xref: /openbmc/linux/arch/powerpc/kernel/process.c (revision d9679d00)
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/interrupt.h>
45 #include <asm/io.h>
46 #include <asm/processor.h>
47 #include <asm/mmu.h>
48 #include <asm/prom.h>
49 #include <asm/machdep.h>
50 #include <asm/time.h>
51 #include <asm/runlatch.h>
52 #include <asm/syscalls.h>
53 #include <asm/switch_to.h>
54 #include <asm/tm.h>
55 #include <asm/debug.h>
56 #ifdef CONFIG_PPC64
57 #include <asm/firmware.h>
58 #include <asm/hw_irq.h>
59 #endif
60 #include <asm/code-patching.h>
61 #include <asm/exec.h>
62 #include <asm/livepatch.h>
63 #include <asm/cpu_has_feature.h>
64 #include <asm/asm-prototypes.h>
65 #include <asm/stacktrace.h>
66 #include <asm/hw_breakpoint.h>
67 
68 #include <linux/kprobes.h>
69 #include <linux/kdebug.h>
70 
71 /* Transactional Memory debug */
72 #ifdef TM_DEBUG_SW
73 #define TM_DEBUG(x...) printk(KERN_INFO x)
74 #else
75 #define TM_DEBUG(x...) do { } while(0)
76 #endif
77 
78 extern unsigned long _get_SP(void);
79 
80 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
81 /*
82  * Are we running in "Suspend disabled" mode? If so we have to block any
83  * sigreturn that would get us into suspended state, and we also warn in some
84  * other paths that we should never reach with suspend disabled.
85  */
86 bool tm_suspend_disabled __ro_after_init = false;
87 
88 static void check_if_tm_restore_required(struct task_struct *tsk)
89 {
90 	/*
91 	 * If we are saving the current thread's registers, and the
92 	 * thread is in a transactional state, set the TIF_RESTORE_TM
93 	 * bit so that we know to restore the registers before
94 	 * returning to userspace.
95 	 */
96 	if (tsk == current && tsk->thread.regs &&
97 	    MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
98 	    !test_thread_flag(TIF_RESTORE_TM)) {
99 		regs_set_return_msr(&tsk->thread.ckpt_regs,
100 						tsk->thread.regs->msr);
101 		set_thread_flag(TIF_RESTORE_TM);
102 	}
103 }
104 
105 #else
106 static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
107 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
108 
109 bool strict_msr_control;
110 EXPORT_SYMBOL(strict_msr_control);
111 
112 static int __init enable_strict_msr_control(char *str)
113 {
114 	strict_msr_control = true;
115 	pr_info("Enabling strict facility control\n");
116 
117 	return 0;
118 }
119 early_param("ppc_strict_facility_enable", enable_strict_msr_control);
120 
121 /* notrace because it's called by restore_math */
122 unsigned long notrace msr_check_and_set(unsigned long bits)
123 {
124 	unsigned long oldmsr = mfmsr();
125 	unsigned long newmsr;
126 
127 	newmsr = oldmsr | bits;
128 
129 	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
130 		newmsr |= MSR_VSX;
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 	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
148 		newmsr &= ~MSR_VSX;
149 
150 	if (oldmsr != newmsr)
151 		mtmsr_isync(newmsr);
152 }
153 EXPORT_SYMBOL(__msr_check_and_clear);
154 
155 #ifdef CONFIG_PPC_FPU
156 static void __giveup_fpu(struct task_struct *tsk)
157 {
158 	unsigned long msr;
159 
160 	save_fpu(tsk);
161 	msr = tsk->thread.regs->msr;
162 	msr &= ~(MSR_FP|MSR_FE0|MSR_FE1);
163 	if (cpu_has_feature(CPU_FTR_VSX))
164 		msr &= ~MSR_VSX;
165 	regs_set_return_msr(tsk->thread.regs, msr);
166 }
167 
168 void giveup_fpu(struct task_struct *tsk)
169 {
170 	check_if_tm_restore_required(tsk);
171 
172 	msr_check_and_set(MSR_FP);
173 	__giveup_fpu(tsk);
174 	msr_check_and_clear(MSR_FP);
175 }
176 EXPORT_SYMBOL(giveup_fpu);
177 
178 /*
179  * Make sure the floating-point register state in the
180  * the thread_struct is up to date for task tsk.
181  */
182 void flush_fp_to_thread(struct task_struct *tsk)
183 {
184 	if (tsk->thread.regs) {
185 		/*
186 		 * We need to disable preemption here because if we didn't,
187 		 * another process could get scheduled after the regs->msr
188 		 * test but before we have finished saving the FP registers
189 		 * to the thread_struct.  That process could take over the
190 		 * FPU, and then when we get scheduled again we would store
191 		 * bogus values for the remaining FP registers.
192 		 */
193 		preempt_disable();
194 		if (tsk->thread.regs->msr & MSR_FP) {
195 			/*
196 			 * This should only ever be called for current or
197 			 * for a stopped child process.  Since we save away
198 			 * the FP register state on context switch,
199 			 * there is something wrong if a stopped child appears
200 			 * to still have its FP state in the CPU registers.
201 			 */
202 			BUG_ON(tsk != current);
203 			giveup_fpu(tsk);
204 		}
205 		preempt_enable();
206 	}
207 }
208 EXPORT_SYMBOL_GPL(flush_fp_to_thread);
209 
210 void enable_kernel_fp(void)
211 {
212 	unsigned long cpumsr;
213 
214 	WARN_ON(preemptible());
215 
216 	cpumsr = msr_check_and_set(MSR_FP);
217 
218 	if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
219 		check_if_tm_restore_required(current);
220 		/*
221 		 * If a thread has already been reclaimed then the
222 		 * checkpointed registers are on the CPU but have definitely
223 		 * been saved by the reclaim code. Don't need to and *cannot*
224 		 * giveup as this would save  to the 'live' structure not the
225 		 * checkpointed structure.
226 		 */
227 		if (!MSR_TM_ACTIVE(cpumsr) &&
228 		     MSR_TM_ACTIVE(current->thread.regs->msr))
229 			return;
230 		__giveup_fpu(current);
231 	}
232 }
233 EXPORT_SYMBOL(enable_kernel_fp);
234 #else
235 static inline void __giveup_fpu(struct task_struct *tsk) { }
236 #endif /* CONFIG_PPC_FPU */
237 
238 #ifdef CONFIG_ALTIVEC
239 static void __giveup_altivec(struct task_struct *tsk)
240 {
241 	unsigned long msr;
242 
243 	save_altivec(tsk);
244 	msr = tsk->thread.regs->msr;
245 	msr &= ~MSR_VEC;
246 	if (cpu_has_feature(CPU_FTR_VSX))
247 		msr &= ~MSR_VSX;
248 	regs_set_return_msr(tsk->thread.regs, msr);
249 }
250 
251 void giveup_altivec(struct task_struct *tsk)
252 {
253 	check_if_tm_restore_required(tsk);
254 
255 	msr_check_and_set(MSR_VEC);
256 	__giveup_altivec(tsk);
257 	msr_check_and_clear(MSR_VEC);
258 }
259 EXPORT_SYMBOL(giveup_altivec);
260 
261 void enable_kernel_altivec(void)
262 {
263 	unsigned long cpumsr;
264 
265 	WARN_ON(preemptible());
266 
267 	cpumsr = msr_check_and_set(MSR_VEC);
268 
269 	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
270 		check_if_tm_restore_required(current);
271 		/*
272 		 * If a thread has already been reclaimed then the
273 		 * checkpointed registers are on the CPU but have definitely
274 		 * been saved by the reclaim code. Don't need to and *cannot*
275 		 * giveup as this would save  to the 'live' structure not the
276 		 * checkpointed structure.
277 		 */
278 		if (!MSR_TM_ACTIVE(cpumsr) &&
279 		     MSR_TM_ACTIVE(current->thread.regs->msr))
280 			return;
281 		__giveup_altivec(current);
282 	}
283 }
284 EXPORT_SYMBOL(enable_kernel_altivec);
285 
286 /*
287  * Make sure the VMX/Altivec register state in the
288  * the thread_struct is up to date for task tsk.
289  */
290 void flush_altivec_to_thread(struct task_struct *tsk)
291 {
292 	if (tsk->thread.regs) {
293 		preempt_disable();
294 		if (tsk->thread.regs->msr & MSR_VEC) {
295 			BUG_ON(tsk != current);
296 			giveup_altivec(tsk);
297 		}
298 		preempt_enable();
299 	}
300 }
301 EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
302 #endif /* CONFIG_ALTIVEC */
303 
304 #ifdef CONFIG_VSX
305 static void __giveup_vsx(struct task_struct *tsk)
306 {
307 	unsigned long msr = tsk->thread.regs->msr;
308 
309 	/*
310 	 * We should never be ssetting MSR_VSX without also setting
311 	 * MSR_FP and MSR_VEC
312 	 */
313 	WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
314 
315 	/* __giveup_fpu will clear MSR_VSX */
316 	if (msr & MSR_FP)
317 		__giveup_fpu(tsk);
318 	if (msr & MSR_VEC)
319 		__giveup_altivec(tsk);
320 }
321 
322 static void giveup_vsx(struct task_struct *tsk)
323 {
324 	check_if_tm_restore_required(tsk);
325 
326 	msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
327 	__giveup_vsx(tsk);
328 	msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
329 }
330 
331 void enable_kernel_vsx(void)
332 {
333 	unsigned long cpumsr;
334 
335 	WARN_ON(preemptible());
336 
337 	cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
338 
339 	if (current->thread.regs &&
340 	    (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
341 		check_if_tm_restore_required(current);
342 		/*
343 		 * If a thread has already been reclaimed then the
344 		 * checkpointed registers are on the CPU but have definitely
345 		 * been saved by the reclaim code. Don't need to and *cannot*
346 		 * giveup as this would save  to the 'live' structure not the
347 		 * checkpointed structure.
