.dbcr0 & DBCR0_IDM)
711 		|| (new_debug->dbcr0 & DBCR0_IDM))
712 			prime_debug_regs(new_debug);
713 }
714 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
715 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
716 #ifndef CONFIG_HAVE_HW_BREAKPOINT
717 static void set_debug_reg_defaults(struct thread_struct *thread)
718 {
719 	thread->hw_brk.address = 0;
720 	thread->hw_brk.type = 0;
721 	set_breakpoint(&thread->hw_brk);
722 }
723 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
724 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
725 
726 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
727 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
728 {
729 	mtspr(SPRN_DAC1, dabr);
730 #ifdef CONFIG_PPC_47x
731 	isync();
732 #endif
733 	return 0;
734 }
735 #elif defined(CONFIG_PPC_BOOK3S)
736 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
737 {
738 	mtspr(SPRN_DABR, dabr);
739 	if (cpu_has_feature(CPU_FTR_DABRX))
740 		mtspr(SPRN_DABRX, dabrx);
741 	return 0;
742 }
743 #elif defined(CONFIG_PPC_8xx)
744 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
745 {
746 	unsigned long addr = dabr & ~HW_BRK_TYPE_DABR;
747 	unsigned long lctrl1 = 0x90000000; /* compare type: equal on E & F */
748 	unsigned long lctrl2 = 0x8e000002; /* watchpoint 1 on cmp E | F */
749 
750 	if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
751 		lctrl1 |= 0xa0000;
752 	else if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
753 		lctrl1 |= 0xf0000;
754 	else if ((dabr & HW_BRK_TYPE_RDWR) == 0)
755 		lctrl2 = 0;
756 
757 	mtspr(SPRN_LCTRL2, 0);
758 	mtspr(SPRN_CMPE, addr);
759 	mtspr(SPRN_CMPF, addr + 4);
760 	mtspr(SPRN_LCTRL1, lctrl1);
761 	mtspr(SPRN_LCTRL2, lctrl2);
762 
763 	return 0;
764 }
765 #else
766 static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
767 {
768 	return -EINVAL;
769 }
770 #endif
771 
772 static inline int set_dabr(struct arch_hw_breakpoint *brk)
773 {
774 	unsigned long dabr, dabrx;
775 
776 	dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
777 	dabrx = ((brk->type >> 3) & 0x7);
778 
779 	if (ppc_md.set_dabr)
780 		return ppc_md.set_dabr(dabr, dabrx);
781 
782 	return __set_dabr(dabr, dabrx);
783 }
784 
785 static inline int set_dawr(struct arch_hw_breakpoint *brk)
786 {
787 	unsigned long dawr, dawrx, mrd;
788 
789 	dawr = brk->address;
790 
791 	dawrx  = (brk->type & (HW_BRK_TYPE_READ | HW_BRK_TYPE_WRITE)) \
792 		                   << (63 - 58); //* read/write bits */
793 	dawrx |= ((brk->type & (HW_BRK_TYPE_TRANSLATE)) >> 2) \
794 		                   << (63 - 59); //* translate */
795 	dawrx |= (brk->type & (HW_BRK_TYPE_PRIV_ALL)) \
796 		                   >> 3; //* PRIM bits */
797 	/* dawr length is stored in field MDR bits 48:53.  Matches range in
798 	   doublewords (64 bits) baised by -1 eg. 0b000000=1DW and
799 	   0b111111=64DW.
800 	   brk->len is in bytes.
801 	   This aligns up to double word size, shifts and does the bias.
802 	*/
803 	mrd = ((brk->len + 7) >> 3) - 1;
804 	dawrx |= (mrd & 0x3f) << (63 - 53);
805 
806 	if (ppc_md.set_dawr)
807 		return ppc_md.set_dawr(dawr, dawrx);
808 	mtspr(SPRN_DAWR, dawr);
809 	mtspr(SPRN_DAWRX, dawrx);
810 	return 0;
811 }
812 
813 void __set_breakpoint(struct arch_hw_breakpoint *brk)
814 {
815 	memcpy(this_cpu_ptr(&current_brk), brk, sizeof(*brk));
816 
817 	if (cpu_has_feature(CPU_FTR_DAWR))
818 		set_dawr(brk);
819 	else
820 		set_dabr(brk);
821 }
822 
823 void set_breakpoint(struct arch_hw_breakpoint *brk)
824 {
825 	preempt_disable();
826 	__set_breakpoint(brk);
827 	preempt_enable();
828 }
829 
830 #ifdef CONFIG_PPC64
831 DEFINE_PER_CPU(struct cpu_usage, cpu_usage_array);
832 #endif
833 
834 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
835 			      struct arch_hw_breakpoint *b)
836 {
837 	if (a->address != b->address)
838 		return false;
839 	if (a->type != b->type)
840 		return false;
841 	if (a->len != b->len)
842 		return false;
843 	return true;
844 }
845 
846 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
847 
848 static inline bool tm_enabled(struct task_struct *tsk)
849 {
850 	return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
851 }
852 
853 static void tm_reclaim_thread(struct thread_struct *thr,
854 			      struct thread_info *ti, uint8_t cause)
855 {
856 	/*
857 	 * Use the current MSR TM suspended bit to track if we have
858 	 * checkpointed state outstanding.
859 	 * On signal delivery, we'd normally reclaim the checkpointed
860 	 * state to obtain stack pointer (see:get_tm_stackpointer()).
861 	 * This will then directly return to userspace without going
862 	 * through __switch_to(). However, if the stack frame is bad,
863 	 * we need to exit this thread which calls __switch_to() which
864 	 * will again attempt to reclaim the already saved tm state.
865 	 * Hence we need to check that we've not already reclaimed
866 	 * this state.
867 	 * We do this using the current MSR, rather tracking it in
868 	 * some specific thread_struct bit, as it has the additional
869 	 * benefit of checking for a potential TM bad thing exception.
870 	 */
871 	if (!MSR_TM_SUSPENDED(mfmsr()))
872 		return;
873 
874 	giveup_all(container_of(thr, struct task_struct, thread));
875 
876 	tm_reclaim(thr, cause);
877 
878 	/*
879 	 * If we are in a transaction and FP is off then we can't have
880 	 * used FP inside that transaction. Hence the checkpointed
881 	 * state is the same as the live state. We need to copy the
882 	 * live state to the checkpointed state so that when the
883 	 * transaction is restored, the checkpointed state is correct
884 	 * and the aborted transaction sees the correct state. We use
885 	 * ckpt_regs.msr here as that's what tm_reclaim will use to
886 	 * determine if it's going to write the checkpointed state or
887 	 * not. So either this will write the checkpointed registers,
888 	 * or reclaim will. Similarly for VMX.