348 		 */
349 		if (!MSR_TM_ACTIVE(cpumsr) &&
350 		     MSR_TM_ACTIVE(current->thread.regs->msr))
351 			return;
352 		__giveup_vsx(current);
353 	}
354 }
355 EXPORT_SYMBOL(enable_kernel_vsx);
356 
357 void flush_vsx_to_thread(struct task_struct *tsk)
358 {
359 	if (tsk->thread.regs) {
360 		preempt_disable();
361 		if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
362 			BUG_ON(tsk != current);
363 			giveup_vsx(tsk);
364 		}
365 		preempt_enable();
366 	}
367 }
368 EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
369 #endif /* CONFIG_VSX */
370 
371 #ifdef CONFIG_SPE
372 void giveup_spe(struct task_struct *tsk)
373 {
374 	check_if_tm_restore_required(tsk);
375 
376 	msr_check_and_set(MSR_SPE);
377 	__giveup_spe(tsk);
378 	msr_check_and_clear(MSR_SPE);
379 }
380 EXPORT_SYMBOL(giveup_spe);
381 
382 void enable_kernel_spe(void)
383 {
384 	WARN_ON(preemptible());
385 
386 	msr_check_and_set(MSR_SPE);
387 
388 	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
389 		check_if_tm_restore_required(current);
390 		__giveup_spe(current);
391 	}
392 }
393 EXPORT_SYMBOL(enable_kernel_spe);
394 
395 void flush_spe_to_thread(struct task_struct *tsk)
396 {
397 	if (tsk->thread.regs) {
398 		preempt_disable();
399 		if (tsk->thread.regs->msr & MSR_SPE) {
400 			BUG_ON(tsk != current);
401 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
402 			giveup_spe(tsk);
403 		}
404 		preempt_enable();
405 	}
406 }
407 #endif /* CONFIG_SPE */
408 
409 static unsigned long msr_all_available;
410 
411 static int __init init_msr_all_available(void)
412 {
413 	if (IS_ENABLED(CONFIG_PPC_FPU))
414 		msr_all_available |= MSR_FP;
415 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
416 		msr_all_available |= MSR_VEC;
417 	if (cpu_has_feature(CPU_FTR_VSX))
418 		msr_all_available |= MSR_VSX;
419 	if (cpu_has_feature(CPU_FTR_SPE))
420 		msr_all_available |= MSR_SPE;
421 
422 	return 0;
423 }
424 early_initcall(init_msr_all_available);
425 
426 void giveup_all(struct task_struct *tsk)
427 {
428 	unsigned long usermsr;
429 
430 	if (!tsk->thread.regs)
431 		return;
432 
433 	check_if_tm_restore_required(tsk);
434 
435 	usermsr = tsk->thread.regs->msr;
436 
437 	if ((usermsr & msr_all_available) == 0)
438 		return;
439 
440 	msr_check_and_set(msr_all_available);
441 
442 	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
443 
444 	if (usermsr & MSR_FP)
445 		__giveup_fpu(tsk);
446 	if (usermsr & MSR_VEC)
447 		__giveup_altivec(tsk);
448 	if (usermsr & MSR_SPE)
449 		__giveup_spe(tsk);
450 
451 	msr_check_and_clear(msr_all_available);
452 }
453 EXPORT_SYMBOL(giveup_all);
454 
455 #ifdef CONFIG_PPC_BOOK3S_64
456 #ifdef CONFIG_PPC_FPU
457 static bool should_restore_fp(void)
458 {
459 	if (current->thread.load_fp) {
460 		current->thread.load_fp++;
461 		return true;
462 	}
463 	return false;
464 }
465 
466 static void do_restore_fp(void)
467 {
468 	load_fp_state(&current->thread.fp_state);
469 }
470 #else
471 static bool should_restore_fp(void) { return false; }
472 static void do_restore_fp(void) { }
473 #endif /* CONFIG_PPC_FPU */
474 
475 #ifdef CONFIG_ALTIVEC
476 static bool should_restore_altivec(void)
477 {
478 	if (cpu_has_feature(CPU_FTR_ALTIVEC) && (current->thread.load_vec)) {
479 		current->thread.load_vec++;
480 		return true;
481 	}
482 	return false;
483 }
484 
485 static void do_restore_altivec(void)
486 {
487 	load_vr_state(&current->thread.vr_state);
488 	current->thread.used_vr = 1;
489 }
490 #else
491 static bool should_restore_altivec(void) { return false; }
492 static void do_restore_altivec(void) { }
493 #endif /* CONFIG_ALTIVEC */
494 
495 static bool should_restore_vsx(void)
496 {
497 	if (cpu_has_feature(CPU_FTR_VSX))
498 		return true;
499 	return false;
500 }
501 #ifdef CONFIG_VSX
502 static void do_restore_vsx(void)
503 {
504 	current->thread.used_vsr = 1;
505 }
506 #else
507 static void do_restore_vsx(void) { }
508 #endif /* CONFIG_VSX */
509 
510 /*
511  * The exception exit path calls restore_math() with interrupts hard disabled
512  * but the soft irq state not "reconciled". ftrace code that calls
513  * local_irq_save/restore causes warnings.
514  *
515  * Rather than complicate the exit path, just don't trace restore_math. This
516  * could be done by having ftrace entry code check for this un-reconciled
517  * condition where MSR[EE]=0 and PACA_IRQ_HARD_DIS is not set, and
518  * temporarily fix it up for the duration of the ftrace call.
519  */
520 void notrace restore_math(struct pt_regs *regs)
521 {
522 	unsigned long msr;
523 	unsigned long new_msr = 0;
524 
525 	msr = regs->msr;
526 
527 	/*
528 	 * new_msr tracks the facilities that are to be restored. Only reload
529 	 * if the bit is not set in the user MSR (if it is set, the registers
530 	 * are live for the user thread).
531 	 */
532 	if ((!(msr & MSR_FP)) && should_restore_fp())
533 		new_msr |= MSR_FP;
534 
535 	if ((!(msr & MSR_VEC)) && should_restore_altivec())
536 		new_msr |= MSR_VEC;
537 
538 	if ((!(msr & MSR_VSX)) && should_restore_vsx()) {
539 		if (((msr | new_msr) & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC))
540 			new_msr |= MSR_VSX;
541 	}
542 
543 	if (new_msr) {
544 		unsigned long fpexc_mode = 0;
545 
546 		msr_check_and_set(new_msr);
547 
548 		if (new_msr & MSR_FP) {
549 			do_restore_fp();
550 
551 			// This also covers VSX, because VSX implies FP
552 			fpexc_mode = current->thread.fpexc_mode;
553 		}
554 
555 		if (new_msr & MSR_VEC)
556 			do_restore_altivec();
557 
558 		if (new_msr & MSR_VSX)
559 			do_restore_vsx();
560 
561 		msr_check_and_clear(new_msr);
562 
563 		regs_set_return_msr(regs, regs->msr | new_msr | fpexc_mode);
564 	}
565 }
566 #endif /* CONFIG_PPC_BOOK3S_64 */
567 
568 static void save_all(struct task_struct *tsk)
569 {
570 	unsigned long usermsr;
571 
572 	if (!tsk->thread.regs)
573 		return;
574 
575 	usermsr = tsk->thread.regs->msr;
576 
577 	if ((usermsr & msr_all_available) == 0)
578 		return;
579 
580 	msr_check_and_set(msr_all_available);
581 
582 	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
583 
584 	if (usermsr & MSR_FP)
585 		save_fpu(tsk);
586 
587 	if (usermsr & MSR_VEC)
588 		save_altivec(tsk);
589 
590 	if (usermsr & MSR_SPE)
591 		__giveup_spe(tsk);
592 
593 	msr_check_and_clear(msr_all_available);
594 }
595 
596 void flush_all_to_thread(struct task_struct *tsk)
597 {
598 	if (tsk->thread.regs) {
599 		preempt_disable();
600 		BUG_ON(tsk != current);
601 #ifdef CONFIG_SPE
602 		if (tsk->thread.regs->msr & MSR_SPE)
603 			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
604 #endif
605 		save_all(tsk);
606 
607 		preempt_enable();
608 	}
609 }
610 EXPORT_SYMBOL(flush_all_to_thread);
611 
612 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
613 void do_send_trap(struct pt_regs *regs, unsigned long address,
614 		  unsigned long error_code, int breakpt)
615 {
616 	current->thread.trap_nr = TRAP_HWBKPT;
617 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
618 			11, SIGSEGV) == NOTIFY_STOP)
619 		return;
620 
621 	/* Deliver the signal to userspace */
622 	force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
623 				    (void __user *)address);
624 }
625 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
626 
627 static void do_break_handler(struct pt_regs *regs)
628 {
629 	struct arch_hw_breakpoint null_brk = {0};
630 	struct arch_hw_breakpoint *info;
631 	struct ppc_inst instr = ppc_inst(0);
632 	int type = 0;
633 	int size = 0;
634 	unsigned long ea;
635 	int i;
636 
637 	/*
638 	 * If underneath hw supports only one watchpoint, we know it
639 	 * caused exception. 8xx also falls into this category.
640 	 */
641 	if (nr_wp_slots() == 1) {
642 		__set_breakpoint(0, &null_brk);
643 		current->thread.hw_brk[0] = null_brk;
644 		current->thread.hw_brk[0].flags |= HW_BRK_FLAG_DISABLED;
645 		return;
646 	}
647 
648 	/* Otherwise findout which DAWR caused exception and disable it. */
649 	wp_get_instr_detail(regs, &instr, &type, &size, &ea);
650 
651 	for (i = 0; i < nr_wp_slots(); i++) {
652 		info = &current->thread.hw_brk[i];
653 		if (!info->address)
654 			continue;
655 
656 		if (wp_check_constraints(regs, instr, ea, type, size, info)) {
657 			__set_breakpoint(i, &null_brk);
658 			current->thread.hw_brk[i] = null_brk;
659 			current->thread.hw_brk[i].flags |= HW_BRK_FLAG_DISABLED;
660 		}
661 	}
662 }
663 
664 DEFINE_INTERRUPT_HANDLER(do_break)
665 {
666 	current->thread.trap_nr = TRAP_HWBKPT;
667 	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, regs->dsisr,
668 			11, SIGSEGV) == NOTIFY_STOP)
669 		return;
670 
671 	if (debugger_break_match(regs))
672 		return;
673 
674 	/*
675 	 * We reach here only when watchpoint exception is generated by ptrace
676 	 * event (or hw is buggy!). Now if CONFIG_HAVE_HW_BREAKPOINT is set,
677 	 * watchpoint is already handled by hw_breakpoint_handler() so we don't
678 	 * have to do anything. But when CONFIG_HAVE_HW_BREAKPOINT is not set,
679 	 * we need to manually handle the watchpoint here.
680 	 */
681 	if (!IS_ENABLED(CONFIG_HAVE_HW_BREAKPOINT))
682 		do_break_handler(regs);
683 
684 	/* Deliver the signal to userspace */
685 	force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)regs->dar);
686 }
687 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
688 
689 static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk[HBP_NUM_MAX]);
690 
691 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
692 /*
693  * Set the debug registers back to their default "safe" values.
694  */
695 static void set_debug_reg_defaults(struct thread_struct *thread)
696 {
697 	thread->debug.iac1 = thread->debug.iac2 = 0;
698 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
699 	thread->debug.iac3 = thread->debug.iac4 = 0;
700 #endif
701 	thread->debug.dac1 = thread->debug.dac2 = 0;
702 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
703 	thread->debug.dvc1 = thread->debug.dvc2 = 0;
704 #endif
705 	thread->debug.dbcr0 = 0;
706 #ifdef CONFIG_BOOKE
707 	/*
708 	 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
709 	 */
710 	thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
711 			DBCR1_IAC3US | DBCR1_IAC4US;
712 	/*
713 	 * Force Data Address Compare User/Supervisor bits to be User-only
714 	 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
715 	 */
716 	thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
717 #else
718 	thread->debug.dbcr1 = 0;
719 #endif
720 }
721 
722 static void prime_debug_regs(struct debug_reg *debug)
723 {
724 	/*
725 	 * We could have inherited MSR_DE from userspace, since
726 	 * it doesn't get cleared on exception entry.  Make sure
727 	 * MSR_DE is clear before we enable any debug events.
728 	 */
729 	mtmsr(mfmsr() & ~MSR_DE);
730 
731 	mtspr(SPRN_IAC1, debug->iac1);
732 	mtspr(SPRN_IAC2, debug->iac2);
733 #if CONFIG_PPC_ADV_DEBUG_IACS > 2
734 	mtspr(SPRN_IAC3, debug->iac3);
735 	mtspr(SPRN_IAC4, debug->iac4);
736 #endif
737 	mtspr(SPRN_DAC1, debug->dac1);
738 	mtspr(SPRN_DAC2, debug->dac2);
739 #if CONFIG_PPC_ADV_DEBUG_DVCS > 0
740 	mtspr(SPRN_DVC1, debug->dvc1);
741 	mtspr(SPRN_DVC2, debug->dvc2);
742 #endif
743 	mtspr(SPRN_DBCR0, debug->dbcr0);
744 	mtspr(SPRN_DBCR1, debug->dbcr1);
745 #ifdef CONFIG_BOOKE
746 	mtspr(SPRN_DBCR2, debug->dbcr2);
747 #endif
748 }
749 /*
750  * Unless neither the old or new thread are making use of the
751  * debug registers, set the debug registers from the values
752  * stored in the new thread.
753  */
754 void switch_booke_debug_regs(struct debug_reg *new_debug)
755 {
756 	if ((current->thread.debug.dbcr0 & DBCR0_IDM)
757 		|| (new_debug->dbcr0 & DBCR0_IDM))
758 			prime_debug_regs(new_debug);
759 }
760 EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
761 #else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
762 #ifndef CONFIG_HAVE_HW_BREAKPOINT
763 static void set_breakpoint(int i, struct arch_hw_breakpoint *brk)
764 {
765 	preempt_disable();
766 	__set_breakpoint(i, brk);
767 	preempt_enable();
768 }
769 
770 static void set_debug_reg_defaults(struct thread_struct *thread)
771 {
772 	int i;
773 	struct arch_hw_breakpoint null_brk = {0};
774 
775 	for (i = 0; i < nr_wp_slots(); i++) {
776 		thread->hw_brk[i] = null_brk;
777 		if (ppc_breakpoint_available())
778 			set_breakpoint(i, &thread->hw_brk[i]);
779 	}
780 }
781 
782 static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
783 				struct arch_hw_breakpoint *b)
784 {
785 	if (a->address != b->address)
786 		return false;
787 	if (a->type != b->type)
788 		return false;
789 	if (a->len != b->len)
790 		return false;
791 	/* no need to check hw_len. it's calculated from address and len */
792 	return true;
793 }
794 
795 static void switch_hw_breakpoint(struct task_struct *new)
796 {
797 	int i;
798 
799 	for (i = 0; i < nr_wp_slots(); i++) {
800 		if (likely(hw_brk_match(this_cpu_ptr(&current_brk[i]),
801 					&new->thread.hw_brk[i])))
802 			continue;
803 
804 		__set_breakpoint(i, &new->thread.hw_brk[i]);
805 	}
806 }
807 #endif /* !CONFIG_HAVE_HW_BREAKPOINT */
808 #endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
809 
810 static inline int set_dabr(struct arch_hw_breakpoint *brk)
811 {
812 	unsigned long dabr, dabrx;
813 
814 	dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
815 	dabrx = ((brk->type >> 3) & 0x7);
816 
817 	if (ppc_md.set_dabr)
818 		return ppc_md.set_dabr(dabr, dabrx);
819 
820 	if (IS_ENABLED(CONFIG_PPC_ADV_DEBUG_REGS)) {
821 		mtspr(SPRN_DAC1, dabr);
822 		if (IS_ENABLED(CONFIG_PPC_47x))
823 			isync();
824 		return 0;
825 	} else if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
826 		mtspr(SPRN_DABR, dabr);
827 		if (cpu_has_feature(CPU_FTR_DABRX))
828 			mtspr(SPRN_DABRX, dabrx);
829 		return 0;
830 	} else {
831 		return -EINVAL;
832 	}
833 }
834 
835 static inline int set_breakpoint_8xx(struct arch_hw_breakpoint *brk)
836 {
837 	unsigned long lctrl1 = LCTRL1_CTE_GT | LCTRL1_CTF_LT | LCTRL1_CRWE_RW |
838 			       LCTRL1_CRWF_RW;
839 	unsigned long lctrl2 = LCTRL2_LW0EN | LCTRL2_LW0LADC | LCTRL2_SLW0EN;
840 	unsigned long start_addr = ALIGN_DOWN(brk->address, HW_BREAKPOINT_SIZE);
841 	unsigned long end_addr = ALIGN(brk->address + brk->len, HW_BREAKPOINT_SIZE);
842 
843 	if (start_addr == 0)
844 		lctrl2 |= LCTRL2_LW0LA_F;
845 	else if (end_addr == 0)
846 		lctrl2 |= LCTRL2_LW0LA_E;
847 	else
848 		lctrl2 |= LCTRL2_LW0LA_EandF;
849 
850 	mtspr(SPRN_LCTRL2, 0);
851 
852 	if ((brk->type & HW_BRK_TYPE_RDWR) == 0)
853 		return 0;
854 
855 	if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
856 		lctrl1 |= LCTRL1_CRWE_RO | LCTRL1_CRWF_RO;
857 	if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
858 		lctrl1 |= LCTRL1_CRWE_WO | LCTRL1_CRWF_WO;
859 
860 	mtspr(SPRN_CMPE, start_addr - 1);
861 	mtspr(SPRN_CMPF, end_addr);
862 	mtspr(SPRN_LCTRL1, lctrl1);
863 	mtspr(SPRN_LCTRL2, lctrl2);
864 
865 	return 0;
866 }
867 
868 void __set_breakpoint(int nr, struct arch_hw_breakpoint *brk)
869 {
870 	memcpy(this_cpu_ptr(&current_brk[nr]), brk, sizeof(*brk));
871 
872 	if (dawr_enabled())
873 		// Power8 or later
874 		set_dawr(nr, brk);
875 	else if (IS_ENABLED(CONFIG_PPC_8xx))
876 		set_breakpoint_8xx(brk);
877 	else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
878 		// Power7 or earlier
879 		set_dabr(brk);
880 	else
881 		// Shouldn't happen due to higher level checks
882 		WARN_ON_ONCE(1);
883 }
884 
885 /* Check if we have DAWR or DABR hardware */
886 bool ppc_breakpoint_available(void)
887 {
888 	if (dawr_enabled())
889 		return true; /* POWER8 DAWR or POWER9 forced DAWR */
890 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
891 		return false; /* POWER9 with DAWR disabled */
892 	/* DABR: Everything but POWER8 and POWER9 */
893 	return true;
894 }
895 EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
896 
897 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
898 
899 static inline bool tm_enabled(struct task_struct *tsk)
900 {
901 	return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
902 }
903 
904 static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
905 {
906 	/*
907 	 * Use the current MSR TM suspended bit to track if we have
908 	 * checkpointed state outstanding.
909 	 * On signal delivery, we'd normally reclaim the checkpointed
910 	 * state to obtain stack pointer (see:get_tm_stackpointer()).
911 	 * This will then directly return to userspace without going
912 	 * through __switch_to(). However, if the stack frame is bad,
913 	 * we need to exit this thread which calls __switch_to() which
914 	 * will again attempt to reclaim the already saved tm state.
915 	 * Hence we need to check that we've not already reclaimed
916 	 * this state.
917 	 * We do this using the current MSR, rather tracking it in
918 	 * some specific thread_struct bit, as it has the additional
919 	 * benefit of checking for a potential TM bad thing exception.