889 	 */
890 	if ((thr->ckpt_regs.msr & MSR_FP) == 0)
891 		memcpy(&thr->ckfp_state, &thr->fp_state,
892 		       sizeof(struct thread_fp_state));
893 	if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
894 		memcpy(&thr->ckvr_state, &thr->vr_state,
895 		       sizeof(struct thread_vr_state));
896 }
897 
898 void tm_reclaim_current(uint8_t cause)
899 {
900 	tm_enable();
901 	tm_reclaim_thread(&current->thread, current_thread_info(), cause);
902 }
903 
904 static inline void tm_reclaim_task(struct task_struct *tsk)
905 {
906 	/* We have to work out if we're switching from/to a task that's in the
907 	 * middle of a transaction.
908 	 *
909 	 * In switching we need to maintain a 2nd register state as
910 	 * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
911 	 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
912 	 * ckvr_state
913 	 *
914 	 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
915 	 */
916 	struct thread_struct *thr = &tsk->thread;
917 
918 	if (!thr->regs)
919 		return;
920 
921 	if (!MSR_TM_ACTIVE(thr->regs->msr))
922 		goto out_and_saveregs;
923 
924 	WARN_ON(tm_suspend_disabled);
925 
926 	TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
927 		 "ccr=%lx, msr=%lx, trap=%lx)\n",
928 		 tsk->pid, thr->regs->nip,
929 		 thr->regs->ccr, thr->regs->msr,
930 		 thr->regs->trap);
931 
932 	tm_reclaim_thread(thr, task_thread_info(tsk), TM_CAUSE_RESCHED);
933 
934 	TM_DEBUG("--- tm_reclaim on pid %d complete\n",
935 		 tsk->pid);
936 
937 out_and_saveregs:
938 	/* Always save the regs here, even if a transaction's not active.
939 	 * This context-switches a thread's TM info SPRs.  We do it here to
940 	 * be consistent with the restore path (in recheckpoint) which
941 	 * cannot happen later in _switch().
942 	 */
943 	tm_save_sprs(thr);
944 }
945 
946 extern void __tm_recheckpoint(struct thread_struct *thread);
947 
948 void tm_recheckpoint(struct thread_struct *thread)
949 {
950 	unsigned long flags;
951 
952 	if (!(thread->regs->msr & MSR_TM))
953 		return;
954 
955 	/* We really can't be interrupted here as the TEXASR registers can't
956 	 * change and later in the trecheckpoint code, we have a userspace R1.
957 	 * So let's hard disable over this region.
958 	 */
959 	local_irq_save(flags);
960 	hard_irq_disable();
961 
962 	/* The TM SPRs are restored here, so that TEXASR.FS can be set
963 	 * before the trecheckpoint and no explosion occurs.
964 	 */
965 	tm_restore_sprs(thread);
966 
967 	__tm_recheckpoint(thread);
968 
969 	local_irq_restore(flags);
970 }
971 
972 static inline void tm_recheckpoint_new_task(struct task_struct *new)
973 {
974 	if (!cpu_has_feature(CPU_FTR_TM))
975 		return;
976 
977 	/* Recheckpoint the registers of the thread we're about to switch to.
978 	 *
979 	 * If the task was using FP, we non-lazily reload both the original and
980 	 * the speculative FP register states.  This is because the kernel
981 	 * doesn't see if/when a TM rollback occurs, so if we take an FP
982 	 * unavailable later, we are unable to determine which set of FP regs
983 	 * need to be restored.
984 	 */
985 	if (!tm_enabled(new))
986 		return;
987 
988 	if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
989 		tm_restore_sprs(&new->thread);
990 		return;
991 	}
992 	/* Recheckpoint to restore original checkpointed register state. */
993 	TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
994 		 new->pid, new->thread.regs->msr);
995 
996 	tm_recheckpoint(&new->thread);
997 
998 	/*
999 	 * The checkpointed state has been restored but the live state has
1000 	 * not, ensure all the math functionality is turned off to trigger
1001 	 * restore_math() to reload.
1002 	 */
1003 	new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1004 
1005 	TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1006 		 "(kernel msr 0x%lx)\n",
1007 		 new->pid, mfmsr());
1008 }
1009 
1010 static inline void __switch_to_tm(struct task_struct *prev,
1011 		struct task_struct *new)
1012 {
1013 	if (cpu_has_feature(CPU_FTR_TM)) {
1014 		if (tm_enabled(prev) || tm_enabled(new))
1015 			tm_enable();
1016 
1017 		if (tm_enabled(prev)) {
1018 			prev->thread.load_tm++;
1019 			tm_reclaim_task(prev);
1020 			if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1021 				prev->thread.regs->msr &= ~MSR_TM;
1022 		}
1023 
1024 		tm_recheckpoint_new_task(new);
1025 	}
1026 }
1027 
1028 /*
1029  * This is called if we are on the way out to userspace and the
1030  * TIF_RESTORE_TM flag is set.  It checks if we need to reload
1031  * FP and/or vector state and does so if necessary.
1032  * If userspace is inside a transaction (whether active or
1033  * suspended) and FP/VMX/VSX instructions have ever been enabled
1034  * inside that transaction, then we have to keep them enabled
1035  * and keep the FP/VMX/VSX state loaded while ever the transaction
1036  * continues.  The reason is that if we didn't, and subsequently
1037  * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1038  * we don't know whether it's the same transaction, and thus we
1039  * don't know which of the checkpointed state and the transactional
1040  * state to use.
1041  */
1042 void restore_tm_state(struct pt_regs *regs)
1043 {
1044 	unsigned long msr_diff;
1045 
1046 	/*
1047 	 * This is the only moment we should clear TIF_RESTORE_TM as
1048 	 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1049 	 * again, anything else could lead to an incorrect ckpt_msr being
1050 	 * saved and therefore incorrect signal contexts.
1051 	 */
1052 	clear_thread_flag(TIF_RESTORE_TM);
1053 	if (!MSR_TM_ACTIVE(regs->msr))
1054 		return;
1055 
1056 	msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1057 	msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1058 
1059 	/* Ensure that restore_math() will restore */
1060 	if (msr_diff & MSR_FP)
1061 		current->thread.load_fp = 1;
1062 #ifdef CONFIG_ALTIVEC
1063 	if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1064 		current->thread.load_vec = 1;
1065 #endif
1066 	restore_math(regs);
1067 
1068 	regs->msr |= msr_diff;
1069 }
1070 
1071 #else
1072 #define tm_recheckpoint_new_task(new)
1073 #define __switch_to_tm(prev, new)
1074 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1075 
1076 static inline void save_sprs(struct thread_struct *t)
1077 {
1078 #ifdef CONFIG_ALTIVEC
1079 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
1080 		t->vrsave = mfspr(SPRN_VRSAVE);
1081 #endif
1082 #ifdef CONFIG_PPC_BOOK3S_64
1083 	if (cpu_has_feature(CPU_FTR_DSCR))
1084 		t->dscr = mfspr(SPRN_DSCR);
1085 
1086 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1087 		t->bescr = mfspr(SPRN_BESCR);
1088 		t->ebbhr = mfspr(SPRN_EBBHR);
1089 		t->ebbrr = mfspr(SPRN_EBBRR);
1090 
1091 		t->fscr = mfspr(SPRN_FSCR);
1092 
1093 		/*
1094 		 * Note that the TAR is not available for use in the kernel.