920 	 */
921 	if (!MSR_TM_SUSPENDED(mfmsr()))
922 		return;
923 
924 	giveup_all(container_of(thr, struct task_struct, thread));
925 
926 	tm_reclaim(thr, cause);
927 
928 	/*
929 	 * If we are in a transaction and FP is off then we can't have
930 	 * used FP inside that transaction. Hence the checkpointed
931 	 * state is the same as the live state. We need to copy the
932 	 * live state to the checkpointed state so that when the
933 	 * transaction is restored, the checkpointed state is correct
934 	 * and the aborted transaction sees the correct state. We use
935 	 * ckpt_regs.msr here as that's what tm_reclaim will use to
936 	 * determine if it's going to write the checkpointed state or
937 	 * not. So either this will write the checkpointed registers,
938 	 * or reclaim will. Similarly for VMX.
939 	 */
940 	if ((thr->ckpt_regs.msr & MSR_FP) == 0)
941 		memcpy(&thr->ckfp_state, &thr->fp_state,
942 		       sizeof(struct thread_fp_state));
943 	if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
944 		memcpy(&thr->ckvr_state, &thr->vr_state,
945 		       sizeof(struct thread_vr_state));
946 }
947 
948 void tm_reclaim_current(uint8_t cause)
949 {
950 	tm_enable();
951 	tm_reclaim_thread(&current->thread, cause);
952 }
953 
954 static inline void tm_reclaim_task(struct task_struct *tsk)
955 {
956 	/* We have to work out if we're switching from/to a task that's in the
957 	 * middle of a transaction.
958 	 *
959 	 * In switching we need to maintain a 2nd register state as
960 	 * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
961 	 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
962 	 * ckvr_state
963 	 *
964 	 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
965 	 */
966 	struct thread_struct *thr = &tsk->thread;
967 
968 	if (!thr->regs)
969 		return;
970 
971 	if (!MSR_TM_ACTIVE(thr->regs->msr))
972 		goto out_and_saveregs;
973 
974 	WARN_ON(tm_suspend_disabled);
975 
976 	TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
977 		 "ccr=%lx, msr=%lx, trap=%lx)\n",
978 		 tsk->pid, thr->regs->nip,
979 		 thr->regs->ccr, thr->regs->msr,
980 		 thr->regs->trap);
981 
982 	tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
983 
984 	TM_DEBUG("--- tm_reclaim on pid %d complete\n",
985 		 tsk->pid);
986 
987 out_and_saveregs:
988 	/* Always save the regs here, even if a transaction's not active.
989 	 * This context-switches a thread's TM info SPRs.  We do it here to
990 	 * be consistent with the restore path (in recheckpoint) which
991 	 * cannot happen later in _switch().
992 	 */
993 	tm_save_sprs(thr);
994 }
995 
996 extern void __tm_recheckpoint(struct thread_struct *thread);
997 
998 void tm_recheckpoint(struct thread_struct *thread)
999 {
1000 	unsigned long flags;
1001 
1002 	if (!(thread->regs->msr & MSR_TM))
1003 		return;
1004 
1005 	/* We really can't be interrupted here as the TEXASR registers can't
1006 	 * change and later in the trecheckpoint code, we have a userspace R1.
1007 	 * So let's hard disable over this region.
1008 	 */
1009 	local_irq_save(flags);
1010 	hard_irq_disable();
1011 
1012 	/* The TM SPRs are restored here, so that TEXASR.FS can be set
1013 	 * before the trecheckpoint and no explosion occurs.
1014 	 */
1015 	tm_restore_sprs(thread);
1016 
1017 	__tm_recheckpoint(thread);
1018 
1019 	local_irq_restore(flags);
1020 }
1021 
1022 static inline void tm_recheckpoint_new_task(struct task_struct *new)
1023 {
1024 	if (!cpu_has_feature(CPU_FTR_TM))
1025 		return;
1026 
1027 	/* Recheckpoint the registers of the thread we're about to switch to.
1028 	 *
1029 	 * If the task was using FP, we non-lazily reload both the original and
1030 	 * the speculative FP register states.  This is because the kernel
1031 	 * doesn't see if/when a TM rollback occurs, so if we take an FP
1032 	 * unavailable later, we are unable to determine which set of FP regs
1033 	 * need to be restored.
1034 	 */
1035 	if (!tm_enabled(new))
1036 		return;
1037 
1038 	if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
1039 		tm_restore_sprs(&new->thread);
1040 		return;
1041 	}
1042 	/* Recheckpoint to restore original checkpointed register state. */
1043 	TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
1044 		 new->pid, new->thread.regs->msr);
1045 
1046 	tm_recheckpoint(&new->thread);
1047 
1048 	/*
1049 	 * The checkpointed state has been restored but the live state has
1050 	 * not, ensure all the math functionality is turned off to trigger
1051 	 * restore_math() to reload.
1052 	 */
1053 	new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1054 
1055 	TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1056 		 "(kernel msr 0x%lx)\n",
1057 		 new->pid, mfmsr());
1058 }
1059 
1060 static inline void __switch_to_tm(struct task_struct *prev,
1061 		struct task_struct *new)
1062 {
1063 	if (cpu_has_feature(CPU_FTR_TM)) {
1064 		if (tm_enabled(prev) || tm_enabled(new))
1065 			tm_enable();
1066 
1067 		if (tm_enabled(prev)) {
1068 			prev->thread.load_tm++;
1069 			tm_reclaim_task(prev);
1070 			if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1071 				prev->thread.regs->msr &= ~MSR_TM;
1072 		}
1073 
1074 		tm_recheckpoint_new_task(new);
1075 	}
1076 }
1077 
1078 /*
1079  * This is called if we are on the way out to userspace and the
1080  * TIF_RESTORE_TM flag is set.  It checks if we need to reload
1081  * FP and/or vector state and does so if necessary.
1082  * If userspace is inside a transaction (whether active or
1083  * suspended) and FP/VMX/VSX instructions have ever been enabled
1084  * inside that transaction, then we have to keep them enabled
1085  * and keep the FP/VMX/VSX state loaded while ever the transaction
1086  * continues.  The reason is that if we didn't, and subsequently
1087  * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1088  * we don't know whether it's the same transaction, and thus we
1089  * don't know which of the checkpointed state and the transactional
1090  * state to use.
1091  */
1092 void restore_tm_state(struct pt_regs *regs)
1093 {
1094 	unsigned long msr_diff;
1095 
1096 	/*
1097 	 * This is the only moment we should clear TIF_RESTORE_TM as
1098 	 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1099 	 * again, anything else could lead to an incorrect ckpt_msr being
1100 	 * saved and therefore incorrect signal contexts.
1101 	 */
1102 	clear_thread_flag(TIF_RESTORE_TM);
1103 	if (!MSR_TM_ACTIVE(regs->msr))
1104 		return;
1105 
1106 	msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1107 	msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1108 
1109 	/* Ensure that restore_math() will restore */
1110 	if (msr_diff & MSR_FP)
1111 		current->thread.load_fp = 1;
1112 #ifdef CONFIG_ALTIVEC
1113 	if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1114 		current->thread.load_vec = 1;
1115 #endif
1116 	restore_math(regs);
1117 
1118 	regs_set_return_msr(regs, regs->msr | msr_diff);
1119 }
1120 
1121 #else /* !CONFIG_PPC_TRANSACTIONAL_MEM */
1122 #define tm_recheckpoint_new_task(new)
1123 #define __switch_to_tm(prev, new)
1124 void tm_reclaim_current(uint8_t cause) {}
1125 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1126 
1127 static inline void save_sprs(struct thread_struct *t)
1128 {
1129 #ifdef CONFIG_ALTIVEC
1130 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
1131 		t->vrsave = mfspr(SPRN_VRSAVE);
1132 #endif
1133 #ifdef CONFIG_SPE
1134 	if (cpu_has_feature(CPU_FTR_SPE))
1135 		t->spefscr = mfspr(SPRN_SPEFSCR);
1136 #endif
1137 #ifdef CONFIG_PPC_BOOK3S_64
1138 	if (cpu_has_feature(CPU_FTR_DSCR))
1139 		t->dscr = mfspr(SPRN_DSCR);
1140 
1141 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1142 		t->bescr = mfspr(SPRN_BESCR);
1143 		t->ebbhr = mfspr(SPRN_EBBHR);
1144 		t->ebbrr = mfspr(SPRN_EBBRR);
1145 
1146 		t->fscr = mfspr(SPRN_FSCR);
1147 
1148 		/*
1149 		 * Note that the TAR is not available for use in the kernel.
1150 		 * (To provide this, the TAR should be backed up/restored on
1151 		 * exception entry/exit instead, and be in pt_regs.  FIXME,
1152 		 * this should be in pt_regs anyway (for debug).)