1095 		 * (To provide this, the TAR should be backed up/restored on
1096 		 * exception entry/exit instead, and be in pt_regs.  FIXME,
1097 		 * this should be in pt_regs anyway (for debug).)
1098 		 */
1099 		t->tar = mfspr(SPRN_TAR);
1100 	}
1101 #endif
1102 
1103 	thread_pkey_regs_save(t);
1104 }
1105 
1106 static inline void restore_sprs(struct thread_struct *old_thread,
1107 				struct thread_struct *new_thread)
1108 {
1109 #ifdef CONFIG_ALTIVEC
1110 	if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1111 	    old_thread->vrsave != new_thread->vrsave)
1112 		mtspr(SPRN_VRSAVE, new_thread->vrsave);
1113 #endif
1114 #ifdef CONFIG_PPC_BOOK3S_64
1115 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1116 		u64 dscr = get_paca()->dscr_default;
1117 		if (new_thread->dscr_inherit)
1118 			dscr = new_thread->dscr;
1119 
1120 		if (old_thread->dscr != dscr)
1121 			mtspr(SPRN_DSCR, dscr);
1122 	}
1123 
1124 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1125 		if (old_thread->bescr != new_thread->bescr)
1126 			mtspr(SPRN_BESCR, new_thread->bescr);
1127 		if (old_thread->ebbhr != new_thread->ebbhr)
1128 			mtspr(SPRN_EBBHR, new_thread->ebbhr);
1129 		if (old_thread->ebbrr != new_thread->ebbrr)
1130 			mtspr(SPRN_EBBRR, new_thread->ebbrr);
1131 
1132 		if (old_thread->fscr != new_thread->fscr)
1133 			mtspr(SPRN_FSCR, new_thread->fscr);
1134 
1135 		if (old_thread->tar != new_thread->tar)
1136 			mtspr(SPRN_TAR, new_thread->tar);
1137 	}
1138 
1139 	if (cpu_has_feature(CPU_FTR_ARCH_300) &&
1140 	    old_thread->tidr != new_thread->tidr)
1141 		mtspr(SPRN_TIDR, new_thread->tidr);
1142 #endif
1143 
1144 	thread_pkey_regs_restore(new_thread, old_thread);
1145 }
1146 
1147 #ifdef CONFIG_PPC_BOOK3S_64
1148 #define CP_SIZE 128
1149 static const u8 dummy_copy_buffer[CP_SIZE] __attribute__((aligned(CP_SIZE)));
1150 #endif
1151 
1152 struct task_struct *__switch_to(struct task_struct *prev,
1153 	struct task_struct *new)
1154 {
1155 	struct thread_struct *new_thread, *old_thread;
1156 	struct task_struct *last;
1157 #ifdef CONFIG_PPC_BOOK3S_64
1158 	struct ppc64_tlb_batch *batch;
1159 #endif
1160 
1161 	new_thread = &new->thread;
1162 	old_thread = &current->thread;
1163 
1164 	WARN_ON(!irqs_disabled());
1165 
1166 #ifdef CONFIG_PPC64
1167 	/*
1168 	 * Collect processor utilization data per process
1169 	 */
1170 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
1171 		struct cpu_usage *cu = this_cpu_ptr(&cpu_usage_array);
1172 		long unsigned start_tb, current_tb;
1173 		start_tb = old_thread->start_tb;
1174 		cu->current_tb = current_tb = mfspr(SPRN_PURR);
1175 		old_thread->accum_tb += (current_tb - start_tb);
1176 		new_thread->start_tb = current_tb;
1177 	}
1178 #endif /* CONFIG_PPC64 */
1179 
1180 #ifdef CONFIG_PPC_BOOK3S_64
1181 	batch = this_cpu_ptr(&ppc64_tlb_batch);
1182 	if (batch->active) {
1183 		current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1184 		if (batch->index)
1185 			__flush_tlb_pending(batch);
1186 		batch->active = 0;
1187 	}
1188 #endif /* CONFIG_PPC_BOOK3S_64 */
1189 
1190 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1191 	switch_booke_debug_regs(&new->thread.debug);
1192 #else
1193 /*
1194  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1195  * schedule DABR
1196  */
1197 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1198 	if (unlikely(!hw_brk_match(this_cpu_ptr(&current_brk), &new->thread.hw_brk)))
1199 		__set_breakpoint(&new->thread.hw_brk);
1200 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1201 #endif
1202 
1203 	/*
1204 	 * We need to save SPRs before treclaim/trecheckpoint as these will
1205 	 * change a number of them.
1206 	 */
1207 	save_sprs(&prev->thread);
1208 
1209 	/* Save FPU, Altivec, VSX and SPE state */
1210 	giveup_all(prev);
1211 
1212 	__switch_to_tm(prev, new);
1213 
1214 	if (!radix_enabled()) {
1215 		/*
1216 		 * We can't take a PMU exception inside _switch() since there
1217 		 * is a window where the kernel stack SLB and the kernel stack
1218 		 * are out of sync. Hard disable here.
1219 		 */
1220 		hard_irq_disable();
1221 	}
1222 
1223 	/*
1224 	 * Call restore_sprs() before calling _switch(). If we move it after
1225 	 * _switch() then we miss out on calling it for new tasks. The reason
1226 	 * for this is we manually create a stack frame for new tasks that
1227 	 * directly returns through ret_from_fork() or
1228 	 * ret_from_kernel_thread(). See copy_thread() for details.
1229 	 */
1230 	restore_sprs(old_thread, new_thread);
1231 
1232 	last = _switch(old_thread, new_thread);
1233 
1234 #ifdef CONFIG_PPC_BOOK3S_64
1235 	if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1236 		current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1237 		batch = this_cpu_ptr(&ppc64_tlb_batch);
1238 		batch->active = 1;
1239 	}
1240 
1241 	if (current_thread_info()->task->thread.regs) {
1242 		restore_math(current_thread_info()->task->thread.regs);
1243 
1244 		/*
1245 		 * The copy-paste buffer can only store into foreign real
1246 		 * addresses, so unprivileged processes can not see the
1247 		 * data or use it in any way unless they have foreign real
1248 		 * mappings. If the new process has the foreign real address
1249 		 * mappings, we must issue a cp_abort to clear any state and
1250 		 * prevent snooping, corruption or a covert channel.
1251 		 *
1252 		 * DD1 allows paste into normal system memory so we do an
1253 		 * unpaired copy, rather than cp_abort, to clear the buffer,
1254 		 * since cp_abort is quite expensive.
1255 		 */
1256 		if (current_thread_info()->task->thread.used_vas) {
1257 			asm volatile(PPC_CP_ABORT);
1258 		} else if (cpu_has_feature(CPU_FTR_POWER9_DD1)) {
1259 			asm volatile(PPC_COPY(%0, %1)
1260 					: : "r"(dummy_copy_buffer), "r"(0));
1261 		}
1262 	}
1263 #endif /* CONFIG_PPC_BOOK3S_64 */
1264 
1265 	return last;
1266 }
1267 
1268 static int instructions_to_print = 16;
1269 
1270 static void show_instructions(struct pt_regs *regs)
1271 {
1272 	int i;
1273 	unsigned long pc = regs->nip - (instructions_to_print * 3 / 4 *
1274 			sizeof(int));
1275 
1276 	printk("Instruction dump:");
1277 
1278 	for (i = 0; i < instructions_to_print; i++) {
1279 		int instr;
1280 
1281 		if (!(i % 8))
1282 			pr_cont("\n");
1283 
1284 #if !defined(CONFIG_BOOKE)
1285 		/* If executing with the IMMU off, adjust pc rather
1286 		 * than print XXXXXXXX.
1287 		 */
1288 		if (!(regs->msr & MSR_IR))
1289 			pc = (unsigned long)phys_to_virt(pc);
1290 #endif
1291 
1292 		if (!__kernel_text_address(pc) ||
1293 		     probe_kernel_address((unsigned int __user *)pc, instr)) {
1294 			pr_cont("XXXXXXXX ");
1295 		} else {
1296 			if (regs->nip == pc)
1297 				pr_cont("<%08x> ", instr);
1298 			else
1299 				pr_cont("%08x ", instr);
1300 		}
1301 
1302 		pc += sizeof(int);
1303 	}
1304 
1305 	pr_cont("\n");
1306 }
1307 
1308 struct regbit {
1309 	unsigned long bit;
1310 	const char *name;
1311 };
1312 
1313 static struct regbit msr_bits[] = {
1314 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1315 	{MSR_SF,	"SF"},
1316 	{MSR_HV,	"HV"},
1317 #endif
1318 	{MSR_VEC,	"VEC"},
1319 	{MSR_VSX,	"VSX"},
1320 #ifdef CONFIG_BOOKE
1321 	{MSR_CE,	"CE"},
1322 #endif
1323 	{MSR_EE,	"EE"},
1324 	{MSR_PR,	"PR"},
1325 	{MSR_FP,	"FP"},
1326 	{MSR_ME,	"ME"},
1327 #ifdef CONFIG_BOOKE
1328 	{MSR_DE,	"DE"},
1329 #else
1330 	{MSR_SE,	"SE"},
1331 	{MSR_BE,	"BE"},
1332 #endif
1333 	{MSR_IR,	"IR"},
1334 	{MSR_DR,	"DR"},
1335 	{MSR_PMM,	"PMM"},
1336 #ifndef CONFIG_BOOKE
1337 	{MSR_RI,	"RI"},
1338 	{MSR_LE,	"LE"},
1339 #endif
1340 	{0,		NULL}
1341 };
1342 
1343 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1344 {
1345 	const char *s = "";
1346 
1347 	for (; bits->bit; ++bits)
1348 		if (val & bits->bit) {
1349 			pr_cont("%s%s", s, bits->name);
1350 			s = sep;
1351 		}
1352 }
1353 
1354 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1355 static struct regbit msr_tm_bits[] = {
1356 	{MSR_TS_T,	"T"},
1357 	{MSR_TS_S,	"S"},
1358 	{MSR_TM,	"E"},
1359 	{0,		NULL}
1360 };
1361 
1362 static void print_tm_bits(unsigned long val)
1363 {
1364 /*
1365  * This only prints something if at least one of the TM bit is set.
1366  * Inside the TM[], the output means:
1367  *   E: Enabled		(bit 32)
1368  *   S: Suspended	(bit 33)
1369  *   T: Transactional	(bit 34)
1370  */
1371 	if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1372 		pr_cont(",TM[");
1373 		print_bits(val, msr_tm_bits, "");
1374 		pr_cont("]");
1375 	}
1376 }
1377 #else
1378 static void print_tm_bits(unsigned long val) {}
1379 #endif
1380 
1381 static void print_msr_bits(unsigned long val)
1382 {
1383 	pr_cont("<");
1384 	print_bits(val, msr_bits, ",");
1385 	print_tm_bits(val);
1386 	pr_cont(">");
1387 }
1388 
1389 #ifdef CONFIG_PPC64
1390 #define REG		"%016lx"
1391 #define REGS_PER_LINE	4
1392 #define LAST_VOLATILE	13
1393 #else
1394 #define REG		"%08lx"
1395 #define REGS_PER_LINE	8
1396 #define LAST_VOLATILE	12
1397 #endif
1398 
1399 void show_regs(struct pt_regs * regs)
1400 {
1401 	int i, trap;
1402 
1403 	show_regs_print_info(KERN_DEFAULT);
1404 
1405 	printk("NIP:  "REG" LR: "REG" CTR: "REG"\n",
1406 	       regs->nip, regs->link, regs->ctr);
1407 	printk("REGS: %px TRAP: %04lx   %s  (%s)\n",
1408 	       regs, regs->trap, print_tainted(), init_utsname()->release);
1409 	printk("MSR:  "REG" ", regs->msr);
1410 	print_msr_bits(regs->msr);
1411 	pr_cont("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1412 	trap = TRAP(regs);
1413 	if ((TRAP(regs) != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
1414 		pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1415 	if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1416 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1417 		pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1418 #else
1419 		pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1420 #endif
1421 #ifdef CONFIG_PPC64
1422 	pr_cont("SOFTE: %ld ", regs->softe);
1423 #endif
1424 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1425 	if (MSR_TM_ACTIVE(regs->msr))
1426 		pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1427 #endif
1428 
1429 	for (i = 0;  i < 32;  i++) {
1430 		if ((i % REGS_PER_LINE) == 0)
1431 			pr_cont("\nGPR%02d: ", i);
1432 		pr_cont(REG " ", regs->gpr[i]);
1433 		if (i == LAST_VOLATILE && !FULL_REGS(regs))
1434 			break;
1435 	}
1436 	pr_cont("\n");
1437 #ifdef CONFIG_KALLSYMS
1438 	/*
1439 	 * Lookup NIP late so we have the best change of getting the
1440 	 * above info out without failing
1441 	 */
1442 	printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1443 	printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1444 #endif
1445 	show_stack(current, (unsigned long *) regs->gpr[1]);
1446 	if (!user_mode(regs))
1447 		show_instructions(regs);
1448 }
1449 
1450 void flush_thread(void)
1451 {
1452 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1453 	flush_ptrace_hw_breakpoint(current);
1454 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1455 	set_debug_reg_defaults(&current->thread);
1456 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1457 }
1458 
1459 int set_thread_uses_vas(void)
1460 {
1461 #ifdef CONFIG_PPC_BOOK3S_64
1462 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1463 		return -EINVAL;
1464 
1465 	current->thread.used_vas = 1;
1466 
1467 	/*
1468 	 * Even a process that has no foreign real address mapping can use
1469 	 * an unpaired COPY instruction (to no real effect). Issue CP_ABORT
1470 	 * to clear any pending COPY and prevent a covert channel.