1153 		 */
1154 		t->tar = mfspr(SPRN_TAR);
1155 	}
1156 #endif
1157 }
1158 
1159 static inline void restore_sprs(struct thread_struct *old_thread,
1160 				struct thread_struct *new_thread)
1161 {
1162 #ifdef CONFIG_ALTIVEC
1163 	if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1164 	    old_thread->vrsave != new_thread->vrsave)
1165 		mtspr(SPRN_VRSAVE, new_thread->vrsave);
1166 #endif
1167 #ifdef CONFIG_SPE
1168 	if (cpu_has_feature(CPU_FTR_SPE) &&
1169 	    old_thread->spefscr != new_thread->spefscr)
1170 		mtspr(SPRN_SPEFSCR, new_thread->spefscr);
1171 #endif
1172 #ifdef CONFIG_PPC_BOOK3S_64
1173 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1174 		u64 dscr = get_paca()->dscr_default;
1175 		if (new_thread->dscr_inherit)
1176 			dscr = new_thread->dscr;
1177 
1178 		if (old_thread->dscr != dscr)
1179 			mtspr(SPRN_DSCR, dscr);
1180 	}
1181 
1182 	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1183 		if (old_thread->bescr != new_thread->bescr)
1184 			mtspr(SPRN_BESCR, new_thread->bescr);
1185 		if (old_thread->ebbhr != new_thread->ebbhr)
1186 			mtspr(SPRN_EBBHR, new_thread->ebbhr);
1187 		if (old_thread->ebbrr != new_thread->ebbrr)
1188 			mtspr(SPRN_EBBRR, new_thread->ebbrr);
1189 
1190 		if (old_thread->fscr != new_thread->fscr)
1191 			mtspr(SPRN_FSCR, new_thread->fscr);
1192 
1193 		if (old_thread->tar != new_thread->tar)
1194 			mtspr(SPRN_TAR, new_thread->tar);
1195 	}
1196 
1197 	if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1198 	    old_thread->tidr != new_thread->tidr)
1199 		mtspr(SPRN_TIDR, new_thread->tidr);
1200 #endif
1201 
1202 }
1203 
1204 struct task_struct *__switch_to(struct task_struct *prev,
1205 	struct task_struct *new)
1206 {
1207 	struct thread_struct *new_thread, *old_thread;
1208 	struct task_struct *last;
1209 #ifdef CONFIG_PPC_BOOK3S_64
1210 	struct ppc64_tlb_batch *batch;
1211 #endif
1212 
1213 	new_thread = &new->thread;
1214 	old_thread = &current->thread;
1215 
1216 	WARN_ON(!irqs_disabled());
1217 
1218 #ifdef CONFIG_PPC_BOOK3S_64
1219 	batch = this_cpu_ptr(&ppc64_tlb_batch);
1220 	if (batch->active) {
1221 		current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1222 		if (batch->index)
1223 			__flush_tlb_pending(batch);
1224 		batch->active = 0;
1225 	}
1226 
1227 	/*
1228 	 * On POWER9 the copy-paste buffer can only paste into
1229 	 * foreign real addresses, so unprivileged processes can not
1230 	 * see the data or use it in any way unless they have
1231 	 * foreign real mappings. If the new process has the foreign
1232 	 * real address mappings, we must issue a cp_abort to clear
1233 	 * any state and prevent snooping, corruption or a covert
1234 	 * channel. ISA v3.1 supports paste into local memory.
1235 	 */
1236 	if (new->mm && (cpu_has_feature(CPU_FTR_ARCH_31) ||
1237 			atomic_read(&new->mm->context.vas_windows)))
1238 		asm volatile(PPC_CP_ABORT);
1239 #endif /* CONFIG_PPC_BOOK3S_64 */
1240 
1241 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
1242 	switch_booke_debug_regs(&new->thread.debug);
1243 #else
1244 /*
1245  * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1246  * schedule DABR
1247  */
1248 #ifndef CONFIG_HAVE_HW_BREAKPOINT
1249 	switch_hw_breakpoint(new);
1250 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1251 #endif
1252 
1253 	/*
1254 	 * We need to save SPRs before treclaim/trecheckpoint as these will
1255 	 * change a number of them.
1256 	 */
1257 	save_sprs(&prev->thread);
1258 
1259 	/* Save FPU, Altivec, VSX and SPE state */
1260 	giveup_all(prev);
1261 
1262 	__switch_to_tm(prev, new);
1263 
1264 	if (!radix_enabled()) {
1265 		/*
1266 		 * We can't take a PMU exception inside _switch() since there
1267 		 * is a window where the kernel stack SLB and the kernel stack
1268 		 * are out of sync. Hard disable here.
1269 		 */
1270 		hard_irq_disable();
1271 	}
1272 
1273 	/*
1274 	 * Call restore_sprs() and set_return_regs_changed() before calling
1275 	 * _switch(). If we move it after _switch() then we miss out on calling
1276 	 * it for new tasks. The reason for this is we manually create a stack
1277 	 * frame for new tasks that directly returns through ret_from_fork() or
1278 	 * ret_from_kernel_thread(). See copy_thread() for details.
1279 	 */
1280 	restore_sprs(old_thread, new_thread);
1281 
1282 	set_return_regs_changed(); /* _switch changes stack (and regs) */
1283 
1284 #ifdef CONFIG_PPC32
1285 	kuap_assert_locked();
1286 #endif
1287 	last = _switch(old_thread, new_thread);
1288 
1289 	/*
1290 	 * Nothing after _switch will be run for newly created tasks,
1291 	 * because they switch directly to ret_from_fork/ret_from_kernel_thread
1292 	 * etc. Code added here should have a comment explaining why that is
1293 	 * okay.
1294 	 */
1295 
1296 #ifdef CONFIG_PPC_BOOK3S_64
1297 	/*
1298 	 * This applies to a process that was context switched while inside
1299 	 * arch_enter_lazy_mmu_mode(), to re-activate the batch that was
1300 	 * deactivated above, before _switch(). This will never be the case
1301 	 * for new tasks.
1302 	 */
1303 	if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1304 		current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1305 		batch = this_cpu_ptr(&ppc64_tlb_batch);
1306 		batch->active = 1;
1307 	}
1308 
1309 	/*
1310 	 * Math facilities are masked out of the child MSR in copy_thread.
1311 	 * A new task does not need to restore_math because it will
1312 	 * demand fault them.
1313 	 */
1314 	if (current->thread.regs)
1315 		restore_math(current->thread.regs);
1316 #endif /* CONFIG_PPC_BOOK3S_64 */
1317 
1318 	return last;
1319 }
1320 
1321 #define NR_INSN_TO_PRINT	16
1322 
1323 static void show_instructions(struct pt_regs *regs)
1324 {
1325 	int i;
1326 	unsigned long nip = regs->nip;
1327 	unsigned long pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1328 
1329 	printk("Instruction dump:");
1330 
1331 	/*
1332 	 * If we were executing with the MMU off for instructions, adjust pc
1333 	 * rather than printing XXXXXXXX.
1334 	 */
1335 	if (!IS_ENABLED(CONFIG_BOOKE) && !(regs->msr & MSR_IR)) {
1336 		pc = (unsigned long)phys_to_virt(pc);
1337 		nip = (unsigned long)phys_to_virt(regs->nip);
1338 	}
1339 
1340 	for (i = 0; i < NR_INSN_TO_PRINT; i++) {
1341 		int instr;
1342 
1343 		if (!(i % 8))
1344 			pr_cont("\n");
1345 
1346 		if (!__kernel_text_address(pc) ||
1347 		    get_kernel_nofault(instr, (const void *)pc)) {
1348 			pr_cont("XXXXXXXX ");
1349 		} else {
1350 			if (nip == pc)
1351 				pr_cont("<%08x> ", instr);
1352 			else
1353 				pr_cont("%08x ", instr);
1354 		}
1355 
1356 		pc += sizeof(int);
1357 	}
1358 
1359 	pr_cont("\n");
1360 }
1361 
1362 void show_user_instructions(struct pt_regs *regs)
1363 {
1364 	unsigned long pc;
1365 	int n = NR_INSN_TO_PRINT;
1366 	struct seq_buf s;
1367 	char buf[96]; /* enough for 8 times 9 + 2 chars */
1368 
1369 	pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1370 
1371 	seq_buf_init(&s, buf, sizeof(buf));
1372 
1373 	while (n) {
1374 		int i;
1375 
1376 		seq_buf_clear(&s);
1377 
1378 		for (i = 0; i < 8 && n; i++, n--, pc += sizeof(int)) {
1379 			int instr;
1380 
1381 			if (copy_from_user_nofault(&instr, (void __user *)pc,
1382 					sizeof(instr))) {
1383 				seq_buf_printf(&s, "XXXXXXXX ");
1384 				continue;
1385 			}
1386 			seq_buf_printf(&s, regs->nip == pc ? "<%08x> " : "%08x ", instr);
1387 		}
1388 
1389 		if (!seq_buf_has_overflowed(&s))
1390 			pr_info("%s[%d]: code: %s\n", current->comm,
1391 				current->pid, s.buffer);
1392 	}
1393 }
1394 
1395 struct regbit {
1396 	unsigned long bit;
1397 	const char *name;
1398 };
1399 
1400 static struct regbit msr_bits[] = {
1401 #if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1402 	{MSR_SF,	"SF"},
1403 	{MSR_HV,	"HV"},
1404 #endif
1405 	{MSR_VEC,	"VEC"},
1406 	{MSR_VSX,	"VSX"},
1407 #ifdef CONFIG_BOOKE
1408 	{MSR_CE,	"CE"},
1409 #endif
1410 	{MSR_EE,	"EE"},
1411 	{MSR_PR,	"PR"},
1412 	{MSR_FP,	"FP"},
1413 	{MSR_ME,	"ME"},
1414 #ifdef CONFIG_BOOKE
1415 	{MSR_DE,	"DE"},
1416 #else
1417 	{MSR_SE,	"SE"},
1418 	{MSR_BE,	"BE"},
1419 #endif
1420 	{MSR_IR,	"IR"},
1421 	{MSR_DR,	"DR"},
1422 	{MSR_PMM,	"PMM"},
1423 #ifndef CONFIG_BOOKE
1424 	{MSR_RI,	"RI"},
1425 	{MSR_LE,	"LE"},
1426 #endif
1427 	{0,		NULL}
1428 };
1429 
1430 static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1431 {
1432 	const char *s = "";
1433 
1434 	for (; bits->bit; ++bits)
1435 		if (val & bits->bit) {
1436 			pr_cont("%s%s", s, bits->name);
1437 			s = sep;
1438 		}
1439 }
1440 
1441 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1442 static struct regbit msr_tm_bits[] = {
1443 	{MSR_TS_T,	"T"},
1444 	{MSR_TS_S,	"S"},
1445 	{MSR_TM,	"E"},
1446 	{0,		NULL}
1447 };
1448 
1449 static void print_tm_bits(unsigned long val)
1450 {
1451 /*
1452  * This only prints something if at least one of the TM bit is set.