1471 	 *
1472 	 * __switch_to() will issue CP_ABORT on future context switches.
1473 	 */
1474 	asm volatile(PPC_CP_ABORT);
1475 
1476 #endif /* CONFIG_PPC_BOOK3S_64 */
1477 	return 0;
1478 }
1479 
1480 #ifdef CONFIG_PPC64
1481 static DEFINE_SPINLOCK(vas_thread_id_lock);
1482 static DEFINE_IDA(vas_thread_ida);
1483 
1484 /*
1485  * We need to assign a unique thread id to each thread in a process.
1486  *
1487  * This thread id, referred to as TIDR, and separate from the Linux's tgid,
1488  * is intended to be used to direct an ASB_Notify from the hardware to the
1489  * thread, when a suitable event occurs in the system.
1490  *
1491  * One such event is a "paste" instruction in the context of Fast Thread
1492  * Wakeup (aka Core-to-core wake up in the Virtual Accelerator Switchboard
1493  * (VAS) in POWER9.
1494  *
1495  * To get a unique TIDR per process we could simply reuse task_pid_nr() but
1496  * the problem is that task_pid_nr() is not yet available copy_thread() is
1497  * called. Fixing that would require changing more intrusive arch-neutral
1498  * code in code path in copy_process()?.
1499  *
1500  * Further, to assign unique TIDRs within each process, we need an atomic
1501  * field (or an IDR) in task_struct, which again intrudes into the arch-
1502  * neutral code. So try to assign globally unique TIDRs for now.
1503  *
1504  * NOTE: TIDR 0 indicates that the thread does not need a TIDR value.
1505  *	 For now, only threads that expect to be notified by the VAS
1506  *	 hardware need a TIDR value and we assign values > 0 for those.
1507  */
1508 #define MAX_THREAD_CONTEXT	((1 << 16) - 1)
1509 static int assign_thread_tidr(void)
1510 {
1511 	int index;
1512 	int err;
1513 	unsigned long flags;
1514 
1515 again:
1516 	if (!ida_pre_get(&vas_thread_ida, GFP_KERNEL))
1517 		return -ENOMEM;
1518 
1519 	spin_lock_irqsave(&vas_thread_id_lock, flags);
1520 	err = ida_get_new_above(&vas_thread_ida, 1, &index);
1521 	spin_unlock_irqrestore(&vas_thread_id_lock, flags);
1522 
1523 	if (err == -EAGAIN)
1524 		goto again;
1525 	else if (err)
1526 		return err;
1527 
1528 	if (index > MAX_THREAD_CONTEXT) {
1529 		spin_lock_irqsave(&vas_thread_id_lock, flags);
1530 		ida_remove(&vas_thread_ida, index);
1531 		spin_unlock_irqrestore(&vas_thread_id_lock, flags);
1532 		return -ENOMEM;
1533 	}
1534 
1535 	return index;
1536 }
1537 
1538 static void free_thread_tidr(int id)
1539 {
1540 	unsigned long flags;
1541 
1542 	spin_lock_irqsave(&vas_thread_id_lock, flags);
1543 	ida_remove(&vas_thread_ida, id);
1544 	spin_unlock_irqrestore(&vas_thread_id_lock, flags);
1545 }
1546 
1547 /*
1548  * Clear any TIDR value assigned to this thread.
1549  */
1550 void clear_thread_tidr(struct task_struct *t)
1551 {
1552 	if (!t->thread.tidr)
1553 		return;
1554 
1555 	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1556 		WARN_ON_ONCE(1);
1557 		return;
1558 	}
1559 
1560 	mtspr(SPRN_TIDR, 0);
1561 	free_thread_tidr(t->thread.tidr);
1562 	t->thread.tidr = 0;
1563 }
1564 
1565 void arch_release_task_struct(struct task_struct *t)
1566 {
1567 	clear_thread_tidr(t);
1568 }
1569 
1570 /*
1571  * Assign a unique TIDR (thread id) for task @t and set it in the thread
1572  * structure. For now, we only support setting TIDR for 'current' task.
1573  */
1574 int set_thread_tidr(struct task_struct *t)
1575 {
1576 	int rc;
1577 
1578 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1579 		return -EINVAL;
1580 
1581 	if (t != current)
1582 		return -EINVAL;
1583 
1584 	if (t->thread.tidr)
1585 		return 0;
1586 
1587 	rc = assign_thread_tidr();
1588 	if (rc < 0)
1589 		return rc;
1590 
1591 	t->thread.tidr = rc;
1592 	mtspr(SPRN_TIDR, t->thread.tidr);
1593 
1594 	return 0;
1595 }
1596 EXPORT_SYMBOL_GPL(set_thread_tidr);
1597 
1598 #endif /* CONFIG_PPC64 */
1599 
1600 void
1601 release_thread(struct task_struct *t)
1602 {
1603 }
1604 
1605 /*
1606  * this gets called so that we can store coprocessor state into memory and
1607  * copy the current task into the new thread.
1608  */
1609 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1610 {
1611 	flush_all_to_thread(src);
1612 	/*
1613 	 * Flush TM state out so we can copy it.  __switch_to_tm() does this
1614 	 * flush but it removes the checkpointed state from the current CPU and
1615 	 * transitions the CPU out of TM mode.  Hence we need to call
1616 	 * tm_recheckpoint_new_task() (on the same task) to restore the
1617 	 * checkpointed state back and the TM mode.
1618 	 *
1619 	 * Can't pass dst because it isn't ready. Doesn't matter, passing
1620 	 * dst is only important for __switch_to()
1621 	 */
1622 	__switch_to_tm(src, src);
1623 
1624 	*dst = *src;
1625 
1626 	clear_task_ebb(dst);
1627 
1628 	return 0;
1629 }
1630 
1631 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1632 {
1633 #ifdef CONFIG_PPC_BOOK3S_64
1634 	unsigned long sp_vsid;
1635 	unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1636 
1637 	if (radix_enabled())
1638 		return;
1639 
1640 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1641 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1642 			<< SLB_VSID_SHIFT_1T;
1643 	else
1644 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1645 			<< SLB_VSID_SHIFT;
1646 	sp_vsid |= SLB_VSID_KERNEL | llp;
1647 	p->thread.ksp_vsid = sp_vsid;
1648 #endif
1649 }
1650 
1651 /*
1652  * Copy a thread..
1653  */
1654 
1655 /*
1656  * Copy architecture-specific thread state
1657  */
1658 int copy_thread(unsigned long clone_flags, unsigned long usp,
1659 		unsigned long kthread_arg, struct task_struct *p)
1660 {
1661 	struct pt_regs *childregs, *kregs;
1662 	extern void ret_from_fork(void);
1663 	extern void ret_from_kernel_thread(void);
1664 	void (*f)(void);
1665 	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1666 	struct thread_info *ti = task_thread_info(p);
1667 
1668 	klp_init_thread_info(ti);
1669 
1670 	/* Copy registers */
1671 	sp -= sizeof(struct pt_regs);
1672 	childregs = (struct pt_regs *) sp;
1673 	if (unlikely(p->flags & PF_KTHREAD)) {
1674 		/* kernel thread */
1675 		memset(childregs, 0, sizeof(struct pt_regs));
1676 		childregs->gpr[1] = sp + sizeof(struct pt_regs);
1677 		/* function */
1678 		if (usp)
1679 			childregs->gpr[14] = ppc_function_entry((void *)usp);
1680 #ifdef CONFIG_PPC64
1681 		clear_tsk_thread_flag(p, TIF_32BIT);
1682 		childregs->softe = IRQS_ENABLED;
1683 #endif
1684 		childregs->gpr[15] = kthread_arg;
1685 		p->thread.regs = NULL;	/* no user register state */
1686 		ti->flags |= _TIF_RESTOREALL;
1687 		f = ret_from_kernel_thread;
1688 	} else {
1689 		/* user thread */
1690 		struct pt_regs *regs = current_pt_regs();
1691 		CHECK_FULL_REGS(regs);
1692 		*childregs = *regs;
1693 		if (usp)
1694 			childregs->gpr[1] = usp;
1695 		p->thread.regs = childregs;
1696 		childregs->gpr[3] = 0;  /* Result from fork() */
1697 		if (clone_flags & CLONE_SETTLS) {
1698 #ifdef CONFIG_PPC64
1699 			if (!is_32bit_task())
1700 				childregs->gpr[13] = childregs->gpr[6];
1701 			else
1702 #endif
1703 				childregs->gpr[2] = childregs->gpr[6];
1704 		}
1705 
1706 		f = ret_from_fork;
1707 	}
1708 	childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1709 	sp -= STACK_FRAME_OVERHEAD;
1710 
1711 	/*
1712 	 * The way this works is that at some point in the future
1713 	 * some task will call _switch to switch to the new task.
1714 	 * That will pop off the stack frame created below and start
1715 	 * the new task running at ret_from_fork.  The new task will
1716 	 * do some house keeping and then return from the fork or clone
1717 	 * system call, using the stack frame created above.
1718 	 */
1719 	((unsigned long *)sp)[0] = 0;
1720 	sp -= sizeof(struct pt_regs);
1721 	kregs = (struct pt_regs *) sp;
1722 	sp -= STACK_FRAME_OVERHEAD;
1723 	p->thread.ksp = sp;
1724 #ifdef CONFIG_PPC32
1725 	p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
1726 				_ALIGN_UP(sizeof(struct thread_info), 16);
1727 #endif
1728 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1729 	p->thread.ptrace_bps[0] = NULL;
1730 #endif
1731 
1732 	p->thread.fp_save_area = NULL;
1733 #ifdef CONFIG_ALTIVEC
1734 	p->thread.vr_save_area = NULL;
1735 #endif
1736 
1737 	setup_ksp_vsid(p, sp);
1738 
1739 #ifdef CONFIG_PPC64
1740 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1741 		p->thread.dscr_inherit = current->thread.dscr_inherit;
1742 		p->thread.dscr = mfspr(SPRN_DSCR);
1743 	}
1744 	if (cpu_has_feature(CPU_FTR_HAS_PPR))
1745 		p->thread.ppr = INIT_PPR;
1746 
1747 	p->thread.tidr = 0;
1748 #endif
1749 	kregs->nip = ppc_function_entry(f);
1750 	return 0;
1751 }
1752 
1753 /*
1754  * Set up a thread for executing a new program
1755  */
1756 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1757 {
1758 #ifdef CONFIG_PPC64
1759 	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
1760 #endif
1761 
1762 	/*
1763 	 * If we exec out of a kernel thread then thread.regs will not be
1764 	 * set.  Do it now.
1765 	 */
1766 	if (!current->thread.regs) {
1767 		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1768 		current->thread.regs = regs - 1;
1769 	}
1770 
1771 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1772 	/*
1773 	 * Clear any transactional state, we're exec()ing. The cause is
1774 	 * not important as there will never be a recheckpoint so it's not
1775 	 * user visible.
1776 	 */
1777 	if (MSR_TM_SUSPENDED(mfmsr()))
1778 		tm_reclaim_current(0);
1779 #endif
1780 
1781 	memset(regs->gpr, 0, sizeof(regs->gpr));
1782 	regs->ctr = 0;
1783 	regs->link = 0;
1784 	regs->xer = 0;
1785 	regs->ccr = 0;
1786 	regs->gpr[1] = sp;
1787 
1788 	/*
1789 	 * We have just cleared all the nonvolatile GPRs, so make
1790 	 * FULL_REGS(regs) return true.  This is necessary to allow
1791 	 * ptrace to examine the thread immediately after exec.
1792 	 */
1793 	regs->trap &= ~1UL;
1794 
1795 #ifdef CONFIG_PPC32
1796 	regs->mq = 0;
1797 	regs->nip = start;
1798 	regs->msr = MSR_USER;
1799 #else
1800 	if (!is_32bit_task()) {
1801 		unsigned long entry;
1802 
1803 		if (is_elf2_task()) {
1804 			/* Look ma, no function descriptors! */
1805 			entry = start;
1806 
1807 			/*
1808 			 * Ulrich says:
1809 			 *   The latest iteration of the ABI requires that when
1810 			 *   calling a function (at its global entry point),
1811 			 *   the caller must ensure r12 holds the entry point
1812 			 *   address (so that the function can quickly
1813 			 *   establish addressability).
1814 			 */
1815 			regs->gpr[12] = start;
1816 			/* Make sure that's restored on entry to userspace. */
1817 			set_thread_flag(TIF_RESTOREALL);
1818 		} else {
1819 			unsigned long toc;
1820 
1821 			/* start is a relocated pointer to the function
1822 			 * descriptor for the elf _start routine.  The first
1823 			 * entry in the function descriptor is the entry
1824 			 * address of _start and the second entry is the TOC
1825 			 * value we need to use.
1826 			 */
1827 			__get_user(entry, (unsigned long __user *)start);
1828 			__get_user(toc, (unsigned long __user *)start+1);
1829 
1830 			/* Check whether the e_entry function descriptor entries
1831 			 * need to be relocated before we can use them.