1453  * Inside the TM[], the output means:
1454  *   E: Enabled		(bit 32)
1455  *   S: Suspended	(bit 33)
1456  *   T: Transactional	(bit 34)
1457  */
1458 	if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1459 		pr_cont(",TM[");
1460 		print_bits(val, msr_tm_bits, "");
1461 		pr_cont("]");
1462 	}
1463 }
1464 #else
1465 static void print_tm_bits(unsigned long val) {}
1466 #endif
1467 
1468 static void print_msr_bits(unsigned long val)
1469 {
1470 	pr_cont("<");
1471 	print_bits(val, msr_bits, ",");
1472 	print_tm_bits(val);
1473 	pr_cont(">");
1474 }
1475 
1476 #ifdef CONFIG_PPC64
1477 #define REG		"%016lx"
1478 #define REGS_PER_LINE	4
1479 #else
1480 #define REG		"%08lx"
1481 #define REGS_PER_LINE	8
1482 #endif
1483 
1484 static void __show_regs(struct pt_regs *regs)
1485 {
1486 	int i, trap;
1487 
1488 	printk("NIP:  "REG" LR: "REG" CTR: "REG"\n",
1489 	       regs->nip, regs->link, regs->ctr);
1490 	printk("REGS: %px TRAP: %04lx   %s  (%s)\n",
1491 	       regs, regs->trap, print_tainted(), init_utsname()->release);
1492 	printk("MSR:  "REG" ", regs->msr);
1493 	print_msr_bits(regs->msr);
1494 	pr_cont("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1495 	trap = TRAP(regs);
1496 	if (!trap_is_syscall(regs) && cpu_has_feature(CPU_FTR_CFAR))
1497 		pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1498 	if (trap == INTERRUPT_MACHINE_CHECK ||
1499 	    trap == INTERRUPT_DATA_STORAGE ||
1500 	    trap == INTERRUPT_ALIGNMENT) {
1501 		if (IS_ENABLED(CONFIG_4xx) || IS_ENABLED(CONFIG_BOOKE))
1502 			pr_cont("DEAR: "REG" ESR: "REG" ", regs->dear, regs->esr);
1503 		else
1504 			pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1505 	}
1506 
1507 #ifdef CONFIG_PPC64
1508 	pr_cont("IRQMASK: %lx ", regs->softe);
1509 #endif
1510 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1511 	if (MSR_TM_ACTIVE(regs->msr))
1512 		pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1513 #endif
1514 
1515 	for (i = 0;  i < 32;  i++) {
1516 		if ((i % REGS_PER_LINE) == 0)
1517 			pr_cont("\nGPR%02d: ", i);
1518 		pr_cont(REG " ", regs->gpr[i]);
1519 	}
1520 	pr_cont("\n");
1521 	/*
1522 	 * Lookup NIP late so we have the best change of getting the
1523 	 * above info out without failing
1524 	 */
1525 	if (IS_ENABLED(CONFIG_KALLSYMS)) {
1526 		printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1527 		printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1528 	}
1529 }
1530 
1531 void show_regs(struct pt_regs *regs)
1532 {
1533 	show_regs_print_info(KERN_DEFAULT);
1534 	__show_regs(regs);
1535 	show_stack(current, (unsigned long *) regs->gpr[1], KERN_DEFAULT);
1536 	if (!user_mode(regs))
1537 		show_instructions(regs);
1538 }
1539 
1540 void flush_thread(void)
1541 {
1542 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1543 	flush_ptrace_hw_breakpoint(current);
1544 #else /* CONFIG_HAVE_HW_BREAKPOINT */
1545 	set_debug_reg_defaults(&current->thread);
1546 #endif /* CONFIG_HAVE_HW_BREAKPOINT */
1547 }
1548 
1549 void arch_setup_new_exec(void)
1550 {
1551 
1552 #ifdef CONFIG_PPC_BOOK3S_64
1553 	if (!radix_enabled())
1554 		hash__setup_new_exec();
1555 #endif
1556 	/*
1557 	 * If we exec out of a kernel thread then thread.regs will not be
1558 	 * set.  Do it now.
1559 	 */
1560 	if (!current->thread.regs) {
1561 		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1562 		current->thread.regs = regs - 1;
1563 	}
1564 
1565 #ifdef CONFIG_PPC_MEM_KEYS
1566 	current->thread.regs->amr  = default_amr;
1567 	current->thread.regs->iamr  = default_iamr;
1568 #endif
1569 }
1570 
1571 #ifdef CONFIG_PPC64
1572 /**
1573  * Assign a TIDR (thread ID) for task @t and set it in the thread
1574  * structure. For now, we only support setting TIDR for 'current' task.
1575  *
1576  * Since the TID value is a truncated form of it PID, it is possible
1577  * (but unlikely) for 2 threads to have the same TID. In the unlikely event
1578  * that 2 threads share the same TID and are waiting, one of the following
1579  * cases will happen:
1580  *
1581  * 1. The correct thread is running, the wrong thread is not
1582  * In this situation, the correct thread is woken and proceeds to pass it's
1583  * condition check.
1584  *
1585  * 2. Neither threads are running
1586  * In this situation, neither thread will be woken. When scheduled, the waiting
1587  * threads will execute either a wait, which will return immediately, followed
1588  * by a condition check, which will pass for the correct thread and fail
1589  * for the wrong thread, or they will execute the condition check immediately.
1590  *
1591  * 3. The wrong thread is running, the correct thread is not
1592  * The wrong thread will be woken, but will fail it's condition check and
1593  * re-execute wait. The correct thread, when scheduled, will execute either
1594  * it's condition check (which will pass), or wait, which returns immediately
1595  * when called the first time after the thread is scheduled, followed by it's
1596  * condition check (which will pass).
1597  *
1598  * 4. Both threads are running
1599  * Both threads will be woken. The wrong thread will fail it's condition check
1600  * and execute another wait, while the correct thread will pass it's condition
1601  * check.
1602  *
1603  * @t: the task to set the thread ID for
1604  */
1605 int set_thread_tidr(struct task_struct *t)
1606 {
1607 	if (!cpu_has_feature(CPU_FTR_P9_TIDR))
1608 		return -EINVAL;
1609 
1610 	if (t != current)
1611 		return -EINVAL;
1612 
1613 	if (t->thread.tidr)
1614 		return 0;
1615 
1616 	t->thread.tidr = (u16)task_pid_nr(t);
1617 	mtspr(SPRN_TIDR, t->thread.tidr);
1618 
1619 	return 0;
1620 }
1621 EXPORT_SYMBOL_GPL(set_thread_tidr);
1622 
1623 #endif /* CONFIG_PPC64 */
1624 
1625 void
1626 release_thread(struct task_struct *t)
1627 {
1628 }
1629 
1630 /*
1631  * this gets called so that we can store coprocessor state into memory and
1632  * copy the current task into the new thread.
1633  */
1634 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1635 {
1636 	flush_all_to_thread(src);
1637 	/*
1638 	 * Flush TM state out so we can copy it.  __switch_to_tm() does this
1639 	 * flush but it removes the checkpointed state from the current CPU and
1640 	 * transitions the CPU out of TM mode.  Hence we need to call
1641 	 * tm_recheckpoint_new_task() (on the same task) to restore the
1642 	 * checkpointed state back and the TM mode.
1643 	 *
1644 	 * Can't pass dst because it isn't ready. Doesn't matter, passing
1645 	 * dst is only important for __switch_to()
1646 	 */
1647 	__switch_to_tm(src, src);
1648 
1649 	*dst = *src;
1650 
1651 	clear_task_ebb(dst);
1652 
1653 	return 0;
1654 }
1655 
1656 static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1657 {
1658 #ifdef CONFIG_PPC_BOOK3S_64
1659 	unsigned long sp_vsid;
1660 	unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1661 
1662 	if (radix_enabled())
1663 		return;
1664 
1665 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1666 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1667 			<< SLB_VSID_SHIFT_1T;
1668 	else
1669 		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1670 			<< SLB_VSID_SHIFT;
1671 	sp_vsid |= SLB_VSID_KERNEL | llp;
1672 	p->thread.ksp_vsid = sp_vsid;
1673 #endif
1674 }
1675 
1676 /*
1677  * Copy a thread..
1678  */
1679 
1680 /*
1681  * Copy architecture-specific thread state
1682  */
1683 int copy_thread(unsigned long clone_flags, unsigned long usp,
1684 		unsigned long kthread_arg, struct task_struct *p,
1685 		unsigned long tls)
1686 {
1687 	struct pt_regs *childregs, *kregs;
1688 	extern void ret_from_fork(void);
1689 	extern void ret_from_fork_scv(void);
1690 	extern void ret_from_kernel_thread(void);
1691 	void (*f)(void);
1692 	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1693 	struct thread_info *ti = task_thread_info(p);
1694 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1695 	int i;
1696 #endif
1697 
1698 	klp_init_thread_info(p);
1699 
1700 	/* Copy registers */
1701 	sp -= sizeof(struct pt_regs);
1702 	childregs = (struct pt_regs *) sp;
1703 	if (unlikely(p->flags & (PF_KTHREAD | PF_IO_WORKER))) {
1704 		/* kernel thread */
1705 		memset(childregs, 0, sizeof(struct pt_regs));
1706 		childregs->gpr[1] = sp + sizeof(struct pt_regs);
1707 		/* function */
1708 		if (usp)
1709 			childregs->gpr[14] = ppc_function_entry((void *)usp);
1710 #ifdef CONFIG_PPC64
1711 		clear_tsk_thread_flag(p, TIF_32BIT);
1712 		childregs->softe = IRQS_ENABLED;
1713 #endif
1714 		childregs->gpr[15] = kthread_arg;
1715 		p->thread.regs = NULL;	/* no user register state */
1716 		ti->flags |= _TIF_RESTOREALL;
1717 		f = ret_from_kernel_thread;
1718 	} else {
1719 		/* user thread */
1720 		struct pt_regs *regs = current_pt_regs();
1721 		*childregs = *regs;
1722 		if (usp)
1723 			childregs->gpr[1] = usp;
1724 		p->thread.regs = childregs;
1725 		/* 64s sets this in ret_from_fork */
1726 		if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64))
1727 			childregs->gpr[3] = 0;  /* Result from fork() */
1728 		if (clone_flags & CLONE_SETTLS) {
1729 			if (!is_32bit_task())
1730 				childregs->gpr[13] = tls;
1731 			else
1732 				childregs->gpr[2] = tls;
1733 		}
1734 
1735 		if (trap_is_scv(regs))
1736 			f = ret_from_fork_scv;
1737 		else
1738 			f = ret_from_fork;
1739 	}
1740 	childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1741 	sp -= STACK_FRAME_OVERHEAD;
1742 
1743 	/*
1744 	 * The way this works is that at some point in the future
1745 	 * some task will call _switch to switch to the new task.