1832 			 */
1833 			if (load_addr != 0) {
1834 				entry += load_addr;
1835 				toc   += load_addr;
1836 			}
1837 			regs->gpr[2] = toc;
1838 		}
1839 		regs->nip = entry;
1840 		regs->msr = MSR_USER64;
1841 	} else {
1842 		regs->nip = start;
1843 		regs->gpr[2] = 0;
1844 		regs->msr = MSR_USER32;
1845 	}
1846 #endif
1847 #ifdef CONFIG_VSX
1848 	current->thread.used_vsr = 0;
1849 #endif
1850 	current->thread.load_fp = 0;
1851 	memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1852 	current->thread.fp_save_area = NULL;
1853 #ifdef CONFIG_ALTIVEC
1854 	memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1855 	current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1856 	current->thread.vr_save_area = NULL;
1857 	current->thread.vrsave = 0;
1858 	current->thread.used_vr = 0;
1859 	current->thread.load_vec = 0;
1860 #endif /* CONFIG_ALTIVEC */
1861 #ifdef CONFIG_SPE
1862 	memset(current->thread.evr, 0, sizeof(current->thread.evr));
1863 	current->thread.acc = 0;
1864 	current->thread.spefscr = 0;
1865 	current->thread.used_spe = 0;
1866 #endif /* CONFIG_SPE */
1867 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1868 	current->thread.tm_tfhar = 0;
1869 	current->thread.tm_texasr = 0;
1870 	current->thread.tm_tfiar = 0;
1871 	current->thread.load_tm = 0;
1872 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1873 
1874 	thread_pkey_regs_init(&current->thread);
1875 }
1876 EXPORT_SYMBOL(start_thread);
1877 
1878 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1879 		| PR_FP_EXC_RES | PR_FP_EXC_INV)
1880 
1881 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1882 {
1883 	struct pt_regs *regs = tsk->thread.regs;
1884 
1885 	/* This is a bit hairy.  If we are an SPE enabled  processor
1886 	 * (have embedded fp) we store the IEEE exception enable flags in
1887 	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
1888 	 * mode (asyn, precise, disabled) for 'Classic' FP. */
1889 	if (val & PR_FP_EXC_SW_ENABLE) {
1890 #ifdef CONFIG_SPE
1891 		if (cpu_has_feature(CPU_FTR_SPE)) {
1892 			/*
1893 			 * When the sticky exception bits are set
1894 			 * directly by userspace, it must call prctl
1895 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1896 			 * in the existing prctl settings) or
1897 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1898 			 * the bits being set).  <fenv.h> functions
1899 			 * saving and restoring the whole
1900 			 * floating-point environment need to do so
1901 			 * anyway to restore the prctl settings from
1902 			 * the saved environment.
1903 			 */
1904 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1905 			tsk->thread.fpexc_mode = val &
1906 				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1907 			return 0;
1908 		} else {
1909 			return -EINVAL;
1910 		}
1911 #else
1912 		return -EINVAL;
1913 #endif
1914 	}
1915 
1916 	/* on a CONFIG_SPE this does not hurt us.  The bits that
1917 	 * __pack_fe01 use do not overlap with bits used for
1918 	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1919 	 * on CONFIG_SPE implementations are reserved so writing to
1920 	 * them does not change anything */
1921 	if (val > PR_FP_EXC_PRECISE)
1922 		return -EINVAL;
1923 	tsk->thread.fpexc_mode = __pack_fe01(val);
1924 	if (regs != NULL && (regs->msr & MSR_FP) != 0)
1925 		regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1926 			| tsk->thread.fpexc_mode;
1927 	return 0;
1928 }
1929 
1930 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1931 {
1932 	unsigned int val;
1933 
1934 	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1935 #ifdef CONFIG_SPE
1936 		if (cpu_has_feature(CPU_FTR_SPE)) {
1937 			/*
1938 			 * When the sticky exception bits are set
1939 			 * directly by userspace, it must call prctl
1940 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1941 			 * in the existing prctl settings) or
1942 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1943 			 * the bits being set).  <fenv.h> functions
1944 			 * saving and restoring the whole
1945 			 * floating-point environment need to do so
1946 			 * anyway to restore the prctl settings from
1947 			 * the saved environment.
1948 			 */
1949 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1950 			val = tsk->thread.fpexc_mode;
1951 		} else
1952 			return -EINVAL;
1953 #else
1954 		return -EINVAL;
1955 #endif
1956 	else
1957 		val = __unpack_fe01(tsk->thread.fpexc_mode);
1958 	return put_user(val, (unsigned int __user *) adr);
1959 }
1960 
1961 int set_endian(struct task_struct *tsk, unsigned int val)
1962 {
1963 	struct pt_regs *regs = tsk->thread.regs;
1964 
1965 	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1966 	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1967 		return -EINVAL;
1968 
1969 	if (regs == NULL)
1970 		return -EINVAL;
1971 
1972 	if (val == PR_ENDIAN_BIG)
1973 		regs->msr &= ~MSR_LE;
1974 	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1975 		regs->msr |= MSR_LE;
1976 	else
1977 		return -EINVAL;
1978 
1979 	return 0;
1980 }
1981 
1982 int get_endian(struct task_struct *tsk, unsigned long adr)
1983 {
1984 	struct pt_regs *regs = tsk->thread.regs;
1985 	unsigned int val;
1986 
1987 	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1988 	    !cpu_has_feature(CPU_FTR_REAL_LE))
1989 		return -EINVAL;
1990 
1991 	if (regs == NULL)
1992 		return -EINVAL;
1993 
1994 	if (regs->msr & MSR_LE) {
1995 		if (cpu_has_feature(CPU_FTR_REAL_LE))
1996 			val = PR_ENDIAN_LITTLE;
1997 		else
1998 			val = PR_ENDIAN_PPC_LITTLE;
1999 	} else
2000 		val = PR_ENDIAN_BIG;
2001 
2002 	return put_user(val, (unsigned int __user *)adr);
2003 }
2004 
2005 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
2006 {
2007 	tsk->thread.align_ctl = val;
2008 	return 0;
2009 }
2010 
2011 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
2012 {
2013 	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
2014 }
2015 
2016 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
2017 				  unsigned long nbytes)
2018 {
2019 	unsigned long stack_page;
2020 	unsigned long cpu = task_cpu(p);
2021 
2022 	/*
2023 	 * Avoid crashing if the stack has overflowed and corrupted
2024 	 * task_cpu(p), which is in the thread_info struct.