1746 	 * That will pop off the stack frame created below and start
1747 	 * the new task running at ret_from_fork.  The new task will
1748 	 * do some house keeping and then return from the fork or clone
1749 	 * system call, using the stack frame created above.
1750 	 */
1751 	((unsigned long *)sp)[0] = 0;
1752 	sp -= sizeof(struct pt_regs);
1753 	kregs = (struct pt_regs *) sp;
1754 	sp -= STACK_FRAME_OVERHEAD;
1755 	p->thread.ksp = sp;
1756 #ifdef CONFIG_HAVE_HW_BREAKPOINT
1757 	for (i = 0; i < nr_wp_slots(); i++)
1758 		p->thread.ptrace_bps[i] = NULL;
1759 #endif
1760 
1761 #ifdef CONFIG_PPC_FPU_REGS
1762 	p->thread.fp_save_area = NULL;
1763 #endif
1764 #ifdef CONFIG_ALTIVEC
1765 	p->thread.vr_save_area = NULL;
1766 #endif
1767 #if defined(CONFIG_PPC_BOOK3S_32) && defined(CONFIG_PPC_KUAP)
1768 	p->thread.kuap = KUAP_NONE;
1769 #endif
1770 
1771 	setup_ksp_vsid(p, sp);
1772 
1773 #ifdef CONFIG_PPC64
1774 	if (cpu_has_feature(CPU_FTR_DSCR)) {
1775 		p->thread.dscr_inherit = current->thread.dscr_inherit;
1776 		p->thread.dscr = mfspr(SPRN_DSCR);
1777 	}
1778 	if (cpu_has_feature(CPU_FTR_HAS_PPR))
1779 		childregs->ppr = DEFAULT_PPR;
1780 
1781 	p->thread.tidr = 0;
1782 #endif
1783 	/*
1784 	 * Run with the current AMR value of the kernel
1785 	 */
1786 #ifdef CONFIG_PPC_PKEY
1787 	if (mmu_has_feature(MMU_FTR_BOOK3S_KUAP))
1788 		kregs->amr = AMR_KUAP_BLOCKED;
1789 
1790 	if (mmu_has_feature(MMU_FTR_BOOK3S_KUEP))
1791 		kregs->iamr = AMR_KUEP_BLOCKED;
1792 #endif
1793 	kregs->nip = ppc_function_entry(f);
1794 	return 0;
1795 }
1796 
1797 void preload_new_slb_context(unsigned long start, unsigned long sp);
1798 
1799 /*
1800  * Set up a thread for executing a new program
1801  */
1802 void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1803 {
1804 #ifdef CONFIG_PPC64
1805 	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
1806 
1807 	if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && !radix_enabled())
1808 		preload_new_slb_context(start, sp);
1809 #endif
1810 
1811 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1812 	/*
1813 	 * Clear any transactional state, we're exec()ing. The cause is
1814 	 * not important as there will never be a recheckpoint so it's not
1815 	 * user visible.
1816 	 */
1817 	if (MSR_TM_SUSPENDED(mfmsr()))
1818 		tm_reclaim_current(0);
1819 #endif
1820 
1821 	memset(regs->gpr, 0, sizeof(regs->gpr));
1822 	regs->ctr = 0;
1823 	regs->link = 0;
1824 	regs->xer = 0;
1825 	regs->ccr = 0;
1826 	regs->gpr[1] = sp;
1827 
1828 #ifdef CONFIG_PPC32
1829 	regs->mq = 0;
1830 	regs->nip = start;
1831 	regs->msr = MSR_USER;
1832 #else
1833 	if (!is_32bit_task()) {
1834 		unsigned long entry;
1835 
1836 		if (is_elf2_task()) {
1837 			/* Look ma, no function descriptors! */
1838 			entry = start;
1839 
1840 			/*
1841 			 * Ulrich says:
1842 			 *   The latest iteration of the ABI requires that when
1843 			 *   calling a function (at its global entry point),
1844 			 *   the caller must ensure r12 holds the entry point
1845 			 *   address (so that the function can quickly
1846 			 *   establish addressability).
1847 			 */
1848 			regs->gpr[12] = start;
1849 			/* Make sure that's restored on entry to userspace. */
1850 			set_thread_flag(TIF_RESTOREALL);
1851 		} else {
1852 			unsigned long toc;
1853 
1854 			/* start is a relocated pointer to the function
1855 			 * descriptor for the elf _start routine.  The first
1856 			 * entry in the function descriptor is the entry
1857 			 * address of _start and the second entry is the TOC
1858 			 * value we need to use.
1859 			 */
1860 			__get_user(entry, (unsigned long __user *)start);
1861 			__get_user(toc, (unsigned long __user *)start+1);
1862 
1863 			/* Check whether the e_entry function descriptor entries
1864 			 * need to be relocated before we can use them.
1865 			 */
1866 			if (load_addr != 0) {
1867 				entry += load_addr;
1868 				toc   += load_addr;
1869 			}
1870 			regs->gpr[2] = toc;
1871 		}
1872 		regs_set_return_ip(regs, entry);
1873 		regs_set_return_msr(regs, MSR_USER64);
1874 	} else {
1875 		regs->gpr[2] = 0;
1876 		regs_set_return_ip(regs, start);
1877 		regs_set_return_msr(regs, MSR_USER32);
1878 	}
1879 
1880 #endif
1881 #ifdef CONFIG_VSX
1882 	current->thread.used_vsr = 0;
1883 #endif
1884 	current->thread.load_slb = 0;
1885 	current->thread.load_fp = 0;
1886 #ifdef CONFIG_PPC_FPU_REGS
1887 	memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1888 	current->thread.fp_save_area = NULL;
1889 #endif
1890 #ifdef CONFIG_ALTIVEC
1891 	memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1892 	current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1893 	current->thread.vr_save_area = NULL;
1894 	current->thread.vrsave = 0;
1895 	current->thread.used_vr = 0;
1896 	current->thread.load_vec = 0;
1897 #endif /* CONFIG_ALTIVEC */
1898 #ifdef CONFIG_SPE
1899 	memset(current->thread.evr, 0, sizeof(current->thread.evr));
1900 	current->thread.acc = 0;
1901 	current->thread.spefscr = 0;
1902 	current->thread.used_spe = 0;
1903 #endif /* CONFIG_SPE */
1904 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1905 	current->thread.tm_tfhar = 0;
1906 	current->thread.tm_texasr = 0;
1907 	current->thread.tm_tfiar = 0;
1908 	current->thread.load_tm = 0;
1909 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1910 }
1911 EXPORT_SYMBOL(start_thread);
1912 
1913 #define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1914 		| PR_FP_EXC_RES | PR_FP_EXC_INV)
1915 
1916 int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1917 {
1918 	struct pt_regs *regs = tsk->thread.regs;
1919 
1920 	/* This is a bit hairy.  If we are an SPE enabled  processor
1921 	 * (have embedded fp) we store the IEEE exception enable flags in
1922 	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
1923 	 * mode (asyn, precise, disabled) for 'Classic' FP. */
1924 	if (val & PR_FP_EXC_SW_ENABLE) {
1925 		if (cpu_has_feature(CPU_FTR_SPE)) {
1926 			/*
1927 			 * When the sticky exception bits are set
1928 			 * directly by userspace, it must call prctl
1929 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1930 			 * in the existing prctl settings) or
1931 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1932 			 * the bits being set).  <fenv.h> functions
1933 			 * saving and restoring the whole
1934 			 * floating-point environment need to do so
1935 			 * anyway to restore the prctl settings from
1936 			 * the saved environment.
1937 			 */
1938 #ifdef CONFIG_SPE
1939 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1940 			tsk->thread.fpexc_mode = val &
1941 				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1942 #endif
1943 			return 0;
1944 		} else {
1945 			return -EINVAL;
1946 		}
1947 	}
1948 
1949 	/* on a CONFIG_SPE this does not hurt us.  The bits that
1950 	 * __pack_fe01 use do not overlap with bits used for
1951 	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1952 	 * on CONFIG_SPE implementations are reserved so writing to
1953 	 * them does not change anything */
1954 	if (val > PR_FP_EXC_PRECISE)
1955 		return -EINVAL;
1956 	tsk->thread.fpexc_mode = __pack_fe01(val);
1957 	if (regs != NULL && (regs->msr & MSR_FP) != 0) {
1958 		regs_set_return_msr(regs, (regs->msr & ~(MSR_FE0|MSR_FE1))
1959 						| tsk->thread.fpexc_mode);
1960 	}
1961 	return 0;
1962 }
1963 
1964 int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1965 {
1966 	unsigned int val = 0;
1967 
1968 	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE) {
1969 		if (cpu_has_feature(CPU_FTR_SPE)) {
1970 			/*
1971 			 * When the sticky exception bits are set
1972 			 * directly by userspace, it must call prctl
1973 			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1974 			 * in the existing prctl settings) or
1975 			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1976 			 * the bits being set).  <fenv.h> functions
1977 			 * saving and restoring the whole
1978 			 * floating-point environment need to do so
1979 			 * anyway to restore the prctl settings from
1980 			 * the saved environment.