2025 	 */
2026 	if (cpu < NR_CPUS && cpu_possible(cpu)) {
2027 		stack_page = (unsigned long) hardirq_ctx[cpu];
2028 		if (sp >= stack_page + sizeof(struct thread_struct)
2029 		    && sp <= stack_page + THREAD_SIZE - nbytes)
2030 			return 1;
2031 
2032 		stack_page = (unsigned long) softirq_ctx[cpu];
2033 		if (sp >= stack_page + sizeof(struct thread_struct)
2034 		    && sp <= stack_page + THREAD_SIZE - nbytes)
2035 			return 1;
2036 	}
2037 	return 0;
2038 }
2039 
2040 int validate_sp(unsigned long sp, struct task_struct *p,
2041 		       unsigned long nbytes)
2042 {
2043 	unsigned long stack_page = (unsigned long)task_stack_page(p);
2044 
2045 	if (sp >= stack_page + sizeof(struct thread_struct)
2046 	    && sp <= stack_page + THREAD_SIZE - nbytes)
2047 		return 1;
2048 
2049 	return valid_irq_stack(sp, p, nbytes);
2050 }
2051 
2052 EXPORT_SYMBOL(validate_sp);
2053 
2054 unsigned long get_wchan(struct task_struct *p)
2055 {
2056 	unsigned long ip, sp;
2057 	int count = 0;
2058 
2059 	if (!p || p == current || p->state == TASK_RUNNING)
2060 		return 0;
2061 
2062 	sp = p->thread.ksp;
2063 	if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2064 		return 0;
2065 
2066 	do {
2067 		sp = *(unsigned long *)sp;
2068 		if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2069 		    p->state == TASK_RUNNING)
2070 			return 0;
2071 		if (count > 0) {
2072 			ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2073 			if (!in_sched_functions(ip))
2074 				return ip;
2075 		}
2076 	} while (count++ < 16);
2077 	return 0;
2078 }
2079 
2080 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2081 
2082 void show_stack(struct task_struct *tsk, unsigned long *stack)
2083 {
2084 	unsigned long sp, ip, lr, newsp;
2085 	int count = 0;
2086 	int firstframe = 1;
2087 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2088 	int curr_frame = current->curr_ret_stack;
2089 	extern void return_to_handler(void);
2090 	unsigned long rth = (unsigned long)return_to_handler;
2091 #endif
2092 
2093 	sp = (unsigned long) stack;
2094 	if (tsk == NULL)
2095 		tsk = current;
2096 	if (sp == 0) {
2097 		if (tsk == current)
2098 			sp = current_stack_pointer();
2099 		else
2100 			sp = tsk->thread.ksp;
2101 	}
2102 
2103 	lr = 0;
2104 	printk("Call Trace:\n");
2105 	do {
2106 		if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2107 			return;
2108 
2109 		stack = (unsigned long *) sp;
2110 		newsp = stack[0];
2111 		ip = stack[STACK_FRAME_LR_SAVE];
2112 		if (!firstframe || ip != lr) {
2113 			printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
2114 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
2115 			if ((ip == rth) && curr_frame >= 0) {
2116 				pr_cont(" (%pS)",
2117 				       (void *)current->ret_stack[curr_frame].ret);
2118 				curr_frame--;
2119 			}
2120 #endif
2121 			if (firstframe)
2122 				pr_cont(" (unreliable)");
2123 			pr_cont("\n");
2124 		}
2125 		firstframe = 0;
2126 
2127 		/*
2128 		 * See if this is an exception frame.
2129 		 * We look for the "regshere" marker in the current frame.
2130 		 */
2131 		if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
2132 		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2133 			struct pt_regs *regs = (struct pt_regs *)
2134 				(sp + STACK_FRAME_OVERHEAD);
2135 			lr = regs->link;
2136 			printk("--- interrupt: %lx at %pS\n    LR = %pS\n",
2137 			       regs->trap, (void *)regs->nip, (void *)lr);
2138 			firstframe = 1;
2139 		}
2140 
2141 		sp = newsp;
2142 	} while (count++ < kstack_depth_to_print);
2143 }
2144 
2145 #ifdef CONFIG_PPC64
2146 /* Called with hard IRQs off */
2147 void notrace __ppc64_runlatch_on(void)
2148 {
2149 	struct thread_info *ti = current_thread_info();
2150 
2151 	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2152 		/*
2153 		 * Least significant bit (RUN) is the only writable bit of
2154 		 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2155 		 * earliest ISA where this is the case, but it's convenient.
2156 		 */
2157 		mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2158 	} else {
2159 		unsigned long ctrl;
2160 
2161 		/*
2162 		 * Some architectures (e.g., Cell) have writable fields other
2163 		 * than RUN, so do the read-modify-write.
2164 		 */
2165 		ctrl = mfspr(SPRN_CTRLF);
2166 		ctrl |= CTRL_RUNLATCH;
2167 		mtspr(SPRN_CTRLT, ctrl);
2168 	}
2169 
2170 	ti->local_flags |= _TLF_RUNLATCH;
2171 }
2172 
2173 /* Called with hard IRQs off */
2174 void notrace __ppc64_runlatch_off(void)
2175 {
2176 	struct thread_info *ti = current_thread_info();
2177 
2178 	ti->local_flags &= ~_TLF_RUNLATCH;
2179 
2180 	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2181 		mtspr(SPRN_CTRLT, 0);
2182 	} else {
2183 		unsigned long ctrl;
2184 
2185 		ctrl = mfspr(SPRN_CTRLF);
2186 		ctrl &= ~CTRL_RUNLATCH;
2187 		mtspr(SPRN_CTRLT, ctrl);
2188 	}
2189 }
2190 #endif /* CONFIG_PPC64 */
2191 
2192 unsigned long arch_align_stack(unsigned long sp)
2193 {
2194 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2195 		sp -= get_random_int() & ~PAGE_MASK;
2196 	return sp & ~0xf;
2197 }
2198 
2199 static inline unsigned long brk_rnd(void)
2200 {
2201         unsigned long rnd = 0;
2202 
2203 	/* 8MB for 32bit, 1GB for 64bit */
2204 	if (is_32bit_task())
2205 		rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2206 	else
2207 		rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2208 
2209 	return rnd << PAGE_SHIFT;
2210 }
2211 
2212 unsigned long arch_randomize_brk(struct mm_struct *mm)
2213 {
2214 	unsigned long base = mm->brk;
2215 	unsigned long ret;
2216 
2217 #ifdef CONFIG_PPC_BOOK3S_64
2218 	/*
2219 	 * If we are using 1TB segments and we are allowed to randomise
2220 	 * the heap, we can put it above 1TB so it is backed by a 1TB
2221 	 * segment. Otherwise the heap will be in the bottom 1TB
2222 	 * which always uses 256MB segments and this may result in a
2223 	 * performance penalty. We don't need to worry about radix. For
2224 	 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2225 	 */
2226 	if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2227 		base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2228 #endif
2229 
2230 	ret = PAGE_ALIGN(base + brk_rnd());
2231 
2232 	if (ret < mm->brk)
2233 		return mm->brk;
2234 
2235 	return ret;
2236 }
2237 
2238