1981 			 */
1982 #ifdef CONFIG_SPE
1983 			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1984 			val = tsk->thread.fpexc_mode;
1985 #endif
1986 		} else
1987 			return -EINVAL;
1988 	} else {
1989 		val = __unpack_fe01(tsk->thread.fpexc_mode);
1990 	}
1991 	return put_user(val, (unsigned int __user *) adr);
1992 }
1993 
1994 int set_endian(struct task_struct *tsk, unsigned int val)
1995 {
1996 	struct pt_regs *regs = tsk->thread.regs;
1997 
1998 	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1999 	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
2000 		return -EINVAL;
2001 
2002 	if (regs == NULL)
2003 		return -EINVAL;
2004 
2005 	if (val == PR_ENDIAN_BIG)
2006 		regs_set_return_msr(regs, regs->msr & ~MSR_LE);
2007 	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
2008 		regs_set_return_msr(regs, regs->msr | MSR_LE);
2009 	else
2010 		return -EINVAL;
2011 
2012 	return 0;
2013 }
2014 
2015 int get_endian(struct task_struct *tsk, unsigned long adr)
2016 {
2017 	struct pt_regs *regs = tsk->thread.regs;
2018 	unsigned int val;
2019 
2020 	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
2021 	    !cpu_has_feature(CPU_FTR_REAL_LE))
2022 		return -EINVAL;
2023 
2024 	if (regs == NULL)
2025 		return -EINVAL;
2026 
2027 	if (regs->msr & MSR_LE) {
2028 		if (cpu_has_feature(CPU_FTR_REAL_LE))
2029 			val = PR_ENDIAN_LITTLE;
2030 		else
2031 			val = PR_ENDIAN_PPC_LITTLE;
2032 	} else
2033 		val = PR_ENDIAN_BIG;
2034 
2035 	return put_user(val, (unsigned int __user *)adr);
2036 }
2037 
2038 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
2039 {
2040 	tsk->thread.align_ctl = val;
2041 	return 0;
2042 }
2043 
2044 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
2045 {
2046 	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
2047 }
2048 
2049 static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
2050 				  unsigned long nbytes)
2051 {
2052 	unsigned long stack_page;
2053 	unsigned long cpu = task_cpu(p);
2054 
2055 	stack_page = (unsigned long)hardirq_ctx[cpu];
2056 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2057 		return 1;
2058 
2059 	stack_page = (unsigned long)softirq_ctx[cpu];
2060 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2061 		return 1;
2062 
2063 	return 0;
2064 }
2065 
2066 static inline int valid_emergency_stack(unsigned long sp, struct task_struct *p,
2067 					unsigned long nbytes)
2068 {
2069 #ifdef CONFIG_PPC64
2070 	unsigned long stack_page;
2071 	unsigned long cpu = task_cpu(p);
2072 
2073 	if (!paca_ptrs)
2074 		return 0;
2075 
2076 	stack_page = (unsigned long)paca_ptrs[cpu]->emergency_sp - THREAD_SIZE;
2077 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2078 		return 1;
2079 
2080 # ifdef CONFIG_PPC_BOOK3S_64
2081 	stack_page = (unsigned long)paca_ptrs[cpu]->nmi_emergency_sp - THREAD_SIZE;
2082 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2083 		return 1;
2084 
2085 	stack_page = (unsigned long)paca_ptrs[cpu]->mc_emergency_sp - THREAD_SIZE;
2086 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2087 		return 1;
2088 # endif
2089 #endif
2090 
2091 	return 0;
2092 }
2093 
2094 
2095 int validate_sp(unsigned long sp, struct task_struct *p,
2096 		       unsigned long nbytes)
2097 {
2098 	unsigned long stack_page = (unsigned long)task_stack_page(p);
2099 
2100 	if (sp < THREAD_SIZE)
2101 		return 0;
2102 
2103 	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2104 		return 1;
2105 
2106 	if (valid_irq_stack(sp, p, nbytes))
2107 		return 1;
2108 
2109 	return valid_emergency_stack(sp, p, nbytes);
2110 }
2111 
2112 EXPORT_SYMBOL(validate_sp);
2113 
2114 static unsigned long ___get_wchan(struct task_struct *p)
2115 {
2116 	unsigned long ip, sp;
2117 	int count = 0;
2118 
2119 	sp = p->thread.ksp;
2120 	if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2121 		return 0;
2122 
2123 	do {
2124 		sp = *(unsigned long *)sp;
2125 		if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2126 		    task_is_running(p))
2127 			return 0;
2128 		if (count > 0) {
2129 			ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2130 			if (!in_sched_functions(ip))
2131 				return ip;
2132 		}
2133 	} while (count++ < 16);
2134 	return 0;
2135 }
2136 
2137 unsigned long __get_wchan(struct task_struct *p)
2138 {
2139 	unsigned long ret;
2140 
2141 	if (!try_get_task_stack(p))
2142 		return 0;
2143 
2144 	ret = ___get_wchan(p);
2145 
2146 	put_task_stack(p);
2147 
2148 	return ret;
2149 }
2150 
2151 static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2152 
2153 void __no_sanitize_address show_stack(struct task_struct *tsk,
2154 				      unsigned long *stack,
2155 				      const char *loglvl)
2156 {
2157 	unsigned long sp, ip, lr, newsp;
2158 	int count = 0;
2159 	int firstframe = 1;
2160 	unsigned long ret_addr;
2161 	int ftrace_idx = 0;
2162 
2163 	if (tsk == NULL)
2164 		tsk = current;
2165 
2166 	if (!try_get_task_stack(tsk))
2167 		return;
2168 
2169 	sp = (unsigned long) stack;
2170 	if (sp == 0) {
2171 		if (tsk == current)
2172 			sp = current_stack_frame();
2173 		else
2174 			sp = tsk->thread.ksp;
2175 	}
2176 
2177 	lr = 0;
2178 	printk("%sCall Trace:\n", loglvl);
2179 	do {
2180 		if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2181 			break;
2182 
2183 		stack = (unsigned long *) sp;
2184 		newsp = stack[0];
2185 		ip = stack[STACK_FRAME_LR_SAVE];
2186 		if (!firstframe || ip != lr) {
2187 			printk("%s["REG"] ["REG"] %pS",
2188 				loglvl, sp, ip, (void *)ip);
2189 			ret_addr = ftrace_graph_ret_addr(current,
2190 						&ftrace_idx, ip, stack);
2191 			if (ret_addr != ip)
2192 				pr_cont(" (%pS)", (void *)ret_addr);
2193 			if (firstframe)
2194 				pr_cont(" (unreliable)");
2195 			pr_cont("\n");
2196 		}
2197 		firstframe = 0;
2198 
2199 		/*
2200 		 * See if this is an exception frame.
2201 		 * We look for the "regshere" marker in the current frame.
2202 		 */
2203 		if (validate_sp(sp, tsk, STACK_FRAME_WITH_PT_REGS)
2204 		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2205 			struct pt_regs *regs = (struct pt_regs *)
2206 				(sp + STACK_FRAME_OVERHEAD);
2207 
2208 			lr = regs->link;
2209 			printk("%s--- interrupt: %lx at %pS\n",
2210 			       loglvl, regs->trap, (void *)regs->nip);
2211 			__show_regs(regs);
2212 			printk("%s--- interrupt: %lx\n",
2213 			       loglvl, regs->trap);
2214 
2215 			firstframe = 1;
2216 		}
2217 
2218 		sp = newsp;
2219 	} while (count++ < kstack_depth_to_print);
2220 
2221 	put_task_stack(tsk);
2222 }
2223 
2224 #ifdef CONFIG_PPC64
2225 /* Called with hard IRQs off */
2226 void notrace __ppc64_runlatch_on(void)
2227 {
2228 	struct thread_info *ti = current_thread_info();
2229 
2230 	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2231 		/*
2232 		 * Least significant bit (RUN) is the only writable bit of
2233 		 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2234 		 * earliest ISA where this is the case, but it's convenient.
2235 		 */
2236 		mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2237 	} else {
2238 		unsigned long ctrl;
2239 
2240 		/*
2241 		 * Some architectures (e.g., Cell) have writable fields other
2242 		 * than RUN, so do the read-modify-write.
2243 		 */
2244 		ctrl = mfspr(SPRN_CTRLF);
2245 		ctrl |= CTRL_RUNLATCH;
2246 		mtspr(SPRN_CTRLT, ctrl);
2247 	}
2248 
2249 	ti->local_flags |= _TLF_RUNLATCH;
2250 }
2251 
2252 /* Called with hard IRQs off */
2253 void notrace __ppc64_runlatch_off(void)
2254 {
2255 	struct thread_info *ti = current_thread_info();
2256 
2257 	ti->local_flags &= ~_TLF_RUNLATCH;
2258 
2259 	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2260 		mtspr(SPRN_CTRLT, 0);
2261 	} else {
2262 		unsigned long ctrl;
2263 
2264 		ctrl = mfspr(SPRN_CTRLF);
2265 		ctrl &= ~CTRL_RUNLATCH;
2266 		mtspr(SPRN_CTRLT, ctrl);
2267 	}
2268 }
2269 #endif /* CONFIG_PPC64 */
2270 
2271 unsigned long arch_align_stack(unsigned long sp)
2272 {
2273 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2274 		sp -= get_random_int() & ~PAGE_MASK;
2275 	return sp & ~0xf;
2276 }
2277 
2278 static inline unsigned long brk_rnd(void)
2279 {
2280         unsigned long rnd = 0;
2281 
2282 	/* 8MB for 32bit, 1GB for 64bit */
2283 	if (is_32bit_task())
2284 		rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2285 	else
2286 		rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2287 
2288 	return rnd << PAGE_SHIFT;
2289 }
2290 
2291 unsigned long arch_randomize_brk(struct mm_struct *mm)
2292 {
2293 	unsigned long base = mm->brk;
2294 	unsigned long ret;
2295 
2296 #ifdef CONFIG_PPC_BOOK3S_64
2297 	/*
2298 	 * If we are using 1TB segments and we are allowed to randomise
2299 	 * the heap, we can put it above 1TB so it is backed by a 1TB
2300 	 * segment. Otherwise the heap will be in the bottom 1TB
2301 	 * which always uses 256MB segments and this may result in a
2302 	 * performance penalty. We don't need to worry about radix. For
2303 	 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2304 	 */
2305 	if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2306 		base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2307 #endif
2308 
2309 	ret = PAGE_ALIGN(base + brk_rnd());
2310 
2311 	if (ret < mm->brk)
2312 		return mm->brk;
2313 
2314 	return ret;
2315 }
2316 
2317